Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2011 STRATO. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/pagemap.h>
8#include <linux/writeback.h>
9#include <linux/blkdev.h>
10#include <linux/rbtree.h>
11#include <linux/slab.h>
12#include <linux/workqueue.h>
13#include <linux/btrfs.h>
14#include <linux/sched/mm.h>
15
16#include "ctree.h"
17#include "transaction.h"
18#include "disk-io.h"
19#include "locking.h"
20#include "ulist.h"
21#include "backref.h"
22#include "extent_io.h"
23#include "qgroup.h"
24#include "block-group.h"
25#include "sysfs.h"
26#include "tree-mod-log.h"
27#include "fs.h"
28#include "accessors.h"
29#include "extent-tree.h"
30#include "root-tree.h"
31#include "tree-checker.h"
32
33/*
34 * Helpers to access qgroup reservation
35 *
36 * Callers should ensure the lock context and type are valid
37 */
38
39static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
40{
41 u64 ret = 0;
42 int i;
43
44 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
45 ret += qgroup->rsv.values[i];
46
47 return ret;
48}
49
50#ifdef CONFIG_BTRFS_DEBUG
51static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
52{
53 if (type == BTRFS_QGROUP_RSV_DATA)
54 return "data";
55 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
56 return "meta_pertrans";
57 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
58 return "meta_prealloc";
59 return NULL;
60}
61#endif
62
63static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
64 struct btrfs_qgroup *qgroup, u64 num_bytes,
65 enum btrfs_qgroup_rsv_type type)
66{
67 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
68 qgroup->rsv.values[type] += num_bytes;
69}
70
71static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
72 struct btrfs_qgroup *qgroup, u64 num_bytes,
73 enum btrfs_qgroup_rsv_type type)
74{
75 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
76 if (qgroup->rsv.values[type] >= num_bytes) {
77 qgroup->rsv.values[type] -= num_bytes;
78 return;
79 }
80#ifdef CONFIG_BTRFS_DEBUG
81 WARN_RATELIMIT(1,
82 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
83 qgroup->qgroupid, qgroup_rsv_type_str(type),
84 qgroup->rsv.values[type], num_bytes);
85#endif
86 qgroup->rsv.values[type] = 0;
87}
88
89static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
90 struct btrfs_qgroup *dest,
91 struct btrfs_qgroup *src)
92{
93 int i;
94
95 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
96 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
97}
98
99static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
100 struct btrfs_qgroup *dest,
101 struct btrfs_qgroup *src)
102{
103 int i;
104
105 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
106 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
107}
108
109static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
110 int mod)
111{
112 if (qg->old_refcnt < seq)
113 qg->old_refcnt = seq;
114 qg->old_refcnt += mod;
115}
116
117static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
118 int mod)
119{
120 if (qg->new_refcnt < seq)
121 qg->new_refcnt = seq;
122 qg->new_refcnt += mod;
123}
124
125static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
126{
127 if (qg->old_refcnt < seq)
128 return 0;
129 return qg->old_refcnt - seq;
130}
131
132static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
133{
134 if (qg->new_refcnt < seq)
135 return 0;
136 return qg->new_refcnt - seq;
137}
138
139/*
140 * glue structure to represent the relations between qgroups.
141 */
142struct btrfs_qgroup_list {
143 struct list_head next_group;
144 struct list_head next_member;
145 struct btrfs_qgroup *group;
146 struct btrfs_qgroup *member;
147};
148
149static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
150{
151 return (u64)(uintptr_t)qg;
152}
153
154static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
155{
156 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
157}
158
159static int
160qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
161 int init_flags);
162static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
163
164/* must be called with qgroup_ioctl_lock held */
165static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
166 u64 qgroupid)
167{
168 struct rb_node *n = fs_info->qgroup_tree.rb_node;
169 struct btrfs_qgroup *qgroup;
170
171 while (n) {
172 qgroup = rb_entry(n, struct btrfs_qgroup, node);
173 if (qgroup->qgroupid < qgroupid)
174 n = n->rb_left;
175 else if (qgroup->qgroupid > qgroupid)
176 n = n->rb_right;
177 else
178 return qgroup;
179 }
180 return NULL;
181}
182
183/* must be called with qgroup_lock held */
184static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
185 u64 qgroupid)
186{
187 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
188 struct rb_node *parent = NULL;
189 struct btrfs_qgroup *qgroup;
190
191 while (*p) {
192 parent = *p;
193 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
194
195 if (qgroup->qgroupid < qgroupid)
196 p = &(*p)->rb_left;
197 else if (qgroup->qgroupid > qgroupid)
198 p = &(*p)->rb_right;
199 else
200 return qgroup;
201 }
202
203 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
204 if (!qgroup)
205 return ERR_PTR(-ENOMEM);
206
207 qgroup->qgroupid = qgroupid;
208 INIT_LIST_HEAD(&qgroup->groups);
209 INIT_LIST_HEAD(&qgroup->members);
210 INIT_LIST_HEAD(&qgroup->dirty);
211
212 rb_link_node(&qgroup->node, parent, p);
213 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
214
215 return qgroup;
216}
217
218static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
219 struct btrfs_qgroup *qgroup)
220{
221 struct btrfs_qgroup_list *list;
222
223 list_del(&qgroup->dirty);
224 while (!list_empty(&qgroup->groups)) {
225 list = list_first_entry(&qgroup->groups,
226 struct btrfs_qgroup_list, next_group);
227 list_del(&list->next_group);
228 list_del(&list->next_member);
229 kfree(list);
230 }
231
232 while (!list_empty(&qgroup->members)) {
233 list = list_first_entry(&qgroup->members,
234 struct btrfs_qgroup_list, next_member);
235 list_del(&list->next_group);
236 list_del(&list->next_member);
237 kfree(list);
238 }
239}
240
241/* must be called with qgroup_lock held */
242static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
243{
244 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
245
246 if (!qgroup)
247 return -ENOENT;
248
249 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
250 __del_qgroup_rb(fs_info, qgroup);
251 return 0;
252}
253
254/*
255 * Add relation specified by two qgroups.
256 *
257 * Must be called with qgroup_lock held.
258 *
259 * Return: 0 on success
260 * -ENOENT if one of the qgroups is NULL
261 * <0 other errors
262 */
263static int __add_relation_rb(struct btrfs_qgroup *member, struct btrfs_qgroup *parent)
264{
265 struct btrfs_qgroup_list *list;
266
267 if (!member || !parent)
268 return -ENOENT;
269
270 list = kzalloc(sizeof(*list), GFP_ATOMIC);
271 if (!list)
272 return -ENOMEM;
273
274 list->group = parent;
275 list->member = member;
276 list_add_tail(&list->next_group, &member->groups);
277 list_add_tail(&list->next_member, &parent->members);
278
279 return 0;
280}
281
282/*
283 * Add relation specified by two qgroup ids.
284 *
285 * Must be called with qgroup_lock held.
286 *
287 * Return: 0 on success
288 * -ENOENT if one of the ids does not exist
289 * <0 other errors
290 */
291static int add_relation_rb(struct btrfs_fs_info *fs_info, u64 memberid, u64 parentid)
292{
293 struct btrfs_qgroup *member;
294 struct btrfs_qgroup *parent;
295
296 member = find_qgroup_rb(fs_info, memberid);
297 parent = find_qgroup_rb(fs_info, parentid);
298
299 return __add_relation_rb(member, parent);
300}
301
302/* Must be called with qgroup_lock held */
303static int del_relation_rb(struct btrfs_fs_info *fs_info,
304 u64 memberid, u64 parentid)
305{
306 struct btrfs_qgroup *member;
307 struct btrfs_qgroup *parent;
308 struct btrfs_qgroup_list *list;
309
310 member = find_qgroup_rb(fs_info, memberid);
311 parent = find_qgroup_rb(fs_info, parentid);
312 if (!member || !parent)
313 return -ENOENT;
314
315 list_for_each_entry(list, &member->groups, next_group) {
316 if (list->group == parent) {
317 list_del(&list->next_group);
318 list_del(&list->next_member);
319 kfree(list);
320 return 0;
321 }
322 }
323 return -ENOENT;
324}
325
326#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
327int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
328 u64 rfer, u64 excl)
329{
330 struct btrfs_qgroup *qgroup;
331
332 qgroup = find_qgroup_rb(fs_info, qgroupid);
333 if (!qgroup)
334 return -EINVAL;
335 if (qgroup->rfer != rfer || qgroup->excl != excl)
336 return -EINVAL;
337 return 0;
338}
339#endif
340
341static void qgroup_mark_inconsistent(struct btrfs_fs_info *fs_info)
342{
343 fs_info->qgroup_flags |= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
344 BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
345 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
346}
347
348/*
349 * The full config is read in one go, only called from open_ctree()
350 * It doesn't use any locking, as at this point we're still single-threaded
351 */
352int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
353{
354 struct btrfs_key key;
355 struct btrfs_key found_key;
356 struct btrfs_root *quota_root = fs_info->quota_root;
357 struct btrfs_path *path = NULL;
358 struct extent_buffer *l;
359 int slot;
360 int ret = 0;
361 u64 flags = 0;
362 u64 rescan_progress = 0;
363
364 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
365 return 0;
366
367 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
368 if (!fs_info->qgroup_ulist) {
369 ret = -ENOMEM;
370 goto out;
371 }
372
373 path = btrfs_alloc_path();
374 if (!path) {
375 ret = -ENOMEM;
376 goto out;
377 }
378
379 ret = btrfs_sysfs_add_qgroups(fs_info);
380 if (ret < 0)
381 goto out;
382 /* default this to quota off, in case no status key is found */
383 fs_info->qgroup_flags = 0;
384
385 /*
386 * pass 1: read status, all qgroup infos and limits
387 */
388 key.objectid = 0;
389 key.type = 0;
390 key.offset = 0;
391 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
392 if (ret)
393 goto out;
394
395 while (1) {
396 struct btrfs_qgroup *qgroup;
397
398 slot = path->slots[0];
399 l = path->nodes[0];
400 btrfs_item_key_to_cpu(l, &found_key, slot);
401
402 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
403 struct btrfs_qgroup_status_item *ptr;
404
405 ptr = btrfs_item_ptr(l, slot,
406 struct btrfs_qgroup_status_item);
407
408 if (btrfs_qgroup_status_version(l, ptr) !=
409 BTRFS_QGROUP_STATUS_VERSION) {
410 btrfs_err(fs_info,
411 "old qgroup version, quota disabled");
412 goto out;
413 }
414 if (btrfs_qgroup_status_generation(l, ptr) !=
415 fs_info->generation) {
416 qgroup_mark_inconsistent(fs_info);
417 btrfs_err(fs_info,
418 "qgroup generation mismatch, marked as inconsistent");
419 }
420 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
421 ptr);
422 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
423 goto next1;
424 }
425
426 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
427 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
428 goto next1;
429
430 qgroup = find_qgroup_rb(fs_info, found_key.offset);
431 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
432 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
433 btrfs_err(fs_info, "inconsistent qgroup config");
434 qgroup_mark_inconsistent(fs_info);
435 }
436 if (!qgroup) {
437 qgroup = add_qgroup_rb(fs_info, found_key.offset);
438 if (IS_ERR(qgroup)) {
439 ret = PTR_ERR(qgroup);
440 goto out;
441 }
442 }
443 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
444 if (ret < 0)
445 goto out;
446
447 switch (found_key.type) {
448 case BTRFS_QGROUP_INFO_KEY: {
449 struct btrfs_qgroup_info_item *ptr;
450
451 ptr = btrfs_item_ptr(l, slot,
452 struct btrfs_qgroup_info_item);
453 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
454 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
455 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
456 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
457 /* generation currently unused */
458 break;
459 }
460 case BTRFS_QGROUP_LIMIT_KEY: {
461 struct btrfs_qgroup_limit_item *ptr;
462
463 ptr = btrfs_item_ptr(l, slot,
464 struct btrfs_qgroup_limit_item);
465 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
466 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
467 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
468 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
469 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
470 break;
471 }
472 }
473next1:
474 ret = btrfs_next_item(quota_root, path);
475 if (ret < 0)
476 goto out;
477 if (ret)
478 break;
479 }
480 btrfs_release_path(path);
481
482 /*
483 * pass 2: read all qgroup relations
484 */
485 key.objectid = 0;
486 key.type = BTRFS_QGROUP_RELATION_KEY;
487 key.offset = 0;
488 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
489 if (ret)
490 goto out;
491 while (1) {
492 slot = path->slots[0];
493 l = path->nodes[0];
494 btrfs_item_key_to_cpu(l, &found_key, slot);
495
496 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
497 goto next2;
498
499 if (found_key.objectid > found_key.offset) {
500 /* parent <- member, not needed to build config */
501 /* FIXME should we omit the key completely? */
502 goto next2;
503 }
504
505 ret = add_relation_rb(fs_info, found_key.objectid,
506 found_key.offset);
507 if (ret == -ENOENT) {
508 btrfs_warn(fs_info,
509 "orphan qgroup relation 0x%llx->0x%llx",
510 found_key.objectid, found_key.offset);
511 ret = 0; /* ignore the error */
512 }
513 if (ret)
514 goto out;
515next2:
516 ret = btrfs_next_item(quota_root, path);
517 if (ret < 0)
518 goto out;
519 if (ret)
520 break;
521 }
522out:
523 btrfs_free_path(path);
524 fs_info->qgroup_flags |= flags;
525 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
526 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
527 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
528 ret >= 0)
529 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
530
531 if (ret < 0) {
532 ulist_free(fs_info->qgroup_ulist);
533 fs_info->qgroup_ulist = NULL;
534 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
535 btrfs_sysfs_del_qgroups(fs_info);
536 }
537
538 return ret < 0 ? ret : 0;
539}
540
541/*
542 * Called in close_ctree() when quota is still enabled. This verifies we don't
543 * leak some reserved space.
544 *
545 * Return false if no reserved space is left.
546 * Return true if some reserved space is leaked.
547 */
548bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
549{
550 struct rb_node *node;
551 bool ret = false;
552
553 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
554 return ret;
555 /*
556 * Since we're unmounting, there is no race and no need to grab qgroup
557 * lock. And here we don't go post-order to provide a more user
558 * friendly sorted result.
559 */
560 for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
561 struct btrfs_qgroup *qgroup;
562 int i;
563
564 qgroup = rb_entry(node, struct btrfs_qgroup, node);
565 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
566 if (qgroup->rsv.values[i]) {
567 ret = true;
568 btrfs_warn(fs_info,
569 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
570 btrfs_qgroup_level(qgroup->qgroupid),
571 btrfs_qgroup_subvolid(qgroup->qgroupid),
572 i, qgroup->rsv.values[i]);
573 }
574 }
575 }
576 return ret;
577}
578
579/*
580 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
581 * first two are in single-threaded paths.And for the third one, we have set
582 * quota_root to be null with qgroup_lock held before, so it is safe to clean
583 * up the in-memory structures without qgroup_lock held.
584 */
585void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
586{
587 struct rb_node *n;
588 struct btrfs_qgroup *qgroup;
589
590 while ((n = rb_first(&fs_info->qgroup_tree))) {
591 qgroup = rb_entry(n, struct btrfs_qgroup, node);
592 rb_erase(n, &fs_info->qgroup_tree);
593 __del_qgroup_rb(fs_info, qgroup);
594 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
595 kfree(qgroup);
596 }
597 /*
598 * We call btrfs_free_qgroup_config() when unmounting
599 * filesystem and disabling quota, so we set qgroup_ulist
600 * to be null here to avoid double free.
601 */
602 ulist_free(fs_info->qgroup_ulist);
603 fs_info->qgroup_ulist = NULL;
604 btrfs_sysfs_del_qgroups(fs_info);
605}
606
607static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
608 u64 dst)
609{
610 int ret;
611 struct btrfs_root *quota_root = trans->fs_info->quota_root;
612 struct btrfs_path *path;
613 struct btrfs_key key;
614
615 path = btrfs_alloc_path();
616 if (!path)
617 return -ENOMEM;
618
619 key.objectid = src;
620 key.type = BTRFS_QGROUP_RELATION_KEY;
621 key.offset = dst;
622
623 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
624
625 btrfs_mark_buffer_dirty(path->nodes[0]);
626
627 btrfs_free_path(path);
628 return ret;
629}
630
631static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
632 u64 dst)
633{
634 int ret;
635 struct btrfs_root *quota_root = trans->fs_info->quota_root;
636 struct btrfs_path *path;
637 struct btrfs_key key;
638
639 path = btrfs_alloc_path();
640 if (!path)
641 return -ENOMEM;
642
643 key.objectid = src;
644 key.type = BTRFS_QGROUP_RELATION_KEY;
645 key.offset = dst;
646
647 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
648 if (ret < 0)
649 goto out;
650
651 if (ret > 0) {
652 ret = -ENOENT;
653 goto out;
654 }
655
656 ret = btrfs_del_item(trans, quota_root, path);
657out:
658 btrfs_free_path(path);
659 return ret;
660}
661
662static int add_qgroup_item(struct btrfs_trans_handle *trans,
663 struct btrfs_root *quota_root, u64 qgroupid)
664{
665 int ret;
666 struct btrfs_path *path;
667 struct btrfs_qgroup_info_item *qgroup_info;
668 struct btrfs_qgroup_limit_item *qgroup_limit;
669 struct extent_buffer *leaf;
670 struct btrfs_key key;
671
672 if (btrfs_is_testing(quota_root->fs_info))
673 return 0;
674
675 path = btrfs_alloc_path();
676 if (!path)
677 return -ENOMEM;
678
679 key.objectid = 0;
680 key.type = BTRFS_QGROUP_INFO_KEY;
681 key.offset = qgroupid;
682
683 /*
684 * Avoid a transaction abort by catching -EEXIST here. In that
685 * case, we proceed by re-initializing the existing structure
686 * on disk.
687 */
688
689 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
690 sizeof(*qgroup_info));
691 if (ret && ret != -EEXIST)
692 goto out;
693
694 leaf = path->nodes[0];
695 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
696 struct btrfs_qgroup_info_item);
697 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
698 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
699 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
700 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
701 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
702
703 btrfs_mark_buffer_dirty(leaf);
704
705 btrfs_release_path(path);
706
707 key.type = BTRFS_QGROUP_LIMIT_KEY;
708 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
709 sizeof(*qgroup_limit));
710 if (ret && ret != -EEXIST)
711 goto out;
712
713 leaf = path->nodes[0];
714 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
715 struct btrfs_qgroup_limit_item);
716 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
717 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
718 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
719 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
720 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
721
722 btrfs_mark_buffer_dirty(leaf);
723
724 ret = 0;
725out:
726 btrfs_free_path(path);
727 return ret;
728}
729
730static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
731{
732 int ret;
733 struct btrfs_root *quota_root = trans->fs_info->quota_root;
734 struct btrfs_path *path;
735 struct btrfs_key key;
736
737 path = btrfs_alloc_path();
738 if (!path)
739 return -ENOMEM;
740
741 key.objectid = 0;
742 key.type = BTRFS_QGROUP_INFO_KEY;
743 key.offset = qgroupid;
744 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
745 if (ret < 0)
746 goto out;
747
748 if (ret > 0) {
749 ret = -ENOENT;
750 goto out;
751 }
752
753 ret = btrfs_del_item(trans, quota_root, path);
754 if (ret)
755 goto out;
756
757 btrfs_release_path(path);
758
759 key.type = BTRFS_QGROUP_LIMIT_KEY;
760 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
761 if (ret < 0)
762 goto out;
763
764 if (ret > 0) {
765 ret = -ENOENT;
766 goto out;
767 }
768
769 ret = btrfs_del_item(trans, quota_root, path);
770
771out:
772 btrfs_free_path(path);
773 return ret;
774}
775
776static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
777 struct btrfs_qgroup *qgroup)
778{
779 struct btrfs_root *quota_root = trans->fs_info->quota_root;
780 struct btrfs_path *path;
781 struct btrfs_key key;
782 struct extent_buffer *l;
783 struct btrfs_qgroup_limit_item *qgroup_limit;
784 int ret;
785 int slot;
786
787 key.objectid = 0;
788 key.type = BTRFS_QGROUP_LIMIT_KEY;
789 key.offset = qgroup->qgroupid;
790
791 path = btrfs_alloc_path();
792 if (!path)
793 return -ENOMEM;
794
795 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
796 if (ret > 0)
797 ret = -ENOENT;
798
799 if (ret)
800 goto out;
801
802 l = path->nodes[0];
803 slot = path->slots[0];
804 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
805 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
806 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
807 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
808 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
809 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
810
811 btrfs_mark_buffer_dirty(l);
812
813out:
814 btrfs_free_path(path);
815 return ret;
816}
817
818static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
819 struct btrfs_qgroup *qgroup)
820{
821 struct btrfs_fs_info *fs_info = trans->fs_info;
822 struct btrfs_root *quota_root = fs_info->quota_root;
823 struct btrfs_path *path;
824 struct btrfs_key key;
825 struct extent_buffer *l;
826 struct btrfs_qgroup_info_item *qgroup_info;
827 int ret;
828 int slot;
829
830 if (btrfs_is_testing(fs_info))
831 return 0;
832
833 key.objectid = 0;
834 key.type = BTRFS_QGROUP_INFO_KEY;
835 key.offset = qgroup->qgroupid;
836
837 path = btrfs_alloc_path();
838 if (!path)
839 return -ENOMEM;
840
841 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
842 if (ret > 0)
843 ret = -ENOENT;
844
845 if (ret)
846 goto out;
847
848 l = path->nodes[0];
849 slot = path->slots[0];
850 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
851 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
852 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
853 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
854 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
855 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
856
857 btrfs_mark_buffer_dirty(l);
858
859out:
860 btrfs_free_path(path);
861 return ret;
862}
863
864static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
865{
866 struct btrfs_fs_info *fs_info = trans->fs_info;
867 struct btrfs_root *quota_root = fs_info->quota_root;
868 struct btrfs_path *path;
869 struct btrfs_key key;
870 struct extent_buffer *l;
871 struct btrfs_qgroup_status_item *ptr;
872 int ret;
873 int slot;
874
875 key.objectid = 0;
876 key.type = BTRFS_QGROUP_STATUS_KEY;
877 key.offset = 0;
878
879 path = btrfs_alloc_path();
880 if (!path)
881 return -ENOMEM;
882
883 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
884 if (ret > 0)
885 ret = -ENOENT;
886
887 if (ret)
888 goto out;
889
890 l = path->nodes[0];
891 slot = path->slots[0];
892 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
893 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags &
894 BTRFS_QGROUP_STATUS_FLAGS_MASK);
895 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
896 btrfs_set_qgroup_status_rescan(l, ptr,
897 fs_info->qgroup_rescan_progress.objectid);
898
899 btrfs_mark_buffer_dirty(l);
900
901out:
902 btrfs_free_path(path);
903 return ret;
904}
905
906/*
907 * called with qgroup_lock held
908 */
909static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
910 struct btrfs_root *root)
911{
912 struct btrfs_path *path;
913 struct btrfs_key key;
914 struct extent_buffer *leaf = NULL;
915 int ret;
916 int nr = 0;
917
918 path = btrfs_alloc_path();
919 if (!path)
920 return -ENOMEM;
921
922 key.objectid = 0;
923 key.offset = 0;
924 key.type = 0;
925
926 while (1) {
927 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
928 if (ret < 0)
929 goto out;
930 leaf = path->nodes[0];
931 nr = btrfs_header_nritems(leaf);
932 if (!nr)
933 break;
934 /*
935 * delete the leaf one by one
936 * since the whole tree is going
937 * to be deleted.
938 */
939 path->slots[0] = 0;
940 ret = btrfs_del_items(trans, root, path, 0, nr);
941 if (ret)
942 goto out;
943
944 btrfs_release_path(path);
945 }
946 ret = 0;
947out:
948 btrfs_free_path(path);
949 return ret;
950}
951
952int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
953{
954 struct btrfs_root *quota_root;
955 struct btrfs_root *tree_root = fs_info->tree_root;
956 struct btrfs_path *path = NULL;
957 struct btrfs_qgroup_status_item *ptr;
958 struct extent_buffer *leaf;
959 struct btrfs_key key;
960 struct btrfs_key found_key;
961 struct btrfs_qgroup *qgroup = NULL;
962 struct btrfs_trans_handle *trans = NULL;
963 struct ulist *ulist = NULL;
964 int ret = 0;
965 int slot;
966
967 /*
968 * We need to have subvol_sem write locked, to prevent races between
969 * concurrent tasks trying to enable quotas, because we will unlock
970 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
971 * and before setting BTRFS_FS_QUOTA_ENABLED.
972 */
973 lockdep_assert_held_write(&fs_info->subvol_sem);
974
975 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
976 btrfs_err(fs_info,
977 "qgroups are currently unsupported in extent tree v2");
978 return -EINVAL;
979 }
980
981 mutex_lock(&fs_info->qgroup_ioctl_lock);
982 if (fs_info->quota_root)
983 goto out;
984
985 ulist = ulist_alloc(GFP_KERNEL);
986 if (!ulist) {
987 ret = -ENOMEM;
988 goto out;
989 }
990
991 ret = btrfs_sysfs_add_qgroups(fs_info);
992 if (ret < 0)
993 goto out;
994
995 /*
996 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
997 * avoid lock acquisition inversion problems (reported by lockdep) between
998 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
999 * start a transaction.
1000 * After we started the transaction lock qgroup_ioctl_lock again and
1001 * check if someone else created the quota root in the meanwhile. If so,
1002 * just return success and release the transaction handle.
1003 *
1004 * Also we don't need to worry about someone else calling
1005 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
1006 * that function returns 0 (success) when the sysfs entries already exist.
1007 */
1008 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1009
1010 /*
1011 * 1 for quota root item
1012 * 1 for BTRFS_QGROUP_STATUS item
1013 *
1014 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1015 * per subvolume. However those are not currently reserved since it
1016 * would be a lot of overkill.
