blob: 87ab7387680f400f83772b220a52f2a490922819 [file] [log] [blame]
Arne Jansenbed92ea2012-06-28 18:03:02 +02001/*
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>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000026#include <linux/btrfs.h>
Arne Jansenbed92ea2012-06-28 18:03:02 +020027
28#include "ctree.h"
29#include "transaction.h"
30#include "disk-io.h"
31#include "locking.h"
32#include "ulist.h"
Arne Jansenbed92ea2012-06-28 18:03:02 +020033#include "backref.h"
Jan Schmidt2f232032013-04-25 16:04:51 +000034#include "extent_io.h"
Josef Bacikfcebe452014-05-13 17:30:47 -070035#include "qgroup.h"
Arne Jansenbed92ea2012-06-28 18:03:02 +020036
Qu Wenruoe69bcee2015-04-17 10:23:16 +080037
Arne Jansenbed92ea2012-06-28 18:03:02 +020038/* TODO XXX FIXME
39 * - subvol delete -> delete when ref goes to 0? delete limits also?
40 * - reorganize keys
41 * - compressed
42 * - sync
Arne Jansenbed92ea2012-06-28 18:03:02 +020043 * - copy also limits on subvol creation
44 * - limit
45 * - caches fuer ulists
46 * - performance benchmarks
47 * - check all ioctl parameters
48 */
49
50/*
51 * one struct for each qgroup, organized in fs_info->qgroup_tree.
52 */
53struct btrfs_qgroup {
54 u64 qgroupid;
55
56 /*
57 * state
58 */
59 u64 rfer; /* referenced */
60 u64 rfer_cmpr; /* referenced compressed */
61 u64 excl; /* exclusive */
62 u64 excl_cmpr; /* exclusive compressed */
63
64 /*
65 * limits
66 */
67 u64 lim_flags; /* which limits are set */
68 u64 max_rfer;
69 u64 max_excl;
70 u64 rsv_rfer;
71 u64 rsv_excl;
72
73 /*
74 * reservation tracking
75 */
76 u64 reserved;
77
78 /*
79 * lists
80 */
81 struct list_head groups; /* groups this group is member of */
82 struct list_head members; /* groups that are members of this group */
83 struct list_head dirty; /* dirty groups */
84 struct rb_node node; /* tree of qgroups */
85
86 /*
87 * temp variables for accounting operations
Nicholas D Steeves01327612016-05-19 21:18:45 -040088 * Refer to qgroup_shared_accounting() for details.
Arne Jansenbed92ea2012-06-28 18:03:02 +020089 */
Josef Bacikfcebe452014-05-13 17:30:47 -070090 u64 old_refcnt;
91 u64 new_refcnt;
Arne Jansenbed92ea2012-06-28 18:03:02 +020092};
93
Qu Wenruo9c542132015-03-12 16:10:13 +080094static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
95 int mod)
96{
97 if (qg->old_refcnt < seq)
98 qg->old_refcnt = seq;
99 qg->old_refcnt += mod;
100}
101
102static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
103 int mod)
104{
105 if (qg->new_refcnt < seq)
106 qg->new_refcnt = seq;
107 qg->new_refcnt += mod;
108}
109
110static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
111{
112 if (qg->old_refcnt < seq)
113 return 0;
114 return qg->old_refcnt - seq;
115}
116
117static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
118{
119 if (qg->new_refcnt < seq)
120 return 0;
121 return qg->new_refcnt - seq;
122}
123
Arne Jansenbed92ea2012-06-28 18:03:02 +0200124/*
125 * glue structure to represent the relations between qgroups.
126 */
127struct btrfs_qgroup_list {
128 struct list_head next_group;
129 struct list_head next_member;
130 struct btrfs_qgroup *group;
131 struct btrfs_qgroup *member;
132};
133
David Sterbaef2fff62016-10-26 16:23:50 +0200134static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
135{
136 return (u64)(uintptr_t)qg;
137}
138
139static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
140{
141 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
142}
Josef Bacikfcebe452014-05-13 17:30:47 -0700143
Jan Schmidtb382a322013-05-28 15:47:24 +0000144static int
145qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
146 int init_flags);
147static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
Jan Schmidt2f232032013-04-25 16:04:51 +0000148
Wang Shilong58400fc2013-04-07 10:50:17 +0000149/* must be called with qgroup_ioctl_lock held */
Arne Jansenbed92ea2012-06-28 18:03:02 +0200150static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
151 u64 qgroupid)
152{
153 struct rb_node *n = fs_info->qgroup_tree.rb_node;
154 struct btrfs_qgroup *qgroup;
155
156 while (n) {
157 qgroup = rb_entry(n, struct btrfs_qgroup, node);
158 if (qgroup->qgroupid < qgroupid)
159 n = n->rb_left;
160 else if (qgroup->qgroupid > qgroupid)
161 n = n->rb_right;
162 else
163 return qgroup;
164 }
165 return NULL;
166}
167
168/* must be called with qgroup_lock held */
169static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
170 u64 qgroupid)
171{
172 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
173 struct rb_node *parent = NULL;
174 struct btrfs_qgroup *qgroup;
175
176 while (*p) {
177 parent = *p;
178 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
179
180 if (qgroup->qgroupid < qgroupid)
181 p = &(*p)->rb_left;
182 else if (qgroup->qgroupid > qgroupid)
183 p = &(*p)->rb_right;
184 else
185 return qgroup;
186 }
187
188 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
189 if (!qgroup)
190 return ERR_PTR(-ENOMEM);
191
192 qgroup->qgroupid = qgroupid;
193 INIT_LIST_HEAD(&qgroup->groups);
194 INIT_LIST_HEAD(&qgroup->members);
195 INIT_LIST_HEAD(&qgroup->dirty);
196
197 rb_link_node(&qgroup->node, parent, p);
198 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
199
200 return qgroup;
201}
202
Wang Shilong4082bd32013-08-14 09:13:36 +0800203static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200204{
Arne Jansenbed92ea2012-06-28 18:03:02 +0200205 struct btrfs_qgroup_list *list;
206
Arne Jansenbed92ea2012-06-28 18:03:02 +0200207 list_del(&qgroup->dirty);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200208 while (!list_empty(&qgroup->groups)) {
209 list = list_first_entry(&qgroup->groups,
210 struct btrfs_qgroup_list, next_group);
211 list_del(&list->next_group);
212 list_del(&list->next_member);
213 kfree(list);
214 }
215
216 while (!list_empty(&qgroup->members)) {
217 list = list_first_entry(&qgroup->members,
218 struct btrfs_qgroup_list, next_member);
219 list_del(&list->next_group);
220 list_del(&list->next_member);
221 kfree(list);
222 }
223 kfree(qgroup);
Wang Shilong4082bd32013-08-14 09:13:36 +0800224}
Arne Jansenbed92ea2012-06-28 18:03:02 +0200225
Wang Shilong4082bd32013-08-14 09:13:36 +0800226/* must be called with qgroup_lock held */
227static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
228{
229 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
230
231 if (!qgroup)
232 return -ENOENT;
233
234 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
235 __del_qgroup_rb(qgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200236 return 0;
237}
238
239/* must be called with qgroup_lock held */
240static int add_relation_rb(struct btrfs_fs_info *fs_info,
241 u64 memberid, u64 parentid)
242{
243 struct btrfs_qgroup *member;
244 struct btrfs_qgroup *parent;
245 struct btrfs_qgroup_list *list;
246
247 member = find_qgroup_rb(fs_info, memberid);
248 parent = find_qgroup_rb(fs_info, parentid);
249 if (!member || !parent)
250 return -ENOENT;
251
252 list = kzalloc(sizeof(*list), GFP_ATOMIC);
253 if (!list)
254 return -ENOMEM;
255
256 list->group = parent;
257 list->member = member;
258 list_add_tail(&list->next_group, &member->groups);
259 list_add_tail(&list->next_member, &parent->members);
260
261 return 0;
262}
263
264/* must be called with qgroup_lock held */
265static int del_relation_rb(struct btrfs_fs_info *fs_info,
266 u64 memberid, u64 parentid)
267{
268 struct btrfs_qgroup *member;
269 struct btrfs_qgroup *parent;
270 struct btrfs_qgroup_list *list;
271
272 member = find_qgroup_rb(fs_info, memberid);
273 parent = find_qgroup_rb(fs_info, parentid);
274 if (!member || !parent)
275 return -ENOENT;
276
277 list_for_each_entry(list, &member->groups, next_group) {
278 if (list->group == parent) {
279 list_del(&list->next_group);
280 list_del(&list->next_member);
281 kfree(list);
282 return 0;
283 }
284 }
285 return -ENOENT;
286}
287
Josef Bacikfaa2dbf2014-05-07 17:06:09 -0400288#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
289int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
290 u64 rfer, u64 excl)
291{
292 struct btrfs_qgroup *qgroup;
293
294 qgroup = find_qgroup_rb(fs_info, qgroupid);
295 if (!qgroup)
296 return -EINVAL;
297 if (qgroup->rfer != rfer || qgroup->excl != excl)
298 return -EINVAL;
299 return 0;
300}
301#endif
302
Arne Jansenbed92ea2012-06-28 18:03:02 +0200303/*
304 * The full config is read in one go, only called from open_ctree()
305 * It doesn't use any locking, as at this point we're still single-threaded
306 */
307int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
308{
309 struct btrfs_key key;
310 struct btrfs_key found_key;
311 struct btrfs_root *quota_root = fs_info->quota_root;
312 struct btrfs_path *path = NULL;
313 struct extent_buffer *l;
314 int slot;
315 int ret = 0;
316 u64 flags = 0;
Jan Schmidtb382a322013-05-28 15:47:24 +0000317 u64 rescan_progress = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200318
Josef Bacikafcdd122016-09-02 15:40:02 -0400319 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Arne Jansenbed92ea2012-06-28 18:03:02 +0200320 return 0;
321
Wang Shilong1e8f9152013-05-06 11:03:27 +0000322 fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
323 if (!fs_info->qgroup_ulist) {
324 ret = -ENOMEM;
325 goto out;
326 }
327
Arne Jansenbed92ea2012-06-28 18:03:02 +0200328 path = btrfs_alloc_path();
329 if (!path) {
330 ret = -ENOMEM;
331 goto out;
332 }
333
334 /* default this to quota off, in case no status key is found */
335 fs_info->qgroup_flags = 0;
336
337 /*
338 * pass 1: read status, all qgroup infos and limits
339 */
340 key.objectid = 0;
341 key.type = 0;
342 key.offset = 0;
343 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
344 if (ret)
345 goto out;
346
347 while (1) {
348 struct btrfs_qgroup *qgroup;
349
350 slot = path->slots[0];
351 l = path->nodes[0];
352 btrfs_item_key_to_cpu(l, &found_key, slot);
353
354 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
355 struct btrfs_qgroup_status_item *ptr;
356
357 ptr = btrfs_item_ptr(l, slot,
358 struct btrfs_qgroup_status_item);
359
360 if (btrfs_qgroup_status_version(l, ptr) !=
361 BTRFS_QGROUP_STATUS_VERSION) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500362 btrfs_err(fs_info,
363 "old qgroup version, quota disabled");
Arne Jansenbed92ea2012-06-28 18:03:02 +0200364 goto out;
365 }
366 if (btrfs_qgroup_status_generation(l, ptr) !=
367 fs_info->generation) {
368 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Frank Holtonefe120a2013-12-20 11:37:06 -0500369 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -0400370 "qgroup generation mismatch, marked as inconsistent");
Arne Jansenbed92ea2012-06-28 18:03:02 +0200371 }
372 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
373 ptr);
Jan Schmidtb382a322013-05-28 15:47:24 +0000374 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200375 goto next1;
376 }
377
378 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
379 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
380 goto next1;
381
382 qgroup = find_qgroup_rb(fs_info, found_key.offset);
383 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
384 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
Geert Uytterhoevend41e36a2015-07-06 15:38:11 +0200385 btrfs_err(fs_info, "inconsistent qgroup config");
Arne Jansenbed92ea2012-06-28 18:03:02 +0200386 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
387 }
388 if (!qgroup) {
389 qgroup = add_qgroup_rb(fs_info, found_key.offset);
390 if (IS_ERR(qgroup)) {
391 ret = PTR_ERR(qgroup);
392 goto out;
393 }
394 }
395 switch (found_key.type) {
396 case BTRFS_QGROUP_INFO_KEY: {
397 struct btrfs_qgroup_info_item *ptr;
398
399 ptr = btrfs_item_ptr(l, slot,
400 struct btrfs_qgroup_info_item);
401 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
402 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
403 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
404 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
405 /* generation currently unused */
406 break;
407 }
408 case BTRFS_QGROUP_LIMIT_KEY: {
409 struct btrfs_qgroup_limit_item *ptr;
410
411 ptr = btrfs_item_ptr(l, slot,
412 struct btrfs_qgroup_limit_item);
413 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
414 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
415 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
416 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
417 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
418 break;
419 }
420 }
421next1:
422 ret = btrfs_next_item(quota_root, path);
423 if (ret < 0)
424 goto out;
425 if (ret)
426 break;
427 }
428 btrfs_release_path(path);
429
430 /*
431 * pass 2: read all qgroup relations
432 */
433 key.objectid = 0;
434 key.type = BTRFS_QGROUP_RELATION_KEY;
435 key.offset = 0;
436 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
437 if (ret)
438 goto out;
439 while (1) {
440 slot = path->slots[0];
441 l = path->nodes[0];
442 btrfs_item_key_to_cpu(l, &found_key, slot);
443
444 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
445 goto next2;
446
447 if (found_key.objectid > found_key.offset) {
448 /* parent <- member, not needed to build config */
449 /* FIXME should we omit the key completely? */
450 goto next2;
451 }
452
453 ret = add_relation_rb(fs_info, found_key.objectid,
454 found_key.offset);
Arne Jansenff248582013-01-17 01:22:08 -0700455 if (ret == -ENOENT) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500456 btrfs_warn(fs_info,
457 "orphan qgroup relation 0x%llx->0x%llx",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200458 found_key.objectid, found_key.offset);
Arne Jansenff248582013-01-17 01:22:08 -0700459 ret = 0; /* ignore the error */
460 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200461 if (ret)
462 goto out;
463next2:
464 ret = btrfs_next_item(quota_root, path);
465 if (ret < 0)
466 goto out;
467 if (ret)
468 break;
469 }
470out:
471 fs_info->qgroup_flags |= flags;
Josef Bacikafcdd122016-09-02 15:40:02 -0400472 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
473 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
474 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
475 ret >= 0)
Jan Schmidtb382a322013-05-28 15:47:24 +0000476 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200477 btrfs_free_path(path);
478
Jan Schmidteb1716a2013-05-28 15:47:23 +0000479 if (ret < 0) {
Wang Shilong1e8f9152013-05-06 11:03:27 +0000480 ulist_free(fs_info->qgroup_ulist);
Jan Schmidteb1716a2013-05-28 15:47:23 +0000481 fs_info->qgroup_ulist = NULL;
Jan Schmidtb382a322013-05-28 15:47:24 +0000482 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
Jan Schmidteb1716a2013-05-28 15:47:23 +0000483 }
Wang Shilong1e8f9152013-05-06 11:03:27 +0000484
Arne Jansenbed92ea2012-06-28 18:03:02 +0200485 return ret < 0 ? ret : 0;
486}
487
488/*
Wang Shilonge685da12013-08-14 09:13:37 +0800489 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
490 * first two are in single-threaded paths.And for the third one, we have set
491 * quota_root to be null with qgroup_lock held before, so it is safe to clean
492 * up the in-memory structures without qgroup_lock held.
