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"
38 * - subvol delete -> delete when ref goes to 0? delete limits also?
42 * - copy also limits on subvol creation
44 * - caches fuer ulists
45 * - performance benchmarks
46 * - check all ioctl parameters
50 * one struct for each qgroup, organized in fs_info->qgroup_tree.
58 u64 rfer
; /* referenced */
59 u64 rfer_cmpr
; /* referenced compressed */
60 u64 excl
; /* exclusive */
61 u64 excl_cmpr
; /* exclusive compressed */
66 u64 lim_flags
; /* which limits are set */
73 * reservation tracking
80 struct list_head groups
; /* groups this group is member of */
81 struct list_head members
; /* groups that are members of this group */
82 struct list_head dirty
; /* dirty groups */
83 struct rb_node node
; /* tree of qgroups */
86 * temp variables for accounting operations
93 * glue structure to represent the relations between qgroups.
95 struct btrfs_qgroup_list
{
96 struct list_head next_group
;
97 struct list_head next_member
;
98 struct btrfs_qgroup
*group
;
99 struct btrfs_qgroup
*member
;
102 #define ptr_to_u64(x) ((u64)(uintptr_t)x)
103 #define u64_to_ptr(x) ((struct btrfs_qgroup *)(uintptr_t)x)
106 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
108 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
);
110 /* must be called with qgroup_ioctl_lock held */
111 static struct btrfs_qgroup
*find_qgroup_rb(struct btrfs_fs_info
*fs_info
,
114 struct rb_node
*n
= fs_info
->qgroup_tree
.rb_node
;
115 struct btrfs_qgroup
*qgroup
;
118 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
119 if (qgroup
->qgroupid
< qgroupid
)
121 else if (qgroup
->qgroupid
> qgroupid
)
129 /* must be called with qgroup_lock held */
130 static struct btrfs_qgroup
*add_qgroup_rb(struct btrfs_fs_info
*fs_info
,
133 struct rb_node
**p
= &fs_info
->qgroup_tree
.rb_node
;
134 struct rb_node
*parent
= NULL
;
135 struct btrfs_qgroup
*qgroup
;
139 qgroup
= rb_entry(parent
, struct btrfs_qgroup
, node
);
141 if (qgroup
->qgroupid
< qgroupid
)
143 else if (qgroup
->qgroupid
> qgroupid
)
149 qgroup
= kzalloc(sizeof(*qgroup
), GFP_ATOMIC
);
151 return ERR_PTR(-ENOMEM
);
153 qgroup
->qgroupid
= qgroupid
;
154 INIT_LIST_HEAD(&qgroup
->groups
);
155 INIT_LIST_HEAD(&qgroup
->members
);
156 INIT_LIST_HEAD(&qgroup
->dirty
);
158 rb_link_node(&qgroup
->node
, parent
, p
);
159 rb_insert_color(&qgroup
->node
, &fs_info
->qgroup_tree
);
164 static void __del_qgroup_rb(struct btrfs_qgroup
*qgroup
)
166 struct btrfs_qgroup_list
*list
;
168 list_del(&qgroup
->dirty
);
169 while (!list_empty(&qgroup
->groups
)) {
170 list
= list_first_entry(&qgroup
->groups
,
171 struct btrfs_qgroup_list
, next_group
);
172 list_del(&list
->next_group
);
173 list_del(&list
->next_member
);
177 while (!list_empty(&qgroup
->members
)) {
178 list
= list_first_entry(&qgroup
->members
,
179 struct btrfs_qgroup_list
, next_member
);
180 list_del(&list
->next_group
);
181 list_del(&list
->next_member
);
187 /* must be called with qgroup_lock held */
188 static int del_qgroup_rb(struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
190 struct btrfs_qgroup
*qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
195 rb_erase(&qgroup
->node
, &fs_info
->qgroup_tree
);
196 __del_qgroup_rb(qgroup
);
200 /* must be called with qgroup_lock held */
201 static int add_relation_rb(struct btrfs_fs_info
*fs_info
,
202 u64 memberid
, u64 parentid
)
204 struct btrfs_qgroup
*member
;
205 struct btrfs_qgroup
*parent
;
206 struct btrfs_qgroup_list
*list
;
208 member
= find_qgroup_rb(fs_info
, memberid
);
209 parent
= find_qgroup_rb(fs_info
, parentid
);
210 if (!member
|| !parent
)
213 list
= kzalloc(sizeof(*list
), GFP_ATOMIC
);
217 list
->group
= parent
;
218 list
->member
= member
;
219 list_add_tail(&list
->next_group
, &member
->groups
);
220 list_add_tail(&list
->next_member
, &parent
->members
);
225 /* must be called with qgroup_lock held */
226 static int del_relation_rb(struct btrfs_fs_info
*fs_info
,
227 u64 memberid
, u64 parentid
)
229 struct btrfs_qgroup
*member
;
230 struct btrfs_qgroup
*parent
;
231 struct btrfs_qgroup_list
*list
;
233 member
= find_qgroup_rb(fs_info
, memberid
);
234 parent
= find_qgroup_rb(fs_info
, parentid
);
235 if (!member
|| !parent
)
238 list_for_each_entry(list
, &member
->groups
, next_group
) {
239 if (list
->group
== parent
) {
240 list_del(&list
->next_group
);
241 list_del(&list
->next_member
);
249 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
250 int btrfs_verify_qgroup_counts(struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
253 struct btrfs_qgroup
*qgroup
;
255 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
258 if (qgroup
->rfer
!= rfer
|| qgroup
->excl
!= excl
)
265 * The full config is read in one go, only called from open_ctree()
266 * It doesn't use any locking, as at this point we're still single-threaded
268 int btrfs_read_qgroup_config(struct btrfs_fs_info
*fs_info
)
270 struct btrfs_key key
;
271 struct btrfs_key found_key
;
272 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
273 struct btrfs_path
*path
= NULL
;
274 struct extent_buffer
*l
;
278 u64 rescan_progress
= 0;
280 if (!fs_info
->quota_enabled
)
283 fs_info
->qgroup_ulist
= ulist_alloc(GFP_NOFS
);
284 if (!fs_info
->qgroup_ulist
) {
289 path
= btrfs_alloc_path();
295 /* default this to quota off, in case no status key is found */
296 fs_info
->qgroup_flags
= 0;
299 * pass 1: read status, all qgroup infos and limits
304 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 1);
309 struct btrfs_qgroup
*qgroup
;
311 slot
= path
->slots
[0];
313 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
315 if (found_key
.type
== BTRFS_QGROUP_STATUS_KEY
) {
316 struct btrfs_qgroup_status_item
*ptr
;
318 ptr
= btrfs_item_ptr(l
, slot
,
319 struct btrfs_qgroup_status_item
);
321 if (btrfs_qgroup_status_version(l
, ptr
) !=
322 BTRFS_QGROUP_STATUS_VERSION
) {
324 "old qgroup version, quota disabled");
327 if (btrfs_qgroup_status_generation(l
, ptr
) !=
328 fs_info
->generation
) {
329 flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
331 "qgroup generation mismatch, "
332 "marked as inconsistent");
334 fs_info
->qgroup_flags
= btrfs_qgroup_status_flags(l
,
336 rescan_progress
= btrfs_qgroup_status_rescan(l
, ptr
);
340 if (found_key
.type
!= BTRFS_QGROUP_INFO_KEY
&&
341 found_key
.type
!= BTRFS_QGROUP_LIMIT_KEY
)
344 qgroup
= find_qgroup_rb(fs_info
, found_key
.offset
);
345 if ((qgroup
&& found_key
.type
== BTRFS_QGROUP_INFO_KEY
) ||
346 (!qgroup
&& found_key
.type
== BTRFS_QGROUP_LIMIT_KEY
)) {
347 btrfs_err(fs_info
, "inconsitent qgroup config");
348 flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
351 qgroup
= add_qgroup_rb(fs_info
, found_key
.offset
);
352 if (IS_ERR(qgroup
)) {
353 ret
= PTR_ERR(qgroup
);
357 switch (found_key
.type
) {
358 case BTRFS_QGROUP_INFO_KEY
: {
359 struct btrfs_qgroup_info_item
*ptr
;
361 ptr
= btrfs_item_ptr(l
, slot
,
362 struct btrfs_qgroup_info_item
);
363 qgroup
->rfer
= btrfs_qgroup_info_rfer(l
, ptr
);
364 qgroup
->rfer_cmpr
= btrfs_qgroup_info_rfer_cmpr(l
, ptr
);
365 qgroup
->excl
= btrfs_qgroup_info_excl(l
, ptr
);
366 qgroup
->excl_cmpr
= btrfs_qgroup_info_excl_cmpr(l
, ptr
);
367 /* generation currently unused */
370 case BTRFS_QGROUP_LIMIT_KEY
: {
371 struct btrfs_qgroup_limit_item
*ptr
;
373 ptr
= btrfs_item_ptr(l
, slot
,
374 struct btrfs_qgroup_limit_item
);
375 qgroup
->lim_flags
= btrfs_qgroup_limit_flags(l
, ptr
);
376 qgroup
->max_rfer
= btrfs_qgroup_limit_max_rfer(l
, ptr
);
377 qgroup
->max_excl
= btrfs_qgroup_limit_max_excl(l
, ptr
);
378 qgroup
->rsv_rfer
= btrfs_qgroup_limit_rsv_rfer(l
, ptr
);
379 qgroup
->rsv_excl
= btrfs_qgroup_limit_rsv_excl(l
, ptr
);
384 ret
= btrfs_next_item(quota_root
, path
);
390 btrfs_release_path(path
);
393 * pass 2: read all qgroup relations
396 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
398 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 0);
402 slot
= path
->slots
[0];
404 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
406 if (found_key
.type
!= BTRFS_QGROUP_RELATION_KEY
)
409 if (found_key
.objectid
> found_key
.offset
) {
410 /* parent <- member, not needed to build config */
411 /* FIXME should we omit the key completely? */
415 ret
= add_relation_rb(fs_info
, found_key
.objectid
,
417 if (ret
== -ENOENT
) {
419 "orphan qgroup relation 0x%llx->0x%llx",
420 found_key
.objectid
, found_key
.offset
);
421 ret
= 0; /* ignore the error */
426 ret
= btrfs_next_item(quota_root
, path
);
433 fs_info
->qgroup_flags
|= flags
;
434 if (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
)) {
435 fs_info
->quota_enabled
= 0;
436 fs_info
->pending_quota_state
= 0;
437 } else if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
&&
439 ret
= qgroup_rescan_init(fs_info
, rescan_progress
, 0);
441 btrfs_free_path(path
);
444 ulist_free(fs_info
->qgroup_ulist
);
445 fs_info
->qgroup_ulist
= NULL
;
446 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
449 return ret
< 0 ? ret
: 0;
453 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
454 * first two are in single-threaded paths.And for the third one, we have set
455 * quota_root to be null with qgroup_lock held before, so it is safe to clean
456 * up the in-memory structures without qgroup_lock held.
