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 if (btrfs_test_is_dummy_root(quota_root
))
545 path
= btrfs_alloc_path();
550 key
.type
= BTRFS_QGROUP_INFO_KEY
;
551 key
.offset
= qgroupid
;
554 * Avoid a transaction abort by catching -EEXIST here. In that
555 * case, we proceed by re-initializing the existing structure
559 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
560 sizeof(*qgroup_info
));
561 if (ret
&& ret
!= -EEXIST
)
564 leaf
= path
->nodes
[0];
565 qgroup_info
= btrfs_item_ptr(leaf
, path
->slots
[0],
566 struct btrfs_qgroup_info_item
);
567 btrfs_set_qgroup_info_generation(leaf
, qgroup_info
, trans
->transid
);
568 btrfs_set_qgroup_info_rfer(leaf
, qgroup_info
, 0);
569 btrfs_set_qgroup_info_rfer_cmpr(leaf
, qgroup_info
, 0);
570 btrfs_set_qgroup_info_excl(leaf
, qgroup_info
, 0);
571 btrfs_set_qgroup_info_excl_cmpr(leaf
, qgroup_info
, 0);
573 btrfs_mark_buffer_dirty(leaf
);
575 btrfs_release_path(path
);
577 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
578 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
579 sizeof(*qgroup_limit
));
580 if (ret
&& ret
!= -EEXIST
)
583 leaf
= path
->nodes
[0];
584 qgroup_limit
= btrfs_item_ptr(leaf
, path
->slots
[0],
585 struct btrfs_qgroup_limit_item
);
586 btrfs_set_qgroup_limit_flags(leaf
, qgroup_limit
, 0);
587 btrfs_set_qgroup_limit_max_rfer(leaf
, qgroup_limit
, 0);
588 btrfs_set_qgroup_limit_max_excl(leaf
, qgroup_limit
, 0);
589 btrfs_set_qgroup_limit_rsv_rfer(leaf
, qgroup_limit
, 0);
590 btrfs_set_qgroup_limit_rsv_excl(leaf
, qgroup_limit
, 0);
592 btrfs_mark_buffer_dirty(leaf
);
596 btrfs_free_path(path
);
600 static int del_qgroup_item(struct btrfs_trans_handle
*trans
,
601 struct btrfs_root
*quota_root
, u64 qgroupid
)
604 struct btrfs_path
*path
;
605 struct btrfs_key key
;
607 path
= btrfs_alloc_path();
612 key
.type
= BTRFS_QGROUP_INFO_KEY
;
613 key
.offset
= qgroupid
;
614 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
623 ret
= btrfs_del_item(trans
, quota_root
, path
);
627 btrfs_release_path(path
);
629 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
630 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
639 ret
= btrfs_del_item(trans
, quota_root
, path
);
642 btrfs_free_path(path
);
646 static int update_qgroup_limit_item(struct btrfs_trans_handle
*trans
,
647 struct btrfs_root
*root
,
648 struct btrfs_qgroup
*qgroup
)
650 struct btrfs_path
*path
;
651 struct btrfs_key key
;
652 struct extent_buffer
*l
;
653 struct btrfs_qgroup_limit_item
*qgroup_limit
;
658 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
659 key
.offset
= qgroup
->qgroupid
;
661 path
= btrfs_alloc_path();
665 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
673 slot
= path
->slots
[0];
674 qgroup_limit
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_limit_item
);
675 btrfs_set_qgroup_limit_flags(l
, qgroup_limit
, qgroup
->lim_flags
);
676 btrfs_set_qgroup_limit_max_rfer(l
, qgroup_limit
, qgroup
->max_rfer
);
677 btrfs_set_qgroup_limit_max_excl(l
, qgroup_limit
, qgroup
->max_excl
);
678 btrfs_set_qgroup_limit_rsv_rfer(l
, qgroup_limit
, qgroup
->rsv_rfer
);
679 btrfs_set_qgroup_limit_rsv_excl(l
, qgroup_limit
, qgroup
->rsv_excl
);
681 btrfs_mark_buffer_dirty(l
);
684 btrfs_free_path(path
);
688 static int update_qgroup_info_item(struct btrfs_trans_handle
*trans
,
689 struct btrfs_root
*root
,
690 struct btrfs_qgroup
*qgroup
)
692 struct btrfs_path
*path
;
693 struct btrfs_key key
;
694 struct extent_buffer
*l
;
695 struct btrfs_qgroup_info_item
*qgroup_info
;
699 if (btrfs_test_is_dummy_root(root
))
703 key
.type
= BTRFS_QGROUP_INFO_KEY
;
704 key
.offset
= qgroup
->qgroupid
;
706 path
= btrfs_alloc_path();
710 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
718 slot
= path
->slots
[0];
719 qgroup_info
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_info_item
);
720 btrfs_set_qgroup_info_generation(l
, qgroup_info
, trans
->transid
);
721 btrfs_set_qgroup_info_rfer(l
, qgroup_info
, qgroup
->rfer
);
722 btrfs_set_qgroup_info_rfer_cmpr(l
, qgroup_info
, qgroup
->rfer_cmpr
);
723 btrfs_set_qgroup_info_excl(l
, qgroup_info
, qgroup
->excl
);
724 btrfs_set_qgroup_info_excl_cmpr(l
, qgroup_info
, qgroup
->excl_cmpr
);
726 btrfs_mark_buffer_dirty(l
);
729 btrfs_free_path(path
);
733 static int update_qgroup_status_item(struct btrfs_trans_handle
*trans
,
734 struct btrfs_fs_info
*fs_info
,
735 struct btrfs_root
*root
)
737 struct btrfs_path
*path
;
738 struct btrfs_key key
;
739 struct extent_buffer
*l
;
740 struct btrfs_qgroup_status_item
*ptr
;
745 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
748 path
= btrfs_alloc_path();
752 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
760 slot
= path
->slots
[0];
761 ptr
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_status_item
);
762 btrfs_set_qgroup_status_flags(l
, ptr
, fs_info
->qgroup_flags
);
763 btrfs_set_qgroup_status_generation(l
, ptr
, trans
->transid
);
764 btrfs_set_qgroup_status_rescan(l
, ptr
,
765 fs_info
->qgroup_rescan_progress
.objectid
);
767 btrfs_mark_buffer_dirty(l
);
770 btrfs_free_path(path
);
775 * called with qgroup_lock held
777 static int btrfs_clean_quota_tree(struct btrfs_trans_handle
*trans
,
778 struct btrfs_root
*root
)
780 struct btrfs_path
*path
;
781 struct btrfs_key key
;
782 struct extent_buffer
*leaf
= NULL
;
786 path
= btrfs_alloc_path();
790 path
->leave_spinning
= 1;
797 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
800 leaf
= path
->nodes
[0];
801 nr
= btrfs_header_nritems(leaf
);
805 * delete the leaf one by one
806 * since the whole tree is going
810 ret
= btrfs_del_items(trans
, root
, path
, 0, nr
);
814 btrfs_release_path(path
);
818 root
->fs_info
->pending_quota_state
= 0;
819 btrfs_free_path(path
);
823 int btrfs_quota_enable(struct btrfs_trans_handle
*trans
,
824 struct btrfs_fs_info
*fs_info
)
826 struct btrfs_root
*quota_root
;
827 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
828 struct btrfs_path
*path
= NULL
;
