projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: introduce a mutex lock for btrfs quota operations
[linux-block.git] / fs / btrfs / qgroup.c
CommitLineData
bed92eae
AJ		 1/*
2 * Copyright (C) 2011 STRATO. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
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>
55e301fd 26#include <linux/btrfs.h>
bed92eae
AJ		 27
28#include "ctree.h"
29#include "transaction.h"
30#include "disk-io.h"
31#include "locking.h"
32#include "ulist.h"
bed92eae
AJ		 33#include "backref.h"
34
35/* TODO XXX FIXME
36 * - subvol delete -> delete when ref goes to 0? delete limits also?
37 * - reorganize keys
38 * - compressed
39 * - sync
40 * - rescan
41 * - copy also limits on subvol creation
42 * - limit
43 * - caches fuer ulists
44 * - performance benchmarks
45 * - check all ioctl parameters
46 */
47
48/*
49 * one struct for each qgroup, organized in fs_info->qgroup_tree.
50 */
51struct btrfs_qgroup {
52 u64 qgroupid;
53
54 /*
55 * state
56 */
57 u64 rfer; /* referenced */
58 u64 rfer_cmpr; /* referenced compressed */
59 u64 excl; /* exclusive */
60 u64 excl_cmpr; /* exclusive compressed */
61
62 /*
63 * limits
64 */
65 u64 lim_flags; /* which limits are set */
66 u64 max_rfer;
67 u64 max_excl;
68 u64 rsv_rfer;
69 u64 rsv_excl;
70
71 /*
72 * reservation tracking
73 */
74 u64 reserved;
75
76 /*
77 * lists
78 */
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 */
83
84 /*
85 * temp variables for accounting operations
86 */
87 u64 tag;
88 u64 refcnt;
89};
90
91/*
92 * glue structure to represent the relations between qgroups.
93 */
94struct 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;
99};
100
101/* must be called with qgroup_lock held */
102static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
103 u64 qgroupid)
104{
105 struct rb_node *n = fs_info->qgroup_tree.rb_node;
106 struct btrfs_qgroup *qgroup;
107
108 while (n) {
109 qgroup = rb_entry(n, struct btrfs_qgroup, node);
110 if (qgroup->qgroupid < qgroupid)
111 n = n->rb_left;
112 else if (qgroup->qgroupid > qgroupid)
113 n = n->rb_right;
114 else
115 return qgroup;
116 }
117 return NULL;
118}
119
120/* must be called with qgroup_lock held */
121static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
122 u64 qgroupid)
123{
124 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
125 struct rb_node *parent = NULL;
126 struct btrfs_qgroup *qgroup;
127
128 while (*p) {
129 parent = *p;
130 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
131
132 if (qgroup->qgroupid < qgroupid)
133 p = &(*p)->rb_left;
134 else if (qgroup->qgroupid > qgroupid)
135 p = &(*p)->rb_right;
136 else
137 return qgroup;
138 }
139
140 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
141 if (!qgroup)
142 return ERR_PTR(-ENOMEM);
143
144 qgroup->qgroupid = qgroupid;
145 INIT_LIST_HEAD(&qgroup->groups);
146 INIT_LIST_HEAD(&qgroup->members);
147 INIT_LIST_HEAD(&qgroup->dirty);
148
149 rb_link_node(&qgroup->node, parent, p);
150 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
151
152 return qgroup;
153}
154
155/* must be called with qgroup_lock held */
156static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
157{
158 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
159 struct btrfs_qgroup_list *list;
160
161 if (!qgroup)
162 return -ENOENT;
163
164 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
165 list_del(&qgroup->dirty);
166
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);
172 kfree(list);
173 }
174
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);
180 kfree(list);
181 }
182 kfree(qgroup);
183
184 return 0;
185}
186
187/* must be called with qgroup_lock held */
188static int add_relation_rb(struct btrfs_fs_info *fs_info,
189 u64 memberid, u64 parentid)
190{
191 struct btrfs_qgroup *member;
192 struct btrfs_qgroup *parent;
193 struct btrfs_qgroup_list *list;
194
195 member = find_qgroup_rb(fs_info, memberid);
196 parent = find_qgroup_rb(fs_info, parentid);
197 if (!member || !parent)
198 return -ENOENT;
199
200 list = kzalloc(sizeof(*list), GFP_ATOMIC);
201 if (!list)
202 return -ENOMEM;
203
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);
208
209 return 0;
210}
211
212/* must be called with qgroup_lock held */
213static int del_relation_rb(struct btrfs_fs_info *fs_info,
214 u64 memberid, u64 parentid)
215{
216 struct btrfs_qgroup *member;
217 struct btrfs_qgroup *parent;
218 struct btrfs_qgroup_list *list;
219
220 member = find_qgroup_rb(fs_info, memberid);
221 parent = find_qgroup_rb(fs_info, parentid);
222 if (!member || !parent)
223 return -ENOENT;
224
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);
229 kfree(list);
230 return 0;
231 }
232 }
233 return -ENOENT;
234}
235
236/*
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
239 */
240int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
241{
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;
247 int slot;
248 int ret = 0;
249 u64 flags = 0;
250
251 if (!fs_info->quota_enabled)
252 return 0;
253
254 path = btrfs_alloc_path();
255 if (!path) {
256 ret = -ENOMEM;
257 goto out;
258 }
259
260 /* default this to quota off, in case no status key is found */
261 fs_info->qgroup_flags = 0;
262
263 /*
264 * pass 1: read status, all qgroup infos and limits
265 */
266 key.objectid = 0;
267 key.type = 0;
268 key.offset = 0;
269 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
270 if (ret)
271 goto out;
272
273 while (1) {
274 struct btrfs_qgroup *qgroup;
275
276 slot = path->slots[0];
277 l = path->nodes[0];
278 btrfs_item_key_to_cpu(l, &found_key, slot);
279
280 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
281 struct btrfs_qgroup_status_item *ptr;
282