1017 */
1018 trans = btrfs_start_transaction(tree_root, 2);
1019
1020 mutex_lock(&fs_info->qgroup_ioctl_lock);
1021 if (IS_ERR(trans)) {
1022 ret = PTR_ERR(trans);
1023 trans = NULL;
1024 goto out;
1025 }
1026
1027 if (fs_info->quota_root)
1028 goto out;
1029
1030 fs_info->qgroup_ulist = ulist;
1031 ulist = NULL;
1032
1033 /*
1034 * initially create the quota tree
1035 */
1036 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1037 if (IS_ERR(quota_root)) {
1038 ret = PTR_ERR(quota_root);
1039 btrfs_abort_transaction(trans, ret);
1040 goto out;
1041 }
1042
1043 path = btrfs_alloc_path();
1044 if (!path) {
1045 ret = -ENOMEM;
1046 btrfs_abort_transaction(trans, ret);
1047 goto out_free_root;
1048 }
1049
1050 key.objectid = 0;
1051 key.type = BTRFS_QGROUP_STATUS_KEY;
1052 key.offset = 0;
1053
1054 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1055 sizeof(*ptr));
1056 if (ret) {
1057 btrfs_abort_transaction(trans, ret);
1058 goto out_free_path;
1059 }
1060
1061 leaf = path->nodes[0];
1062 ptr = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_qgroup_status_item);
1064 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1065 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1066 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1067 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1068 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
1069 BTRFS_QGROUP_STATUS_FLAGS_MASK);
1070 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1071
1072 btrfs_mark_buffer_dirty(leaf);
1073
1074 key.objectid = 0;
1075 key.type = BTRFS_ROOT_REF_KEY;
1076 key.offset = 0;
1077
1078 btrfs_release_path(path);
1079 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1080 if (ret > 0)
1081 goto out_add_root;
1082 if (ret < 0) {
1083 btrfs_abort_transaction(trans, ret);
1084 goto out_free_path;
1085 }
1086
1087 while (1) {
1088 slot = path->slots[0];
1089 leaf = path->nodes[0];
1090 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1091
1092 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1093
1094 /* Release locks on tree_root before we access quota_root */
1095 btrfs_release_path(path);
1096
1097 ret = add_qgroup_item(trans, quota_root,
1098 found_key.offset);
1099 if (ret) {
1100 btrfs_abort_transaction(trans, ret);
1101 goto out_free_path;
1102 }
1103
1104 qgroup = add_qgroup_rb(fs_info, found_key.offset);
1105 if (IS_ERR(qgroup)) {
1106 ret = PTR_ERR(qgroup);
1107 btrfs_abort_transaction(trans, ret);
1108 goto out_free_path;
1109 }
1110 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1111 if (ret < 0) {
1112 btrfs_abort_transaction(trans, ret);
1113 goto out_free_path;
1114 }
1115 ret = btrfs_search_slot_for_read(tree_root, &found_key,
1116 path, 1, 0);
1117 if (ret < 0) {
1118 btrfs_abort_transaction(trans, ret);
1119 goto out_free_path;
1120 }
1121 if (ret > 0) {
1122 /*
1123 * Shouldn't happen, but in case it does we
1124 * don't need to do the btrfs_next_item, just
1125 * continue.
1126 */
1127 continue;
1128 }
1129 }
1130 ret = btrfs_next_item(tree_root, path);
1131 if (ret < 0) {
1132 btrfs_abort_transaction(trans, ret);
1133 goto out_free_path;
1134 }
1135 if (ret)
1136 break;
1137 }
1138
1139out_add_root:
1140 btrfs_release_path(path);
1141 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1142 if (ret) {
1143 btrfs_abort_transaction(trans, ret);
1144 goto out_free_path;
1145 }
1146
1147 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1148 if (IS_ERR(qgroup)) {
1149 ret = PTR_ERR(qgroup);
1150 btrfs_abort_transaction(trans, ret);
1151 goto out_free_path;
1152 }
1153 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1154 if (ret < 0) {
1155 btrfs_abort_transaction(trans, ret);
1156 goto out_free_path;
1157 }
1158
1159 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1160 /*
1161 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1162 * a deadlock with tasks concurrently doing other qgroup operations, such
1163 * adding/removing qgroups or adding/deleting qgroup relations for example,
1164 * because all qgroup operations first start or join a transaction and then
1165 * lock the qgroup_ioctl_lock mutex.
1166 * We are safe from a concurrent task trying to enable quotas, by calling
1167 * this function, since we are serialized by fs_info->subvol_sem.
1168 */
1169 ret = btrfs_commit_transaction(trans);
1170 trans = NULL;
1171 mutex_lock(&fs_info->qgroup_ioctl_lock);
1172 if (ret)
1173 goto out_free_path;
1174
1175 /*
1176 * Set quota enabled flag after committing the transaction, to avoid
1177 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1178 * creation.
1179 */
1180 spin_lock(&fs_info->qgroup_lock);
1181 fs_info->quota_root = quota_root;
1182 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1183 spin_unlock(&fs_info->qgroup_lock);
1184
1185 ret = qgroup_rescan_init(fs_info, 0, 1);
1186 if (!ret) {
1187 qgroup_rescan_zero_tracking(fs_info);
1188 fs_info->qgroup_rescan_running = true;
1189 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1190 &fs_info->qgroup_rescan_work);
1191 } else {
1192 /*
1193 * We have set both BTRFS_FS_QUOTA_ENABLED and
1194 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1195 * -EINPROGRESS. That can happen because someone started the
1196 * rescan worker by calling quota rescan ioctl before we
1197 * attempted to initialize the rescan worker. Failure due to
1198 * quotas disabled in the meanwhile is not possible, because
1199 * we are holding a write lock on fs_info->subvol_sem, which
1200 * is also acquired when disabling quotas.
1201 * Ignore such error, and any other error would need to undo
1202 * everything we did in the transaction we just committed.
1203 */
1204 ASSERT(ret == -EINPROGRESS);
1205 ret = 0;
1206 }
1207
1208out_free_path:
1209 btrfs_free_path(path);
1210out_free_root:
1211 if (ret)
1212 btrfs_put_root(quota_root);
1213out:
1214 if (ret) {
1215 ulist_free(fs_info->qgroup_ulist);
1216 fs_info->qgroup_ulist = NULL;
1217 btrfs_sysfs_del_qgroups(fs_info);
1218 }
1219 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1220 if (ret && trans)
1221 btrfs_end_transaction(trans);
1222 else if (trans)
1223 ret = btrfs_end_transaction(trans);
1224 ulist_free(ulist);
1225 return ret;
1226}
1227
1228int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1229{
1230 struct btrfs_root *quota_root;
1231 struct btrfs_trans_handle *trans = NULL;
1232 int ret = 0;
1233
1234 /*
1235 * We need to have subvol_sem write locked, to prevent races between
1236 * concurrent tasks trying to disable quotas, because we will unlock
1237 * and relock qgroup_ioctl_lock across BTRFS_FS_QUOTA_ENABLED changes.
1238 */
1239 lockdep_assert_held_write(&fs_info->subvol_sem);
1240
1241 mutex_lock(&fs_info->qgroup_ioctl_lock);
1242 if (!fs_info->quota_root)
1243 goto out;
1244
1245 /*
1246 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1247 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1248 * to lock that mutex while holding a transaction handle and the rescan
1249 * worker needs to commit a transaction.
1250 */
1251 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1252
1253 /*
1254 * Request qgroup rescan worker to complete and wait for it. This wait
1255 * must be done before transaction start for quota disable since it may
1256 * deadlock with transaction by the qgroup rescan worker.
1257 */
1258 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1259 btrfs_qgroup_wait_for_completion(fs_info, false);
1260
1261 /*
1262 * 1 For the root item
1263 *
1264 * We should also reserve enough items for the quota tree deletion in
1265 * btrfs_clean_quota_tree but this is not done.
1266 *
1267 * Also, we must always start a transaction without holding the mutex
1268 * qgroup_ioctl_lock, see btrfs_quota_enable().
1269 */
1270 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1271
1272 mutex_lock(&fs_info->qgroup_ioctl_lock);
1273 if (IS_ERR(trans)) {
1274 ret = PTR_ERR(trans);
1275 trans = NULL;
1276 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1277 goto out;
1278 }
1279
1280 if (!fs_info->quota_root)
1281 goto out;
1282
1283 spin_lock(&fs_info->qgroup_lock);
1284 quota_root = fs_info->quota_root;
1285 fs_info->quota_root = NULL;
1286 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1287 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1288 spin_unlock(&fs_info->qgroup_lock);
1289
1290 btrfs_free_qgroup_config(fs_info);
1291
1292 ret = btrfs_clean_quota_tree(trans, quota_root);
1293 if (ret) {
1294 btrfs_abort_transaction(trans, ret);
1295 goto out;
1296 }
1297
1298 ret = btrfs_del_root(trans, "a_root->root_key);
1299 if (ret) {
1300 btrfs_abort_transaction(trans, ret);
1301 goto out;
1302 }
1303
1304 list_del("a_root->dirty_list);
1305
1306 btrfs_tree_lock(quota_root->node);
1307 btrfs_clean_tree_block(quota_root->node);
1308 btrfs_tree_unlock(quota_root->node);
1309 btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1310 quota_root->node, 0, 1);
1311
1312 btrfs_put_root(quota_root);
1313
1314out:
1315 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1316 if (ret && trans)
1317 btrfs_end_transaction(trans);
1318 else if (trans)
1319 ret = btrfs_end_transaction(trans);
1320
1321 return ret;
1322}
1323
1324static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1325 struct btrfs_qgroup *qgroup)
1326{
1327 if (list_empty(&qgroup->dirty))
1328 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1329}
1330
1331/*
1332 * The easy accounting, we're updating qgroup relationship whose child qgroup
1333 * only has exclusive extents.
1334 *
1335 * In this case, all exclusive extents will also be exclusive for parent, so
1336 * excl/rfer just get added/removed.
1337 *
1338 * So is qgroup reservation space, which should also be added/removed to
1339 * parent.
1340 * Or when child tries to release reservation space, parent will underflow its
1341 * reservation (for relationship adding case).
1342 *
1343 * Caller should hold fs_info->qgroup_lock.
1344 */
1345static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1346 struct ulist *tmp, u64 ref_root,
1347 struct btrfs_qgroup *src, int sign)
1348{
1349 struct btrfs_qgroup *qgroup;
1350 struct btrfs_qgroup_list *glist;
1351 struct ulist_node *unode;
1352 struct ulist_iterator uiter;
1353 u64 num_bytes = src->excl;
1354 int ret = 0;
1355
1356 qgroup = find_qgroup_rb(fs_info, ref_root);
1357 if (!qgroup)
1358 goto out;
1359
1360 qgroup->rfer += sign * num_bytes;
1361 qgroup->rfer_cmpr += sign * num_bytes;
1362
1363 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1364 qgroup->excl += sign * num_bytes;
1365 qgroup->excl_cmpr += sign * num_bytes;
1366
1367 if (sign > 0)
1368 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1369 else
1370 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1371
1372 qgroup_dirty(fs_info, qgroup);
1373
1374 /* Get all of the parent groups that contain this qgroup */
1375 list_for_each_entry(glist, &qgroup->groups, next_group) {
1376 ret = ulist_add(tmp, glist->group->qgroupid,
1377 qgroup_to_aux(glist->group), GFP_ATOMIC);
1378 if (ret < 0)
1379 goto out;
1380 }
1381
1382 /* Iterate all of the parents and adjust their reference counts */
1383 ULIST_ITER_INIT(&uiter);
1384 while ((unode = ulist_next(tmp, &uiter))) {
1385 qgroup = unode_aux_to_qgroup(unode);
1386 qgroup->rfer += sign * num_bytes;
1387 qgroup->rfer_cmpr += sign * num_bytes;
1388 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1389 qgroup->excl += sign * num_bytes;
1390 if (sign > 0)
1391 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1392 else
1393 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1394 qgroup->excl_cmpr += sign * num_bytes;
1395 qgroup_dirty(fs_info, qgroup);
1396
1397 /* Add any parents of the parents */
1398 list_for_each_entry(glist, &qgroup->groups, next_group) {
1399 ret = ulist_add(tmp, glist->group->qgroupid,
1400 qgroup_to_aux(glist->group), GFP_ATOMIC);
1401 if (ret < 0)
1402 goto out;
1403 }
1404 }
1405 ret = 0;
1406out:
1407 return ret;
1408}
1409
1410
1411/*
1412 * Quick path for updating qgroup with only excl refs.
1413 *
1414 * In that case, just update all parent will be enough.
1415 * Or we needs to do a full rescan.
1416 * Caller should also hold fs_info->qgroup_lock.
1417 *
1418 * Return 0 for quick update, return >0 for need to full rescan
1419 * and mark INCONSISTENT flag.
1420 * Return < 0 for other error.
1421 */
1422static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1423 struct ulist *tmp, u64 src, u64 dst,
1424 int sign)
1425{
1426 struct btrfs_qgroup *qgroup;
1427 int ret = 1;
1428 int err = 0;
1429
1430 qgroup = find_qgroup_rb(fs_info, src);
1431 if (!qgroup)
1432 goto out;
1433 if (qgroup->excl == qgroup->rfer) {
1434 ret = 0;
1435 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1436 qgroup, sign);
1437 if (err < 0) {
1438 ret = err;
1439 goto out;
1440 }
1441 }
1442out:
1443 if (ret)
1444 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1445 return ret;
1446}
1447
1448int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1449 u64 dst)
1450{
1451 struct btrfs_fs_info *fs_info = trans->fs_info;
1452 struct btrfs_qgroup *parent;
1453 struct btrfs_qgroup *member;
1454 struct btrfs_qgroup_list *list;
1455 struct ulist *tmp;
1456 unsigned int nofs_flag;
1457 int ret = 0;
1458
1459 /* Check the level of src and dst first */
1460 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1461 return -EINVAL;
1462
1463 /* We hold a transaction handle open, must do a NOFS allocation. */
1464 nofs_flag = memalloc_nofs_save();
1465 tmp = ulist_alloc(GFP_KERNEL);
1466 memalloc_nofs_restore(nofs_flag);
1467 if (!tmp)
1468 return -ENOMEM;
1469
1470 mutex_lock(&fs_info->qgroup_ioctl_lock);
1471 if (!fs_info->quota_root) {
1472 ret = -ENOTCONN;
1473 goto out;
1474 }
1475 member = find_qgroup_rb(fs_info, src);
1476 parent = find_qgroup_rb(fs_info, dst);
1477 if (!member || !parent) {
1478 ret = -EINVAL;
1479 goto out;
1480 }
1481
1482 /* check if such qgroup relation exist firstly */
1483 list_for_each_entry(list, &member->groups, next_group) {
1484 if (list->group == parent) {
1485 ret = -EEXIST;
1486 goto out;
1487 }
1488 }
1489
1490 ret = add_qgroup_relation_item(trans, src, dst);
1491 if (ret)
1492 goto out;
1493
1494 ret = add_qgroup_relation_item(trans, dst, src);
1495 if (ret) {
1496 del_qgroup_relation_item(trans, src, dst);
1497 goto out;
1498 }
1499
1500 spin_lock(&fs_info->qgroup_lock);
1501 ret = __add_relation_rb(member, parent);
1502 if (ret < 0) {
1503 spin_unlock(&fs_info->qgroup_lock);
1504 goto out;
1505 }
1506 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1507 spin_unlock(&fs_info->qgroup_lock);
1508out:
1509 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1510 ulist_free(tmp);
1511 return ret;
1512}
1513
1514static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1515 u64 dst)
1516{
1517 struct btrfs_fs_info *fs_info = trans->fs_info;
1518 struct btrfs_qgroup *parent;
1519 struct btrfs_qgroup *member;
1520 struct btrfs_qgroup_list *list;
1521 struct ulist *tmp;
1522 bool found = false;
1523 unsigned int nofs_flag;
1524 int ret = 0;
1525 int ret2;
1526
1527 /* We hold a transaction handle open, must do a NOFS allocation. */
1528 nofs_flag = memalloc_nofs_save();
1529 tmp = ulist_alloc(GFP_KERNEL);
1530 memalloc_nofs_restore(nofs_flag);
1531 if (!tmp)
1532 return -ENOMEM;
1533
1534 if (!fs_info->quota_root) {
1535 ret = -ENOTCONN;
1536 goto out;
1537 }
1538
1539 member = find_qgroup_rb(fs_info, src);
1540 parent = find_qgroup_rb(fs_info, dst);
1541 /*
1542 * The parent/member pair doesn't exist, then try to delete the dead
1543 * relation items only.
1544 */
1545 if (!member || !parent)
1546 goto delete_item;
1547
1548 /* check if such qgroup relation exist firstly */
1549 list_for_each_entry(list, &member->groups, next_group) {
1550 if (list->group == parent) {
1551 found = true;
1552 break;
1553 }
1554 }
1555
1556delete_item:
1557 ret = del_qgroup_relation_item(trans, src, dst);
1558 if (ret < 0 && ret != -ENOENT)
1559 goto out;
1560 ret2 = del_qgroup_relation_item(trans, dst, src);
1561 if (ret2 < 0 && ret2 != -ENOENT)
1562 goto out;
1563
1564 /* At least one deletion succeeded, return 0 */
1565 if (!ret || !ret2)
1566 ret = 0;
1567
1568 if (found) {
1569 spin_lock(&fs_info->qgroup_lock);
1570 del_relation_rb(fs_info, src, dst);
1571 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1572 spin_unlock(&fs_info->qgroup_lock);
1573 }
1574out:
1575 ulist_free(tmp);
1576 return ret;
1577}
1578
1579int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1580 u64 dst)
1581{
1582 struct btrfs_fs_info *fs_info = trans->fs_info;
1583 int ret = 0;
1584
1585 mutex_lock(&fs_info->qgroup_ioctl_lock);
1586 ret = __del_qgroup_relation(trans, src, dst);
1587 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1588
1589 return ret;
1590}
1591
1592int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1593{
1594 struct btrfs_fs_info *fs_info = trans->fs_info;
1595 struct btrfs_root *quota_root;
1596 struct btrfs_qgroup *qgroup;
1597 int ret = 0;
1598
1599 mutex_lock(&fs_info->qgroup_ioctl_lock);
1600 if (!fs_info->quota_root) {
1601 ret = -ENOTCONN;
1602 goto out;
1603 }
1604 quota_root = fs_info->quota_root;
1605 qgroup = find_qgroup_rb(fs_info, qgroupid);
1606 if (qgroup) {
1607 ret = -EEXIST;
1608 goto out;
1609 }
1610
1611 ret = add_qgroup_item(trans, quota_root, qgroupid);
1612 if (ret)
1613 goto out;
1614
1615 spin_lock(&fs_info->qgroup_lock);
1616 qgroup = add_qgroup_rb(fs_info, qgroupid);
1617 spin_unlock(&fs_info->qgroup_lock);
1618
1619 if (IS_ERR(qgroup)) {
1620 ret = PTR_ERR(qgroup);
1621 goto out;
1622 }
1623 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1624out:
1625 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1626 return ret;
1627}
1628
1629int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1630{
1631 struct btrfs_fs_info *fs_info = trans->fs_info;
1632 struct btrfs_qgroup *qgroup;
1633 struct btrfs_qgroup_list *list;
1634 int ret = 0;
1635
1636 mutex_lock(&fs_info->qgroup_ioctl_lock);
1637 if (!fs_info->quota_root) {
1638 ret = -ENOTCONN;
1639 goto out;
1640 }
1641
1642 qgroup = find_qgroup_rb(fs_info, qgroupid);
1643 if (!qgroup) {
1644 ret = -ENOENT;
1645 goto out;
1646 }
1647
1648 /* Check if there are no children of this qgroup */
1649 if (!list_empty(&qgroup->members)) {
1650 ret = -EBUSY;
1651 goto out;
1652 }
1653
1654 ret = del_qgroup_item(trans, qgroupid);
1655 if (ret && ret != -ENOENT)
1656 goto out;
1657
1658 while (!list_empty(&qgroup->groups)) {
1659 list = list_first_entry(&qgroup->groups,
1660 struct btrfs_qgroup_list, next_group);
1661 ret = __del_qgroup_relation(trans, qgroupid,
1662 list->group->qgroupid);
1663 if (ret)
1664 goto out;
1665 }
1666
1667 spin_lock(&fs_info->qgroup_lock);
1668 del_qgroup_rb(fs_info, qgroupid);
1669 spin_unlock(&fs_info->qgroup_lock);
1670
1671 /*
1672 * Remove the qgroup from sysfs now without holding the qgroup_lock
1673 * spinlock, since the sysfs_remove_group() function needs to take
1674 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1675 */
1676 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1677 kfree(qgroup);
1678out:
1679 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1680 return ret;
1681}
1682
1683int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1684 struct btrfs_qgroup_limit *limit)
1685{
1686 struct btrfs_fs_info *fs_info = trans->fs_info;
1687 struct btrfs_qgroup *qgroup;
1688 int ret = 0;
1689 /* Sometimes we would want to clear the limit on this qgroup.
1690 * To meet this requirement, we treat the -1 as a special value
1691 * which tell kernel to clear the limit on this qgroup.
1692 */
1693 const u64 CLEAR_VALUE = -1;
1694
1695 mutex_lock(&fs_info->qgroup_ioctl_lock);
1696 if (!fs_info->quota_root) {
1697 ret = -ENOTCONN;
1698 goto out;
1699 }
1700
1701 qgroup = find_qgroup_rb(fs_info, qgroupid);
1702 if (!qgroup) {
1703 ret = -ENOENT;
1704 goto out;
1705 }
1706
1707 spin_lock(&fs_info->qgroup_lock);
1708 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1709 if (limit->max_rfer == CLEAR_VALUE) {
1710 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1711 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1712 qgroup->max_rfer = 0;
1713 } else {
1714 qgroup->max_rfer = limit->max_rfer;
1715 }
1716 }
1717 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1718 if (limit->max_excl == CLEAR_VALUE) {
1719 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1720 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1721 qgroup->max_excl = 0;
1722 } else {
1723 qgroup->max_excl = limit->max_excl;
1724 }
1725 }
1726 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1727 if (limit->rsv_rfer == CLEAR_VALUE) {
1728 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1729 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1730 qgroup->rsv_rfer = 0;
1731 } else {
1732 qgroup->rsv_rfer = limit->rsv_rfer;
1733 }
1734 }
1735 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1736 if (limit->rsv_excl == CLEAR_VALUE) {
1737 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1738 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1739 qgroup->rsv_excl = 0;
1740 } else {
1741 qgroup->rsv_excl = limit->rsv_excl;
1742 }
1743 }
1744 qgroup->lim_flags |= limit->flags;
1745
1746 spin_unlock(&fs_info->qgroup_lock);
1747
1748 ret = update_qgroup_limit_item(trans, qgroup);
1749 if (ret) {
1750 qgroup_mark_inconsistent(fs_info);
1751 btrfs_info(fs_info, "unable to update quota limit for %llu",
1752 qgroupid);
1753 }
1754
1755out:
1756 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1757 return ret;
1758}
1759
1760int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1761 struct btrfs_delayed_ref_root *delayed_refs,
1762 struct btrfs_qgroup_extent_record *record)
1763{
1764 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1765 struct rb_node *parent_node = NULL;
1766 struct btrfs_qgroup_extent_record *entry;
1767 u64 bytenr = record->bytenr;
1768
1769 lockdep_assert_held(&delayed_refs->lock);
1770 trace_btrfs_qgroup_trace_extent(fs_info, record);
1771
1772 while (*p) {
1773 parent_node = *p;
1774 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1775 node);
1776 if (bytenr < entry->bytenr) {
1777 p = &(*p)->rb_left;
1778 } else if (bytenr > entry->bytenr) {
1779 p = &(*p)->rb_right;
1780 } else {
1781 if (record->data_rsv && !entry->data_rsv) {
1782 entry->data_rsv = record->data_rsv;
1783 entry->data_rsv_refroot =
1784 record->data_rsv_refroot;
1785 }
1786 return 1;
1787 }
1788 }
1789
1790 rb_link_node(&record->node, parent_node, p);
1791 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1792 return 0;
1793}
1794
1795int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1796 struct btrfs_qgroup_extent_record *qrecord)
1797{
1798 struct btrfs_backref_walk_ctx ctx = { 0 };
1799 int ret;
1800
1801 /*
1802 * We are always called in a context where we are already holding a
1803 * transaction handle. Often we are called when adding a data delayed
1804 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1805 * in which case we will be holding a write lock on extent buffer from a
1806 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1807 * acquire fs_info->commit_root_sem, because that is a higher level lock
1808 * that must be acquired before locking any extent buffers.
1809 *
1810 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1811 * but we can't pass it a non-NULL transaction handle, because otherwise
1812 * it would not use commit roots and would lock extent buffers, causing
1813 * a deadlock if it ends up trying to read lock the same extent buffer
1814 * that was previously write locked at btrfs_truncate_inode_items().
1815 *
1816 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1817 * explicitly tell it to not acquire the commit_root_sem - if we are
1818 * holding a transaction handle we don't need its protection.
1819 */
1820 ASSERT(trans != NULL);
1821
1822 if (trans->fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
1823 return 0;
1824
1825 ctx.bytenr = qrecord->bytenr;
1826 ctx.fs_info = trans->fs_info;
1827
1828 ret = btrfs_find_all_roots(&ctx, true);
1829 if (ret < 0) {
1830 qgroup_mark_inconsistent(trans->fs_info);
1831 btrfs_warn(trans->fs_info,
1832"error accounting new delayed refs extent (err code: %d), quota inconsistent",
1833 ret);
1834 return 0;
1835 }
1836
1837 /*
1838 * Here we don't need to get the lock of
1839 * trans->transaction->delayed_refs, since inserted qrecord won't
1840 * be deleted, only qrecord->node may be modified (new qrecord insert)
1841 *
1842 * So modifying qrecord->old_roots is safe here
1843 */
1844 qrecord->old_roots = ctx.roots;
1845 return 0;
1846}
1847
1848int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1849 u64 num_bytes)
1850{
1851 struct btrfs_fs_info *fs_info = trans->fs_info;
1852 struct btrfs_qgroup_extent_record *record;
1853 struct btrfs_delayed_ref_root *delayed_refs;
1854 int ret;
1855
1856 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1857 || bytenr == 0 || num_bytes == 0)
1858 return 0;
1859 record = kzalloc(sizeof(*record), GFP_NOFS);
1860 if (!record)
1861 return -ENOMEM;
1862
1863 delayed_refs = &trans->transaction->delayed_refs;
1864 record->bytenr = bytenr;
1865 record->num_bytes = num_bytes;
1866 record->old_roots = NULL;
1867
1868 spin_lock(&delayed_refs->lock);
1869 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1870 spin_unlock(&delayed_refs->lock);
1871 if (ret > 0) {
1872 kfree(record);
1873 return 0;
1874 }
1875 return btrfs_qgroup_trace_extent_post(trans, record);
1876}
1877
1878int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1879 struct extent_buffer *eb)
1880{
1881 struct btrfs_fs_info *fs_info = trans->fs_info;
1882 int nr = btrfs_header_nritems(eb);
1883 int i, extent_type, ret;
1884 struct btrfs_key key;
1885 struct btrfs_file_extent_item *fi;
1886 u64 bytenr, num_bytes;
1887
1888 /* We can be called directly from walk_up_proc() */
1889 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1890 return 0;
1891
1892 for (i = 0; i < nr; i++) {
1893 btrfs_item_key_to_cpu(eb, &key, i);
1894
1895 if (key.type != BTRFS_EXTENT_DATA_KEY)
1896 continue;
1897
1898 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1899 /* filter out non qgroup-accountable extents */
1900 extent_type = btrfs_file_extent_type(eb, fi);
1901
1902 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1903 continue;
1904
1905 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1906 if (!bytenr)
1907 continue;
1908
1909 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1910
1911 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes);
1912 if (ret)
1913 return ret;
1914 }
1915 cond_resched();
1916 return 0;
1917}
1918
1919/*
1920 * Walk up the tree from the bottom, freeing leaves and any interior
1921 * nodes which have had all slots visited. If a node (leaf or
1922 * interior) is freed, the node above it will have it's slot
1923 * incremented. The root node will never be freed.