829 struct btrfs_qgroup_status_item
*ptr
;
830 struct extent_buffer
*leaf
;
831 struct btrfs_key key
;
832 struct btrfs_key found_key
;
833 struct btrfs_qgroup
*qgroup
= NULL
;
837 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
838 if (fs_info
->quota_root
) {
839 fs_info
->pending_quota_state
= 1;
843 fs_info
->qgroup_ulist
= ulist_alloc(GFP_NOFS
);
844 if (!fs_info
->qgroup_ulist
) {
850 * initially create the quota tree
852 quota_root
= btrfs_create_tree(trans
, fs_info
,
853 BTRFS_QUOTA_TREE_OBJECTID
);
854 if (IS_ERR(quota_root
)) {
855 ret
= PTR_ERR(quota_root
);
859 path
= btrfs_alloc_path();
866 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
869 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
874 leaf
= path
->nodes
[0];
875 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0],
876 struct btrfs_qgroup_status_item
);
877 btrfs_set_qgroup_status_generation(leaf
, ptr
, trans
->transid
);
878 btrfs_set_qgroup_status_version(leaf
, ptr
, BTRFS_QGROUP_STATUS_VERSION
);
879 fs_info
->qgroup_flags
= BTRFS_QGROUP_STATUS_FLAG_ON
|
880 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
881 btrfs_set_qgroup_status_flags(leaf
, ptr
, fs_info
->qgroup_flags
);
882 btrfs_set_qgroup_status_rescan(leaf
, ptr
, 0);
884 btrfs_mark_buffer_dirty(leaf
);
887 key
.type
= BTRFS_ROOT_REF_KEY
;
890 btrfs_release_path(path
);
891 ret
= btrfs_search_slot_for_read(tree_root
, &key
, path
, 1, 0);
899 slot
= path
->slots
[0];
900 leaf
= path
->nodes
[0];
901 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
903 if (found_key
.type
== BTRFS_ROOT_REF_KEY
) {
904 ret
= add_qgroup_item(trans
, quota_root
,
909 qgroup
= add_qgroup_rb(fs_info
, found_key
.offset
);
910 if (IS_ERR(qgroup
)) {
911 ret
= PTR_ERR(qgroup
);
915 ret
= btrfs_next_item(tree_root
, path
);
923 btrfs_release_path(path
);
924 ret
= add_qgroup_item(trans
, quota_root
, BTRFS_FS_TREE_OBJECTID
);
928 qgroup
= add_qgroup_rb(fs_info
, BTRFS_FS_TREE_OBJECTID
);
929 if (IS_ERR(qgroup
)) {
930 ret
= PTR_ERR(qgroup
);
933 spin_lock(&fs_info
->qgroup_lock
);
934 fs_info
->quota_root
= quota_root
;
935 fs_info
->pending_quota_state
= 1;
936 spin_unlock(&fs_info
->qgroup_lock
);
938 btrfs_free_path(path
);
941 free_extent_buffer(quota_root
->node
);
942 free_extent_buffer(quota_root
->commit_root
);
947 ulist_free(fs_info
->qgroup_ulist
);
948 fs_info
->qgroup_ulist
= NULL
;
950 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
954 int btrfs_quota_disable(struct btrfs_trans_handle
*trans
,
955 struct btrfs_fs_info
*fs_info
)
957 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
958 struct btrfs_root
*quota_root
;
961 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
962 if (!fs_info
->quota_root
)
964 spin_lock(&fs_info
->qgroup_lock
);
965 fs_info
->quota_enabled
= 0;
966 fs_info
->pending_quota_state
= 0;
967 quota_root
= fs_info
->quota_root
;
968 fs_info
->quota_root
= NULL
;
969 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
970 spin_unlock(&fs_info
->qgroup_lock
);
972 btrfs_free_qgroup_config(fs_info
);
974 ret
= btrfs_clean_quota_tree(trans
, quota_root
);
978 ret
= btrfs_del_root(trans
, tree_root
, "a_root
->root_key
);
982 list_del("a_root
->dirty_list
);
984 btrfs_tree_lock(quota_root
->node
);
985 clean_tree_block(trans
, tree_root
->fs_info
, quota_root
->node
);
986 btrfs_tree_unlock(quota_root
->node
);
987 btrfs_free_tree_block(trans
, quota_root
, quota_root
->node
, 0, 1);
989 free_extent_buffer(quota_root
->node
);
990 free_extent_buffer(quota_root
->commit_root
);
993 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
997 static void qgroup_dirty(struct btrfs_fs_info
*fs_info
,
998 struct btrfs_qgroup
*qgroup
)
1000 if (list_empty(&qgroup
->dirty
))
1001 list_add(&qgroup
->dirty
, &fs_info
->dirty_qgroups
);
1005 * The easy accounting, if we are adding/removing the only ref for an extent
1006 * then this qgroup and all of the parent qgroups get their refrence and
1007 * exclusive counts adjusted.
1009 * Caller should hold fs_info->qgroup_lock.
1011 static int __qgroup_excl_accounting(struct btrfs_fs_info
*fs_info
,
1012 struct ulist
*tmp
, u64 ref_root
,
1013 u64 num_bytes
, int sign
)
1015 struct btrfs_qgroup
*qgroup
;
1016 struct btrfs_qgroup_list
*glist
;
1017 struct ulist_node
*unode
;
1018 struct ulist_iterator uiter
;
1021 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
1025 qgroup
->rfer
+= sign
* num_bytes
;
1026 qgroup
->rfer_cmpr
+= sign
* num_bytes
;
1028 WARN_ON(sign
< 0 && qgroup
->excl
< num_bytes
);
1029 qgroup
->excl
+= sign
* num_bytes
;
1030 qgroup
->excl_cmpr
+= sign
* num_bytes
;
1032 qgroup
->reserved
-= num_bytes
;
1034 qgroup_dirty(fs_info
, qgroup
);
1036 /* Get all of the parent groups that contain this qgroup */
1037 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
1038 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1039 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1044 /* Iterate all of the parents and adjust their reference counts */
1045 ULIST_ITER_INIT(&uiter
);
1046 while ((unode
= ulist_next(tmp
, &uiter
))) {
1047 qgroup
= u64_to_ptr(unode
->aux
);
1048 qgroup
->rfer
+= sign
* num_bytes
;
1049 qgroup
->rfer_cmpr
+= sign
* num_bytes
;
1050 WARN_ON(sign
< 0 && qgroup
->excl
< num_bytes
);
1051 qgroup
->excl
+= sign
* num_bytes
;
1053 qgroup
->reserved
-= num_bytes
;
1054 qgroup
->excl_cmpr
+= sign
* num_bytes
;
1055 qgroup_dirty(fs_info
, qgroup
);
1057 /* Add any parents of the parents */
1058 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
1059 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1060 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1072 * Quick path for updating qgroup with only excl refs.
1074 * In that case, just update all parent will be enough.
1075 * Or we needs to do a full rescan.
1076 * Caller should also hold fs_info->qgroup_lock.
1078 * Return 0 for quick update, return >0 for need to full rescan
1079 * and mark INCONSISTENT flag.
1080 * Return < 0 for other error.
