projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix memory patcher through fs_info->qgroup_ulist
[linux-2.6-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 33#include "backref.h"
2f232036 34#include "extent_io.h"
bed92eae
AJ		 35
36/* TODO XXX FIXME
37 * - subvol delete -> delete when ref goes to 0? delete limits also?
38 * - reorganize keys
39 * - compressed
40 * - sync
bed92eae
AJ		 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
2f232036
JS		 101struct qgroup_rescan {
102 struct btrfs_work work;
103 struct btrfs_fs_info *fs_info;
104};
105
106static void qgroup_rescan_start(struct btrfs_fs_info *fs_info,
107 struct qgroup_rescan *qscan);
108
58400fce 109/* must be called with qgroup_ioctl_lock held */
bed92eae
AJ		 110static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
111 u64 qgroupid)
112{
113 struct rb_node *n = fs_info->qgroup_tree.rb_node;
114 struct btrfs_qgroup *qgroup;
115
116 while (n) {
117 qgroup = rb_entry(n, struct btrfs_qgroup, node);
118 if (qgroup->qgroupid < qgroupid)
119 n = n->rb_left;
120 else if (qgroup->qgroupid > qgroupid)
121 n = n->rb_right;
122 else
123 return qgroup;
124 }
125 return NULL;
126}
127
128/* must be called with qgroup_lock held */
129static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
130 u64 qgroupid)
131{
132 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
133 struct rb_node *parent = NULL;
134 struct btrfs_qgroup *qgroup;
135
136 while (*p) {
137 parent = *p;
138 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
139
140 if (qgroup->qgroupid < qgroupid)
141 p = &(*p)->rb_left;
142 else if (qgroup->qgroupid > qgroupid)
143 p = &(*p)->rb_right;
144 else
145 return qgroup;
146 }
147
148 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
149 if (!qgroup)
150 return ERR_PTR(-ENOMEM);
151
152 qgroup->qgroupid = qgroupid;
153 INIT_LIST_HEAD(&qgroup->groups);
154 INIT_LIST_HEAD(&qgroup->members);
155 INIT_LIST_HEAD(&qgroup->dirty);
156
157 rb_link_node(&qgroup->node, parent, p);
158 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
159
160 return qgroup;
161}
162
163/* must be called with qgroup_lock held */
164static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
165{
166 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
167 struct btrfs_qgroup_list *list;
168
169 if (!qgroup)
170 return -ENOENT;
171
172 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
173 list_del(&qgroup->dirty);
174
175 while (!list_empty(&qgroup->groups)) {
176 list = list_first_entry(&qgroup->groups,
177 struct btrfs_qgroup_list, next_group);
178 list_del(&list->next_group);
179 list_del(&list->next_member);
180 kfree(list);
181 }
182
183 while (!list_empty(&qgroup->members)) {
184 list = list_first_entry(&qgroup->members,
185 struct btrfs_qgroup_list, next_member);
186 list_del(&list->next_group);
187 list_del(&list->next_member);
188 kfree(list);
189 }
190 kfree(qgroup);
191
192 return 0;
193}
194
195/* must be called with qgroup_lock held */
196static int add_relation_rb(struct btrfs_fs_info *fs_info,
197 u64 memberid, u64 parentid)
198{
199 struct btrfs_qgroup *member;
200 struct btrfs_qgroup *parent;
201 struct btrfs_qgroup_list *list;
202
203 member = find_qgroup_rb(fs_info, memberid);
204 parent = find_qgroup_rb(fs_info, parentid);
205 if (!member || !parent)
206 return -ENOENT;
207
208 list = kzalloc(sizeof(*list), GFP_ATOMIC);
209 if (!list)
210 return -ENOMEM;
211
212 list->group = parent;
213 list->member = member;
214 list_add_tail(&list->next_group, &member->groups);
215 list_add_tail(&list->next_member, &parent->members);
216
217 return 0;
218}
219
220/* must be called with qgroup_lock held */
221static int del_relation_rb(struct btrfs_fs_info *fs_info,
222 u64 memberid, u64 parentid)
223{
224 struct btrfs_qgroup *member;
225 struct btrfs_qgroup *parent;
226 struct btrfs_qgroup_list *list;
227
228 member = find_qgroup_rb(fs_info, memberid);
229 parent = find_qgroup_rb(fs_info, parentid);
230 if (!member || !parent)
231 return -ENOENT;
232
233 list_for_each_entry(list, &member->groups, next_group) {
234 if (list->group == parent) {
235 list_del(&list->next_group);
236 list_del(&list->next_member);
237 kfree(list);
238 return 0;
239 }
240 }
241 return -ENOENT;
242}
243
244/*
245 * The full config is read in one go, only called from open_ctree()
246 * It doesn't use any locking, as at this point we're still single-threaded
247 */
248int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
249{
250 struct btrfs_key key;
251 struct btrfs_key found_key;
252 struct btrfs_root *quota_root = fs_info->quota_root;
253 struct btrfs_path *path = NULL;
254 struct extent_buffer *l;
255 int slot;
256 int ret = 0;
257 u64 flags = 0;
258
259 if (!fs_info->quota_enabled)
260 return 0;
261
1e8f9158
WS		 262 fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
263 if (!fs_info->qgroup_ulist) {
264 ret = -ENOMEM;
265 goto out;
266 }
267
bed92eae
AJ		 268 path = btrfs_alloc_path();
269 if (!path) {
270 ret = -ENOMEM;
271 goto out;
272 }
273
274 /* default this to quota off, in case no status key is found */
275 fs_info->qgroup_flags = 0;
276
277 /*
278 * pass 1: read status, all qgroup infos and limits
279 */
280 key.objectid = 0;
281 key.type = 0;
282 key.offset = 0;
283 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
284 if (ret)
285 goto out;
286
287 while (1) {
288 struct btrfs_qgroup *qgroup;
289
290 slot = path->slots[0];
291 l = path->nodes[0];
292 btrfs_item_key_to_cpu(l, &found_key, slot);
293
294 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
295 struct btrfs_qgroup_status_item *ptr;
296
297 ptr = btrfs_item_ptr(l, slot,
298 struct btrfs_qgroup_status_item);
299
300 if (btrfs_qgroup_status_version(l, ptr) !=
301 BTRFS_QGROUP_STATUS_VERSION) {
302 printk(KERN_ERR
303 "btrfs: old qgroup version, quota disabled\n");
304 goto out;
305 }
306 if (btrfs_qgroup_status_generation(l, ptr) !=
307 fs_info->generation) {
308 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
309 printk(KERN_ERR
310 "btrfs: qgroup generation mismatch, "
311 "marked as inconsistent\n");
312 }
313 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
314 ptr);
2f232036
JS		 315 fs_info->qgroup_rescan_progress.objectid =
316 btrfs_qgroup_status_rescan(l, ptr);
317 if (fs_info->qgroup_flags &
318 BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
319 struct qgroup_rescan *qscan =
320 kmalloc(sizeof(*qscan), GFP_NOFS);
321 if (!qscan) {
322 ret = -ENOMEM;
323 goto out;
324 }
325 fs_info->qgroup_rescan_progress.type = 0;
326 fs_info->qgroup_rescan_progress.offset = 0;
327 qgroup_rescan_start(fs_info, qscan);
328 }
bed92eae
AJ		 329 goto next1;
330 }
331
332 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
333 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
334 goto next1;
335
336 qgroup = find_qgroup_rb(fs_info, found_key.offset);
337 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
338 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
339 printk(KERN_ERR "btrfs: inconsitent qgroup config\n");
340 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
341 }
342 if (!qgroup) {
343 qgroup = add_qgroup_rb(fs_info, found_key.offset);
344 if (IS_ERR(qgroup)) {
345 ret = PTR_ERR(qgroup);
346 goto out;
347 }
348 }
349 switch (found_key.type) {
350 case BTRFS_QGROUP_INFO_KEY: {
351 struct btrfs_qgroup_info_item *ptr;
352
353 ptr = btrfs_item_ptr(l, slot,
354 struct btrfs_qgroup_info_item);
355 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
356 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
357 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
358 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
359 /* generation currently unused */
360 break;
361 }
362 case BTRFS_QGROUP_LIMIT_KEY: {
363 struct btrfs_qgroup_limit_item *ptr;
364
365 ptr = btrfs_item_ptr(l, slot,
366 struct btrfs_qgroup_limit_item);
367 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
368 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
369 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
370 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
371 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
372 break;
373 }
374 }
375next1:
376 ret = btrfs_next_item(quota_root, path);
377 if (ret < 0)
378 goto out;
379 if (ret)
380 break;
381 }
382 btrfs_release_path(path);
383
384 /*
385 * pass 2: read all qgroup relations
386 */
387 key.objectid = 0;
388 key.type = BTRFS_QGROUP_RELATION_KEY;
389 key.offset = 0;
390 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
391 if (ret)
392 goto out;
393 while (1) {
394 slot = path->slots[0];
395 l = path->nodes[0];
396 btrfs_item_key_to_cpu(l, &found_key, slot);
397
398 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
399 goto next2;
400
401 if (found_key.objectid > found_key.offset) {
402 /* parent <- member, not needed to build config */
403 /* FIXME should we omit the key completely? */
404 goto next2;
405 }
406
407 ret = add_relation_rb(fs_info, found_key.objectid,
408 found_key.offset);
ff24858c
AJ		 409 if (ret == -ENOENT) {
410 printk(KERN_WARNING
411 "btrfs: orphan qgroup relation 0x%llx->0x%llx\n",
412 (unsigned long long)found_key.objectid,
413 (unsigned long long)found_key.offset);
414 ret = 0; /* ignore the error */
415 }
bed92eae
AJ		 416 if (ret)
417 goto out;
418next2:
419 ret = btrfs_next_item(quota_root, path);
420 if (ret < 0)
421 goto out;
422 if (ret)
423 break;
424 }
425out:
426 fs_info->qgroup_flags |= flags;
427 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
428 fs_info->quota_enabled = 0;
429 fs_info->pending_quota_state = 0;
430 }
431 btrfs_free_path(path);
432
4373519d 433 if (ret < 0)
1e8f9158
WS		 434 ulist_free(fs_info->qgroup_ulist);
435
bed92eae
AJ		 436 return ret < 0 ? ret : 0;
437}
438
439/*
440 * This is only called from close_ctree() or open_ctree(), both in single-
441 * treaded paths. Clean up the in-memory structures. No locking needed.