Arne Jansenbed92ea2012-06-28 18:03:02 +0200493 */
494void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
495{
496 struct rb_node *n;
497 struct btrfs_qgroup *qgroup;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200498
499 while ((n = rb_first(&fs_info->qgroup_tree))) {
500 qgroup = rb_entry(n, struct btrfs_qgroup, node);
501 rb_erase(n, &fs_info->qgroup_tree);
Wang Shilong4082bd32013-08-14 09:13:36 +0800502 __del_qgroup_rb(qgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200503 }
Wang Shilong1e7bac12013-07-13 21:02:54 +0800504 /*
505 * we call btrfs_free_qgroup_config() when umounting
Nicholas D Steeves01327612016-05-19 21:18:45 -0400506 * filesystem and disabling quota, so we set qgroup_ulist
Wang Shilong1e7bac12013-07-13 21:02:54 +0800507 * to be null here to avoid double free.
508 */
Wang Shilong1e8f9152013-05-06 11:03:27 +0000509 ulist_free(fs_info->qgroup_ulist);
Wang Shilong1e7bac12013-07-13 21:02:54 +0800510 fs_info->qgroup_ulist = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200511}
512
513static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
514 struct btrfs_root *quota_root,
515 u64 src, u64 dst)
516{
517 int ret;
518 struct btrfs_path *path;
519 struct btrfs_key key;
520
521 path = btrfs_alloc_path();
522 if (!path)
523 return -ENOMEM;
524
525 key.objectid = src;
526 key.type = BTRFS_QGROUP_RELATION_KEY;
527 key.offset = dst;
528
529 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
530
531 btrfs_mark_buffer_dirty(path->nodes[0]);
532
533 btrfs_free_path(path);
534 return ret;
535}
536
537static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
538 struct btrfs_root *quota_root,
539 u64 src, u64 dst)
540{
541 int ret;
542 struct btrfs_path *path;
543 struct btrfs_key key;
544
545 path = btrfs_alloc_path();
546 if (!path)
547 return -ENOMEM;
548
549 key.objectid = src;
550 key.type = BTRFS_QGROUP_RELATION_KEY;
551 key.offset = dst;
552
553 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
554 if (ret < 0)
555 goto out;
556
557 if (ret > 0) {
558 ret = -ENOENT;
559 goto out;
560 }
561
562 ret = btrfs_del_item(trans, quota_root, path);
563out:
564 btrfs_free_path(path);
565 return ret;
566}
567
568static int add_qgroup_item(struct btrfs_trans_handle *trans,
569 struct btrfs_root *quota_root, u64 qgroupid)
570{
571 int ret;
572 struct btrfs_path *path;
573 struct btrfs_qgroup_info_item *qgroup_info;
574 struct btrfs_qgroup_limit_item *qgroup_limit;
575 struct extent_buffer *leaf;
576 struct btrfs_key key;
577
Jeff Mahoneyf5ee5c92016-06-21 09:52:41 -0400578 if (btrfs_is_testing(quota_root->fs_info))
Josef Bacikfaa2dbf2014-05-07 17:06:09 -0400579 return 0;
David Sterbafccb84c2014-09-29 23:53:21 +0200580
Arne Jansenbed92ea2012-06-28 18:03:02 +0200581 path = btrfs_alloc_path();
582 if (!path)
583 return -ENOMEM;
584
585 key.objectid = 0;
586 key.type = BTRFS_QGROUP_INFO_KEY;
587 key.offset = qgroupid;
588
Mark Fasheh0b4699d2014-08-18 14:01:17 -0700589 /*
590 * Avoid a transaction abort by catching -EEXIST here. In that
591 * case, we proceed by re-initializing the existing structure
592 * on disk.
593 */
594
Arne Jansenbed92ea2012-06-28 18:03:02 +0200595 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
596 sizeof(*qgroup_info));
Mark Fasheh0b4699d2014-08-18 14:01:17 -0700597 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200598 goto out;
599
600 leaf = path->nodes[0];
601 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
602 struct btrfs_qgroup_info_item);
603 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
604 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
605 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
606 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
607 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
608
609 btrfs_mark_buffer_dirty(leaf);
610
611 btrfs_release_path(path);
612
613 key.type = BTRFS_QGROUP_LIMIT_KEY;
614 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
615 sizeof(*qgroup_limit));
Mark Fasheh0b4699d2014-08-18 14:01:17 -0700616 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200617 goto out;
618
619 leaf = path->nodes[0];
620 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
621 struct btrfs_qgroup_limit_item);
622 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
623 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
624 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
625 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
626 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
627
628 btrfs_mark_buffer_dirty(leaf);
629
630 ret = 0;
631out:
632 btrfs_free_path(path);
633 return ret;
634}
635
636static int del_qgroup_item(struct btrfs_trans_handle *trans,
637 struct btrfs_root *quota_root, u64 qgroupid)
638{
639 int ret;
640 struct btrfs_path *path;
641 struct btrfs_key key;
642
643 path = btrfs_alloc_path();
644 if (!path)
645 return -ENOMEM;
646
647 key.objectid = 0;
648 key.type = BTRFS_QGROUP_INFO_KEY;
649 key.offset = qgroupid;
650 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
651 if (ret < 0)
652 goto out;
653
654 if (ret > 0) {
655 ret = -ENOENT;
656 goto out;
657 }
658
659 ret = btrfs_del_item(trans, quota_root, path);
660 if (ret)
661 goto out;
662
663 btrfs_release_path(path);
664
665 key.type = BTRFS_QGROUP_LIMIT_KEY;
666 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
667 if (ret < 0)
668 goto out;
669
670 if (ret > 0) {
671 ret = -ENOENT;
672 goto out;
673 }
674
675 ret = btrfs_del_item(trans, quota_root, path);
676
677out:
678 btrfs_free_path(path);
679 return ret;
680}
681
682static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
Dongsheng Yang1510e712014-11-20 21:01:41 -0500683 struct btrfs_root *root,
684 struct btrfs_qgroup *qgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200685{
686 struct btrfs_path *path;
687 struct btrfs_key key;
688 struct extent_buffer *l;
689 struct btrfs_qgroup_limit_item *qgroup_limit;
690 int ret;
691 int slot;
692
693 key.objectid = 0;
694 key.type = BTRFS_QGROUP_LIMIT_KEY;
Dongsheng Yang1510e712014-11-20 21:01:41 -0500695 key.offset = qgroup->qgroupid;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200696
697 path = btrfs_alloc_path();
Wang Shilong84cbe2f2013-02-27 11:20:56 +0000698 if (!path)
699 return -ENOMEM;
700
Arne Jansenbed92ea2012-06-28 18:03:02 +0200701 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
702 if (ret > 0)
703 ret = -ENOENT;
704
705 if (ret)
706 goto out;
707
708 l = path->nodes[0];
709 slot = path->slots[0];
Valentina Giustia3df41e2013-11-04 22:34:29 +0100710 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
Dongsheng Yang1510e712014-11-20 21:01:41 -0500711 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
712 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
713 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
714 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
715 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200716
717 btrfs_mark_buffer_dirty(l);
718
719out:
720 btrfs_free_path(path);
721 return ret;
722}
723
724static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
725 struct btrfs_root *root,
726 struct btrfs_qgroup *qgroup)
727{
728 struct btrfs_path *path;
729 struct btrfs_key key;
730 struct extent_buffer *l;
731 struct btrfs_qgroup_info_item *qgroup_info;
732 int ret;
733 int slot;
734
Jeff Mahoneyf5ee5c92016-06-21 09:52:41 -0400735 if (btrfs_is_testing(root->fs_info))
Josef Bacikfaa2dbf2014-05-07 17:06:09 -0400736 return 0;
David Sterbafccb84c2014-09-29 23:53:21 +0200737
Arne Jansenbed92ea2012-06-28 18:03:02 +0200738 key.objectid = 0;
739 key.type = BTRFS_QGROUP_INFO_KEY;
740 key.offset = qgroup->qgroupid;
741
742 path = btrfs_alloc_path();
Wang Shilong84cbe2f2013-02-27 11:20:56 +0000743 if (!path)
744 return -ENOMEM;
745
Arne Jansenbed92ea2012-06-28 18:03:02 +0200746 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
747 if (ret > 0)
748 ret = -ENOENT;
749
750 if (ret)
751 goto out;
752
753 l = path->nodes[0];
754 slot = path->slots[0];
Valentina Giustia3df41e2013-11-04 22:34:29 +0100755 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200756 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
757 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
758 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
759 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
760 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
761
762 btrfs_mark_buffer_dirty(l);
763
764out:
765 btrfs_free_path(path);
766 return ret;
767}
768
769static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
770 struct btrfs_fs_info *fs_info,
771 struct btrfs_root *root)
772{
773 struct btrfs_path *path;
774 struct btrfs_key key;
775 struct extent_buffer *l;
776 struct btrfs_qgroup_status_item *ptr;
777 int ret;
778 int slot;
779
780 key.objectid = 0;
781 key.type = BTRFS_QGROUP_STATUS_KEY;
782 key.offset = 0;
783
784 path = btrfs_alloc_path();
Wang Shilong84cbe2f2013-02-27 11:20:56 +0000785 if (!path)
786 return -ENOMEM;
787
Arne Jansenbed92ea2012-06-28 18:03:02 +0200788 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
789 if (ret > 0)
790 ret = -ENOENT;
791
792 if (ret)
793 goto out;
794
795 l = path->nodes[0];
796 slot = path->slots[0];
797 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
798 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
799 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
Jan Schmidt2f232032013-04-25 16:04:51 +0000800 btrfs_set_qgroup_status_rescan(l, ptr,
801 fs_info->qgroup_rescan_progress.objectid);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200802
803 btrfs_mark_buffer_dirty(l);
804
805out:
806 btrfs_free_path(path);
807 return ret;
808}
809
810/*
811 * called with qgroup_lock held
812 */
813static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
814 struct btrfs_root *root)
815{
816 struct btrfs_path *path;
817 struct btrfs_key key;
Wang Shilong06b3a862013-02-27 11:16:57 +0000818 struct extent_buffer *leaf = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200819 int ret;
Wang Shilong06b3a862013-02-27 11:16:57 +0000820 int nr = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200821
Arne Jansenbed92ea2012-06-28 18:03:02 +0200822 path = btrfs_alloc_path();
823 if (!path)
824 return -ENOMEM;
825
Wang Shilong06b3a862013-02-27 11:16:57 +0000826 path->leave_spinning = 1;
827
828 key.objectid = 0;
829 key.offset = 0;
830 key.type = 0;
831
Arne Jansenbed92ea2012-06-28 18:03:02 +0200832 while (1) {
Arne Jansenbed92ea2012-06-28 18:03:02 +0200833 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Wang Shilong06b3a862013-02-27 11:16:57 +0000834 if (ret < 0)
835 goto out;
836 leaf = path->nodes[0];
837 nr = btrfs_header_nritems(leaf);
838 if (!nr)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200839 break;
Wang Shilong06b3a862013-02-27 11:16:57 +0000840 /*
841 * delete the leaf one by one
842 * since the whole tree is going
843 * to be deleted.
