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"
36 * - subvol delete -> delete when ref goes to 0? delete limits also?
41 * - copy also limits on subvol creation
43 * - caches fuer ulists
44 * - performance benchmarks
45 * - check all ioctl parameters
49 * one struct for each qgroup, organized in fs_info->qgroup_tree.
57 u64 rfer
; /* referenced */
58 u64 rfer_cmpr
; /* referenced compressed */
59 u64 excl
; /* exclusive */
60 u64 excl_cmpr
; /* exclusive compressed */
65 u64 lim_flags
; /* which limits are set */
72 * reservation tracking
79 struct list_head groups
; /* groups this group is member of */
80 struct list_head members
; /* groups that are members of this group */
81 struct list_head dirty
; /* dirty groups */
82 struct rb_node node
; /* tree of qgroups */
85 * temp variables for accounting operations
92 * glue structure to represent the relations between qgroups.
94 struct btrfs_qgroup_list
{
95 struct list_head next_group
;
96 struct list_head next_member
;
97 struct btrfs_qgroup
*group
;
98 struct btrfs_qgroup
*member
;
101 /* must be called with qgroup_lock held */
102 static struct btrfs_qgroup
*find_qgroup_rb(struct btrfs_fs_info
*fs_info
,
105 struct rb_node
*n
= fs_info
->qgroup_tree
.rb_node
;
106 struct btrfs_qgroup
*qgroup
;
109 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
110 if (qgroup
->qgroupid
< qgroupid
)
112 else if (qgroup
->qgroupid
> qgroupid
)
120 /* must be called with qgroup_lock held */
121 static struct btrfs_qgroup
*add_qgroup_rb(struct btrfs_fs_info
*fs_info
,
124 struct rb_node
**p
= &fs_info
->qgroup_tree
.rb_node
;
125 struct rb_node
*parent
= NULL
;
126 struct btrfs_qgroup
*qgroup
;
130 qgroup
= rb_entry(parent
, struct btrfs_qgroup
, node
);
132 if (qgroup
->qgroupid
< qgroupid
)
134 else if (qgroup
->qgroupid
> qgroupid
)
140 qgroup
= kzalloc(sizeof(*qgroup
), GFP_ATOMIC
);
142 return ERR_PTR(-ENOMEM
);
144 qgroup
->qgroupid
= qgroupid
;
145 INIT_LIST_HEAD(&qgroup
->groups
);
146 INIT_LIST_HEAD(&qgroup
->members
);
147 INIT_LIST_HEAD(&qgroup
->dirty
);
149 rb_link_node(&qgroup
->node
, parent
, p
);
150 rb_insert_color(&qgroup
->node
, &fs_info
->qgroup_tree
);
155 /* must be called with qgroup_lock held */
156 static int del_qgroup_rb(struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
158 struct btrfs_qgroup
*qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
159 struct btrfs_qgroup_list
*list
;
164 rb_erase(&qgroup
->node
, &fs_info
->qgroup_tree
);
165 list_del(&qgroup
->dirty
);
167 while (!list_empty(&qgroup
->groups
)) {
168 list
= list_first_entry(&qgroup
->groups
,
169 struct btrfs_qgroup_list
, next_group
);
170 list_del(&list
->next_group
);
171 list_del(&list
->next_member
);
175 while (!list_empty(&qgroup
->members
)) {
176 list
= list_first_entry(&qgroup
->members
,
177 struct btrfs_qgroup_list
, next_member
);
178 list_del(&list
->next_group
);
179 list_del(&list
->next_member
);
187 /* must be called with qgroup_lock held */
188 static int add_relation_rb(struct btrfs_fs_info
*fs_info
,
189 u64 memberid
, u64 parentid
)
191 struct btrfs_qgroup
*member
;
192 struct btrfs_qgroup
*parent
;
193 struct btrfs_qgroup_list
*list
;
195 member
= find_qgroup_rb(fs_info
, memberid
);
196 parent
= find_qgroup_rb(fs_info
, parentid
);
197 if (!member
|| !parent
)
200 list
= kzalloc(sizeof(*list
), GFP_ATOMIC
);
204 list
->group
= parent
;
205 list
->member
= member
;
206 list_add_tail(&list
->next_group
, &member
->groups
);
207 list_add_tail(&list
->next_member
, &parent
->members
);
212 /* must be called with qgroup_lock held */
213 static int del_relation_rb(struct btrfs_fs_info
*fs_info
,
214 u64 memberid
, u64 parentid
)
216 struct btrfs_qgroup
*member
;
217 struct btrfs_qgroup
*parent
;
218 struct btrfs_qgroup_list
*list
;
220 member
= find_qgroup_rb(fs_info
, memberid
);
221 parent
= find_qgroup_rb(fs_info
, parentid
);
222 if (!member
|| !parent
)
225 list_for_each_entry(list
, &member
->groups
, next_group
) {
226 if (list
->group
== parent
) {
227 list_del(&list
->next_group
);
228 list_del(&list
->next_member
);