1082 static int quick_update_accounting(struct btrfs_fs_info
*fs_info
,
1083 struct ulist
*tmp
, u64 src
, u64 dst
,
1086 struct btrfs_qgroup
*qgroup
;
1090 qgroup
= find_qgroup_rb(fs_info
, src
);
1093 if (qgroup
->excl
== qgroup
->rfer
) {
1095 err
= __qgroup_excl_accounting(fs_info
, tmp
, dst
,
1096 qgroup
->excl
, sign
);
1104 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1108 int btrfs_add_qgroup_relation(struct btrfs_trans_handle
*trans
,
1109 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1111 struct btrfs_root
*quota_root
;
1112 struct btrfs_qgroup
*parent
;
1113 struct btrfs_qgroup
*member
;
1114 struct btrfs_qgroup_list
*list
;
1118 /* Check the level of src and dst first */
1119 if (btrfs_qgroup_level(src
) >= btrfs_qgroup_level(dst
))
1122 tmp
= ulist_alloc(GFP_NOFS
);
1126 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1127 quota_root
= fs_info
->quota_root
;
1132 member
= find_qgroup_rb(fs_info
, src
);
1133 parent
= find_qgroup_rb(fs_info
, dst
);
1134 if (!member
|| !parent
) {
1139 /* check if such qgroup relation exist firstly */
1140 list_for_each_entry(list
, &member
->groups
, next_group
) {
1141 if (list
->group
== parent
) {
1147 ret
= add_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1151 ret
= add_qgroup_relation_item(trans
, quota_root
, dst
, src
);
1153 del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1157 spin_lock(&fs_info
->qgroup_lock
);
1158 ret
= add_relation_rb(quota_root
->fs_info
, src
, dst
);
1160 spin_unlock(&fs_info
->qgroup_lock
);
1163 ret
= quick_update_accounting(fs_info
, tmp
, src
, dst
, 1);
1164 spin_unlock(&fs_info
->qgroup_lock
);
1166 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1171 int __del_qgroup_relation(struct btrfs_trans_handle
*trans
,
1172 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1174 struct btrfs_root
*quota_root
;
1175 struct btrfs_qgroup
*parent
;
1176 struct btrfs_qgroup
*member
;
1177 struct btrfs_qgroup_list
*list
;
1182 tmp
= ulist_alloc(GFP_NOFS
);
1186 quota_root
= fs_info
->quota_root
;
1192 member
= find_qgroup_rb(fs_info
, src
);
1193 parent
= find_qgroup_rb(fs_info
, dst
);
1194 if (!member
|| !parent
) {
1199 /* check if such qgroup relation exist firstly */
1200 list_for_each_entry(list
, &member
->groups
, next_group
) {
1201 if (list
->group
== parent
)
1207 ret
= del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1208 err
= del_qgroup_relation_item(trans
, quota_root
, dst
, src
);
1212 spin_lock(&fs_info
->qgroup_lock
);
1213 del_relation_rb(fs_info
, src
, dst
);
1214 ret
= quick_update_accounting(fs_info
, tmp
, src
, dst
, -1);
1215 spin_unlock(&fs_info
->qgroup_lock
);
1221 int btrfs_del_qgroup_relation(struct btrfs_trans_handle
*trans
,
1222 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1226 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1227 ret
= __del_qgroup_relation(trans
, fs_info
, src
, dst
);
1228 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1233 int btrfs_create_qgroup(struct btrfs_trans_handle
*trans
,
1234 struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
1236 struct btrfs_root
*quota_root
;
1237 struct btrfs_qgroup
*qgroup
;
1240 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1241 quota_root
= fs_info
->quota_root
;
1246 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1252 ret
= add_qgroup_item(trans
, quota_root
, qgroupid
);
1256 spin_lock(&fs_info
->qgroup_lock
);
1257 qgroup
= add_qgroup_rb(fs_info
, qgroupid
);
1258 spin_unlock(&fs_info
->qgroup_lock
);
1261 ret
= PTR_ERR(qgroup
);
1263 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1267 int btrfs_remove_qgroup(struct btrfs_trans_handle
*trans
,
1268 struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
1270 struct btrfs_root
*quota_root
;
1271 struct btrfs_qgroup
*qgroup
;
1272 struct btrfs_qgroup_list
*list
;
1275 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1276 quota_root
= fs_info
->quota_root
;
1282 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1287 /* check if there are no children of this qgroup */
1288 if (!list_empty(&qgroup
->members
)) {
1293 ret
= del_qgroup_item(trans
, quota_root
, qgroupid
);
1295 while (!list_empty(&qgroup
->groups
)) {
1296 list
= list_first_entry(&qgroup
->groups
,
1297 struct btrfs_qgroup_list
, next_group
);
1298 ret
= __del_qgroup_relation(trans
, fs_info
,
1300 list
->group
->qgroupid
);
1305 spin_lock(&fs_info
->qgroup_lock
);
1306 del_qgroup_rb(quota_root
->fs_info
, qgroupid
);
1307 spin_unlock(&fs_info
->qgroup_lock
);
1309 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1313 int btrfs_limit_qgroup(struct btrfs_trans_handle
*trans
,
1314 struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
1315 struct btrfs_qgroup_limit
*limit
)
1317 struct btrfs_root
*quota_root
;
1318 struct btrfs_qgroup
*qgroup
;
1321 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1322 quota_root
= fs_info
->quota_root
;
1328 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1334 spin_lock(&fs_info
->qgroup_lock
);
1335 if (limit
->flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
)
1336 qgroup
->max_rfer
= limit
->max_rfer
;
1337 if (limit
->flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
)
1338 qgroup
->max_excl
= limit
->max_excl
;
1339 if (limit
->flags
& BTRFS_QGROUP_LIMIT_RSV_RFER
)
1340 qgroup
->rsv_rfer
= limit
->rsv_rfer
;
1341 if (limit
->flags
& BTRFS_QGROUP_LIMIT_RSV_EXCL
)
1342 qgroup
->rsv_excl
= limit
->rsv_excl
;
1343 qgroup
->lim_flags
|= limit
->flags
;
1345 spin_unlock(&fs_info
->qgroup_lock
);
1347 ret
= update_qgroup_limit_item(trans
, quota_root
, qgroup
);
1349 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1350 btrfs_info(fs_info
, "unable to update quota limit for %llu",
1355 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1359 static int comp_oper_exist(struct btrfs_qgroup_operation
*oper1
,
1360 struct btrfs_qgroup_operation
*oper2
)
1363 * Ignore seq and type here, we're looking for any operation
1364 * at all related to this extent on that root.
1366 if (oper1
->bytenr
< oper2
->bytenr
)
1368 if (oper1
->bytenr
> oper2
->bytenr
)
1370 if (oper1
->ref_root
< oper2
->ref_root
)
1372 if (oper1
->ref_root
> oper2
->ref_root
)
1377 static int qgroup_oper_exists(struct btrfs_fs_info
*fs_info
,
1378 struct btrfs_qgroup_operation
*oper
)
1381 struct btrfs_qgroup_operation
*cur
;
1384 spin_lock(&fs_info
->qgroup_op_lock
);
1385 n
= fs_info
->qgroup_op_tree
.rb_node
;
1387 cur
= rb_entry(n
, struct btrfs_qgroup_operation
, n
);
1388 cmp
= comp_oper_exist(cur
, oper
);
1394 spin_unlock(&fs_info
->qgroup_op_lock
);
1398 spin_unlock(&fs_info
->qgroup_op_lock
);
1402 static int comp_oper(struct btrfs_qgroup_operation
*oper1
,
1403 struct btrfs_qgroup_operation
*oper2
)
1405 if (oper1
->bytenr
< oper2
->bytenr
)
1407 if (oper1
->bytenr
> oper2
->bytenr
)
1409 if (oper1
->ref_root
< oper2
->ref_root
)
1411 if (oper1
->ref_root
> oper2
->ref_root
)
1413 if (oper1
->seq
< oper2
->seq
)
1415 if (oper1
->seq
> oper2
->seq
)
1417 if (oper1
->type
< oper2
->type
)
1419 if (oper1
->type
> oper2
->type
)
1424 static int insert_qgroup_oper(struct btrfs_fs_info
*fs_info
,
1425 struct btrfs_qgroup_operation
*oper
)
1428 struct rb_node
*parent
= NULL
;
1429 struct btrfs_qgroup_operation
*cur
;
1432 spin_lock(&fs_info
->qgroup_op_lock
);
1433 p
= &fs_info
->qgroup_op_tree
.rb_node
;
1436 cur
= rb_entry(parent
, struct btrfs_qgroup_operation
, n
);
1437 cmp
= comp_oper(cur
, oper
);
1439 p
= &(*p
)->rb_right
;
1443 spin_unlock(&fs_info
->qgroup_op_lock
);
1447 rb_link_node(&oper
->n
, parent
, p
);
1448 rb_insert_color(&oper
->n
, &fs_info
->qgroup_op_tree
);
1449 spin_unlock(&fs_info
->qgroup_op_lock
);
1454 * Record a quota operation for processing later on.
1455 * @trans: the transaction we are adding the delayed op to.
1456 * @fs_info: the fs_info for this fs.
1457 * @ref_root: the root of the reference we are acting on,
1458 * @bytenr: the bytenr we are acting on.
1459 * @num_bytes: the number of bytes in the reference.
1460 * @type: the type of operation this is.
1461 * @mod_seq: do we need to get a sequence number for looking up roots.
1463 * We just add it to our trans qgroup_ref_list and carry on and process these
1464 * operations in order at some later point. If the reference root isn't a fs
1465 * root then we don't bother with doing anything.
1467 * MUST BE HOLDING THE REF LOCK.
1469 int btrfs_qgroup_record_ref(struct btrfs_trans_handle
*trans
,
1470 struct btrfs_fs_info
*fs_info
, u64 ref_root
,
1471 u64 bytenr
, u64 num_bytes
,
1472 enum btrfs_qgroup_operation_type type
, int mod_seq
)
1474 struct btrfs_qgroup_operation
*oper
;
1477 if (!is_fstree(ref_root
) || !fs_info
->quota_enabled
)
1480 oper
= kmalloc(sizeof(*oper
), GFP_NOFS
);
1484 oper
->ref_root
= ref_root
;
1485 oper
->bytenr
= bytenr
;
1486 oper
->num_bytes
= num_bytes
;
1488 oper
->seq
= atomic_inc_return(&fs_info
->qgroup_op_seq
);
1489 INIT_LIST_HEAD(&oper
->elem
.list
);
1492 trace_btrfs_qgroup_record_ref(oper
);
1494 if (type
== BTRFS_QGROUP_OPER_SUB_SUBTREE
) {
1496 * If any operation for this bytenr/ref_root combo
1497 * exists, then we know it's not exclusively owned and
1498 * shouldn't be queued up.
