projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: creating the subvolume qgroup automatically when enabling quota
[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
AJ		 793
794 spin_lock(&fs_info->qgroup_lock);
795 if (fs_info->quota_root) {
796 fs_info->pending_quota_state = 1;
797 spin_unlock(&fs_info->qgroup_lock);
798 goto out;
799 }
800 spin_unlock(&fs_info->qgroup_lock);
801
802 /*
803 * initially create the quota tree
804 */
805 quota_root = btrfs_create_tree(trans, fs_info,
806 BTRFS_QUOTA_TREE_OBJECTID);
807 if (IS_ERR(quota_root)) {
808 ret = PTR_ERR(quota_root);
809 goto out;
810 }
811
812 path = btrfs_alloc_path();
5b7ff5b3
TI		 813 if (!path) {
814 ret = -ENOMEM;
815 goto out_free_root;
816 }
bed92eae
AJ		 817
818 key.objectid = 0;
819 key.type = BTRFS_QGROUP_STATUS_KEY;
820 key.offset = 0;
821
822 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
823 sizeof(*ptr));
824 if (ret)
5b7ff5b3 825 goto out_free_path;
bed92eae
AJ		 826
827 leaf = path->nodes[0];
828 ptr = btrfs_item_ptr(leaf, path->slots[0],
829 struct btrfs_qgroup_status_item);
830 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
831 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
832 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
833 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
834 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
835 btrfs_set_qgroup_status_scan(leaf, ptr, 0);
836
837 btrfs_mark_buffer_dirty(leaf);
838
7708f029
WS		 839 key.objectid = 0;
840 key.type = BTRFS_ROOT_REF_KEY;
841 key.offset = 0;
842
843 btrfs_release_path(path);
844 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
845 if (ret > 0)
846 goto out_add_root;
847 if (ret < 0)
848 goto out_free_path;
849
850
851 while (1) {
852 slot = path->slots[0];
853 leaf = path->nodes[0];
854 btrfs_item_key_to_cpu(leaf, &found_key, slot);
855
856 if (found_key.type == BTRFS_ROOT_REF_KEY) {
857 ret = add_qgroup_item(trans, quota_root,
858 found_key.offset);
859 if (ret)
860 goto out_free_path;
861
862 spin_lock(&fs_info->qgroup_lock);
863 qgroup = add_qgroup_rb(fs_info, found_key.offset);
864 if (IS_ERR(qgroup)) {
865 spin_unlock(&fs_info->qgroup_lock);
866 ret = PTR_ERR(qgroup);
867 goto out_free_path;
868 }
869 spin_unlock(&fs_info->qgroup_lock);
870 }
871 ret = btrfs_next_item(tree_root, path);
872 if (ret < 0)
873 goto out_free_path;
874 if (ret)
875 break;
876 }
877
878out_add_root:
879 btrfs_release_path(path);
880 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
881 if (ret)
882 goto out_free_path;
883
bed92eae 884 spin_lock(&fs_info->qgroup_lock);
7708f029
WS		 885 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
886 if (IS_ERR(qgroup)) {
887 spin_unlock(&fs_info->qgroup_lock);
888 ret = PTR_ERR(qgroup);
889 goto out_free_path;
890 }
bed92eae
AJ		 891 fs_info->quota_root = quota_root;
892 fs_info->pending_quota_state = 1;
893 spin_unlock(&fs_info->qgroup_lock);
5b7ff5b3 894out_free_path:
bed92eae 895 btrfs_free_path(path);
5b7ff5b3
TI		 896out_free_root:
897 if (ret) {
898 free_extent_buffer(quota_root->node);
899 free_extent_buffer(quota_root->commit_root);
900 kfree(quota_root);
901 }
902out:
bed92eae
AJ		 903 return ret;
904}
905
906int btrfs_quota_disable(struct btrfs_trans_handle *trans,
907 struct btrfs_fs_info *fs_info)
908{
909 struct btrfs_root *tree_root = fs_info->tree_root;
910 struct btrfs_root *quota_root;
911 int ret = 0;
912
913 spin_lock(&fs_info->qgroup_lock);
683cebda
WS		 914 if (!fs_info->quota_root) {
915 spin_unlock(&fs_info->qgroup_lock);
916 return 0;
917 }
bed92eae
AJ		 918 fs_info->quota_enabled = 0;
919 fs_info->pending_quota_state = 0;
920 quota_root = fs_info->quota_root;
921 fs_info->quota_root = NULL;
922 btrfs_free_qgroup_config(fs_info);
923 spin_unlock(&fs_info->qgroup_lock);
924
925 if (!quota_root)
926 return -EINVAL;
927
928 ret = btrfs_clean_quota_tree(trans, quota_root);