237 * The full config is read in one go, only called from open_ctree()
238 * It doesn't use any locking, as at this point we're still single-threaded
240 int btrfs_read_qgroup_config(struct btrfs_fs_info
*fs_info
)
242 struct btrfs_key key
;
243 struct btrfs_key found_key
;
244 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
245 struct btrfs_path
*path
= NULL
;
246 struct extent_buffer
*l
;
251 if (!fs_info
->quota_enabled
)
254 path
= btrfs_alloc_path();
260 /* default this to quota off, in case no status key is found */
261 fs_info
->qgroup_flags
= 0;
264 * pass 1: read status, all qgroup infos and limits
269 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 1);
274 struct btrfs_qgroup
*qgroup
;
276 slot
= path
->slots
[0];
278 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
280 if (found_key
.type
== BTRFS_QGROUP_STATUS_KEY
) {
281 struct btrfs_qgroup_status_item
*ptr
;
283 ptr
= btrfs_item_ptr(l
, slot
,
284 struct btrfs_qgroup_status_item
);
286 if (btrfs_qgroup_status_version(l
, ptr
) !=
287 BTRFS_QGROUP_STATUS_VERSION
) {
289 "btrfs: old qgroup version, quota disabled\n");
292 if (btrfs_qgroup_status_generation(l
, ptr
) !=
293 fs_info
->generation
) {
294 flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
296 "btrfs: qgroup generation mismatch, "
297 "marked as inconsistent\n");
299 fs_info
->qgroup_flags
= btrfs_qgroup_status_flags(l
,
301 /* FIXME read scan element */
305 if (found_key
.type
!= BTRFS_QGROUP_INFO_KEY
&&
306 found_key
.type
!= BTRFS_QGROUP_LIMIT_KEY
)
309 qgroup
= find_qgroup_rb(fs_info
, found_key
.offset
);
310 if ((qgroup
&& found_key
.type
== BTRFS_QGROUP_INFO_KEY
) ||
311 (!qgroup
&& found_key
.type
== BTRFS_QGROUP_LIMIT_KEY
)) {
312 printk(KERN_ERR
"btrfs: inconsitent qgroup config\n");
313 flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
316 qgroup
= add_qgroup_rb(fs_info
, found_key
.offset
);
317 if (IS_ERR(qgroup
)) {
318 ret
= PTR_ERR(qgroup
);
322 switch (found_key
.type
) {
323 case BTRFS_QGROUP_INFO_KEY
: {
324 struct btrfs_qgroup_info_item
*ptr
;
326 ptr
= btrfs_item_ptr(l
, slot
,
327 struct btrfs_qgroup_info_item
);
328 qgroup
->rfer
= btrfs_qgroup_info_rfer(l
, ptr
);
329 qgroup
->rfer_cmpr
= btrfs_qgroup_info_rfer_cmpr(l
, ptr
);
330 qgroup
->excl
= btrfs_qgroup_info_excl(l
, ptr
);
331 qgroup
->excl_cmpr
= btrfs_qgroup_info_excl_cmpr(l
, ptr
);
332 /* generation currently unused */
335 case BTRFS_QGROUP_LIMIT_KEY
: {
336 struct btrfs_qgroup_limit_item
*ptr
;
338 ptr
= btrfs_item_ptr(l
, slot
,
339 struct btrfs_qgroup_limit_item
);
340 qgroup
->lim_flags
= btrfs_qgroup_limit_flags(l
, ptr
);
341 qgroup
->max_rfer
= btrfs_qgroup_limit_max_rfer(l
, ptr
);
342 qgroup
->max_excl
= btrfs_qgroup_limit_max_excl(l
, ptr
);
343 qgroup
->rsv_rfer
= btrfs_qgroup_limit_rsv_rfer(l
, ptr
);
344 qgroup
->rsv_excl
= btrfs_qgroup_limit_rsv_excl(l
, ptr
);
349 ret
= btrfs_next_item(quota_root
, path
);
355 btrfs_release_path(path
);
358 * pass 2: read all qgroup relations
361 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
363 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 0);
367 slot
= path
->slots
[0];
369 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
371 if (found_key
.type
!= BTRFS_QGROUP_RELATION_KEY
)
374 if (found_key
.objectid
> found_key
.offset
) {
375 /* parent <- member, not needed to build config */
376 /* FIXME should we omit the key completely? */
380 ret
= add_relation_rb(fs_info
, found_key
.objectid
,
382 if (ret
== -ENOENT
) {
384 "btrfs: orphan qgroup relation 0x%llx->0x%llx\n",
385 (unsigned long long)found_key
.objectid
,
386 (unsigned long long)found_key
.offset
);
387 ret
= 0; /* ignore the error */
392 ret
= btrfs_next_item(quota_root
, path
);
399 fs_info
->qgroup_flags
|= flags
;
400 if (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
)) {
401 fs_info
->quota_enabled
= 0;
402 fs_info
->pending_quota_state
= 0;
404 btrfs_free_path(path
);
406 return ret
< 0 ? ret
: 0;
410 * This is only called from close_ctree() or open_ctree(), both in single-
411 * treaded paths. Clean up the in-memory structures. No locking needed.