442 */
443void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
444{
445 struct rb_node *n;
446 struct btrfs_qgroup *qgroup;
447 struct btrfs_qgroup_list *list;
448
449 while ((n = rb_first(&fs_info->qgroup_tree))) {
450 qgroup = rb_entry(n, struct btrfs_qgroup, node);
451 rb_erase(n, &fs_info->qgroup_tree);
452
bed92eae
AJ		 453 while (!list_empty(&qgroup->groups)) {
454 list = list_first_entry(&qgroup->groups,
455 struct btrfs_qgroup_list,
456 next_group);
457 list_del(&list->next_group);
458 list_del(&list->next_member);
459 kfree(list);
460 }
461
462 while (!list_empty(&qgroup->members)) {
463 list = list_first_entry(&qgroup->members,
464 struct btrfs_qgroup_list,
465 next_member);
466 list_del(&list->next_group);
467 list_del(&list->next_member);
468 kfree(list);
469 }
470 kfree(qgroup);
471 }
1e8f9158 472 ulist_free(fs_info->qgroup_ulist);
bed92eae
AJ		 473}
474
475static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
476 struct btrfs_root *quota_root,
477 u64 src, u64 dst)
478{
479 int ret;
480 struct btrfs_path *path;
481 struct btrfs_key key;
482
483 path = btrfs_alloc_path();
484 if (!path)
485 return -ENOMEM;
486
487 key.objectid = src;
488 key.type = BTRFS_QGROUP_RELATION_KEY;
489 key.offset = dst;
490
491 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
492
493 btrfs_mark_buffer_dirty(path->nodes[0]);
494
495 btrfs_free_path(path);
496 return ret;
497}
498
499static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
500 struct btrfs_root *quota_root,
501 u64 src, u64 dst)
502{
503 int ret;
504 struct btrfs_path *path;
505 struct btrfs_key key;
506
507 path = btrfs_alloc_path();
508 if (!path)
509 return -ENOMEM;
510
511 key.objectid = src;
512 key.type = BTRFS_QGROUP_RELATION_KEY;
513 key.offset = dst;
514
515 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
516 if (ret < 0)
517 goto out;
518
519 if (ret > 0) {
520 ret = -ENOENT;
521 goto out;
522 }
523
524 ret = btrfs_del_item(trans, quota_root, path);
525out:
526 btrfs_free_path(path);
527 return ret;
528}
529
530static int add_qgroup_item(struct btrfs_trans_handle *trans,
531 struct btrfs_root *quota_root, u64 qgroupid)
532{
533 int ret;
534 struct btrfs_path *path;
535 struct btrfs_qgroup_info_item *qgroup_info;
536 struct btrfs_qgroup_limit_item *qgroup_limit;
537 struct extent_buffer *leaf;
538 struct btrfs_key key;
539
540 path = btrfs_alloc_path();
541 if (!path)
542 return -ENOMEM;
543
544 key.objectid = 0;
545 key.type = BTRFS_QGROUP_INFO_KEY;
546 key.offset = qgroupid;
547
548 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
549 sizeof(*qgroup_info));
550 if (ret)
551 goto out;
552
553 leaf = path->nodes[0];
554 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
555 struct btrfs_qgroup_info_item);
556 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
557 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
558 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
559 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
560 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
561
562 btrfs_mark_buffer_dirty(leaf);
563
564 btrfs_release_path(path);
565
566 key.type = BTRFS_QGROUP_LIMIT_KEY;
567 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
568 sizeof(*qgroup_limit));
569 if (ret)
570 goto out;
571
572 leaf = path->nodes[0];
573 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
574 struct btrfs_qgroup_limit_item);
575 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
576 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
577 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
578 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
579 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
580
581 btrfs_mark_buffer_dirty(leaf);
582
583 ret = 0;
584out:
585 btrfs_free_path(path);
586 return ret;
587}
588
589static int del_qgroup_item(struct btrfs_trans_handle *trans,
590 struct btrfs_root *quota_root, u64 qgroupid)
591{
592 int ret;
593 struct btrfs_path *path;
594 struct btrfs_key key;
595
596 path = btrfs_alloc_path();
597 if (!path)
598 return -ENOMEM;
599
600 key.objectid = 0;
601 key.type = BTRFS_QGROUP_INFO_KEY;
602 key.offset = qgroupid;
603 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
604 if (ret < 0)
605 goto out;
606
607 if (ret > 0) {
608 ret = -ENOENT;
609 goto out;
610 }
611
612 ret = btrfs_del_item(trans, quota_root, path);
613 if (ret)
614 goto out;
615
616 btrfs_release_path(path);
617
618 key.type = BTRFS_QGROUP_LIMIT_KEY;
619 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
620 if (ret < 0)
621 goto out;
622
623 if (ret > 0) {
624 ret = -ENOENT;
625 goto out;
626 }
627
628 ret = btrfs_del_item(trans, quota_root, path);
629
630out:
631 btrfs_free_path(path);
632 return ret;
633}
634
635static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
636 struct btrfs_root *root, u64 qgroupid,
637 u64 flags, u64 max_rfer, u64 max_excl,
638 u64 rsv_rfer, u64 rsv_excl)
639{
640 struct btrfs_path *path;
641 struct btrfs_key key;
642 struct extent_buffer *l;
643 struct btrfs_qgroup_limit_item *qgroup_limit;
644 int ret;
645 int slot;
646
647 key.objectid = 0;
648 key.type = BTRFS_QGROUP_LIMIT_KEY;
649 key.offset = qgroupid;
650
651 path = btrfs_alloc_path();
84cbe2f7
WS		 652 if (!path)
653 return -ENOMEM;
654
bed92eae
AJ		 655 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
656 if (ret > 0)
657 ret = -ENOENT;
658
659 if (ret)
660 goto out;
661
662 l = path->nodes[0];
663 slot = path->slots[0];
664 qgroup_limit = btrfs_item_ptr(l, path->slots[0],
665 struct btrfs_qgroup_limit_item);
666 btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
667 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
668 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
669 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
670 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
671
672 btrfs_mark_buffer_dirty(l);
673
674out:
675 btrfs_free_path(path);
676 return ret;
677}
678
679static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
680 struct btrfs_root *root,
681 struct btrfs_qgroup *qgroup)
682{
683 struct btrfs_path *path;
684 struct btrfs_key key;
685 struct extent_buffer *l;
686 struct btrfs_qgroup_info_item *qgroup_info;
687 int ret;
688 int slot;
689
690 key.objectid = 0;
691 key.type = BTRFS_QGROUP_INFO_KEY;
692 key.offset = qgroup->qgroupid;
693
694 path = btrfs_alloc_path();
84cbe2f7
WS		 695 if (!path)
696 return -ENOMEM;
697
bed92eae
AJ		 698 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
699 if (ret > 0)
700 ret = -ENOENT;
701
702 if (ret)
703 goto out;
704
705 l = path->nodes[0];
706 slot = path->slots[0];
707 qgroup_info = btrfs_item_ptr(l, path->slots[0],
708 struct btrfs_qgroup_info_item);
709 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
710 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
711 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
712 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
713 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
714