929 if (ret)
930 goto out;
931
932 ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
933 if (ret)
934 goto out;
935
936 list_del(&quota_root->dirty_list);
937
938 btrfs_tree_lock(quota_root->node);
939 clean_tree_block(trans, tree_root, quota_root->node);
940 btrfs_tree_unlock(quota_root->node);
941 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
942
943 free_extent_buffer(quota_root->node);
944 free_extent_buffer(quota_root->commit_root);
945 kfree(quota_root);
946out:
947 return ret;
948}
949
950int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
951{
952 /* FIXME */
953 return 0;
954}
955
956int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
957 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
958{
959 struct btrfs_root *quota_root;
960 int ret = 0;
961
962 quota_root = fs_info->quota_root;
963 if (!quota_root)
964 return -EINVAL;
965
966 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
967 if (ret)
968 return ret;
969
970 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
971 if (ret) {
972 del_qgroup_relation_item(trans, quota_root, src, dst);
973 return ret;
974 }
975
976 spin_lock(&fs_info->qgroup_lock);
977 ret = add_relation_rb(quota_root->fs_info, src, dst);
978 spin_unlock(&fs_info->qgroup_lock);
979
980 return ret;
981}
982
983int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
984 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
985{
986 struct btrfs_root *quota_root;
987 int ret = 0;
988 int err;
989
990 quota_root = fs_info->quota_root;
991 if (!quota_root)
992 return -EINVAL;
993
994 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
995 err = del_qgroup_relation_item(trans, quota_root, dst, src);
996 if (err && !ret)
997 ret = err;
998
999 spin_lock(&fs_info->qgroup_lock);
1000 del_relation_rb(fs_info, src, dst);
1001
1002 spin_unlock(&fs_info->qgroup_lock);
1003
1004 return ret;
1005}
1006
1007int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1008 struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
1009{
1010 struct btrfs_root *quota_root;
1011 struct btrfs_qgroup *qgroup;
1012 int ret = 0;
1013
1014 quota_root = fs_info->quota_root;
1015 if (!quota_root)
1016 return -EINVAL;
1017
1018 ret = add_qgroup_item(trans, quota_root, qgroupid);
1019
1020 spin_lock(&fs_info->qgroup_lock);
1021 qgroup = add_qgroup_rb(fs_info, qgroupid);
1022 spin_unlock(&fs_info->qgroup_lock);
1023
1024 if (IS_ERR(qgroup))
1025 ret = PTR_ERR(qgroup);
1026
1027 return ret;
1028}
1029
1030int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1031 struct btrfs_fs_info *fs_info, u64 qgroupid)
1032{
1033 struct btrfs_root *quota_root;
2cf68703 1034 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 1035 int ret = 0;
1036
1037 quota_root = fs_info->quota_root;
1038 if (!quota_root)
1039 return -EINVAL;
1040
2cf68703
AJ		 1041 /* check if there are no relations to this qgroup */
1042 spin_lock(&fs_info->qgroup_lock);
1043 qgroup = find_qgroup_rb(fs_info, qgroupid);
1044 if (qgroup) {
1045 if (!list_empty(&qgroup->groups) || !list_empty(&qgroup->members)) {
1046 spin_unlock(&fs_info->qgroup_lock);
1047 return -EBUSY;
1048 }
1049 }
1050 spin_unlock(&fs_info->qgroup_lock);
1051
bed92eae
AJ		 1052 ret = del_qgroup_item(trans, quota_root, qgroupid);
1053
1054 spin_lock(&fs_info->qgroup_lock);
1055 del_qgroup_rb(quota_root->fs_info, qgroupid);
bed92eae
AJ		 1056 spin_unlock(&fs_info->qgroup_lock);
1057
1058 return ret;
1059}
1060
1061int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1062 struct btrfs_fs_info *fs_info, u64 qgroupid,
1063 struct btrfs_qgroup_limit *limit)
1064{
1065 struct btrfs_root *quota_root = fs_info->quota_root;
1066 struct btrfs_qgroup *qgroup;
1067 int ret = 0;
1068
1069 if (!quota_root)
1070 return -EINVAL;
1071
1072 ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1073 limit->flags, limit->max_rfer,
1074 limit->max_excl, limit->rsv_rfer,