283 ptr = btrfs_item_ptr(l, slot,
284 struct btrfs_qgroup_status_item);
285
286 if (btrfs_qgroup_status_version(l, ptr) !=
287 BTRFS_QGROUP_STATUS_VERSION) {
288 printk(KERN_ERR
289 "btrfs: old qgroup version, quota disabled\n");
290 goto out;
291 }
292 if (btrfs_qgroup_status_generation(l, ptr) !=
293 fs_info->generation) {
294 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
295 printk(KERN_ERR
296 "btrfs: qgroup generation mismatch, "
297 "marked as inconsistent\n");
298 }
299 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
300 ptr);
301 /* FIXME read scan element */
302 goto next1;
303 }
304
305 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
306 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
307 goto next1;
308
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;
314 }
315 if (!qgroup) {
316 qgroup = add_qgroup_rb(fs_info, found_key.offset);
317 if (IS_ERR(qgroup)) {
318 ret = PTR_ERR(qgroup);
319 goto out;
320 }
321 }
322 switch (found_key.type) {
323 case BTRFS_QGROUP_INFO_KEY: {
324 struct btrfs_qgroup_info_item *ptr;
325
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 */
333 break;
334 }
335 case BTRFS_QGROUP_LIMIT_KEY: {
336 struct btrfs_qgroup_limit_item *ptr;
337
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);
345 break;
346 }
347 }
348next1:
349 ret = btrfs_next_item(quota_root, path);
350 if (ret < 0)
351 goto out;
352 if (ret)
353 break;
354 }
355 btrfs_release_path(path);
356
357 /*
358 * pass 2: read all qgroup relations
359 */
360 key.objectid = 0;
361 key.type = BTRFS_QGROUP_RELATION_KEY;
362 key.offset = 0;
363 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
364 if (ret)
365 goto out;
366 while (1) {
367 slot = path->slots[0];
368 l = path->nodes[0];
369 btrfs_item_key_to_cpu(l, &found_key, slot);
370
371 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
372 goto next2;
373
374 if (found_key.objectid > found_key.offset) {
375 /* parent <- member, not needed to build config */
376 /* FIXME should we omit the key completely? */
377 goto next2;
378 }
379
380 ret = add_relation_rb(fs_info, found_key.objectid,
381 found_key.offset);
ff24858c
AJ		 382 if (ret == -ENOENT) {
383 printk(KERN_WARNING
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 */
388 }
bed92eae
AJ		 389 if (ret)
390 goto out;
391next2:
392 ret = btrfs_next_item(quota_root, path);
393 if (ret < 0)
394 goto out;
395 if (ret)
396 break;
397 }
398out:
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;
403 }
404 btrfs_free_path(path);
405
406 return ret < 0 ? ret : 0;
407}
408
409/*
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.
412 */
413void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
414{
415 struct rb_node *n;
416 struct btrfs_qgroup *qgroup;
417 struct btrfs_qgroup_list *list;
418
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);
422
bed92eae
AJ		 423 while (!list_empty(&qgroup->groups)) {
424 list = list_first_entry(&qgroup->groups,
425 struct btrfs_qgroup_list,
426 next_group);
427 list_del(&list->next_group);
428 list_del(&list->next_member);
429 kfree(list);
430 }
431
432 while (!list_empty(&qgroup->members)) {
433 list = list_first_entry(&qgroup->members,
434 struct btrfs_qgroup_list,
435 next_member);
436 list_del(&list->next_group);
437 list_del(&list->next_member);
438 kfree(list);
439 }
440 kfree(qgroup);
441 }
442}
443
444static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
445 struct btrfs_root *quota_root,
446 u64 src, u64 dst)
447{
448 int ret;
449 struct btrfs_path *path;
450 struct btrfs_key key;
451
452 path = btrfs_alloc_path();
453 if (!path)
454 return -ENOMEM;
455
456 key.objectid = src;
457 key.type = BTRFS_QGROUP_RELATION_KEY;
458 key.offset = dst;
459
460 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
461
462 btrfs_mark_buffer_dirty(path->nodes[0]);
463
464 btrfs_free_path(path);
465 return ret;
466}
467
468static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
469 struct btrfs_root *quota_root,
470 u64 src, u64 dst)
471{
472 int ret;
473 struct btrfs_path *path;
474 struct btrfs_key key;
475
476 path = btrfs_alloc_path();
477 if (!path)
478 return -ENOMEM;
479
480 key.objectid = src;
481 key.type = BTRFS_QGROUP_RELATION_KEY;
482 key.offset = dst;
483
484 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
485 if (ret < 0)
486 goto out;
487
488 if (ret > 0) {
489 ret = -ENOENT;
490 goto out;
491 }
492
493 ret = btrfs_del_item(trans, quota_root, path);
494out:
495 btrfs_free_path(path);
496 return ret;
497}
498
499static int add_qgroup_item(struct btrfs_trans_handle *trans,
500 struct btrfs_root *quota_root, u64 qgroupid)
501{
502 int ret;
503 struct btrfs_path *path;
504 struct btrfs_qgroup_info_item *qgroup_info;
505 struct btrfs_qgroup_limit_item *qgroup_limit;
506 struct extent_buffer *leaf;
507 struct btrfs_key key;
508
509 path = btrfs_alloc_path();
510 if (!path)
511 return -ENOMEM;
512
513 key.objectid = 0;
514 key.type = BTRFS_QGROUP_INFO_KEY;
515 key.offset = qgroupid;
516
517 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
518 sizeof(*qgroup_info));
519 if (ret)
520 goto out;
521
522 leaf = path->nodes[0];
523 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
524 struct btrfs_qgroup_info_item);
525 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
526 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
527 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
528 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
529 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
530
531 btrfs_mark_buffer_dirty(leaf);
532
533 btrfs_release_path(path);
534
535 key.type = BTRFS_QGROUP_LIMIT_KEY;
536 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