715 btrfs_mark_buffer_dirty(l);
716
717out:
718 btrfs_free_path(path);
719 return ret;
720}
721
722static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
723 struct btrfs_fs_info *fs_info,
724 struct btrfs_root *root)
725{
726 struct btrfs_path *path;
727 struct btrfs_key key;
728 struct extent_buffer *l;
729 struct btrfs_qgroup_status_item *ptr;
730 int ret;
731 int slot;
732
733 key.objectid = 0;
734 key.type = BTRFS_QGROUP_STATUS_KEY;
735 key.offset = 0;
736
737 path = btrfs_alloc_path();
84cbe2f7
WS		 738 if (!path)
739 return -ENOMEM;
740
bed92eae
AJ		 741 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
742 if (ret > 0)
743 ret = -ENOENT;
744
745 if (ret)
746 goto out;
747
748 l = path->nodes[0];
749 slot = path->slots[0];
750 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
751 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
752 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
2f232036
JS		 753 btrfs_set_qgroup_status_rescan(l, ptr,
754 fs_info->qgroup_rescan_progress.objectid);
bed92eae
AJ		 755
756 btrfs_mark_buffer_dirty(l);
757
758out:
759 btrfs_free_path(path);
760 return ret;
761}
762
763/*
764 * called with qgroup_lock held
765 */
766static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
767 struct btrfs_root *root)
768{
769 struct btrfs_path *path;
770 struct btrfs_key key;
06b3a860 771 struct extent_buffer *leaf = NULL;
bed92eae 772 int ret;
06b3a860 773 int nr = 0;
bed92eae 774
bed92eae
AJ		 775 path = btrfs_alloc_path();
776 if (!path)
777 return -ENOMEM;
778
06b3a860
WS		 779 path->leave_spinning = 1;
780
781 key.objectid = 0;
782 key.offset = 0;
783 key.type = 0;
bed92eae 784
06b3a860 785 while (1) {
bed92eae 786 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
06b3a860
WS		 787 if (ret < 0)
788 goto out;
789 leaf = path->nodes[0];
790 nr = btrfs_header_nritems(leaf);
791 if (!nr)
bed92eae 792 break;
06b3a860
WS		 793 /*
794 * delete the leaf one by one
795 * since the whole tree is going
796 * to be deleted.
797 */
798 path->slots[0] = 0;
799 ret = btrfs_del_items(trans, root, path, 0, nr);
bed92eae
AJ		 800 if (ret)
801 goto out;
06b3a860 802
bed92eae
AJ		 803 btrfs_release_path(path);
804 }
805 ret = 0;
806out:
807 root->fs_info->pending_quota_state = 0;
808 btrfs_free_path(path);
809 return ret;
810}
811
812int btrfs_quota_enable(struct btrfs_trans_handle *trans,
813 struct btrfs_fs_info *fs_info)
814{
815 struct btrfs_root *quota_root;
7708f029 816 struct btrfs_root *tree_root = fs_info->tree_root;
bed92eae
AJ		 817 struct btrfs_path *path = NULL;
818 struct btrfs_qgroup_status_item *ptr;
819 struct extent_buffer *leaf;
820 struct btrfs_key key;
7708f029
WS		 821 struct btrfs_key found_key;
822 struct btrfs_qgroup *qgroup = NULL;
bed92eae 823 int ret = 0;
7708f029 824 int slot;
bed92eae 825
f2f6ed3d 826 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 827 if (fs_info->quota_root) {
828 fs_info->pending_quota_state = 1;
bed92eae
AJ		 829 goto out;
830 }
bed92eae 831
1e8f9158
WS		 832 fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
833 if (!fs_info->qgroup_ulist) {
834 ret = -ENOMEM;
835 goto out;
836 }
837
bed92eae
AJ		 838 /*
839 * initially create the quota tree
840 */
841 quota_root = btrfs_create_tree(trans, fs_info,
842 BTRFS_QUOTA_TREE_OBJECTID);
843 if (IS_ERR(quota_root)) {
844 ret = PTR_ERR(quota_root);
845 goto out;
846 }
847
848 path = btrfs_alloc_path();
5b7ff5b3
TI		 849 if (!path) {
850 ret = -ENOMEM;
851 goto out_free_root;
852 }
bed92eae
AJ		 853
854 key.objectid = 0;
855 key.type = BTRFS_QGROUP_STATUS_KEY;
856 key.offset = 0;
857
858 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
859 sizeof(*ptr));
860 if (ret)
5b7ff5b3 861 goto out_free_path;
bed92eae
AJ		 862
863 leaf = path->nodes[0];
864 ptr = btrfs_item_ptr(leaf, path->slots[0],
865 struct btrfs_qgroup_status_item);
866 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
867 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
868 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
869 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
870 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
2f232036 871 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
bed92eae
AJ		 872
873 btrfs_mark_buffer_dirty(leaf);
874
7708f029
WS		 875 key.objectid = 0;
876 key.type = BTRFS_ROOT_REF_KEY;
877 key.offset = 0;
878
879 btrfs_release_path(path);
880 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
881 if (ret > 0)
882 goto out_add_root;
883 if (ret < 0)
884 goto out_free_path;
885
886
887 while (1) {
888 slot = path->slots[0];
889 leaf = path->nodes[0];
890 btrfs_item_key_to_cpu(leaf, &found_key, slot);
891
892 if (found_key.type == BTRFS_ROOT_REF_KEY) {
893 ret = add_qgroup_item(trans, quota_root,
894 found_key.offset);
895 if (ret)
896 goto out_free_path;
897
7708f029
WS		 898 qgroup = add_qgroup_rb(fs_info, found_key.offset);
899 if (IS_ERR(qgroup)) {
7708f029
WS		 900 ret = PTR_ERR(qgroup);
901 goto out_free_path;
902 }
7708f029
WS		 903 }
904 ret = btrfs_next_item(tree_root, path);
905 if (ret < 0)
906 goto out_free_path;
907 if (ret)
908 break;
909 }
910
911out_add_root:
912 btrfs_release_path(path);
913 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
914 if (ret)
915 goto out_free_path;
916
7708f029
WS		 917 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
918 if (IS_ERR(qgroup)) {
7708f029
WS		 919 ret = PTR_ERR(qgroup);
920 goto out_free_path;
921 }
58400fce 922 spin_lock(&fs_info->qgroup_lock);
bed92eae
AJ		 923 fs_info->quota_root = quota_root;
924 fs_info->pending_quota_state = 1;
925 spin_unlock(&fs_info->qgroup_lock);
5b7ff5b3 926out_free_path:
bed92eae 927 btrfs_free_path(path);
5b7ff5b3
TI		 928out_free_root:
929 if (ret) {
930 free_extent_buffer(quota_root->node);
931 free_extent_buffer(quota_root->commit_root);
932 kfree(quota_root);
933 }
934out:
1e8f9158
WS		 935 if (ret)
936 ulist_free(fs_info->qgroup_ulist);
f2f6ed3d 937 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 938 return ret;
939}
940
941int btrfs_quota_disable(struct btrfs_trans_handle *trans,
942 struct btrfs_fs_info *fs_info)
943{
944 struct btrfs_root *tree_root = fs_info->tree_root;
945 struct btrfs_root *quota_root;
946 int ret = 0;
947
f2f6ed3d 948 mutex_lock(&fs_info->qgroup_ioctl_lock);
58400fce 949 if (!fs_info->quota_root)
f2f6ed3d 950 goto out;
58400fce 951 spin_lock(&fs_info->qgroup_lock);
bed92eae
AJ		 952 fs_info->quota_enabled = 0;
953 fs_info->pending_quota_state = 0;
954 quota_root = fs_info->quota_root;
955 fs_info->quota_root = NULL;
956 btrfs_free_qgroup_config(fs_info);
957 spin_unlock(&fs_info->qgroup_lock);
958
f2f6ed3d
WS		 959 if (!quota_root) {
960 ret = -EINVAL;
961 goto out;
962 }
bed92eae
AJ		 963
964 ret = btrfs_clean_quota_tree(trans, quota_root);
965 if (ret)
966 goto out;
967
968 ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
969 if (ret)
970 goto out;
971
972 list_del(&quota_root->dirty_list);
973
974 btrfs_tree_lock(quota_root->node);
975 clean_tree_block(trans, tree_root, quota_root->node);