1924 *
1925 * At the end of this function, we should have a path which has all
1926 * slots incremented to the next position for a search. If we need to
1927 * read a new node it will be NULL and the node above it will have the
1928 * correct slot selected for a later read.
1929 *
1930 * If we increment the root nodes slot counter past the number of
1931 * elements, 1 is returned to signal completion of the search.
1932 */
1933static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1934{
1935 int level = 0;
1936 int nr, slot;
1937 struct extent_buffer *eb;
1938
1939 if (root_level == 0)
1940 return 1;
1941
1942 while (level <= root_level) {
1943 eb = path->nodes[level];
1944 nr = btrfs_header_nritems(eb);
1945 path->slots[level]++;
1946 slot = path->slots[level];
1947 if (slot >= nr || level == 0) {
1948 /*
1949 * Don't free the root - we will detect this
1950 * condition after our loop and return a
1951 * positive value for caller to stop walking the tree.
1952 */
1953 if (level != root_level) {
1954 btrfs_tree_unlock_rw(eb, path->locks[level]);
1955 path->locks[level] = 0;
1956
1957 free_extent_buffer(eb);
1958 path->nodes[level] = NULL;
1959 path->slots[level] = 0;
1960 }
1961 } else {
1962 /*
1963 * We have a valid slot to walk back down
1964 * from. Stop here so caller can process these
1965 * new nodes.
1966 */
1967 break;
1968 }
1969
1970 level++;
1971 }
1972
1973 eb = path->nodes[root_level];
1974 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1975 return 1;
1976
1977 return 0;
1978}
1979
1980/*
1981 * Helper function to trace a subtree tree block swap.
1982 *
1983 * The swap will happen in highest tree block, but there may be a lot of
1984 * tree blocks involved.
1985 *
1986 * For example:
1987 * OO = Old tree blocks
1988 * NN = New tree blocks allocated during balance
1989 *
1990 * File tree (257) Reloc tree for 257
1991 * L2 OO NN
1992 * / \ / \
1993 * L1 OO OO (a) OO NN (a)
1994 * / \ / \ / \ / \
1995 * L0 OO OO OO OO OO OO NN NN
1996 * (b) (c) (b) (c)
1997 *
1998 * When calling qgroup_trace_extent_swap(), we will pass:
1999 * @src_eb = OO(a)
2000 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
2001 * @dst_level = 0
2002 * @root_level = 1
2003 *
2004 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
2005 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
2006 *
2007 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
2008 *
2009 * 1) Tree search from @src_eb
2010 * It should acts as a simplified btrfs_search_slot().
2011 * The key for search can be extracted from @dst_path->nodes[dst_level]
2012 * (first key).
2013 *
2014 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
2015 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
2016 * They should be marked during previous (@dst_level = 1) iteration.
2017 *
2018 * 3) Mark file extents in leaves dirty
2019 * We don't have good way to pick out new file extents only.
2020 * So we still follow the old method by scanning all file extents in
2021 * the leave.
2022 *
2023 * This function can free us from keeping two paths, thus later we only need
2024 * to care about how to iterate all new tree blocks in reloc tree.
2025 */
2026static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
2027 struct extent_buffer *src_eb,
2028 struct btrfs_path *dst_path,
2029 int dst_level, int root_level,
2030 bool trace_leaf)
2031{
2032 struct btrfs_key key;
2033 struct btrfs_path *src_path;
2034 struct btrfs_fs_info *fs_info = trans->fs_info;
2035 u32 nodesize = fs_info->nodesize;
2036 int cur_level = root_level;
2037 int ret;
2038
2039 BUG_ON(dst_level > root_level);
2040 /* Level mismatch */
2041 if (btrfs_header_level(src_eb) != root_level)
2042 return -EINVAL;
2043
2044 src_path = btrfs_alloc_path();
2045 if (!src_path) {
2046 ret = -ENOMEM;
2047 goto out;
2048 }
2049
2050 if (dst_level)
2051 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2052 else
2053 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2054
2055 /* For src_path */
2056 atomic_inc(&src_eb->refs);
2057 src_path->nodes[root_level] = src_eb;
2058 src_path->slots[root_level] = dst_path->slots[root_level];
2059 src_path->locks[root_level] = 0;
2060
2061 /* A simplified version of btrfs_search_slot() */
2062 while (cur_level >= dst_level) {
2063 struct btrfs_key src_key;
2064 struct btrfs_key dst_key;
2065
2066 if (src_path->nodes[cur_level] == NULL) {
2067 struct extent_buffer *eb;
2068 int parent_slot;
2069
2070 eb = src_path->nodes[cur_level + 1];
2071 parent_slot = src_path->slots[cur_level + 1];
2072
2073 eb = btrfs_read_node_slot(eb, parent_slot);
2074 if (IS_ERR(eb)) {
2075 ret = PTR_ERR(eb);
2076 goto out;
2077 }
2078
2079 src_path->nodes[cur_level] = eb;
2080
2081 btrfs_tree_read_lock(eb);
2082 src_path->locks[cur_level] = BTRFS_READ_LOCK;
2083 }
2084
2085 src_path->slots[cur_level] = dst_path->slots[cur_level];
2086 if (cur_level) {
2087 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2088 &dst_key, dst_path->slots[cur_level]);
2089 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2090 &src_key, src_path->slots[cur_level]);
2091 } else {
2092 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2093 &dst_key, dst_path->slots[cur_level]);
2094 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2095 &src_key, src_path->slots[cur_level]);
2096 }
2097 /* Content mismatch, something went wrong */
2098 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2099 ret = -ENOENT;
2100 goto out;
2101 }
2102 cur_level--;
2103 }
2104
2105 /*
2106 * Now both @dst_path and @src_path have been populated, record the tree
2107 * blocks for qgroup accounting.
2108 */
2109 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2110 nodesize);
2111 if (ret < 0)
2112 goto out;
2113 ret = btrfs_qgroup_trace_extent(trans, dst_path->nodes[dst_level]->start,
2114 nodesize);
2115 if (ret < 0)
2116 goto out;
2117
2118 /* Record leaf file extents */
2119 if (dst_level == 0 && trace_leaf) {
2120 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2121 if (ret < 0)
2122 goto out;
2123 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2124 }
2125out:
2126 btrfs_free_path(src_path);
2127 return ret;
2128}
2129
2130/*
2131 * Helper function to do recursive generation-aware depth-first search, to
2132 * locate all new tree blocks in a subtree of reloc tree.
2133 *
2134 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2135 * reloc tree
2136 * L2 NN (a)
2137 * / \
2138 * L1 OO NN (b)
2139 * / \ / \
2140 * L0 OO OO OO NN
2141 * (c) (d)
2142 * If we pass:
2143 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2144 * @cur_level = 1
2145 * @root_level = 1
2146 *
2147 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2148 * above tree blocks along with their counter parts in file tree.
2149 * While during search, old tree blocks OO(c) will be skipped as tree block swap
2150 * won't affect OO(c).
2151 */
2152static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2153 struct extent_buffer *src_eb,
2154 struct btrfs_path *dst_path,
2155 int cur_level, int root_level,
2156 u64 last_snapshot, bool trace_leaf)
2157{
2158 struct btrfs_fs_info *fs_info = trans->fs_info;
2159 struct extent_buffer *eb;
2160 bool need_cleanup = false;
2161 int ret = 0;
2162 int i;
2163
2164 /* Level sanity check */
2165 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2166 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2167 root_level < cur_level) {
2168 btrfs_err_rl(fs_info,
2169 "%s: bad levels, cur_level=%d root_level=%d",
2170 __func__, cur_level, root_level);
2171 return -EUCLEAN;
2172 }
2173
2174 /* Read the tree block if needed */
2175 if (dst_path->nodes[cur_level] == NULL) {
2176 int parent_slot;
2177 u64 child_gen;
2178
2179 /*
2180 * dst_path->nodes[root_level] must be initialized before
2181 * calling this function.
2182 */
2183 if (cur_level == root_level) {
2184 btrfs_err_rl(fs_info,
2185 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2186 __func__, root_level, root_level, cur_level);
2187 return -EUCLEAN;
2188 }
2189
2190 /*
2191 * We need to get child blockptr/gen from parent before we can
2192 * read it.
2193 */
2194 eb = dst_path->nodes[cur_level + 1];
2195 parent_slot = dst_path->slots[cur_level + 1];
2196 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2197
2198 /* This node is old, no need to trace */
2199 if (child_gen < last_snapshot)
2200 goto out;
2201
2202 eb = btrfs_read_node_slot(eb, parent_slot);
2203 if (IS_ERR(eb)) {
2204 ret = PTR_ERR(eb);
2205 goto out;
2206 }
2207
2208 dst_path->nodes[cur_level] = eb;
2209 dst_path->slots[cur_level] = 0;
2210
2211 btrfs_tree_read_lock(eb);
2212 dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2213 need_cleanup = true;
2214 }
2215
2216 /* Now record this tree block and its counter part for qgroups */
2217 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2218 root_level, trace_leaf);
2219 if (ret < 0)
2220 goto cleanup;
2221
2222 eb = dst_path->nodes[cur_level];
2223
2224 if (cur_level > 0) {
2225 /* Iterate all child tree blocks */
2226 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2227 /* Skip old tree blocks as they won't be swapped */
2228 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2229 continue;
2230 dst_path->slots[cur_level] = i;
2231
2232 /* Recursive call (at most 7 times) */
2233 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2234 dst_path, cur_level - 1, root_level,
2235 last_snapshot, trace_leaf);
2236 if (ret < 0)
2237 goto cleanup;
2238 }
2239 }
2240
2241cleanup:
2242 if (need_cleanup) {
2243 /* Clean up */
2244 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2245 dst_path->locks[cur_level]);
2246 free_extent_buffer(dst_path->nodes[cur_level]);
2247 dst_path->nodes[cur_level] = NULL;
2248 dst_path->slots[cur_level] = 0;
2249 dst_path->locks[cur_level] = 0;
2250 }
2251out:
2252 return ret;
2253}
2254
2255static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2256 struct extent_buffer *src_eb,
2257 struct extent_buffer *dst_eb,
2258 u64 last_snapshot, bool trace_leaf)
2259{
2260 struct btrfs_fs_info *fs_info = trans->fs_info;
2261 struct btrfs_path *dst_path = NULL;
2262 int level;
2263 int ret;
2264
2265 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2266 return 0;
2267
2268 /* Wrong parameter order */
2269 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2270 btrfs_err_rl(fs_info,
2271 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2272 btrfs_header_generation(src_eb),
2273 btrfs_header_generation(dst_eb));
2274 return -EUCLEAN;
2275 }
2276
2277 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2278 ret = -EIO;
2279 goto out;
2280 }
2281
2282 level = btrfs_header_level(dst_eb);
2283 dst_path = btrfs_alloc_path();
2284 if (!dst_path) {
2285 ret = -ENOMEM;
2286 goto out;
2287 }
2288 /* For dst_path */
2289 atomic_inc(&dst_eb->refs);
2290 dst_path->nodes[level] = dst_eb;
2291 dst_path->slots[level] = 0;
2292 dst_path->locks[level] = 0;
2293
2294 /* Do the generation aware breadth-first search */
2295 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2296 level, last_snapshot, trace_leaf);
2297 if (ret < 0)
2298 goto out;
2299 ret = 0;
2300
2301out:
2302 btrfs_free_path(dst_path);
2303 if (ret < 0)
2304 qgroup_mark_inconsistent(fs_info);
2305 return ret;
2306}
2307
2308int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2309 struct extent_buffer *root_eb,
2310 u64 root_gen, int root_level)
2311{
2312 struct btrfs_fs_info *fs_info = trans->fs_info;
2313 int ret = 0;
2314 int level;
2315 u8 drop_subptree_thres;
2316 struct extent_buffer *eb = root_eb;
2317 struct btrfs_path *path = NULL;
2318
2319 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2320 BUG_ON(root_eb == NULL);
2321
2322 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2323 return 0;
2324
2325 spin_lock(&fs_info->qgroup_lock);
2326 drop_subptree_thres = fs_info->qgroup_drop_subtree_thres;
2327 spin_unlock(&fs_info->qgroup_lock);
2328
2329 /*
2330 * This function only gets called for snapshot drop, if we hit a high
2331 * node here, it means we are going to change ownership for quite a lot
2332 * of extents, which will greatly slow down btrfs_commit_transaction().
2333 *
2334 * So here if we find a high tree here, we just skip the accounting and
2335 * mark qgroup inconsistent.
2336 */
2337 if (root_level >= drop_subptree_thres) {
2338 qgroup_mark_inconsistent(fs_info);
2339 return 0;
2340 }
2341
2342 if (!extent_buffer_uptodate(root_eb)) {
2343 struct btrfs_tree_parent_check check = {
2344 .has_first_key = false,
2345 .transid = root_gen,
2346 .level = root_level
2347 };
2348
2349 ret = btrfs_read_extent_buffer(root_eb, &check);
2350 if (ret)
2351 goto out;
2352 }
2353
2354 if (root_level == 0) {
2355 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2356 goto out;
2357 }
2358
2359 path = btrfs_alloc_path();
2360 if (!path)
2361 return -ENOMEM;
2362
2363 /*
2364 * Walk down the tree. Missing extent blocks are filled in as
2365 * we go. Metadata is accounted every time we read a new
2366 * extent block.
2367 *
2368 * When we reach a leaf, we account for file extent items in it,
2369 * walk back up the tree (adjusting slot pointers as we go)
2370 * and restart the search process.
2371 */
2372 atomic_inc(&root_eb->refs); /* For path */
2373 path->nodes[root_level] = root_eb;
2374 path->slots[root_level] = 0;
2375 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2376walk_down:
2377 level = root_level;
2378 while (level >= 0) {
2379 if (path->nodes[level] == NULL) {
2380 int parent_slot;
2381 u64 child_bytenr;
2382
2383 /*
2384 * We need to get child blockptr from parent before we
2385 * can read it.
2386 */
2387 eb = path->nodes[level + 1];
2388 parent_slot = path->slots[level + 1];
2389 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2390
2391 eb = btrfs_read_node_slot(eb, parent_slot);
2392 if (IS_ERR(eb)) {
2393 ret = PTR_ERR(eb);
2394 goto out;
2395 }
2396
2397 path->nodes[level] = eb;
2398 path->slots[level] = 0;
2399
2400 btrfs_tree_read_lock(eb);
2401 path->locks[level] = BTRFS_READ_LOCK;
2402
2403 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2404 fs_info->nodesize);
2405 if (ret)
2406 goto out;
2407 }
2408
2409 if (level == 0) {
2410 ret = btrfs_qgroup_trace_leaf_items(trans,
2411 path->nodes[level]);
2412 if (ret)
2413 goto out;
2414
2415 /* Nonzero return here means we completed our search */
2416 ret = adjust_slots_upwards(path, root_level);
2417 if (ret)
2418 break;
2419
2420 /* Restart search with new slots */
2421 goto walk_down;
2422 }
2423
2424 level--;
2425 }
2426
2427 ret = 0;
2428out:
2429 btrfs_free_path(path);
2430
2431 return ret;
2432}
2433
2434#define UPDATE_NEW 0
2435#define UPDATE_OLD 1
2436/*
2437 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2438 */
2439static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2440 struct ulist *roots, struct ulist *tmp,
2441 struct ulist *qgroups, u64 seq, int update_old)
2442{
2443 struct ulist_node *unode;
2444 struct ulist_iterator uiter;
2445 struct ulist_node *tmp_unode;
2446 struct ulist_iterator tmp_uiter;
2447 struct btrfs_qgroup *qg;
2448 int ret = 0;
2449
2450 if (!roots)
2451 return 0;
2452 ULIST_ITER_INIT(&uiter);
2453 while ((unode = ulist_next(roots, &uiter))) {
2454 qg = find_qgroup_rb(fs_info, unode->val);
2455 if (!qg)
2456 continue;
2457
2458 ulist_reinit(tmp);
2459 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2460 GFP_ATOMIC);
2461 if (ret < 0)
2462 return ret;
2463 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2464 if (ret < 0)
2465 return ret;
2466 ULIST_ITER_INIT(&tmp_uiter);
2467 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2468 struct btrfs_qgroup_list *glist;
2469
2470 qg = unode_aux_to_qgroup(tmp_unode);
2471 if (update_old)
2472 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2473 else
2474 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2475 list_for_each_entry(glist, &qg->groups, next_group) {
2476 ret = ulist_add(qgroups, glist->group->qgroupid,
2477 qgroup_to_aux(glist->group),
2478 GFP_ATOMIC);
2479 if (ret < 0)
2480 return ret;
2481 ret = ulist_add(tmp, glist->group->qgroupid,
2482 qgroup_to_aux(glist->group),
2483 GFP_ATOMIC);
2484 if (ret < 0)
2485 return ret;
2486 }
2487 }
2488 }
2489 return 0;
2490}
2491
2492/*
2493 * Update qgroup rfer/excl counters.
2494 * Rfer update is easy, codes can explain themselves.
2495 *
2496 * Excl update is tricky, the update is split into 2 parts.
2497 * Part 1: Possible exclusive <-> sharing detect:
2498 * | A | !A |
2499 * -------------------------------------
2500 * B | * | - |
2501 * -------------------------------------
2502 * !B | + | ** |
2503 * -------------------------------------
2504 *
2505 * Conditions:
2506 * A: cur_old_roots < nr_old_roots (not exclusive before)
2507 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2508 * B: cur_new_roots < nr_new_roots (not exclusive now)
2509 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2510 *
2511 * Results:
2512 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2513 * *: Definitely not changed. **: Possible unchanged.
2514 *
2515 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2516 *
2517 * To make the logic clear, we first use condition A and B to split
2518 * combination into 4 results.
2519 *
2520 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2521 * only on variant maybe 0.
2522 *
2523 * Lastly, check result **, since there are 2 variants maybe 0, split them
2524 * again(2x2).
2525 * But this time we don't need to consider other things, the codes and logic
2526 * is easy to understand now.
2527 */
2528static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2529 struct ulist *qgroups,
2530 u64 nr_old_roots,
2531 u64 nr_new_roots,
2532 u64 num_bytes, u64 seq)
2533{
2534 struct ulist_node *unode;
2535 struct ulist_iterator uiter;
2536 struct btrfs_qgroup *qg;
2537 u64 cur_new_count, cur_old_count;
2538
2539 ULIST_ITER_INIT(&uiter);
2540 while ((unode = ulist_next(qgroups, &uiter))) {
2541 bool dirty = false;
2542
2543 qg = unode_aux_to_qgroup(unode);
2544 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2545 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2546
2547 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2548 cur_new_count);
2549
2550 /* Rfer update part */
2551 if (cur_old_count == 0 && cur_new_count > 0) {
2552 qg->rfer += num_bytes;
2553 qg->rfer_cmpr += num_bytes;
2554 dirty = true;
2555 }
2556 if (cur_old_count > 0 && cur_new_count == 0) {
2557 qg->rfer -= num_bytes;
2558 qg->rfer_cmpr -= num_bytes;
2559 dirty = true;
2560 }
2561
2562 /* Excl update part */
2563 /* Exclusive/none -> shared case */
2564 if (cur_old_count == nr_old_roots &&
2565 cur_new_count < nr_new_roots) {
2566 /* Exclusive -> shared */
2567 if (cur_old_count != 0) {
2568 qg->excl -= num_bytes;
2569 qg->excl_cmpr -= num_bytes;
2570 dirty = true;
2571 }
2572 }
2573
2574 /* Shared -> exclusive/none case */
2575 if (cur_old_count < nr_old_roots &&
2576 cur_new_count == nr_new_roots) {
2577 /* Shared->exclusive */
2578 if (cur_new_count != 0) {
2579 qg->excl += num_bytes;
2580 qg->excl_cmpr += num_bytes;
2581 dirty = true;
2582 }
2583 }
2584
2585 /* Exclusive/none -> exclusive/none case */
2586 if (cur_old_count == nr_old_roots &&
2587 cur_new_count == nr_new_roots) {
2588 if (cur_old_count == 0) {
2589 /* None -> exclusive/none */
2590
2591 if (cur_new_count != 0) {
2592 /* None -> exclusive */
2593 qg->excl += num_bytes;
2594 qg->excl_cmpr += num_bytes;
2595 dirty = true;
2596 }
2597 /* None -> none, nothing changed */
2598 } else {
2599 /* Exclusive -> exclusive/none */
2600
2601 if (cur_new_count == 0) {
2602 /* Exclusive -> none */
2603 qg->excl -= num_bytes;
2604 qg->excl_cmpr -= num_bytes;
2605 dirty = true;
2606 }
2607 /* Exclusive -> exclusive, nothing changed */
2608 }
2609 }
2610
2611 if (dirty)
2612 qgroup_dirty(fs_info, qg);
2613 }
2614 return 0;
2615}
2616
2617/*
2618 * Check if the @roots potentially is a list of fs tree roots
2619 *
2620 * Return 0 for definitely not a fs/subvol tree roots ulist
2621 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2622 * one as well)
2623 */
2624static int maybe_fs_roots(struct ulist *roots)
2625{
2626 struct ulist_node *unode;
2627 struct ulist_iterator uiter;
2628
2629 /* Empty one, still possible for fs roots */
2630 if (!roots || roots->nnodes == 0)
2631 return 1;
2632
2633 ULIST_ITER_INIT(&uiter);
2634 unode = ulist_next(roots, &uiter);
2635 if (!unode)
2636 return 1;
2637
2638 /*
2639 * If it contains fs tree roots, then it must belong to fs/subvol
2640 * trees.
2641 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2642 */
2643 return is_fstree(unode->val);
2644}
2645
2646int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2647 u64 num_bytes, struct ulist *old_roots,
2648 struct ulist *new_roots)
2649{
2650 struct btrfs_fs_info *fs_info = trans->fs_info;
2651 struct ulist *qgroups = NULL;
2652 struct ulist *tmp = NULL;
2653 u64 seq;
2654 u64 nr_new_roots = 0;
2655 u64 nr_old_roots = 0;
2656 int ret = 0;
2657
2658 /*
2659 * If quotas get disabled meanwhile, the resources need to be freed and
2660 * we can't just exit here.
2661 */
2662 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2663 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
2664 goto out_free;
2665
2666 if (new_roots) {
2667 if (!maybe_fs_roots(new_roots))
2668 goto out_free;
2669 nr_new_roots = new_roots->nnodes;
2670 }
2671 if (old_roots) {
2672 if (!maybe_fs_roots(old_roots))
2673 goto out_free;
2674 nr_old_roots = old_roots->nnodes;
2675 }
2676
2677 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2678 if (nr_old_roots == 0 && nr_new_roots == 0)
2679 goto out_free;
2680
2681 BUG_ON(!fs_info->quota_root);
2682
2683 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2684 num_bytes, nr_old_roots, nr_new_roots);
2685
2686 qgroups = ulist_alloc(GFP_NOFS);
2687 if (!qgroups) {
2688 ret = -ENOMEM;
2689 goto out_free;
2690 }
2691 tmp = ulist_alloc(GFP_NOFS);
2692 if (!tmp) {
2693 ret = -ENOMEM;
2694 goto out_free;
2695 }
2696
2697 mutex_lock(&fs_info->qgroup_rescan_lock);
2698 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2699 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2700 mutex_unlock(&fs_info->qgroup_rescan_lock);
2701 ret = 0;
2702 goto out_free;
2703 }
2704 }
2705 mutex_unlock(&fs_info->qgroup_rescan_lock);
2706
2707 spin_lock(&fs_info->qgroup_lock);
2708 seq = fs_info->qgroup_seq;
2709
2710 /* Update old refcnts using old_roots */
2711 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2712 UPDATE_OLD);
2713 if (ret < 0)
2714 goto out;
2715
2716 /* Update new refcnts using new_roots */
2717 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2718 UPDATE_NEW);
2719 if (ret < 0)
2720 goto out;
2721
2722 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2723 num_bytes, seq);
2724
2725 /*
2726 * Bump qgroup_seq to avoid seq overlap
2727 */
2728 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2729out:
2730 spin_unlock(&fs_info->qgroup_lock);
2731out_free:
2732 ulist_free(tmp);
2733 ulist_free(qgroups);
2734 ulist_free(old_roots);
2735 ulist_free(new_roots);
2736 return ret;
2737}
2738
2739int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2740{
2741 struct btrfs_fs_info *fs_info = trans->fs_info;
2742 struct btrfs_qgroup_extent_record *record;
2743 struct btrfs_delayed_ref_root *delayed_refs;
2744 struct ulist *new_roots = NULL;
2745 struct rb_node *node;
2746 u64 num_dirty_extents = 0;
2747 u64 qgroup_to_skip;
2748 int ret = 0;
2749
2750 delayed_refs = &trans->transaction->delayed_refs;
2751 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2752 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2753 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2754 node);
2755
2756 num_dirty_extents++;
2757 trace_btrfs_qgroup_account_extents(fs_info, record);
2758
2759 if (!ret && !(fs_info->qgroup_flags &
2760 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)) {
2761 struct btrfs_backref_walk_ctx ctx = { 0 };
2762
2763 ctx.bytenr = record->bytenr;
2764 ctx.fs_info = fs_info;
2765
2766 /*
2767 * Old roots should be searched when inserting qgroup
2768 * extent record.
2769 *
2770 * But for INCONSISTENT (NO_ACCOUNTING) -> rescan case,
2771 * we may have some record inserted during
2772 * NO_ACCOUNTING (thus no old_roots populated), but
2773 * later we start rescan, which clears NO_ACCOUNTING,
2774 * leaving some inserted records without old_roots
2775 * populated.
2776 *
2777 * Those cases are rare and should not cause too much
2778 * time spent during commit_transaction().
2779 */
2780 if (!record->old_roots) {
2781 /* Search commit root to find old_roots */
2782 ret = btrfs_find_all_roots(&ctx, false);
2783 if (ret < 0)
2784 goto cleanup;
2785 record->old_roots = ctx.roots;
2786 ctx.roots = NULL;
2787 }
2788
2789 /* Free the reserved data space */
2790 btrfs_qgroup_free_refroot(fs_info,
2791 record->data_rsv_refroot,
2792 record->data_rsv,
2793 BTRFS_QGROUP_RSV_DATA);
2794 /*
2795 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2796 * which doesn't lock tree or delayed_refs and search
2797 * current root. It's safe inside commit_transaction().
2798 */
2799 ctx.trans = trans;
2800 ctx.time_seq = BTRFS_SEQ_LAST;
2801 ret = btrfs_find_all_roots(&ctx, false);
2802 if (ret < 0)
2803 goto cleanup;
2804 new_roots = ctx.roots;
2805 if (qgroup_to_skip) {
2806 ulist_del(new_roots, qgroup_to_skip, 0);
2807 ulist_del(record->old_roots, qgroup_to_skip,
2808 0);
2809 }
2810 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2811 record->num_bytes,
2812 record->old_roots,
2813 new_roots);
2814 record->old_roots = NULL;
2815 new_roots = NULL;
2816 }
2817cleanup:
2818 ulist_free(record->old_roots);
2819 ulist_free(new_roots);
2820 new_roots = NULL;
2821 rb_erase(node, &delayed_refs->dirty_extent_root);
2822 kfree(record);
2823
2824 }
2825 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2826 num_dirty_extents);
2827 return ret;
2828}
2829
2830/*
2831 * called from commit_transaction. Writes all changed qgroups to disk.