458 void btrfs_free_qgroup_config(struct btrfs_fs_info
*fs_info
)
461 struct btrfs_qgroup
*qgroup
;
463 while ((n
= rb_first(&fs_info
->qgroup_tree
))) {
464 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
465 rb_erase(n
, &fs_info
->qgroup_tree
);
466 __del_qgroup_rb(qgroup
);
469 * we call btrfs_free_qgroup_config() when umounting
470 * filesystem and disabling quota, so we set qgroup_ulit
471 * to be null here to avoid double free.
473 ulist_free(fs_info
->qgroup_ulist
);
474 fs_info
->qgroup_ulist
= NULL
;
477 static int add_qgroup_relation_item(struct btrfs_trans_handle
*trans
,
478 struct btrfs_root
*quota_root
,
482 struct btrfs_path
*path
;
483 struct btrfs_key key
;
485 path
= btrfs_alloc_path();
490 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
493 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
, 0);
495 btrfs_mark_buffer_dirty(path
->nodes
[0]);
497 btrfs_free_path(path
);
501 static int del_qgroup_relation_item(struct btrfs_trans_handle
*trans
,
502 struct btrfs_root
*quota_root
,
506 struct btrfs_path
*path
;
507 struct btrfs_key key
;
509 path
= btrfs_alloc_path();
514 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
517 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
526 ret
= btrfs_del_item(trans
, quota_root
, path
);
528 btrfs_free_path(path
);
532 static int add_qgroup_item(struct btrfs_trans_handle
*trans
,
533 struct btrfs_root
*quota_root
, u64 qgroupid
)
536 struct btrfs_path
*path
;
537 struct btrfs_qgroup_info_item
*qgroup_info
;
538 struct btrfs_qgroup_limit_item
*qgroup_limit
;
539 struct extent_buffer
*leaf
;
540 struct btrfs_key key
;
542 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
543 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT
, "a_root
->state
)))
546 path
= btrfs_alloc_path();
551 key
.type
= BTRFS_QGROUP_INFO_KEY
;
552 key
.offset
= qgroupid
;
554 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
555 sizeof(*qgroup_info
));
559 leaf
= path
->nodes
[0];
560 qgroup_info
= btrfs_item_ptr(leaf
, path
->slots
[0],
561 struct btrfs_qgroup_info_item
);
562 btrfs_set_qgroup_info_generation(leaf
, qgroup_info
, trans
->transid
);
563 btrfs_set_qgroup_info_rfer(leaf
, qgroup_info
, 0);
564 btrfs_set_qgroup_info_rfer_cmpr(leaf
, qgroup_info
, 0);
565 btrfs_set_qgroup_info_excl(leaf
, qgroup_info
, 0);
566 btrfs_set_qgroup_info_excl_cmpr(leaf
, qgroup_info
, 0);
568 btrfs_mark_buffer_dirty(leaf
);
570 btrfs_release_path(path
);
572 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
573 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
574 sizeof(*qgroup_limit
));
578 leaf
= path
->nodes
[0];
579 qgroup_limit
= btrfs_item_ptr(leaf
, path
->slots
[0],
580 struct btrfs_qgroup_limit_item
);
581 btrfs_set_qgroup_limit_flags(leaf
, qgroup_limit
, 0);
582 btrfs_set_qgroup_limit_max_rfer(leaf
, qgroup_limit
, 0);
583 btrfs_set_qgroup_limit_max_excl(leaf
, qgroup_limit
, 0);
584 btrfs_set_qgroup_limit_rsv_rfer(leaf
, qgroup_limit
, 0);
585 btrfs_set_qgroup_limit_rsv_excl(leaf
, qgroup_limit
, 0);
587 btrfs_mark_buffer_dirty(leaf
);
591 btrfs_free_path(path
);
595 static int del_qgroup_item(struct btrfs_trans_handle
*trans
,
596 struct btrfs_root
*quota_root
, u64 qgroupid
)
599 struct btrfs_path
*path
;
600 struct btrfs_key key
;
602 path
= btrfs_alloc_path();
607 key
.type
= BTRFS_QGROUP_INFO_KEY
;
608 key
.offset
= qgroupid
;
609 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
618 ret
= btrfs_del_item(trans
, quota_root
, path
);
622 btrfs_release_path(path
);
624 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
625 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
634 ret
= btrfs_del_item(trans
, quota_root
, path
);
637 btrfs_free_path(path
);
641 static int update_qgroup_limit_item(struct btrfs_trans_handle
*trans
,
642 struct btrfs_root
*root
, u64 qgroupid
,
643 u64 flags
, u64 max_rfer
, u64 max_excl
,
644 u64 rsv_rfer
, u64 rsv_excl
)
646 struct btrfs_path
*path
;
647 struct btrfs_key key
;
648 struct extent_buffer
*l
;
649 struct btrfs_qgroup_limit_item
*qgroup_limit
;
654 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
655 key
.offset
= qgroupid
;
657 path
= btrfs_alloc_path();
661 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
669 slot
= path
->slots
[0];
670 qgroup_limit
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_limit_item
);
671 btrfs_set_qgroup_limit_flags(l
, qgroup_limit
, flags
);
672 btrfs_set_qgroup_limit_max_rfer(l
, qgroup_limit
, max_rfer
);
673 btrfs_set_qgroup_limit_max_excl(l
, qgroup_limit
, max_excl
);
674 btrfs_set_qgroup_limit_rsv_rfer(l
, qgroup_limit
, rsv_rfer
);
675 btrfs_set_qgroup_limit_rsv_excl(l
, qgroup_limit
, rsv_excl
);
677 btrfs_mark_buffer_dirty(l
);
680 btrfs_free_path(path
);
684 static int update_qgroup_info_item(struct btrfs_trans_handle
*trans
,
685 struct btrfs_root
*root
,
686 struct btrfs_qgroup
*qgroup
)
688 struct btrfs_path
*path
;
689 struct btrfs_key key
;
690 struct extent_buffer
*l
;
691 struct btrfs_qgroup_info_item
*qgroup_info
;
695 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
696 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
)))
700 key
.type
= BTRFS_QGROUP_INFO_KEY
;
701 key
.offset
= qgroup
->qgroupid
;
703 path
= btrfs_alloc_path();
707 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
715 slot
= path
->slots
[0];
716 qgroup_info
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_info_item
);
717 btrfs_set_qgroup_info_generation(l
, qgroup_info
, trans
->transid
);
718 btrfs_set_qgroup_info_rfer(l
, qgroup_info
, qgroup
->rfer
);
719 btrfs_set_qgroup_info_rfer_cmpr(l
, qgroup_info
, qgroup
->rfer_cmpr
);
720 btrfs_set_qgroup_info_excl(l
, qgroup_info
, qgroup
->excl
);
721 btrfs_set_qgroup_info_excl_cmpr(l
, qgroup_info
, qgroup
->excl_cmpr
);
723 btrfs_mark_buffer_dirty(l
);
726 btrfs_free_path(path
);
730 static int update_qgroup_status_item(struct btrfs_trans_handle
*trans
,
731 struct btrfs_fs_info
*fs_info
,
732 struct btrfs_root
*root
)
734 struct btrfs_path
*path
;
735 struct btrfs_key key
;
736 struct extent_buffer
*l
;
737 struct btrfs_qgroup_status_item
*ptr
;
742 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
745 path
= btrfs_alloc_path();
749 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
757 slot
= path
->slots
[0];
758 ptr
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_status_item
);
759 btrfs_set_qgroup_status_flags(l
, ptr
, fs_info
->qgroup_flags
);
760 btrfs_set_qgroup_status_generation(l
, ptr
, trans
->transid
);
761 btrfs_set_qgroup_status_rescan(l
, ptr
,
762 fs_info
->qgroup_rescan_progress
.objectid
);
764 btrfs_mark_buffer_dirty(l
);
767 btrfs_free_path(path
);
772 * called with qgroup_lock held
774 static int btrfs_clean_quota_tree(struct btrfs_trans_handle
*trans
,
775 struct btrfs_root
*root
)
777 struct btrfs_path
*path
;
778 struct btrfs_key key
;
779 struct extent_buffer
*leaf
= NULL
;
783 path
= btrfs_alloc_path();
787 path
->leave_spinning
= 1;
794 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
797 leaf
= path
->nodes
[0];
798 nr
= btrfs_header_nritems(leaf
);
802 * delete the leaf one by one
803 * since the whole tree is going