537 sizeof(*qgroup_limit));
538 if (ret)
539 goto out;
540
541 leaf = path->nodes[0];
542 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
543 struct btrfs_qgroup_limit_item);
544 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
545 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
546 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
547 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
548 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
549
550 btrfs_mark_buffer_dirty(leaf);
551
552 ret = 0;
553out:
554 btrfs_free_path(path);
555 return ret;
556}
557
558static int del_qgroup_item(struct btrfs_trans_handle *trans,
559 struct btrfs_root *quota_root, u64 qgroupid)
560{
561 int ret;
562 struct btrfs_path *path;
563 struct btrfs_key key;
564
565 path = btrfs_alloc_path();
566 if (!path)
567 return -ENOMEM;
568
569 key.objectid = 0;
570 key.type = BTRFS_QGROUP_INFO_KEY;
571 key.offset = qgroupid;
572 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
573 if (ret < 0)
574 goto out;
575
576 if (ret > 0) {
577 ret = -ENOENT;
578 goto out;
579 }
580
581 ret = btrfs_del_item(trans, quota_root, path);
582 if (ret)
583 goto out;
584
585 btrfs_release_path(path);
586
587 key.type = BTRFS_QGROUP_LIMIT_KEY;
588 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
589 if (ret < 0)
590 goto out;
591
592 if (ret > 0) {
593 ret = -ENOENT;
594 goto out;
595 }
596
597 ret = btrfs_del_item(trans, quota_root, path);
598
599out:
600 btrfs_free_path(path);
601 return ret;
602}
603
604static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
605 struct btrfs_root *root, u64 qgroupid,
606 u64 flags, u64 max_rfer, u64 max_excl,
607 u64 rsv_rfer, u64 rsv_excl)
608{
609 struct btrfs_path *path;
610 struct btrfs_key key;
611 struct extent_buffer *l;
612 struct btrfs_qgroup_limit_item *qgroup_limit;
613 int ret;
614 int slot;
615
616 key.objectid = 0;
617 key.type = BTRFS_QGROUP_LIMIT_KEY;
618 key.offset = qgroupid;
619
620 path = btrfs_alloc_path();
84cbe2f7
WS		 621 if (!path)
622 return -ENOMEM;
623
bed92eae
AJ		 624 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
625 if (ret > 0)
626 ret = -ENOENT;
627
628 if (ret)
629 goto out;
630
631 l = path->nodes[0];
632 slot = path->slots[0];
633 qgroup_limit = btrfs_item_ptr(l, path->slots[0],
634 struct btrfs_qgroup_limit_item);
635 btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
636 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
637 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
638 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
639 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
640
641 btrfs_mark_buffer_dirty(l);
642
643out:
644 btrfs_free_path(path);
645 return ret;
646}
647
648static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
649 struct btrfs_root *root,
650 struct btrfs_qgroup *qgroup)
651{
652 struct btrfs_path *path;
653 struct btrfs_key key;
654 struct extent_buffer *l;
655 struct btrfs_qgroup_info_item *qgroup_info;
656 int ret;
657 int slot;
658
659 key.objectid = 0;
660 key.type = BTRFS_QGROUP_INFO_KEY;
661 key.offset = qgroup->qgroupid;
662
663 path = btrfs_alloc_path();
84cbe2f7
WS		 664 if (!path)
665 return -ENOMEM;
666
bed92eae
AJ		 667 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
668 if (ret > 0)
669 ret = -ENOENT;
670
671 if (ret)
672 goto out;
673
674 l = path->nodes[0];
675 slot = path->slots[0];
676 qgroup_info = btrfs_item_ptr(l, path->slots[0],
677 struct btrfs_qgroup_info_item);
678 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
679 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
680 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
681 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
682 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
683
684 btrfs_mark_buffer_dirty(l);
685
686out:
687 btrfs_free_path(path);
688 return ret;
689}
690
691static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
692 struct btrfs_fs_info *fs_info,
693 struct btrfs_root *root)
694{
695 struct btrfs_path *path;
696 struct btrfs_key key;
697 struct extent_buffer *l;
698 struct btrfs_qgroup_status_item *ptr;
699 int ret;
700 int slot;
701
702 key.objectid = 0;
703 key.type = BTRFS_QGROUP_STATUS_KEY;
704 key.offset = 0;
705
706 path = btrfs_alloc_path();
84cbe2f7
WS		 707 if (!path)
708 return -ENOMEM;
709
bed92eae
AJ		 710 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
711 if (ret > 0)
712 ret = -ENOENT;
713
714 if (ret)
715 goto out;
716
717 l = path->nodes[0];
718 slot = path->slots[0];
719 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
720 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
721 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
722 /* XXX scan */
723
724 btrfs_mark_buffer_dirty(l);
725
726out:
727 btrfs_free_path(path);
728 return ret;
729}
730
731/*
732 * called with qgroup_lock held
733 */
734static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root)
736{
737 struct btrfs_path *path;
738 struct btrfs_key key;
06b3a860 739 struct extent_buffer *leaf = NULL;
bed92eae 740 int ret;
06b3a860 741 int nr = 0;
bed92eae 742
bed92eae
AJ		 743 path = btrfs_alloc_path();
744 if (!path)
745 return -ENOMEM;
746
06b3a860
WS		 747 path->leave_spinning = 1;
748
749 key.objectid = 0;
750 key.offset = 0;
751 key.type = 0;
bed92eae 752
06b3a860 753 while (1) {
bed92eae 754 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
06b3a860
WS		 755 if (ret < 0)
756 goto out;
757 leaf = path->nodes[0];
758 nr = btrfs_header_nritems(leaf);
759 if (!nr)
bed92eae 760 break;
06b3a860
WS		 761 /*
762 * delete the leaf one by one
763 * since the whole tree is going
764 * to be deleted.