2832 */
2833int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2834{
2835 struct btrfs_fs_info *fs_info = trans->fs_info;
2836 int ret = 0;
2837
2838 if (!fs_info->quota_root)
2839 return ret;
2840
2841 spin_lock(&fs_info->qgroup_lock);
2842 while (!list_empty(&fs_info->dirty_qgroups)) {
2843 struct btrfs_qgroup *qgroup;
2844 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2845 struct btrfs_qgroup, dirty);
2846 list_del_init(&qgroup->dirty);
2847 spin_unlock(&fs_info->qgroup_lock);
2848 ret = update_qgroup_info_item(trans, qgroup);
2849 if (ret)
2850 qgroup_mark_inconsistent(fs_info);
2851 ret = update_qgroup_limit_item(trans, qgroup);
2852 if (ret)
2853 qgroup_mark_inconsistent(fs_info);
2854 spin_lock(&fs_info->qgroup_lock);
2855 }
2856 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2857 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2858 else
2859 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2860 spin_unlock(&fs_info->qgroup_lock);
2861
2862 ret = update_qgroup_status_item(trans);
2863 if (ret)
2864 qgroup_mark_inconsistent(fs_info);
2865
2866 return ret;
2867}
2868
2869/*
2870 * Copy the accounting information between qgroups. This is necessary
2871 * when a snapshot or a subvolume is created. Throwing an error will
2872 * cause a transaction abort so we take extra care here to only error
2873 * when a readonly fs is a reasonable outcome.
2874 */
2875int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2876 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2877{
2878 int ret = 0;
2879 int i;
2880 u64 *i_qgroups;
2881 bool committing = false;
2882 struct btrfs_fs_info *fs_info = trans->fs_info;
2883 struct btrfs_root *quota_root;
2884 struct btrfs_qgroup *srcgroup;
2885 struct btrfs_qgroup *dstgroup;
2886 bool need_rescan = false;
2887 u32 level_size = 0;
2888 u64 nums;
2889
2890 /*
2891 * There are only two callers of this function.
2892 *
2893 * One in create_subvol() in the ioctl context, which needs to hold
2894 * the qgroup_ioctl_lock.
2895 *
2896 * The other one in create_pending_snapshot() where no other qgroup
2897 * code can modify the fs as they all need to either start a new trans
2898 * or hold a trans handler, thus we don't need to hold
2899 * qgroup_ioctl_lock.
2900 * This would avoid long and complex lock chain and make lockdep happy.
2901 */
2902 spin_lock(&fs_info->trans_lock);
2903 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2904 committing = true;
2905 spin_unlock(&fs_info->trans_lock);
2906
2907 if (!committing)
2908 mutex_lock(&fs_info->qgroup_ioctl_lock);
2909 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2910 goto out;
2911
2912 quota_root = fs_info->quota_root;
2913 if (!quota_root) {
2914 ret = -EINVAL;
2915 goto out;
2916 }
2917
2918 if (inherit) {
2919 i_qgroups = (u64 *)(inherit + 1);
2920 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2921 2 * inherit->num_excl_copies;
2922 for (i = 0; i < nums; ++i) {
2923 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2924
2925 /*
2926 * Zero out invalid groups so we can ignore
2927 * them later.
2928 */
2929 if (!srcgroup ||
2930 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2931 *i_qgroups = 0ULL;
2932
2933 ++i_qgroups;
2934 }
2935 }
2936
2937 /*
2938 * create a tracking group for the subvol itself
2939 */
2940 ret = add_qgroup_item(trans, quota_root, objectid);
2941 if (ret)
2942 goto out;
2943
2944 /*
2945 * add qgroup to all inherited groups
2946 */
2947 if (inherit) {
2948 i_qgroups = (u64 *)(inherit + 1);
2949 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2950 if (*i_qgroups == 0)
2951 continue;
2952 ret = add_qgroup_relation_item(trans, objectid,
2953 *i_qgroups);
2954 if (ret && ret != -EEXIST)
2955 goto out;
2956 ret = add_qgroup_relation_item(trans, *i_qgroups,
2957 objectid);
2958 if (ret && ret != -EEXIST)
2959 goto out;
2960 }
2961 ret = 0;
2962 }
2963
2964
2965 spin_lock(&fs_info->qgroup_lock);
2966
2967 dstgroup = add_qgroup_rb(fs_info, objectid);
2968 if (IS_ERR(dstgroup)) {
2969 ret = PTR_ERR(dstgroup);
2970 goto unlock;
2971 }
2972
2973 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2974 dstgroup->lim_flags = inherit->lim.flags;
2975 dstgroup->max_rfer = inherit->lim.max_rfer;
2976 dstgroup->max_excl = inherit->lim.max_excl;
2977 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2978 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2979
2980 qgroup_dirty(fs_info, dstgroup);
2981 }
2982
2983 if (srcid) {
2984 srcgroup = find_qgroup_rb(fs_info, srcid);
2985 if (!srcgroup)
2986 goto unlock;
2987
2988 /*
2989 * We call inherit after we clone the root in order to make sure
2990 * our counts don't go crazy, so at this point the only
2991 * difference between the two roots should be the root node.
2992 */
2993 level_size = fs_info->nodesize;
2994 dstgroup->rfer = srcgroup->rfer;
2995 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2996 dstgroup->excl = level_size;
2997 dstgroup->excl_cmpr = level_size;
2998 srcgroup->excl = level_size;
2999 srcgroup->excl_cmpr = level_size;
3000
3001 /* inherit the limit info */
3002 dstgroup->lim_flags = srcgroup->lim_flags;
3003 dstgroup->max_rfer = srcgroup->max_rfer;
3004 dstgroup->max_excl = srcgroup->max_excl;
3005 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
3006 dstgroup->rsv_excl = srcgroup->rsv_excl;
3007
3008 qgroup_dirty(fs_info, dstgroup);
3009 qgroup_dirty(fs_info, srcgroup);
3010 }
3011
3012 if (!inherit)
3013 goto unlock;
3014
3015 i_qgroups = (u64 *)(inherit + 1);
3016 for (i = 0; i < inherit->num_qgroups; ++i) {
3017 if (*i_qgroups) {
3018 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
3019 if (ret)
3020 goto unlock;
3021 }
3022 ++i_qgroups;
3023
3024 /*
3025 * If we're doing a snapshot, and adding the snapshot to a new
3026 * qgroup, the numbers are guaranteed to be incorrect.
3027 */
3028 if (srcid)
3029 need_rescan = true;
3030 }
3031
3032 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
3033 struct btrfs_qgroup *src;
3034 struct btrfs_qgroup *dst;
3035
3036 if (!i_qgroups[0] || !i_qgroups[1])
3037 continue;
3038
3039 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3040 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3041
3042 if (!src || !dst) {
3043 ret = -EINVAL;
3044 goto unlock;
3045 }
3046
3047 dst->rfer = src->rfer - level_size;
3048 dst->rfer_cmpr = src->rfer_cmpr - level_size;
3049
3050 /* Manually tweaking numbers certainly needs a rescan */
3051 need_rescan = true;
3052 }
3053 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
3054 struct btrfs_qgroup *src;
3055 struct btrfs_qgroup *dst;
3056
3057 if (!i_qgroups[0] || !i_qgroups[1])
3058 continue;
3059
3060 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3061 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3062
3063 if (!src || !dst) {
3064 ret = -EINVAL;
3065 goto unlock;
3066 }
3067
3068 dst->excl = src->excl + level_size;
3069 dst->excl_cmpr = src->excl_cmpr + level_size;
3070 need_rescan = true;
3071 }
3072
3073unlock:
3074 spin_unlock(&fs_info->qgroup_lock);
3075 if (!ret)
3076 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3077out:
3078 if (!committing)
3079 mutex_unlock(&fs_info->qgroup_ioctl_lock);
3080 if (need_rescan)
3081 qgroup_mark_inconsistent(fs_info);
3082 return ret;
3083}
3084
3085static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3086{
3087 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3088 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3089 return false;
3090
3091 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3092 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3093 return false;
3094
3095 return true;
3096}
3097
3098static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3099 enum btrfs_qgroup_rsv_type type)
3100{
3101 struct btrfs_qgroup *qgroup;
3102 struct btrfs_fs_info *fs_info = root->fs_info;
3103 u64 ref_root = root->root_key.objectid;
3104 int ret = 0;
3105 struct ulist_node *unode;
3106 struct ulist_iterator uiter;
3107
3108 if (!is_fstree(ref_root))
3109 return 0;
3110
3111 if (num_bytes == 0)
3112 return 0;
3113
3114 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3115 capable(CAP_SYS_RESOURCE))
3116 enforce = false;
3117
3118 spin_lock(&fs_info->qgroup_lock);
3119 if (!fs_info->quota_root)
3120 goto out;
3121
3122 qgroup = find_qgroup_rb(fs_info, ref_root);
3123 if (!qgroup)
3124 goto out;
3125
3126 /*
3127 * in a first step, we check all affected qgroups if any limits would
3128 * be exceeded
3129 */
3130 ulist_reinit(fs_info->qgroup_ulist);
3131 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3132 qgroup_to_aux(qgroup), GFP_ATOMIC);
3133 if (ret < 0)
3134 goto out;
3135 ULIST_ITER_INIT(&uiter);
3136 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3137 struct btrfs_qgroup *qg;
3138 struct btrfs_qgroup_list *glist;
3139
3140 qg = unode_aux_to_qgroup(unode);
3141
3142 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3143 ret = -EDQUOT;
3144 goto out;
3145 }
3146
3147 list_for_each_entry(glist, &qg->groups, next_group) {
3148 ret = ulist_add(fs_info->qgroup_ulist,
3149 glist->group->qgroupid,
3150 qgroup_to_aux(glist->group), GFP_ATOMIC);
3151 if (ret < 0)
3152 goto out;
3153 }
3154 }
3155 ret = 0;
3156 /*
3157 * no limits exceeded, now record the reservation into all qgroups
3158 */
3159 ULIST_ITER_INIT(&uiter);
3160 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3161 struct btrfs_qgroup *qg;
3162
3163 qg = unode_aux_to_qgroup(unode);
3164
3165 qgroup_rsv_add(fs_info, qg, num_bytes, type);
3166 }
3167
3168out:
3169 spin_unlock(&fs_info->qgroup_lock);
3170 return ret;
3171}
3172
3173/*
3174 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
3175 * qgroup).
3176 *
3177 * Will handle all higher level qgroup too.
3178 *
3179 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3180 * This special case is only used for META_PERTRANS type.
3181 */
3182void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3183 u64 ref_root, u64 num_bytes,
3184 enum btrfs_qgroup_rsv_type type)
3185{
3186 struct btrfs_qgroup *qgroup;
3187 struct ulist_node *unode;
3188 struct ulist_iterator uiter;
3189 int ret = 0;
3190
3191 if (!is_fstree(ref_root))
3192 return;
3193
3194 if (num_bytes == 0)
3195 return;
3196
3197 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3198 WARN(1, "%s: Invalid type to free", __func__);
3199 return;
3200 }
3201 spin_lock(&fs_info->qgroup_lock);
3202
3203 if (!fs_info->quota_root)
3204 goto out;
3205
3206 qgroup = find_qgroup_rb(fs_info, ref_root);
3207 if (!qgroup)
3208 goto out;
3209
3210 if (num_bytes == (u64)-1)
3211 /*
3212 * We're freeing all pertrans rsv, get reserved value from
3213 * level 0 qgroup as real num_bytes to free.
3214 */
3215 num_bytes = qgroup->rsv.values[type];
3216
3217 ulist_reinit(fs_info->qgroup_ulist);
3218 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3219 qgroup_to_aux(qgroup), GFP_ATOMIC);
3220 if (ret < 0)
3221 goto out;
3222 ULIST_ITER_INIT(&uiter);
3223 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3224 struct btrfs_qgroup *qg;
3225 struct btrfs_qgroup_list *glist;
3226
3227 qg = unode_aux_to_qgroup(unode);
3228
3229 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3230
3231 list_for_each_entry(glist, &qg->groups, next_group) {
3232 ret = ulist_add(fs_info->qgroup_ulist,
3233 glist->group->qgroupid,
3234 qgroup_to_aux(glist->group), GFP_ATOMIC);
3235 if (ret < 0)
3236 goto out;
3237 }
3238 }
3239
3240out:
3241 spin_unlock(&fs_info->qgroup_lock);
3242}
3243
3244/*
3245 * Check if the leaf is the last leaf. Which means all node pointers
3246 * are at their last position.
3247 */
3248static bool is_last_leaf(struct btrfs_path *path)
3249{
3250 int i;
3251
3252 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3253 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3254 return false;
3255 }
3256 return true;
3257}
3258
3259/*
3260 * returns < 0 on error, 0 when more leafs are to be scanned.
3261 * returns 1 when done.
3262 */
3263static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3264 struct btrfs_path *path)
3265{
3266 struct btrfs_fs_info *fs_info = trans->fs_info;
3267 struct btrfs_root *extent_root;
3268 struct btrfs_key found;
3269 struct extent_buffer *scratch_leaf = NULL;
3270 u64 num_bytes;
3271 bool done;
3272 int slot;
3273 int ret;
3274
3275 mutex_lock(&fs_info->qgroup_rescan_lock);
3276 extent_root = btrfs_extent_root(fs_info,
3277 fs_info->qgroup_rescan_progress.objectid);
3278 ret = btrfs_search_slot_for_read(extent_root,
3279 &fs_info->qgroup_rescan_progress,
3280 path, 1, 0);
3281
3282 btrfs_debug(fs_info,
3283 "current progress key (%llu %u %llu), search_slot ret %d",
3284 fs_info->qgroup_rescan_progress.objectid,
3285 fs_info->qgroup_rescan_progress.type,
3286 fs_info->qgroup_rescan_progress.offset, ret);
3287
3288 if (ret) {
3289 /*
3290 * The rescan is about to end, we will not be scanning any
3291 * further blocks. We cannot unset the RESCAN flag here, because
3292 * we want to commit the transaction if everything went well.
3293 * To make the live accounting work in this phase, we set our
3294 * scan progress pointer such that every real extent objectid
3295 * will be smaller.
3296 */
3297 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3298 btrfs_release_path(path);
3299 mutex_unlock(&fs_info->qgroup_rescan_lock);
3300 return ret;
3301 }
3302 done = is_last_leaf(path);
3303
3304 btrfs_item_key_to_cpu(path->nodes[0], &found,
3305 btrfs_header_nritems(path->nodes[0]) - 1);
3306 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3307
3308 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3309 if (!scratch_leaf) {
3310 ret = -ENOMEM;
3311 mutex_unlock(&fs_info->qgroup_rescan_lock);
3312 goto out;
3313 }
3314 slot = path->slots[0];
3315 btrfs_release_path(path);
3316 mutex_unlock(&fs_info->qgroup_rescan_lock);
3317
3318 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3319 struct btrfs_backref_walk_ctx ctx = { 0 };
3320
3321 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3322 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3323 found.type != BTRFS_METADATA_ITEM_KEY)
3324 continue;
3325 if (found.type == BTRFS_METADATA_ITEM_KEY)
3326 num_bytes = fs_info->nodesize;
3327 else
3328 num_bytes = found.offset;
3329
3330 ctx.bytenr = found.objectid;
3331 ctx.fs_info = fs_info;
3332
3333 ret = btrfs_find_all_roots(&ctx, false);
3334 if (ret < 0)
3335 goto out;
3336 /* For rescan, just pass old_roots as NULL */
3337 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3338 num_bytes, NULL, ctx.roots);
3339 if (ret < 0)
3340 goto out;
3341 }
3342out:
3343 if (scratch_leaf)
3344 free_extent_buffer(scratch_leaf);
3345
3346 if (done && !ret) {
3347 ret = 1;
3348 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3349 }
3350 return ret;
3351}
3352
3353static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3354{
3355 return btrfs_fs_closing(fs_info) ||
3356 test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3357 !test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3358 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3359}
3360
3361static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3362{
3363 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3364 qgroup_rescan_work);
3365 struct btrfs_path *path;
3366 struct btrfs_trans_handle *trans = NULL;
3367 int err = -ENOMEM;
3368 int ret = 0;
3369 bool stopped = false;
3370 bool did_leaf_rescans = false;
3371
3372 path = btrfs_alloc_path();
3373 if (!path)
3374 goto out;
3375 /*
3376 * Rescan should only search for commit root, and any later difference
3377 * should be recorded by qgroup
3378 */
3379 path->search_commit_root = 1;
3380 path->skip_locking = 1;
3381
3382 err = 0;
3383 while (!err && !(stopped = rescan_should_stop(fs_info))) {
3384 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3385 if (IS_ERR(trans)) {
3386 err = PTR_ERR(trans);
3387 break;
3388 }
3389
3390 err = qgroup_rescan_leaf(trans, path);
3391 did_leaf_rescans = true;
3392
3393 if (err > 0)
3394 btrfs_commit_transaction(trans);
3395 else
3396 btrfs_end_transaction(trans);
3397 }
3398
3399out:
3400 btrfs_free_path(path);
3401
3402 mutex_lock(&fs_info->qgroup_rescan_lock);
3403 if (err > 0 &&
3404 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3405 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3406 } else if (err < 0 || stopped) {
3407 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3408 }
3409 mutex_unlock(&fs_info->qgroup_rescan_lock);
3410
3411 /*
3412 * Only update status, since the previous part has already updated the
3413 * qgroup info, and only if we did any actual work. This also prevents
3414 * race with a concurrent quota disable, which has already set
3415 * fs_info->quota_root to NULL and cleared BTRFS_FS_QUOTA_ENABLED at
3416 * btrfs_quota_disable().
3417 */
3418 if (did_leaf_rescans) {
3419 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3420 if (IS_ERR(trans)) {
3421 err = PTR_ERR(trans);
3422 trans = NULL;
3423 btrfs_err(fs_info,
3424 "fail to start transaction for status update: %d",
3425 err);
3426 }
3427 } else {
3428 trans = NULL;
3429 }
3430
3431 mutex_lock(&fs_info->qgroup_rescan_lock);
3432 if (!stopped ||
3433 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN)
3434 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3435 if (trans) {
3436 ret = update_qgroup_status_item(trans);
3437 if (ret < 0) {
3438 err = ret;
3439 btrfs_err(fs_info, "fail to update qgroup status: %d",
3440 err);
3441 }
3442 }
3443 fs_info->qgroup_rescan_running = false;
3444 fs_info->qgroup_flags &= ~BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3445 complete_all(&fs_info->qgroup_rescan_completion);
3446 mutex_unlock(&fs_info->qgroup_rescan_lock);
3447
3448 if (!trans)
3449 return;
3450
3451 btrfs_end_transaction(trans);
3452
3453 if (stopped) {
3454 btrfs_info(fs_info, "qgroup scan paused");
3455 } else if (fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN) {
3456 btrfs_info(fs_info, "qgroup scan cancelled");
3457 } else if (err >= 0) {
3458 btrfs_info(fs_info, "qgroup scan completed%s",
3459 err > 0 ? " (inconsistency flag cleared)" : "");
3460 } else {
3461 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3462 }
3463}
3464
3465/*
3466 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3467 * memory required for the rescan context.
3468 */
3469static int
3470qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3471 int init_flags)
3472{
3473 int ret = 0;
3474
3475 if (!init_flags) {
3476 /* we're resuming qgroup rescan at mount time */
3477 if (!(fs_info->qgroup_flags &
3478 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3479 btrfs_warn(fs_info,
3480 "qgroup rescan init failed, qgroup rescan is not queued");
3481 ret = -EINVAL;
3482 } else if (!(fs_info->qgroup_flags &
3483 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3484 btrfs_warn(fs_info,
3485 "qgroup rescan init failed, qgroup is not enabled");
3486 ret = -EINVAL;
3487 }
3488
3489 if (ret)
3490 return ret;
3491 }
3492
3493 mutex_lock(&fs_info->qgroup_rescan_lock);
3494
3495 if (init_flags) {
3496 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3497 btrfs_warn(fs_info,
3498 "qgroup rescan is already in progress");
3499 ret = -EINPROGRESS;
3500 } else if (!(fs_info->qgroup_flags &
3501 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3502 btrfs_warn(fs_info,
3503 "qgroup rescan init failed, qgroup is not enabled");
3504 ret = -EINVAL;
3505 } else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3506 /* Quota disable is in progress */
3507 ret = -EBUSY;
3508 }
3509
3510 if (ret) {
3511 mutex_unlock(&fs_info->qgroup_rescan_lock);
3512 return ret;
3513 }
3514 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3515 }
3516
3517 memset(&fs_info->qgroup_rescan_progress, 0,
3518 sizeof(fs_info->qgroup_rescan_progress));
3519 fs_info->qgroup_flags &= ~(BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
3520 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
3521 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3522 init_completion(&fs_info->qgroup_rescan_completion);
3523 mutex_unlock(&fs_info->qgroup_rescan_lock);
3524
3525 btrfs_init_work(&fs_info->qgroup_rescan_work,
3526 btrfs_qgroup_rescan_worker, NULL, NULL);
3527 return 0;
3528}
3529
3530static void
3531qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3532{
3533 struct rb_node *n;
3534 struct btrfs_qgroup *qgroup;
3535
3536 spin_lock(&fs_info->qgroup_lock);
3537 /* clear all current qgroup tracking information */
3538 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3539 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3540 qgroup->rfer = 0;
3541 qgroup->rfer_cmpr = 0;
3542 qgroup->excl = 0;
3543 qgroup->excl_cmpr = 0;
3544 qgroup_dirty(fs_info, qgroup);
3545 }
3546 spin_unlock(&fs_info->qgroup_lock);
3547}
3548
3549int
3550btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3551{
3552 int ret = 0;
3553 struct btrfs_trans_handle *trans;
3554
3555 ret = qgroup_rescan_init(fs_info, 0, 1);
3556 if (ret)
3557 return ret;
3558
3559 /*
3560 * We have set the rescan_progress to 0, which means no more
3561 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3562 * However, btrfs_qgroup_account_ref may be right after its call
3563 * to btrfs_find_all_roots, in which case it would still do the
3564 * accounting.
3565 * To solve this, we're committing the transaction, which will
3566 * ensure we run all delayed refs and only after that, we are
3567 * going to clear all tracking information for a clean start.
3568 */
3569
3570 trans = btrfs_join_transaction(fs_info->fs_root);
3571 if (IS_ERR(trans)) {
3572 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3573 return PTR_ERR(trans);
3574 }
3575 ret = btrfs_commit_transaction(trans);
3576 if (ret) {
3577 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3578 return ret;
3579 }
3580
3581 qgroup_rescan_zero_tracking(fs_info);
3582
3583 mutex_lock(&fs_info->qgroup_rescan_lock);
3584 fs_info->qgroup_rescan_running = true;
3585 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3586 &fs_info->qgroup_rescan_work);
3587 mutex_unlock(&fs_info->qgroup_rescan_lock);
3588
3589 return 0;
3590}
3591
3592int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3593 bool interruptible)
3594{
3595 int running;
3596 int ret = 0;
3597
3598 mutex_lock(&fs_info->qgroup_rescan_lock);
3599 running = fs_info->qgroup_rescan_running;
3600 mutex_unlock(&fs_info->qgroup_rescan_lock);
3601
3602 if (!running)
3603 return 0;
3604
3605 if (interruptible)
3606 ret = wait_for_completion_interruptible(
3607 &fs_info->qgroup_rescan_completion);
3608 else
3609 wait_for_completion(&fs_info->qgroup_rescan_completion);
3610
3611 return ret;
3612}
3613
3614/*
3615 * this is only called from open_ctree where we're still single threaded, thus
3616 * locking is omitted here.
3617 */
3618void
3619btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3620{
3621 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3622 mutex_lock(&fs_info->qgroup_rescan_lock);
3623 fs_info->qgroup_rescan_running = true;
3624 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3625 &fs_info->qgroup_rescan_work);
3626 mutex_unlock(&fs_info->qgroup_rescan_lock);
3627 }
3628}
3629
3630#define rbtree_iterate_from_safe(node, next, start) \
3631 for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3632
3633static int qgroup_unreserve_range(struct btrfs_inode *inode,
3634 struct extent_changeset *reserved, u64 start,
3635 u64 len)
3636{
3637 struct rb_node *node;
3638 struct rb_node *next;
3639 struct ulist_node *entry;
3640 int ret = 0;
3641
3642 node = reserved->range_changed.root.rb_node;
3643 if (!node)
3644 return 0;
3645 while (node) {
3646 entry = rb_entry(node, struct ulist_node, rb_node);
3647 if (entry->val < start)
3648 node = node->rb_right;
3649 else
3650 node = node->rb_left;
3651 }
3652
3653 if (entry->val > start && rb_prev(&entry->rb_node))
3654 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3655 rb_node);
3656
3657 rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3658 u64 entry_start;
3659 u64 entry_end;
3660 u64 entry_len;
3661 int clear_ret;
3662
3663 entry = rb_entry(node, struct ulist_node, rb_node);
3664 entry_start = entry->val;
3665 entry_end = entry->aux;
3666 entry_len = entry_end - entry_start + 1;
3667
3668 if (entry_start >= start + len)
3669 break;
3670 if (entry_start + entry_len <= start)
3671 continue;
3672 /*
3673 * Now the entry is in [start, start + len), revert the
3674 * EXTENT_QGROUP_RESERVED bit.
3675 */
3676 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3677 entry_end, EXTENT_QGROUP_RESERVED);
3678 if (!ret && clear_ret < 0)
3679 ret = clear_ret;
3680
3681 ulist_del(&reserved->range_changed, entry->val, entry->aux);
3682 if (likely(reserved->bytes_changed >= entry_len)) {
3683 reserved->bytes_changed -= entry_len;
3684 } else {
3685 WARN_ON(1);
3686 reserved->bytes_changed = 0;
3687 }
3688 }
3689
3690 return ret;
3691}
3692
3693/*
3694 * Try to free some space for qgroup.
3695 *
3696 * For qgroup, there are only 3 ways to free qgroup space:
3697 * - Flush nodatacow write
3698 * Any nodatacow write will free its reserved data space at run_delalloc_range().
3699 * In theory, we should only flush nodatacow inodes, but it's not yet
3700 * possible, so we need to flush the whole root.
3701 *
3702 * - Wait for ordered extents
3703 * When ordered extents are finished, their reserved metadata is finally
3704 * converted to per_trans status, which can be freed by later commit
3705 * transaction.
3706 *
3707 * - Commit transaction
3708 * This would free the meta_per_trans space.
3709 * In theory this shouldn't provide much space, but any more qgroup space
3710 * is needed.
3711 */
3712static int try_flush_qgroup(struct btrfs_root *root)
3713{
3714 struct btrfs_trans_handle *trans;
3715 int ret;
3716
3717 /* Can't hold an open transaction or we run the risk of deadlocking. */
3718 ASSERT(current->journal_info == NULL);
3719 if (WARN_ON(current->journal_info))
3720 return 0;
3721
3722 /*
3723 * We don't want to run flush again and again, so if there is a running
3724 * one, we won't try to start a new flush, but exit directly.