844 */
845 path->slots[0] = 0;
846 ret = btrfs_del_items(trans, root, path, 0, nr);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200847 if (ret)
848 goto out;
Wang Shilong06b3a862013-02-27 11:16:57 +0000849
Arne Jansenbed92ea2012-06-28 18:03:02 +0200850 btrfs_release_path(path);
851 }
852 ret = 0;
853out:
Josef Bacikafcdd122016-09-02 15:40:02 -0400854 set_bit(BTRFS_FS_QUOTA_DISABLING, &root->fs_info->flags);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200855 btrfs_free_path(path);
856 return ret;
857}
858
859int btrfs_quota_enable(struct btrfs_trans_handle *trans,
860 struct btrfs_fs_info *fs_info)
861{
862 struct btrfs_root *quota_root;
Wang Shilong7708f022013-04-07 10:24:57 +0000863 struct btrfs_root *tree_root = fs_info->tree_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200864 struct btrfs_path *path = NULL;
865 struct btrfs_qgroup_status_item *ptr;
866 struct extent_buffer *leaf;
867 struct btrfs_key key;
Wang Shilong7708f022013-04-07 10:24:57 +0000868 struct btrfs_key found_key;
869 struct btrfs_qgroup *qgroup = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200870 int ret = 0;
Wang Shilong7708f022013-04-07 10:24:57 +0000871 int slot;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200872
Wang Shilongf2f6ed32013-04-07 10:50:16 +0000873 mutex_lock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200874 if (fs_info->quota_root) {
Josef Bacikafcdd122016-09-02 15:40:02 -0400875 set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200876 goto out;
877 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200878
Wang Shilong1e8f9152013-05-06 11:03:27 +0000879 fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
880 if (!fs_info->qgroup_ulist) {
881 ret = -ENOMEM;
882 goto out;
883 }
884
Arne Jansenbed92ea2012-06-28 18:03:02 +0200885 /*
886 * initially create the quota tree
887 */
888 quota_root = btrfs_create_tree(trans, fs_info,
889 BTRFS_QUOTA_TREE_OBJECTID);
890 if (IS_ERR(quota_root)) {
891 ret = PTR_ERR(quota_root);
892 goto out;
893 }
894
895 path = btrfs_alloc_path();
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000896 if (!path) {
897 ret = -ENOMEM;
898 goto out_free_root;
899 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200900
901 key.objectid = 0;
902 key.type = BTRFS_QGROUP_STATUS_KEY;
903 key.offset = 0;
904
905 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
906 sizeof(*ptr));
907 if (ret)
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000908 goto out_free_path;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200909
910 leaf = path->nodes[0];
911 ptr = btrfs_item_ptr(leaf, path->slots[0],
912 struct btrfs_qgroup_status_item);
913 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
914 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
915 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
916 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
917 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
Jan Schmidt2f232032013-04-25 16:04:51 +0000918 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200919
920 btrfs_mark_buffer_dirty(leaf);
921
Wang Shilong7708f022013-04-07 10:24:57 +0000922 key.objectid = 0;
923 key.type = BTRFS_ROOT_REF_KEY;
924 key.offset = 0;
925
926 btrfs_release_path(path);
927 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
928 if (ret > 0)
929 goto out_add_root;
930 if (ret < 0)
931 goto out_free_path;
932
933
934 while (1) {
935 slot = path->slots[0];
936 leaf = path->nodes[0];
937 btrfs_item_key_to_cpu(leaf, &found_key, slot);
938
939 if (found_key.type == BTRFS_ROOT_REF_KEY) {
940 ret = add_qgroup_item(trans, quota_root,
941 found_key.offset);
942 if (ret)
943 goto out_free_path;
944
Wang Shilong7708f022013-04-07 10:24:57 +0000945 qgroup = add_qgroup_rb(fs_info, found_key.offset);
946 if (IS_ERR(qgroup)) {
Wang Shilong7708f022013-04-07 10:24:57 +0000947 ret = PTR_ERR(qgroup);
948 goto out_free_path;
949 }
Wang Shilong7708f022013-04-07 10:24:57 +0000950 }
951 ret = btrfs_next_item(tree_root, path);
952 if (ret < 0)
953 goto out_free_path;
954 if (ret)
955 break;
956 }
957
958out_add_root:
959 btrfs_release_path(path);
960 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
961 if (ret)
962 goto out_free_path;
963
Wang Shilong7708f022013-04-07 10:24:57 +0000964 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
965 if (IS_ERR(qgroup)) {
Wang Shilong7708f022013-04-07 10:24:57 +0000966 ret = PTR_ERR(qgroup);
967 goto out_free_path;
968 }
Wang Shilong58400fc2013-04-07 10:50:17 +0000969 spin_lock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200970 fs_info->quota_root = quota_root;
Josef Bacikafcdd122016-09-02 15:40:02 -0400971 set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200972 spin_unlock(&fs_info->qgroup_lock);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000973out_free_path:
Arne Jansenbed92ea2012-06-28 18:03:02 +0200974 btrfs_free_path(path);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000975out_free_root:
976 if (ret) {
977 free_extent_buffer(quota_root->node);
978 free_extent_buffer(quota_root->commit_root);
979 kfree(quota_root);
980 }
981out:
Jan Schmidteb1716a2013-05-28 15:47:23 +0000982 if (ret) {
Wang Shilong1e8f9152013-05-06 11:03:27 +0000983 ulist_free(fs_info->qgroup_ulist);
Jan Schmidteb1716a2013-05-28 15:47:23 +0000984 fs_info->qgroup_ulist = NULL;
985 }
Wang Shilongf2f6ed32013-04-07 10:50:16 +0000986 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200987 return ret;
988}
989
990int btrfs_quota_disable(struct btrfs_trans_handle *trans,
991 struct btrfs_fs_info *fs_info)
992{
993 struct btrfs_root *tree_root = fs_info->tree_root;
994 struct btrfs_root *quota_root;
995 int ret = 0;
996
Wang Shilongf2f6ed32013-04-07 10:50:16 +0000997 mutex_lock(&fs_info->qgroup_ioctl_lock);
Wang Shilong58400fc2013-04-07 10:50:17 +0000998 if (!fs_info->quota_root)
Wang Shilongf2f6ed32013-04-07 10:50:16 +0000999 goto out;
Josef Bacikafcdd122016-09-02 15:40:02 -04001000 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1001 set_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags);
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04001002 btrfs_qgroup_wait_for_completion(fs_info, false);
Justin Maggard967ef512015-11-06 10:36:42 -08001003 spin_lock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001004 quota_root = fs_info->quota_root;
1005 fs_info->quota_root = NULL;
Dongsheng Yang8ea0ec92015-02-27 16:24:26 +08001006 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001007 spin_unlock(&fs_info->qgroup_lock);
1008
Wang Shilonge685da12013-08-14 09:13:37 +08001009 btrfs_free_qgroup_config(fs_info);
1010
Arne Jansenbed92ea2012-06-28 18:03:02 +02001011 ret = btrfs_clean_quota_tree(trans, quota_root);
1012 if (ret)
1013 goto out;
1014
1015 ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
1016 if (ret)
1017 goto out;
1018
1019 list_del(&quota_root->dirty_list);
1020
1021 btrfs_tree_lock(quota_root->node);
Daniel Dressler01d58472014-11-21 17:15:07 +09001022 clean_tree_block(trans, tree_root->fs_info, quota_root->node);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001023 btrfs_tree_unlock(quota_root->node);
1024 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1025
1026 free_extent_buffer(quota_root->node);
1027 free_extent_buffer(quota_root->commit_root);
1028 kfree(quota_root);
1029out:
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001030 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001031 return ret;
1032}
1033
Jan Schmidt2f232032013-04-25 16:04:51 +00001034static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1035 struct btrfs_qgroup *qgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001036{
Jan Schmidt2f232032013-04-25 16:04:51 +00001037 if (list_empty(&qgroup->dirty))
1038 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001039}
1040
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001041/*
1042 * The easy accounting, if we are adding/removing the only ref for an extent
Nicholas D Steeves01327612016-05-19 21:18:45 -04001043 * then this qgroup and all of the parent qgroups get their reference and
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001044 * exclusive counts adjusted.
1045 *
1046 * Caller should hold fs_info->qgroup_lock.
1047 */
1048static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1049 struct ulist *tmp, u64 ref_root,
1050 u64 num_bytes, int sign)
1051{
1052 struct btrfs_qgroup *qgroup;
1053 struct btrfs_qgroup_list *glist;
1054 struct ulist_node *unode;
1055 struct ulist_iterator uiter;
1056 int ret = 0;
1057
1058 qgroup = find_qgroup_rb(fs_info, ref_root);
1059 if (!qgroup)
1060 goto out;
1061
1062 qgroup->rfer += sign * num_bytes;
1063 qgroup->rfer_cmpr += sign * num_bytes;
1064
1065 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1066 qgroup->excl += sign * num_bytes;
1067 qgroup->excl_cmpr += sign * num_bytes;
1068 if (sign > 0)
1069 qgroup->reserved -= num_bytes;
1070
1071 qgroup_dirty(fs_info, qgroup);
1072
1073 /* Get all of the parent groups that contain this qgroup */
1074 list_for_each_entry(glist, &qgroup->groups, next_group) {
1075 ret = ulist_add(tmp, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02001076 qgroup_to_aux(glist->group), GFP_ATOMIC);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001077 if (ret < 0)
1078 goto out;
1079 }
1080
1081 /* Iterate all of the parents and adjust their reference counts */
1082 ULIST_ITER_INIT(&uiter);
1083 while ((unode = ulist_next(tmp, &uiter))) {
David Sterbaef2fff62016-10-26 16:23:50 +02001084 qgroup = unode_aux_to_qgroup(unode);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001085 qgroup->rfer += sign * num_bytes;
1086 qgroup->rfer_cmpr += sign * num_bytes;
1087 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1088 qgroup->excl += sign * num_bytes;
1089 if (sign > 0)
1090 qgroup->reserved -= num_bytes;
1091 qgroup->excl_cmpr += sign * num_bytes;
1092 qgroup_dirty(fs_info, qgroup);
1093
1094 /* Add any parents of the parents */
1095 list_for_each_entry(glist, &qgroup->groups, next_group) {
1096 ret = ulist_add(tmp, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02001097 qgroup_to_aux(glist->group), GFP_ATOMIC);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001098 if (ret < 0)
1099 goto out;
1100 }
1101 }
1102 ret = 0;
1103out:
1104 return ret;
1105}
1106
1107
1108/*
1109 * Quick path for updating qgroup with only excl refs.
1110 *
1111 * In that case, just update all parent will be enough.
1112 * Or we needs to do a full rescan.
1113 * Caller should also hold fs_info->qgroup_lock.
1114 *
1115 * Return 0 for quick update, return >0 for need to full rescan
1116 * and mark INCONSISTENT flag.
1117 * Return < 0 for other error.