807 ret
= btrfs_del_items(trans
, root
, path
, 0, nr
);
811 btrfs_release_path(path
);
815 root
->fs_info
->pending_quota_state
= 0;
816 btrfs_free_path(path
);
820 int btrfs_quota_enable(struct btrfs_trans_handle
*trans
,
821 struct btrfs_fs_info
*fs_info
)
823 struct btrfs_root
*quota_root
;
824 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
825 struct btrfs_path
*path
= NULL
;
826 struct btrfs_qgroup_status_item
*ptr
;
827 struct extent_buffer
*leaf
;
828 struct btrfs_key key
;
829 struct btrfs_key found_key
;
830 struct btrfs_qgroup
*qgroup
= NULL
;
834 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
835 if (fs_info
->quota_root
) {
836 fs_info
->pending_quota_state
= 1;
840 fs_info
->qgroup_ulist
= ulist_alloc(GFP_NOFS
);
841 if (!fs_info
->qgroup_ulist
) {
847 * initially create the quota tree
849 quota_root
= btrfs_create_tree(trans
, fs_info
,
850 BTRFS_QUOTA_TREE_OBJECTID
);
851 if (IS_ERR(quota_root
)) {
852 ret
= PTR_ERR(quota_root
);
856 path
= btrfs_alloc_path();
863 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
866 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
871 leaf
= path
->nodes
[0];
872 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0],
873 struct btrfs_qgroup_status_item
);
874 btrfs_set_qgroup_status_generation(leaf
, ptr
, trans
->transid
);
875 btrfs_set_qgroup_status_version(leaf
, ptr
, BTRFS_QGROUP_STATUS_VERSION
);
876 fs_info
->qgroup_flags
= BTRFS_QGROUP_STATUS_FLAG_ON
|
877 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
878 btrfs_set_qgroup_status_flags(leaf
, ptr
, fs_info
->qgroup_flags
);
879 btrfs_set_qgroup_status_rescan(leaf
, ptr
, 0);
881 btrfs_mark_buffer_dirty(leaf
);
884 key
.type
= BTRFS_ROOT_REF_KEY
;
887 btrfs_release_path(path
);
888 ret
= btrfs_search_slot_for_read(tree_root
, &key
, path
, 1, 0);
896 slot
= path
->slots
[0];
897 leaf
= path
->nodes
[0];
898 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
900 if (found_key
.type
== BTRFS_ROOT_REF_KEY
) {
901 ret
= add_qgroup_item(trans
, quota_root
,
906 qgroup
= add_qgroup_rb(fs_info
, found_key
.offset
);
907 if (IS_ERR(qgroup
)) {
908 ret
= PTR_ERR(qgroup
);
912 ret
= btrfs_next_item(tree_root
, path
);
920 btrfs_release_path(path
);
921 ret
= add_qgroup_item(trans
, quota_root
, BTRFS_FS_TREE_OBJECTID
);
925 qgroup
= add_qgroup_rb(fs_info
, BTRFS_FS_TREE_OBJECTID
);
926 if (IS_ERR(qgroup
)) {
927 ret
= PTR_ERR(qgroup
);
930 spin_lock(&fs_info
->qgroup_lock
);
931 fs_info
->quota_root
= quota_root
;
932 fs_info
->pending_quota_state
= 1;
933 spin_unlock(&fs_info
->qgroup_lock
);
935 btrfs_free_path(path
);
938 free_extent_buffer(quota_root
->node
);
939 free_extent_buffer(quota_root
->commit_root
);
944 ulist_free(fs_info
->qgroup_ulist
);
945 fs_info
->qgroup_ulist
= NULL
;
947 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
951 int btrfs_quota_disable(struct btrfs_trans_handle
*trans
,
952 struct btrfs_fs_info
*fs_info
)
954 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
955 struct btrfs_root
*quota_root
;
958 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
959 if (!fs_info
->quota_root
)
961 spin_lock(&fs_info
->qgroup_lock
);
962 fs_info
->quota_enabled
= 0;
963 fs_info
->pending_quota_state
= 0;
964 quota_root
= fs_info
->quota_root
;
965 fs_info
->quota_root
= NULL
;
966 spin_unlock(&fs_info
->qgroup_lock
);
968 btrfs_free_qgroup_config(fs_info
);
970 ret
= btrfs_clean_quota_tree(trans
, quota_root
);
974 ret
= btrfs_del_root(trans
, tree_root
, "a_root
->root_key
);
978 list_del("a_root
->dirty_list
);
980 btrfs_tree_lock(quota_root
->node
);
981 clean_tree_block(trans
, tree_root
, quota_root
->node
);
982 btrfs_tree_unlock(quota_root
->node
);
983 btrfs_free_tree_block(trans
, quota_root
, quota_root
->node
, 0, 1);
985 free_extent_buffer(quota_root
->node
);
986 free_extent_buffer(quota_root
->commit_root
);
989 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
993 static void qgroup_dirty(struct btrfs_fs_info
*fs_info
,
994 struct btrfs_qgroup
*qgroup
)
996 if (list_empty(&qgroup
->dirty
))
997 list_add(&qgroup
->dirty
, &fs_info
->dirty_qgroups
);
1000 int btrfs_add_qgroup_relation(struct btrfs_trans_handle
*trans
,
1001 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1003 struct btrfs_root
*quota_root
;
1004 struct btrfs_qgroup
*parent
;
1005 struct btrfs_qgroup
*member
;
1006 struct btrfs_qgroup_list
*list
;
1009 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1010 quota_root
= fs_info
->quota_root
;
1015 member
= find_qgroup_rb(fs_info
, src
);
1016 parent
= find_qgroup_rb(fs_info
, dst
);
1017 if (!member
|| !parent
) {
1022 /* check if such qgroup relation exist firstly */
1023 list_for_each_entry(list
, &member
->groups
, next_group
) {
1024 if (list
->group
== parent
) {
1030 ret
= add_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1034 ret
= add_qgroup_relation_item(trans
, quota_root
, dst
, src
);
1036 del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1040 spin_lock(&fs_info
->qgroup_lock
);
1041 ret
= add_relation_rb(quota_root
->fs_info
, src
, dst
);
1042 spin_unlock(&fs_info
->qgroup_lock
);
1044 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1048 int btrfs_del_qgroup_relation(struct btrfs_trans_handle
*trans
,
1049 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1051 struct btrfs_root
*quota_root
;
1052 struct btrfs_qgroup
*parent
;
1053 struct btrfs_qgroup
*member
;
1054 struct btrfs_qgroup_list
*list
;
1058 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1059 quota_root
= fs_info
->quota_root
;
1065 member
= find_qgroup_rb(fs_info
, src
);
1066 parent
= find_qgroup_rb(fs_info
, dst
);
1067 if (!member
|| !parent
) {
1072 /* check if such qgroup relation exist firstly */
1073 list_for_each_entry(list
, &member
->groups
, next_group
) {
1074 if (list
->group
== parent
)
1080 ret
= del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1081 err
= del_qgroup_relation_item(trans
, quota_root
, dst
, src
);
1085 spin_lock(&fs_info
->qgroup_lock
);
1086 del_relation_rb(fs_info
, src
, dst
);
1087 spin_unlock(&fs_info
->qgroup_lock
);
1089 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1093 int btrfs_create_qgroup(struct btrfs_trans_handle
*trans
,
1094 struct btrfs_fs_info
*fs_info
, u64 qgroupid
, char *name
)
1096 struct btrfs_root
*quota_root
;
1097 struct btrfs_qgroup
*qgroup
;
1100 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1101 quota_root
= fs_info
->quota_root
;
1106 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1112 ret
= add_qgroup_item(trans
, quota_root
, qgroupid
);
1116 spin_lock(&fs_info
->qgroup_lock
);
1117 qgroup
= add_qgroup_rb(fs_info
, qgroupid
);
1118 spin_unlock(&fs_info
->qgroup_lock
);
1121 ret
= PTR_ERR(qgroup
);
1123 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1127 int btrfs_remove_qgroup(struct btrfs_trans_handle
*trans
,
1128 struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
1130 struct btrfs_root
*quota_root
;
1131 struct btrfs_qgroup
*qgroup
;
1134 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1135 quota_root
= fs_info
->quota_root
;