3725 */
3726 if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3727 wait_event(root->qgroup_flush_wait,
3728 !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3729 return 0;
3730 }
3731
3732 ret = btrfs_start_delalloc_snapshot(root, true);
3733 if (ret < 0)
3734 goto out;
3735 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3736
3737 trans = btrfs_join_transaction(root);
3738 if (IS_ERR(trans)) {
3739 ret = PTR_ERR(trans);
3740 goto out;
3741 }
3742
3743 ret = btrfs_commit_transaction(trans);
3744out:
3745 clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3746 wake_up(&root->qgroup_flush_wait);
3747 return ret;
3748}
3749
3750static int qgroup_reserve_data(struct btrfs_inode *inode,
3751 struct extent_changeset **reserved_ret, u64 start,
3752 u64 len)
3753{
3754 struct btrfs_root *root = inode->root;
3755 struct extent_changeset *reserved;
3756 bool new_reserved = false;
3757 u64 orig_reserved;
3758 u64 to_reserve;
3759 int ret;
3760
3761 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3762 !is_fstree(root->root_key.objectid) || len == 0)
3763 return 0;
3764
3765 /* @reserved parameter is mandatory for qgroup */
3766 if (WARN_ON(!reserved_ret))
3767 return -EINVAL;
3768 if (!*reserved_ret) {
3769 new_reserved = true;
3770 *reserved_ret = extent_changeset_alloc();
3771 if (!*reserved_ret)
3772 return -ENOMEM;
3773 }
3774 reserved = *reserved_ret;
3775 /* Record already reserved space */
3776 orig_reserved = reserved->bytes_changed;
3777 ret = set_record_extent_bits(&inode->io_tree, start,
3778 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3779
3780 /* Newly reserved space */
3781 to_reserve = reserved->bytes_changed - orig_reserved;
3782 trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3783 to_reserve, QGROUP_RESERVE);
3784 if (ret < 0)
3785 goto out;
3786 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3787 if (ret < 0)
3788 goto cleanup;
3789
3790 return ret;
3791
3792cleanup:
3793 qgroup_unreserve_range(inode, reserved, start, len);
3794out:
3795 if (new_reserved) {
3796 extent_changeset_free(reserved);
3797 *reserved_ret = NULL;
3798 }
3799 return ret;
3800}
3801
3802/*
3803 * Reserve qgroup space for range [start, start + len).
3804 *
3805 * This function will either reserve space from related qgroups or do nothing
3806 * if the range is already reserved.
3807 *
3808 * Return 0 for successful reservation
3809 * Return <0 for error (including -EQUOT)
3810 *
3811 * NOTE: This function may sleep for memory allocation, dirty page flushing and
3812 * commit transaction. So caller should not hold any dirty page locked.
3813 */
3814int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3815 struct extent_changeset **reserved_ret, u64 start,
3816 u64 len)
3817{
3818 int ret;
3819
3820 ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3821 if (ret <= 0 && ret != -EDQUOT)
3822 return ret;
3823
3824 ret = try_flush_qgroup(inode->root);
3825 if (ret < 0)
3826 return ret;
3827 return qgroup_reserve_data(inode, reserved_ret, start, len);
3828}
3829
3830/* Free ranges specified by @reserved, normally in error path */
3831static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3832 struct extent_changeset *reserved, u64 start, u64 len)
3833{
3834 struct btrfs_root *root = inode->root;
3835 struct ulist_node *unode;
3836 struct ulist_iterator uiter;
3837 struct extent_changeset changeset;
3838 int freed = 0;
3839 int ret;
3840
3841 extent_changeset_init(&changeset);
3842 len = round_up(start + len, root->fs_info->sectorsize);
3843 start = round_down(start, root->fs_info->sectorsize);
3844
3845 ULIST_ITER_INIT(&uiter);
3846 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3847 u64 range_start = unode->val;
3848 /* unode->aux is the inclusive end */
3849 u64 range_len = unode->aux - range_start + 1;
3850 u64 free_start;
3851 u64 free_len;
3852
3853 extent_changeset_release(&changeset);
3854
3855 /* Only free range in range [start, start + len) */
3856 if (range_start >= start + len ||
3857 range_start + range_len <= start)
3858 continue;
3859 free_start = max(range_start, start);
3860 free_len = min(start + len, range_start + range_len) -
3861 free_start;
3862 /*
3863 * TODO: To also modify reserved->ranges_reserved to reflect
3864 * the modification.
3865 *
3866 * However as long as we free qgroup reserved according to
3867 * EXTENT_QGROUP_RESERVED, we won't double free.
3868 * So not need to rush.
3869 */
3870 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3871 free_start + free_len - 1,
3872 EXTENT_QGROUP_RESERVED, &changeset);
3873 if (ret < 0)
3874 goto out;
3875 freed += changeset.bytes_changed;
3876 }
3877 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3878 BTRFS_QGROUP_RSV_DATA);
3879 ret = freed;
3880out:
3881 extent_changeset_release(&changeset);
3882 return ret;
3883}
3884
3885static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3886 struct extent_changeset *reserved, u64 start, u64 len,
3887 int free)
3888{
3889 struct extent_changeset changeset;
3890 int trace_op = QGROUP_RELEASE;
3891 int ret;
3892
3893 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3894 return 0;
3895
3896 /* In release case, we shouldn't have @reserved */
3897 WARN_ON(!free && reserved);
3898 if (free && reserved)
3899 return qgroup_free_reserved_data(inode, reserved, start, len);
3900 extent_changeset_init(&changeset);
3901 ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3902 EXTENT_QGROUP_RESERVED, &changeset);
3903 if (ret < 0)
3904 goto out;
3905
3906 if (free)
3907 trace_op = QGROUP_FREE;
3908 trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3909 changeset.bytes_changed, trace_op);
3910 if (free)
3911 btrfs_qgroup_free_refroot(inode->root->fs_info,
3912 inode->root->root_key.objectid,
3913 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3914 ret = changeset.bytes_changed;
3915out:
3916 extent_changeset_release(&changeset);
3917 return ret;
3918}
3919
3920/*
3921 * Free a reserved space range from io_tree and related qgroups
3922 *
3923 * Should be called when a range of pages get invalidated before reaching disk.
3924 * Or for error cleanup case.
3925 * if @reserved is given, only reserved range in [@start, @start + @len) will
3926 * be freed.
3927 *
3928 * For data written to disk, use btrfs_qgroup_release_data().
3929 *
3930 * NOTE: This function may sleep for memory allocation.
3931 */
3932int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3933 struct extent_changeset *reserved, u64 start, u64 len)
3934{
3935 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3936}
3937
3938/*
3939 * Release a reserved space range from io_tree only.
3940 *
3941 * Should be called when a range of pages get written to disk and corresponding
3942 * FILE_EXTENT is inserted into corresponding root.
3943 *
3944 * Since new qgroup accounting framework will only update qgroup numbers at
3945 * commit_transaction() time, its reserved space shouldn't be freed from
3946 * related qgroups.
3947 *
3948 * But we should release the range from io_tree, to allow further write to be
3949 * COWed.
3950 *
3951 * NOTE: This function may sleep for memory allocation.
3952 */
3953int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3954{
3955 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3956}
3957
3958static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3959 enum btrfs_qgroup_rsv_type type)
3960{
3961 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3962 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3963 return;
3964 if (num_bytes == 0)
3965 return;
3966
3967 spin_lock(&root->qgroup_meta_rsv_lock);
3968 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3969 root->qgroup_meta_rsv_prealloc += num_bytes;
3970 else
3971 root->qgroup_meta_rsv_pertrans += num_bytes;
3972 spin_unlock(&root->qgroup_meta_rsv_lock);
3973}
3974
3975static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3976 enum btrfs_qgroup_rsv_type type)
3977{
3978 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3979 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3980 return 0;
3981 if (num_bytes == 0)
3982 return 0;
3983
3984 spin_lock(&root->qgroup_meta_rsv_lock);
3985 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3986 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3987 num_bytes);
3988 root->qgroup_meta_rsv_prealloc -= num_bytes;
3989 } else {
3990 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3991 num_bytes);
3992 root->qgroup_meta_rsv_pertrans -= num_bytes;
3993 }
3994 spin_unlock(&root->qgroup_meta_rsv_lock);
3995 return num_bytes;
3996}
3997
3998int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3999 enum btrfs_qgroup_rsv_type type, bool enforce)
4000{
4001 struct btrfs_fs_info *fs_info = root->fs_info;
4002 int ret;
4003
4004 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4005 !is_fstree(root->root_key.objectid) || num_bytes == 0)
4006 return 0;
4007
4008 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4009 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
4010 ret = qgroup_reserve(root, num_bytes, enforce, type);
4011 if (ret < 0)
4012 return ret;
4013 /*
4014 * Record what we have reserved into root.
4015 *
4016 * To avoid quota disabled->enabled underflow.
4017 * In that case, we may try to free space we haven't reserved
4018 * (since quota was disabled), so record what we reserved into root.
4019 * And ensure later release won't underflow this number.
4020 */
4021 add_root_meta_rsv(root, num_bytes, type);
4022 return ret;
4023}
4024
4025int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4026 enum btrfs_qgroup_rsv_type type, bool enforce,
4027 bool noflush)
4028{
4029 int ret;
4030
4031 ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4032 if ((ret <= 0 && ret != -EDQUOT) || noflush)
4033 return ret;
4034
4035 ret = try_flush_qgroup(root);
4036 if (ret < 0)
4037 return ret;
4038 return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4039}
4040
4041void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
4042{
4043 struct btrfs_fs_info *fs_info = root->fs_info;
4044
4045 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4046 !is_fstree(root->root_key.objectid))
4047 return;
4048
4049 /* TODO: Update trace point to handle such free */
4050 trace_qgroup_meta_free_all_pertrans(root);
4051 /* Special value -1 means to free all reserved space */
4052 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
4053 BTRFS_QGROUP_RSV_META_PERTRANS);
4054}
4055
4056void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
4057 enum btrfs_qgroup_rsv_type type)
4058{
4059 struct btrfs_fs_info *fs_info = root->fs_info;
4060
4061 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4062 !is_fstree(root->root_key.objectid))
4063 return;
4064
4065 /*
4066 * reservation for META_PREALLOC can happen before quota is enabled,
4067 * which can lead to underflow.
4068 * Here ensure we will only free what we really have reserved.
4069 */
4070 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
4071 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4072 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
4073 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
4074 num_bytes, type);
4075}
4076
4077static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
4078 int num_bytes)
4079{
4080 struct btrfs_qgroup *qgroup;
4081 struct ulist_node *unode;
4082 struct ulist_iterator uiter;
4083 int ret = 0;
4084
4085 if (num_bytes == 0)
4086 return;
4087 if (!fs_info->quota_root)
4088 return;
4089
4090 spin_lock(&fs_info->qgroup_lock);
4091 qgroup = find_qgroup_rb(fs_info, ref_root);
4092 if (!qgroup)
4093 goto out;
4094 ulist_reinit(fs_info->qgroup_ulist);
4095 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
4096 qgroup_to_aux(qgroup), GFP_ATOMIC);
4097 if (ret < 0)
4098 goto out;
4099 ULIST_ITER_INIT(&uiter);
4100 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
4101 struct btrfs_qgroup *qg;
4102 struct btrfs_qgroup_list *glist;
4103
4104 qg = unode_aux_to_qgroup(unode);
4105
4106 qgroup_rsv_release(fs_info, qg, num_bytes,
4107 BTRFS_QGROUP_RSV_META_PREALLOC);
4108 qgroup_rsv_add(fs_info, qg, num_bytes,
4109 BTRFS_QGROUP_RSV_META_PERTRANS);
4110 list_for_each_entry(glist, &qg->groups, next_group) {
4111 ret = ulist_add(fs_info->qgroup_ulist,
4112 glist->group->qgroupid,
4113 qgroup_to_aux(glist->group), GFP_ATOMIC);
4114 if (ret < 0)
4115 goto out;
4116 }
4117 }
4118out:
4119 spin_unlock(&fs_info->qgroup_lock);
4120}
4121
4122void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4123{
4124 struct btrfs_fs_info *fs_info = root->fs_info;
4125
4126 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4127 !is_fstree(root->root_key.objectid))
4128 return;
4129 /* Same as btrfs_qgroup_free_meta_prealloc() */
4130 num_bytes = sub_root_meta_rsv(root, num_bytes,
4131 BTRFS_QGROUP_RSV_META_PREALLOC);
4132 trace_qgroup_meta_convert(root, num_bytes);
4133 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4134}
4135
4136/*
4137 * Check qgroup reserved space leaking, normally at destroy inode
4138 * time
4139 */
4140void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4141{
4142 struct extent_changeset changeset;
4143 struct ulist_node *unode;
4144 struct ulist_iterator iter;
4145 int ret;
4146
4147 extent_changeset_init(&changeset);
4148 ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4149 EXTENT_QGROUP_RESERVED, &changeset);
4150
4151 WARN_ON(ret < 0);
4152 if (WARN_ON(changeset.bytes_changed)) {
4153 ULIST_ITER_INIT(&iter);
4154 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4155 btrfs_warn(inode->root->fs_info,
4156 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4157 btrfs_ino(inode), unode->val, unode->aux);
4158 }
4159 btrfs_qgroup_free_refroot(inode->root->fs_info,
4160 inode->root->root_key.objectid,
4161 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4162
4163 }
4164 extent_changeset_release(&changeset);
4165}
4166
4167void btrfs_qgroup_init_swapped_blocks(
4168 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4169{
4170 int i;
4171
4172 spin_lock_init(&swapped_blocks->lock);
4173 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4174 swapped_blocks->blocks[i] = RB_ROOT;
4175 swapped_blocks->swapped = false;
4176}
4177
4178/*
4179 * Delete all swapped blocks record of @root.
4180 * Every record here means we skipped a full subtree scan for qgroup.
4181 *
4182 * Gets called when committing one transaction.
4183 */
4184void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4185{
4186 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4187 int i;
4188
4189 swapped_blocks = &root->swapped_blocks;
4190
4191 spin_lock(&swapped_blocks->lock);
4192 if (!swapped_blocks->swapped)
4193 goto out;
4194 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4195 struct rb_root *cur_root = &swapped_blocks->blocks[i];
4196 struct btrfs_qgroup_swapped_block *entry;
4197 struct btrfs_qgroup_swapped_block *next;
4198
4199 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4200 node)
4201 kfree(entry);
4202 swapped_blocks->blocks[i] = RB_ROOT;
4203 }
4204 swapped_blocks->swapped = false;
4205out:
4206 spin_unlock(&swapped_blocks->lock);
4207}
4208
4209/*
4210 * Add subtree roots record into @subvol_root.
4211 *
4212 * @subvol_root: tree root of the subvolume tree get swapped
4213 * @bg: block group under balance
4214 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4215 * @reloc_parent/slot: pointer to the subtree root in reloc tree
4216 * BOTH POINTERS ARE BEFORE TREE SWAP
4217 * @last_snapshot: last snapshot generation of the subvolume tree
4218 */
4219int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4220 struct btrfs_root *subvol_root,
4221 struct btrfs_block_group *bg,
4222 struct extent_buffer *subvol_parent, int subvol_slot,
4223 struct extent_buffer *reloc_parent, int reloc_slot,
4224 u64 last_snapshot)
4225{
4226 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4227 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4228 struct btrfs_qgroup_swapped_block *block;
4229 struct rb_node **cur;
4230 struct rb_node *parent = NULL;
4231 int level = btrfs_header_level(subvol_parent) - 1;
4232 int ret = 0;
4233
4234 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4235 return 0;
4236
4237 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4238 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4239 btrfs_err_rl(fs_info,
4240 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4241 __func__,
4242 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4243 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4244 return -EUCLEAN;
4245 }
4246
4247 block = kmalloc(sizeof(*block), GFP_NOFS);
4248 if (!block) {
4249 ret = -ENOMEM;
4250 goto out;
4251 }
4252
4253 /*
4254 * @reloc_parent/slot is still before swap, while @block is going to
4255 * record the bytenr after swap, so we do the swap here.
4256 */
4257 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4258 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4259 reloc_slot);
4260 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4261 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4262 subvol_slot);
4263 block->last_snapshot = last_snapshot;
4264 block->level = level;
4265
4266 /*
4267 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4268 * no one else can modify tree blocks thus we qgroup will not change
4269 * no matter the value of trace_leaf.
4270 */
4271 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4272 block->trace_leaf = true;
4273 else
4274 block->trace_leaf = false;
4275 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4276
4277 /* Insert @block into @blocks */
4278 spin_lock(&blocks->lock);
4279 cur = &blocks->blocks[level].rb_node;
4280 while (*cur) {
4281 struct btrfs_qgroup_swapped_block *entry;
4282
4283 parent = *cur;
4284 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4285 node);
4286
4287 if (entry->subvol_bytenr < block->subvol_bytenr) {
4288 cur = &(*cur)->rb_left;
4289 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4290 cur = &(*cur)->rb_right;
4291 } else {
4292 if (entry->subvol_generation !=
4293 block->subvol_generation ||
4294 entry->reloc_bytenr != block->reloc_bytenr ||
4295 entry->reloc_generation !=
4296 block->reloc_generation) {
4297 /*
4298 * Duplicated but mismatch entry found.
4299 * Shouldn't happen.
4300 *
4301 * Marking qgroup inconsistent should be enough
4302 * for end users.
4303 */
4304 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4305 ret = -EEXIST;
4306 }
4307 kfree(block);
4308 goto out_unlock;
4309 }
4310 }
4311 rb_link_node(&block->node, parent, cur);
4312 rb_insert_color(&block->node, &blocks->blocks[level]);
4313 blocks->swapped = true;
4314out_unlock:
4315 spin_unlock(&blocks->lock);
4316out:
4317 if (ret < 0)
4318 qgroup_mark_inconsistent(fs_info);
4319 return ret;
4320}
4321
4322/*
4323 * Check if the tree block is a subtree root, and if so do the needed
4324 * delayed subtree trace for qgroup.
4325 *
4326 * This is called during btrfs_cow_block().
4327 */
4328int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4329 struct btrfs_root *root,
4330 struct extent_buffer *subvol_eb)
4331{
4332 struct btrfs_fs_info *fs_info = root->fs_info;
4333 struct btrfs_tree_parent_check check = { 0 };
4334 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4335 struct btrfs_qgroup_swapped_block *block;
4336 struct extent_buffer *reloc_eb = NULL;
4337 struct rb_node *node;
4338 bool found = false;
4339 bool swapped = false;
4340 int level = btrfs_header_level(subvol_eb);
4341 int ret = 0;
4342 int i;
4343
4344 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4345 return 0;
4346 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4347 return 0;
4348
4349 spin_lock(&blocks->lock);
4350 if (!blocks->swapped) {
4351 spin_unlock(&blocks->lock);
4352 return 0;
4353 }
4354 node = blocks->blocks[level].rb_node;
4355
4356 while (node) {
4357 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4358 if (block->subvol_bytenr < subvol_eb->start) {
4359 node = node->rb_left;
4360 } else if (block->subvol_bytenr > subvol_eb->start) {
4361 node = node->rb_right;
4362 } else {
4363 found = true;
4364 break;
4365 }
4366 }
4367 if (!found) {
4368 spin_unlock(&blocks->lock);
4369 goto out;
4370 }
4371 /* Found one, remove it from @blocks first and update blocks->swapped */
4372 rb_erase(&block->node, &blocks->blocks[level]);
4373 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4374 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4375 swapped = true;
4376 break;
4377 }
4378 }
4379 blocks->swapped = swapped;
4380 spin_unlock(&blocks->lock);
4381
4382 check.level = block->level;
4383 check.transid = block->reloc_generation;
4384 check.has_first_key = true;
4385 memcpy(&check.first_key, &block->first_key, sizeof(check.first_key));
4386
4387 /* Read out reloc subtree root */
4388 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, &check);
4389 if (IS_ERR(reloc_eb)) {
4390 ret = PTR_ERR(reloc_eb);
4391 reloc_eb = NULL;
4392 goto free_out;
4393 }
4394 if (!extent_buffer_uptodate(reloc_eb)) {
4395 ret = -EIO;
4396 goto free_out;
4397 }
4398
4399 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4400 block->last_snapshot, block->trace_leaf);
4401free_out:
4402 kfree(block);
4403 free_extent_buffer(reloc_eb);
4404out:
4405 if (ret < 0) {
4406 btrfs_err_rl(fs_info,
4407 "failed to account subtree at bytenr %llu: %d",
4408 subvol_eb->start, ret);
4409 qgroup_mark_inconsistent(fs_info);
4410 }
4411 return ret;
4412}
4413
4414void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4415{
4416 struct btrfs_qgroup_extent_record *entry;
4417 struct btrfs_qgroup_extent_record *next;
4418 struct rb_root *root;
4419
4420 root = &trans->delayed_refs.dirty_extent_root;
4421 rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4422 ulist_free(entry->old_roots);
4423 kfree(entry);
4424 }
4425}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2011 STRATO. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/pagemap.h>
8#include <linux/writeback.h>
9#include <linux/blkdev.h>
10#include <linux/rbtree.h>
11#include <linux/slab.h>
12#include <linux/workqueue.h>
13#include <linux/btrfs.h>
14#include <linux/sizes.h>
15
16#include "ctree.h"
17#include "transaction.h"
18#include "disk-io.h"
19#include "locking.h"
20#include "ulist.h"
21#include "backref.h"
22#include "extent_io.h"
23#include "qgroup.h"
24#include "block-group.h"
25
26/* TODO XXX FIXME
27 * - subvol delete -> delete when ref goes to 0? delete limits also?
28 * - reorganize keys
29 * - compressed
30 * - sync
31 * - copy also limits on subvol creation
32 * - limit
33 * - caches for ulists
34 * - performance benchmarks
35 * - check all ioctl parameters
36 */
37
38/*
39 * Helpers to access qgroup reservation
40 *
41 * Callers should ensure the lock context and type are valid
42 */
43
44static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45{
46 u64 ret = 0;
47 int i;
48
49 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 ret += qgroup->rsv.values[i];
51
52 return ret;
53}
54
55#ifdef CONFIG_BTRFS_DEBUG
56static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57{
58 if (type == BTRFS_QGROUP_RSV_DATA)
59 return "data";
60 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 return "meta_pertrans";
62 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 return "meta_prealloc";
64 return NULL;
65}
66#endif
67
68static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 struct btrfs_qgroup *qgroup, u64 num_bytes,
70 enum btrfs_qgroup_rsv_type type)
71{
72 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 qgroup->rsv.values[type] += num_bytes;
74}
75
76static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 struct btrfs_qgroup *qgroup, u64 num_bytes,
78 enum btrfs_qgroup_rsv_type type)
79{
80 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 if (qgroup->rsv.values[type] >= num_bytes) {
82 qgroup->rsv.values[type] -= num_bytes;
83 return;
84 }
85#ifdef CONFIG_BTRFS_DEBUG
86 WARN_RATELIMIT(1,
87 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 qgroup->qgroupid, qgroup_rsv_type_str(type),
89 qgroup->rsv.values[type], num_bytes);
90#endif
91 qgroup->rsv.values[type] = 0;
92}
93
94static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 struct btrfs_qgroup *dest,
96 struct btrfs_qgroup *src)
97{
98 int i;
99
100 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
102}
103
104static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 struct btrfs_qgroup *dest,
106 struct btrfs_qgroup *src)
107{
108 int i;
109
110 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
112}
113
114static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115 int mod)
116{
117 if (qg->old_refcnt < seq)
118 qg->old_refcnt = seq;
119 qg->old_refcnt += mod;
120}
121
122static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123 int mod)
124{
125 if (qg->new_refcnt < seq)
126 qg->new_refcnt = seq;
127 qg->new_refcnt += mod;
128}
129
130static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131{
132 if (qg->old_refcnt < seq)
133 return 0;
134 return qg->old_refcnt - seq;
135}
136
137static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138{
139 if (qg->new_refcnt < seq)
140 return 0;
141 return qg->new_refcnt - seq;
142}
143
144/*
145 * glue structure to represent the relations between qgroups.