1118 */
1119static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1120 struct ulist *tmp, u64 src, u64 dst,
1121 int sign)
1122{
1123 struct btrfs_qgroup *qgroup;
1124 int ret = 1;
1125 int err = 0;
1126
1127 qgroup = find_qgroup_rb(fs_info, src);
1128 if (!qgroup)
1129 goto out;
1130 if (qgroup->excl == qgroup->rfer) {
1131 ret = 0;
1132 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1133 qgroup->excl, sign);
1134 if (err < 0) {
1135 ret = err;
1136 goto out;
1137 }
1138 }
1139out:
1140 if (ret)
1141 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1142 return ret;
1143}
1144
Arne Jansenbed92ea2012-06-28 18:03:02 +02001145int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1146 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1147{
1148 struct btrfs_root *quota_root;
Wang Shilongb7fef4f2013-04-07 10:50:18 +00001149 struct btrfs_qgroup *parent;
1150 struct btrfs_qgroup *member;
Wang Shilong534e6622013-04-17 14:49:51 +00001151 struct btrfs_qgroup_list *list;
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001152 struct ulist *tmp;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001153 int ret = 0;
1154
Qu Wenruo8465ece2015-02-27 16:24:22 +08001155 /* Check the level of src and dst first */
1156 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1157 return -EINVAL;
1158
Christian Engelmayerab3680d2015-05-02 17:19:55 +02001159 tmp = ulist_alloc(GFP_NOFS);
1160 if (!tmp)
1161 return -ENOMEM;
1162
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001163 mutex_lock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001164 quota_root = fs_info->quota_root;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001165 if (!quota_root) {
1166 ret = -EINVAL;
1167 goto out;
1168 }
Wang Shilongb7fef4f2013-04-07 10:50:18 +00001169 member = find_qgroup_rb(fs_info, src);
1170 parent = find_qgroup_rb(fs_info, dst);
1171 if (!member || !parent) {
1172 ret = -EINVAL;
1173 goto out;
1174 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001175
Wang Shilong534e6622013-04-17 14:49:51 +00001176 /* check if such qgroup relation exist firstly */
1177 list_for_each_entry(list, &member->groups, next_group) {
1178 if (list->group == parent) {
1179 ret = -EEXIST;
1180 goto out;
1181 }
1182 }
1183
Arne Jansenbed92ea2012-06-28 18:03:02 +02001184 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1185 if (ret)
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001186 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001187
1188 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1189 if (ret) {
1190 del_qgroup_relation_item(trans, quota_root, src, dst);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001191 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001192 }
1193
1194 spin_lock(&fs_info->qgroup_lock);
1195 ret = add_relation_rb(quota_root->fs_info, src, dst);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001196 if (ret < 0) {
1197 spin_unlock(&fs_info->qgroup_lock);
1198 goto out;
1199 }
1200 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001201 spin_unlock(&fs_info->qgroup_lock);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001202out:
1203 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001204 ulist_free(tmp);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001205 return ret;
1206}
1207
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001208int __del_qgroup_relation(struct btrfs_trans_handle *trans,
Arne Jansenbed92ea2012-06-28 18:03:02 +02001209 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1210{
1211 struct btrfs_root *quota_root;
Wang Shilong534e6622013-04-17 14:49:51 +00001212 struct btrfs_qgroup *parent;
1213 struct btrfs_qgroup *member;
1214 struct btrfs_qgroup_list *list;
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001215 struct ulist *tmp;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001216 int ret = 0;
1217 int err;
1218
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001219 tmp = ulist_alloc(GFP_NOFS);
1220 if (!tmp)
1221 return -ENOMEM;
1222
Arne Jansenbed92ea2012-06-28 18:03:02 +02001223 quota_root = fs_info->quota_root;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001224 if (!quota_root) {
1225 ret = -EINVAL;
1226 goto out;
1227 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001228
Wang Shilong534e6622013-04-17 14:49:51 +00001229 member = find_qgroup_rb(fs_info, src);
1230 parent = find_qgroup_rb(fs_info, dst);
1231 if (!member || !parent) {
1232 ret = -EINVAL;
1233 goto out;
1234 }
1235
1236 /* check if such qgroup relation exist firstly */
1237 list_for_each_entry(list, &member->groups, next_group) {
1238 if (list->group == parent)
1239 goto exist;
1240 }
1241 ret = -ENOENT;
1242 goto out;
1243exist:
Arne Jansenbed92ea2012-06-28 18:03:02 +02001244 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1245 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1246 if (err && !ret)
1247 ret = err;
1248
1249 spin_lock(&fs_info->qgroup_lock);
1250 del_relation_rb(fs_info, src, dst);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001251 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001252 spin_unlock(&fs_info->qgroup_lock);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001253out:
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001254 ulist_free(tmp);
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001255 return ret;
1256}
1257
1258int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1260{
1261 int ret = 0;
1262
1263 mutex_lock(&fs_info->qgroup_ioctl_lock);
1264 ret = __del_qgroup_relation(trans, fs_info, src, dst);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001265 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001266
Arne Jansenbed92ea2012-06-28 18:03:02 +02001267 return ret;
1268}
1269
1270int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
Dongsheng Yang4087cf22015-01-18 10:59:23 -05001271 struct btrfs_fs_info *fs_info, u64 qgroupid)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001272{
1273 struct btrfs_root *quota_root;
1274 struct btrfs_qgroup *qgroup;
1275 int ret = 0;
1276
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001277 mutex_lock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001278 quota_root = fs_info->quota_root;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001279 if (!quota_root) {
1280 ret = -EINVAL;
1281 goto out;
1282 }
Wang Shilong534e6622013-04-17 14:49:51 +00001283 qgroup = find_qgroup_rb(fs_info, qgroupid);
1284 if (qgroup) {
1285 ret = -EEXIST;
1286 goto out;
1287 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001288
1289 ret = add_qgroup_item(trans, quota_root, qgroupid);
Wang Shilong534e6622013-04-17 14:49:51 +00001290 if (ret)
1291 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001292
1293 spin_lock(&fs_info->qgroup_lock);
1294 qgroup = add_qgroup_rb(fs_info, qgroupid);
1295 spin_unlock(&fs_info->qgroup_lock);
1296
1297 if (IS_ERR(qgroup))
1298 ret = PTR_ERR(qgroup);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001299out:
1300 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001301 return ret;
1302}
1303
1304int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1305 struct btrfs_fs_info *fs_info, u64 qgroupid)
1306{
1307 struct btrfs_root *quota_root;
Arne Jansen2cf68702013-01-17 01:22:09 -07001308 struct btrfs_qgroup *qgroup;
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001309 struct btrfs_qgroup_list *list;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001310 int ret = 0;
1311
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001312 mutex_lock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001313 quota_root = fs_info->quota_root;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001314 if (!quota_root) {
1315 ret = -EINVAL;
1316 goto out;
1317 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001318
Arne Jansen2cf68702013-01-17 01:22:09 -07001319 qgroup = find_qgroup_rb(fs_info, qgroupid);
Wang Shilong534e6622013-04-17 14:49:51 +00001320 if (!qgroup) {
1321 ret = -ENOENT;
1322 goto out;
1323 } else {
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001324 /* check if there are no children of this qgroup */
1325 if (!list_empty(&qgroup->members)) {
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001326 ret = -EBUSY;
1327 goto out;
Arne Jansen2cf68702013-01-17 01:22:09 -07001328 }
1329 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001330 ret = del_qgroup_item(trans, quota_root, qgroupid);
1331
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001332 while (!list_empty(&qgroup->groups)) {
1333 list = list_first_entry(&qgroup->groups,
1334 struct btrfs_qgroup_list, next_group);
1335 ret = __del_qgroup_relation(trans, fs_info,
1336 qgroupid,
1337 list->group->qgroupid);
1338 if (ret)
1339 goto out;
1340 }
1341
Arne Jansenbed92ea2012-06-28 18:03:02 +02001342 spin_lock(&fs_info->qgroup_lock);
1343 del_qgroup_rb(quota_root->fs_info, qgroupid);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001344 spin_unlock(&fs_info->qgroup_lock);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001345out:
1346 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001347 return ret;
1348}
1349
1350int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1351 struct btrfs_fs_info *fs_info, u64 qgroupid,
1352 struct btrfs_qgroup_limit *limit)
1353{
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001354 struct btrfs_root *quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001355 struct btrfs_qgroup *qgroup;
1356 int ret = 0;
Yang Dongshengfe759902015-06-03 14:57:32 +08001357 /* Sometimes we would want to clear the limit on this qgroup.
1358 * To meet this requirement, we treat the -1 as a special value
1359 * which tell kernel to clear the limit on this qgroup.
1360 */
1361 const u64 CLEAR_VALUE = -1;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001362
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001363 mutex_lock(&fs_info->qgroup_ioctl_lock);
1364 quota_root = fs_info->quota_root;
1365 if (!quota_root) {
1366 ret = -EINVAL;
1367 goto out;
1368 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001369
Wang Shilongddb47af2013-04-07 10:50:20 +00001370 qgroup = find_qgroup_rb(fs_info, qgroupid);
1371 if (!qgroup) {
1372 ret = -ENOENT;
1373 goto out;
1374 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001375
Wang Shilong58400fc2013-04-07 10:50:17 +00001376 spin_lock(&fs_info->qgroup_lock);
Yang Dongshengfe759902015-06-03 14:57:32 +08001377 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1378 if (limit->max_rfer == CLEAR_VALUE) {
1379 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1380 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1381 qgroup->max_rfer = 0;
1382 } else {
1383 qgroup->max_rfer = limit->max_rfer;
1384 }
1385 }
1386 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1387 if (limit->max_excl == CLEAR_VALUE) {
1388 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1389 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1390 qgroup->max_excl = 0;
1391 } else {
1392 qgroup->max_excl = limit->max_excl;
1393 }
1394 }
1395 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1396 if (limit->rsv_rfer == CLEAR_VALUE) {
1397 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1398 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1399 qgroup->rsv_rfer = 0;
1400 } else {
1401 qgroup->rsv_rfer = limit->rsv_rfer;
1402 }
1403 }
1404 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1405 if (limit->rsv_excl == CLEAR_VALUE) {
1406 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1407 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1408 qgroup->rsv_excl = 0;
1409 } else {
1410 qgroup->rsv_excl = limit->rsv_excl;
1411 }
1412 }
Dongsheng Yang03477d92015-02-06 11:06:25 -05001413 qgroup->lim_flags |= limit->flags;
1414
Arne Jansenbed92ea2012-06-28 18:03:02 +02001415 spin_unlock(&fs_info->qgroup_lock);
Dongsheng Yang1510e712014-11-20 21:01:41 -05001416
1417 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1418 if (ret) {
1419 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1420 btrfs_info(fs_info, "unable to update quota limit for %llu",
1421 qgroupid);
1422 }
1423
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001424out:
1425 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001426 return ret;
1427}
Mark Fasheh11526512014-07-17 12:39:01 -07001428
Qu Wenruo3b7d00f2015-04-16 16:40:39 +08001429int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1430 struct btrfs_fs_info *fs_info)
1431{
1432 struct btrfs_qgroup_extent_record *record;
1433 struct btrfs_delayed_ref_root *delayed_refs;
1434 struct rb_node *node;
Qu Wenruo9086db82015-04-20 09:53:50 +08001435 u64 qgroup_to_skip;
Qu Wenruo3b7d00f2015-04-16 16:40:39 +08001436 int ret = 0;
1437
1438 delayed_refs = &trans->transaction->delayed_refs;
Qu Wenruo9086db82015-04-20 09:53:50 +08001439 qgroup_to_skip = delayed_refs->qgroup_to_skip;
Qu Wenruo3b7d00f2015-04-16 16:40:39 +08001440
1441 /*
1442 * No need to do lock, since this function will only be called in
Nicholas D Steeves01327612016-05-19 21:18:45 -04001443 * btrfs_commit_transaction().