765 */
766 path->slots[0] = 0;
767 ret = btrfs_del_items(trans, root, path, 0, nr);
bed92eae
AJ		 768 if (ret)
769 goto out;
06b3a860 770
bed92eae
AJ		 771 btrfs_release_path(path);
772 }
773 ret = 0;
774out:
775 root->fs_info->pending_quota_state = 0;
776 btrfs_free_path(path);
777 return ret;
778}
779
780int btrfs_quota_enable(struct btrfs_trans_handle *trans,
781 struct btrfs_fs_info *fs_info)
782{
783 struct btrfs_root *quota_root;
7708f029 784 struct btrfs_root *tree_root = fs_info->tree_root;
bed92eae
AJ		 785 struct btrfs_path *path = NULL;
786 struct btrfs_qgroup_status_item *ptr;
787 struct extent_buffer *leaf;
788 struct btrfs_key key;
7708f029
WS		 789 struct btrfs_key found_key;
790 struct btrfs_qgroup *qgroup = NULL;
bed92eae 791 int ret = 0;
7708f029 792 int slot;
bed92eae 793
f2f6ed3d 794 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 795 spin_lock(&fs_info->qgroup_lock);
796 if (fs_info->quota_root) {
797 fs_info->pending_quota_state = 1;
798 spin_unlock(&fs_info->qgroup_lock);
799 goto out;
800 }
801 spin_unlock(&fs_info->qgroup_lock);
802
803 /*
804 * initially create the quota tree
805 */
806 quota_root = btrfs_create_tree(trans, fs_info,
807 BTRFS_QUOTA_TREE_OBJECTID);
808 if (IS_ERR(quota_root)) {
809 ret = PTR_ERR(quota_root);
810 goto out;
811 }
812
813 path = btrfs_alloc_path();
5b7ff5b3
TI		 814 if (!path) {
815 ret = -ENOMEM;
816 goto out_free_root;
817 }
bed92eae
AJ		 818
819 key.objectid = 0;
820 key.type = BTRFS_QGROUP_STATUS_KEY;
821 key.offset = 0;
822
823 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
824 sizeof(*ptr));
825 if (ret)
5b7ff5b3 826 goto out_free_path;
bed92eae
AJ		 827
828 leaf = path->nodes[0];
829 ptr = btrfs_item_ptr(leaf, path->slots[0],
830 struct btrfs_qgroup_status_item);
831 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
832 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
833 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
834 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
835 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
836 btrfs_set_qgroup_status_scan(leaf, ptr, 0);
837
838 btrfs_mark_buffer_dirty(leaf);
839
7708f029
WS		 840 key.objectid = 0;
841 key.type = BTRFS_ROOT_REF_KEY;
842 key.offset = 0;
843
844 btrfs_release_path(path);
845 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
846 if (ret > 0)
847 goto out_add_root;
848 if (ret < 0)
849 goto out_free_path;
850
851
852 while (1) {
853 slot = path->slots[0];
854 leaf = path->nodes[0];
855 btrfs_item_key_to_cpu(leaf, &found_key, slot);
856
857 if (found_key.type == BTRFS_ROOT_REF_KEY) {
858 ret = add_qgroup_item(trans, quota_root,
859 found_key.offset);
860 if (ret)
861 goto out_free_path;
862
863 spin_lock(&fs_info->qgroup_lock);
864 qgroup = add_qgroup_rb(fs_info, found_key.offset);
865 if (IS_ERR(qgroup)) {
866 spin_unlock(&fs_info->qgroup_lock);
867 ret = PTR_ERR(qgroup);
868 goto out_free_path;
869 }
870 spin_unlock(&fs_info->qgroup_lock);
871 }
872 ret = btrfs_next_item(tree_root, path);
873 if (ret < 0)
874 goto out_free_path;
875 if (ret)
876 break;
877 }
878
879out_add_root:
880 btrfs_release_path(path);
881 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
882 if (ret)
883 goto out_free_path;
884
bed92eae 885 spin_lock(&fs_info->qgroup_lock);
7708f029
WS		 886 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
887 if (IS_ERR(qgroup)) {
888 spin_unlock(&fs_info->qgroup_lock);
889 ret = PTR_ERR(qgroup);
890 goto out_free_path;
891 }
bed92eae
AJ		 892 fs_info->quota_root = quota_root;
893 fs_info->pending_quota_state = 1;
894 spin_unlock(&fs_info->qgroup_lock);
5b7ff5b3 895out_free_path:
bed92eae 896 btrfs_free_path(path);
5b7ff5b3
TI		 897out_free_root:
898 if (ret) {
899 free_extent_buffer(quota_root->node);
900 free_extent_buffer(quota_root->commit_root);
901 kfree(quota_root);
902 }
903out:
f2f6ed3d 904 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 905 return ret;
906}
907
908int btrfs_quota_disable(struct btrfs_trans_handle *trans,
909 struct btrfs_fs_info *fs_info)
910{
911 struct btrfs_root *tree_root = fs_info->tree_root;
912 struct btrfs_root *quota_root;
913 int ret = 0;
914
f2f6ed3d 915 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 916 spin_lock(&fs_info->qgroup_lock);
683cebda
WS		 917 if (!fs_info->quota_root) {
918 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d 919 goto out;
683cebda 920 }
bed92eae
AJ		 921 fs_info->quota_enabled = 0;
922 fs_info->pending_quota_state = 0;
923 quota_root = fs_info->quota_root;
924 fs_info->quota_root = NULL;
925 btrfs_free_qgroup_config(fs_info);
926 spin_unlock(&fs_info->qgroup_lock);
927
f2f6ed3d
WS		 928 if (!quota_root) {
929 ret = -EINVAL;
930 goto out;
931 }
bed92eae
AJ		 932
933 ret = btrfs_clean_quota_tree(trans, quota_root);
934 if (ret)
935 goto out;
936
937 ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
938 if (ret)
939 goto out;
940
941 list_del(&quota_root->dirty_list);
942
943 btrfs_tree_lock(quota_root->node);
944 clean_tree_block(trans, tree_root, quota_root->node);
945 btrfs_tree_unlock(quota_root->node);
946 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
947
948 free_extent_buffer(quota_root->node);
949 free_extent_buffer(quota_root->commit_root);
950 kfree(quota_root);
951out:
f2f6ed3d 952 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 953 return ret;
954}
955
956int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
957{
958 /* FIXME */
959 return 0;
960}
961
962int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
963 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
964{
965 struct btrfs_root *quota_root;
966 int ret = 0;
967
f2f6ed3d 968 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 969 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 970 if (!quota_root) {
971 ret = -EINVAL;
972 goto out;
973 }