1500 * This also catches the case where we have a cloned
1501 * extent that gets queued up multiple times during
1504 if (qgroup_oper_exists(fs_info
, oper
)) {
1510 ret
= insert_qgroup_oper(fs_info
, oper
);
1512 /* Shouldn't happen so have an assert for developers */
1517 list_add_tail(&oper
->list
, &trans
->qgroup_ref_list
);
1520 btrfs_get_tree_mod_seq(fs_info
, &oper
->elem
);
1525 static int qgroup_excl_accounting(struct btrfs_fs_info
*fs_info
,
1526 struct btrfs_qgroup_operation
*oper
)
1532 tmp
= ulist_alloc(GFP_NOFS
);
1536 spin_lock(&fs_info
->qgroup_lock
);
1537 if (!fs_info
->quota_root
)
1540 switch (oper
->type
) {
1541 case BTRFS_QGROUP_OPER_ADD_EXCL
:
1544 case BTRFS_QGROUP_OPER_SUB_EXCL
:
1550 ret
= __qgroup_excl_accounting(fs_info
, tmp
, oper
->ref_root
,
1551 oper
->num_bytes
, sign
);
1553 spin_unlock(&fs_info
->qgroup_lock
);
1559 * Walk all of the roots that pointed to our bytenr and adjust their refcnts as
1562 static int qgroup_calc_old_refcnt(struct btrfs_fs_info
*fs_info
,
1563 u64 root_to_skip
, struct ulist
*tmp
,
1564 struct ulist
*roots
, struct ulist
*qgroups
,
1565 u64 seq
, int *old_roots
, int rescan
)
1567 struct ulist_node
*unode
;
1568 struct ulist_iterator uiter
;
1569 struct ulist_node
*tmp_unode
;
1570 struct ulist_iterator tmp_uiter
;
1571 struct btrfs_qgroup
*qg
;
1574 ULIST_ITER_INIT(&uiter
);
1575 while ((unode
= ulist_next(roots
, &uiter
))) {
1576 /* We don't count our current root here */
1577 if (unode
->val
== root_to_skip
)
1579 qg
= find_qgroup_rb(fs_info
, unode
->val
);
1583 * We could have a pending removal of this same ref so we may
1584 * not have actually found our ref root when doing
1585 * btrfs_find_all_roots, so we need to keep track of how many
1586 * old roots we find in case we removed ours and added a
1587 * different one at the same time. I don't think this could
1588 * happen in practice but that sort of thinking leads to pain
1589 * and suffering and to the dark side.
1594 ret
= ulist_add(qgroups
, qg
->qgroupid
, ptr_to_u64(qg
),
1598 ret
= ulist_add(tmp
, qg
->qgroupid
, ptr_to_u64(qg
), GFP_ATOMIC
);
1601 ULIST_ITER_INIT(&tmp_uiter
);
1602 while ((tmp_unode
= ulist_next(tmp
, &tmp_uiter
))) {
1603 struct btrfs_qgroup_list
*glist
;
1605 qg
= u64_to_ptr(tmp_unode
->aux
);
1607 * We use this sequence number to keep from having to
1608 * run the whole list and 0 out the refcnt every time.
1609 * We basically use sequnce as the known 0 count and
1610 * then add 1 everytime we see a qgroup. This is how we
1611 * get how many of the roots actually point up to the
1612 * upper level qgroups in order to determine exclusive
1615 * For rescan we want to set old_refcnt to seq so our
1616 * exclusive calculations end up correct.
1619 qg
->old_refcnt
= seq
;
1620 else if (qg
->old_refcnt
< seq
)
1621 qg
->old_refcnt
= seq
+ 1;
1625 if (qg
->new_refcnt
< seq
)
1626 qg
->new_refcnt
= seq
+ 1;
1629 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1630 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1631 ptr_to_u64(glist
->group
),
1635 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1636 ptr_to_u64(glist
->group
),
1647 * We need to walk forward in our operation tree and account for any roots that
1648 * were deleted after we made this operation.
1650 static int qgroup_account_deleted_refs(struct btrfs_fs_info
*fs_info
,
1651 struct btrfs_qgroup_operation
*oper
,
1653 struct ulist
*qgroups
, u64 seq
,
1656 struct ulist_node
*unode
;
1657 struct ulist_iterator uiter
;
1658 struct btrfs_qgroup
*qg
;
1659 struct btrfs_qgroup_operation
*tmp_oper
;
1666 * We only walk forward in the tree since we're only interested in
1667 * removals that happened _after_ our operation.
1669 spin_lock(&fs_info
->qgroup_op_lock
);
1670 n
= rb_next(&oper
->n
);
1671 spin_unlock(&fs_info
->qgroup_op_lock
);
1674 tmp_oper
= rb_entry(n
, struct btrfs_qgroup_operation
, n
);
1675 while (tmp_oper
->bytenr
== oper
->bytenr
) {
1677 * If it's not a removal we don't care, additions work out
1678 * properly with our refcnt tracking.
1680 if (tmp_oper
->type
!= BTRFS_QGROUP_OPER_SUB_SHARED
&&
1681 tmp_oper
->type
!= BTRFS_QGROUP_OPER_SUB_EXCL
)
1683 qg
= find_qgroup_rb(fs_info
, tmp_oper
->ref_root
);
1686 ret
= ulist_add(qgroups
, qg
->qgroupid
, ptr_to_u64(qg
),
1692 * We only want to increase old_roots if this qgroup is
1693 * not already in the list of qgroups. If it is already
1694 * there then that means it must have been re-added or
1695 * the delete will be discarded because we had an
1696 * existing ref that we haven't looked up yet. In this
1697 * case we don't want to increase old_roots. So if ret
1698 * == 1 then we know that this is the first time we've
1699 * seen this qgroup and we can bump the old_roots.
1702 ret
= ulist_add(tmp
, qg
->qgroupid
, ptr_to_u64(qg
),
1708 spin_lock(&fs_info
->qgroup_op_lock
);
1709 n
= rb_next(&tmp_oper
->n
);
1710 spin_unlock(&fs_info
->qgroup_op_lock
);
1713 tmp_oper
= rb_entry(n
, struct btrfs_qgroup_operation
, n
);
1716 /* Ok now process the qgroups we found */
1717 ULIST_ITER_INIT(&uiter
);
1718 while ((unode
= ulist_next(tmp
, &uiter
))) {
1719 struct btrfs_qgroup_list
*glist
;
1721 qg
= u64_to_ptr(unode
->aux
);
1722 if (qg
->old_refcnt
< seq
)
1723 qg
->old_refcnt
= seq
+ 1;
1726 if (qg
->new_refcnt
< seq
)
1727 qg
->new_refcnt
= seq
+ 1;
1730 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1731 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1732 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1735 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1736 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1744 /* Add refcnt for the newly added reference. */
1745 static int qgroup_calc_new_refcnt(struct btrfs_fs_info
*fs_info
,
1746 struct btrfs_qgroup_operation
*oper
,
1747 struct btrfs_qgroup
*qgroup
,
1748 struct ulist
*tmp
, struct ulist
*qgroups
,
1751 struct ulist_node
*unode
;
1752 struct ulist_iterator uiter
;
1753 struct btrfs_qgroup
*qg
;
1757 ret
= ulist_add(qgroups
, qgroup
->qgroupid
, ptr_to_u64(qgroup
),
1761 ret
= ulist_add(tmp
, qgroup
->qgroupid
, ptr_to_u64(qgroup
),
1765 ULIST_ITER_INIT(&uiter
);
1766 while ((unode
= ulist_next(tmp
, &uiter
))) {
1767 struct btrfs_qgroup_list
*glist
;
1769 qg
= u64_to_ptr(unode
->aux
);
1770 if (oper
->type
== BTRFS_QGROUP_OPER_ADD_SHARED
) {
1771 if (qg
->new_refcnt
< seq
)
1772 qg
->new_refcnt
= seq
+ 1;
1776 if (qg
->old_refcnt
< seq
)
1777 qg
->old_refcnt
= seq
+ 1;
1781 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1782 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1783 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1786 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1787 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
1796 * This adjusts the counters for all referenced qgroups if need be.