Qu Wenruo3b7d00f2015-04-16 16:40:39 +08001444 */
1445 node = rb_first(&delayed_refs->dirty_extent_root);
1446 while (node) {
1447 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1448 node);
1449 ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0,
1450 &record->old_roots);
1451 if (ret < 0)
1452 break;
Qu Wenruo9086db82015-04-20 09:53:50 +08001453 if (qgroup_to_skip)
1454 ulist_del(record->old_roots, qgroup_to_skip, 0);
Qu Wenruo3b7d00f2015-04-16 16:40:39 +08001455 node = rb_next(node);
1456 }
1457 return ret;
1458}
1459
Qu Wenruo50b3e042016-10-18 09:31:27 +08001460int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
Qu Wenruocb93b522016-08-15 10:36:50 +08001461 struct btrfs_delayed_ref_root *delayed_refs,
1462 struct btrfs_qgroup_extent_record *record)
Qu Wenruo3368d002015-04-16 14:34:17 +08001463{
1464 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1465 struct rb_node *parent_node = NULL;
1466 struct btrfs_qgroup_extent_record *entry;
1467 u64 bytenr = record->bytenr;
1468
Mark Fasheh82bd1012015-11-05 14:38:00 -08001469 assert_spin_locked(&delayed_refs->lock);
Qu Wenruo50b3e042016-10-18 09:31:27 +08001470 trace_btrfs_qgroup_trace_extent(fs_info, record);
Mark Fasheh82bd1012015-11-05 14:38:00 -08001471
Qu Wenruo3368d002015-04-16 14:34:17 +08001472 while (*p) {
1473 parent_node = *p;
1474 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1475 node);
1476 if (bytenr < entry->bytenr)
1477 p = &(*p)->rb_left;
1478 else if (bytenr > entry->bytenr)
1479 p = &(*p)->rb_right;
1480 else
Qu Wenruocb93b522016-08-15 10:36:50 +08001481 return 1;
Qu Wenruo3368d002015-04-16 14:34:17 +08001482 }
1483
1484 rb_link_node(&record->node, parent_node, p);
1485 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
Qu Wenruocb93b522016-08-15 10:36:50 +08001486 return 0;
1487}
1488
Qu Wenruo50b3e042016-10-18 09:31:27 +08001489int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
Qu Wenruocb93b522016-08-15 10:36:50 +08001490 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1491 gfp_t gfp_flag)
1492{
1493 struct btrfs_qgroup_extent_record *record;
1494 struct btrfs_delayed_ref_root *delayed_refs;
1495 int ret;
1496
Josef Bacikafcdd122016-09-02 15:40:02 -04001497 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1498 || bytenr == 0 || num_bytes == 0)
Qu Wenruocb93b522016-08-15 10:36:50 +08001499 return 0;
1500 if (WARN_ON(trans == NULL))
1501 return -EINVAL;
1502 record = kmalloc(sizeof(*record), gfp_flag);
1503 if (!record)
1504 return -ENOMEM;
1505
1506 delayed_refs = &trans->transaction->delayed_refs;
1507 record->bytenr = bytenr;
1508 record->num_bytes = num_bytes;
1509 record->old_roots = NULL;
1510
1511 spin_lock(&delayed_refs->lock);
Qu Wenruo50b3e042016-10-18 09:31:27 +08001512 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs,
Qu Wenruocb93b522016-08-15 10:36:50 +08001513 record);
1514 spin_unlock(&delayed_refs->lock);
1515 if (ret > 0)
1516 kfree(record);
1517 return 0;
Qu Wenruo3368d002015-04-16 14:34:17 +08001518}
1519
Qu Wenruod810ef22015-04-12 16:52:34 +08001520#define UPDATE_NEW 0
1521#define UPDATE_OLD 1
1522/*
1523 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1524 */
1525static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1526 struct ulist *roots, struct ulist *tmp,
1527 struct ulist *qgroups, u64 seq, int update_old)
1528{
1529 struct ulist_node *unode;
1530 struct ulist_iterator uiter;
1531 struct ulist_node *tmp_unode;
1532 struct ulist_iterator tmp_uiter;
1533 struct btrfs_qgroup *qg;
1534 int ret = 0;
1535
1536 if (!roots)
1537 return 0;
1538 ULIST_ITER_INIT(&uiter);
1539 while ((unode = ulist_next(roots, &uiter))) {
1540 qg = find_qgroup_rb(fs_info, unode->val);
1541 if (!qg)
1542 continue;
1543
1544 ulist_reinit(tmp);
David Sterbaef2fff62016-10-26 16:23:50 +02001545 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
Qu Wenruod810ef22015-04-12 16:52:34 +08001546 GFP_ATOMIC);
1547 if (ret < 0)
1548 return ret;
David Sterbaef2fff62016-10-26 16:23:50 +02001549 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
Qu Wenruod810ef22015-04-12 16:52:34 +08001550 if (ret < 0)
1551 return ret;
1552 ULIST_ITER_INIT(&tmp_uiter);
1553 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1554 struct btrfs_qgroup_list *glist;
1555
David Sterbaef2fff62016-10-26 16:23:50 +02001556 qg = unode_aux_to_qgroup(tmp_unode);
Qu Wenruod810ef22015-04-12 16:52:34 +08001557 if (update_old)
1558 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1559 else
1560 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1561 list_for_each_entry(glist, &qg->groups, next_group) {
1562 ret = ulist_add(qgroups, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02001563 qgroup_to_aux(glist->group),
Qu Wenruod810ef22015-04-12 16:52:34 +08001564 GFP_ATOMIC);
1565 if (ret < 0)
1566 return ret;
1567 ret = ulist_add(tmp, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02001568 qgroup_to_aux(glist->group),
Qu Wenruod810ef22015-04-12 16:52:34 +08001569 GFP_ATOMIC);
1570 if (ret < 0)
1571 return ret;
1572 }
1573 }
1574 }
1575 return 0;
1576}
1577
Josef Bacikfcebe452014-05-13 17:30:47 -07001578/*
Qu Wenruo823ae5b2015-04-12 16:59:57 +08001579 * Update qgroup rfer/excl counters.
1580 * Rfer update is easy, codes can explain themselves.
Qu Wenruoe69bcee2015-04-17 10:23:16 +08001581 *
Qu Wenruo823ae5b2015-04-12 16:59:57 +08001582 * Excl update is tricky, the update is split into 2 part.
1583 * Part 1: Possible exclusive <-> sharing detect:
1584 * | A | !A |
1585 * -------------------------------------
1586 * B | * | - |
1587 * -------------------------------------
1588 * !B | + | ** |
1589 * -------------------------------------
1590 *
1591 * Conditions:
1592 * A: cur_old_roots < nr_old_roots (not exclusive before)
1593 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1594 * B: cur_new_roots < nr_new_roots (not exclusive now)
Nicholas D Steeves01327612016-05-19 21:18:45 -04001595 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
Qu Wenruo823ae5b2015-04-12 16:59:57 +08001596 *
1597 * Results:
1598 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1599 * *: Definitely not changed. **: Possible unchanged.
1600 *
1601 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1602 *
1603 * To make the logic clear, we first use condition A and B to split
1604 * combination into 4 results.
1605 *
1606 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1607 * only on variant maybe 0.
1608 *
1609 * Lastly, check result **, since there are 2 variants maybe 0, split them
1610 * again(2x2).
1611 * But this time we don't need to consider other things, the codes and logic
1612 * is easy to understand now.
1613 */
1614static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1615 struct ulist *qgroups,
1616 u64 nr_old_roots,
1617 u64 nr_new_roots,
1618 u64 num_bytes, u64 seq)
1619{
1620 struct ulist_node *unode;
1621 struct ulist_iterator uiter;
1622 struct btrfs_qgroup *qg;
1623 u64 cur_new_count, cur_old_count;
1624
1625 ULIST_ITER_INIT(&uiter);
1626 while ((unode = ulist_next(qgroups, &uiter))) {
1627 bool dirty = false;
1628
David Sterbaef2fff62016-10-26 16:23:50 +02001629 qg = unode_aux_to_qgroup(unode);
Qu Wenruo823ae5b2015-04-12 16:59:57 +08001630 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1631 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1632
Jeff Mahoneybc074522016-06-09 17:27:55 -04001633 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1634 cur_old_count, cur_new_count);
Mark Fasheh0f5dcf82016-03-29 17:19:55 -07001635
Qu Wenruo823ae5b2015-04-12 16:59:57 +08001636 /* Rfer update part */
1637 if (cur_old_count == 0 && cur_new_count > 0) {
1638 qg->rfer += num_bytes;
1639 qg->rfer_cmpr += num_bytes;
1640 dirty = true;
1641 }
1642 if (cur_old_count > 0 && cur_new_count == 0) {
1643 qg->rfer -= num_bytes;
1644 qg->rfer_cmpr -= num_bytes;
1645 dirty = true;
1646 }
1647
1648 /* Excl update part */
1649 /* Exclusive/none -> shared case */
1650 if (cur_old_count == nr_old_roots &&
1651 cur_new_count < nr_new_roots) {
1652 /* Exclusive -> shared */
1653 if (cur_old_count != 0) {
1654 qg->excl -= num_bytes;
1655 qg->excl_cmpr -= num_bytes;
1656 dirty = true;
1657 }
1658 }
1659
1660 /* Shared -> exclusive/none case */
1661 if (cur_old_count < nr_old_roots &&
1662 cur_new_count == nr_new_roots) {
1663 /* Shared->exclusive */
1664 if (cur_new_count != 0) {
1665 qg->excl += num_bytes;
1666 qg->excl_cmpr += num_bytes;
1667 dirty = true;
1668 }
1669 }
1670
1671 /* Exclusive/none -> exclusive/none case */
1672 if (cur_old_count == nr_old_roots &&
1673 cur_new_count == nr_new_roots) {
1674 if (cur_old_count == 0) {
1675 /* None -> exclusive/none */
1676
1677 if (cur_new_count != 0) {
1678 /* None -> exclusive */
1679 qg->excl += num_bytes;
1680 qg->excl_cmpr += num_bytes;
1681 dirty = true;
1682 }
1683 /* None -> none, nothing changed */
1684 } else {
1685 /* Exclusive -> exclusive/none */
1686
1687 if (cur_new_count == 0) {
1688 /* Exclusive -> none */
1689 qg->excl -= num_bytes;
1690 qg->excl_cmpr -= num_bytes;
1691 dirty = true;
1692 }
1693 /* Exclusive -> exclusive, nothing changed */
1694 }
1695 }
Qu Wenruoc05f9422015-08-03 14:44:29 +08001696
Qu Wenruo823ae5b2015-04-12 16:59:57 +08001697 if (dirty)
1698 qgroup_dirty(fs_info, qg);
1699 }
1700 return 0;
1701}
1702
Qu Wenruo442244c2015-04-16 17:18:36 +08001703int
Qu Wenruo550d7a22015-04-16 15:37:33 +08001704btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1705 struct btrfs_fs_info *fs_info,
1706 u64 bytenr, u64 num_bytes,
1707 struct ulist *old_roots, struct ulist *new_roots)
1708{
1709 struct ulist *qgroups = NULL;
1710 struct ulist *tmp = NULL;
1711 u64 seq;
1712 u64 nr_new_roots = 0;
1713 u64 nr_old_roots = 0;
1714 int ret = 0;
1715
1716 if (new_roots)
1717 nr_new_roots = new_roots->nnodes;
1718 if (old_roots)
1719 nr_old_roots = old_roots->nnodes;
1720
Josef Bacikafcdd122016-09-02 15:40:02 -04001721 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Qu Wenruo550d7a22015-04-16 15:37:33 +08001722 goto out_free;
1723 BUG_ON(!fs_info->quota_root);
1724
Jeff Mahoneybc074522016-06-09 17:27:55 -04001725 trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1726 nr_old_roots, nr_new_roots);
Mark Fasheh0f5dcf82016-03-29 17:19:55 -07001727
Qu Wenruo550d7a22015-04-16 15:37:33 +08001728 qgroups = ulist_alloc(GFP_NOFS);
1729 if (!qgroups) {
1730 ret = -ENOMEM;
1731 goto out_free;
1732 }
1733 tmp = ulist_alloc(GFP_NOFS);
1734 if (!tmp) {
1735 ret = -ENOMEM;
1736 goto out_free;
1737 }
1738
1739 mutex_lock(&fs_info->qgroup_rescan_lock);
1740 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1741 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1742 mutex_unlock(&fs_info->qgroup_rescan_lock);
1743 ret = 0;
1744 goto out_free;
1745 }
1746 }
1747 mutex_unlock(&fs_info->qgroup_rescan_lock);
1748
1749 spin_lock(&fs_info->qgroup_lock);
1750 seq = fs_info->qgroup_seq;
1751
1752 /* Update old refcnts using old_roots */
1753 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1754 UPDATE_OLD);
1755 if (ret < 0)
1756 goto out;
1757
1758 /* Update new refcnts using new_roots */
1759 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1760 UPDATE_NEW);
1761 if (ret < 0)
1762 goto out;
1763
1764 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1765 num_bytes, seq);
1766
1767 /*
1768 * Bump qgroup_seq to avoid seq overlap
1769 */
1770 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1771out:
1772 spin_unlock(&fs_info->qgroup_lock);
1773out_free:
1774 ulist_free(tmp);
1775 ulist_free(qgroups);
1776 ulist_free(old_roots);
1777 ulist_free(new_roots);
1778 return ret;
1779}
1780
1781int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1782 struct btrfs_fs_info *fs_info)
1783{
1784 struct btrfs_qgroup_extent_record *record;
1785 struct btrfs_delayed_ref_root *delayed_refs;
1786 struct ulist *new_roots = NULL;
1787 struct rb_node *node;
Qu Wenruo9086db82015-04-20 09:53:50 +08001788 u64 qgroup_to_skip;
Qu Wenruo550d7a22015-04-16 15:37:33 +08001789 int ret = 0;
1790
1791 delayed_refs = &trans->transaction->delayed_refs;
Qu Wenruo9086db82015-04-20 09:53:50 +08001792 qgroup_to_skip = delayed_refs->qgroup_to_skip;
Qu Wenruo550d7a22015-04-16 15:37:33 +08001793 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
1794 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1795 node);
1796
Jeff Mahoneybc074522016-06-09 17:27:55 -04001797 trace_btrfs_qgroup_account_extents(fs_info, record);
Mark Fasheh0f5dcf82016-03-29 17:19:55 -07001798
Qu Wenruo550d7a22015-04-16 15:37:33 +08001799 if (!ret) {
1800 /*
1801 * Use (u64)-1 as time_seq to do special search, which
1802 * doesn't lock tree or delayed_refs and search current
1803 * root. It's safe inside commit_transaction().
1804 */
1805 ret = btrfs_find_all_roots(trans, fs_info,
1806 record->bytenr, (u64)-1, &new_roots);
1807 if (ret < 0)
1808 goto cleanup;
Qu Wenruo9086db82015-04-20 09:53:50 +08001809 if (qgroup_to_skip)
1810 ulist_del(new_roots, qgroup_to_skip, 0);
Qu Wenruo550d7a22015-04-16 15:37:33 +08001811 ret = btrfs_qgroup_account_extent(trans, fs_info,
1812 record->bytenr, record->num_bytes,
1813 record->old_roots, new_roots);
1814 record->old_roots = NULL;
1815 new_roots = NULL;
1816 }
1817cleanup:
1818 ulist_free(record->old_roots);
1819 ulist_free(new_roots);
1820 new_roots = NULL;
1821 rb_erase(node, &delayed_refs->dirty_extent_root);
1822 kfree(record);
1823
1824 }
1825 return ret;
1826}
1827
Josef Bacikfcebe452014-05-13 17:30:47 -07001828/*
Arne Jansenbed92ea2012-06-28 18:03:02 +02001829 * called from commit_transaction. Writes all changed qgroups to disk.