1075 limit->rsv_excl);
1076 if (ret) {
1077 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1078 printk(KERN_INFO "unable to update quota limit for %llu\n",
1079 (unsigned long long)qgroupid);
1080 }
1081
1082 spin_lock(&fs_info->qgroup_lock);
1083
1084 qgroup = find_qgroup_rb(fs_info, qgroupid);
1085 if (!qgroup) {
1086 ret = -ENOENT;
1087 goto unlock;
1088 }
1089 qgroup->lim_flags = limit->flags;
1090 qgroup->max_rfer = limit->max_rfer;
1091 qgroup->max_excl = limit->max_excl;
1092 qgroup->rsv_rfer = limit->rsv_rfer;
1093 qgroup->rsv_excl = limit->rsv_excl;
1094
1095unlock:
1096 spin_unlock(&fs_info->qgroup_lock);
1097
1098 return ret;
1099}
1100
1101static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1102 struct btrfs_qgroup *qgroup)
1103{
1104 if (list_empty(&qgroup->dirty))
1105 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1106}
1107
1108/*
1109 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1110 * the modification into a list that's later used by btrfs_end_transaction to
1111 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1112 */
1113int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1114 struct btrfs_delayed_ref_node *node,
1115 struct btrfs_delayed_extent_op *extent_op)
1116{
1117 struct qgroup_update *u;
1118
1119 BUG_ON(!trans->delayed_ref_elem.seq);
1120 u = kmalloc(sizeof(*u), GFP_NOFS);
1121 if (!u)
1122 return -ENOMEM;
1123
1124 u->node = node;
1125 u->extent_op = extent_op;
1126 list_add_tail(&u->list, &trans->qgroup_ref_list);
1127
1128 return 0;
1129}
1130
1131/*
1132 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1133 * from the fs. First, all roots referencing the extent are searched, and
1134 * then the space is accounted accordingly to the different roots. The
1135 * accounting algorithm works in 3 steps documented inline.
1136 */
1137int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1138 struct btrfs_fs_info *fs_info,
1139 struct btrfs_delayed_ref_node *node,
1140 struct btrfs_delayed_extent_op *extent_op)
1141{
1142 struct btrfs_key ins;
1143 struct btrfs_root *quota_root;
1144 u64 ref_root;
1145 struct btrfs_qgroup *qgroup;
1146 struct ulist_node *unode;
1147 struct ulist *roots = NULL;
1148 struct ulist *tmp = NULL;
1149 struct ulist_iterator uiter;
1150 u64 seq;
1151 int ret = 0;
1152 int sgn;
1153
1154 if (!fs_info->quota_enabled)
1155 return 0;
1156
1157 BUG_ON(!fs_info->quota_root);
1158
1159 ins.objectid = node->bytenr;
1160 ins.offset = node->num_bytes;
1161 ins.type = BTRFS_EXTENT_ITEM_KEY;
1162
1163 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1164 node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1165 struct btrfs_delayed_tree_ref *ref;
1166 ref = btrfs_delayed_node_to_tree_ref(node);
1167 ref_root = ref->root;
1168 } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1169 node->type == BTRFS_SHARED_DATA_REF_KEY) {
1170 struct btrfs_delayed_data_ref *ref;
1171 ref = btrfs_delayed_node_to_data_ref(node);
1172 ref_root = ref->root;
1173 } else {
1174 BUG();
1175 }
1176
1177 if (!is_fstree(ref_root)) {
1178 /*
1179 * non-fs-trees are not being accounted
1180 */
1181 return 0;
1182 }
1183
1184 switch (node->action) {
1185 case BTRFS_ADD_DELAYED_REF:
1186 case BTRFS_ADD_DELAYED_EXTENT:
1187 sgn = 1;
1188 break;
1189 case BTRFS_DROP_DELAYED_REF:
1190 sgn = -1;
1191 break;
1192 case BTRFS_UPDATE_DELAYED_HEAD:
1193 return 0;
1194 default:
1195 BUG();
1196 }
1197
1198 /*
1199 * the delayed ref sequence number we pass depends on the direction of
1200 * the operation. for add operations, we pass (node->seq - 1) to skip
1201 * the delayed ref's current sequence number, because we need the state
1202 * of the tree before the add operation. for delete operations, we pass
1203 * (node->seq) to include the delayed ref's current sequence number,
1204 * because we need the state of the tree after the delete operation.