1798 static int qgroup_adjust_counters(struct btrfs_fs_info
*fs_info
,
1799 u64 root_to_skip
, u64 num_bytes
,
1800 struct ulist
*qgroups
, u64 seq
,
1801 int old_roots
, int new_roots
, int rescan
)
1803 struct ulist_node
*unode
;
1804 struct ulist_iterator uiter
;
1805 struct btrfs_qgroup
*qg
;
1806 u64 cur_new_count
, cur_old_count
;
1808 ULIST_ITER_INIT(&uiter
);
1809 while ((unode
= ulist_next(qgroups
, &uiter
))) {
1812 qg
= u64_to_ptr(unode
->aux
);
1814 * Wasn't referenced before but is now, add to the reference
1817 if (qg
->old_refcnt
<= seq
&& qg
->new_refcnt
> seq
) {
1818 qg
->rfer
+= num_bytes
;
1819 qg
->rfer_cmpr
+= num_bytes
;
1824 * Was referenced before but isn't now, subtract from the
1825 * reference counters.
1827 if (qg
->old_refcnt
> seq
&& qg
->new_refcnt
<= seq
) {
1828 qg
->rfer
-= num_bytes
;
1829 qg
->rfer_cmpr
-= num_bytes
;
1833 if (qg
->old_refcnt
< seq
)
1836 cur_old_count
= qg
->old_refcnt
- seq
;
1837 if (qg
->new_refcnt
< seq
)
1840 cur_new_count
= qg
->new_refcnt
- seq
;
1843 * If our refcount was the same as the roots previously but our
1844 * new count isn't the same as the number of roots now then we
1845 * went from having a exclusive reference on this range to not.
1847 if (old_roots
&& cur_old_count
== old_roots
&&
1848 (cur_new_count
!= new_roots
|| new_roots
== 0)) {
1849 WARN_ON(cur_new_count
!= new_roots
&& new_roots
== 0);
1850 qg
->excl
-= num_bytes
;
1851 qg
->excl_cmpr
-= num_bytes
;
1856 * If we didn't reference all the roots before but now we do we
1857 * have an exclusive reference to this range.
1859 if ((!old_roots
|| (old_roots
&& cur_old_count
!= old_roots
))
1860 && cur_new_count
== new_roots
) {
1861 qg
->excl
+= num_bytes
;
1862 qg
->excl_cmpr
+= num_bytes
;
1867 qgroup_dirty(fs_info
, qg
);
1873 * If we removed a data extent and there were other references for that bytenr
1874 * then we need to lookup all referenced roots to make sure we still don't
1875 * reference this bytenr. If we do then we can just discard this operation.
1877 static int check_existing_refs(struct btrfs_trans_handle
*trans
,
1878 struct btrfs_fs_info
*fs_info
,
1879 struct btrfs_qgroup_operation
*oper
)
1881 struct ulist
*roots
= NULL
;
1882 struct ulist_node
*unode
;
1883 struct ulist_iterator uiter
;
1886 ret
= btrfs_find_all_roots(trans
, fs_info
, oper
->bytenr
,
1887 oper
->elem
.seq
, &roots
);
1892 ULIST_ITER_INIT(&uiter
);
1893 while ((unode
= ulist_next(roots
, &uiter
))) {
1894 if (unode
->val
== oper
->ref_root
) {
1900 btrfs_put_tree_mod_seq(fs_info
, &oper
->elem
);
1906 * If we share a reference across multiple roots then we may need to adjust
1907 * various qgroups referenced and exclusive counters. The basic premise is this
1909 * 1) We have seq to represent a 0 count. Instead of looping through all of the
1910 * qgroups and resetting their refcount to 0 we just constantly bump this
1911 * sequence number to act as the base reference count. This means that if
1912 * anybody is equal to or below this sequence they were never referenced. We
1913 * jack this sequence up by the number of roots we found each time in order to
1914 * make sure we don't have any overlap.
1916 * 2) We first search all the roots that reference the area _except_ the root
1917 * we're acting on currently. This makes up the old_refcnt of all the qgroups
1920 * 3) We walk all of the qgroups referenced by the root we are currently acting
1921 * on, and will either adjust old_refcnt in the case of a removal or the
1922 * new_refcnt in the case of an addition.
1924 * 4) Finally we walk all the qgroups that are referenced by this range
1925 * including the root we are acting on currently. We will adjust the counters
1926 * based on the number of roots we had and will have after this operation.
1928 * Take this example as an illustration
1932 * [qg 0/0] [qg 0/1] [qg 0/2]
1936 * Say we are adding a reference that is covered by qg 0/0. The first step
1937 * would give a refcnt of 1 to qg 0/1 and 0/2 and a refcnt of 2 to qg 1/0 with
1938 * old_roots being 2. Because it is adding new_roots will be 1. We then go
1939 * through qg 0/0 which will get the new_refcnt set to 1 and add 1 to qg 1/0's
1940 * new_refcnt, bringing it to 3. We then walk through all of the qgroups, we
1941 * notice that the old refcnt for qg 0/0 < the new refcnt, so we added a
1942 * reference and thus must add the size to the referenced bytes. Everything
1943 * else is the same so nothing else changes.
1945 static int qgroup_shared_accounting(struct btrfs_trans_handle
*trans
,
1946 struct btrfs_fs_info
*fs_info
,
1947 struct btrfs_qgroup_operation
*oper
)
1949 struct ulist
*roots
= NULL
;
1950 struct ulist
*qgroups
, *tmp
;
1951 struct btrfs_qgroup
*qgroup
;
1952 struct seq_list elem
= SEQ_LIST_INIT(elem
);
1958 if (oper
->elem
.seq
) {
1959 ret
= check_existing_refs(trans
, fs_info
, oper
);
1966 qgroups
= ulist_alloc(GFP_NOFS
);
1970 tmp
= ulist_alloc(GFP_NOFS
);
1972 ulist_free(qgroups
);
1976 btrfs_get_tree_mod_seq(fs_info
, &elem
);
1977 ret
= btrfs_find_all_roots(trans
, fs_info
, oper
->bytenr
, elem
.seq
,
1979 btrfs_put_tree_mod_seq(fs_info
, &elem
);
1981 ulist_free(qgroups
);
1985 spin_lock(&fs_info
->qgroup_lock
);
1986 qgroup
= find_qgroup_rb(fs_info
, oper
->ref_root
);
1989 seq
= fs_info
->qgroup_seq
;
1992 * So roots is the list of all the roots currently pointing at the
1993 * bytenr, including the ref we are adding if we are adding, or not if
1994 * we are removing a ref. So we pass in the ref_root to skip that root
1995 * in our calculations. We set old_refnct and new_refcnt cause who the
1996 * hell knows what everything looked like before, and it doesn't matter
1999 ret
= qgroup_calc_old_refcnt(fs_info
, oper
->ref_root
, tmp
, roots
, qgroups
,
2000 seq
, &old_roots
, 0);
2005 * Now adjust the refcounts of the qgroups that care about this
2006 * reference, either the old_count in the case of removal or new_count
2007 * in the case of an addition.
2009 ret
= qgroup_calc_new_refcnt(fs_info
, oper
, qgroup
, tmp
, qgroups
,
2015 * ...in the case of removals. If we had a removal before we got around
2016 * to processing this operation then we need to find that guy and count
2017 * his references as if they really existed so we don't end up screwing
2018 * up the exclusive counts. Then whenever we go to process the delete
2019 * everything will be grand and we can account for whatever exclusive
2020 * changes need to be made there. We also have to pass in old_roots so
2021 * we have an accurate count of the roots as it pertains to this
2022 * operations view of the world.
2024 ret
= qgroup_account_deleted_refs(fs_info
, oper
, tmp
, qgroups
, seq
,
2030 * We are adding our root, need to adjust up the number of roots,
2031 * otherwise old_roots is the number of roots we want.
2033 if (oper
->type
== BTRFS_QGROUP_OPER_ADD_SHARED
) {
2034 new_roots
= old_roots
+ 1;
2036 new_roots
= old_roots
;
2039 fs_info
->qgroup_seq
+= old_roots
+ 1;
2043 * And now the magic happens, bless Arne for having a pretty elegant
2044 * solution for this.
2046 qgroup_adjust_counters(fs_info
, oper
->ref_root
, oper
->num_bytes
,
2047 qgroups
, seq
, old_roots
, new_roots
, 0);
2049 spin_unlock(&fs_info
->qgroup_lock
);
2050 ulist_free(qgroups
);
2057 * Process a reference to a shared subtree. This type of operation is
2058 * queued during snapshot removal when we encounter extents which are
2059 * shared between more than one root.