1830 */
1831int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1832 struct btrfs_fs_info *fs_info)
1833{
1834 struct btrfs_root *quota_root = fs_info->quota_root;
1835 int ret = 0;
Jan Schmidt3d7b5a22013-04-25 16:04:52 +00001836 int start_rescan_worker = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001837
1838 if (!quota_root)
1839 goto out;
1840
Josef Bacikafcdd122016-09-02 15:40:02 -04001841 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
1842 test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
Jan Schmidt3d7b5a22013-04-25 16:04:52 +00001843 start_rescan_worker = 1;
1844
Josef Bacikafcdd122016-09-02 15:40:02 -04001845 if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
1846 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1847 if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags))
1848 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001849
1850 spin_lock(&fs_info->qgroup_lock);
1851 while (!list_empty(&fs_info->dirty_qgroups)) {
1852 struct btrfs_qgroup *qgroup;
1853 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1854 struct btrfs_qgroup, dirty);
1855 list_del_init(&qgroup->dirty);
1856 spin_unlock(&fs_info->qgroup_lock);
1857 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1858 if (ret)
1859 fs_info->qgroup_flags |=
1860 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Dongsheng Yangd3001ed2014-11-20 21:04:56 -05001861 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1862 if (ret)
1863 fs_info->qgroup_flags |=
1864 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001865 spin_lock(&fs_info->qgroup_lock);
1866 }
Josef Bacikafcdd122016-09-02 15:40:02 -04001867 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Arne Jansenbed92ea2012-06-28 18:03:02 +02001868 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1869 else
1870 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1871 spin_unlock(&fs_info->qgroup_lock);
1872
1873 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1874 if (ret)
1875 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1876
Jan Schmidt3d7b5a22013-04-25 16:04:52 +00001877 if (!ret && start_rescan_worker) {
Jan Schmidtb382a322013-05-28 15:47:24 +00001878 ret = qgroup_rescan_init(fs_info, 0, 1);
1879 if (!ret) {
1880 qgroup_rescan_zero_tracking(fs_info);
Qu Wenruofc97fab2014-02-28 10:46:16 +08001881 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1882 &fs_info->qgroup_rescan_work);
Jan Schmidtb382a322013-05-28 15:47:24 +00001883 }
Jan Schmidt3d7b5a22013-04-25 16:04:52 +00001884 ret = 0;
1885 }
1886
Arne Jansenbed92ea2012-06-28 18:03:02 +02001887out:
1888
1889 return ret;
1890}
1891
1892/*
Nicholas D Steeves01327612016-05-19 21:18:45 -04001893 * Copy the accounting information between qgroups. This is necessary
Mark Fasheh918c2ee2016-03-30 17:57:48 -07001894 * when a snapshot or a subvolume is created. Throwing an error will
1895 * cause a transaction abort so we take extra care here to only error
1896 * when a readonly fs is a reasonable outcome.
Arne Jansenbed92ea2012-06-28 18:03:02 +02001897 */
1898int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1899 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1900 struct btrfs_qgroup_inherit *inherit)
1901{
1902 int ret = 0;
1903 int i;
1904 u64 *i_qgroups;
1905 struct btrfs_root *quota_root = fs_info->quota_root;
1906 struct btrfs_qgroup *srcgroup;
1907 struct btrfs_qgroup *dstgroup;
1908 u32 level_size = 0;
Wang Shilong3f5e2d32013-04-07 10:50:19 +00001909 u64 nums;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001910
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001911 mutex_lock(&fs_info->qgroup_ioctl_lock);
Josef Bacikafcdd122016-09-02 15:40:02 -04001912 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001913 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001914
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001915 if (!quota_root) {
1916 ret = -EINVAL;
1917 goto out;
1918 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001919
Wang Shilong3f5e2d32013-04-07 10:50:19 +00001920 if (inherit) {
1921 i_qgroups = (u64 *)(inherit + 1);
1922 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1923 2 * inherit->num_excl_copies;
1924 for (i = 0; i < nums; ++i) {
1925 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
Dongsheng Yang09870d22014-11-11 07:18:22 -05001926
Mark Fasheh918c2ee2016-03-30 17:57:48 -07001927 /*
1928 * Zero out invalid groups so we can ignore
1929 * them later.
1930 */
1931 if (!srcgroup ||
1932 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
1933 *i_qgroups = 0ULL;
1934
Wang Shilong3f5e2d32013-04-07 10:50:19 +00001935 ++i_qgroups;
1936 }
1937 }
1938
Arne Jansenbed92ea2012-06-28 18:03:02 +02001939 /*
1940 * create a tracking group for the subvol itself
1941 */
1942 ret = add_qgroup_item(trans, quota_root, objectid);
1943 if (ret)
1944 goto out;
1945
Arne Jansenbed92ea2012-06-28 18:03:02 +02001946 if (srcid) {
1947 struct btrfs_root *srcroot;
1948 struct btrfs_key srckey;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001949
1950 srckey.objectid = srcid;
1951 srckey.type = BTRFS_ROOT_ITEM_KEY;
1952 srckey.offset = (u64)-1;
1953 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1954 if (IS_ERR(srcroot)) {
1955 ret = PTR_ERR(srcroot);
1956 goto out;
1957 }
1958
1959 rcu_read_lock();
David Sterba707e8a02014-06-04 19:22:26 +02001960 level_size = srcroot->nodesize;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001961 rcu_read_unlock();
1962 }
1963
1964 /*
1965 * add qgroup to all inherited groups
1966 */
1967 if (inherit) {
1968 i_qgroups = (u64 *)(inherit + 1);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07001969 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
1970 if (*i_qgroups == 0)
1971 continue;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001972 ret = add_qgroup_relation_item(trans, quota_root,
1973 objectid, *i_qgroups);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07001974 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001975 goto out;
1976 ret = add_qgroup_relation_item(trans, quota_root,
1977 *i_qgroups, objectid);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07001978 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001979 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001980 }
Mark Fasheh918c2ee2016-03-30 17:57:48 -07001981 ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001982 }
1983
1984
1985 spin_lock(&fs_info->qgroup_lock);
1986
1987 dstgroup = add_qgroup_rb(fs_info, objectid);
Dan Carpenter57a5a882012-07-30 02:15:43 -06001988 if (IS_ERR(dstgroup)) {
1989 ret = PTR_ERR(dstgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001990 goto unlock;
Dan Carpenter57a5a882012-07-30 02:15:43 -06001991 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001992
Dongsheng Yange8c85412014-11-20 20:58:34 -05001993 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
Dongsheng Yange8c85412014-11-20 20:58:34 -05001994 dstgroup->lim_flags = inherit->lim.flags;
1995 dstgroup->max_rfer = inherit->lim.max_rfer;
1996 dstgroup->max_excl = inherit->lim.max_excl;
1997 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
1998 dstgroup->rsv_excl = inherit->lim.rsv_excl;
Dongsheng Yang1510e712014-11-20 21:01:41 -05001999
2000 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2001 if (ret) {
2002 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Jeff Mahoney5d163e02016-09-20 10:05:00 -04002003 btrfs_info(fs_info,
2004 "unable to update quota limit for %llu",
2005 dstgroup->qgroupid);
Dongsheng Yang1510e712014-11-20 21:01:41 -05002006 goto unlock;
2007 }
Dongsheng Yange8c85412014-11-20 20:58:34 -05002008 }
2009
Arne Jansenbed92ea2012-06-28 18:03:02 +02002010 if (srcid) {
2011 srcgroup = find_qgroup_rb(fs_info, srcid);
Chris Masonf3a87f12012-09-14 20:06:30 -04002012 if (!srcgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002013 goto unlock;
Josef Bacikfcebe452014-05-13 17:30:47 -07002014
2015 /*
2016 * We call inherit after we clone the root in order to make sure
2017 * our counts don't go crazy, so at this point the only
2018 * difference between the two roots should be the root node.
2019 */
2020 dstgroup->rfer = srcgroup->rfer;
2021 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2022 dstgroup->excl = level_size;
2023 dstgroup->excl_cmpr = level_size;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002024 srcgroup->excl = level_size;
2025 srcgroup->excl_cmpr = level_size;
Dongsheng Yang3eeb4d52014-11-20 20:14:38 -05002026
2027 /* inherit the limit info */
2028 dstgroup->lim_flags = srcgroup->lim_flags;
2029 dstgroup->max_rfer = srcgroup->max_rfer;
2030 dstgroup->max_excl = srcgroup->max_excl;
2031 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2032 dstgroup->rsv_excl = srcgroup->rsv_excl;
2033
Arne Jansenbed92ea2012-06-28 18:03:02 +02002034 qgroup_dirty(fs_info, dstgroup);
2035 qgroup_dirty(fs_info, srcgroup);
2036 }
2037
Chris Masonf3a87f12012-09-14 20:06:30 -04002038 if (!inherit)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002039 goto unlock;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002040
2041 i_qgroups = (u64 *)(inherit + 1);
2042 for (i = 0; i < inherit->num_qgroups; ++i) {
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002043 if (*i_qgroups) {
2044 ret = add_relation_rb(quota_root->fs_info, objectid,
2045 *i_qgroups);
2046 if (ret)
2047 goto unlock;
2048 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002049 ++i_qgroups;
2050 }
2051
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002052 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002053 struct btrfs_qgroup *src;
2054 struct btrfs_qgroup *dst;
2055
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002056 if (!i_qgroups[0] || !i_qgroups[1])
2057 continue;
2058
Arne Jansenbed92ea2012-06-28 18:03:02 +02002059 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2060 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2061
2062 if (!src || !dst) {
2063 ret = -EINVAL;
2064 goto unlock;
2065 }
2066
2067 dst->rfer = src->rfer - level_size;
2068 dst->rfer_cmpr = src->rfer_cmpr - level_size;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002069 }
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002070 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002071 struct btrfs_qgroup *src;
2072 struct btrfs_qgroup *dst;
2073
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002074 if (!i_qgroups[0] || !i_qgroups[1])
2075 continue;
2076
Arne Jansenbed92ea2012-06-28 18:03:02 +02002077 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2078 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2079
2080 if (!src || !dst) {
2081 ret = -EINVAL;
2082 goto unlock;
2083 }
2084
2085 dst->excl = src->excl + level_size;
2086 dst->excl_cmpr = src->excl_cmpr + level_size;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002087 }
2088
2089unlock:
2090 spin_unlock(&fs_info->qgroup_lock);
2091out:
Wang Shilongf2f6ed32013-04-07 10:50:16 +00002092 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002093 return ret;
2094}
2095
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002096static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002097{
2098 struct btrfs_root *quota_root;
2099 struct btrfs_qgroup *qgroup;
2100 struct btrfs_fs_info *fs_info = root->fs_info;
2101 u64 ref_root = root->root_key.objectid;
2102 int ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002103 struct ulist_node *unode;
2104 struct ulist_iterator uiter;
2105
2106 if (!is_fstree(ref_root))
2107 return 0;
2108
2109 if (num_bytes == 0)
2110 return 0;
2111
2112 spin_lock(&fs_info->qgroup_lock);
2113 quota_root = fs_info->quota_root;
2114 if (!quota_root)
2115 goto out;
2116
2117 qgroup = find_qgroup_rb(fs_info, ref_root);
2118 if (!qgroup)
2119 goto out;
2120
2121 /*
2122 * in a first step, we check all affected qgroups if any limits would
2123 * be exceeded
2124 */
Wang Shilong1e8f9152013-05-06 11:03:27 +00002125 ulist_reinit(fs_info->qgroup_ulist);
2126 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
Wang Shilong3c971852013-04-17 14:00:36 +00002127 (uintptr_t)qgroup, GFP_ATOMIC);
2128 if (ret < 0)
2129 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002130 ULIST_ITER_INIT(&uiter);
Wang Shilong1e8f9152013-05-06 11:03:27 +00002131 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002132 struct btrfs_qgroup *qg;
2133 struct btrfs_qgroup_list *glist;
2134
David Sterbaef2fff62016-10-26 16:23:50 +02002135 qg = unode_aux_to_qgroup(unode);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002136
2137 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
Dongsheng Yange2d1f922015-02-06 10:26:52 -05002138 qg->reserved + (s64)qg->rfer + num_bytes >
Wang Shilong720f1e22013-03-06 11:51:47 +00002139 qg->max_rfer) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002140 ret = -EDQUOT;
Wang Shilong720f1e22013-03-06 11:51:47 +00002141 goto out;
2142 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002143
2144 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
Dongsheng Yange2d1f922015-02-06 10:26:52 -05002145 qg->reserved + (s64)qg->excl + num_bytes >
Wang Shilong720f1e22013-03-06 11:51:47 +00002146 qg->max_excl) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002147 ret = -EDQUOT;
Wang Shilong720f1e22013-03-06 11:51:47 +00002148 goto out;
2149 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002150
2151 list_for_each_entry(glist, &qg->groups, next_group) {
Wang Shilong1e8f9152013-05-06 11:03:27 +00002152 ret = ulist_add(fs_info->qgroup_ulist,
2153 glist->group->qgroupid,
Wang Shilong3c971852013-04-17 14:00:36 +00002154 (uintptr_t)glist->group, GFP_ATOMIC);
2155 if (ret < 0)
2156 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002157 }
2158 }
Wang Shilong3c971852013-04-17 14:00:36 +00002159 ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002160 /*
2161 * no limits exceeded, now record the reservation into all qgroups
2162 */
2163 ULIST_ITER_INIT(&uiter);
Wang Shilong1e8f9152013-05-06 11:03:27 +00002164 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002165 struct btrfs_qgroup *qg;
2166
David Sterbaef2fff62016-10-26 16:23:50 +02002167 qg = unode_aux_to_qgroup(unode);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002168
Dongsheng Yange2d1f922015-02-06 10:26:52 -05002169 qg->reserved += num_bytes;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002170 }
2171
2172out:
2173 spin_unlock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002174 return ret;
2175}
2176
Qu Wenruo297d7502015-09-08 17:08:37 +08002177void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2178 u64 ref_root, u64 num_bytes)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002179{
2180 struct btrfs_root *quota_root;
2181 struct btrfs_qgroup *qgroup;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002182 struct ulist_node *unode;
2183 struct ulist_iterator uiter;
Wang Shilong3c971852013-04-17 14:00:36 +00002184 int ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002185
2186 if (!is_fstree(ref_root))
2187 return;
2188
2189 if (num_bytes == 0)
2190 return;
2191
2192 spin_lock(&fs_info->qgroup_lock);
2193
2194 quota_root = fs_info->quota_root;
2195 if (!quota_root)
2196 goto out;
2197
2198 qgroup = find_qgroup_rb(fs_info, ref_root);
2199 if (!qgroup)
2200 goto out;
2201
Wang Shilong1e8f9152013-05-06 11:03:27 +00002202 ulist_reinit(fs_info->qgroup_ulist);
2203 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
Wang Shilong3c971852013-04-17 14:00:36 +00002204 (uintptr_t)qgroup, GFP_ATOMIC);
2205 if (ret < 0)
2206 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002207 ULIST_ITER_INIT(&uiter);
Wang Shilong1e8f9152013-05-06 11:03:27 +00002208 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002209 struct btrfs_qgroup *qg;
2210 struct btrfs_qgroup_list *glist;
2211
David Sterbaef2fff62016-10-26 16:23:50 +02002212 qg = unode_aux_to_qgroup(unode);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002213
Dongsheng Yange2d1f922015-02-06 10:26:52 -05002214 qg->reserved -= num_bytes;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002215
2216 list_for_each_entry(glist, &qg->groups, next_group) {
Wang Shilong1e8f9152013-05-06 11:03:27 +00002217 ret = ulist_add(fs_info->qgroup_ulist,
2218 glist->group->qgroupid,
Wang Shilong3c971852013-04-17 14:00:36 +00002219 (uintptr_t)glist->group, GFP_ATOMIC);
2220 if (ret < 0)
2221 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002222 }
2223 }
2224
2225out:
2226 spin_unlock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002227}
2228
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002229static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2230{
2231 return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2232 num_bytes);
2233}
Arne Jansenbed92ea2012-06-28 18:03:02 +02002234void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2235{
2236 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2237 return;
Jeff Mahoney64b63582016-06-20 17:23:41 -04002238 btrfs_err(trans->fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04002239 "qgroups not uptodate in trans handle %p: list is%s empty, seq is %#x.%x",
Arne Jansenbed92ea2012-06-28 18:03:02 +02002240 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +00002241 (u32)(trans->delayed_ref_elem.seq >> 32),
2242 (u32)trans->delayed_ref_elem.seq);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002243 BUG();
2244}
Jan Schmidt2f232032013-04-25 16:04:51 +00002245
2246/*
2247 * returns < 0 on error, 0 when more leafs are to be scanned.