976 btrfs_tree_unlock(quota_root->node);
977 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
978
979 free_extent_buffer(quota_root->node);
980 free_extent_buffer(quota_root->commit_root);
981 kfree(quota_root);
982out:
f2f6ed3d 983 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 984 return ret;
985}
986
2f232036
JS		 987static void qgroup_dirty(struct btrfs_fs_info *fs_info,
988 struct btrfs_qgroup *qgroup)
bed92eae 989{
2f232036
JS		 990 if (list_empty(&qgroup->dirty))
991 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
bed92eae
AJ		 992}
993
994int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
995 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
996{
997 struct btrfs_root *quota_root;
b7fef4f5
WS		 998 struct btrfs_qgroup *parent;
999 struct btrfs_qgroup *member;
534e6623 1000 struct btrfs_qgroup_list *list;
bed92eae
AJ		 1001 int ret = 0;
1002
f2f6ed3d 1003 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1004 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1005 if (!quota_root) {
1006 ret = -EINVAL;
1007 goto out;
1008 }
b7fef4f5
WS		 1009 member = find_qgroup_rb(fs_info, src);
1010 parent = find_qgroup_rb(fs_info, dst);
1011 if (!member || !parent) {
1012 ret = -EINVAL;
1013 goto out;
1014 }
bed92eae 1015
534e6623
WS		 1016 /* check if such qgroup relation exist firstly */
1017 list_for_each_entry(list, &member->groups, next_group) {
1018 if (list->group == parent) {
1019 ret = -EEXIST;
1020 goto out;
1021 }
1022 }
1023
bed92eae
AJ		 1024 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1025 if (ret)
f2f6ed3d 1026 goto out;
bed92eae
AJ		 1027
1028 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1029 if (ret) {
1030 del_qgroup_relation_item(trans, quota_root, src, dst);
f2f6ed3d 1031 goto out;
bed92eae
AJ		 1032 }
1033
1034 spin_lock(&fs_info->qgroup_lock);
1035 ret = add_relation_rb(quota_root->fs_info, src, dst);
1036 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1037out:
1038 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1039 return ret;
1040}
1041
1042int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1043 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1044{
1045 struct btrfs_root *quota_root;
534e6623
WS		 1046 struct btrfs_qgroup *parent;
1047 struct btrfs_qgroup *member;
1048 struct btrfs_qgroup_list *list;
bed92eae
AJ		 1049 int ret = 0;
1050 int err;
1051
f2f6ed3d 1052 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1053 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1054 if (!quota_root) {
1055 ret = -EINVAL;
1056 goto out;
1057 }
bed92eae 1058
534e6623
WS		 1059 member = find_qgroup_rb(fs_info, src);
1060 parent = find_qgroup_rb(fs_info, dst);
1061 if (!member || !parent) {
1062 ret = -EINVAL;
1063 goto out;
1064 }
1065
1066 /* check if such qgroup relation exist firstly */
1067 list_for_each_entry(list, &member->groups, next_group) {
1068 if (list->group == parent)
1069 goto exist;
1070 }
1071 ret = -ENOENT;
1072 goto out;
1073exist:
bed92eae
AJ		 1074 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1075 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1076 if (err && !ret)
1077 ret = err;
1078
1079 spin_lock(&fs_info->qgroup_lock);
1080 del_relation_rb(fs_info, src, dst);
bed92eae 1081 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1082out:
1083 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1084 return ret;
1085}
1086
1087int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1088 struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
1089{
1090 struct btrfs_root *quota_root;
1091 struct btrfs_qgroup *qgroup;
1092 int ret = 0;
1093
f2f6ed3d 1094 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1095 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1096 if (!quota_root) {
1097 ret = -EINVAL;
1098 goto out;
1099 }
534e6623
WS		 1100 qgroup = find_qgroup_rb(fs_info, qgroupid);
1101 if (qgroup) {
1102 ret = -EEXIST;
1103 goto out;
1104 }
bed92eae
AJ		 1105
1106 ret = add_qgroup_item(trans, quota_root, qgroupid);
534e6623
WS		 1107 if (ret)
1108 goto out;
bed92eae
AJ		 1109
1110 spin_lock(&fs_info->qgroup_lock);
1111 qgroup = add_qgroup_rb(fs_info, qgroupid);
1112 spin_unlock(&fs_info->qgroup_lock);
1113
1114 if (IS_ERR(qgroup))
1115 ret = PTR_ERR(qgroup);
f2f6ed3d
WS		 1116out:
1117 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1118 return ret;
1119}
1120
1121int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1122 struct btrfs_fs_info *fs_info, u64 qgroupid)
1123{
1124 struct btrfs_root *quota_root;
2cf68703 1125 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 1126 int ret = 0;
1127
f2f6ed3d 1128 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1129 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1130 if (!quota_root) {
1131 ret = -EINVAL;
1132 goto out;
1133 }
bed92eae 1134
2cf68703 1135 qgroup = find_qgroup_rb(fs_info, qgroupid);
534e6623
WS		 1136 if (!qgroup) {
1137 ret = -ENOENT;
1138 goto out;
1139 } else {
1140 /* check if there are no relations to this qgroup */
1141 if (!list_empty(&qgroup->groups) ||
1142 !list_empty(&qgroup->members)) {
f2f6ed3d
WS		 1143 ret = -EBUSY;
1144 goto out;
2cf68703
AJ		 1145 }
1146 }
bed92eae
AJ		 1147 ret = del_qgroup_item(trans, quota_root, qgroupid);
1148
1149 spin_lock(&fs_info->qgroup_lock);
1150 del_qgroup_rb(quota_root->fs_info, qgroupid);
bed92eae 1151 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1152out:
1153 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1154 return ret;
1155}
1156
1157int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1158 struct btrfs_fs_info *fs_info, u64 qgroupid,
1159 struct btrfs_qgroup_limit *limit)
1160{
f2f6ed3d 1161 struct btrfs_root *quota_root;
bed92eae
AJ		 1162 struct btrfs_qgroup *qgroup;
1163 int ret = 0;
1164
f2f6ed3d
WS		 1165 mutex_lock(&fs_info->qgroup_ioctl_lock);
1166 quota_root = fs_info->quota_root;
1167 if (!quota_root) {
1168 ret = -EINVAL;
1169 goto out;
1170 }
bed92eae 1171
ddb47afa
WS		 1172 qgroup = find_qgroup_rb(fs_info, qgroupid);
1173 if (!qgroup) {
1174 ret = -ENOENT;
1175 goto out;
1176 }
bed92eae
AJ		 1177 ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1178 limit->flags, limit->max_rfer,
1179 limit->max_excl, limit->rsv_rfer,
1180 limit->rsv_excl);
1181 if (ret) {
1182 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1183 printk(KERN_INFO "unable to update quota limit for %llu\n",
1184 (unsigned long long)qgroupid);
1185 }
1186
58400fce 1187 spin_lock(&fs_info->qgroup_lock);
bed92eae
AJ		 1188 qgroup->lim_flags = limit->flags;
1189 qgroup->max_rfer = limit->max_rfer;
1190 qgroup->max_excl = limit->max_excl;
1191 qgroup->rsv_rfer = limit->rsv_rfer;
1192 qgroup->rsv_excl = limit->rsv_excl;
bed92eae 1193 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1194out:
1195 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1196 return ret;
1197}
1198
bed92eae
AJ		 1199/*
1200 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1201 * the modification into a list that's later used by btrfs_end_transaction to