413 void btrfs_free_qgroup_config(struct btrfs_fs_info
*fs_info
)
416 struct btrfs_qgroup
*qgroup
;
417 struct btrfs_qgroup_list
*list
;
419 while ((n
= rb_first(&fs_info
->qgroup_tree
))) {
420 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
421 rb_erase(n
, &fs_info
->qgroup_tree
);
423 WARN_ON(!list_empty(&qgroup
->dirty
));
425 while (!list_empty(&qgroup
->groups
)) {
426 list
= list_first_entry(&qgroup
->groups
,
427 struct btrfs_qgroup_list
,
429 list_del(&list
->next_group
);
430 list_del(&list
->next_member
);
434 while (!list_empty(&qgroup
->members
)) {
435 list
= list_first_entry(&qgroup
->members
,
436 struct btrfs_qgroup_list
,
438 list_del(&list
->next_group
);
439 list_del(&list
->next_member
);
446 static int add_qgroup_relation_item(struct btrfs_trans_handle
*trans
,
447 struct btrfs_root
*quota_root
,
451 struct btrfs_path
*path
;
452 struct btrfs_key key
;
454 path
= btrfs_alloc_path();
459 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
462 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
, 0);
464 btrfs_mark_buffer_dirty(path
->nodes
[0]);
466 btrfs_free_path(path
);
470 static int del_qgroup_relation_item(struct btrfs_trans_handle
*trans
,
471 struct btrfs_root
*quota_root
,
475 struct btrfs_path
*path
;
476 struct btrfs_key key
;
478 path
= btrfs_alloc_path();
483 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
486 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
495 ret
= btrfs_del_item(trans
, quota_root
, path
);
497 btrfs_free_path(path
);
501 static int add_qgroup_item(struct btrfs_trans_handle
*trans
,
502 struct btrfs_root
*quota_root
, u64 qgroupid
)
505 struct btrfs_path
*path
;
506 struct btrfs_qgroup_info_item
*qgroup_info
;
507 struct btrfs_qgroup_limit_item
*qgroup_limit
;
508 struct extent_buffer
*leaf
;
509 struct btrfs_key key
;
511 path
= btrfs_alloc_path();
516 key
.type
= BTRFS_QGROUP_INFO_KEY
;
517 key
.offset
= qgroupid
;
519 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
520 sizeof(*qgroup_info
));
524 leaf
= path
->nodes
[0];
525 qgroup_info
= btrfs_item_ptr(leaf
, path
->slots
[0],
526 struct btrfs_qgroup_info_item
);
527 btrfs_set_qgroup_info_generation(leaf
, qgroup_info
, trans
->transid
);
528 btrfs_set_qgroup_info_rfer(leaf
, qgroup_info
, 0);
529 btrfs_set_qgroup_info_rfer_cmpr(leaf
, qgroup_info
, 0);
530 btrfs_set_qgroup_info_excl(leaf
, qgroup_info
, 0);
531 btrfs_set_qgroup_info_excl_cmpr(leaf
, qgroup_info
, 0);
533 btrfs_mark_buffer_dirty(leaf
);
535 btrfs_release_path(path
);
537 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
538 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
539 sizeof(*qgroup_limit
));
543 leaf
= path
->nodes
[0];
544 qgroup_limit
= btrfs_item_ptr(leaf
, path
->slots
[0],
545 struct btrfs_qgroup_limit_item
);
546 btrfs_set_qgroup_limit_flags(leaf
, qgroup_limit
, 0);
547 btrfs_set_qgroup_limit_max_rfer(leaf
, qgroup_limit
, 0);
548 btrfs_set_qgroup_limit_max_excl(leaf
, qgroup_limit
, 0);
549 btrfs_set_qgroup_limit_rsv_rfer(leaf
, qgroup_limit
, 0);
550 btrfs_set_qgroup_limit_rsv_excl(leaf
, qgroup_limit
, 0);
552 btrfs_mark_buffer_dirty(leaf
);
556 btrfs_free_path(path
);
560 static int del_qgroup_item(struct btrfs_trans_handle
*trans
,
561 struct btrfs_root
*quota_root
, u64 qgroupid
)
564 struct btrfs_path
*path
;
565 struct btrfs_key key
;
567 path
= btrfs_alloc_path();
572 key
.type
= BTRFS_QGROUP_INFO_KEY
;
573 key
.offset
= qgroupid
;
574 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
583 ret
= btrfs_del_item(trans
, quota_root
, path
);
587 btrfs_release_path(path
);
589 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
590 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
599 ret
= btrfs_del_item(trans
, quota_root
, path
);
602 btrfs_free_path(path
);
606 static int update_qgroup_limit_item(struct btrfs_trans_handle
*trans
,
607 struct btrfs_root
*root
, u64 qgroupid
,
608 u64 flags
, u64 max_rfer
, u64 max_excl
,
609 u64 rsv_rfer
, u64 rsv_excl
)
611 struct btrfs_path
*path
;
612 struct btrfs_key key
;
613 struct extent_buffer
*l
;
614 struct btrfs_qgroup_limit_item
*qgroup_limit
;
619 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
620 key
.offset
= qgroupid
;
622 path
= btrfs_alloc_path();
626 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
634 slot
= path
->slots
[0];
635 qgroup_limit
= btrfs_item_ptr(l
, path
->slots
[0],
636 struct btrfs_qgroup_limit_item
);
637 btrfs_set_qgroup_limit_flags(l
, qgroup_limit
, flags
);
638 btrfs_set_qgroup_limit_max_rfer(l
, qgroup_limit
, max_rfer
);
639 btrfs_set_qgroup_limit_max_excl(l
, qgroup_limit
, max_excl
);
640 btrfs_set_qgroup_limit_rsv_rfer(l
, qgroup_limit
, rsv_rfer
);