2061 static int qgroup_subtree_accounting(struct btrfs_trans_handle
*trans
,
2062 struct btrfs_fs_info
*fs_info
,
2063 struct btrfs_qgroup_operation
*oper
)
2065 struct ulist
*roots
= NULL
;
2066 struct ulist_node
*unode
;
2067 struct ulist_iterator uiter
;
2068 struct btrfs_qgroup_list
*glist
;
2069 struct ulist
*parents
;
2072 struct btrfs_qgroup
*qg
;
2074 struct seq_list elem
= SEQ_LIST_INIT(elem
);
2076 parents
= ulist_alloc(GFP_NOFS
);
2080 btrfs_get_tree_mod_seq(fs_info
, &elem
);
2081 ret
= btrfs_find_all_roots(trans
, fs_info
, oper
->bytenr
,
2083 btrfs_put_tree_mod_seq(fs_info
, &elem
);
2087 if (roots
->nnodes
!= 1)
2090 ULIST_ITER_INIT(&uiter
);
2091 unode
= ulist_next(roots
, &uiter
); /* Only want 1 so no need to loop */
2093 * If we find our ref root then that means all refs
2094 * this extent has to the root have not yet been
2095 * deleted. In that case, we do nothing and let the
2096 * last ref for this bytenr drive our update.
2098 * This can happen for example if an extent is
2099 * referenced multiple times in a snapshot (clone,
2100 * etc). If we are in the middle of snapshot removal,
2101 * queued updates for such an extent will find the
2102 * root if we have not yet finished removing the
2105 if (unode
->val
== oper
->ref_root
)
2108 root_obj
= unode
->val
;
2111 spin_lock(&fs_info
->qgroup_lock
);
2112 qg
= find_qgroup_rb(fs_info
, root_obj
);
2116 qg
->excl
+= oper
->num_bytes
;
2117 qg
->excl_cmpr
+= oper
->num_bytes
;
2118 qgroup_dirty(fs_info
, qg
);
2121 * Adjust counts for parent groups. First we find all
2122 * parents, then in the 2nd loop we do the adjustment
2123 * while adding parents of the parents to our ulist.
2125 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2126 err
= ulist_add(parents
, glist
->group
->qgroupid
,
2127 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
2134 ULIST_ITER_INIT(&uiter
);
2135 while ((unode
= ulist_next(parents
, &uiter
))) {
2136 qg
= u64_to_ptr(unode
->aux
);
2137 qg
->excl
+= oper
->num_bytes
;
2138 qg
->excl_cmpr
+= oper
->num_bytes
;
2139 qgroup_dirty(fs_info
, qg
);
2141 /* Add any parents of the parents */
2142 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2143 err
= ulist_add(parents
, glist
->group
->qgroupid
,
2144 ptr_to_u64(glist
->group
), GFP_ATOMIC
);
2153 spin_unlock(&fs_info
->qgroup_lock
);
2157 ulist_free(parents
);
2162 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
2163 * from the fs. First, all roots referencing the extent are searched, and
2164 * then the space is accounted accordingly to the different roots. The
2165 * accounting algorithm works in 3 steps documented inline.
2167 static int btrfs_qgroup_account(struct btrfs_trans_handle
*trans
,
2168 struct btrfs_fs_info
*fs_info
,
2169 struct btrfs_qgroup_operation
*oper
)
2173 if (!fs_info
->quota_enabled
)
2176 BUG_ON(!fs_info
->quota_root
);
2178 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2179 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) {
2180 if (fs_info
->qgroup_rescan_progress
.objectid
<= oper
->bytenr
) {
2181 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2185 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2187 ASSERT(is_fstree(oper
->ref_root
));
2189 trace_btrfs_qgroup_account(oper
);
2191 switch (oper
->type
) {
2192 case BTRFS_QGROUP_OPER_ADD_EXCL
:
2193 case BTRFS_QGROUP_OPER_SUB_EXCL
:
2194 ret
= qgroup_excl_accounting(fs_info
, oper
);
2196 case BTRFS_QGROUP_OPER_ADD_SHARED
:
2197 case BTRFS_QGROUP_OPER_SUB_SHARED
:
2198 ret
= qgroup_shared_accounting(trans
, fs_info
, oper
);
2200 case BTRFS_QGROUP_OPER_SUB_SUBTREE
:
2201 ret
= qgroup_subtree_accounting(trans
, fs_info
, oper
);
2210 * Needs to be called everytime we run delayed refs, even if there is an error
2211 * in order to cleanup outstanding operations.
2213 int btrfs_delayed_qgroup_accounting(struct btrfs_trans_handle
*trans
,
2214 struct btrfs_fs_info
*fs_info
)
2216 struct btrfs_qgroup_operation
*oper
;
2219 while (!list_empty(&trans
->qgroup_ref_list
)) {
2220 oper
= list_first_entry(&trans
->qgroup_ref_list
,
2221 struct btrfs_qgroup_operation
, list
);
2222 list_del_init(&oper
->list
);
2223 if (!ret
|| !trans
->aborted
)
2224 ret
= btrfs_qgroup_account(trans
, fs_info
, oper
);
2225 spin_lock(&fs_info
->qgroup_op_lock
);
2226 rb_erase(&oper
->n
, &fs_info
->qgroup_op_tree
);
2227 spin_unlock(&fs_info
->qgroup_op_lock
);
2228 btrfs_put_tree_mod_seq(fs_info
, &oper
->elem
);
2235 * called from commit_transaction. Writes all changed qgroups to disk.
2237 int btrfs_run_qgroups(struct btrfs_trans_handle
*trans
,
2238 struct btrfs_fs_info
*fs_info
)
2240 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
2242 int start_rescan_worker
= 0;
2247 if (!fs_info
->quota_enabled
&& fs_info
->pending_quota_state
)
2248 start_rescan_worker
= 1;
2250 fs_info
->quota_enabled
= fs_info
->pending_quota_state
;
2252 spin_lock(&fs_info
->qgroup_lock
);
2253 while (!list_empty(&fs_info
->dirty_qgroups
)) {
2254 struct btrfs_qgroup
*qgroup
;
2255 qgroup
= list_first_entry(&fs_info
->dirty_qgroups
,
2256 struct btrfs_qgroup
, dirty
);
2257 list_del_init(&qgroup
->dirty
);
2258 spin_unlock(&fs_info
->qgroup_lock
);
2259 ret
= update_qgroup_info_item(trans
, quota_root
, qgroup
);
2261 fs_info
->qgroup_flags
|=
2262 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2263 ret
= update_qgroup_limit_item(trans
, quota_root
, qgroup
);
2265 fs_info
->qgroup_flags
|=
2266 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2267 spin_lock(&fs_info
->qgroup_lock
);
2269 if (fs_info
->quota_enabled
)
2270 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_ON
;
2272 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
2273 spin_unlock(&fs_info
->qgroup_lock
);
2275 ret
= update_qgroup_status_item(trans
, fs_info
, quota_root
);
2277 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2279 if (!ret
&& start_rescan_worker
) {
2280 ret
= qgroup_rescan_init(fs_info
, 0, 1);
2282 qgroup_rescan_zero_tracking(fs_info
);
2283 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2284 &fs_info
->qgroup_rescan_work
);
2295 * copy the acounting information between qgroups. This is necessary when a
2296 * snapshot or a subvolume is created
2298 int btrfs_qgroup_inherit(struct btrfs_trans_handle
*trans
,
2299 struct btrfs_fs_info
*fs_info
, u64 srcid
, u64 objectid
,
2300 struct btrfs_qgroup_inherit
*inherit
)
2305 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
2306 struct btrfs_qgroup
*srcgroup
;
2307 struct btrfs_qgroup
*dstgroup
;
2311 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
2312 if (!fs_info
->quota_enabled
)
2321 i_qgroups
= (u64
*)(inherit
+ 1);
2322 nums
= inherit
->num_qgroups
+ 2 * inherit
->num_ref_copies
+
2323 2 * inherit
->num_excl_copies
;
2324 for (i
= 0; i
< nums
; ++i
) {
2325 srcgroup
= find_qgroup_rb(fs_info
, *i_qgroups
);
2331 if ((srcgroup
->qgroupid
>> 48) <= (objectid
>> 48)) {
2340 * create a tracking group for the subvol itself
2342 ret
= add_qgroup_item(trans
, quota_root
, objectid
);
2347 struct btrfs_root
*srcroot
;
2348 struct btrfs_key srckey
;
2350 srckey
.objectid
= srcid
;
2351 srckey
.type
= BTRFS_ROOT_ITEM_KEY
;
2352 srckey
.offset
= (u64
)-1;
2353 srcroot
= btrfs_read_fs_root_no_name(fs_info
, &srckey
);
2354 if (IS_ERR(srcroot
)) {
2355 ret
= PTR_ERR(srcroot
);
2360 level_size
= srcroot
->nodesize
;
2365 * add qgroup to all inherited groups
2368 i_qgroups
= (u64
*)(inherit
+ 1);
2369 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
2370 ret
= add_qgroup_relation_item(trans
, quota_root
,
2371 objectid
, *i_qgroups
);
2374 ret
= add_qgroup_relation_item(trans
, quota_root
,
2375 *i_qgroups
, objectid
);
2383 spin_lock(&fs_info
->qgroup_lock
);
2385 dstgroup
= add_qgroup_rb(fs_info
, objectid
);
2386 if (IS_ERR(dstgroup
)) {
2387 ret
= PTR_ERR(dstgroup
);
2391 if (inherit
&& inherit
->flags
& BTRFS_QGROUP_INHERIT_SET_LIMITS
) {
2392 dstgroup
->lim_flags
= inherit
->lim
.flags
;
2393 dstgroup
->max_rfer
= inherit
->lim
.max_rfer
;
2394 dstgroup
->max_excl
= inherit
->lim
.max_excl
;
2395 dstgroup
->rsv_rfer
= inherit
->lim
.rsv_rfer
;
2396 dstgroup
->rsv_excl
= inherit
->lim
.rsv_excl
;
2398 ret
= update_qgroup_limit_item(trans
, quota_root
, dstgroup
);
2400 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2401 btrfs_info(fs_info
, "unable to update quota limit for %llu",
2402 dstgroup
->qgroupid
);
2408 srcgroup
= find_qgroup_rb(fs_info
, srcid
);
2413 * We call inherit after we clone the root in order to make sure
2414 * our counts don't go crazy, so at this point the only
2415 * difference between the two roots should be the root node.