1202 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1203 */
1204int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1205 struct btrfs_delayed_ref_node *node,
1206 struct btrfs_delayed_extent_op *extent_op)
1207{
1208 struct qgroup_update *u;
1209
1210 BUG_ON(!trans->delayed_ref_elem.seq);
1211 u = kmalloc(sizeof(*u), GFP_NOFS);
1212 if (!u)
1213 return -ENOMEM;
1214
1215 u->node = node;
1216 u->extent_op = extent_op;
1217 list_add_tail(&u->list, &trans->qgroup_ref_list);
1218
1219 return 0;
1220}
1221
46b665ce
JS		 1222static int qgroup_account_ref_step1(struct btrfs_fs_info *fs_info,
1223 struct ulist *roots, struct ulist *tmp,
1224 u64 seq)
1225{
1226 struct ulist_node *unode;
1227 struct ulist_iterator uiter;
1228 struct ulist_node *tmp_unode;
1229 struct ulist_iterator tmp_uiter;
1230 struct btrfs_qgroup *qg;
1231 int ret;
1232
1233 ULIST_ITER_INIT(&uiter);
1234 while ((unode = ulist_next(roots, &uiter))) {
1235 qg = find_qgroup_rb(fs_info, unode->val);
1236 if (!qg)
1237 continue;
1238
1239 ulist_reinit(tmp);
1240 /* XXX id not needed */
1241 ret = ulist_add(tmp, qg->qgroupid,
1242 (u64)(uintptr_t)qg, GFP_ATOMIC);
1243 if (ret < 0)
1244 return ret;
1245 ULIST_ITER_INIT(&tmp_uiter);
1246 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1247 struct btrfs_qgroup_list *glist;
1248
1249 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1250 if (qg->refcnt < seq)
1251 qg->refcnt = seq + 1;
1252 else
1253 ++qg->refcnt;
1254
1255 list_for_each_entry(glist, &qg->groups, next_group) {
1256 ret = ulist_add(tmp, glist->group->qgroupid,
1257 (u64)(uintptr_t)glist->group,
1258 GFP_ATOMIC);
1259 if (ret < 0)
1260 return ret;
1261 }
1262 }
1263 }
1264
1265 return 0;
1266}
1267
1268static int qgroup_account_ref_step2(struct btrfs_fs_info *fs_info,
1269 struct ulist *roots, struct ulist *tmp,
1270 u64 seq, int sgn, u64 num_bytes,
1271 struct btrfs_qgroup *qgroup)
1272{
1273 struct ulist_node *unode;
1274 struct ulist_iterator uiter;
1275 struct btrfs_qgroup *qg;
1276 struct btrfs_qgroup_list *glist;
1277 int ret;
1278
1279 ulist_reinit(tmp);
1280 ret = ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
1281 if (ret < 0)
1282 return ret;
1283
1284 ULIST_ITER_INIT(&uiter);
1285 while ((unode = ulist_next(tmp, &uiter))) {
1286 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1287 if (qg->refcnt < seq) {
1288 /* not visited by step 1 */
1289 qg->rfer += sgn * num_bytes;
1290 qg->rfer_cmpr += sgn * num_bytes;
1291 if (roots->nnodes == 0) {
1292 qg->excl += sgn * num_bytes;
1293 qg->excl_cmpr += sgn * num_bytes;
1294 }
1295 qgroup_dirty(fs_info, qg);
1296 }
1297 WARN_ON(qg->tag >= seq);
1298 qg->tag = seq;
1299
1300 list_for_each_entry(glist, &qg->groups, next_group) {
1301 ret = ulist_add(tmp, glist->group->qgroupid,
1302 (uintptr_t)glist->group, GFP_ATOMIC);
1303 if (ret < 0)
1304 return ret;
1305 }
1306 }
1307
1308 return 0;
1309}
1310
1311static int qgroup_account_ref_step3(struct btrfs_fs_info *fs_info,
1312 struct ulist *roots, struct ulist *tmp,
1313 u64 seq, int sgn, u64 num_bytes)
1314{
1315 struct ulist_node *unode;
1316 struct ulist_iterator uiter;
1317 struct btrfs_qgroup *qg;
1318 struct ulist_node *tmp_unode;
1319 struct ulist_iterator tmp_uiter;
1320 int ret;
1321
1322 ULIST_ITER_INIT(&uiter);
1323 while ((unode = ulist_next(roots, &uiter))) {
1324 qg = find_qgroup_rb(fs_info, unode->val);
1325 if (!qg)
1326 continue;
1327
1328 ulist_reinit(tmp);
1329 ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
1330 if (ret < 0)
1331 return ret;
1332
1333 ULIST_ITER_INIT(&tmp_uiter);
1334 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1335 struct btrfs_qgroup_list *glist;
1336
1337 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1338 if (qg->tag == seq)
1339 continue;
1340
1341 if (qg->refcnt - seq == roots->nnodes) {
1342 qg->excl -= sgn * num_bytes;
1343 qg->excl_cmpr -= sgn * num_bytes;
1344 qgroup_dirty(fs_info, qg);
1345 }
1346
1347 list_for_each_entry(glist, &qg->groups, next_group) {
1348 ret = ulist_add(tmp, glist->group->qgroupid,
1349 (uintptr_t)glist->group,
1350 GFP_ATOMIC);
1351 if (ret < 0)
1352 return ret;
1353 }
1354 }
1355 }
1356
1357 return 0;
1358}
1359
bed92eae
AJ		 1360/*
1361 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1362 * from the fs. First, all roots referencing the extent are searched, and
1363 * then the space is accounted accordingly to the different roots. The
1364 * accounting algorithm works in 3 steps documented inline.
1365 */
1366int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1367 struct btrfs_fs_info *fs_info,
1368 struct btrfs_delayed_ref_node *node,
1369 struct btrfs_delayed_extent_op *extent_op)
1370{
1371 struct btrfs_key ins;
1372 struct btrfs_root *quota_root;
1373 u64 ref_root;
1374 struct btrfs_qgroup *qgroup;
bed92eae 1375 struct ulist *roots = NULL;
bed92eae
AJ		 1376 u64 seq;
1377 int ret = 0;
1378 int sgn;
1379
1380 if (!fs_info->quota_enabled)
1381 return 0;
1382
1383 BUG_ON(!fs_info->quota_root);
1384
1385 ins.objectid = node->bytenr;
1386 ins.offset = node->num_bytes;
1387 ins.type = BTRFS_EXTENT_ITEM_KEY;
1388
1389 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1390 node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1391 struct btrfs_delayed_tree_ref *ref;
1392 ref = btrfs_delayed_node_to_tree_ref(node);
1393 ref_root = ref->root;
1394 } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1395 node->type == BTRFS_SHARED_DATA_REF_KEY) {
1396 struct btrfs_delayed_data_ref *ref;
1397 ref = btrfs_delayed_node_to_data_ref(node);
1398 ref_root = ref->root;
1399 } else {
1400 BUG();
1401 }
1402
1403 if (!is_fstree(ref_root)) {
1404 /*
1405 * non-fs-trees are not being accounted
1406 */
1407 return 0;
1408 }
1409
1410 switch (node->action) {
1411 case BTRFS_ADD_DELAYED_REF:
1412 case BTRFS_ADD_DELAYED_EXTENT:
1413 sgn = 1;
fc36ed7e 1414 seq = btrfs_tree_mod_seq_prev(node->seq);
bed92eae
AJ		 1415 break;
1416 case BTRFS_DROP_DELAYED_REF:
1417 sgn = -1;
fc36ed7e 1418 seq = node->seq;
bed92eae
AJ		 1419 break;
1420 case BTRFS_UPDATE_DELAYED_HEAD:
1421 return 0;
1422 default:
1423 BUG();
1424 }
1425
2f232036
JS		 1426 mutex_lock(&fs_info->qgroup_rescan_lock);
1427 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1428 if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
1429 mutex_unlock(&fs_info->qgroup_rescan_lock);
1430 return 0;
1431 }
1432 }
1433 mutex_unlock(&fs_info->qgroup_rescan_lock);
1434
bed92eae
AJ		 1435 /*
1436 * the delayed ref sequence number we pass depends on the direction of
fc36ed7e
JS		 1437 * the operation. for add operations, we pass
1438 * tree_mod_log_prev_seq(node->seq) to skip
bed92eae
AJ		 1439 * the delayed ref's current sequence number, because we need the state
1440 * of the tree before the add operation. for delete operations, we pass
1441 * (node->seq) to include the delayed ref's current sequence number,
1442 * because we need the state of the tree after the delete operation.