146 */
147struct btrfs_qgroup_list {
148 struct list_head next_group;
149 struct list_head next_member;
150 struct btrfs_qgroup *group;
151 struct btrfs_qgroup *member;
152};
153
154static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155{
156 return (u64)(uintptr_t)qg;
157}
158
159static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160{
161 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162}
163
164static int
165qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 int init_flags);
167static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168
169/* must be called with qgroup_ioctl_lock held */
170static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171 u64 qgroupid)
172{
173 struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 struct btrfs_qgroup *qgroup;
175
176 while (n) {
177 qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 if (qgroup->qgroupid < qgroupid)
179 n = n->rb_left;
180 else if (qgroup->qgroupid > qgroupid)
181 n = n->rb_right;
182 else
183 return qgroup;
184 }
185 return NULL;
186}
187
188/* must be called with qgroup_lock held */
189static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190 u64 qgroupid)
191{
192 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 struct rb_node *parent = NULL;
194 struct btrfs_qgroup *qgroup;
195
196 while (*p) {
197 parent = *p;
198 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199
200 if (qgroup->qgroupid < qgroupid)
201 p = &(*p)->rb_left;
202 else if (qgroup->qgroupid > qgroupid)
203 p = &(*p)->rb_right;
204 else
205 return qgroup;
206 }
207
208 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 if (!qgroup)
210 return ERR_PTR(-ENOMEM);
211
212 qgroup->qgroupid = qgroupid;
213 INIT_LIST_HEAD(&qgroup->groups);
214 INIT_LIST_HEAD(&qgroup->members);
215 INIT_LIST_HEAD(&qgroup->dirty);
216
217 rb_link_node(&qgroup->node, parent, p);
218 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219
220 return qgroup;
221}
222
223static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
224{
225 struct btrfs_qgroup_list *list;
226
227 list_del(&qgroup->dirty);
228 while (!list_empty(&qgroup->groups)) {
229 list = list_first_entry(&qgroup->groups,
230 struct btrfs_qgroup_list, next_group);
231 list_del(&list->next_group);
232 list_del(&list->next_member);
233 kfree(list);
234 }
235
236 while (!list_empty(&qgroup->members)) {
237 list = list_first_entry(&qgroup->members,
238 struct btrfs_qgroup_list, next_member);
239 list_del(&list->next_group);
240 list_del(&list->next_member);
241 kfree(list);
242 }
243 kfree(qgroup);
244}
245
246/* must be called with qgroup_lock held */
247static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248{
249 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250
251 if (!qgroup)
252 return -ENOENT;
253
254 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 __del_qgroup_rb(qgroup);
256 return 0;
257}
258
259/* must be called with qgroup_lock held */
260static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 u64 memberid, u64 parentid)
262{
263 struct btrfs_qgroup *member;
264 struct btrfs_qgroup *parent;
265 struct btrfs_qgroup_list *list;
266
267 member = find_qgroup_rb(fs_info, memberid);
268 parent = find_qgroup_rb(fs_info, parentid);
269 if (!member || !parent)
270 return -ENOENT;
271
272 list = kzalloc(sizeof(*list), GFP_ATOMIC);
273 if (!list)
274 return -ENOMEM;
275
276 list->group = parent;
277 list->member = member;
278 list_add_tail(&list->next_group, &member->groups);
279 list_add_tail(&list->next_member, &parent->members);
280
281 return 0;
282}
283
284/* must be called with qgroup_lock held */
285static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 u64 memberid, u64 parentid)
287{
288 struct btrfs_qgroup *member;
289 struct btrfs_qgroup *parent;
290 struct btrfs_qgroup_list *list;
291
292 member = find_qgroup_rb(fs_info, memberid);
293 parent = find_qgroup_rb(fs_info, parentid);
294 if (!member || !parent)
295 return -ENOENT;
296
297 list_for_each_entry(list, &member->groups, next_group) {
298 if (list->group == parent) {
299 list_del(&list->next_group);
300 list_del(&list->next_member);
301 kfree(list);
302 return 0;
303 }
304 }
305 return -ENOENT;
306}
307
308#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310 u64 rfer, u64 excl)
311{
312 struct btrfs_qgroup *qgroup;
313
314 qgroup = find_qgroup_rb(fs_info, qgroupid);
315 if (!qgroup)
316 return -EINVAL;
317 if (qgroup->rfer != rfer || qgroup->excl != excl)
318 return -EINVAL;
319 return 0;
320}
321#endif
322
323/*
324 * The full config is read in one go, only called from open_ctree()
325 * It doesn't use any locking, as at this point we're still single-threaded
326 */
327int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328{
329 struct btrfs_key key;
330 struct btrfs_key found_key;
331 struct btrfs_root *quota_root = fs_info->quota_root;
332 struct btrfs_path *path = NULL;
333 struct extent_buffer *l;
334 int slot;
335 int ret = 0;
336 u64 flags = 0;
337 u64 rescan_progress = 0;
338
339 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
340 return 0;
341
342 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 if (!fs_info->qgroup_ulist) {
344 ret = -ENOMEM;
345 goto out;
346 }
347
348 path = btrfs_alloc_path();
349 if (!path) {
350 ret = -ENOMEM;
351 goto out;
352 }
353
354 /* default this to quota off, in case no status key is found */
355 fs_info->qgroup_flags = 0;
356
357 /*
358 * pass 1: read status, all qgroup infos and limits
359 */
360 key.objectid = 0;
361 key.type = 0;
362 key.offset = 0;
363 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
364 if (ret)
365 goto out;
366
367 while (1) {
368 struct btrfs_qgroup *qgroup;
369
370 slot = path->slots[0];
371 l = path->nodes[0];
372 btrfs_item_key_to_cpu(l, &found_key, slot);
373
374 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 struct btrfs_qgroup_status_item *ptr;
376
377 ptr = btrfs_item_ptr(l, slot,
378 struct btrfs_qgroup_status_item);
379
380 if (btrfs_qgroup_status_version(l, ptr) !=
381 BTRFS_QGROUP_STATUS_VERSION) {
382 btrfs_err(fs_info,
383 "old qgroup version, quota disabled");
384 goto out;
385 }
386 if (btrfs_qgroup_status_generation(l, ptr) !=
387 fs_info->generation) {
388 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 btrfs_err(fs_info,
390 "qgroup generation mismatch, marked as inconsistent");
391 }
392 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 ptr);
394 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
395 goto next1;
396 }
397
398 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
400 goto next1;
401
402 qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 btrfs_err(fs_info, "inconsistent qgroup config");
406 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407 }
408 if (!qgroup) {
409 qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 if (IS_ERR(qgroup)) {
411 ret = PTR_ERR(qgroup);
412 goto out;
413 }
414 }
415 switch (found_key.type) {
416 case BTRFS_QGROUP_INFO_KEY: {
417 struct btrfs_qgroup_info_item *ptr;
418
419 ptr = btrfs_item_ptr(l, slot,
420 struct btrfs_qgroup_info_item);
421 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 /* generation currently unused */
426 break;
427 }
428 case BTRFS_QGROUP_LIMIT_KEY: {
429 struct btrfs_qgroup_limit_item *ptr;
430
431 ptr = btrfs_item_ptr(l, slot,
432 struct btrfs_qgroup_limit_item);
433 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438 break;
439 }
440 }
441next1:
442 ret = btrfs_next_item(quota_root, path);
443 if (ret < 0)
444 goto out;
445 if (ret)
446 break;
447 }
448 btrfs_release_path(path);
449
450 /*
451 * pass 2: read all qgroup relations
452 */
453 key.objectid = 0;
454 key.type = BTRFS_QGROUP_RELATION_KEY;
455 key.offset = 0;
456 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457 if (ret)
458 goto out;
459 while (1) {
460 slot = path->slots[0];
461 l = path->nodes[0];
462 btrfs_item_key_to_cpu(l, &found_key, slot);
463
464 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465 goto next2;
466
467 if (found_key.objectid > found_key.offset) {
468 /* parent <- member, not needed to build config */
469 /* FIXME should we omit the key completely? */
470 goto next2;
471 }
472
473 ret = add_relation_rb(fs_info, found_key.objectid,
474 found_key.offset);
475 if (ret == -ENOENT) {
476 btrfs_warn(fs_info,
477 "orphan qgroup relation 0x%llx->0x%llx",
478 found_key.objectid, found_key.offset);
479 ret = 0; /* ignore the error */
480 }
481 if (ret)
482 goto out;
483next2:
484 ret = btrfs_next_item(quota_root, path);
485 if (ret < 0)
486 goto out;
487 if (ret)
488 break;
489 }
490out:
491 fs_info->qgroup_flags |= flags;
492 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495 ret >= 0)
496 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 btrfs_free_path(path);
498
499 if (ret < 0) {
500 ulist_free(fs_info->qgroup_ulist);
501 fs_info->qgroup_ulist = NULL;
502 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
503 }
504
505 return ret < 0 ? ret : 0;
506}
507
508/*
509 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510 * first two are in single-threaded paths.And for the third one, we have set
511 * quota_root to be null with qgroup_lock held before, so it is safe to clean
512 * up the in-memory structures without qgroup_lock held.
513 */
514void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515{
516 struct rb_node *n;
517 struct btrfs_qgroup *qgroup;
518
519 while ((n = rb_first(&fs_info->qgroup_tree))) {
520 qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 rb_erase(n, &fs_info->qgroup_tree);
522 __del_qgroup_rb(qgroup);
523 }
524 /*
525 * We call btrfs_free_qgroup_config() when unmounting
526 * filesystem and disabling quota, so we set qgroup_ulist
527 * to be null here to avoid double free.
528 */
529 ulist_free(fs_info->qgroup_ulist);
530 fs_info->qgroup_ulist = NULL;
531}
532
533static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 u64 dst)
535{
536 int ret;
537 struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 struct btrfs_path *path;
539 struct btrfs_key key;
540
541 path = btrfs_alloc_path();
542 if (!path)
543 return -ENOMEM;
544
545 key.objectid = src;
546 key.type = BTRFS_QGROUP_RELATION_KEY;
547 key.offset = dst;
548
549 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550
551 btrfs_mark_buffer_dirty(path->nodes[0]);
552
553 btrfs_free_path(path);
554 return ret;
555}
556
557static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 u64 dst)
559{
560 int ret;
561 struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 struct btrfs_path *path;
563 struct btrfs_key key;
564
565 path = btrfs_alloc_path();
566 if (!path)
567 return -ENOMEM;
568
569 key.objectid = src;
570 key.type = BTRFS_QGROUP_RELATION_KEY;
571 key.offset = dst;
572
573 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574 if (ret < 0)
575 goto out;
576
577 if (ret > 0) {
578 ret = -ENOENT;
579 goto out;
580 }
581
582 ret = btrfs_del_item(trans, quota_root, path);
583out:
584 btrfs_free_path(path);
585 return ret;
586}
587
588static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 struct btrfs_root *quota_root, u64 qgroupid)
590{
591 int ret;
592 struct btrfs_path *path;
593 struct btrfs_qgroup_info_item *qgroup_info;
594 struct btrfs_qgroup_limit_item *qgroup_limit;
595 struct extent_buffer *leaf;
596 struct btrfs_key key;
597
598 if (btrfs_is_testing(quota_root->fs_info))
599 return 0;
600
601 path = btrfs_alloc_path();
602 if (!path)
603 return -ENOMEM;
604
605 key.objectid = 0;
606 key.type = BTRFS_QGROUP_INFO_KEY;
607 key.offset = qgroupid;
608
609 /*
610 * Avoid a transaction abort by catching -EEXIST here. In that
611 * case, we proceed by re-initializing the existing structure
612 * on disk.
613 */
614
615 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 sizeof(*qgroup_info));
617 if (ret && ret != -EEXIST)
618 goto out;
619
620 leaf = path->nodes[0];
621 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 struct btrfs_qgroup_info_item);
623 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628
629 btrfs_mark_buffer_dirty(leaf);
630
631 btrfs_release_path(path);
632
633 key.type = BTRFS_QGROUP_LIMIT_KEY;
634 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 sizeof(*qgroup_limit));
636 if (ret && ret != -EEXIST)
637 goto out;
638
639 leaf = path->nodes[0];
640 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_qgroup_limit_item);
642 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647
648 btrfs_mark_buffer_dirty(leaf);
649
650 ret = 0;
651out:
652 btrfs_free_path(path);
653 return ret;
654}
655
656static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
657{
658 int ret;
659 struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 struct btrfs_path *path;
661 struct btrfs_key key;
662
663 path = btrfs_alloc_path();
664 if (!path)
665 return -ENOMEM;
666
667 key.objectid = 0;
668 key.type = BTRFS_QGROUP_INFO_KEY;
669 key.offset = qgroupid;
670 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 if (ret < 0)
672 goto out;
673
674 if (ret > 0) {
675 ret = -ENOENT;
676 goto out;
677 }
678
679 ret = btrfs_del_item(trans, quota_root, path);
680 if (ret)
681 goto out;
682
683 btrfs_release_path(path);
684
685 key.type = BTRFS_QGROUP_LIMIT_KEY;
686 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687 if (ret < 0)
688 goto out;
689
690 if (ret > 0) {
691 ret = -ENOENT;
692 goto out;
693 }
694
695 ret = btrfs_del_item(trans, quota_root, path);
696
697out:
698 btrfs_free_path(path);
699 return ret;
700}
701
702static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 struct btrfs_qgroup *qgroup)
704{
705 struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 struct btrfs_path *path;
707 struct btrfs_key key;
708 struct extent_buffer *l;
709 struct btrfs_qgroup_limit_item *qgroup_limit;
710 int ret;
711 int slot;
712
713 key.objectid = 0;
714 key.type = BTRFS_QGROUP_LIMIT_KEY;
715 key.offset = qgroup->qgroupid;
716
717 path = btrfs_alloc_path();
718 if (!path)
719 return -ENOMEM;
720
721 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
722 if (ret > 0)
723 ret = -ENOENT;
724
725 if (ret)
726 goto out;
727
728 l = path->nodes[0];
729 slot = path->slots[0];
730 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736
737 btrfs_mark_buffer_dirty(l);
738
739out:
740 btrfs_free_path(path);
741 return ret;
742}
743
744static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 struct btrfs_qgroup *qgroup)
746{
747 struct btrfs_fs_info *fs_info = trans->fs_info;
748 struct btrfs_root *quota_root = fs_info->quota_root;
749 struct btrfs_path *path;
750 struct btrfs_key key;
751 struct extent_buffer *l;
752 struct btrfs_qgroup_info_item *qgroup_info;
753 int ret;
754 int slot;
755
756 if (btrfs_is_testing(fs_info))
757 return 0;
758
759 key.objectid = 0;
760 key.type = BTRFS_QGROUP_INFO_KEY;
761 key.offset = qgroup->qgroupid;
762
763 path = btrfs_alloc_path();
764 if (!path)
765 return -ENOMEM;
766
767 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
768 if (ret > 0)
769 ret = -ENOENT;
770
771 if (ret)
772 goto out;
773
774 l = path->nodes[0];
775 slot = path->slots[0];
776 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782
783 btrfs_mark_buffer_dirty(l);
784
785out:
786 btrfs_free_path(path);
787 return ret;
788}
789
790static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
791{
792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *quota_root = fs_info->quota_root;
794 struct btrfs_path *path;
795 struct btrfs_key key;
796 struct extent_buffer *l;
797 struct btrfs_qgroup_status_item *ptr;
798 int ret;
799 int slot;
800
801 key.objectid = 0;
802 key.type = BTRFS_QGROUP_STATUS_KEY;
803 key.offset = 0;
804
805 path = btrfs_alloc_path();
806 if (!path)
807 return -ENOMEM;
808
809 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
810 if (ret > 0)
811 ret = -ENOENT;
812
813 if (ret)
814 goto out;
815
816 l = path->nodes[0];
817 slot = path->slots[0];
818 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 btrfs_set_qgroup_status_rescan(l, ptr,
822 fs_info->qgroup_rescan_progress.objectid);
823
824 btrfs_mark_buffer_dirty(l);
825
826out:
827 btrfs_free_path(path);
828 return ret;
829}
830
831/*
832 * called with qgroup_lock held
833 */
834static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root)
836{
837 struct btrfs_path *path;
838 struct btrfs_key key;
839 struct extent_buffer *leaf = NULL;
840 int ret;
841 int nr = 0;
842
843 path = btrfs_alloc_path();
844 if (!path)
845 return -ENOMEM;
846
847 path->leave_spinning = 1;
848
849 key.objectid = 0;
850 key.offset = 0;
851 key.type = 0;
852
853 while (1) {
854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855 if (ret < 0)
856 goto out;
857 leaf = path->nodes[0];
858 nr = btrfs_header_nritems(leaf);
859 if (!nr)
860 break;
861 /*
862 * delete the leaf one by one
863 * since the whole tree is going
864 * to be deleted.
865 */
866 path->slots[0] = 0;
867 ret = btrfs_del_items(trans, root, path, 0, nr);
868 if (ret)
869 goto out;
870
871 btrfs_release_path(path);
872 }
873 ret = 0;
874out:
875 btrfs_free_path(path);
876 return ret;
877}
878
879int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
880{
881 struct btrfs_root *quota_root;
882 struct btrfs_root *tree_root = fs_info->tree_root;
883 struct btrfs_path *path = NULL;
884 struct btrfs_qgroup_status_item *ptr;
885 struct extent_buffer *leaf;
886 struct btrfs_key key;
887 struct btrfs_key found_key;
888 struct btrfs_qgroup *qgroup = NULL;
889 struct btrfs_trans_handle *trans = NULL;
890 int ret = 0;
891 int slot;
892
893 mutex_lock(&fs_info->qgroup_ioctl_lock);
894 if (fs_info->quota_root)
895 goto out;
896
897 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 if (!fs_info->qgroup_ulist) {
899 ret = -ENOMEM;
900 goto out;
901 }
902
903 /*
904 * 1 for quota root item
905 * 1 for BTRFS_QGROUP_STATUS item
906 *
907 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
908 * per subvolume. However those are not currently reserved since it
909 * would be a lot of overkill.
910 */
911 trans = btrfs_start_transaction(tree_root, 2);
912 if (IS_ERR(trans)) {
913 ret = PTR_ERR(trans);
914 trans = NULL;
915 goto out;
916 }
917
918 /*
919 * initially create the quota tree
920 */
921 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
922 if (IS_ERR(quota_root)) {
923 ret = PTR_ERR(quota_root);
924 btrfs_abort_transaction(trans, ret);
925 goto out;
926 }
927
928 path = btrfs_alloc_path();
929 if (!path) {
930 ret = -ENOMEM;
931 btrfs_abort_transaction(trans, ret);
932 goto out_free_root;
933 }
934
935 key.objectid = 0;
936 key.type = BTRFS_QGROUP_STATUS_KEY;
937 key.offset = 0;
938
939 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
940 sizeof(*ptr));
941 if (ret) {
942 btrfs_abort_transaction(trans, ret);
943 goto out_free_path;
944 }
945
946 leaf = path->nodes[0];
947 ptr = btrfs_item_ptr(leaf, path->slots[0],
948 struct btrfs_qgroup_status_item);
949 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
950 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
951 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
952 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
953 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
954 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
955
956 btrfs_mark_buffer_dirty(leaf);
957
958 key.objectid = 0;
959 key.type = BTRFS_ROOT_REF_KEY;
960 key.offset = 0;
961
962 btrfs_release_path(path);
963 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
964 if (ret > 0)
965 goto out_add_root;
966 if (ret < 0) {
967 btrfs_abort_transaction(trans, ret);
968 goto out_free_path;
969 }
970
971 while (1) {
972 slot = path->slots[0];
973 leaf = path->nodes[0];
974 btrfs_item_key_to_cpu(leaf, &found_key, slot);
975
976 if (found_key.type == BTRFS_ROOT_REF_KEY) {
977 ret = add_qgroup_item(trans, quota_root,
978 found_key.offset);
979 if (ret) {
980 btrfs_abort_transaction(trans, ret);
981 goto out_free_path;
982 }
983
984 qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 if (IS_ERR(qgroup)) {
986 ret = PTR_ERR(qgroup);
987 btrfs_abort_transaction(trans, ret);
988 goto out_free_path;
989 }
990 }
991 ret = btrfs_next_item(tree_root, path);
992 if (ret < 0) {
993 btrfs_abort_transaction(trans, ret);
994 goto out_free_path;
995 }
996 if (ret)
997 break;
998 }
999
1000out_add_root:
1001 btrfs_release_path(path);
1002 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1003 if (ret) {
1004 btrfs_abort_transaction(trans, ret);
1005 goto out_free_path;
1006 }
1007
1008 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 if (IS_ERR(qgroup)) {
1010 ret = PTR_ERR(qgroup);
1011 btrfs_abort_transaction(trans, ret);
1012 goto out_free_path;
1013 }
1014
1015 ret = btrfs_commit_transaction(trans);
1016 trans = NULL;
1017 if (ret)
1018 goto out_free_path;
1019
1020 /*
1021 * Set quota enabled flag after committing the transaction, to avoid
1022 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1023 * creation.
1024 */
1025 spin_lock(&fs_info->qgroup_lock);
1026 fs_info->quota_root = quota_root;
1027 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1028 spin_unlock(&fs_info->qgroup_lock);
1029
1030 ret = qgroup_rescan_init(fs_info, 0, 1);
1031 if (!ret) {
1032 qgroup_rescan_zero_tracking(fs_info);
1033 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1034 &fs_info->qgroup_rescan_work);
1035 }
1036
1037out_free_path:
1038 btrfs_free_path(path);
1039out_free_root:
1040 if (ret) {
1041 free_extent_buffer(quota_root->node);
1042 free_extent_buffer(quota_root->commit_root);
1043 kfree(quota_root);
1044 }
1045out:
1046 if (ret) {
1047 ulist_free(fs_info->qgroup_ulist);
1048 fs_info->qgroup_ulist = NULL;
1049 if (trans)
1050 btrfs_end_transaction(trans);
1051 }
1052 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1053 return ret;
1054}
1055
1056int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1057{
1058 struct btrfs_root *quota_root;
1059 struct btrfs_trans_handle *trans = NULL;
1060 int ret = 0;
1061
1062 mutex_lock(&fs_info->qgroup_ioctl_lock);
1063 if (!fs_info->quota_root)
1064 goto out;
1065
1066 /*
1067 * 1 For the root item
1068 *
1069 * We should also reserve enough items for the quota tree deletion in
1070 * btrfs_clean_quota_tree but this is not done.
1071 */
1072 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1073 if (IS_ERR(trans)) {
1074 ret = PTR_ERR(trans);
1075 goto out;
1076 }
1077
1078 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1079 btrfs_qgroup_wait_for_completion(fs_info, false);
1080 spin_lock(&fs_info->qgroup_lock);
1081 quota_root = fs_info->quota_root;
1082 fs_info->quota_root = NULL;
1083 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1084 spin_unlock(&fs_info->qgroup_lock);
1085
1086 btrfs_free_qgroup_config(fs_info);
1087
1088 ret = btrfs_clean_quota_tree(trans, quota_root);
1089 if (ret) {
1090 btrfs_abort_transaction(trans, ret);
1091 goto end_trans;
1092 }
1093
1094 ret = btrfs_del_root(trans, "a_root->root_key);
1095 if (ret) {
1096 btrfs_abort_transaction(trans, ret);
1097 goto end_trans;
1098 }
1099
1100 list_del("a_root->dirty_list);
1101
1102 btrfs_tree_lock(quota_root->node);
1103 btrfs_clean_tree_block(quota_root->node);
1104 btrfs_tree_unlock(quota_root->node);
1105 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1106
1107 free_extent_buffer(quota_root->node);
1108 free_extent_buffer(quota_root->commit_root);
1109 kfree(quota_root);
1110
1111end_trans:
1112 ret = btrfs_end_transaction(trans);
1113out:
1114 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1115 return ret;
1116}
1117
1118static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1119 struct btrfs_qgroup *qgroup)
1120{
1121 if (list_empty(&qgroup->dirty))
1122 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1123}
1124
1125/*
1126 * The easy accounting, we're updating qgroup relationship whose child qgroup
1127 * only has exclusive extents.
1128 *
1129 * In this case, all exclusive extents will also be exclusive for parent, so
1130 * excl/rfer just get added/removed.
1131 *
1132 * So is qgroup reservation space, which should also be added/removed to
1133 * parent.
1134 * Or when child tries to release reservation space, parent will underflow its
1135 * reservation (for relationship adding case).
1136 *
1137 * Caller should hold fs_info->qgroup_lock.
1138 */
1139static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1140 struct ulist *tmp, u64 ref_root,
1141 struct btrfs_qgroup *src, int sign)
1142{
1143 struct btrfs_qgroup *qgroup;
1144 struct btrfs_qgroup_list *glist;
1145 struct ulist_node *unode;
1146 struct ulist_iterator uiter;
1147 u64 num_bytes = src->excl;
1148 int ret = 0;
1149
1150 qgroup = find_qgroup_rb(fs_info, ref_root);
1151 if (!qgroup)
1152 goto out;
1153
1154 qgroup->rfer += sign * num_bytes;
1155 qgroup->rfer_cmpr += sign * num_bytes;
1156
1157 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1158 qgroup->excl += sign * num_bytes;
1159 qgroup->excl_cmpr += sign * num_bytes;
1160
1161 if (sign > 0)
1162 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1163 else
1164 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1165
1166 qgroup_dirty(fs_info, qgroup);
1167
1168 /* Get all of the parent groups that contain this qgroup */
1169 list_for_each_entry(glist, &qgroup->groups, next_group) {
1170 ret = ulist_add(tmp, glist->group->qgroupid,
1171 qgroup_to_aux(glist->group), GFP_ATOMIC);
1172 if (ret < 0)
1173 goto out;
1174 }
1175
1176 /* Iterate all of the parents and adjust their reference counts */
1177 ULIST_ITER_INIT(&uiter);
1178 while ((unode = ulist_next(tmp, &uiter))) {
1179 qgroup = unode_aux_to_qgroup(unode);
1180 qgroup->rfer += sign * num_bytes;
1181 qgroup->rfer_cmpr += sign * num_bytes;
1182 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1183 qgroup->excl += sign * num_bytes;
1184 if (sign > 0)
1185 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1186 else
1187 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1188 qgroup->excl_cmpr += sign * num_bytes;
1189 qgroup_dirty(fs_info, qgroup);
1190
1191 /* Add any parents of the parents */
1192 list_for_each_entry(glist, &qgroup->groups, next_group) {
1193 ret = ulist_add(tmp, glist->group->qgroupid,
1194 qgroup_to_aux(glist->group), GFP_ATOMIC);
1195 if (ret < 0)
1196 goto out;
1197 }
1198 }
1199 ret = 0;
1200out:
1201 return ret;
1202}
1203
1204
1205/*
1206 * Quick path for updating qgroup with only excl refs.
1207 *
1208 * In that case, just update all parent will be enough.
1209 * Or we needs to do a full rescan.
1210 * Caller should also hold fs_info->qgroup_lock.
1211 *
1212 * Return 0 for quick update, return >0 for need to full rescan
1213 * and mark INCONSISTENT flag.
1214 * Return < 0 for other error.
1215 */
1216static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1217 struct ulist *tmp, u64 src, u64 dst,
1218 int sign)
1219{
1220 struct btrfs_qgroup *qgroup;
1221 int ret = 1;
1222 int err = 0;
1223
1224 qgroup = find_qgroup_rb(fs_info, src);
1225 if (!qgroup)
1226 goto out;
1227 if (qgroup->excl == qgroup->rfer) {
1228 ret = 0;
1229 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1230 qgroup, sign);
1231 if (err < 0) {
1232 ret = err;
1233 goto out;
1234 }
1235 }
1236out:
1237 if (ret)
1238 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1239 return ret;
1240}
1241
1242int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1243 u64 dst)
1244{
1245 struct btrfs_fs_info *fs_info = trans->fs_info;
1246 struct btrfs_root *quota_root;
1247 struct btrfs_qgroup *parent;
1248 struct btrfs_qgroup *member;
1249 struct btrfs_qgroup_list *list;
1250 struct ulist *tmp;
1251 int ret = 0;
1252
1253 /* Check the level of src and dst first */
1254 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1255 return -EINVAL;
1256
1257 tmp = ulist_alloc(GFP_KERNEL);
1258 if (!tmp)
1259 return -ENOMEM;
1260
1261 mutex_lock(&fs_info->qgroup_ioctl_lock);
1262 quota_root = fs_info->quota_root;
1263 if (!quota_root) {
1264 ret = -EINVAL;
1265 goto out;
1266 }
1267 member = find_qgroup_rb(fs_info, src);
1268 parent = find_qgroup_rb(fs_info, dst);
1269 if (!member || !parent) {
1270 ret = -EINVAL;
1271 goto out;
1272 }
1273
1274 /* check if such qgroup relation exist firstly */
1275 list_for_each_entry(list, &member->groups, next_group) {
1276 if (list->group == parent) {
1277 ret = -EEXIST;
1278 goto out;
1279 }
1280 }
1281
1282 ret = add_qgroup_relation_item(trans, src, dst);
1283 if (ret)
1284 goto out;
1285
1286 ret = add_qgroup_relation_item(trans, dst, src);
1287 if (ret) {
1288 del_qgroup_relation_item(trans, src, dst);
1289 goto out;
1290 }
1291
1292 spin_lock(&fs_info->qgroup_lock);
1293 ret = add_relation_rb(fs_info, src, dst);
1294 if (ret < 0) {
1295 spin_unlock(&fs_info->qgroup_lock);
1296 goto out;
1297 }
1298 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1299 spin_unlock(&fs_info->qgroup_lock);
1300out:
1301 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1302 ulist_free(tmp);
1303 return ret;
1304}
1305
1306static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1307 u64 dst)
1308{
1309 struct btrfs_fs_info *fs_info = trans->fs_info;
1310 struct btrfs_root *quota_root;
1311 struct btrfs_qgroup *parent;
1312 struct btrfs_qgroup *member;
1313 struct btrfs_qgroup_list *list;
1314 struct ulist *tmp;
1315 bool found = false;
1316 int ret = 0;
1317 int ret2;
1318
1319 tmp = ulist_alloc(GFP_KERNEL);
1320 if (!tmp)
1321 return -ENOMEM;
1322
1323 quota_root = fs_info->quota_root;
1324 if (!quota_root) {
1325 ret = -EINVAL;
1326 goto out;
1327 }
1328
1329 member = find_qgroup_rb(fs_info, src);
1330 parent = find_qgroup_rb(fs_info, dst);
1331 /*
1332 * The parent/member pair doesn't exist, then try to delete the dead
1333 * relation items only.