1205 */
1206 ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1207 sgn > 0 ? node->seq - 1 : node->seq, &roots);
1208 if (ret < 0)
a7975026 1209 return ret;
bed92eae
AJ		 1210
1211 spin_lock(&fs_info->qgroup_lock);
1212 quota_root = fs_info->quota_root;
1213 if (!quota_root)
1214 goto unlock;
1215
1216 qgroup = find_qgroup_rb(fs_info, ref_root);
1217 if (!qgroup)
1218 goto unlock;
1219
1220 /*
1221 * step 1: for each old ref, visit all nodes once and inc refcnt
1222 */
1223 tmp = ulist_alloc(GFP_ATOMIC);
1224 if (!tmp) {
1225 ret = -ENOMEM;
1226 goto unlock;
1227 }
1228 seq = fs_info->qgroup_seq;
1229 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1230
1231 ULIST_ITER_INIT(&uiter);
1232 while ((unode = ulist_next(roots, &uiter))) {
1233 struct ulist_node *tmp_unode;
1234 struct ulist_iterator tmp_uiter;
1235 struct btrfs_qgroup *qg;
1236
1237 qg = find_qgroup_rb(fs_info, unode->val);
1238 if (!qg)
1239 continue;
1240
1241 ulist_reinit(tmp);
1242 /* XXX id not needed */
995e01b7 1243 ulist_add(tmp, qg->qgroupid, (u64)(uintptr_t)qg, GFP_ATOMIC);
bed92eae
AJ		 1244 ULIST_ITER_INIT(&tmp_uiter);
1245 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1246 struct btrfs_qgroup_list *glist;
1247
995e01b7 1248 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
bed92eae
AJ		 1249 if (qg->refcnt < seq)
1250 qg->refcnt = seq + 1;
1251 else
1252 ++qg->refcnt;
1253
1254 list_for_each_entry(glist, &qg->groups, next_group) {
1255 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1256 (u64)(uintptr_t)glist->group,
bed92eae
AJ		 1257 GFP_ATOMIC);
1258 }
1259 }
1260 }
1261
1262 /*
1263 * step 2: walk from the new root
1264 */
1265 ulist_reinit(tmp);
995e01b7 1266 ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1267 ULIST_ITER_INIT(&uiter);
1268 while ((unode = ulist_next(tmp, &uiter))) {
1269 struct btrfs_qgroup *qg;
1270 struct btrfs_qgroup_list *glist;
1271
995e01b7 1272 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1273 if (qg->refcnt < seq) {
1274 /* not visited by step 1 */
1275 qg->rfer += sgn * node->num_bytes;
1276 qg->rfer_cmpr += sgn * node->num_bytes;
1277 if (roots->nnodes == 0) {
1278 qg->excl += sgn * node->num_bytes;
1279 qg->excl_cmpr += sgn * node->num_bytes;
1280 }
1281 qgroup_dirty(fs_info, qg);
1282 }
1283 WARN_ON(qg->tag >= seq);
1284 qg->tag = seq;
1285
1286 list_for_each_entry(glist, &qg->groups, next_group) {
1287 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1288 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1289 }
1290 }
1291
1292 /*
1293 * step 3: walk again from old refs
1294 */
1295 ULIST_ITER_INIT(&uiter);
1296 while ((unode = ulist_next(roots, &uiter))) {
1297 struct btrfs_qgroup *qg;
1298 struct ulist_node *tmp_unode;
1299 struct ulist_iterator tmp_uiter;
1300
1301 qg = find_qgroup_rb(fs_info, unode->val);
1302 if (!qg)
1303 continue;
1304
1305 ulist_reinit(tmp);
995e01b7 1306 ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
bed92eae
AJ		 1307 ULIST_ITER_INIT(&tmp_uiter);
1308 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1309 struct btrfs_qgroup_list *glist;