1141 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1146 /* check if there are no relations to this qgroup */
1147 if (!list_empty(&qgroup
->groups
) ||
1148 !list_empty(&qgroup
->members
)) {
1153 ret
= del_qgroup_item(trans
, quota_root
, qgroupid
);
1155 spin_lock(&fs_info
->qgroup_lock
);
1156 del_qgroup_rb(quota_root
->fs_info
, qgroupid
);
1157 spin_unlock(&fs_info
->qgroup_lock
);
1159 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1163 int btrfs_limit_qgroup(struct btrfs_trans_handle
*trans
,
1164 struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
1165 struct btrfs_qgroup_limit
*limit
)
1167 struct btrfs_root
*quota_root
;
1168 struct btrfs_qgroup
*qgroup
;
1171 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1172 quota_root
= fs_info
->quota_root
;
1178 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1183 ret
= update_qgroup_limit_item(trans
, quota_root
, qgroupid
,
1184 limit
->flags
, limit
->max_rfer
,
1185 limit
->max_excl
, limit
->rsv_rfer
,
1188 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1189 btrfs_info(fs_info
, "unable to update quota limit for %llu",
1193 spin_lock(&fs_info
->qgroup_lock
);
1194 qgroup
->lim_flags
= limit
->flags
;
1195 qgroup
->max_rfer
= limit
->max_rfer
;
1196 qgroup
->max_excl
= limit
->max_excl
;
1197 qgroup
->rsv_rfer
= limit
->rsv_rfer
;
1198 qgroup
->rsv_excl
= limit
->rsv_excl
;
1199 spin_unlock(&fs_info
->qgroup_lock
);
1201 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1204 static int comp_oper(struct btrfs_qgroup_operation
*oper1
,
1205 struct btrfs_qgroup_operation
*oper2
)
1207 if (oper1
->bytenr
< oper2
->bytenr
)
1209 if (oper1
->bytenr
> oper2
->bytenr
)
1211 if (oper1
->seq
< oper2
->seq
)
1213 if (oper1
->seq
> oper2
->seq
)
1215 if (oper1
->ref_root
< oper2
->ref_root
)
1217 if (oper1
->ref_root
> oper2
->ref_root
)
1219 if (oper1
->type
< oper2
->type
)
1221 if (oper1
->type
> oper2
->type
)
1226 static int insert_qgroup_oper(struct btrfs_fs_info
*fs_info
,
1227 struct btrfs_qgroup_operation
*oper
)
1230 struct rb_node
*parent
= NULL
;
1231 struct btrfs_qgroup_operation
*cur
;
1234 spin_lock(&fs_info
->qgroup_op_lock
);
1235 p
= &fs_info
->qgroup_op_tree
.rb_node
;
1238 cur
= rb_entry(parent
, struct btrfs_qgroup_operation
, n
);
1239 cmp
= comp_oper(cur
, oper
);
1241 p
= &(*p
)->rb_right
;
1245 spin_unlock(&fs_info
->qgroup_op_lock
);
1249 rb_link_node(&oper
->n
, parent
, p
);
1250 rb_insert_color(&oper
->n
, &fs_info
->qgroup_op_tree
);
1251 spin_unlock(&fs_info
->qgroup_op_lock
);
1256 * Record a quota operation for processing later on.
1257 * @trans: the transaction we are adding the delayed op to.
1258 * @fs_info: the fs_info for this fs.
1259 * @ref_root: the root of the reference we are acting on,
1260 * @bytenr: the bytenr we are acting on.
1261 * @num_bytes: the number of bytes in the reference.
1262 * @type: the type of operation this is.
1263 * @mod_seq: do we need to get a sequence number for looking up roots.
1265 * We just add it to our trans qgroup_ref_list and carry on and process these
1266 * operations in order at some later point. If the reference root isn't a fs
1267 * root then we don't bother with doing anything.
1269 * MUST BE HOLDING THE REF LOCK.
1271 int btrfs_qgroup_record_ref(struct btrfs_trans_handle
*trans
,
1272 struct btrfs_fs_info
*fs_info
, u64 ref_root
,
1273 u64 bytenr
, u64 num_bytes
,
1274 enum btrfs_qgroup_operation_type type
, int mod_seq
)
1276 struct btrfs_qgroup_operation
*oper
;
1279 if (!is_fstree(ref_root
) || !fs_info
->quota_enabled
)
1282 oper
= kmalloc(sizeof(*oper
), GFP_NOFS
);
1286 oper
->ref_root
= ref_root
;
1287 oper
->bytenr
= bytenr
;
1288 oper
->num_bytes
= num_bytes
;
1290 oper
->seq
= atomic_inc_return(&fs_info
->qgroup_op_seq
);
1291 INIT_LIST_HEAD(&oper
->elem
.list
);
1293 ret
= insert_qgroup_oper(fs_info
, oper
);
1295 /* Shouldn't happen so have an assert for developers */
1300 list_add_tail(&oper
->list
, &trans
->qgroup_ref_list
);
1303 btrfs_get_tree_mod_seq(fs_info
, &oper
->elem
);
1309 * The easy accounting, if we are adding/removing the only ref for an extent
1310 * then this qgroup and all of the parent qgroups get their refrence and
1311 * exclusive counts adjusted.
1313 static int qgroup_excl_accounting(struct btrfs_fs_info
*fs_info
,
1314 struct btrfs_qgroup_operation
*oper
)
1316 struct btrfs_qgroup
*qgroup
;
1318 struct btrfs_qgroup_list
*glist
;
1319 struct ulist_node
*unode
;
1320 struct ulist_iterator uiter
;
1324 tmp
= ulist_alloc(GFP_NOFS
);
1328 spin_lock(&fs_info
->qgroup_lock
);
1329 if (!fs_info
->quota_root
)
1331 qgroup
= find_qgroup_rb(fs_info
, oper
->ref_root
);
1334 switch (oper
->type
) {
1335 case BTRFS_QGROUP_OPER_ADD_EXCL
:
1338 case BTRFS_QGROUP_OPER_SUB_EXCL
:
1344 qgroup
->rfer
+= sign
* oper
->num_bytes
;
1345 qgroup
->rfer_cmpr
+= sign
* oper
->num_bytes
;
1347 WARN_ON(sign
< 0 && qgroup
->excl
< oper
->num_bytes
);
1348 qgroup
->excl
+= sign
* oper
->num_bytes
;
1349 qgroup
->excl_cmpr
+= sign
* oper
->num_bytes
;
1351 qgroup_dirty(fs_info
, qgroup
);
1353 /* Get all of the parent groups that contain this qgroup */
1354 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
1355 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1356 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1361 /* Iterate all of the parents and adjust their reference counts */
1362 ULIST_ITER_INIT(&uiter
);
1363 while ((unode
= ulist_next(tmp
, &uiter
))) {
1364 qgroup
= u64_to_ptr(unode
->aux
);
1365 qgroup
->rfer
+= sign
* oper
->num_bytes
;
1366 qgroup
->rfer_cmpr
+= sign
* oper
->num_bytes
;
1367 qgroup
->excl
+= sign
* oper
->num_bytes
;
1369 WARN_ON(qgroup
->excl
< oper
->num_bytes
);
1370 qgroup
->excl_cmpr
+= sign
* oper
->num_bytes
;
1371 qgroup_dirty(fs_info
, qgroup
);
1373 /* Add any parents of the parents */
1374 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
1375 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1376 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1383 spin_unlock(&fs_info
->qgroup_lock
);
1389 * Walk all of the roots that pointed to our bytenr and adjust their refcnts as
1392 static int qgroup_calc_old_refcnt(struct btrfs_fs_info
*fs_info
,
1393 u64 root_to_skip
, struct ulist
*tmp
,
1394 struct ulist
*roots
, struct ulist
*qgroups
,
1395 u64 seq
, int *old_roots
, int rescan
)
1397 struct ulist_node
*unode
;
1398 struct ulist_iterator uiter
;
1399 struct ulist_node
*tmp_unode
;
1400 struct ulist_iterator tmp_uiter
;
1401 struct btrfs_qgroup
*qg
;
1404 ULIST_ITER_INIT(&uiter
);
1405 while ((unode
= ulist_next(roots
, &uiter
))) {
1406 /* We don't count our current root here */
1407 if (unode
->val
== root_to_skip
)
1409 qg
= find_qgroup_rb(fs_info
, unode
->val
);
1413 * We could have a pending removal of this same ref so we may
1414 * not have actually found our ref root when doing
1415 * btrfs_find_all_roots, so we need to keep track of how many
1416 * old roots we find in case we removed ours and added a
1417 * different one at the same time. I don't think this could
1418 * happen in practice but that sort of thinking leads to pain
1419 * and suffering and to the dark side.