1334 */
1335 if (!member || !parent)
1336 goto delete_item;
1337
1338 /* check if such qgroup relation exist firstly */
1339 list_for_each_entry(list, &member->groups, next_group) {
1340 if (list->group == parent) {
1341 found = true;
1342 break;
1343 }
1344 }
1345
1346delete_item:
1347 ret = del_qgroup_relation_item(trans, src, dst);
1348 if (ret < 0 && ret != -ENOENT)
1349 goto out;
1350 ret2 = del_qgroup_relation_item(trans, dst, src);
1351 if (ret2 < 0 && ret2 != -ENOENT)
1352 goto out;
1353
1354 /* At least one deletion succeeded, return 0 */
1355 if (!ret || !ret2)
1356 ret = 0;
1357
1358 if (found) {
1359 spin_lock(&fs_info->qgroup_lock);
1360 del_relation_rb(fs_info, src, dst);
1361 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1362 spin_unlock(&fs_info->qgroup_lock);
1363 }
1364out:
1365 ulist_free(tmp);
1366 return ret;
1367}
1368
1369int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1370 u64 dst)
1371{
1372 struct btrfs_fs_info *fs_info = trans->fs_info;
1373 int ret = 0;
1374
1375 mutex_lock(&fs_info->qgroup_ioctl_lock);
1376 ret = __del_qgroup_relation(trans, src, dst);
1377 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1378
1379 return ret;
1380}
1381
1382int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1383{
1384 struct btrfs_fs_info *fs_info = trans->fs_info;
1385 struct btrfs_root *quota_root;
1386 struct btrfs_qgroup *qgroup;
1387 int ret = 0;
1388
1389 mutex_lock(&fs_info->qgroup_ioctl_lock);
1390 quota_root = fs_info->quota_root;
1391 if (!quota_root) {
1392 ret = -EINVAL;
1393 goto out;
1394 }
1395 qgroup = find_qgroup_rb(fs_info, qgroupid);
1396 if (qgroup) {
1397 ret = -EEXIST;
1398 goto out;
1399 }
1400
1401 ret = add_qgroup_item(trans, quota_root, qgroupid);
1402 if (ret)
1403 goto out;
1404
1405 spin_lock(&fs_info->qgroup_lock);
1406 qgroup = add_qgroup_rb(fs_info, qgroupid);
1407 spin_unlock(&fs_info->qgroup_lock);
1408
1409 if (IS_ERR(qgroup))
1410 ret = PTR_ERR(qgroup);
1411out:
1412 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1413 return ret;
1414}
1415
1416int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1417{
1418 struct btrfs_fs_info *fs_info = trans->fs_info;
1419 struct btrfs_root *quota_root;
1420 struct btrfs_qgroup *qgroup;
1421 struct btrfs_qgroup_list *list;
1422 int ret = 0;
1423
1424 mutex_lock(&fs_info->qgroup_ioctl_lock);
1425 quota_root = fs_info->quota_root;
1426 if (!quota_root) {
1427 ret = -EINVAL;
1428 goto out;
1429 }
1430
1431 qgroup = find_qgroup_rb(fs_info, qgroupid);
1432 if (!qgroup) {
1433 ret = -ENOENT;
1434 goto out;
1435 }
1436
1437 /* Check if there are no children of this qgroup */
1438 if (!list_empty(&qgroup->members)) {
1439 ret = -EBUSY;
1440 goto out;
1441 }
1442
1443 ret = del_qgroup_item(trans, qgroupid);
1444 if (ret && ret != -ENOENT)
1445 goto out;
1446
1447 while (!list_empty(&qgroup->groups)) {
1448 list = list_first_entry(&qgroup->groups,
1449 struct btrfs_qgroup_list, next_group);
1450 ret = __del_qgroup_relation(trans, qgroupid,
1451 list->group->qgroupid);
1452 if (ret)
1453 goto out;
1454 }
1455
1456 spin_lock(&fs_info->qgroup_lock);
1457 del_qgroup_rb(fs_info, qgroupid);
1458 spin_unlock(&fs_info->qgroup_lock);
1459out:
1460 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1461 return ret;
1462}
1463
1464int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1465 struct btrfs_qgroup_limit *limit)
1466{
1467 struct btrfs_fs_info *fs_info = trans->fs_info;
1468 struct btrfs_root *quota_root;
1469 struct btrfs_qgroup *qgroup;
1470 int ret = 0;
1471 /* Sometimes we would want to clear the limit on this qgroup.
1472 * To meet this requirement, we treat the -1 as a special value
1473 * which tell kernel to clear the limit on this qgroup.
1474 */
1475 const u64 CLEAR_VALUE = -1;
1476
1477 mutex_lock(&fs_info->qgroup_ioctl_lock);
1478 quota_root = fs_info->quota_root;
1479 if (!quota_root) {
1480 ret = -EINVAL;
1481 goto out;
1482 }
1483
1484 qgroup = find_qgroup_rb(fs_info, qgroupid);
1485 if (!qgroup) {
1486 ret = -ENOENT;
1487 goto out;
1488 }
1489
1490 spin_lock(&fs_info->qgroup_lock);
1491 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1492 if (limit->max_rfer == CLEAR_VALUE) {
1493 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1494 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1495 qgroup->max_rfer = 0;
1496 } else {
1497 qgroup->max_rfer = limit->max_rfer;
1498 }
1499 }
1500 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1501 if (limit->max_excl == CLEAR_VALUE) {
1502 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1503 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1504 qgroup->max_excl = 0;
1505 } else {
1506 qgroup->max_excl = limit->max_excl;
1507 }
1508 }
1509 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1510 if (limit->rsv_rfer == CLEAR_VALUE) {
1511 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1512 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1513 qgroup->rsv_rfer = 0;
1514 } else {
1515 qgroup->rsv_rfer = limit->rsv_rfer;
1516 }
1517 }
1518 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1519 if (limit->rsv_excl == CLEAR_VALUE) {
1520 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1521 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1522 qgroup->rsv_excl = 0;
1523 } else {
1524 qgroup->rsv_excl = limit->rsv_excl;
1525 }
1526 }
1527 qgroup->lim_flags |= limit->flags;
1528
1529 spin_unlock(&fs_info->qgroup_lock);
1530
1531 ret = update_qgroup_limit_item(trans, qgroup);
1532 if (ret) {
1533 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1534 btrfs_info(fs_info, "unable to update quota limit for %llu",
1535 qgroupid);
1536 }
1537
1538out:
1539 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1540 return ret;
1541}
1542
1543int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1544 struct btrfs_delayed_ref_root *delayed_refs,
1545 struct btrfs_qgroup_extent_record *record)
1546{
1547 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1548 struct rb_node *parent_node = NULL;
1549 struct btrfs_qgroup_extent_record *entry;
1550 u64 bytenr = record->bytenr;
1551
1552 lockdep_assert_held(&delayed_refs->lock);
1553 trace_btrfs_qgroup_trace_extent(fs_info, record);
1554
1555 while (*p) {
1556 parent_node = *p;
1557 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1558 node);
1559 if (bytenr < entry->bytenr) {
1560 p = &(*p)->rb_left;
1561 } else if (bytenr > entry->bytenr) {
1562 p = &(*p)->rb_right;
1563 } else {
1564 if (record->data_rsv && !entry->data_rsv) {
1565 entry->data_rsv = record->data_rsv;
1566 entry->data_rsv_refroot =
1567 record->data_rsv_refroot;
1568 }
1569 return 1;
1570 }
1571 }
1572
1573 rb_link_node(&record->node, parent_node, p);
1574 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1575 return 0;
1576}
1577
1578int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1579 struct btrfs_qgroup_extent_record *qrecord)
1580{
1581 struct ulist *old_root;
1582 u64 bytenr = qrecord->bytenr;
1583 int ret;
1584
1585 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1586 if (ret < 0) {
1587 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1588 btrfs_warn(fs_info,
1589"error accounting new delayed refs extent (err code: %d), quota inconsistent",
1590 ret);
1591 return 0;
1592 }
1593
1594 /*
1595 * Here we don't need to get the lock of
1596 * trans->transaction->delayed_refs, since inserted qrecord won't
1597 * be deleted, only qrecord->node may be modified (new qrecord insert)
1598 *
1599 * So modifying qrecord->old_roots is safe here
1600 */
1601 qrecord->old_roots = old_root;
1602 return 0;
1603}
1604
1605int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1606 u64 num_bytes, gfp_t gfp_flag)
1607{
1608 struct btrfs_fs_info *fs_info = trans->fs_info;
1609 struct btrfs_qgroup_extent_record *record;
1610 struct btrfs_delayed_ref_root *delayed_refs;
1611 int ret;
1612
1613 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1614 || bytenr == 0 || num_bytes == 0)
1615 return 0;
1616 record = kzalloc(sizeof(*record), gfp_flag);
1617 if (!record)
1618 return -ENOMEM;
1619
1620 delayed_refs = &trans->transaction->delayed_refs;
1621 record->bytenr = bytenr;
1622 record->num_bytes = num_bytes;
1623 record->old_roots = NULL;
1624
1625 spin_lock(&delayed_refs->lock);
1626 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1627 spin_unlock(&delayed_refs->lock);
1628 if (ret > 0) {
1629 kfree(record);
1630 return 0;
1631 }
1632 return btrfs_qgroup_trace_extent_post(fs_info, record);
1633}
1634
1635int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1636 struct extent_buffer *eb)
1637{
1638 struct btrfs_fs_info *fs_info = trans->fs_info;
1639 int nr = btrfs_header_nritems(eb);
1640 int i, extent_type, ret;
1641 struct btrfs_key key;
1642 struct btrfs_file_extent_item *fi;
1643 u64 bytenr, num_bytes;
1644
1645 /* We can be called directly from walk_up_proc() */
1646 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1647 return 0;
1648
1649 for (i = 0; i < nr; i++) {
1650 btrfs_item_key_to_cpu(eb, &key, i);
1651
1652 if (key.type != BTRFS_EXTENT_DATA_KEY)
1653 continue;
1654
1655 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1656 /* filter out non qgroup-accountable extents */
1657 extent_type = btrfs_file_extent_type(eb, fi);
1658
1659 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1660 continue;
1661
1662 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1663 if (!bytenr)
1664 continue;
1665
1666 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1667
1668 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1669 GFP_NOFS);
1670 if (ret)
1671 return ret;
1672 }
1673 cond_resched();
1674 return 0;
1675}
1676
1677/*
1678 * Walk up the tree from the bottom, freeing leaves and any interior
1679 * nodes which have had all slots visited. If a node (leaf or
1680 * interior) is freed, the node above it will have it's slot
1681 * incremented. The root node will never be freed.
1682 *
1683 * At the end of this function, we should have a path which has all
1684 * slots incremented to the next position for a search. If we need to
1685 * read a new node it will be NULL and the node above it will have the
1686 * correct slot selected for a later read.
1687 *
1688 * If we increment the root nodes slot counter past the number of
1689 * elements, 1 is returned to signal completion of the search.
1690 */
1691static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1692{
1693 int level = 0;
1694 int nr, slot;
1695 struct extent_buffer *eb;
1696
1697 if (root_level == 0)
1698 return 1;
1699
1700 while (level <= root_level) {
1701 eb = path->nodes[level];
1702 nr = btrfs_header_nritems(eb);
1703 path->slots[level]++;
1704 slot = path->slots[level];
1705 if (slot >= nr || level == 0) {
1706 /*
1707 * Don't free the root - we will detect this
1708 * condition after our loop and return a
1709 * positive value for caller to stop walking the tree.
1710 */
1711 if (level != root_level) {
1712 btrfs_tree_unlock_rw(eb, path->locks[level]);
1713 path->locks[level] = 0;
1714
1715 free_extent_buffer(eb);
1716 path->nodes[level] = NULL;
1717 path->slots[level] = 0;
1718 }
1719 } else {
1720 /*
1721 * We have a valid slot to walk back down
1722 * from. Stop here so caller can process these
1723 * new nodes.
1724 */
1725 break;
1726 }
1727
1728 level++;
1729 }
1730
1731 eb = path->nodes[root_level];
1732 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1733 return 1;
1734
1735 return 0;
1736}
1737
1738/*
1739 * Helper function to trace a subtree tree block swap.
1740 *
1741 * The swap will happen in highest tree block, but there may be a lot of
1742 * tree blocks involved.
1743 *
1744 * For example:
1745 * OO = Old tree blocks
1746 * NN = New tree blocks allocated during balance
1747 *
1748 * File tree (257) Reloc tree for 257
1749 * L2 OO NN
1750 * / \ / \
1751 * L1 OO OO (a) OO NN (a)
1752 * / \ / \ / \ / \
1753 * L0 OO OO OO OO OO OO NN NN
1754 * (b) (c) (b) (c)
1755 *
1756 * When calling qgroup_trace_extent_swap(), we will pass:
1757 * @src_eb = OO(a)
1758 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1759 * @dst_level = 0
1760 * @root_level = 1
1761 *
1762 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1763 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1764 *
1765 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1766 *
1767 * 1) Tree search from @src_eb
1768 * It should acts as a simplified btrfs_search_slot().
1769 * The key for search can be extracted from @dst_path->nodes[dst_level]
1770 * (first key).
1771 *
1772 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1773 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1774 * They should be marked during previous (@dst_level = 1) iteration.
1775 *
1776 * 3) Mark file extents in leaves dirty
1777 * We don't have good way to pick out new file extents only.
1778 * So we still follow the old method by scanning all file extents in
1779 * the leave.
1780 *
1781 * This function can free us from keeping two paths, thus later we only need
1782 * to care about how to iterate all new tree blocks in reloc tree.
1783 */
1784static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1785 struct extent_buffer *src_eb,
1786 struct btrfs_path *dst_path,
1787 int dst_level, int root_level,
1788 bool trace_leaf)
1789{
1790 struct btrfs_key key;
1791 struct btrfs_path *src_path;
1792 struct btrfs_fs_info *fs_info = trans->fs_info;
1793 u32 nodesize = fs_info->nodesize;
1794 int cur_level = root_level;
1795 int ret;
1796
1797 BUG_ON(dst_level > root_level);
1798 /* Level mismatch */
1799 if (btrfs_header_level(src_eb) != root_level)
1800 return -EINVAL;
1801
1802 src_path = btrfs_alloc_path();
1803 if (!src_path) {
1804 ret = -ENOMEM;
1805 goto out;
1806 }
1807
1808 if (dst_level)
1809 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1810 else
1811 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1812
1813 /* For src_path */
1814 extent_buffer_get(src_eb);
1815 src_path->nodes[root_level] = src_eb;
1816 src_path->slots[root_level] = dst_path->slots[root_level];
1817 src_path->locks[root_level] = 0;
1818
1819 /* A simplified version of btrfs_search_slot() */
1820 while (cur_level >= dst_level) {
1821 struct btrfs_key src_key;
1822 struct btrfs_key dst_key;
1823
1824 if (src_path->nodes[cur_level] == NULL) {
1825 struct btrfs_key first_key;
1826 struct extent_buffer *eb;
1827 int parent_slot;
1828 u64 child_gen;
1829 u64 child_bytenr;
1830
1831 eb = src_path->nodes[cur_level + 1];
1832 parent_slot = src_path->slots[cur_level + 1];
1833 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1834 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1835 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1836
1837 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1838 cur_level, &first_key);
1839 if (IS_ERR(eb)) {
1840 ret = PTR_ERR(eb);
1841 goto out;
1842 } else if (!extent_buffer_uptodate(eb)) {
1843 free_extent_buffer(eb);
1844 ret = -EIO;
1845 goto out;
1846 }
1847
1848 src_path->nodes[cur_level] = eb;
1849
1850 btrfs_tree_read_lock(eb);
1851 btrfs_set_lock_blocking_read(eb);
1852 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1853 }
1854
1855 src_path->slots[cur_level] = dst_path->slots[cur_level];
1856 if (cur_level) {
1857 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1858 &dst_key, dst_path->slots[cur_level]);
1859 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1860 &src_key, src_path->slots[cur_level]);
1861 } else {
1862 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1863 &dst_key, dst_path->slots[cur_level]);
1864 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1865 &src_key, src_path->slots[cur_level]);
1866 }
1867 /* Content mismatch, something went wrong */
1868 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1869 ret = -ENOENT;
1870 goto out;
1871 }
1872 cur_level--;
1873 }
1874
1875 /*
1876 * Now both @dst_path and @src_path have been populated, record the tree
1877 * blocks for qgroup accounting.
1878 */
1879 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1880 nodesize, GFP_NOFS);
1881 if (ret < 0)
1882 goto out;
1883 ret = btrfs_qgroup_trace_extent(trans,
1884 dst_path->nodes[dst_level]->start,
1885 nodesize, GFP_NOFS);
1886 if (ret < 0)
1887 goto out;
1888
1889 /* Record leaf file extents */
1890 if (dst_level == 0 && trace_leaf) {
1891 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1892 if (ret < 0)
1893 goto out;
1894 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1895 }
1896out:
1897 btrfs_free_path(src_path);
1898 return ret;
1899}
1900
1901/*
1902 * Helper function to do recursive generation-aware depth-first search, to
1903 * locate all new tree blocks in a subtree of reloc tree.
1904 *
1905 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1906 * reloc tree
1907 * L2 NN (a)
1908 * / \
1909 * L1 OO NN (b)
1910 * / \ / \
1911 * L0 OO OO OO NN
1912 * (c) (d)
1913 * If we pass:
1914 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1915 * @cur_level = 1
1916 * @root_level = 1
1917 *
1918 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1919 * above tree blocks along with their counter parts in file tree.
1920 * While during search, old tree blocks OO(c) will be skipped as tree block swap
1921 * won't affect OO(c).
1922 */
1923static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1924 struct extent_buffer *src_eb,
1925 struct btrfs_path *dst_path,
1926 int cur_level, int root_level,
1927 u64 last_snapshot, bool trace_leaf)
1928{
1929 struct btrfs_fs_info *fs_info = trans->fs_info;
1930 struct extent_buffer *eb;
1931 bool need_cleanup = false;
1932 int ret = 0;
1933 int i;
1934
1935 /* Level sanity check */
1936 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1937 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1938 root_level < cur_level) {
1939 btrfs_err_rl(fs_info,
1940 "%s: bad levels, cur_level=%d root_level=%d",
1941 __func__, cur_level, root_level);
1942 return -EUCLEAN;
1943 }
1944
1945 /* Read the tree block if needed */
1946 if (dst_path->nodes[cur_level] == NULL) {
1947 struct btrfs_key first_key;
1948 int parent_slot;
1949 u64 child_gen;
1950 u64 child_bytenr;
1951
1952 /*
1953 * dst_path->nodes[root_level] must be initialized before
1954 * calling this function.
1955 */
1956 if (cur_level == root_level) {
1957 btrfs_err_rl(fs_info,
1958 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1959 __func__, root_level, root_level, cur_level);
1960 return -EUCLEAN;
1961 }
1962
1963 /*
1964 * We need to get child blockptr/gen from parent before we can
1965 * read it.
1966 */
1967 eb = dst_path->nodes[cur_level + 1];
1968 parent_slot = dst_path->slots[cur_level + 1];
1969 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1970 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1971 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1972
1973 /* This node is old, no need to trace */
1974 if (child_gen < last_snapshot)
1975 goto out;
1976
1977 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1978 cur_level, &first_key);
1979 if (IS_ERR(eb)) {
1980 ret = PTR_ERR(eb);
1981 goto out;
1982 } else if (!extent_buffer_uptodate(eb)) {
1983 free_extent_buffer(eb);
1984 ret = -EIO;
1985 goto out;
1986 }
1987
1988 dst_path->nodes[cur_level] = eb;
1989 dst_path->slots[cur_level] = 0;
1990
1991 btrfs_tree_read_lock(eb);
1992 btrfs_set_lock_blocking_read(eb);
1993 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1994 need_cleanup = true;
1995 }
1996
1997 /* Now record this tree block and its counter part for qgroups */
1998 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1999 root_level, trace_leaf);
2000 if (ret < 0)
2001 goto cleanup;
2002
2003 eb = dst_path->nodes[cur_level];
2004
2005 if (cur_level > 0) {
2006 /* Iterate all child tree blocks */
2007 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2008 /* Skip old tree blocks as they won't be swapped */
2009 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2010 continue;
2011 dst_path->slots[cur_level] = i;
2012
2013 /* Recursive call (at most 7 times) */
2014 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2015 dst_path, cur_level - 1, root_level,
2016 last_snapshot, trace_leaf);
2017 if (ret < 0)
2018 goto cleanup;
2019 }
2020 }
2021
2022cleanup:
2023 if (need_cleanup) {
2024 /* Clean up */
2025 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2026 dst_path->locks[cur_level]);
2027 free_extent_buffer(dst_path->nodes[cur_level]);
2028 dst_path->nodes[cur_level] = NULL;
2029 dst_path->slots[cur_level] = 0;
2030 dst_path->locks[cur_level] = 0;
2031 }
2032out:
2033 return ret;
2034}
2035
2036static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2037 struct extent_buffer *src_eb,
2038 struct extent_buffer *dst_eb,
2039 u64 last_snapshot, bool trace_leaf)
2040{
2041 struct btrfs_fs_info *fs_info = trans->fs_info;
2042 struct btrfs_path *dst_path = NULL;
2043 int level;
2044 int ret;
2045
2046 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2047 return 0;
2048
2049 /* Wrong parameter order */
2050 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2051 btrfs_err_rl(fs_info,
2052 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2053 btrfs_header_generation(src_eb),
2054 btrfs_header_generation(dst_eb));
2055 return -EUCLEAN;
2056 }
2057
2058 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2059 ret = -EIO;
2060 goto out;
2061 }
2062
2063 level = btrfs_header_level(dst_eb);
2064 dst_path = btrfs_alloc_path();
2065 if (!dst_path) {
2066 ret = -ENOMEM;
2067 goto out;
2068 }
2069 /* For dst_path */
2070 extent_buffer_get(dst_eb);
2071 dst_path->nodes[level] = dst_eb;
2072 dst_path->slots[level] = 0;
2073 dst_path->locks[level] = 0;
2074
2075 /* Do the generation aware breadth-first search */
2076 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2077 level, last_snapshot, trace_leaf);
2078 if (ret < 0)
2079 goto out;
2080 ret = 0;
2081
2082out:
2083 btrfs_free_path(dst_path);
2084 if (ret < 0)
2085 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2086 return ret;
2087}
2088
2089int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2090 struct extent_buffer *root_eb,
2091 u64 root_gen, int root_level)
2092{
2093 struct btrfs_fs_info *fs_info = trans->fs_info;
2094 int ret = 0;
2095 int level;
2096 struct extent_buffer *eb = root_eb;
2097 struct btrfs_path *path = NULL;
2098
2099 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2100 BUG_ON(root_eb == NULL);
2101
2102 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2103 return 0;
2104
2105 if (!extent_buffer_uptodate(root_eb)) {
2106 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2107 if (ret)
2108 goto out;
2109 }
2110
2111 if (root_level == 0) {
2112 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2113 goto out;
2114 }
2115
2116 path = btrfs_alloc_path();
2117 if (!path)
2118 return -ENOMEM;
2119
2120 /*
2121 * Walk down the tree. Missing extent blocks are filled in as
2122 * we go. Metadata is accounted every time we read a new
2123 * extent block.
2124 *
2125 * When we reach a leaf, we account for file extent items in it,
2126 * walk back up the tree (adjusting slot pointers as we go)
2127 * and restart the search process.
2128 */
2129 extent_buffer_get(root_eb); /* For path */
2130 path->nodes[root_level] = root_eb;
2131 path->slots[root_level] = 0;
2132 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2133walk_down:
2134 level = root_level;
2135 while (level >= 0) {
2136 if (path->nodes[level] == NULL) {
2137 struct btrfs_key first_key;
2138 int parent_slot;
2139 u64 child_gen;
2140 u64 child_bytenr;
2141
2142 /*
2143 * We need to get child blockptr/gen from parent before
2144 * we can read it.
2145 */
2146 eb = path->nodes[level + 1];
2147 parent_slot = path->slots[level + 1];
2148 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2149 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2150 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2151
2152 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2153 level, &first_key);
2154 if (IS_ERR(eb)) {
2155 ret = PTR_ERR(eb);
2156 goto out;
2157 } else if (!extent_buffer_uptodate(eb)) {
2158 free_extent_buffer(eb);
2159 ret = -EIO;
2160 goto out;
2161 }
2162
2163 path->nodes[level] = eb;
2164 path->slots[level] = 0;
2165
2166 btrfs_tree_read_lock(eb);
2167 btrfs_set_lock_blocking_read(eb);
2168 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2169
2170 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2171 fs_info->nodesize,
2172 GFP_NOFS);
2173 if (ret)
2174 goto out;
2175 }
2176
2177 if (level == 0) {
2178 ret = btrfs_qgroup_trace_leaf_items(trans,
2179 path->nodes[level]);
2180 if (ret)
2181 goto out;
2182
2183 /* Nonzero return here means we completed our search */
2184 ret = adjust_slots_upwards(path, root_level);
2185 if (ret)
2186 break;
2187
2188 /* Restart search with new slots */
2189 goto walk_down;
2190 }
2191
2192 level--;
2193 }
2194
2195 ret = 0;
2196out:
2197 btrfs_free_path(path);
2198
2199 return ret;
2200}
2201
2202#define UPDATE_NEW 0
2203#define UPDATE_OLD 1
2204/*
2205 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2206 */
2207static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2208 struct ulist *roots, struct ulist *tmp,
2209 struct ulist *qgroups, u64 seq, int update_old)
2210{
2211 struct ulist_node *unode;
2212 struct ulist_iterator uiter;
2213 struct ulist_node *tmp_unode;
2214 struct ulist_iterator tmp_uiter;
2215 struct btrfs_qgroup *qg;
2216 int ret = 0;
2217
2218 if (!roots)
2219 return 0;
2220 ULIST_ITER_INIT(&uiter);
2221 while ((unode = ulist_next(roots, &uiter))) {
2222 qg = find_qgroup_rb(fs_info, unode->val);
2223 if (!qg)
2224 continue;
2225
2226 ulist_reinit(tmp);
2227 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2228 GFP_ATOMIC);
2229 if (ret < 0)
2230 return ret;
2231 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2232 if (ret < 0)
2233 return ret;
2234 ULIST_ITER_INIT(&tmp_uiter);
2235 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2236 struct btrfs_qgroup_list *glist;
2237
2238 qg = unode_aux_to_qgroup(tmp_unode);
2239 if (update_old)
2240 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2241 else
2242 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2243 list_for_each_entry(glist, &qg->groups, next_group) {
2244 ret = ulist_add(qgroups, glist->group->qgroupid,
2245 qgroup_to_aux(glist->group),
2246 GFP_ATOMIC);
2247 if (ret < 0)
2248 return ret;
2249 ret = ulist_add(tmp, glist->group->qgroupid,
2250 qgroup_to_aux(glist->group),
2251 GFP_ATOMIC);
2252 if (ret < 0)
2253 return ret;
2254 }
2255 }
2256 }
2257 return 0;
2258}
2259
2260/*
2261 * Update qgroup rfer/excl counters.