1310
995e01b7 1311 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
bed92eae
AJ		 1312 if (qg->tag == seq)
1313 continue;
1314
1315 if (qg->refcnt - seq == roots->nnodes) {
1316 qg->excl -= sgn * node->num_bytes;
1317 qg->excl_cmpr -= sgn * node->num_bytes;
1318 qgroup_dirty(fs_info, qg);
1319 }
1320
1321 list_for_each_entry(glist, &qg->groups, next_group) {
1322 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1323 (uintptr_t)glist->group,
bed92eae
AJ		 1324 GFP_ATOMIC);
1325 }
1326 }
1327 }
1328 ret = 0;
1329unlock:
1330 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 1331 ulist_free(roots);
1332 ulist_free(tmp);
1333
1334 return ret;
1335}
1336
1337/*
1338 * called from commit_transaction. Writes all changed qgroups to disk.
1339 */
1340int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1341 struct btrfs_fs_info *fs_info)
1342{
1343 struct btrfs_root *quota_root = fs_info->quota_root;
1344 int ret = 0;
1345
1346 if (!quota_root)
1347 goto out;
1348
1349 fs_info->quota_enabled = fs_info->pending_quota_state;
1350
1351 spin_lock(&fs_info->qgroup_lock);
1352 while (!list_empty(&fs_info->dirty_qgroups)) {
1353 struct btrfs_qgroup *qgroup;
1354 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1355 struct btrfs_qgroup, dirty);
1356 list_del_init(&qgroup->dirty);
1357 spin_unlock(&fs_info->qgroup_lock);
1358 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1359 if (ret)
1360 fs_info->qgroup_flags |=
1361 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1362 spin_lock(&fs_info->qgroup_lock);
1363 }
1364 if (fs_info->quota_enabled)
1365 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1366 else
1367 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1368 spin_unlock(&fs_info->qgroup_lock);
1369
1370 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1371 if (ret)
1372 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1373
1374out:
1375
1376 return ret;
1377}
1378
1379/*
1380 * copy the acounting information between qgroups. This is necessary when a
1381 * snapshot or a subvolume is created
1382 */
1383int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1384 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1385 struct btrfs_qgroup_inherit *inherit)
1386{
1387 int ret = 0;
1388 int i;
1389 u64 *i_qgroups;
1390 struct btrfs_root *quota_root = fs_info->quota_root;
1391 struct btrfs_qgroup *srcgroup;
1392 struct btrfs_qgroup *dstgroup;
1393 u32 level_size = 0;
1394
1395 if (!fs_info->quota_enabled)
1396 return 0;
1397
1398 if (!quota_root)
1399 return -EINVAL;
1400
1401 /*
1402 * create a tracking group for the subvol itself
1403 */
1404 ret = add_qgroup_item(trans, quota_root, objectid);
1405 if (ret)
1406 goto out;
1407
1408 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1409 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1410 inherit->lim.flags,
1411 inherit->lim.max_rfer,
1412 inherit->lim.max_excl,
1413 inherit->lim.rsv_rfer,
1414 inherit->lim.rsv_excl);
1415 if (ret)
1416 goto out;
1417 }
1418
1419 if (srcid) {
1420 struct btrfs_root *srcroot;