Qu Wenruo33931682015-02-27 16:24:24 +08002248 * returns 1 when done.
Jan Schmidt2f232032013-04-25 16:04:51 +00002249 */
2250static int
Jan Schmidtb382a322013-05-28 15:47:24 +00002251qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08002252 struct btrfs_trans_handle *trans)
Jan Schmidt2f232032013-04-25 16:04:51 +00002253{
2254 struct btrfs_key found;
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08002255 struct extent_buffer *scratch_leaf = NULL;
Jan Schmidt2f232032013-04-25 16:04:51 +00002256 struct ulist *roots = NULL;
David Sterba3284da72015-02-25 15:47:32 +01002257 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
Josef Bacikfcebe452014-05-13 17:30:47 -07002258 u64 num_bytes;
Jan Schmidt2f232032013-04-25 16:04:51 +00002259 int slot;
2260 int ret;
2261
Jan Schmidt2f232032013-04-25 16:04:51 +00002262 mutex_lock(&fs_info->qgroup_rescan_lock);
2263 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2264 &fs_info->qgroup_rescan_progress,
2265 path, 1, 0);
2266
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04002267 btrfs_debug(fs_info,
2268 "current progress key (%llu %u %llu), search_slot ret %d",
2269 fs_info->qgroup_rescan_progress.objectid,
2270 fs_info->qgroup_rescan_progress.type,
2271 fs_info->qgroup_rescan_progress.offset, ret);
Jan Schmidt2f232032013-04-25 16:04:51 +00002272
2273 if (ret) {
2274 /*
2275 * The rescan is about to end, we will not be scanning any
2276 * further blocks. We cannot unset the RESCAN flag here, because
2277 * we want to commit the transaction if everything went well.
2278 * To make the live accounting work in this phase, we set our
2279 * scan progress pointer such that every real extent objectid
2280 * will be smaller.
2281 */
2282 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2283 btrfs_release_path(path);
2284 mutex_unlock(&fs_info->qgroup_rescan_lock);
2285 return ret;
2286 }
2287
2288 btrfs_item_key_to_cpu(path->nodes[0], &found,
2289 btrfs_header_nritems(path->nodes[0]) - 1);
2290 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2291
2292 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08002293 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2294 if (!scratch_leaf) {
2295 ret = -ENOMEM;
2296 mutex_unlock(&fs_info->qgroup_rescan_lock);
2297 goto out;
2298 }
2299 extent_buffer_get(scratch_leaf);
2300 btrfs_tree_read_lock(scratch_leaf);
2301 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
Jan Schmidt2f232032013-04-25 16:04:51 +00002302 slot = path->slots[0];
2303 btrfs_release_path(path);
2304 mutex_unlock(&fs_info->qgroup_rescan_lock);
2305
2306 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2307 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
Josef Bacik3a6d75e2014-01-23 16:45:10 -05002308 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2309 found.type != BTRFS_METADATA_ITEM_KEY)
Jan Schmidt2f232032013-04-25 16:04:51 +00002310 continue;
Josef Bacik3a6d75e2014-01-23 16:45:10 -05002311 if (found.type == BTRFS_METADATA_ITEM_KEY)
David Sterba707e8a02014-06-04 19:22:26 +02002312 num_bytes = fs_info->extent_root->nodesize;
Josef Bacik3a6d75e2014-01-23 16:45:10 -05002313 else
2314 num_bytes = found.offset;
2315
Josef Bacikfcebe452014-05-13 17:30:47 -07002316 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2317 &roots);
Jan Schmidt2f232032013-04-25 16:04:51 +00002318 if (ret < 0)
2319 goto out;
Qu Wenruo9d220c92015-04-13 11:02:16 +08002320 /* For rescan, just pass old_roots as NULL */
2321 ret = btrfs_qgroup_account_extent(trans, fs_info,
2322 found.objectid, num_bytes, NULL, roots);
2323 if (ret < 0)
Jan Schmidt2f232032013-04-25 16:04:51 +00002324 goto out;
Jan Schmidt2f232032013-04-25 16:04:51 +00002325 }
Jan Schmidt2f232032013-04-25 16:04:51 +00002326out:
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08002327 if (scratch_leaf) {
2328 btrfs_tree_read_unlock_blocking(scratch_leaf);
2329 free_extent_buffer(scratch_leaf);
2330 }
Jan Schmidt2f232032013-04-25 16:04:51 +00002331 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2332
2333 return ret;
2334}
2335
Qu Wenruod458b052014-02-28 10:46:19 +08002336static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
Jan Schmidt2f232032013-04-25 16:04:51 +00002337{
Jan Schmidtb382a322013-05-28 15:47:24 +00002338 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2339 qgroup_rescan_work);
Jan Schmidt2f232032013-04-25 16:04:51 +00002340 struct btrfs_path *path;
2341 struct btrfs_trans_handle *trans = NULL;
Jan Schmidt2f232032013-04-25 16:04:51 +00002342 int err = -ENOMEM;
Qu Wenruo53b7cde2015-02-27 16:24:25 +08002343 int ret = 0;
Jan Schmidt2f232032013-04-25 16:04:51 +00002344
Jeff Mahoneyd2c609b2016-08-15 12:10:33 -04002345 mutex_lock(&fs_info->qgroup_rescan_lock);
2346 fs_info->qgroup_rescan_running = true;
2347 mutex_unlock(&fs_info->qgroup_rescan_lock);
2348
Jan Schmidt2f232032013-04-25 16:04:51 +00002349 path = btrfs_alloc_path();
2350 if (!path)
2351 goto out;
Jan Schmidt2f232032013-04-25 16:04:51 +00002352
2353 err = 0;
Justin Maggard7343dd62015-11-04 15:56:16 -08002354 while (!err && !btrfs_fs_closing(fs_info)) {
Jan Schmidt2f232032013-04-25 16:04:51 +00002355 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2356 if (IS_ERR(trans)) {
2357 err = PTR_ERR(trans);
2358 break;
2359 }
Josef Bacikafcdd122016-09-02 15:40:02 -04002360 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
Jan Schmidt2f232032013-04-25 16:04:51 +00002361 err = -EINTR;
2362 } else {
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08002363 err = qgroup_rescan_leaf(fs_info, path, trans);
Jan Schmidt2f232032013-04-25 16:04:51 +00002364 }
2365 if (err > 0)
2366 btrfs_commit_transaction(trans, fs_info->fs_root);
2367 else
2368 btrfs_end_transaction(trans, fs_info->fs_root);
2369 }
2370
2371out:
Jan Schmidt2f232032013-04-25 16:04:51 +00002372 btrfs_free_path(path);
Jan Schmidt2f232032013-04-25 16:04:51 +00002373
2374 mutex_lock(&fs_info->qgroup_rescan_lock);
Justin Maggard7343dd62015-11-04 15:56:16 -08002375 if (!btrfs_fs_closing(fs_info))
2376 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
Jan Schmidt2f232032013-04-25 16:04:51 +00002377
Qu Wenruo33931682015-02-27 16:24:24 +08002378 if (err > 0 &&
Jan Schmidt2f232032013-04-25 16:04:51 +00002379 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2380 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2381 } else if (err < 0) {
2382 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2383 }
2384 mutex_unlock(&fs_info->qgroup_rescan_lock);
2385
Qu Wenruo53b7cde2015-02-27 16:24:25 +08002386 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04002387 * only update status, since the previous part has already updated the
Qu Wenruo53b7cde2015-02-27 16:24:25 +08002388 * qgroup info.
2389 */
2390 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2391 if (IS_ERR(trans)) {
2392 err = PTR_ERR(trans);
2393 btrfs_err(fs_info,
2394 "fail to start transaction for status update: %d\n",
2395 err);
2396 goto done;
2397 }
2398 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2399 if (ret < 0) {
2400 err = ret;
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04002401 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
Qu Wenruo53b7cde2015-02-27 16:24:25 +08002402 }
2403 btrfs_end_transaction(trans, fs_info->quota_root);
2404
Justin Maggard7343dd62015-11-04 15:56:16 -08002405 if (btrfs_fs_closing(fs_info)) {
2406 btrfs_info(fs_info, "qgroup scan paused");
2407 } else if (err >= 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05002408 btrfs_info(fs_info, "qgroup scan completed%s",
Qu Wenruo33931682015-02-27 16:24:24 +08002409 err > 0 ? " (inconsistency flag cleared)" : "");
Jan Schmidt2f232032013-04-25 16:04:51 +00002410 } else {
Frank Holtonefe120a2013-12-20 11:37:06 -05002411 btrfs_err(fs_info, "qgroup scan failed with %d", err);
Jan Schmidt2f232032013-04-25 16:04:51 +00002412 }
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002413
Qu Wenruo53b7cde2015-02-27 16:24:25 +08002414done:
Jeff Mahoneyd2c609b2016-08-15 12:10:33 -04002415 mutex_lock(&fs_info->qgroup_rescan_lock);
2416 fs_info->qgroup_rescan_running = false;
2417 mutex_unlock(&fs_info->qgroup_rescan_lock);
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002418 complete_all(&fs_info->qgroup_rescan_completion);
Jan Schmidt2f232032013-04-25 16:04:51 +00002419}
2420
Jan Schmidtb382a322013-05-28 15:47:24 +00002421/*
2422 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2423 * memory required for the rescan context.