1443 */
fc36ed7e 1444 ret = btrfs_find_all_roots(trans, fs_info, node->bytenr, seq, &roots);
bed92eae 1445 if (ret < 0)
a7975026 1446 return ret;
bed92eae 1447
2f232036 1448 mutex_lock(&fs_info->qgroup_rescan_lock);
bed92eae 1449 spin_lock(&fs_info->qgroup_lock);
2f232036
JS		 1450 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1451 if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
1452 ret = 0;
1453 goto unlock;
1454 }
1455 }
1456
bed92eae
AJ		 1457 quota_root = fs_info->quota_root;
1458 if (!quota_root)
1459 goto unlock;
1460
1461 qgroup = find_qgroup_rb(fs_info, ref_root);
1462 if (!qgroup)
1463 goto unlock;
1464
1465 /*
1466 * step 1: for each old ref, visit all nodes once and inc refcnt
1467 */
1e8f9158 1468 ulist_reinit(fs_info->qgroup_ulist);
bed92eae
AJ		 1469 seq = fs_info->qgroup_seq;
1470 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1471
1e8f9158
WS		 1472 ret = qgroup_account_ref_step1(fs_info, roots, fs_info->qgroup_ulist,
1473 seq);
46b665ce
JS		 1474 if (ret)
1475 goto unlock;
bed92eae
AJ		 1476
1477 /*
1478 * step 2: walk from the new root
1479 */
1e8f9158
WS		 1480 ret = qgroup_account_ref_step2(fs_info, roots, fs_info->qgroup_ulist,
1481 seq, sgn, node->num_bytes, qgroup);
46b665ce 1482 if (ret)
3c97185c 1483 goto unlock;
bed92eae
AJ		 1484
1485 /*
1486 * step 3: walk again from old refs
1487 */
1e8f9158
WS		 1488 ret = qgroup_account_ref_step3(fs_info, roots, fs_info->qgroup_ulist,
1489 seq, sgn, node->num_bytes);
46b665ce
JS		 1490 if (ret)
1491 goto unlock;
bed92eae 1492
bed92eae
AJ		 1493unlock:
1494 spin_unlock(&fs_info->qgroup_lock);
2f232036 1495 mutex_unlock(&fs_info->qgroup_rescan_lock);
bed92eae 1496 ulist_free(roots);
bed92eae
AJ		 1497
1498 return ret;
1499}
1500
1501/*
1502 * called from commit_transaction. Writes all changed qgroups to disk.
1503 */
1504int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1505 struct btrfs_fs_info *fs_info)
1506{
1507 struct btrfs_root *quota_root = fs_info->quota_root;
1508 int ret = 0;
3d7b5a28 1509 int start_rescan_worker = 0;
bed92eae
AJ		 1510
1511 if (!quota_root)
1512 goto out;
1513
3d7b5a28
JS		 1514 if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1515 start_rescan_worker = 1;
1516
bed92eae
AJ		 1517 fs_info->quota_enabled = fs_info->pending_quota_state;
1518
1519 spin_lock(&fs_info->qgroup_lock);
1520 while (!list_empty(&fs_info->dirty_qgroups)) {
1521 struct btrfs_qgroup *qgroup;
1522 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1523 struct btrfs_qgroup, dirty);
1524 list_del_init(&qgroup->dirty);
1525 spin_unlock(&fs_info->qgroup_lock);
1526 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1527 if (ret)
1528 fs_info->qgroup_flags |=
1529 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1530 spin_lock(&fs_info->qgroup_lock);
1531 }
1532 if (fs_info->quota_enabled)
1533 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1534 else
1535 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1536 spin_unlock(&fs_info->qgroup_lock);
1537
1538 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1539 if (ret)
1540 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1541
3d7b5a28
JS		 1542 if (!ret && start_rescan_worker) {
1543 ret = btrfs_qgroup_rescan(fs_info);
1544 if (ret)
1545 pr_err("btrfs: start rescan quota failed: %d\n", ret);
1546 ret = 0;
1547 }
1548
bed92eae
AJ		 1549out:
1550
1551 return ret;
1552}
1553
1554/*
1555 * copy the acounting information between qgroups. This is necessary when a
1556 * snapshot or a subvolume is created
1557 */
1558int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1559 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1560 struct btrfs_qgroup_inherit *inherit)
1561{
1562 int ret = 0;
1563 int i;
1564 u64 *i_qgroups;
1565 struct btrfs_root *quota_root = fs_info->quota_root;
1566 struct btrfs_qgroup *srcgroup;
1567 struct btrfs_qgroup *dstgroup;
1568 u32 level_size = 0;
3f5e2d3b 1569 u64 nums;
bed92eae 1570
f2f6ed3d 1571 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1572 if (!fs_info->quota_enabled)
f2f6ed3d 1573 goto out;
bed92eae 1574
f2f6ed3d
WS		 1575 if (!quota_root) {
1576 ret = -EINVAL;
1577 goto out;
1578 }
bed92eae 1579
3f5e2d3b
WS		 1580 if (inherit) {
1581 i_qgroups = (u64 *)(inherit + 1);
1582 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1583 2 * inherit->num_excl_copies;
1584 for (i = 0; i < nums; ++i) {
1585 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1586 if (!srcgroup) {
1587 ret = -EINVAL;
1588 goto out;
1589 }
1590 ++i_qgroups;
1591 }
1592 }
1593
bed92eae
AJ		 1594 /*
1595 * create a tracking group for the subvol itself
1596 */
1597 ret = add_qgroup_item(trans, quota_root, objectid);
1598 if (ret)
1599 goto out;
1600
1601 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1602 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1603 inherit->lim.flags,
1604 inherit->lim.max_rfer,
1605 inherit->lim.max_excl,
1606 inherit->lim.rsv_rfer,
1607 inherit->lim.rsv_excl);
1608 if (ret)
1609 goto out;
1610 }
1611
1612 if (srcid) {
1613 struct btrfs_root *srcroot;
1614 struct btrfs_key srckey;
1615 int srcroot_level;
1616
1617 srckey.objectid = srcid;
1618 srckey.type = BTRFS_ROOT_ITEM_KEY;
1619 srckey.offset = (u64)-1;
1620 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1621 if (IS_ERR(srcroot)) {
1622 ret = PTR_ERR(srcroot);
1623 goto out;
1624 }
1625
1626 rcu_read_lock();
1627 srcroot_level = btrfs_header_level(srcroot->node);
1628 level_size = btrfs_level_size(srcroot, srcroot_level);
1629 rcu_read_unlock();
1630 }
1631
1632 /*
1633 * add qgroup to all inherited groups
1634 */
1635 if (inherit) {
1636 i_qgroups = (u64 *)(inherit + 1);
1637 for (i = 0; i < inherit->num_qgroups; ++i) {
1638 ret = add_qgroup_relation_item(trans, quota_root,
1639 objectid, *i_qgroups);
1640 if (ret)
1641 goto out;
1642 ret = add_qgroup_relation_item(trans, quota_root,
1643 *i_qgroups, objectid);
1644 if (ret)
1645 goto out;
1646 ++i_qgroups;
1647 }
1648 }
1649
1650
1651 spin_lock(&fs_info->qgroup_lock);
1652
1653 dstgroup = add_qgroup_rb(fs_info, objectid);
57a5a882
DC		 1654 if (IS_ERR(dstgroup)) {
1655 ret = PTR_ERR(dstgroup);
bed92eae 1656 goto unlock;
57a5a882 1657 }
bed92eae
AJ		 1658
1659 if (srcid) {
1660 srcgroup = find_qgroup_rb(fs_info, srcid);
f3a87f1b 1661 if (!srcgroup)
bed92eae
AJ		 1662 goto unlock;
1663 dstgroup->rfer = srcgroup->rfer - level_size;
1664 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1665 srcgroup->excl = level_size;
1666 srcgroup->excl_cmpr = level_size;
1667 qgroup_dirty(fs_info, dstgroup);
1668 qgroup_dirty(fs_info, srcgroup);
1669 }
1670
f3a87f1b 1671 if (!inherit)
bed92eae
AJ		 1672 goto unlock;
1673
1674 i_qgroups = (u64 *)(inherit + 1);
1675 for (i = 0; i < inherit->num_qgroups; ++i) {
1676 ret = add_relation_rb(quota_root->fs_info, objectid,
1677 *i_qgroups);
1678 if (ret)
1679 goto unlock;
1680 ++i_qgroups;
1681 }
1682
1683 for (i = 0; i < inherit->num_ref_copies; ++i) {
1684 struct btrfs_qgroup *src;
1685 struct btrfs_qgroup *dst;
1686
1687 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1688 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1689
1690 if (!src || !dst) {
1691 ret = -EINVAL;
1692 goto unlock;
1693 }
1694
1695 dst->rfer = src->rfer - level_size;
1696 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1697 i_qgroups += 2;
1698 }
1699 for (i = 0; i < inherit->num_excl_copies; ++i) {
1700 struct btrfs_qgroup *src;
1701 struct btrfs_qgroup *dst;
1702
1703 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1704 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1705
1706 if (!src || !dst) {
1707 ret = -EINVAL;
1708 goto unlock;
1709 }
1710
1711 dst->excl = src->excl + level_size;
1712 dst->excl_cmpr = src->excl_cmpr + level_size;
1713 i_qgroups += 2;
1714 }
1715
1716unlock:
1717 spin_unlock(&fs_info->qgroup_lock);
1718out:
f2f6ed3d 1719 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1720 return ret;
1721}
1722
1723/*
1724 * reserve some space for a qgroup and all its parents. The reservation takes
1725 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1726 * accounting. If not enough space is available, EDQUOT is returned.