bed92eae
AJ		 974
975 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
976 if (ret)
f2f6ed3d 977 goto out;
bed92eae
AJ		 978
979 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
980 if (ret) {
981 del_qgroup_relation_item(trans, quota_root, src, dst);
f2f6ed3d 982 goto out;
bed92eae
AJ		 983 }
984
985 spin_lock(&fs_info->qgroup_lock);
986 ret = add_relation_rb(quota_root->fs_info, src, dst);
987 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 988out:
989 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 990 return ret;
991}
992
993int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
994 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
995{
996 struct btrfs_root *quota_root;
997 int ret = 0;
998 int err;
999
f2f6ed3d 1000 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1001 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1002 if (!quota_root) {
1003 ret = -EINVAL;
1004 goto out;
1005 }
bed92eae
AJ		 1006
1007 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1008 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1009 if (err && !ret)
1010 ret = err;
1011
1012 spin_lock(&fs_info->qgroup_lock);
1013 del_relation_rb(fs_info, src, dst);
1014
1015 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1016out:
1017 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1018 return ret;
1019}
1020
1021int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1022 struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
1023{
1024 struct btrfs_root *quota_root;
1025 struct btrfs_qgroup *qgroup;
1026 int ret = 0;
1027
f2f6ed3d 1028 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1029 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1030 if (!quota_root) {
1031 ret = -EINVAL;
1032 goto out;
1033 }
bed92eae
AJ		 1034
1035 ret = add_qgroup_item(trans, quota_root, qgroupid);
1036
1037 spin_lock(&fs_info->qgroup_lock);
1038 qgroup = add_qgroup_rb(fs_info, qgroupid);
1039 spin_unlock(&fs_info->qgroup_lock);
1040
1041 if (IS_ERR(qgroup))
1042 ret = PTR_ERR(qgroup);
f2f6ed3d
WS		 1043out:
1044 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1045 return ret;
1046}
1047
1048int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1049 struct btrfs_fs_info *fs_info, u64 qgroupid)
1050{
1051 struct btrfs_root *quota_root;
2cf68703 1052 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 1053 int ret = 0;
1054
f2f6ed3d 1055 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1056 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1057 if (!quota_root) {
1058 ret = -EINVAL;
1059 goto out;
1060 }
bed92eae 1061
2cf68703
AJ		 1062 /* check if there are no relations to this qgroup */
1063 spin_lock(&fs_info->qgroup_lock);
1064 qgroup = find_qgroup_rb(fs_info, qgroupid);
1065 if (qgroup) {
1066 if (!list_empty(&qgroup->groups) || !list_empty(&qgroup->members)) {
1067 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1068 ret = -EBUSY;
1069 goto out;
2cf68703
AJ		 1070 }
1071 }
1072 spin_unlock(&fs_info->qgroup_lock);
1073
bed92eae
AJ		 1074 ret = del_qgroup_item(trans, quota_root, qgroupid);
1075
1076 spin_lock(&fs_info->qgroup_lock);
1077 del_qgroup_rb(quota_root->fs_info, qgroupid);
bed92eae 1078 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1079out:
1080 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1081 return ret;
1082}
1083
1084int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1085 struct btrfs_fs_info *fs_info, u64 qgroupid,
1086 struct btrfs_qgroup_limit *limit)
1087{
f2f6ed3d 1088 struct btrfs_root *quota_root;
bed92eae
AJ		 1089 struct btrfs_qgroup *qgroup;
1090 int ret = 0;
1091
f2f6ed3d
WS		 1092 mutex_lock(&fs_info->qgroup_ioctl_lock);
1093 quota_root = fs_info->quota_root;
1094 if (!quota_root) {
1095 ret = -EINVAL;
1096 goto out;
1097 }
bed92eae
AJ		 1098
1099 ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1100 limit->flags, limit->max_rfer,
1101 limit->max_excl, limit->rsv_rfer,
1102 limit->rsv_excl);
1103 if (ret) {
1104 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1105 printk(KERN_INFO "unable to update quota limit for %llu\n",
1106 (unsigned long long)qgroupid);
1107 }
1108
1109 spin_lock(&fs_info->qgroup_lock);
1110
1111 qgroup = find_qgroup_rb(fs_info, qgroupid);
1112 if (!qgroup) {
1113 ret = -ENOENT;
1114 goto unlock;
1115 }
1116 qgroup->lim_flags = limit->flags;
1117 qgroup->max_rfer = limit->max_rfer;
1118 qgroup->max_excl = limit->max_excl;
1119 qgroup->rsv_rfer = limit->rsv_rfer;
1120 qgroup->rsv_excl = limit->rsv_excl;
1121
1122unlock:
1123 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1124out:
1125 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1126 return ret;
1127}
1128
1129static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1130 struct btrfs_qgroup *qgroup)
1131{
1132 if (list_empty(&qgroup->dirty))
1133 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1134}
1135
1136/*
1137 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1138 * the modification into a list that's later used by btrfs_end_transaction to
1139 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1140 */
1141int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1142 struct btrfs_delayed_ref_node *node,
1143 struct btrfs_delayed_extent_op *extent_op)
1144{
1145 struct qgroup_update *u;
1146
1147 BUG_ON(!trans->delayed_ref_elem.seq);
1148 u = kmalloc(sizeof(*u), GFP_NOFS);
1149 if (!u)
1150 return -ENOMEM;
1151
1152 u->node = node;
1153 u->extent_op = extent_op;
1154 list_add_tail(&u->list, &trans->qgroup_ref_list);
1155
1156 return 0;
1157}
1158
1159/*
1160 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1161 * from the fs. First, all roots referencing the extent are searched, and
1162 * then the space is accounted accordingly to the different roots. The
1163 * accounting algorithm works in 3 steps documented inline.