2417 dstgroup
->rfer
= srcgroup
->rfer
;
2418 dstgroup
->rfer_cmpr
= srcgroup
->rfer_cmpr
;
2419 dstgroup
->excl
= level_size
;
2420 dstgroup
->excl_cmpr
= level_size
;
2421 srcgroup
->excl
= level_size
;
2422 srcgroup
->excl_cmpr
= level_size
;
2424 /* inherit the limit info */
2425 dstgroup
->lim_flags
= srcgroup
->lim_flags
;
2426 dstgroup
->max_rfer
= srcgroup
->max_rfer
;
2427 dstgroup
->max_excl
= srcgroup
->max_excl
;
2428 dstgroup
->rsv_rfer
= srcgroup
->rsv_rfer
;
2429 dstgroup
->rsv_excl
= srcgroup
->rsv_excl
;
2431 qgroup_dirty(fs_info
, dstgroup
);
2432 qgroup_dirty(fs_info
, srcgroup
);
2438 i_qgroups
= (u64
*)(inherit
+ 1);
2439 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
2440 ret
= add_relation_rb(quota_root
->fs_info
, objectid
,
2447 for (i
= 0; i
< inherit
->num_ref_copies
; ++i
) {
2448 struct btrfs_qgroup
*src
;
2449 struct btrfs_qgroup
*dst
;
2451 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
2452 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
2459 dst
->rfer
= src
->rfer
- level_size
;
2460 dst
->rfer_cmpr
= src
->rfer_cmpr
- level_size
;
2463 for (i
= 0; i
< inherit
->num_excl_copies
; ++i
) {
2464 struct btrfs_qgroup
*src
;
2465 struct btrfs_qgroup
*dst
;
2467 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
2468 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
2475 dst
->excl
= src
->excl
+ level_size
;
2476 dst
->excl_cmpr
= src
->excl_cmpr
+ level_size
;
2481 spin_unlock(&fs_info
->qgroup_lock
);
2483 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
2487 int btrfs_qgroup_reserve(struct btrfs_root
*root
, u64 num_bytes
)
2489 struct btrfs_root
*quota_root
;
2490 struct btrfs_qgroup
*qgroup
;
2491 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2492 u64 ref_root
= root
->root_key
.objectid
;
2494 struct ulist_node
*unode
;
2495 struct ulist_iterator uiter
;
2497 if (!is_fstree(ref_root
))
2503 spin_lock(&fs_info
->qgroup_lock
);
2504 quota_root
= fs_info
->quota_root
;
2508 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
2513 * in a first step, we check all affected qgroups if any limits would
2516 ulist_reinit(fs_info
->qgroup_ulist
);
2517 ret
= ulist_add(fs_info
->qgroup_ulist
, qgroup
->qgroupid
,
2518 (uintptr_t)qgroup
, GFP_ATOMIC
);
2521 ULIST_ITER_INIT(&uiter
);
2522 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2523 struct btrfs_qgroup
*qg
;
2524 struct btrfs_qgroup_list
*glist
;
2526 qg
= u64_to_ptr(unode
->aux
);
2528 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) &&
2529 qg
->reserved
+ (s64
)qg
->rfer
+ num_bytes
>
2535 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) &&
2536 qg
->reserved
+ (s64
)qg
->excl
+ num_bytes
>
2542 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2543 ret
= ulist_add(fs_info
->qgroup_ulist
,
2544 glist
->group
->qgroupid
,
2545 (uintptr_t)glist
->group
, GFP_ATOMIC
);
2552 * no limits exceeded, now record the reservation into all qgroups
2554 ULIST_ITER_INIT(&uiter
);
2555 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2556 struct btrfs_qgroup
*qg
;
2558 qg
= u64_to_ptr(unode
->aux
);
2560 qg
->reserved
+= num_bytes
;
2564 spin_unlock(&fs_info
->qgroup_lock
);
2568 void btrfs_qgroup_free(struct btrfs_root
*root
, u64 num_bytes
)
2570 struct btrfs_root
*quota_root
;
2571 struct btrfs_qgroup
*qgroup
;
2572 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2573 struct ulist_node
*unode
;
2574 struct ulist_iterator uiter
;
2575 u64 ref_root
= root
->root_key
.objectid
;
2578 if (!is_fstree(ref_root
))
2584 spin_lock(&fs_info
->qgroup_lock
);
2586 quota_root
= fs_info
->quota_root
;
2590 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
2594 ulist_reinit(fs_info
->qgroup_ulist
);
2595 ret
= ulist_add(fs_info
->qgroup_ulist
, qgroup
->qgroupid
,
2596 (uintptr_t)qgroup
, GFP_ATOMIC
);
2599 ULIST_ITER_INIT(&uiter
);
2600 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2601 struct btrfs_qgroup
*qg
;
2602 struct btrfs_qgroup_list
*glist
;
2604 qg
= u64_to_ptr(unode
->aux
);
2606 qg
->reserved
-= num_bytes
;
2608 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2609 ret
= ulist_add(fs_info
->qgroup_ulist
,
2610 glist
->group
->qgroupid
,
2611 (uintptr_t)glist
->group
, GFP_ATOMIC
);
2618 spin_unlock(&fs_info
->qgroup_lock
);
2621 void assert_qgroups_uptodate(struct btrfs_trans_handle
*trans
)
2623 if (list_empty(&trans
->qgroup_ref_list
) && !trans
->delayed_ref_elem
.seq
)
2625 btrfs_err(trans
->root
->fs_info
,
2626 "qgroups not uptodate in trans handle %p: list is%s empty, "
2628 trans
, list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
2629 (u32
)(trans
->delayed_ref_elem
.seq
>> 32),
2630 (u32
)trans
->delayed_ref_elem
.seq
);
2635 * returns < 0 on error, 0 when more leafs are to be scanned.
2636 * returns 1 when done.