641 btrfs_set_qgroup_limit_rsv_excl(l
, qgroup_limit
, rsv_excl
);
643 btrfs_mark_buffer_dirty(l
);
646 btrfs_free_path(path
);
650 static int update_qgroup_info_item(struct btrfs_trans_handle
*trans
,
651 struct btrfs_root
*root
,
652 struct btrfs_qgroup
*qgroup
)
654 struct btrfs_path
*path
;
655 struct btrfs_key key
;
656 struct extent_buffer
*l
;
657 struct btrfs_qgroup_info_item
*qgroup_info
;
662 key
.type
= BTRFS_QGROUP_INFO_KEY
;
663 key
.offset
= qgroup
->qgroupid
;
665 path
= btrfs_alloc_path();
669 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
677 slot
= path
->slots
[0];
678 qgroup_info
= btrfs_item_ptr(l
, path
->slots
[0],
679 struct btrfs_qgroup_info_item
);
680 btrfs_set_qgroup_info_generation(l
, qgroup_info
, trans
->transid
);
681 btrfs_set_qgroup_info_rfer(l
, qgroup_info
, qgroup
->rfer
);
682 btrfs_set_qgroup_info_rfer_cmpr(l
, qgroup_info
, qgroup
->rfer_cmpr
);
683 btrfs_set_qgroup_info_excl(l
, qgroup_info
, qgroup
->excl
);
684 btrfs_set_qgroup_info_excl_cmpr(l
, qgroup_info
, qgroup
->excl_cmpr
);
686 btrfs_mark_buffer_dirty(l
);
689 btrfs_free_path(path
);
693 static int update_qgroup_status_item(struct btrfs_trans_handle
*trans
,
694 struct btrfs_fs_info
*fs_info
,
695 struct btrfs_root
*root
)
697 struct btrfs_path
*path
;
698 struct btrfs_key key
;
699 struct extent_buffer
*l
;
700 struct btrfs_qgroup_status_item
*ptr
;
705 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
708 path
= btrfs_alloc_path();
712 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
720 slot
= path
->slots
[0];
721 ptr
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_status_item
);
722 btrfs_set_qgroup_status_flags(l
, ptr
, fs_info
->qgroup_flags
);
723 btrfs_set_qgroup_status_generation(l
, ptr
, trans
->transid
);
726 btrfs_mark_buffer_dirty(l
);
729 btrfs_free_path(path
);
734 * called with qgroup_lock held
736 static int btrfs_clean_quota_tree(struct btrfs_trans_handle
*trans
,
737 struct btrfs_root
*root
)
739 struct btrfs_path
*path
;
740 struct btrfs_key key
;
741 struct extent_buffer
*leaf
= NULL
;
745 path
= btrfs_alloc_path();
749 path
->leave_spinning
= 1;
756 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
759 leaf
= path
->nodes
[0];
760 nr
= btrfs_header_nritems(leaf
);
764 * delete the leaf one by one
765 * since the whole tree is going
769 ret
= btrfs_del_items(trans
, root
, path
, 0, nr
);
773 btrfs_release_path(path
);
777 root
->fs_info
->pending_quota_state
= 0;
778 btrfs_free_path(path
);
782 int btrfs_quota_enable(struct btrfs_trans_handle
*trans
,
783 struct btrfs_fs_info
*fs_info
)
785 struct btrfs_root
*quota_root
;
786 struct btrfs_path
*path
= NULL
;
787 struct btrfs_qgroup_status_item
*ptr
;
788 struct extent_buffer
*leaf
;
789 struct btrfs_key key
;
792 spin_lock(&fs_info
->qgroup_lock
);
793 if (fs_info
->quota_root
) {
794 fs_info
->pending_quota_state
= 1;
795 spin_unlock(&fs_info
->qgroup_lock
);
798 spin_unlock(&fs_info
->qgroup_lock
);
801 * initially create the quota tree
803 quota_root
= btrfs_create_tree(trans
, fs_info
,
804 BTRFS_QUOTA_TREE_OBJECTID
);
805 if (IS_ERR(quota_root
)) {
806 ret
= PTR_ERR(quota_root
);
810 path
= btrfs_alloc_path();
817 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
820 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
825 leaf
= path
->nodes
[0];
826 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0],
827 struct btrfs_qgroup_status_item
);
828 btrfs_set_qgroup_status_generation(leaf
, ptr
, trans
->transid
);
829 btrfs_set_qgroup_status_version(leaf
, ptr
, BTRFS_QGROUP_STATUS_VERSION
);
830 fs_info
->qgroup_flags
= BTRFS_QGROUP_STATUS_FLAG_ON
|
831 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
832 btrfs_set_qgroup_status_flags(leaf
, ptr
, fs_info
->qgroup_flags
);
833 btrfs_set_qgroup_status_scan(leaf
, ptr
, 0);
835 btrfs_mark_buffer_dirty(leaf
);
837 spin_lock(&fs_info
->qgroup_lock
);
838 fs_info
->quota_root
= quota_root
;
839 fs_info
->pending_quota_state
= 1;
840 spin_unlock(&fs_info
->qgroup_lock
);
842 btrfs_free_path(path
);
845 free_extent_buffer(quota_root
->node
);
846 free_extent_buffer(quota_root
->commit_root
);
853 int btrfs_quota_disable(struct btrfs_trans_handle
*trans
,
854 struct btrfs_fs_info
*fs_info
)
856 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
857 struct btrfs_root
*quota_root
;
860 spin_lock(&fs_info
->qgroup_lock
);
861 if (!fs_info
->quota_root
) {
862 spin_unlock(&fs_info
->qgroup_lock
);
865 fs_info
->quota_enabled
= 0;
866 fs_info
->pending_quota_state
= 0;
867 quota_root
= fs_info
->quota_root
;
868 fs_info
->quota_root
= NULL
;
869 btrfs_free_qgroup_config(fs_info
);
870 spin_unlock(&fs_info
->qgroup_lock
);
875 ret
= btrfs_clean_quota_tree(trans