1424 ret
= ulist_add(qgroups
, qg
->qgroupid
, ptr_to_u64(qg
),
1428 ret
= ulist_add(tmp
, qg
->qgroupid
, ptr_to_u64(qg
), GFP_ATOMIC
);
1431 ULIST_ITER_INIT(&tmp_uiter
);
1432 while ((tmp_unode
= ulist_next(tmp
, &tmp_uiter
))) {
1433 struct btrfs_qgroup_list
*glist
;
1435 qg
= u64_to_ptr(tmp_unode
->aux
);
1437 * We use this sequence number to keep from having to
1438 * run the whole list and 0 out the refcnt every time.
1439 * We basically use sequnce as the known 0 count and
1440 * then add 1 everytime we see a qgroup. This is how we
1441 * get how many of the roots actually point up to the
1442 * upper level qgroups in order to determine exclusive
1445 * For rescan we want to set old_refcnt to seq so our
1446 * exclusive calculations end up correct.
1449 qg
->old_refcnt
= seq
;
1450 else if (qg
->old_refcnt
< seq
)
1451 qg
->old_refcnt
= seq
+ 1;
1455 if (qg
->new_refcnt
< seq
)
1456 qg
->new_refcnt
= seq
+ 1;
1459 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1460 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1461 ptr_to_u64(glist
->group
),
1465 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1466 ptr_to_u64(glist
->group
),
1477 * We need to walk forward in our operation tree and account for any roots that
1478 * were deleted after we made this operation.
1480 static int qgroup_account_deleted_refs(struct btrfs_fs_info
*fs_info
,
1481 struct btrfs_qgroup_operation
*oper
,
1483 struct ulist
*qgroups
, u64 seq
,
1486 struct ulist_node
*unode
;
1487 struct ulist_iterator uiter
;
1488 struct btrfs_qgroup
*qg
;
1489 struct btrfs_qgroup_operation
*tmp_oper
;
1496 * We only walk forward in the tree since we're only interested in
1497 * removals that happened _after_ our operation.
1499 spin_lock(&fs_info
->qgroup_op_lock
);
1500 n
= rb_next(&oper
->n
);
1501 spin_unlock(&fs_info
->qgroup_op_lock
);
1504 tmp_oper
= rb_entry(n
, struct btrfs_qgroup_operation
, n
);
1505 while (tmp_oper
->bytenr
== oper
->bytenr
) {
1507 * If it's not a removal we don't care, additions work out
1508 * properly with our refcnt tracking.
1510 if (tmp_oper
->type
!= BTRFS_QGROUP_OPER_SUB_SHARED
&&
1511 tmp_oper
->type
!= BTRFS_QGROUP_OPER_SUB_EXCL
)
1513 qg
= find_qgroup_rb(fs_info
, tmp_oper
->ref_root
);
1516 ret
= ulist_add(qgroups
, qg
->qgroupid
, ptr_to_u64(qg
),
1522 * We only want to increase old_roots if this qgroup is
1523 * not already in the list of qgroups. If it is already
1524 * there then that means it must have been re-added or
1525 * the delete will be discarded because we had an
1526 * existing ref that we haven't looked up yet. In this
1527 * case we don't want to increase old_roots. So if ret
1528 * == 1 then we know that this is the first time we've
1529 * seen this qgroup and we can bump the old_roots.
1532 ret
= ulist_add(tmp
, qg
->qgroupid
, ptr_to_u64(qg
),
1538 spin_lock(&fs_info
->qgroup_op_lock
);
1539 n
= rb_next(&tmp_oper
->n
);
1540 spin_unlock(&fs_info
->qgroup_op_lock
);
1543 tmp_oper
= rb_entry(n
, struct btrfs_qgroup_operation
, n
);
1546 /* Ok now process the qgroups we found */
1547 ULIST_ITER_INIT(&uiter
);
1548 while ((unode
= ulist_next(tmp
, &uiter
))) {
1549 struct btrfs_qgroup_list
*glist
;
1551 qg
= u64_to_ptr(unode
->aux
);
1552 if (qg
->old_refcnt
< seq
)
1553 qg
->old_refcnt
= seq
+ 1;
1556 if (qg
->new_refcnt
< seq
)
1557 qg
->new_refcnt
= seq
+ 1;
1560 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1561 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1562 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1565 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1566 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1574 /* Add refcnt for the newly added reference. */
1575 static int qgroup_calc_new_refcnt(struct btrfs_fs_info
*fs_info
,
1576 struct btrfs_qgroup_operation
*oper
,
1577 struct btrfs_qgroup
*qgroup
,
1578 struct ulist
*tmp
, struct ulist
*qgroups
,
1581 struct ulist_node
*unode
;
1582 struct ulist_iterator uiter
;
1583 struct btrfs_qgroup
*qg
;
1587 ret
= ulist_add(qgroups
, qgroup
->qgroupid
, ptr_to_u64(qgroup
),
1591 ret
= ulist_add(tmp
, qgroup
->qgroupid
, ptr_to_u64(qgroup
),
1595 ULIST_ITER_INIT(&uiter
);
1596 while ((unode
= ulist_next(tmp
, &uiter
))) {
1597 struct btrfs_qgroup_list
*glist
;
1599 qg
= u64_to_ptr(unode
->aux
);
1600 if (oper
->type
== BTRFS_QGROUP_OPER_ADD_SHARED
) {
1601 if (qg
->new_refcnt
< seq
)
1602 qg
->new_refcnt
= seq
+ 1;
1606 if (qg
->old_refcnt
< seq
)
1607 qg
->old_refcnt
= seq
+ 1;
1611 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1612 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1613 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1616 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1617 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1626 * This adjusts the counters for all referenced qgroups if need be.
1628 static int qgroup_adjust_counters(struct btrfs_fs_info
*fs_info
,
1629 u64 root_to_skip
, u64 num_bytes
,
1630 struct ulist
*qgroups
, u64 seq
,
1631 int old_roots
, int new_roots
, int rescan
)
1633 struct ulist_node
*unode
;
1634 struct ulist_iterator uiter
;
1635 struct btrfs_qgroup
*qg
;
1636 u64 cur_new_count
, cur_old_count
;
1638 ULIST_ITER_INIT(&uiter
);
1639 while ((unode
= ulist_next(qgroups
, &uiter
))) {
1642 qg
= u64_to_ptr(unode
->aux
);
1644 * Wasn't referenced before but is now, add to the reference
1647 if (qg
->old_refcnt
<= seq
&& qg
->new_refcnt
> seq
) {
1648 qg
->rfer
+= num_bytes
;
1649 qg
->rfer_cmpr
+= num_bytes
;
1654 * Was referenced before but isn't now, subtract from the
1655 * reference counters.
1657 if (qg
->old_refcnt
> seq
&& qg
->new_refcnt
<= seq
) {
1658 qg
->rfer
-= num_bytes
;
1659 qg
->rfer_cmpr
-= num_bytes
;
1663 if (qg
->old_refcnt
< seq
)
1666 cur_old_count
= qg
->old_refcnt
- seq
;
1667 if (qg
->new_refcnt
< seq
)
1670 cur_new_count
= qg
->new_refcnt
- seq
;
1673 * If our refcount was the same as the roots previously but our
1674 * new count isn't the same as the number of roots now then we
1675 * went from having a exclusive reference on this range to not.
1677 if (old_roots
&& cur_old_count
== old_roots
&&
1678 (cur_new_count
!= new_roots
|| new_roots
== 0)) {
1679 WARN_ON(cur_new_count
!= new_roots
&& new_roots
== 0);
1680 qg
->excl
-= num_bytes
;
1681 qg
->excl_cmpr
-= num_bytes
;
1686 * If we didn't reference all the roots before but now we do we
1687 * have an exclusive reference to this range.
1689 if ((!old_roots
|| (old_roots
&& cur_old_count
!= old_roots
))
1690 && cur_new_count
== new_roots
) {
1691 qg
->excl
+= num_bytes
;
1692 qg
->excl_cmpr
+= num_bytes
;
1697 qgroup_dirty(fs_info
, qg
);
1703 * If we removed a data extent and there were other references for that bytenr
1704 * then we need to lookup all referenced roots to make sure we still don't
1705 * reference this bytenr. If we do then we can just discard this operation.