1727 * We assume that the requested space is new for all qgroups.
1728 */
1729int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1730{
1731 struct btrfs_root *quota_root;
1732 struct btrfs_qgroup *qgroup;
1733 struct btrfs_fs_info *fs_info = root->fs_info;
1734 u64 ref_root = root->root_key.objectid;
1735 int ret = 0;
bed92eae
AJ		 1736 struct ulist_node *unode;
1737 struct ulist_iterator uiter;
1738
1739 if (!is_fstree(ref_root))
1740 return 0;
1741
1742 if (num_bytes == 0)
1743 return 0;
1744
1745 spin_lock(&fs_info->qgroup_lock);
1746 quota_root = fs_info->quota_root;
1747 if (!quota_root)
1748 goto out;
1749
1750 qgroup = find_qgroup_rb(fs_info, ref_root);
1751 if (!qgroup)
1752 goto out;
1753
1754 /*
1755 * in a first step, we check all affected qgroups if any limits would
1756 * be exceeded
1757 */
1e8f9158
WS		 1758 ulist_reinit(fs_info->qgroup_ulist);
1759 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3c97185c
WS		 1760 (uintptr_t)qgroup, GFP_ATOMIC);
1761 if (ret < 0)
1762 goto out;
bed92eae 1763 ULIST_ITER_INIT(&uiter);
1e8f9158 1764 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 1765 struct btrfs_qgroup *qg;
1766 struct btrfs_qgroup_list *glist;
1767
995e01b7 1768 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1769
1770 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
b4fcd6be 1771 qg->reserved + (s64)qg->rfer + num_bytes >
720f1e20 1772 qg->max_rfer) {
bed92eae 1773 ret = -EDQUOT;
720f1e20
WS		 1774 goto out;
1775 }
bed92eae
AJ		 1776
1777 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
b4fcd6be 1778 qg->reserved + (s64)qg->excl + num_bytes >
720f1e20 1779 qg->max_excl) {
bed92eae 1780 ret = -EDQUOT;
720f1e20
WS		 1781 goto out;
1782 }
bed92eae
AJ		 1783
1784 list_for_each_entry(glist, &qg->groups, next_group) {
1e8f9158
WS		 1785 ret = ulist_add(fs_info->qgroup_ulist,
1786 glist->group->qgroupid,
3c97185c
WS		 1787 (uintptr_t)glist->group, GFP_ATOMIC);
1788 if (ret < 0)
1789 goto out;
bed92eae
AJ		 1790 }
1791 }
3c97185c 1792 ret = 0;
bed92eae
AJ		 1793 /*
1794 * no limits exceeded, now record the reservation into all qgroups
1795 */
1796 ULIST_ITER_INIT(&uiter);
1e8f9158 1797 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 1798 struct btrfs_qgroup *qg;
1799
995e01b7 1800 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1801
1802 qg->reserved += num_bytes;
1803 }
1804
1805out:
1806 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 1807 return ret;
1808}
1809
1810void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1811{
1812 struct btrfs_root *quota_root;
1813 struct btrfs_qgroup *qgroup;
1814 struct btrfs_fs_info *fs_info = root->fs_info;
bed92eae
AJ		 1815 struct ulist_node *unode;
1816 struct ulist_iterator uiter;
1817 u64 ref_root = root->root_key.objectid;
3c97185c 1818 int ret = 0;
bed92eae
AJ		 1819
1820 if (!is_fstree(ref_root))
1821 return;
1822
1823 if (num_bytes == 0)
1824 return;
1825
1826 spin_lock(&fs_info->qgroup_lock);
1827
1828 quota_root = fs_info->quota_root;
1829 if (!quota_root)
1830 goto out;
1831
1832 qgroup = find_qgroup_rb(fs_info, ref_root);
1833 if (!qgroup)
1834 goto out;
1835
1e8f9158
WS		 1836 ulist_reinit(fs_info->qgroup_ulist);
1837 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3c97185c
WS		 1838 (uintptr_t)qgroup, GFP_ATOMIC);
1839 if (ret < 0)
1840 goto out;
bed92eae 1841 ULIST_ITER_INIT(&uiter);
1e8f9158 1842 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 1843 struct btrfs_qgroup *qg;
1844 struct btrfs_qgroup_list *glist;
1845
995e01b7 1846 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1847
1848 qg->reserved -= num_bytes;
1849
1850 list_for_each_entry(glist, &qg->groups, next_group) {
1e8f9158
WS		 1851 ret = ulist_add(fs_info->qgroup_ulist,
1852 glist->group->qgroupid,
3c97185c
WS		 1853 (uintptr_t)glist->group, GFP_ATOMIC);
1854 if (ret < 0)
1855 goto out;
bed92eae
AJ		 1856 }
1857 }
1858
1859out:
1860 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 1861}
1862
1863void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1864{
1865 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1866 return;
fc36ed7e 1867 pr_err("btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %#x.%x\n",
bed92eae 1868 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
fc36ed7e
JS		 1869 (u32)(trans->delayed_ref_elem.seq >> 32),
1870 (u32)trans->delayed_ref_elem.seq);
bed92eae
AJ		 1871 BUG();
1872}
2f232036
JS		 1873
1874/*
1875 * returns < 0 on error, 0 when more leafs are to be scanned.
1876 * returns 1 when done, 2 when done and FLAG_INCONSISTENT was cleared.
1877 */
1878static int
1879qgroup_rescan_leaf(struct qgroup_rescan *qscan, struct btrfs_path *path,
1880 struct btrfs_trans_handle *trans, struct ulist *tmp,
1881 struct extent_buffer *scratch_leaf)
1882{
1883 struct btrfs_key found;
1884 struct btrfs_fs_info *fs_info = qscan->fs_info;
1885 struct ulist *roots = NULL;
1886 struct ulist_node *unode;
1887 struct ulist_iterator uiter;
1888 struct seq_list tree_mod_seq_elem = {};
1889 u64 seq;
1890 int slot;
1891 int ret;
1892
1893 path->leave_spinning = 1;
1894 mutex_lock(&fs_info->qgroup_rescan_lock);
1895 ret = btrfs_search_slot_for_read(fs_info->extent_root,
1896 &fs_info->qgroup_rescan_progress,
1897 path, 1, 0);
1898
1899 pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
1900 (unsigned long long)fs_info->qgroup_rescan_progress.objectid,
1901 fs_info->qgroup_rescan_progress.type,
1902 (unsigned long long)fs_info->qgroup_rescan_progress.offset,
1903 ret);
1904
1905 if (ret) {
1906 /*
1907 * The rescan is about to end, we will not be scanning any
1908 * further blocks. We cannot unset the RESCAN flag here, because
1909 * we want to commit the transaction if everything went well.
1910 * To make the live accounting work in this phase, we set our
1911 * scan progress pointer such that every real extent objectid
1912 * will be smaller.