, quota_root
);
879 ret
= btrfs_del_root(trans
, tree_root
, "a_root
->root_key
);
883 list_del("a_root
->dirty_list
);
885 btrfs_tree_lock(quota_root
->node
);
886 clean_tree_block(trans
, tree_root
, quota_root
->node
);
887 btrfs_tree_unlock(quota_root
->node
);
888 btrfs_free_tree_block(trans
, quota_root
, quota_root
->node
, 0, 1);
890 free_extent_buffer(quota_root
->node
);
891 free_extent_buffer(quota_root
->commit_root
);
897 int btrfs_quota_rescan(struct btrfs_fs_info
*fs_info
)
903 int btrfs_add_qgroup_relation(struct btrfs_trans_handle
*trans
,
904 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
906 struct btrfs_root
*quota_root
;
909 quota_root
= fs_info
->quota_root
;
913 ret
= add_qgroup_relation_item(trans
, quota_root
, src
, dst
);
917 ret
= add_qgroup_relation_item(trans
, quota_root
, dst
, src
);
919 del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
923 spin_lock(&fs_info
->qgroup_lock
);
924 ret
= add_relation_rb(quota_root
->fs_info
, src
, dst
);
925 spin_unlock(&fs_info
->qgroup_lock
);
930 int btrfs_del_qgroup_relation(struct btrfs_trans_handle
*trans
,
931 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
933 struct btrfs_root
*quota_root
;
937 quota_root
= fs_info
->quota_root
;
941 ret
= del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
942 err
= del_qgroup_relation_item(trans
, quota_root
, dst
, src
);
946 spin_lock(&fs_info
->qgroup_lock
);
947 del_relation_rb(fs_info
, src
, dst
);
949 spin_unlock(&fs_info
->qgroup_lock
);
954 int btrfs_create_qgroup(struct btrfs_trans_handle
*trans
,
955 struct btrfs_fs_info
*fs_info
, u64 qgroupid
, char *name
)
957 struct btrfs_root
*quota_root
;
958 struct btrfs_qgroup
*qgroup
;
961 quota_root
= fs_info
->quota_root
;
965 ret
= add_qgroup_item(trans
, quota_root
, qgroupid
);
967 spin_lock(&fs_info
->qgroup_lock
);
968 qgroup
= add_qgroup_rb(fs_info
, qgroupid
);
969 spin_unlock(&fs_info
->qgroup_lock
);
972 ret
= PTR_ERR(qgroup
);
977 int btrfs_remove_qgroup(struct btrfs_trans_handle
*trans
,
978 struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
980 struct btrfs_root
*quota_root
;
981 struct btrfs_qgroup
*qgroup
;
984 quota_root
= fs_info
->quota_root
;
988 /* check if there are no relations to this qgroup */
989 spin_lock(&fs_info
->qgroup_lock
);
990 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
992 if (!list_empty(&qgroup
->groups
) || !list_empty(&qgroup
->members
)) {
993 spin_unlock(&fs_info
->qgroup_lock
);
997 spin_unlock(&fs_info
->qgroup_lock
);
999 ret
= del_qgroup_item(trans
, quota_root
, qgroupid
);
1001 spin_lock(&fs_info
->qgroup_lock
);
1002 del_qgroup_rb(quota_root
->fs_info
, qgroupid
);
1003 spin_unlock(&fs_info
->qgroup_lock
);
1008 int btrfs_limit_qgroup(struct btrfs_trans_handle
*trans
,
1009 struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
1010 struct btrfs_qgroup_limit
*limit
)
1012 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
1013 struct btrfs_qgroup
*qgroup
;
1019 ret
= update_qgroup_limit_item(trans
, quota_root
, qgroupid
,
1020 limit
->flags
, limit
->max_rfer
,
1021 limit
->max_excl
, limit
->rsv_rfer
,
1024 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1025 printk(KERN_INFO
"unable to update quota limit for %llu\n",
1026 (unsigned long long)qgroupid
);
1029 spin_lock(&fs_info
->qgroup_lock
);
1031 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1036 qgroup
->lim_flags
= limit
->flags
;
1037 qgroup
->max_rfer
= limit
->max_rfer
;
1038 qgroup
->max_excl
= limit
->max_excl
;
1039 qgroup
->rsv_rfer
= limit
->rsv_rfer
;
1040 qgroup
->rsv_excl
= limit
->rsv_excl
;
1043 spin_unlock(&fs_info
->qgroup_lock
);
1048 static void qgroup_dirty(struct btrfs_fs_info
*fs_info
,
1049 struct btrfs_qgroup
*qgroup
)
1051 if (list_empty(&qgroup
->dirty
))
1052 list_add(&qgroup
->dirty
, &fs_info
->dirty_qgroups
);
1056 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1057 * the modification into a list that's later used by btrfs_end_transaction to
1058 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1060 int btrfs_qgroup_record_ref(struct btrfs_trans_handle
*trans
,
1061 struct btrfs_delayed_ref_node
*node
,
1062 struct btrfs_delayed_extent_op
*extent_op
)
1064 struct qgroup_update
*u
;
1066 BUG_ON(!trans
->delayed_ref_elem
.seq
);
1067 u
= kmalloc(sizeof(*u
), GFP_NOFS
);
1072 u
->extent_op
= extent_op
;
1073 list_add_tail(&u
->list
, &trans
->qgroup_ref_list
);
1079 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1080 * from the fs. First, all roots referencing the extent are searched, and
1081 * then the space is accounted accordingly to the different roots. The
1082 * accounting algorithm works in 3 steps documented inline.