2424 */
2425static int
2426qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2427 int init_flags)
Jan Schmidt2f232032013-04-25 16:04:51 +00002428{
2429 int ret = 0;
Jan Schmidt2f232032013-04-25 16:04:51 +00002430
Jan Schmidtb382a322013-05-28 15:47:24 +00002431 if (!init_flags &&
2432 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2433 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2434 ret = -EINVAL;
2435 goto err;
2436 }
Jan Schmidt2f232032013-04-25 16:04:51 +00002437
2438 mutex_lock(&fs_info->qgroup_rescan_lock);
2439 spin_lock(&fs_info->qgroup_lock);
Jan Schmidtb382a322013-05-28 15:47:24 +00002440
2441 if (init_flags) {
2442 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2443 ret = -EINPROGRESS;
2444 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2445 ret = -EINVAL;
2446
2447 if (ret) {
2448 spin_unlock(&fs_info->qgroup_lock);
2449 mutex_unlock(&fs_info->qgroup_rescan_lock);
2450 goto err;
2451 }
Jan Schmidtb382a322013-05-28 15:47:24 +00002452 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2453 }
2454
2455 memset(&fs_info->qgroup_rescan_progress, 0,
2456 sizeof(fs_info->qgroup_rescan_progress));
2457 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
Filipe Manana190631f2015-11-05 10:06:23 +00002458 init_completion(&fs_info->qgroup_rescan_completion);
Jan Schmidtb382a322013-05-28 15:47:24 +00002459
2460 spin_unlock(&fs_info->qgroup_lock);
2461 mutex_unlock(&fs_info->qgroup_rescan_lock);
2462
Jan Schmidtb382a322013-05-28 15:47:24 +00002463 memset(&fs_info->qgroup_rescan_work, 0,
2464 sizeof(fs_info->qgroup_rescan_work));
Qu Wenruofc97fab2014-02-28 10:46:16 +08002465 btrfs_init_work(&fs_info->qgroup_rescan_work,
Liu Bo9e0af232014-08-15 23:36:53 +08002466 btrfs_qgroup_rescan_helper,
Qu Wenruofc97fab2014-02-28 10:46:16 +08002467 btrfs_qgroup_rescan_worker, NULL, NULL);
Jan Schmidtb382a322013-05-28 15:47:24 +00002468
Jan Schmidt2f232032013-04-25 16:04:51 +00002469 if (ret) {
Jan Schmidtb382a322013-05-28 15:47:24 +00002470err:
Frank Holtonefe120a2013-12-20 11:37:06 -05002471 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
Jan Schmidt2f232032013-04-25 16:04:51 +00002472 return ret;
2473 }
2474
Jan Schmidtb382a322013-05-28 15:47:24 +00002475 return 0;
2476}
Jan Schmidt2f232032013-04-25 16:04:51 +00002477
Jan Schmidtb382a322013-05-28 15:47:24 +00002478static void
2479qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2480{
2481 struct rb_node *n;
2482 struct btrfs_qgroup *qgroup;
2483
2484 spin_lock(&fs_info->qgroup_lock);
Jan Schmidt2f232032013-04-25 16:04:51 +00002485 /* clear all current qgroup tracking information */
2486 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2487 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2488 qgroup->rfer = 0;
2489 qgroup->rfer_cmpr = 0;
2490 qgroup->excl = 0;
2491 qgroup->excl_cmpr = 0;
2492 }
2493 spin_unlock(&fs_info->qgroup_lock);
Jan Schmidtb382a322013-05-28 15:47:24 +00002494}
Jan Schmidt2f232032013-04-25 16:04:51 +00002495
Jan Schmidtb382a322013-05-28 15:47:24 +00002496int
2497btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2498{
2499 int ret = 0;
2500 struct btrfs_trans_handle *trans;
2501
2502 ret = qgroup_rescan_init(fs_info, 0, 1);
2503 if (ret)
2504 return ret;
2505
2506 /*
2507 * We have set the rescan_progress to 0, which means no more
2508 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2509 * However, btrfs_qgroup_account_ref may be right after its call
2510 * to btrfs_find_all_roots, in which case it would still do the
2511 * accounting.
2512 * To solve this, we're committing the transaction, which will
2513 * ensure we run all delayed refs and only after that, we are
2514 * going to clear all tracking information for a clean start.
2515 */
2516
2517 trans = btrfs_join_transaction(fs_info->fs_root);
2518 if (IS_ERR(trans)) {
2519 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2520 return PTR_ERR(trans);
2521 }
2522 ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2523 if (ret) {
2524 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2525 return ret;
2526 }
2527
2528 qgroup_rescan_zero_tracking(fs_info);
2529
Qu Wenruofc97fab2014-02-28 10:46:16 +08002530 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2531 &fs_info->qgroup_rescan_work);
Jan Schmidt2f232032013-04-25 16:04:51 +00002532
2533 return 0;
2534}
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002535
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04002536int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2537 bool interruptible)
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002538{
2539 int running;
2540 int ret = 0;
2541
2542 mutex_lock(&fs_info->qgroup_rescan_lock);
2543 spin_lock(&fs_info->qgroup_lock);
Jeff Mahoneyd2c609b2016-08-15 12:10:33 -04002544 running = fs_info->qgroup_rescan_running;
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002545 spin_unlock(&fs_info->qgroup_lock);
2546 mutex_unlock(&fs_info->qgroup_rescan_lock);
2547
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04002548 if (!running)
2549 return 0;
2550
2551 if (interruptible)
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002552 ret = wait_for_completion_interruptible(
2553 &fs_info->qgroup_rescan_completion);
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04002554 else
2555 wait_for_completion(&fs_info->qgroup_rescan_completion);
Jan Schmidt57254b6e2013-05-06 19:14:17 +00002556
2557 return ret;
2558}
Jan Schmidtb382a322013-05-28 15:47:24 +00002559
2560/*
2561 * this is only called from open_ctree where we're still single threaded, thus
2562 * locking is omitted here.
2563 */
2564void
2565btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2566{
2567 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
Qu Wenruofc97fab2014-02-28 10:46:16 +08002568 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2569 &fs_info->qgroup_rescan_work);
Jan Schmidtb382a322013-05-28 15:47:24 +00002570}
Qu Wenruo52472552015-10-12 16:05:40 +08002571
2572/*
2573 * Reserve qgroup space for range [start, start + len).
2574 *
2575 * This function will either reserve space from related qgroups or doing
2576 * nothing if the range is already reserved.
2577 *
2578 * Return 0 for successful reserve
2579 * Return <0 for error (including -EQUOT)
2580 *
2581 * NOTE: this function may sleep for memory allocation.
2582 */
2583int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2584{
2585 struct btrfs_root *root = BTRFS_I(inode)->root;
2586 struct extent_changeset changeset;
2587 struct ulist_node *unode;
2588 struct ulist_iterator uiter;
2589 int ret;
2590
Josef Bacikafcdd122016-09-02 15:40:02 -04002591 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2592 !is_fstree(root->objectid) || len == 0)
Qu Wenruo52472552015-10-12 16:05:40 +08002593 return 0;
2594
2595 changeset.bytes_changed = 0;
2596 changeset.range_changed = ulist_alloc(GFP_NOFS);
Qu Wenruo52472552015-10-12 16:05:40 +08002597 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
David Sterba2c53b912016-04-26 23:54:39 +02002598 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
Qu Wenruo81fb6f72015-09-28 16:57:53 +08002599 trace_btrfs_qgroup_reserve_data(inode, start, len,
2600 changeset.bytes_changed,
2601 QGROUP_RESERVE);
Qu Wenruo52472552015-10-12 16:05:40 +08002602 if (ret < 0)
2603 goto cleanup;
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002604 ret = qgroup_reserve(root, changeset.bytes_changed);
Qu Wenruo52472552015-10-12 16:05:40 +08002605 if (ret < 0)
2606 goto cleanup;
2607
2608 ulist_free(changeset.range_changed);
2609 return ret;
2610
2611cleanup:
2612 /* cleanup already reserved ranges */
2613 ULIST_ITER_INIT(&uiter);
2614 while ((unode = ulist_next(changeset.range_changed, &uiter)))
2615 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2616 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2617 GFP_NOFS);
2618 ulist_free(changeset.range_changed);
2619 return ret;
2620}
Qu Wenruof695fdc2015-10-12 16:28:06 +08002621
2622static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2623 int free)
2624{
2625 struct extent_changeset changeset;
Qu Wenruo81fb6f72015-09-28 16:57:53 +08002626 int trace_op = QGROUP_RELEASE;
Qu Wenruof695fdc2015-10-12 16:28:06 +08002627 int ret;
2628
2629 changeset.bytes_changed = 0;
2630 changeset.range_changed = ulist_alloc(GFP_NOFS);
2631 if (!changeset.range_changed)
2632 return -ENOMEM;
2633
2634 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
David Sterbaf734c442016-04-26 23:54:39 +02002635 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
Qu Wenruof695fdc2015-10-12 16:28:06 +08002636 if (ret < 0)
2637 goto out;
2638
Qu Wenruo81fb6f72015-09-28 16:57:53 +08002639 if (free) {
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002640 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
Qu Wenruo81fb6f72015-09-28 16:57:53 +08002641 trace_op = QGROUP_FREE;
2642 }
2643 trace_btrfs_qgroup_release_data(inode, start, len,
2644 changeset.bytes_changed, trace_op);
Qu Wenruof695fdc2015-10-12 16:28:06 +08002645out:
2646 ulist_free(changeset.range_changed);
2647 return ret;
2648}
2649
2650/*
2651 * Free a reserved space range from io_tree and related qgroups
2652 *
2653 * Should be called when a range of pages get invalidated before reaching disk.
2654 * Or for error cleanup case.
2655 *
2656 * For data written to disk, use btrfs_qgroup_release_data().
2657 *
2658 * NOTE: This function may sleep for memory allocation.
2659 */
2660int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2661{
2662 return __btrfs_qgroup_release_data(inode, start, len, 1);
2663}
2664
2665/*
2666 * Release a reserved space range from io_tree only.
2667 *
2668 * Should be called when a range of pages get written to disk and corresponding
2669 * FILE_EXTENT is inserted into corresponding root.
2670 *
2671 * Since new qgroup accounting framework will only update qgroup numbers at
2672 * commit_transaction() time, its reserved space shouldn't be freed from
2673 * related qgroups.
2674 *
2675 * But we should release the range from io_tree, to allow further write to be
2676 * COWed.
2677 *
2678 * NOTE: This function may sleep for memory allocation.
2679 */
2680int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2681{
2682 return __btrfs_qgroup_release_data(inode, start, len, 0);
2683}
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002684
2685int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2686{
2687 int ret;
2688
Josef Bacikafcdd122016-09-02 15:40:02 -04002689 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2690 !is_fstree(root->objectid) || num_bytes == 0)
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002691 return 0;
2692
2693 BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002694 ret = qgroup_reserve(root, num_bytes);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002695 if (ret < 0)
2696 return ret;
2697 atomic_add(num_bytes, &root->qgroup_meta_rsv);
2698 return ret;
2699}
2700
2701void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2702{
2703 int reserved;
2704
Josef Bacikafcdd122016-09-02 15:40:02 -04002705 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2706 !is_fstree(root->objectid))
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002707 return;
2708
2709 reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2710 if (reserved == 0)
2711 return;
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002712 qgroup_free(root, reserved);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002713}
2714
2715void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2716{
Josef Bacikafcdd122016-09-02 15:40:02 -04002717 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2718 !is_fstree(root->objectid))
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002719 return;
2720
2721 BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2722 WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2723 atomic_sub(num_bytes, &root->qgroup_meta_rsv);
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002724 qgroup_free(root, num_bytes);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08002725}
Qu Wenruo56fa9d02015-10-13 09:53:10 +08002726
2727/*
Nicholas D Steeves01327612016-05-19 21:18:45 -04002728 * Check qgroup reserved space leaking, normally at destroy inode
Qu Wenruo56fa9d02015-10-13 09:53:10 +08002729 * time
2730 */
2731void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2732{
2733 struct extent_changeset changeset;
2734 struct ulist_node *unode;
2735 struct ulist_iterator iter;
2736 int ret;
2737
2738 changeset.bytes_changed = 0;
2739 changeset.range_changed = ulist_alloc(GFP_NOFS);
2740 if (WARN_ON(!changeset.range_changed))
2741 return;
2742
2743 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
David Sterbaf734c442016-04-26 23:54:39 +02002744 EXTENT_QGROUP_RESERVED, &changeset);
Qu Wenruo56fa9d02015-10-13 09:53:10 +08002745
2746 WARN_ON(ret < 0);
2747 if (WARN_ON(changeset.bytes_changed)) {
2748 ULIST_ITER_INIT(&iter);
2749 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2750 btrfs_warn(BTRFS_I(inode)->root->fs_info,
2751 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2752 inode->i_ino, unode->val, unode->aux);
2753 }
2754 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2755 }
2756 ulist_free(changeset.range_changed);
2757}