1164 */
1165int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1166 struct btrfs_fs_info *fs_info,
1167 struct btrfs_delayed_ref_node *node,
1168 struct btrfs_delayed_extent_op *extent_op)
1169{
1170 struct btrfs_key ins;
1171 struct btrfs_root *quota_root;
1172 u64 ref_root;
1173 struct btrfs_qgroup *qgroup;
1174 struct ulist_node *unode;
1175 struct ulist *roots = NULL;
1176 struct ulist *tmp = NULL;
1177 struct ulist_iterator uiter;
1178 u64 seq;
1179 int ret = 0;
1180 int sgn;
1181
1182 if (!fs_info->quota_enabled)
1183 return 0;
1184
1185 BUG_ON(!fs_info->quota_root);
1186
1187 ins.objectid = node->bytenr;
1188 ins.offset = node->num_bytes;
1189 ins.type = BTRFS_EXTENT_ITEM_KEY;
1190
1191 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1192 node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1193 struct btrfs_delayed_tree_ref *ref;
1194 ref = btrfs_delayed_node_to_tree_ref(node);
1195 ref_root = ref->root;
1196 } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1197 node->type == BTRFS_SHARED_DATA_REF_KEY) {
1198 struct btrfs_delayed_data_ref *ref;
1199 ref = btrfs_delayed_node_to_data_ref(node);
1200 ref_root = ref->root;
1201 } else {
1202 BUG();
1203 }
1204
1205 if (!is_fstree(ref_root)) {
1206 /*
1207 * non-fs-trees are not being accounted
1208 */
1209 return 0;
1210 }
1211
1212 switch (node->action) {
1213 case BTRFS_ADD_DELAYED_REF:
1214 case BTRFS_ADD_DELAYED_EXTENT:
1215 sgn = 1;
1216 break;
1217 case BTRFS_DROP_DELAYED_REF:
1218 sgn = -1;
1219 break;
1220 case BTRFS_UPDATE_DELAYED_HEAD:
1221 return 0;
1222 default:
1223 BUG();
1224 }
1225
1226 /*
1227 * the delayed ref sequence number we pass depends on the direction of
1228 * the operation. for add operations, we pass (node->seq - 1) to skip
1229 * the delayed ref's current sequence number, because we need the state
1230 * of the tree before the add operation. for delete operations, we pass
1231 * (node->seq) to include the delayed ref's current sequence number,
1232 * because we need the state of the tree after the delete operation.
1233 */
1234 ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1235 sgn > 0 ? node->seq - 1 : node->seq, &roots);
1236 if (ret < 0)
a7975026 1237 return ret;
bed92eae
AJ		 1238
1239 spin_lock(&fs_info->qgroup_lock);
1240 quota_root = fs_info->quota_root;
1241 if (!quota_root)
1242 goto unlock;
1243
1244 qgroup = find_qgroup_rb(fs_info, ref_root);
1245 if (!qgroup)
1246 goto unlock;
1247
1248 /*
1249 * step 1: for each old ref, visit all nodes once and inc refcnt
1250 */
1251 tmp = ulist_alloc(GFP_ATOMIC);
1252 if (!tmp) {
1253 ret = -ENOMEM;
1254 goto unlock;
1255 }
1256 seq = fs_info->qgroup_seq;
1257 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1258
1259 ULIST_ITER_INIT(&uiter);
1260 while ((unode = ulist_next(roots, &uiter))) {
1261 struct ulist_node *tmp_unode;
1262 struct ulist_iterator tmp_uiter;
1263 struct btrfs_qgroup *qg;
1264
1265 qg = find_qgroup_rb(fs_info, unode->val);
1266 if (!qg)
1267 continue;
1268
1269 ulist_reinit(tmp);
1270 /* XXX id not needed */
995e01b7 1271 ulist_add(tmp, qg->qgroupid, (u64)(uintptr_t)qg, GFP_ATOMIC);
bed92eae
AJ		 1272 ULIST_ITER_INIT(&tmp_uiter);
1273 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1274 struct btrfs_qgroup_list *glist;
1275
995e01b7 1276 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
bed92eae
AJ		 1277 if (qg->refcnt < seq)
1278 qg->refcnt = seq + 1;
1279 else
1280 ++qg->refcnt;
1281
1282 list_for_each_entry(glist, &qg->groups, next_group) {
1283 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1284 (u64)(uintptr_t)glist->group,
bed92eae
AJ		 1285 GFP_ATOMIC);
1286 }
1287 }
1288 }
1289
1290 /*
1291 * step 2: walk from the new root
1292 */
1293 ulist_reinit(tmp);
995e01b7 1294 ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1295 ULIST_ITER_INIT(&uiter);
1296 while ((unode = ulist_next(tmp, &uiter))) {
1297 struct btrfs_qgroup *qg;
1298 struct btrfs_qgroup_list *glist;
1299
995e01b7 1300 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1301 if (qg->refcnt < seq) {
1302 /* not visited by step 1 */
1303 qg->rfer += sgn * node->num_bytes;
1304 qg->rfer_cmpr += sgn * node->num_bytes;
1305 if (roots->nnodes == 0) {
1306 qg->excl += sgn * node->num_bytes;
1307 qg->excl_cmpr += sgn * node->num_bytes;
1308 }
1309 qgroup_dirty(fs_info, qg);
1310 }
1311 WARN_ON(qg->tag >= seq);
1312 qg->tag = seq;
1313
1314 list_for_each_entry(glist, &qg->groups, next_group) {
1315 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1316 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1317 }
1318 }
1319
1320 /*
1321 * step 3: walk again from old refs
1322 */
1323 ULIST_ITER_INIT(&uiter);
1324 while ((unode = ulist_next(roots, &uiter))) {
1325 struct btrfs_qgroup *qg;
1326 struct ulist_node *tmp_unode;
1327 struct ulist_iterator tmp_uiter;
1328
1329 qg = find_qgroup_rb(fs_info, unode->val);
1330 if (!qg)
1331 continue;
1332
1333 ulist_reinit(tmp);
995e01b7 1334 ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
bed92eae
AJ		 1335 ULIST_ITER_INIT(&tmp_uiter);
1336 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1337 struct btrfs_qgroup_list *glist;
1338
995e01b7 1339 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
bed92eae
AJ		 1340 if (qg->tag == seq)
1341 continue;
1342
1343 if (qg->refcnt - seq == roots->nnodes) {
1344 qg->excl -= sgn * node->num_bytes;
1345 qg->excl_cmpr -= sgn * node->num_bytes;
1346 qgroup_dirty(fs_info, qg);
1347 }
1348
1349 list_for_each_entry(glist, &qg->groups, next_group) {
1350 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1351 (uintptr_t)glist->group,
bed92eae
AJ		 1352 GFP_ATOMIC);
1353 }
1354 }
1355 }
1356 ret = 0;
1357unlock:
1358 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 1359 ulist_free(roots);
1360 ulist_free(tmp);
1361
1362 return ret;
1363}
1364
1365/*
1366 * called from commit_transaction. Writes all changed qgroups to disk.