1084 int btrfs_qgroup_account_ref(struct btrfs_trans_handle
*trans
,
1085 struct btrfs_fs_info
*fs_info
,
1086 struct btrfs_delayed_ref_node
*node
,
1087 struct btrfs_delayed_extent_op
*extent_op
)
1089 struct btrfs_key ins
;
1090 struct btrfs_root
*quota_root
;
1092 struct btrfs_qgroup
*qgroup
;
1093 struct ulist_node
*unode
;
1094 struct ulist
*roots
= NULL
;
1095 struct ulist
*tmp
= NULL
;
1096 struct ulist_iterator uiter
;
1101 if (!fs_info
->quota_enabled
)
1104 BUG_ON(!fs_info
->quota_root
);
1106 ins
.objectid
= node
->bytenr
;
1107 ins
.offset
= node
->num_bytes
;
1108 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1110 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
1111 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1112 struct btrfs_delayed_tree_ref
*ref
;
1113 ref
= btrfs_delayed_node_to_tree_ref(node
);
1114 ref_root
= ref
->root
;
1115 } else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
1116 node
->type
== BTRFS_SHARED_DATA_REF_KEY
) {
1117 struct btrfs_delayed_data_ref
*ref
;
1118 ref
= btrfs_delayed_node_to_data_ref(node
);
1119 ref_root
= ref
->root
;
1124 if (!is_fstree(ref_root
)) {
1126 * non-fs-trees are not being accounted
1131 switch (node
->action
) {
1132 case BTRFS_ADD_DELAYED_REF
:
1133 case BTRFS_ADD_DELAYED_EXTENT
:
1136 case BTRFS_DROP_DELAYED_REF
:
1139 case BTRFS_UPDATE_DELAYED_HEAD
:
1146 * the delayed ref sequence number we pass depends on the direction of
1147 * the operation. for add operations, we pass (node->seq - 1) to skip
1148 * the delayed ref's current sequence number, because we need the state
1149 * of the tree before the add operation. for delete operations, we pass
1150 * (node->seq) to include the delayed ref's current sequence number,
1151 * because we need the state of the tree after the delete operation.
1153 ret
= btrfs_find_all_roots(trans
, fs_info
, node
->bytenr
,
1154 sgn
> 0 ? node
->seq
- 1 : node
->seq
, &roots
);
1158 spin_lock(&fs_info
->qgroup_lock
);
1159 quota_root
= fs_info
->quota_root
;
1163 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
1168 * step 1: for each old ref, visit all nodes once and inc refcnt
1170 tmp
= ulist_alloc(GFP_ATOMIC
);
1175 seq
= fs_info
->qgroup_seq
;
1176 fs_info
->qgroup_seq
+= roots
->nnodes
+ 1; /* max refcnt */
1178 ULIST_ITER_INIT(&uiter
);
1179 while ((unode
= ulist_next(roots
, &uiter
))) {
1180 struct ulist_node
*tmp_unode
;
1181 struct ulist_iterator tmp_uiter
;
1182 struct btrfs_qgroup
*qg
;
1184 qg
= find_qgroup_rb(fs_info
, unode
->val
);
1189 /* XXX id not needed */
1190 ulist_add(tmp
, qg
->qgroupid
, (u64
)(uintptr_t)qg
, GFP_ATOMIC
);
1191 ULIST_ITER_INIT(&tmp_uiter
);
1192 while ((tmp_unode
= ulist_next(tmp
, &tmp_uiter
))) {
1193 struct btrfs_qgroup_list
*glist
;
1195 qg
= (struct btrfs_qgroup
*)(uintptr_t)tmp_unode
->aux
;
1196 if (qg
->refcnt
< seq
)
1197 qg
->refcnt
= seq
+ 1;
1201 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1202 ulist_add(tmp
, glist
->group
->qgroupid
,
1203 (u64
)(uintptr_t)glist
->group
,
1210 * step 2: walk from the new root
1213 ulist_add(tmp
, qgroup
->qgroupid
, (uintptr_t)qgroup
, GFP_ATOMIC
);
1214 ULIST_ITER_INIT(&uiter
);
1215 while ((unode
= ulist_next(tmp
, &uiter
))) {
1216 struct btrfs_qgroup
*qg
;
1217 struct btrfs_qgroup_list
*glist
;
1219 qg
= (struct btrfs_qgroup
*)(uintptr_t)unode
->aux
;
1220 if (qg
->refcnt
< seq
) {
1221 /* not visited by step 1 */
1222 qg
->rfer
+= sgn
* node
->num_bytes
;
1223 qg
->rfer_cmpr
+= sgn
* node
->num_bytes
;
1224 if (roots
->nnodes
== 0) {
1225 qg
->excl
+= sgn
* node
->num_bytes
;
1226 qg
->excl_cmpr
+= sgn
* node
->num_bytes
;
1228 qgroup_dirty(fs_info
, qg
);
1230 WARN_ON(qg
->tag
>= seq
);
1233 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1234 ulist_add(tmp
, glist
->group
->qgroupid
,
1235 (uintptr_t)glist
->group
, GFP_ATOMIC
);
1240 * step 3: walk again from old refs
1242 ULIST_ITER_INIT(&uiter
);
1243 while ((unode
= ulist_next(roots
, &uiter
))) {
1244 struct btrfs_qgroup
*qg
;
1245 struct ulist_node
*tmp_unode
;
1246 struct ulist_iterator tmp_uiter
;
1248 qg
= find_qgroup_rb(fs_info
, unode
->val
);
1253 ulist_add(tmp
, qg
->qgroupid
, (uintptr_t)qg
, GFP_ATOMIC
);
1254 ULIST_ITER_INIT(&tmp_uiter
);
1255 while ((tmp_unode
= ulist_next(tmp
, &tmp_uiter
))) {
1256 struct btrfs_qgroup_list
*glist
;
1258 qg
= (struct btrfs_qgroup
*)(uintptr_t)tmp_unode
->aux
;
1262 if (qg
->refcnt
- seq
== roots
->nnodes
) {
1263 qg
->excl
-= sgn
* node
->num_bytes
;
1264 qg
->excl_cmpr
-= sgn