1707 static int check_existing_refs(struct btrfs_trans_handle
*trans
,
1708 struct btrfs_fs_info
*fs_info
,
1709 struct btrfs_qgroup_operation
*oper
)
1711 struct ulist
*roots
= NULL
;
1712 struct ulist_node
*unode
;
1713 struct ulist_iterator uiter
;
1716 ret
= btrfs_find_all_roots(trans
, fs_info
, oper
->bytenr
,
1717 oper
->elem
.seq
, &roots
);
1722 ULIST_ITER_INIT(&uiter
);
1723 while ((unode
= ulist_next(roots
, &uiter
))) {
1724 if (unode
->val
== oper
->ref_root
) {
1730 btrfs_put_tree_mod_seq(fs_info
, &oper
->elem
);
1736 * If we share a reference across multiple roots then we may need to adjust
1737 * various qgroups referenced and exclusive counters. The basic premise is this
1739 * 1) We have seq to represent a 0 count. Instead of looping through all of the
1740 * qgroups and resetting their refcount to 0 we just constantly bump this
1741 * sequence number to act as the base reference count. This means that if
1742 * anybody is equal to or below this sequence they were never referenced. We
1743 * jack this sequence up by the number of roots we found each time in order to
1744 * make sure we don't have any overlap.
1746 * 2) We first search all the roots that reference the area _except_ the root
1747 * we're acting on currently. This makes up the old_refcnt of all the qgroups
1750 * 3) We walk all of the qgroups referenced by the root we are currently acting
1751 * on, and will either adjust old_refcnt in the case of a removal or the
1752 * new_refcnt in the case of an addition.
1754 * 4) Finally we walk all the qgroups that are referenced by this range
1755 * including the root we are acting on currently. We will adjust the counters
1756 * based on the number of roots we had and will have after this operation.
1758 * Take this example as an illustration
1762 * [qg 0/0] [qg 0/1] [qg 0/2]
1766 * Say we are adding a reference that is covered by qg 0/0. The first step
1767 * would give a refcnt of 1 to qg 0/1 and 0/2 and a refcnt of 2 to qg 1/0 with
1768 * old_roots being 2. Because it is adding new_roots will be 1. We then go
1769 * through qg 0/0 which will get the new_refcnt set to 1 and add 1 to qg 1/0's
1770 * new_refcnt, bringing it to 3. We then walk through all of the qgroups, we
1771 * notice that the old refcnt for qg 0/0 < the new refcnt, so we added a
1772 * reference and thus must add the size to the referenced bytes. Everything
1773 * else is the same so nothing else changes.
1775 static int qgroup_shared_accounting(struct btrfs_trans_handle
*trans
,
1776 struct btrfs_fs_info
*fs_info
,
1777 struct btrfs_qgroup_operation
*oper
)
1779 struct ulist
*roots
= NULL
;
1780 struct ulist
*qgroups
, *tmp
;
1781 struct btrfs_qgroup
*qgroup
;
1782 struct seq_list elem
= {};
1788 if (oper
->elem
.seq
) {
1789 ret
= check_existing_refs(trans
, fs_info
, oper
);
1796 qgroups
= ulist_alloc(GFP_NOFS
);
1800 tmp
= ulist_alloc(GFP_NOFS
);
1802 ulist_free(qgroups
);
1806 btrfs_get_tree_mod_seq(fs_info
, &elem
);
1807 ret
= btrfs_find_all_roots(trans
, fs_info
, oper
->bytenr
, elem
.seq
,
1809 btrfs_put_tree_mod_seq(fs_info
, &elem
);
1811 ulist_free(qgroups
);
1815 spin_lock(&fs_info
->qgroup_lock
);
1816 qgroup
= find_qgroup_rb(fs_info
, oper
->ref_root
);
1819 seq
= fs_info
->qgroup_seq
;
1822 * So roots is the list of all the roots currently pointing at the
1823 * bytenr, including the ref we are adding if we are adding, or not if
1824 * we are removing a ref. So we pass in the ref_root to skip that root
1825 * in our calculations. We set old_refnct and new_refcnt cause who the
1826 * hell knows what everything looked like before, and it doesn't matter
1829 ret
= qgroup_calc_old_refcnt(fs_info
, oper
->ref_root
, tmp
, roots
, qgroups
,
1830 seq
, &old_roots
, 0);
1835 * Now adjust the refcounts of the qgroups that care about this
1836 * reference, either the old_count in the case of removal or new_count
1837 * in the case of an addition.
1839 ret
= qgroup_calc_new_refcnt(fs_info
, oper
, qgroup
, tmp
, qgroups
,
1845 * ...in the case of removals. If we had a removal before we got around
1846 * to processing this operation then we need to find that guy and count
1847 * his references as if they really existed so we don't end up screwing
1848 * up the exclusive counts. Then whenever we go to process the delete
1849 * everything will be grand and we can account for whatever exclusive
1850 * changes need to be made there. We also have to pass in old_roots so
1851 * we have an accurate count of the roots as it pertains to this
1852 * operations view of the world.
1854 ret
= qgroup_account_deleted_refs(fs_info
, oper
, tmp
, qgroups
, seq
,
1860 * We are adding our root, need to adjust up the number of roots,
1861 * otherwise old_roots is the number of roots we want.
1863 if (oper
->type
== BTRFS_QGROUP_OPER_ADD_SHARED
) {
1864 new_roots
= old_roots
+ 1;
1866 new_roots
= old_roots
;
1869 fs_info
->qgroup_seq
+= old_roots
+ 1;
1873 * And now the magic happens, bless Arne for having a pretty elegant
1874 * solution for this.
1876 qgroup_adjust_counters(fs_info
, oper
->ref_root
, oper
->num_bytes
,
1877 qgroups
, seq
, old_roots
, new_roots
, 0);
1879 spin_unlock(&fs_info
->qgroup_lock
);
1880 ulist_free(qgroups
);
1887 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1888 * from the fs. First, all roots referencing the extent are searched, and
1889 * then the space is accounted accordingly to the different roots. The
1890 * accounting algorithm works in 3 steps documented inline.
1892 static int btrfs_qgroup_account(struct btrfs_trans_handle
*trans
,
1893 struct btrfs_fs_info
*fs_info
,
1894 struct btrfs_qgroup_operation
*oper
)
1898 if (!fs_info
->quota_enabled
)
1901 BUG_ON(!fs_info
->quota_root
);
1903 mutex_lock(&fs_info
->qgroup_rescan_lock
);
1904 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) {
1905 if (fs_info
->qgroup_rescan_progress
.objectid
<= oper
->bytenr
) {
1906 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
1910 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
1912 ASSERT(is_fstree(oper
->ref_root
));
1914 switch (oper
->type
) {
1915 case BTRFS_QGROUP_OPER_ADD_EXCL
:
1916 case BTRFS_QGROUP_OPER_SUB_EXCL
:
1917 ret
= qgroup_excl_accounting(fs_info
, oper
);
1919 case BTRFS_QGROUP_OPER_ADD_SHARED
:
1920 case BTRFS_QGROUP_OPER_SUB_SHARED
:
1921 ret
= qgroup_shared_accounting(trans
, fs_info
, oper
);
1930 * Needs to be called everytime we run delayed refs, even if there is an error
1931 * in order to cleanup outstanding operations.
1933 int btrfs_delayed_qgroup_accounting(struct btrfs_trans_handle
*trans
,
1934 struct btrfs_fs_info
*fs_info
)
1936 struct btrfs_qgroup_operation
*oper
;
1939 while (!list_empty(&trans
->qgroup_ref_list
)) {
1940 oper
= list_first_entry(&trans
->qgroup_ref_list
,
1941 struct btrfs_qgroup_operation
, list
);
1942 list_del_init(&oper
->list
);
1943 if (!ret
|| !trans
->aborted
)
1944 ret
= btrfs_qgroup_account(trans
, fs_info
, oper
);
1945 spin_lock(&fs_info
->qgroup_op_lock
);
1946 rb_erase(&oper
->n
, &fs_info
->qgroup_op_tree
);
1947 spin_unlock(&fs_info
->qgroup_op_lock
);
1948 btrfs_put_tree_mod_seq(fs_info
, &oper
->elem
);
1955 * called from commit_transaction. Writes all changed qgroups to disk.