1421 struct btrfs_key srckey;
1422 int srcroot_level;
1423
1424 srckey.objectid = srcid;
1425 srckey.type = BTRFS_ROOT_ITEM_KEY;
1426 srckey.offset = (u64)-1;
1427 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1428 if (IS_ERR(srcroot)) {
1429 ret = PTR_ERR(srcroot);
1430 goto out;
1431 }
1432
1433 rcu_read_lock();
1434 srcroot_level = btrfs_header_level(srcroot->node);
1435 level_size = btrfs_level_size(srcroot, srcroot_level);
1436 rcu_read_unlock();
1437 }
1438
1439 /*
1440 * add qgroup to all inherited groups
1441 */
1442 if (inherit) {
1443 i_qgroups = (u64 *)(inherit + 1);
1444 for (i = 0; i < inherit->num_qgroups; ++i) {
1445 ret = add_qgroup_relation_item(trans, quota_root,
1446 objectid, *i_qgroups);
1447 if (ret)
1448 goto out;
1449 ret = add_qgroup_relation_item(trans, quota_root,
1450 *i_qgroups, objectid);
1451 if (ret)
1452 goto out;
1453 ++i_qgroups;
1454 }
1455 }
1456
1457
1458 spin_lock(&fs_info->qgroup_lock);
1459
1460 dstgroup = add_qgroup_rb(fs_info, objectid);
57a5a882
DC		 1461 if (IS_ERR(dstgroup)) {
1462 ret = PTR_ERR(dstgroup);
bed92eae 1463 goto unlock;
57a5a882 1464 }
bed92eae
AJ		 1465
1466 if (srcid) {
1467 srcgroup = find_qgroup_rb(fs_info, srcid);
f3a87f1b 1468 if (!srcgroup)
bed92eae
AJ		 1469 goto unlock;
1470 dstgroup->rfer = srcgroup->rfer - level_size;
1471 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1472 srcgroup->excl = level_size;
1473 srcgroup->excl_cmpr = level_size;
1474 qgroup_dirty(fs_info, dstgroup);
1475 qgroup_dirty(fs_info, srcgroup);
1476 }
1477
f3a87f1b 1478 if (!inherit)
bed92eae
AJ		 1479 goto unlock;
1480
1481 i_qgroups = (u64 *)(inherit + 1);
1482 for (i = 0; i < inherit->num_qgroups; ++i) {
1483 ret = add_relation_rb(quota_root->fs_info, objectid,
1484 *i_qgroups);
1485 if (ret)
1486 goto unlock;
1487 ++i_qgroups;
1488 }
1489
1490 for (i = 0; i < inherit->num_ref_copies; ++i) {
1491 struct btrfs_qgroup *src;
1492 struct btrfs_qgroup *dst;
1493
1494 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1495 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1496
1497 if (!src || !dst) {
1498 ret = -EINVAL;
1499 goto unlock;
1500 }
1501
1502 dst->rfer = src->rfer - level_size;
1503 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1504 i_qgroups += 2;
1505 }
1506 for (i = 0; i < inherit->num_excl_copies; ++i) {
1507 struct btrfs_qgroup *src;
1508 struct btrfs_qgroup *dst;
1509
1510 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1511 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1512
1513 if (!src || !dst) {
1514 ret = -EINVAL;
1515 goto unlock;
1516 }
1517
1518 dst->excl = src->excl + level_size;
1519 dst->excl_cmpr = src->excl_cmpr + level_size;
1520 i_qgroups += 2;
1521 }
1522
1523unlock:
1524 spin_unlock(&fs_info->qgroup_lock);
1525out:
1526 return ret;
1527}
1528
1529/*
1530 * reserve some space for a qgroup and all its parents. The reservation takes
1531 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1532 * accounting. If not enough space is available, EDQUOT is returned.
1533 * We assume that the requested space is new for all qgroups.