* node
->num_bytes
;
1265 qgroup_dirty(fs_info
, qg
);
1268 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1269 ulist_add(tmp
, glist
->group
->qgroupid
,
1270 (uintptr_t)glist
->group
,
1277 spin_unlock(&fs_info
->qgroup_lock
);
1285 * called from commit_transaction. Writes all changed qgroups to disk.
1287 int btrfs_run_qgroups(struct btrfs_trans_handle
*trans
,
1288 struct btrfs_fs_info
*fs_info
)
1290 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
1296 fs_info
->quota_enabled
= fs_info
->pending_quota_state
;
1298 spin_lock(&fs_info
->qgroup_lock
);
1299 while (!list_empty(&fs_info
->dirty_qgroups
)) {
1300 struct btrfs_qgroup
*qgroup
;
1301 qgroup
= list_first_entry(&fs_info
->dirty_qgroups
,
1302 struct btrfs_qgroup
, dirty
);
1303 list_del_init(&qgroup
->dirty
);
1304 spin_unlock(&fs_info
->qgroup_lock
);
1305 ret
= update_qgroup_info_item(trans
, quota_root
, qgroup
);
1307 fs_info
->qgroup_flags
|=
1308 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1309 spin_lock(&fs_info
->qgroup_lock
);
1311 if (fs_info
->quota_enabled
)
1312 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_ON
;
1314 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
1315 spin_unlock(&fs_info
->qgroup_lock
);
1317 ret
= update_qgroup_status_item(trans
, fs_info
, quota_root
);
1319 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1327 * copy the acounting information between qgroups. This is necessary when a
1328 * snapshot or a subvolume is created
1330 int btrfs_qgroup_inherit(struct btrfs_trans_handle
*trans
,
1331 struct btrfs_fs_info
*fs_info
, u64 srcid
, u64 objectid
,
1332 struct btrfs_qgroup_inherit
*inherit
)
1337 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
1338 struct btrfs_qgroup
*srcgroup
;
1339 struct btrfs_qgroup
*dstgroup
;
1342 if (!fs_info
->quota_enabled
)
1349 * create a tracking group for the subvol itself
1351 ret
= add_qgroup_item(trans
, quota_root
, objectid
);
1355 if (inherit
&& inherit
->flags
& BTRFS_QGROUP_INHERIT_SET_LIMITS
) {
1356 ret
= update_qgroup_limit_item(trans
, quota_root
, objectid
,
1358 inherit
->lim
.max_rfer
,
1359 inherit
->lim
.max_excl
,
1360 inherit
->lim
.rsv_rfer
,
1361 inherit
->lim
.rsv_excl
);
1367 struct btrfs_root
*srcroot
;
1368 struct btrfs_key srckey
;
1371 srckey
.objectid
= srcid
;
1372 srckey
.type
= BTRFS_ROOT_ITEM_KEY
;
1373 srckey
.offset
= (u64
)-1;
1374 srcroot
= btrfs_read_fs_root_no_name(fs_info
, &srckey
);
1375 if (IS_ERR(srcroot
)) {
1376 ret
= PTR_ERR(srcroot
);
1381 srcroot_level
= btrfs_header_level(srcroot
->node
);
1382 level_size
= btrfs_level_size(srcroot
, srcroot_level
);
1387 * add qgroup to all inherited groups
1390 i_qgroups
= (u64
*)(inherit
+ 1);
1391 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
1392 ret
= add_qgroup_relation_item(trans
, quota_root
,
1393 objectid
, *i_qgroups
);
1396 ret
= add_qgroup_relation_item(trans
, quota_root
,
1397 *i_qgroups
, objectid
);
1405 spin_lock(&fs_info
->qgroup_lock
);
1407 dstgroup
= add_qgroup_rb(fs_info
, objectid
);
1408 if (IS_ERR(dstgroup
)) {
1409 ret
= PTR_ERR(dstgroup
);
1414 srcgroup
= find_qgroup_rb(fs_info
, srcid
);
1417 dstgroup
->rfer
= srcgroup
->rfer
- level_size
;
1418 dstgroup
->rfer_cmpr
= srcgroup
->rfer_cmpr
- level_size
;
1419 srcgroup
->excl
= level_size
;
1420 srcgroup
->excl_cmpr
= level_size
;
1421 qgroup_dirty(fs_info
, dstgroup
);
1422 qgroup_dirty(fs_info
, srcgroup
);
1428 i_qgroups
= (u64
*)(inherit
+ 1);
1429 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
1430 ret
= add_relation_rb(quota_root
->fs_info
, objectid
,
1437 for (i
= 0; i
< inherit
->num_ref_copies
; ++i
) {
1438 struct btrfs_qgroup
*src
;
1439 struct btrfs_qgroup
*dst
;
1441 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
1442 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
1449 dst
->rfer
= src
->rfer
- level_size
;
1450 dst
->rfer_cmpr
= src
->rfer_cmpr
- level_size
;
1453 for (i
= 0; i
< inherit
->num_excl_copies
; ++i
) {
1454 struct btrfs_qgroup
*src
;
1455 struct btrfs_qgroup
*dst
;
1457 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
1458 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
1465 dst
->excl
= src
->excl
+ level_size
;
1466 dst
->excl_cmpr
= src
->excl_cmpr
+ level_size
;
1471 spin_unlock(&fs_info
->qgroup_lock
);
1477 * reserve some space for a qgroup and all its parents. The reservation takes
1478 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1479 * accounting. If not enough space is available, EDQUOT is returned.