1957 int btrfs_run_qgroups(struct btrfs_trans_handle
*trans
,
1958 struct btrfs_fs_info
*fs_info
)
1960 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
1962 int start_rescan_worker
= 0;
1967 if (!fs_info
->quota_enabled
&& fs_info
->pending_quota_state
)
1968 start_rescan_worker
= 1;
1970 fs_info
->quota_enabled
= fs_info
->pending_quota_state
;
1972 spin_lock(&fs_info
->qgroup_lock
);
1973 while (!list_empty(&fs_info
->dirty_qgroups
)) {
1974 struct btrfs_qgroup
*qgroup
;
1975 qgroup
= list_first_entry(&fs_info
->dirty_qgroups
,
1976 struct btrfs_qgroup
, dirty
);
1977 list_del_init(&qgroup
->dirty
);
1978 spin_unlock(&fs_info
->qgroup_lock
);
1979 ret
= update_qgroup_info_item(trans
, quota_root
, qgroup
);
1981 fs_info
->qgroup_flags
|=
1982 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1983 spin_lock(&fs_info
->qgroup_lock
);
1985 if (fs_info
->quota_enabled
)
1986 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_ON
;
1988 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
1989 spin_unlock(&fs_info
->qgroup_lock
);
1991 ret
= update_qgroup_status_item(trans
, fs_info
, quota_root
);
1993 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1995 if (!ret
&& start_rescan_worker
) {
1996 ret
= qgroup_rescan_init(fs_info
, 0, 1);
1998 qgroup_rescan_zero_tracking(fs_info
);
1999 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2000 &fs_info
->qgroup_rescan_work
);
2011 * copy the acounting information between qgroups. This is necessary when a
2012 * snapshot or a subvolume is created
2014 int btrfs_qgroup_inherit(struct btrfs_trans_handle
*trans
,
2015 struct btrfs_fs_info
*fs_info
, u64 srcid
, u64 objectid
,
2016 struct btrfs_qgroup_inherit
*inherit
)
2021 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
2022 struct btrfs_qgroup
*srcgroup
;
2023 struct btrfs_qgroup
*dstgroup
;
2027 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
2028 if (!fs_info
->quota_enabled
)
2037 i_qgroups
= (u64
*)(inherit
+ 1);
2038 nums
= inherit
->num_qgroups
+ 2 * inherit
->num_ref_copies
+
2039 2 * inherit
->num_excl_copies
;
2040 for (i
= 0; i
< nums
; ++i
) {
2041 srcgroup
= find_qgroup_rb(fs_info
, *i_qgroups
);
2051 * create a tracking group for the subvol itself
2053 ret
= add_qgroup_item(trans
, quota_root
, objectid
);
2057 if (inherit
&& inherit
->flags
& BTRFS_QGROUP_INHERIT_SET_LIMITS
) {
2058 ret
= update_qgroup_limit_item(trans
, quota_root
, objectid
,
2060 inherit
->lim
.max_rfer
,
2061 inherit
->lim
.max_excl
,
2062 inherit
->lim
.rsv_rfer
,
2063 inherit
->lim
.rsv_excl
);
2069 struct btrfs_root
*srcroot
;
2070 struct btrfs_key srckey
;
2073 srckey
.objectid
= srcid
;
2074 srckey
.type
= BTRFS_ROOT_ITEM_KEY
;
2075 srckey
.offset
= (u64
)-1;
2076 srcroot
= btrfs_read_fs_root_no_name(fs_info
, &srckey
);
2077 if (IS_ERR(srcroot
)) {
2078 ret
= PTR_ERR(srcroot
);
2083 srcroot_level
= btrfs_header_level(srcroot
->node
);
2084 level_size
= btrfs_level_size(srcroot
, srcroot_level
);
2089 * add qgroup to all inherited groups
2092 i_qgroups
= (u64
*)(inherit
+ 1);
2093 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
2094 ret
= add_qgroup_relation_item(trans
, quota_root
,
2095 objectid
, *i_qgroups
);
2098 ret
= add_qgroup_relation_item(trans
, quota_root
,
2099 *i_qgroups
, objectid
);
2107 spin_lock(&fs_info
->qgroup_lock
);
2109 dstgroup
= add_qgroup_rb(fs_info
, objectid
);
2110 if (IS_ERR(dstgroup
)) {
2111 ret
= PTR_ERR(dstgroup
);
2116 srcgroup
= find_qgroup_rb(fs_info
, srcid
);
2121 * We call inherit after we clone the root in order to make sure
2122 * our counts don't go crazy, so at this point the only
2123 * difference between the two roots should be the root node.
2125 dstgroup
->rfer
= srcgroup
->rfer
;
2126 dstgroup
->rfer_cmpr
= srcgroup
->rfer_cmpr
;
2127 dstgroup
->excl
= level_size
;
2128 dstgroup
->excl_cmpr
= level_size
;
2129 srcgroup
->excl
= level_size
;
2130 srcgroup
->excl_cmpr
= level_size
;
2131 qgroup_dirty(fs_info
, dstgroup
);
2132 qgroup_dirty(fs_info
, srcgroup
);
2138 i_qgroups
= (u64
*)(inherit
+ 1);
2139 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
2140 ret
= add_relation_rb(quota_root
->fs_info
, objectid
,
2147 for (i
= 0; i
< inherit
->num_ref_copies
; ++i
) {
2148 struct btrfs_qgroup
*src
;
2149 struct btrfs_qgroup
*dst
;
2151 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
2152 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
2159 dst
->rfer
= src
->rfer
- level_size
;
2160 dst
->rfer_cmpr
= src
->rfer_cmpr
- level_size
;
2163 for (i
= 0; i
< inherit
->num_excl_copies
; ++i
) {
2164 struct btrfs_qgroup
*src
;
2165 struct btrfs_qgroup
*dst
;
2167 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
2168 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
2175 dst
->excl
= src
->excl
+ level_size
;
2176 dst
->excl_cmpr
= src
->excl_cmpr
+ level_size
;
2181 spin_unlock(&fs_info
->qgroup_lock
);
2183 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
2188 * reserve some space for a qgroup and all its parents. The reservation takes
2189 * place with start_transaction or dealloc_reserve, similar to ENOSPC
2190 * accounting. If not enough space is available, EDQUOT is returned.
2191 * We assume that the requested space is new for all qgroups.
2193 int btrfs_qgroup_reserve(struct btrfs_root
*root
, u64 num_bytes
)
2195 struct btrfs_root
*quota_root
;
2196 struct btrfs_qgroup
*qgroup
;
2197 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2198 u64 ref_root
= root
->root_key
.objectid
;
2200 struct ulist_node
*unode
;
2201 struct ulist_iterator uiter
;
2203 if (!is_fstree(ref_root
))
2209 spin_lock(&fs_info
->qgroup_lock
);
2210 quota_root
= fs_info
->quota_root
;
2214 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
2219 * in a first step, we check all affected qgroups if any limits would
2222 ulist_reinit(fs_info
->qgroup_ulist
);
2223 ret
= ulist_add(fs_info
->qgroup_ulist
, qgroup
->qgroupid
,
2224 (uintptr_t)qgroup
, GFP_ATOMIC
);
2227 ULIST_ITER_INIT(&uiter
);
2228 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2229 struct btrfs_qgroup
*qg
;
2230 struct btrfs_qgroup_list
*glist
;
2232 qg
= u64_to_ptr(unode
->aux
);
2234 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) &&
2235 qg
->reserved
+ (s64
)qg
->rfer
+ num_bytes
>
2241 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) &&
2242 qg
->reserved
+ (s64
)qg
->excl
+ num_bytes
>
2248 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2249 ret
= ulist_add(fs_info
->qgroup_ulist
,
2250 glist
->group
->qgroupid
,
2251 (uintptr_t)glist
->group
, GFP_ATOMIC
);
2258 * no limits exceeded, now record the reservation into all qgroups
2260 ULIST_ITER_INIT(&uiter
);
2261 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2262 struct btrfs_qgroup
*qg
;
2264 qg
= u64_to_ptr(unode
->aux
);
2266 qg
->reserved
+= num_bytes
;
2270 spin_unlock(&fs_info
->qgroup_lock
);
2274 void btrfs_qgroup_free(struct btrfs_root
*root
, u64 num_bytes
)
2276 struct btrfs_root
*quota_root
;
2277 struct btrfs_qgroup
*qgroup
;
2278 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2279 struct ulist_node
*unode
;
2280 struct ulist_iterator uiter
;
2281 u64 ref_root
= root
->root_key
.objectid
;
2284 if (!is_fstree(ref_root
))
2290 spin_lock(&fs_info
->qgroup_lock
);
2292 quota_root
= fs_info
->quota_root
;
2296 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
2300 ulist_reinit(fs_info
->qgroup_ulist
);
2301 ret
= ulist_add(fs_info
->qgroup_ulist
, qgroup
->qgroupid
,
2302 (uintptr_t)qgroup
, GFP_ATOMIC
);
2305 ULIST_ITER_INIT(&uiter
);
2306 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2307 struct btrfs_qgroup
*qg
;
2308 struct btrfs_qgroup_list
*glist
;
2310 qg
= u64_to_ptr(unode
->aux
);
2312 qg
->reserved
-= num_bytes
;
2314 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2315 ret
= ulist_add(fs_info
->qgroup_ulist
,
2316 glist
->group
->qgroupid
,
2317 (uintptr_t)glist
->group
, GFP_ATOMIC
);
2324 spin_unlock(&fs_info
->qgroup_lock
);
2327 void assert_qgroups_uptodate(struct btrfs_trans_handle
*trans
)
2329 if (list_empty(&trans
->qgroup_ref_list
) && !trans
->delayed_ref_elem
.seq
)
2331 btrfs_err(trans
->root
->fs_info
,
2332 "qgroups not uptodate in trans handle %p: list is%s empty, "
2334 trans
, list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
2335 (u32
)(trans
->delayed_ref_elem
.seq
>> 32),
2336 (u32
)trans
->delayed_ref_elem
.seq
);
2341 * returns < 0 on error, 0 when more leafs are to be scanned.