1367 */
1368int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1369 struct btrfs_fs_info *fs_info)
1370{
1371 struct btrfs_root *quota_root = fs_info->quota_root;
1372 int ret = 0;
1373
1374 if (!quota_root)
1375 goto out;
1376
1377 fs_info->quota_enabled = fs_info->pending_quota_state;
1378
1379 spin_lock(&fs_info->qgroup_lock);
1380 while (!list_empty(&fs_info->dirty_qgroups)) {
1381 struct btrfs_qgroup *qgroup;
1382 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1383 struct btrfs_qgroup, dirty);
1384 list_del_init(&qgroup->dirty);
1385 spin_unlock(&fs_info->qgroup_lock);
1386 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1387 if (ret)
1388 fs_info->qgroup_flags |=
1389 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1390 spin_lock(&fs_info->qgroup_lock);
1391 }
1392 if (fs_info->quota_enabled)
1393 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1394 else
1395 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1396 spin_unlock(&fs_info->qgroup_lock);
1397
1398 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1399 if (ret)
1400 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1401
1402out:
1403
1404 return ret;
1405}
1406
1407/*
1408 * copy the acounting information between qgroups. This is necessary when a
1409 * snapshot or a subvolume is created
1410 */
1411int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1412 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1413 struct btrfs_qgroup_inherit *inherit)
1414{
1415 int ret = 0;
1416 int i;
1417 u64 *i_qgroups;
1418 struct btrfs_root *quota_root = fs_info->quota_root;
1419 struct btrfs_qgroup *srcgroup;
1420 struct btrfs_qgroup *dstgroup;
1421 u32 level_size = 0;
1422
f2f6ed3d 1423 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1424 if (!fs_info->quota_enabled)
f2f6ed3d 1425 goto out;
bed92eae 1426
f2f6ed3d
WS		 1427 if (!quota_root) {
1428 ret = -EINVAL;
1429 goto out;
1430 }
bed92eae
AJ		 1431
1432 /*
1433 * create a tracking group for the subvol itself
1434 */
1435 ret = add_qgroup_item(trans, quota_root, objectid);
1436 if (ret)
1437 goto out;
1438
1439 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1440 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1441 inherit->lim.flags,
1442 inherit->lim.max_rfer,
1443 inherit->lim.max_excl,
1444 inherit->lim.rsv_rfer,
1445 inherit->lim.rsv_excl);
1446 if (ret)
1447 goto out;
1448 }
1449
1450 if (srcid) {
1451 struct btrfs_root *srcroot;
1452 struct btrfs_key srckey;
1453 int srcroot_level;
1454
1455 srckey.objectid = srcid;
1456 srckey.type = BTRFS_ROOT_ITEM_KEY;
1457 srckey.offset = (u64)-1;
1458 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1459 if (IS_ERR(srcroot)) {
1460 ret = PTR_ERR(srcroot);
1461 goto out;
1462 }
1463
1464 rcu_read_lock();
1465 srcroot_level = btrfs_header_level(srcroot->node);
1466 level_size = btrfs_level_size(srcroot, srcroot_level);
1467 rcu_read_unlock();
1468 }
1469
1470 /*
1471 * add qgroup to all inherited groups
1472 */
1473 if (inherit) {
1474 i_qgroups = (u64 *)(inherit + 1);
1475 for (i = 0; i < inherit->num_qgroups; ++i) {
1476 ret = add_qgroup_relation_item(trans, quota_root,
1477 objectid, *i_qgroups);
1478 if (ret)
1479 goto out;
1480 ret = add_qgroup_relation_item(trans, quota_root,
1481 *i_qgroups, objectid);
1482 if (ret)
1483 goto out;
1484 ++i_qgroups;
1485 }
1486 }
1487
1488
1489 spin_lock(&fs_info->qgroup_lock);
1490
1491 dstgroup = add_qgroup_rb(fs_info, objectid);
57a5a882
DC		 1492 if (IS_ERR(dstgroup)) {
1493 ret = PTR_ERR(dstgroup);
bed92eae 1494 goto unlock;
57a5a882 1495 }
bed92eae
AJ		 1496
1497 if (srcid) {
1498 srcgroup = find_qgroup_rb(fs_info, srcid);
f3a87f1b 1499 if (!srcgroup)
bed92eae
AJ		 1500 goto unlock;
1501 dstgroup->rfer = srcgroup->rfer - level_size;
1502 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1503 srcgroup->excl = level_size;
1504 srcgroup->excl_cmpr = level_size;
1505 qgroup_dirty(fs_info, dstgroup);
1506 qgroup_dirty(fs_info, srcgroup);
1507 }
1508
f3a87f1b 1509 if (!inherit)
bed92eae
AJ		 1510 goto unlock;
1511
1512 i_qgroups = (u64 *)(inherit + 1);
1513 for (i = 0; i < inherit->num_qgroups; ++i) {
1514 ret = add_relation_rb(quota_root->fs_info, objectid,
1515 *i_qgroups);
1516 if (ret)
1517 goto unlock;
1518 ++i_qgroups;
1519 }
1520
1521 for (i = 0; i < inherit->num_ref_copies; ++i) {
1522 struct btrfs_qgroup *src;
1523 struct btrfs_qgroup *dst;
1524
1525 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1526 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1527
1528 if (!src || !dst) {
1529 ret = -EINVAL;
1530 goto unlock;
1531 }
1532
1533 dst->rfer = src->rfer - level_size;
1534 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1535 i_qgroups += 2;
1536 }
1537 for (i = 0; i < inherit->num_excl_copies; ++i) {
1538 struct btrfs_qgroup *src;
1539 struct btrfs_qgroup *dst;
1540
1541 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1542 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1543
1544 if (!src || !dst) {
1545 ret = -EINVAL;
1546 goto unlock;
1547 }
1548
1549 dst->excl = src->excl + level_size;
1550 dst->excl_cmpr = src->excl_cmpr + level_size;
1551 i_qgroups += 2;
1552 }
1553
1554unlock:
1555 spin_unlock(&fs_info->qgroup_lock);
1556out:
f2f6ed3d 1557 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1558 return ret;
1559}
1560
1561/*
1562 * reserve some space for a qgroup and all its parents. The reservation takes
1563 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1564 * accounting. If not enough space is available, EDQUOT is returned.