1913 */
1914 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
1915 btrfs_release_path(path);
1916 mutex_unlock(&fs_info->qgroup_rescan_lock);
1917 return ret;
1918 }
1919
1920 btrfs_item_key_to_cpu(path->nodes[0], &found,
1921 btrfs_header_nritems(path->nodes[0]) - 1);
1922 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
1923
1924 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1925 memcpy(scratch_leaf, path->nodes[0], sizeof(*scratch_leaf));
1926 slot = path->slots[0];
1927 btrfs_release_path(path);
1928 mutex_unlock(&fs_info->qgroup_rescan_lock);
1929
1930 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
1931 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
1932 if (found.type != BTRFS_EXTENT_ITEM_KEY)
1933 continue;
1934 ret = btrfs_find_all_roots(trans, fs_info, found.objectid,
1935 tree_mod_seq_elem.seq, &roots);
1936 if (ret < 0)
1937 goto out;
1938 spin_lock(&fs_info->qgroup_lock);
1939 seq = fs_info->qgroup_seq;
1940 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1941
1942 ret = qgroup_account_ref_step1(fs_info, roots, tmp, seq);
1943 if (ret) {
1944 spin_unlock(&fs_info->qgroup_lock);
1945 ulist_free(roots);
1946 goto out;
1947 }
1948
1949 /*
1950 * step2 of btrfs_qgroup_account_ref works from a single root,
1951 * we're doing all at once here.
1952 */
1953 ulist_reinit(tmp);
1954 ULIST_ITER_INIT(&uiter);
1955 while ((unode = ulist_next(roots, &uiter))) {
1956 struct btrfs_qgroup *qg;
1957
1958 qg = find_qgroup_rb(fs_info, unode->val);
1959 if (!qg)
1960 continue;
1961
1962 ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg,
1963 GFP_ATOMIC);
1964 if (ret < 0) {
1965 spin_unlock(&fs_info->qgroup_lock);
1966 ulist_free(roots);
1967 goto out;
1968 }
1969 }
1970
1971 /* this loop is similar to step 2 of btrfs_qgroup_account_ref */
1972 ULIST_ITER_INIT(&uiter);
1973 while ((unode = ulist_next(tmp, &uiter))) {
1974 struct btrfs_qgroup *qg;
1975 struct btrfs_qgroup_list *glist;
1976
1977 qg = (struct btrfs_qgroup *)(uintptr_t) unode->aux;
1978 qg->rfer += found.offset;
1979 qg->rfer_cmpr += found.offset;
1980 WARN_ON(qg->tag >= seq);
1981 if (qg->refcnt - seq == roots->nnodes) {
1982 qg->excl += found.offset;
1983 qg->excl_cmpr += found.offset;
1984 }
1985 qgroup_dirty(fs_info, qg);
1986
1987 list_for_each_entry(glist, &qg->groups, next_group) {
1988 ret = ulist_add(tmp, glist->group->qgroupid,
1989 (uintptr_t)glist->group,
1990 GFP_ATOMIC);
1991 if (ret < 0) {
1992 spin_unlock(&fs_info->qgroup_lock);
1993 ulist_free(roots);
1994 goto out;
1995 }
1996 }
1997 }
1998
1999 spin_unlock(&fs_info->qgroup_lock);
2000 ulist_free(roots);
2001 ret = 0;
2002 }
2003
2004out:
2005 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2006
2007 return ret;
2008}
2009
2010static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2011{
2012 struct qgroup_rescan *qscan = container_of(work, struct qgroup_rescan,
2013 work);
2014 struct btrfs_path *path;
2015 struct btrfs_trans_handle *trans = NULL;
2016 struct btrfs_fs_info *fs_info = qscan->fs_info;
2017 struct ulist *tmp = NULL;
2018 struct extent_buffer *scratch_leaf = NULL;
2019 int err = -ENOMEM;
2020
2021 path = btrfs_alloc_path();
2022 if (!path)
2023 goto out;
2024 tmp = ulist_alloc(GFP_NOFS);
2025 if (!tmp)
2026 goto out;
2027 scratch_leaf = kmalloc(sizeof(*scratch_leaf), GFP_NOFS);
2028 if (!scratch_leaf)
2029 goto out;
2030
2031 err = 0;
2032 while (!err) {
2033 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2034 if (IS_ERR(trans)) {
2035 err = PTR_ERR(trans);
2036 break;
2037 }
2038 if (!fs_info->quota_enabled) {
2039 err = -EINTR;
2040 } else {
2041 err = qgroup_rescan_leaf(qscan, path, trans,
2042 tmp, scratch_leaf);
2043 }
2044 if (err > 0)
2045 btrfs_commit_transaction(trans, fs_info->fs_root);
2046 else
2047 btrfs_end_transaction(trans, fs_info->fs_root);
2048 }
2049
2050out:
2051 kfree(scratch_leaf);
2052 ulist_free(tmp);
2053 btrfs_free_path(path);
2054 kfree(qscan);
2055
2056 mutex_lock(&fs_info->qgroup_rescan_lock);
2057 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2058
2059 if (err == 2 &&
2060 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2061 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2062 } else if (err < 0) {
2063 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2064 }
2065 mutex_unlock(&fs_info->qgroup_rescan_lock);
2066
2067 if (err >= 0) {
2068 pr_info("btrfs: qgroup scan completed%s\n",
2069 err == 2 ? " (inconsistency flag cleared)" : "");
2070 } else {
2071 pr_err("btrfs: qgroup scan failed with %d\n", err);
2072 }
57254b6e
JS		 2073
2074 complete_all(&fs_info->qgroup_rescan_completion);
2f232036
JS		 2075}
2076
2077static void
2078qgroup_rescan_start(struct btrfs_fs_info *fs_info, struct qgroup_rescan *qscan)
2079{
2080 memset(&qscan->work, 0, sizeof(qscan->work));
2081 qscan->work.func = btrfs_qgroup_rescan_worker;
2082 qscan->fs_info = fs_info;
2083
2084 pr_info("btrfs: qgroup scan started\n");
2085 btrfs_queue_worker(&fs_info->qgroup_rescan_workers, &qscan->work);
2086}
2087
2088int
2089btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2090{
2091 int ret = 0;
2092 struct rb_node *n;
2093 struct btrfs_qgroup *qgroup;
2094 struct qgroup_rescan *qscan = kmalloc(sizeof(*qscan), GFP_NOFS);
2095
2096 if (!qscan)
2097 return -ENOMEM;
2098
2099 mutex_lock(&fs_info->qgroup_rescan_lock);
2100 spin_lock(&fs_info->qgroup_lock);
2101 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2102 ret = -EINPROGRESS;
2103 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2104 ret = -EINVAL;
2105 if (ret) {
2106 spin_unlock(&fs_info->qgroup_lock);
2107 mutex_unlock(&fs_info->qgroup_rescan_lock);
2108 kfree(qscan);
2109 return ret;
2110 }
2111
2112 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2113 memset(&fs_info->qgroup_rescan_progress, 0,
2114 sizeof(fs_info->qgroup_rescan_progress));
57254b6e 2115 init_completion(&fs_info->qgroup_rescan_completion);
2f232036
JS		 2116
2117 /* clear all current qgroup tracking information */
2118 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2119 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2120 qgroup->rfer = 0;
2121 qgroup->rfer_cmpr = 0;
2122 qgroup->excl = 0;
2123 qgroup->excl_cmpr = 0;
2124 }
2125 spin_unlock(&fs_info->qgroup_lock);
2126 mutex_unlock(&fs_info->qgroup_rescan_lock);
2127
2128 qgroup_rescan_start(fs_info, qscan);
2129
2130 return 0;
2131}
57254b6e
JS		 2132
2133int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info)
2134{
2135 int running;
2136 int ret = 0;
2137
2138 mutex_lock(&fs_info->qgroup_rescan_lock);
2139 spin_lock(&fs_info->qgroup_lock);
2140 running = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2141 spin_unlock(&fs_info->qgroup_lock);
2142 mutex_unlock(&fs_info->qgroup_rescan_lock);
2143
2144 if (running)
2145 ret = wait_for_completion_interruptible(
2146 &fs_info->qgroup_rescan_completion);
2147
2148 return ret;
2149}
Linux 6.x block layer and io_uring tree(s)