1480 * We assume that the requested space is new for all qgroups.
1482 int btrfs_qgroup_reserve(struct btrfs_root
*root
, u64 num_bytes
)
1484 struct btrfs_root
*quota_root
;
1485 struct btrfs_qgroup
*qgroup
;
1486 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1487 u64 ref_root
= root
->root_key
.objectid
;
1489 struct ulist
*ulist
= NULL
;
1490 struct ulist_node
*unode
;
1491 struct ulist_iterator uiter
;
1493 if (!is_fstree(ref_root
))
1499 spin_lock(&fs_info
->qgroup_lock
);
1500 quota_root
= fs_info
->quota_root
;
1504 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
1509 * in a first step, we check all affected qgroups if any limits would
1512 ulist
= ulist_alloc(GFP_ATOMIC
);
1517 ulist_add(ulist
, qgroup
->qgroupid
, (uintptr_t)qgroup
, GFP_ATOMIC
);
1518 ULIST_ITER_INIT(&uiter
);
1519 while ((unode
= ulist_next(ulist
, &uiter
))) {
1520 struct btrfs_qgroup
*qg
;
1521 struct btrfs_qgroup_list
*glist
;
1523 qg
= (struct btrfs_qgroup
*)(uintptr_t)unode
->aux
;
1525 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) &&
1526 qg
->reserved
+ qg
->rfer
+ num_bytes
>
1532 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) &&
1533 qg
->reserved
+ qg
->excl
+ num_bytes
>
1539 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1540 ulist_add(ulist
, glist
->group
->qgroupid
,
1541 (uintptr_t)glist
->group
, GFP_ATOMIC
);
1546 * no limits exceeded, now record the reservation into all qgroups
1548 ULIST_ITER_INIT(&uiter
);
1549 while ((unode
= ulist_next(ulist
, &uiter
))) {
1550 struct btrfs_qgroup
*qg
;
1552 qg
= (struct btrfs_qgroup
*)(uintptr_t)unode
->aux
;
1554 qg
->reserved
+= num_bytes
;
1558 spin_unlock(&fs_info
->qgroup_lock
);
1564 void btrfs_qgroup_free(struct btrfs_root
*root
, u64 num_bytes
)
1566 struct btrfs_root
*quota_root
;
1567 struct btrfs_qgroup
*qgroup
;
1568 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1569 struct ulist
*ulist
= NULL
;
1570 struct ulist_node
*unode
;
1571 struct ulist_iterator uiter
;
1572 u64 ref_root
= root
->root_key
.objectid
;
1574 if (!is_fstree(ref_root
))
1580 spin_lock(&fs_info
->qgroup_lock
);
1582 quota_root
= fs_info
->quota_root
;
1586 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
1590 ulist
= ulist_alloc(GFP_ATOMIC
);
1592 btrfs_std_error(fs_info
, -ENOMEM
);
1595 ulist_add(ulist
, qgroup
->qgroupid
, (uintptr_t)qgroup
, GFP_ATOMIC
);
1596 ULIST_ITER_INIT(&uiter
);
1597 while ((unode
= ulist_next(ulist
, &uiter
))) {
1598 struct btrfs_qgroup
*qg
;
1599 struct btrfs_qgroup_list
*glist
;
1601 qg
= (struct btrfs_qgroup
*)(uintptr_t)unode
->aux
;
1603 qg
->reserved
-= num_bytes
;
1605 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1606 ulist_add(ulist
, glist
->group
->qgroupid
,
1607 (uintptr_t)glist
->group
, GFP_ATOMIC
);
1612 spin_unlock(&fs_info
->qgroup_lock
);
1616 void assert_qgroups_uptodate(struct btrfs_trans_handle
*trans
)
1618 if (list_empty(&trans
->qgroup_ref_list
) && !trans
->delayed_ref_elem
.seq
)
1620 printk(KERN_ERR
"btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1621 trans
, list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
1622 trans
->delayed_ref_elem
.seq
);