1565 * We assume that the requested space is new for all qgroups.
1566 */
1567int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1568{
1569 struct btrfs_root *quota_root;
1570 struct btrfs_qgroup *qgroup;
1571 struct btrfs_fs_info *fs_info = root->fs_info;
1572 u64 ref_root = root->root_key.objectid;
1573 int ret = 0;
1574 struct ulist *ulist = NULL;
1575 struct ulist_node *unode;
1576 struct ulist_iterator uiter;
1577
1578 if (!is_fstree(ref_root))
1579 return 0;
1580
1581 if (num_bytes == 0)
1582 return 0;
1583
1584 spin_lock(&fs_info->qgroup_lock);
1585 quota_root = fs_info->quota_root;
1586 if (!quota_root)
1587 goto out;
1588
1589 qgroup = find_qgroup_rb(fs_info, ref_root);
1590 if (!qgroup)
1591 goto out;
1592
1593 /*
1594 * in a first step, we check all affected qgroups if any limits would
1595 * be exceeded
1596 */
1597 ulist = ulist_alloc(GFP_ATOMIC);
3d6b5c3b
TI		 1598 if (!ulist) {
1599 ret = -ENOMEM;
1600 goto out;
1601 }
995e01b7 1602 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1603 ULIST_ITER_INIT(&uiter);
1604 while ((unode = ulist_next(ulist, &uiter))) {
1605 struct btrfs_qgroup *qg;
1606 struct btrfs_qgroup_list *glist;
1607
995e01b7 1608 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1609
1610 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1611 qg->reserved + qg->rfer + num_bytes >
720f1e20 1612 qg->max_rfer) {
bed92eae 1613 ret = -EDQUOT;
720f1e20
WS		 1614 goto out;
1615 }
bed92eae
AJ		 1616
1617 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1618 qg->reserved + qg->excl + num_bytes >
720f1e20 1619 qg->max_excl) {
bed92eae 1620 ret = -EDQUOT;
720f1e20
WS		 1621 goto out;
1622 }
bed92eae
AJ		 1623
1624 list_for_each_entry(glist, &qg->groups, next_group) {
1625 ulist_add(ulist, glist->group->qgroupid,
995e01b7 1626 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1627 }
1628 }
bed92eae
AJ		 1629
1630 /*
1631 * no limits exceeded, now record the reservation into all qgroups
1632 */
1633 ULIST_ITER_INIT(&uiter);
1634 while ((unode = ulist_next(ulist, &uiter))) {
1635 struct btrfs_qgroup *qg;
1636
995e01b7 1637 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1638
1639 qg->reserved += num_bytes;
1640 }
1641
1642out:
1643 spin_unlock(&fs_info->qgroup_lock);
1644 ulist_free(ulist);
1645
1646 return ret;
1647}
1648
1649void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1650{
1651 struct btrfs_root *quota_root;
1652 struct btrfs_qgroup *qgroup;
1653 struct btrfs_fs_info *fs_info = root->fs_info;
1654 struct ulist *ulist = NULL;
1655 struct ulist_node *unode;
1656 struct ulist_iterator uiter;
1657 u64 ref_root = root->root_key.objectid;
1658
1659 if (!is_fstree(ref_root))
1660 return;
1661
1662 if (num_bytes == 0)
1663 return;
1664
1665 spin_lock(&fs_info->qgroup_lock);
1666
1667 quota_root = fs_info->quota_root;
1668 if (!quota_root)
1669 goto out;
1670
1671 qgroup = find_qgroup_rb(fs_info, ref_root);
1672 if (!qgroup)
1673 goto out;
1674
1675 ulist = ulist_alloc(GFP_ATOMIC);
3d6b5c3b
TI		 1676 if (!ulist) {
1677 btrfs_std_error(fs_info, -ENOMEM);
1678 goto out;
1679 }
995e01b7 1680 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1681 ULIST_ITER_INIT(&uiter);
1682 while ((unode = ulist_next(ulist, &uiter))) {
1683 struct btrfs_qgroup *qg;
1684 struct btrfs_qgroup_list *glist;
1685
995e01b7 1686 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1687
1688 qg->reserved -= num_bytes;
1689
1690 list_for_each_entry(glist, &qg->groups, next_group) {
1691 ulist_add(ulist, glist->group->qgroupid,
995e01b7 1692 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1693 }
1694 }
1695
1696out:
1697 spin_unlock(&fs_info->qgroup_lock);
1698 ulist_free(ulist);
1699}
1700
1701void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1702{
1703 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1704 return;
1705 printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1706 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1707 trans->delayed_ref_elem.seq);
1708 BUG();
1709}
Linux 6.x block layer and io_uring tree(s)