2639 qgroup_rescan_leaf(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
2640 struct btrfs_trans_handle
*trans
, struct ulist
*qgroups
,
2641 struct ulist
*tmp
, struct extent_buffer
*scratch_leaf
)
2643 struct btrfs_key found
;
2644 struct ulist
*roots
= NULL
;
2645 struct seq_list tree_mod_seq_elem
= SEQ_LIST_INIT(tree_mod_seq_elem
);
2652 path
->leave_spinning
= 1;
2653 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2654 ret
= btrfs_search_slot_for_read(fs_info
->extent_root
,
2655 &fs_info
->qgroup_rescan_progress
,
2658 pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2659 fs_info
->qgroup_rescan_progress
.objectid
,
2660 fs_info
->qgroup_rescan_progress
.type
,
2661 fs_info
->qgroup_rescan_progress
.offset
, ret
);
2665 * The rescan is about to end, we will not be scanning any
2666 * further blocks. We cannot unset the RESCAN flag here, because
2667 * we want to commit the transaction if everything went well.
2668 * To make the live accounting work in this phase, we set our
2669 * scan progress pointer such that every real extent objectid
2672 fs_info
->qgroup_rescan_progress
.objectid
= (u64
)-1;
2673 btrfs_release_path(path
);
2674 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2678 btrfs_item_key_to_cpu(path
->nodes
[0], &found
,
2679 btrfs_header_nritems(path
->nodes
[0]) - 1);
2680 fs_info
->qgroup_rescan_progress
.objectid
= found
.objectid
+ 1;
2682 btrfs_get_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2683 memcpy(scratch_leaf
, path
->nodes
[0], sizeof(*scratch_leaf
));
2684 slot
= path
->slots
[0];
2685 btrfs_release_path(path
);
2686 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2688 for (; slot
< btrfs_header_nritems(scratch_leaf
); ++slot
) {
2689 btrfs_item_key_to_cpu(scratch_leaf
, &found
, slot
);
2690 if (found
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
2691 found
.type
!= BTRFS_METADATA_ITEM_KEY
)
2693 if (found
.type
== BTRFS_METADATA_ITEM_KEY
)
2694 num_bytes
= fs_info
->extent_root
->nodesize
;
2696 num_bytes
= found
.offset
;
2698 ulist_reinit(qgroups
);
2699 ret
= btrfs_find_all_roots(NULL
, fs_info
, found
.objectid
, 0,
2703 spin_lock(&fs_info
->qgroup_lock
);
2704 seq
= fs_info
->qgroup_seq
;
2705 fs_info
->qgroup_seq
+= roots
->nnodes
+ 1; /* max refcnt */
2708 ret
= qgroup_calc_old_refcnt(fs_info
, 0, tmp
, roots
, qgroups
,
2709 seq
, &new_roots
, 1);
2711 spin_unlock(&fs_info
->qgroup_lock
);
2716 ret
= qgroup_adjust_counters(fs_info
, 0, num_bytes
, qgroups
,
2717 seq
, 0, new_roots
, 1);
2719 spin_unlock(&fs_info
->qgroup_lock
);
2723 spin_unlock(&fs_info
->qgroup_lock
);
2727 btrfs_put_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2732 static void btrfs_qgroup_rescan_worker(struct btrfs_work
*work
)
2734 struct btrfs_fs_info
*fs_info
= container_of(work
, struct btrfs_fs_info
,
2735 qgroup_rescan_work
);
2736 struct btrfs_path
*path
;
2737 struct btrfs_trans_handle
*trans
= NULL
;
2738 struct ulist
*tmp
= NULL
, *qgroups
= NULL
;
2739 struct extent_buffer
*scratch_leaf
= NULL
;
2743 path
= btrfs_alloc_path();
2746 qgroups
= ulist_alloc(GFP_NOFS
);
2749 tmp
= ulist_alloc(GFP_NOFS
);
2752 scratch_leaf
= kmalloc(sizeof(*scratch_leaf
), GFP_NOFS
);
2758 trans
= btrfs_start_transaction(fs_info
->fs_root
, 0);
2759 if (IS_ERR(trans
)) {
2760 err
= PTR_ERR(trans
);
2763 if (!fs_info
->quota_enabled
) {
2766 err
= qgroup_rescan_leaf(fs_info
, path
, trans
,
2767 qgroups
, tmp
, scratch_leaf
);
2770 btrfs_commit_transaction(trans
, fs_info
->fs_root
);
2772 btrfs_end_transaction(trans
, fs_info
->fs_root
);
2776 kfree(scratch_leaf
);
2777 ulist_free(qgroups
);
2779 btrfs_free_path(path
);
2781 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2782 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2785 fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
) {
2786 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2787 } else if (err
< 0) {
2788 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2790 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2793 * only update status, since the previous part has alreay updated the
2796 trans
= btrfs_start_transaction(fs_info
->quota_root
, 1);
2797 if (IS_ERR(trans
)) {
2798 err
= PTR_ERR(trans
);
2800 "fail to start transaction for status update: %d\n",
2804 ret
= update_qgroup_status_item(trans
, fs_info
, fs_info
->quota_root
);
2807 btrfs_err(fs_info
, "fail to update qgroup status: %d\n", err
);
2809 btrfs_end_transaction(trans
, fs_info
->quota_root
);
2812 btrfs_info(fs_info
, "qgroup scan completed%s",
2813 err
> 0 ? " (inconsistency flag cleared)" : "");
2815 btrfs_err(fs_info
, "qgroup scan failed with %d", err
);
2819 complete_all(&fs_info
->qgroup_rescan_completion
);
2823 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2824 * memory required for the rescan context.
2827 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
2833 (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) ||
2834 !(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))) {
2839 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2840 spin_lock(&fs_info
->qgroup_lock
);
2843 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
2845 else if (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))
2849 spin_unlock(&fs_info
->qgroup_lock
);
2850 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2853 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2856 memset(&fs_info
->qgroup_rescan_progress
, 0,
2857 sizeof(fs_info
->qgroup_rescan_progress
));
2858 fs_info
->qgroup_rescan_progress
.objectid
= progress_objectid
;
2860 spin_unlock(&fs_info
->qgroup_lock
);
2861 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2863 init_completion(&fs_info
->qgroup_rescan_completion
);
2865 memset(&fs_info
->qgroup_rescan_work
, 0,
2866 sizeof(fs_info
->qgroup_rescan_work
));
2867 btrfs_init_work(&fs_info
->qgroup_rescan_work
,
2868 btrfs_qgroup_rescan_helper
,
2869 btrfs_qgroup_rescan_worker
, NULL
, NULL
);
2873 btrfs_info(fs_info
, "qgroup_rescan_init failed with %d", ret
);
2881 qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
)
2884 struct btrfs_qgroup
*qgroup
;
2886 spin_lock(&fs_info
->qgroup_lock
);
2887 /* clear all current qgroup tracking information */
2888 for (n
= rb_first(&fs_info
->qgroup_tree
); n
; n
= rb_next(n
)) {
2889 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
2891 qgroup
->rfer_cmpr
= 0;
2893 qgroup
->excl_cmpr
= 0;
2895 spin_unlock(&fs_info
->qgroup_lock
);
2899 btrfs_qgroup_rescan(struct btrfs_fs_info
*fs_info
)
2902 struct btrfs_trans_handle
*trans
;
2904 ret
= qgroup_rescan_init(fs_info
, 0, 1);
2909 * We have set the rescan_progress to 0, which means no more
2910 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2911 * However, btrfs_qgroup_account_ref may be right after its call
2912 * to btrfs_find_all_roots, in which case it would still do the
2914 * To solve this, we're committing the transaction, which will
2915 * ensure we run all delayed refs and only after that, we are
2916 * going to clear all tracking information for a clean start.
2919 trans
= btrfs_join_transaction(fs_info
->fs_root
);
2920 if (IS_ERR(trans
)) {
2921 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2922 return PTR_ERR(trans
);
2924 ret
= btrfs_commit_transaction(trans
, fs_info
->fs_root
);
2926 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2930 qgroup_rescan_zero_tracking(fs_info
);
2932 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2933 &fs_info
->qgroup_rescan_work
);
2938 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info
*fs_info
)
2943 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2944 spin_lock(&fs_info
->qgroup_lock
);
2945 running
= fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2946 spin_unlock(&fs_info
->qgroup_lock
);
2947 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2950 ret
= wait_for_completion_interruptible(
2951 &fs_info
->qgroup_rescan_completion
);
2957 * this is only called from open_ctree where we're still single threaded, thus
2958 * locking is omitted here.
2961 btrfs_qgroup_rescan_resume(struct btrfs_fs_info
*fs_info
)
2963 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
2964 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2965 &fs_info
->qgroup_rescan_work
);