2342 * returns 1 when done, 2 when done and FLAG_INCONSISTENT was cleared.
2345 qgroup_rescan_leaf(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
2346 struct btrfs_trans_handle
*trans
, struct ulist
*qgroups
,
2347 struct ulist
*tmp
, struct extent_buffer
*scratch_leaf
)
2349 struct btrfs_key found
;
2350 struct ulist
*roots
= NULL
;
2351 struct seq_list tree_mod_seq_elem
= {};
2358 path
->leave_spinning
= 1;
2359 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2360 ret
= btrfs_search_slot_for_read(fs_info
->extent_root
,
2361 &fs_info
->qgroup_rescan_progress
,
2364 pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2365 fs_info
->qgroup_rescan_progress
.objectid
,
2366 fs_info
->qgroup_rescan_progress
.type
,
2367 fs_info
->qgroup_rescan_progress
.offset
, ret
);
2371 * The rescan is about to end, we will not be scanning any
2372 * further blocks. We cannot unset the RESCAN flag here, because
2373 * we want to commit the transaction if everything went well.
2374 * To make the live accounting work in this phase, we set our
2375 * scan progress pointer such that every real extent objectid
2378 fs_info
->qgroup_rescan_progress
.objectid
= (u64
)-1;
2379 btrfs_release_path(path
);
2380 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2384 btrfs_item_key_to_cpu(path
->nodes
[0], &found
,
2385 btrfs_header_nritems(path
->nodes
[0]) - 1);
2386 fs_info
->qgroup_rescan_progress
.objectid
= found
.objectid
+ 1;
2388 btrfs_get_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2389 memcpy(scratch_leaf
, path
->nodes
[0], sizeof(*scratch_leaf
));
2390 slot
= path
->slots
[0];
2391 btrfs_release_path(path
);
2392 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2394 for (; slot
< btrfs_header_nritems(scratch_leaf
); ++slot
) {
2395 btrfs_item_key_to_cpu(scratch_leaf
, &found
, slot
);
2396 if (found
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
2397 found
.type
!= BTRFS_METADATA_ITEM_KEY
)
2399 if (found
.type
== BTRFS_METADATA_ITEM_KEY
)
2400 num_bytes
= fs_info
->extent_root
->leafsize
;
2402 num_bytes
= found
.offset
;
2404 ulist_reinit(qgroups
);
2405 ret
= btrfs_find_all_roots(NULL
, fs_info
, found
.objectid
, 0,
2409 spin_lock(&fs_info
->qgroup_lock
);
2410 seq
= fs_info
->qgroup_seq
;
2411 fs_info
->qgroup_seq
+= roots
->nnodes
+ 1; /* max refcnt */
2414 ret
= qgroup_calc_old_refcnt(fs_info
, 0, tmp
, roots
, qgroups
,
2415 seq
, &new_roots
, 1);
2417 spin_unlock(&fs_info
->qgroup_lock
);
2422 ret
= qgroup_adjust_counters(fs_info
, 0, num_bytes
, qgroups
,
2423 seq
, 0, new_roots
, 1);
2425 spin_unlock(&fs_info
->qgroup_lock
);
2429 spin_unlock(&fs_info
->qgroup_lock
);
2433 btrfs_put_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2438 static void btrfs_qgroup_rescan_worker(struct btrfs_work
*work
)
2440 struct btrfs_fs_info
*fs_info
= container_of(work
, struct btrfs_fs_info
,
2441 qgroup_rescan_work
);
2442 struct btrfs_path
*path
;
2443 struct btrfs_trans_handle
*trans
= NULL
;
2444 struct ulist
*tmp
= NULL
, *qgroups
= NULL
;
2445 struct extent_buffer
*scratch_leaf
= NULL
;
2448 path
= btrfs_alloc_path();
2451 qgroups
= ulist_alloc(GFP_NOFS
);
2454 tmp
= ulist_alloc(GFP_NOFS
);
2457 scratch_leaf
= kmalloc(sizeof(*scratch_leaf
), GFP_NOFS
);
2463 trans
= btrfs_start_transaction(fs_info
->fs_root
, 0);
2464 if (IS_ERR(trans
)) {
2465 err
= PTR_ERR(trans
);
2468 if (!fs_info
->quota_enabled
) {
2471 err
= qgroup_rescan_leaf(fs_info
, path
, trans
,
2472 qgroups
, tmp
, scratch_leaf
);
2475 btrfs_commit_transaction(trans
, fs_info
->fs_root
);
2477 btrfs_end_transaction(trans
, fs_info
->fs_root
);
2481 kfree(scratch_leaf
);
2482 ulist_free(qgroups
);
2484 btrfs_free_path(path
);
2486 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2487 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2490 fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
) {
2491 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2492 } else if (err
< 0) {
2493 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2495 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2498 btrfs_info(fs_info
, "qgroup scan completed%s",
2499 err
== 2 ? " (inconsistency flag cleared)" : "");
2501 btrfs_err(fs_info
, "qgroup scan failed with %d", err
);
2504 complete_all(&fs_info
->qgroup_rescan_completion
);
2508 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2509 * memory required for the rescan context.
2512 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
2518 (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) ||
2519 !(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))) {
2524 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2525 spin_lock(&fs_info
->qgroup_lock
);
2528 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
2530 else if (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))
2534 spin_unlock(&fs_info
->qgroup_lock
);
2535 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2539 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2542 memset(&fs_info
->qgroup_rescan_progress
, 0,
2543 sizeof(fs_info
->qgroup_rescan_progress
));
2544 fs_info
->qgroup_rescan_progress
.objectid
= progress_objectid
;
2546 spin_unlock(&fs_info
->qgroup_lock
);
2547 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2549 init_completion(&fs_info
->qgroup_rescan_completion
);
2551 memset(&fs_info
->qgroup_rescan_work
, 0,
2552 sizeof(fs_info
->qgroup_rescan_work
));
2553 btrfs_init_work(&fs_info
->qgroup_rescan_work
,
2554 btrfs_qgroup_rescan_worker
, NULL
, NULL
);
2558 btrfs_info(fs_info
, "qgroup_rescan_init failed with %d", ret
);
2566 qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
)
2569 struct btrfs_qgroup
*qgroup
;
2571 spin_lock(&fs_info
->qgroup_lock
);
2572 /* clear all current qgroup tracking information */
2573 for (n
= rb_first(&fs_info
->qgroup_tree
); n
; n
= rb_next(n
)) {
2574 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
2576 qgroup
->rfer_cmpr
= 0;
2578 qgroup
->excl_cmpr
= 0;
2580 spin_unlock(&fs_info
->qgroup_lock
);
2584 btrfs_qgroup_rescan(struct btrfs_fs_info
*fs_info
)
2587 struct btrfs_trans_handle
*trans
;
2589 ret
= qgroup_rescan_init(fs_info
, 0, 1);
2594 * We have set the rescan_progress to 0, which means no more
2595 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2596 * However, btrfs_qgroup_account_ref may be right after its call
2597 * to btrfs_find_all_roots, in which case it would still do the
2599 * To solve this, we're committing the transaction, which will
2600 * ensure we run all delayed refs and only after that, we are
2601 * going to clear all tracking information for a clean start.
2604 trans
= btrfs_join_transaction(fs_info
->fs_root
);
2605 if (IS_ERR(trans
)) {
2606 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2607 return PTR_ERR(trans
);
2609 ret
= btrfs_commit_transaction(trans
, fs_info
->fs_root
);
2611 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2615 qgroup_rescan_zero_tracking(fs_info
);
2617 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2618 &fs_info
->qgroup_rescan_work
);
2623 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info
*fs_info
)
2628 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2629 spin_lock(&fs_info
->qgroup_lock
);
2630 running
= fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2631 spin_unlock(&fs_info
->qgroup_lock
);
2632 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2635 ret
= wait_for_completion_interruptible(
2636 &fs_info
->qgroup_rescan_completion
);
2642 * this is only called from open_ctree where we're still single threaded, thus
2643 * locking is omitted here.
2646 btrfs_qgroup_rescan_resume(struct btrfs_fs_info
*fs_info
)
2648 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
2649 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2650 &fs_info
->qgroup_rescan_work
);