1534 */
1535int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1536{
1537 struct btrfs_root *quota_root;
1538 struct btrfs_qgroup *qgroup;
1539 struct btrfs_fs_info *fs_info = root->fs_info;
1540 u64 ref_root = root->root_key.objectid;
1541 int ret = 0;
1542 struct ulist *ulist = NULL;
1543 struct ulist_node *unode;
1544 struct ulist_iterator uiter;
1545
1546 if (!is_fstree(ref_root))
1547 return 0;
1548
1549 if (num_bytes == 0)
1550 return 0;
1551
1552 spin_lock(&fs_info->qgroup_lock);
1553 quota_root = fs_info->quota_root;
1554 if (!quota_root)
1555 goto out;
1556
1557 qgroup = find_qgroup_rb(fs_info, ref_root);
1558 if (!qgroup)
1559 goto out;
1560
1561 /*
1562 * in a first step, we check all affected qgroups if any limits would
1563 * be exceeded
1564 */
1565 ulist = ulist_alloc(GFP_ATOMIC);
3d6b5c3b
TI		 1566 if (!ulist) {
1567 ret = -ENOMEM;
1568 goto out;
1569 }
995e01b7 1570 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1571 ULIST_ITER_INIT(&uiter);
1572 while ((unode = ulist_next(ulist, &uiter))) {
1573 struct btrfs_qgroup *qg;
1574 struct btrfs_qgroup_list *glist;
1575
995e01b7 1576 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1577
1578 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1579 qg->reserved + qg->rfer + num_bytes >
720f1e20 1580 qg->max_rfer) {
bed92eae 1581 ret = -EDQUOT;
720f1e20
WS		 1582 goto out;
1583 }
bed92eae
AJ		 1584
1585 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1586 qg->reserved + qg->excl + num_bytes >
720f1e20 1587 qg->max_excl) {
bed92eae 1588 ret = -EDQUOT;
720f1e20
WS		 1589 goto out;
1590 }
bed92eae
AJ		 1591
1592 list_for_each_entry(glist, &qg->groups, next_group) {
1593 ulist_add(ulist, glist->group->qgroupid,
995e01b7 1594 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1595 }
1596 }
bed92eae
AJ		 1597
1598 /*
1599 * no limits exceeded, now record the reservation into all qgroups
1600 */
1601 ULIST_ITER_INIT(&uiter);
1602 while ((unode = ulist_next(ulist, &uiter))) {
1603 struct btrfs_qgroup *qg;
1604
995e01b7 1605 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1606
1607 qg->reserved += num_bytes;
1608 }
1609
1610out:
1611 spin_unlock(&fs_info->qgroup_lock);
1612 ulist_free(ulist);
1613
1614 return ret;
1615}
1616
1617void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1618{
1619 struct btrfs_root *quota_root;
1620 struct btrfs_qgroup *qgroup;
1621 struct btrfs_fs_info *fs_info = root->fs_info;
1622 struct ulist *ulist = NULL;
1623 struct ulist_node *unode;
1624 struct ulist_iterator uiter;
1625 u64 ref_root = root->root_key.objectid;
1626
1627 if (!is_fstree(ref_root))
1628 return;
1629
1630 if (num_bytes == 0)
1631 return;
1632
1633 spin_lock(&fs_info->qgroup_lock);
1634
1635 quota_root = fs_info->quota_root;
1636 if (!quota_root)
1637 goto out;
1638
1639 qgroup = find_qgroup_rb(fs_info, ref_root);
1640 if (!qgroup)
1641 goto out;
1642
1643 ulist = ulist_alloc(GFP_ATOMIC);
3d6b5c3b
TI		 1644 if (!ulist) {
1645 btrfs_std_error(fs_info, -ENOMEM);
1646 goto out;
1647 }
995e01b7 1648 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1649 ULIST_ITER_INIT(&uiter);
1650 while ((unode = ulist_next(ulist, &uiter))) {
1651 struct btrfs_qgroup *qg;
1652 struct btrfs_qgroup_list *glist;
1653
995e01b7 1654 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1655
1656 qg->reserved -= num_bytes;
1657
1658 list_for_each_entry(glist, &qg->groups, next_group) {
1659 ulist_add(ulist, glist->group->qgroupid,
995e01b7 1660 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1661 }
1662 }
1663
1664out:
1665 spin_unlock(&fs_info->qgroup_lock);
1666 ulist_free(ulist);
1667}
1668
1669void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1670{
1671 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1672 return;
1673 printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1674 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1675 trans->delayed_ref_elem.seq);
1676 BUG();
1677}
Linux 6.x block layer and io_uring tree(s)