2262 * Rfer update is easy, codes can explain themselves.
2263 *
2264 * Excl update is tricky, the update is split into 2 part.
2265 * Part 1: Possible exclusive <-> sharing detect:
2266 * | A | !A |
2267 * -------------------------------------
2268 * B | * | - |
2269 * -------------------------------------
2270 * !B | + | ** |
2271 * -------------------------------------
2272 *
2273 * Conditions:
2274 * A: cur_old_roots < nr_old_roots (not exclusive before)
2275 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2276 * B: cur_new_roots < nr_new_roots (not exclusive now)
2277 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2278 *
2279 * Results:
2280 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2281 * *: Definitely not changed. **: Possible unchanged.
2282 *
2283 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2284 *
2285 * To make the logic clear, we first use condition A and B to split
2286 * combination into 4 results.
2287 *
2288 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2289 * only on variant maybe 0.
2290 *
2291 * Lastly, check result **, since there are 2 variants maybe 0, split them
2292 * again(2x2).
2293 * But this time we don't need to consider other things, the codes and logic
2294 * is easy to understand now.
2295 */
2296static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2297 struct ulist *qgroups,
2298 u64 nr_old_roots,
2299 u64 nr_new_roots,
2300 u64 num_bytes, u64 seq)
2301{
2302 struct ulist_node *unode;
2303 struct ulist_iterator uiter;
2304 struct btrfs_qgroup *qg;
2305 u64 cur_new_count, cur_old_count;
2306
2307 ULIST_ITER_INIT(&uiter);
2308 while ((unode = ulist_next(qgroups, &uiter))) {
2309 bool dirty = false;
2310
2311 qg = unode_aux_to_qgroup(unode);
2312 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2313 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2314
2315 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2316 cur_new_count);
2317
2318 /* Rfer update part */
2319 if (cur_old_count == 0 && cur_new_count > 0) {
2320 qg->rfer += num_bytes;
2321 qg->rfer_cmpr += num_bytes;
2322 dirty = true;
2323 }
2324 if (cur_old_count > 0 && cur_new_count == 0) {
2325 qg->rfer -= num_bytes;
2326 qg->rfer_cmpr -= num_bytes;
2327 dirty = true;
2328 }
2329
2330 /* Excl update part */
2331 /* Exclusive/none -> shared case */
2332 if (cur_old_count == nr_old_roots &&
2333 cur_new_count < nr_new_roots) {
2334 /* Exclusive -> shared */
2335 if (cur_old_count != 0) {
2336 qg->excl -= num_bytes;
2337 qg->excl_cmpr -= num_bytes;
2338 dirty = true;
2339 }
2340 }
2341
2342 /* Shared -> exclusive/none case */
2343 if (cur_old_count < nr_old_roots &&
2344 cur_new_count == nr_new_roots) {
2345 /* Shared->exclusive */
2346 if (cur_new_count != 0) {
2347 qg->excl += num_bytes;
2348 qg->excl_cmpr += num_bytes;
2349 dirty = true;
2350 }
2351 }
2352
2353 /* Exclusive/none -> exclusive/none case */
2354 if (cur_old_count == nr_old_roots &&
2355 cur_new_count == nr_new_roots) {
2356 if (cur_old_count == 0) {
2357 /* None -> exclusive/none */
2358
2359 if (cur_new_count != 0) {
2360 /* None -> exclusive */
2361 qg->excl += num_bytes;
2362 qg->excl_cmpr += num_bytes;
2363 dirty = true;
2364 }
2365 /* None -> none, nothing changed */
2366 } else {
2367 /* Exclusive -> exclusive/none */
2368
2369 if (cur_new_count == 0) {
2370 /* Exclusive -> none */
2371 qg->excl -= num_bytes;
2372 qg->excl_cmpr -= num_bytes;
2373 dirty = true;
2374 }
2375 /* Exclusive -> exclusive, nothing changed */
2376 }
2377 }
2378
2379 if (dirty)
2380 qgroup_dirty(fs_info, qg);
2381 }
2382 return 0;
2383}
2384
2385/*
2386 * Check if the @roots potentially is a list of fs tree roots
2387 *
2388 * Return 0 for definitely not a fs/subvol tree roots ulist
2389 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2390 * one as well)
2391 */
2392static int maybe_fs_roots(struct ulist *roots)
2393{
2394 struct ulist_node *unode;
2395 struct ulist_iterator uiter;
2396
2397 /* Empty one, still possible for fs roots */
2398 if (!roots || roots->nnodes == 0)
2399 return 1;
2400
2401 ULIST_ITER_INIT(&uiter);
2402 unode = ulist_next(roots, &uiter);
2403 if (!unode)
2404 return 1;
2405
2406 /*
2407 * If it contains fs tree roots, then it must belong to fs/subvol
2408 * trees.
2409 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2410 */
2411 return is_fstree(unode->val);
2412}
2413
2414int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2415 u64 num_bytes, struct ulist *old_roots,
2416 struct ulist *new_roots)
2417{
2418 struct btrfs_fs_info *fs_info = trans->fs_info;
2419 struct ulist *qgroups = NULL;
2420 struct ulist *tmp = NULL;
2421 u64 seq;
2422 u64 nr_new_roots = 0;
2423 u64 nr_old_roots = 0;
2424 int ret = 0;
2425
2426 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2427 return 0;
2428
2429 if (new_roots) {
2430 if (!maybe_fs_roots(new_roots))
2431 goto out_free;
2432 nr_new_roots = new_roots->nnodes;
2433 }
2434 if (old_roots) {
2435 if (!maybe_fs_roots(old_roots))
2436 goto out_free;
2437 nr_old_roots = old_roots->nnodes;
2438 }
2439
2440 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2441 if (nr_old_roots == 0 && nr_new_roots == 0)
2442 goto out_free;
2443
2444 BUG_ON(!fs_info->quota_root);
2445
2446 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2447 num_bytes, nr_old_roots, nr_new_roots);
2448
2449 qgroups = ulist_alloc(GFP_NOFS);
2450 if (!qgroups) {
2451 ret = -ENOMEM;
2452 goto out_free;
2453 }
2454 tmp = ulist_alloc(GFP_NOFS);
2455 if (!tmp) {
2456 ret = -ENOMEM;
2457 goto out_free;
2458 }
2459
2460 mutex_lock(&fs_info->qgroup_rescan_lock);
2461 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2462 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2463 mutex_unlock(&fs_info->qgroup_rescan_lock);
2464 ret = 0;
2465 goto out_free;
2466 }
2467 }
2468 mutex_unlock(&fs_info->qgroup_rescan_lock);
2469
2470 spin_lock(&fs_info->qgroup_lock);
2471 seq = fs_info->qgroup_seq;
2472
2473 /* Update old refcnts using old_roots */
2474 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2475 UPDATE_OLD);
2476 if (ret < 0)
2477 goto out;
2478
2479 /* Update new refcnts using new_roots */
2480 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2481 UPDATE_NEW);
2482 if (ret < 0)
2483 goto out;
2484
2485 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2486 num_bytes, seq);
2487
2488 /*
2489 * Bump qgroup_seq to avoid seq overlap
2490 */
2491 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2492out:
2493 spin_unlock(&fs_info->qgroup_lock);
2494out_free:
2495 ulist_free(tmp);
2496 ulist_free(qgroups);
2497 ulist_free(old_roots);
2498 ulist_free(new_roots);
2499 return ret;
2500}
2501
2502int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2503{
2504 struct btrfs_fs_info *fs_info = trans->fs_info;
2505 struct btrfs_qgroup_extent_record *record;
2506 struct btrfs_delayed_ref_root *delayed_refs;
2507 struct ulist *new_roots = NULL;
2508 struct rb_node *node;
2509 u64 num_dirty_extents = 0;
2510 u64 qgroup_to_skip;
2511 int ret = 0;
2512
2513 delayed_refs = &trans->transaction->delayed_refs;
2514 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2515 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2516 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2517 node);
2518
2519 num_dirty_extents++;
2520 trace_btrfs_qgroup_account_extents(fs_info, record);
2521
2522 if (!ret) {
2523 /*
2524 * Old roots should be searched when inserting qgroup
2525 * extent record
2526 */
2527 if (WARN_ON(!record->old_roots)) {
2528 /* Search commit root to find old_roots */
2529 ret = btrfs_find_all_roots(NULL, fs_info,
2530 record->bytenr, 0,
2531 &record->old_roots, false);
2532 if (ret < 0)
2533 goto cleanup;
2534 }
2535
2536 /* Free the reserved data space */
2537 btrfs_qgroup_free_refroot(fs_info,
2538 record->data_rsv_refroot,
2539 record->data_rsv,
2540 BTRFS_QGROUP_RSV_DATA);
2541 /*
2542 * Use SEQ_LAST as time_seq to do special search, which
2543 * doesn't lock tree or delayed_refs and search current
2544 * root. It's safe inside commit_transaction().
2545 */
2546 ret = btrfs_find_all_roots(trans, fs_info,
2547 record->bytenr, SEQ_LAST, &new_roots, false);
2548 if (ret < 0)
2549 goto cleanup;
2550 if (qgroup_to_skip) {
2551 ulist_del(new_roots, qgroup_to_skip, 0);
2552 ulist_del(record->old_roots, qgroup_to_skip,
2553 0);
2554 }
2555 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2556 record->num_bytes,
2557 record->old_roots,
2558 new_roots);
2559 record->old_roots = NULL;
2560 new_roots = NULL;
2561 }
2562cleanup:
2563 ulist_free(record->old_roots);
2564 ulist_free(new_roots);
2565 new_roots = NULL;
2566 rb_erase(node, &delayed_refs->dirty_extent_root);
2567 kfree(record);
2568
2569 }
2570 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2571 num_dirty_extents);
2572 return ret;
2573}
2574
2575/*
2576 * called from commit_transaction. Writes all changed qgroups to disk.
2577 */
2578int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2579{
2580 struct btrfs_fs_info *fs_info = trans->fs_info;
2581 struct btrfs_root *quota_root = fs_info->quota_root;
2582 int ret = 0;
2583
2584 if (!quota_root)
2585 return ret;
2586
2587 spin_lock(&fs_info->qgroup_lock);
2588 while (!list_empty(&fs_info->dirty_qgroups)) {
2589 struct btrfs_qgroup *qgroup;
2590 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2591 struct btrfs_qgroup, dirty);
2592 list_del_init(&qgroup->dirty);
2593 spin_unlock(&fs_info->qgroup_lock);
2594 ret = update_qgroup_info_item(trans, qgroup);
2595 if (ret)
2596 fs_info->qgroup_flags |=
2597 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2598 ret = update_qgroup_limit_item(trans, qgroup);
2599 if (ret)
2600 fs_info->qgroup_flags |=
2601 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2602 spin_lock(&fs_info->qgroup_lock);
2603 }
2604 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2605 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2606 else
2607 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2608 spin_unlock(&fs_info->qgroup_lock);
2609
2610 ret = update_qgroup_status_item(trans);
2611 if (ret)
2612 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2613
2614 return ret;
2615}
2616
2617/*
2618 * Copy the accounting information between qgroups. This is necessary
2619 * when a snapshot or a subvolume is created. Throwing an error will
2620 * cause a transaction abort so we take extra care here to only error
2621 * when a readonly fs is a reasonable outcome.
2622 */
2623int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2624 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2625{
2626 int ret = 0;
2627 int i;
2628 u64 *i_qgroups;
2629 bool committing = false;
2630 struct btrfs_fs_info *fs_info = trans->fs_info;
2631 struct btrfs_root *quota_root;
2632 struct btrfs_qgroup *srcgroup;
2633 struct btrfs_qgroup *dstgroup;
2634 u32 level_size = 0;
2635 u64 nums;
2636
2637 /*
2638 * There are only two callers of this function.
2639 *
2640 * One in create_subvol() in the ioctl context, which needs to hold
2641 * the qgroup_ioctl_lock.
2642 *
2643 * The other one in create_pending_snapshot() where no other qgroup
2644 * code can modify the fs as they all need to either start a new trans
2645 * or hold a trans handler, thus we don't need to hold
2646 * qgroup_ioctl_lock.
2647 * This would avoid long and complex lock chain and make lockdep happy.
2648 */
2649 spin_lock(&fs_info->trans_lock);
2650 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2651 committing = true;
2652 spin_unlock(&fs_info->trans_lock);
2653
2654 if (!committing)
2655 mutex_lock(&fs_info->qgroup_ioctl_lock);
2656 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2657 goto out;
2658
2659 quota_root = fs_info->quota_root;
2660 if (!quota_root) {
2661 ret = -EINVAL;
2662 goto out;
2663 }
2664
2665 if (inherit) {
2666 i_qgroups = (u64 *)(inherit + 1);
2667 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2668 2 * inherit->num_excl_copies;
2669 for (i = 0; i < nums; ++i) {
2670 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2671
2672 /*
2673 * Zero out invalid groups so we can ignore
2674 * them later.
2675 */
2676 if (!srcgroup ||
2677 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2678 *i_qgroups = 0ULL;
2679
2680 ++i_qgroups;
2681 }
2682 }
2683
2684 /*
2685 * create a tracking group for the subvol itself
2686 */
2687 ret = add_qgroup_item(trans, quota_root, objectid);
2688 if (ret)
2689 goto out;
2690
2691 /*
2692 * add qgroup to all inherited groups
2693 */
2694 if (inherit) {
2695 i_qgroups = (u64 *)(inherit + 1);
2696 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2697 if (*i_qgroups == 0)
2698 continue;
2699 ret = add_qgroup_relation_item(trans, objectid,
2700 *i_qgroups);
2701 if (ret && ret != -EEXIST)
2702 goto out;
2703 ret = add_qgroup_relation_item(trans, *i_qgroups,
2704 objectid);
2705 if (ret && ret != -EEXIST)
2706 goto out;
2707 }
2708 ret = 0;
2709 }
2710
2711
2712 spin_lock(&fs_info->qgroup_lock);
2713
2714 dstgroup = add_qgroup_rb(fs_info, objectid);
2715 if (IS_ERR(dstgroup)) {
2716 ret = PTR_ERR(dstgroup);
2717 goto unlock;
2718 }
2719
2720 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2721 dstgroup->lim_flags = inherit->lim.flags;
2722 dstgroup->max_rfer = inherit->lim.max_rfer;
2723 dstgroup->max_excl = inherit->lim.max_excl;
2724 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2725 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2726
2727 ret = update_qgroup_limit_item(trans, dstgroup);
2728 if (ret) {
2729 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2730 btrfs_info(fs_info,
2731 "unable to update quota limit for %llu",
2732 dstgroup->qgroupid);
2733 goto unlock;
2734 }
2735 }
2736
2737 if (srcid) {
2738 srcgroup = find_qgroup_rb(fs_info, srcid);
2739 if (!srcgroup)
2740 goto unlock;
2741
2742 /*
2743 * We call inherit after we clone the root in order to make sure
2744 * our counts don't go crazy, so at this point the only
2745 * difference between the two roots should be the root node.
2746 */
2747 level_size = fs_info->nodesize;
2748 dstgroup->rfer = srcgroup->rfer;
2749 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2750 dstgroup->excl = level_size;
2751 dstgroup->excl_cmpr = level_size;
2752 srcgroup->excl = level_size;
2753 srcgroup->excl_cmpr = level_size;
2754
2755 /* inherit the limit info */
2756 dstgroup->lim_flags = srcgroup->lim_flags;
2757 dstgroup->max_rfer = srcgroup->max_rfer;
2758 dstgroup->max_excl = srcgroup->max_excl;
2759 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2760 dstgroup->rsv_excl = srcgroup->rsv_excl;
2761
2762 qgroup_dirty(fs_info, dstgroup);
2763 qgroup_dirty(fs_info, srcgroup);
2764 }
2765
2766 if (!inherit)
2767 goto unlock;
2768
2769 i_qgroups = (u64 *)(inherit + 1);
2770 for (i = 0; i < inherit->num_qgroups; ++i) {
2771 if (*i_qgroups) {
2772 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2773 if (ret)
2774 goto unlock;
2775 }
2776 ++i_qgroups;
2777 }
2778
2779 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2780 struct btrfs_qgroup *src;
2781 struct btrfs_qgroup *dst;
2782
2783 if (!i_qgroups[0] || !i_qgroups[1])
2784 continue;
2785
2786 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2787 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2788
2789 if (!src || !dst) {
2790 ret = -EINVAL;
2791 goto unlock;
2792 }
2793
2794 dst->rfer = src->rfer - level_size;
2795 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2796 }
2797 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2798 struct btrfs_qgroup *src;
2799 struct btrfs_qgroup *dst;
2800
2801 if (!i_qgroups[0] || !i_qgroups[1])
2802 continue;
2803
2804 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2805 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2806
2807 if (!src || !dst) {
2808 ret = -EINVAL;
2809 goto unlock;
2810 }
2811
2812 dst->excl = src->excl + level_size;
2813 dst->excl_cmpr = src->excl_cmpr + level_size;
2814 }
2815
2816unlock:
2817 spin_unlock(&fs_info->qgroup_lock);
2818out:
2819 if (!committing)
2820 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2821 return ret;
2822}
2823
2824/*
2825 * Two limits to commit transaction in advance.
2826 *
2827 * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2828 * For SIZE, it will be in byte unit as threshold.
2829 */
2830#define QGROUP_FREE_RATIO 32
2831#define QGROUP_FREE_SIZE SZ_32M
2832static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2833 const struct btrfs_qgroup *qg, u64 num_bytes)
2834{
2835 u64 free;
2836 u64 threshold;
2837
2838 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2839 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2840 return false;
2841
2842 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2843 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2844 return false;
2845
2846 /*
2847 * Even if we passed the check, it's better to check if reservation
2848 * for meta_pertrans is pushing us near limit.
2849 * If there is too much pertrans reservation or it's near the limit,
2850 * let's try commit transaction to free some, using transaction_kthread
2851 */
2852 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2853 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2854 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2855 free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2856 threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2857 QGROUP_FREE_SIZE);
2858 } else {
2859 free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2860 threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2861 QGROUP_FREE_SIZE);
2862 }
2863
2864 /*
2865 * Use transaction_kthread to commit transaction, so we no
2866 * longer need to bother nested transaction nor lock context.
2867 */
2868 if (free < threshold)
2869 btrfs_commit_transaction_locksafe(fs_info);
2870 }
2871
2872 return true;
2873}
2874
2875static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2876 enum btrfs_qgroup_rsv_type type)
2877{
2878 struct btrfs_root *quota_root;
2879 struct btrfs_qgroup *qgroup;
2880 struct btrfs_fs_info *fs_info = root->fs_info;
2881 u64 ref_root = root->root_key.objectid;
2882 int ret = 0;
2883 struct ulist_node *unode;
2884 struct ulist_iterator uiter;
2885
2886 if (!is_fstree(ref_root))
2887 return 0;
2888
2889 if (num_bytes == 0)
2890 return 0;
2891
2892 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2893 capable(CAP_SYS_RESOURCE))
2894 enforce = false;
2895
2896 spin_lock(&fs_info->qgroup_lock);
2897 quota_root = fs_info->quota_root;
2898 if (!quota_root)
2899 goto out;
2900
2901 qgroup = find_qgroup_rb(fs_info, ref_root);
2902 if (!qgroup)
2903 goto out;
2904
2905 /*
2906 * in a first step, we check all affected qgroups if any limits would
2907 * be exceeded
2908 */
2909 ulist_reinit(fs_info->qgroup_ulist);
2910 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2911 qgroup_to_aux(qgroup), GFP_ATOMIC);
2912 if (ret < 0)
2913 goto out;
2914 ULIST_ITER_INIT(&uiter);
2915 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2916 struct btrfs_qgroup *qg;
2917 struct btrfs_qgroup_list *glist;
2918
2919 qg = unode_aux_to_qgroup(unode);
2920
2921 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2922 ret = -EDQUOT;
2923 goto out;
2924 }
2925
2926 list_for_each_entry(glist, &qg->groups, next_group) {
2927 ret = ulist_add(fs_info->qgroup_ulist,
2928 glist->group->qgroupid,
2929 qgroup_to_aux(glist->group), GFP_ATOMIC);
2930 if (ret < 0)
2931 goto out;
2932 }
2933 }
2934 ret = 0;
2935 /*
2936 * no limits exceeded, now record the reservation into all qgroups
2937 */
2938 ULIST_ITER_INIT(&uiter);
2939 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2940 struct btrfs_qgroup *qg;
2941
2942 qg = unode_aux_to_qgroup(unode);
2943
2944 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2945 }
2946
2947out:
2948 spin_unlock(&fs_info->qgroup_lock);
2949 return ret;
2950}
2951
2952/*
2953 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2954 * qgroup).
2955 *
2956 * Will handle all higher level qgroup too.
2957 *
2958 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2959 * This special case is only used for META_PERTRANS type.
2960 */
2961void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2962 u64 ref_root, u64 num_bytes,
2963 enum btrfs_qgroup_rsv_type type)
2964{
2965 struct btrfs_root *quota_root;
2966 struct btrfs_qgroup *qgroup;
2967 struct ulist_node *unode;
2968 struct ulist_iterator uiter;
2969 int ret = 0;
2970
2971 if (!is_fstree(ref_root))
2972 return;
2973
2974 if (num_bytes == 0)
2975 return;
2976
2977 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2978 WARN(1, "%s: Invalid type to free", __func__);
2979 return;
2980 }
2981 spin_lock(&fs_info->qgroup_lock);
2982
2983 quota_root = fs_info->quota_root;
2984 if (!quota_root)
2985 goto out;
2986
2987 qgroup = find_qgroup_rb(fs_info, ref_root);
2988 if (!qgroup)
2989 goto out;
2990
2991 if (num_bytes == (u64)-1)
2992 /*
2993 * We're freeing all pertrans rsv, get reserved value from
2994 * level 0 qgroup as real num_bytes to free.
2995 */
2996 num_bytes = qgroup->rsv.values[type];
2997
2998 ulist_reinit(fs_info->qgroup_ulist);
2999 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3000 qgroup_to_aux(qgroup), GFP_ATOMIC);
3001 if (ret < 0)
3002 goto out;
3003 ULIST_ITER_INIT(&uiter);
3004 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3005 struct btrfs_qgroup *qg;
3006 struct btrfs_qgroup_list *glist;
3007
3008 qg = unode_aux_to_qgroup(unode);
3009
3010 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3011
3012 list_for_each_entry(glist, &qg->groups, next_group) {
3013 ret = ulist_add(fs_info->qgroup_ulist,
3014 glist->group->qgroupid,
3015 qgroup_to_aux(glist->group), GFP_ATOMIC);
3016 if (ret < 0)
3017 goto out;
3018 }
3019 }
3020
3021out:
3022 spin_unlock(&fs_info->qgroup_lock);
3023}
3024
3025/*
3026 * Check if the leaf is the last leaf. Which means all node pointers
3027 * are at their last position.
3028 */
3029static bool is_last_leaf(struct btrfs_path *path)
3030{
3031 int i;
3032
3033 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3034 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3035 return false;
3036 }
3037 return true;
3038}
3039
3040/*
3041 * returns < 0 on error, 0 when more leafs are to be scanned.
3042 * returns 1 when done.
3043 */
3044static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3045 struct btrfs_path *path)
3046{
3047 struct btrfs_fs_info *fs_info = trans->fs_info;
3048 struct btrfs_key found;
3049 struct extent_buffer *scratch_leaf = NULL;
3050 struct ulist *roots = NULL;
3051 u64 num_bytes;
3052 bool done;
3053 int slot;
3054 int ret;
3055
3056 mutex_lock(&fs_info->qgroup_rescan_lock);
3057 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3058 &fs_info->qgroup_rescan_progress,
3059 path, 1, 0);
3060
3061 btrfs_debug(fs_info,
3062 "current progress key (%llu %u %llu), search_slot ret %d",
3063 fs_info->qgroup_rescan_progress.objectid,
3064 fs_info->qgroup_rescan_progress.type,
3065 fs_info->qgroup_rescan_progress.offset, ret);
3066
3067 if (ret) {
3068 /*
3069 * The rescan is about to end, we will not be scanning any
3070 * further blocks. We cannot unset the RESCAN flag here, because
3071 * we want to commit the transaction if everything went well.
3072 * To make the live accounting work in this phase, we set our
3073 * scan progress pointer such that every real extent objectid
3074 * will be smaller.
3075 */
3076 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3077 btrfs_release_path(path);
3078 mutex_unlock(&fs_info->qgroup_rescan_lock);
3079 return ret;
3080 }
3081 done = is_last_leaf(path);
3082
3083 btrfs_item_key_to_cpu(path->nodes[0], &found,
3084 btrfs_header_nritems(path->nodes[0]) - 1);
3085 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3086
3087 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3088 if (!scratch_leaf) {
3089 ret = -ENOMEM;
3090 mutex_unlock(&fs_info->qgroup_rescan_lock);
3091 goto out;
3092 }
3093 slot = path->slots[0];
3094 btrfs_release_path(path);
3095 mutex_unlock(&fs_info->qgroup_rescan_lock);
3096
3097 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3098 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3099 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3100 found.type != BTRFS_METADATA_ITEM_KEY)
3101 continue;
3102 if (found.type == BTRFS_METADATA_ITEM_KEY)
3103 num_bytes = fs_info->nodesize;
3104 else
3105 num_bytes = found.offset;
3106
3107 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3108 &roots, false);
3109 if (ret < 0)
3110 goto out;
3111 /* For rescan, just pass old_roots as NULL */
3112 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3113 num_bytes, NULL, roots);
3114 if (ret < 0)
3115 goto out;
3116 }
3117out:
3118 if (scratch_leaf)
3119 free_extent_buffer(scratch_leaf);
3120
3121 if (done && !ret) {
3122 ret = 1;
3123 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3124 }
3125 return ret;
3126}
3127
3128static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3129{
3130 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3131 qgroup_rescan_work);
3132 struct btrfs_path *path;
3133 struct btrfs_trans_handle *trans = NULL;
3134 int err = -ENOMEM;
3135 int ret = 0;
3136
3137 path = btrfs_alloc_path();
3138 if (!path)
3139 goto out;
3140 /*
3141 * Rescan should only search for commit root, and any later difference
3142 * should be recorded by qgroup
3143 */
3144 path->search_commit_root = 1;
3145 path->skip_locking = 1;
3146
3147 err = 0;
3148 while (!err && !btrfs_fs_closing(fs_info)) {
3149 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3150 if (IS_ERR(trans)) {
3151 err = PTR_ERR(trans);
3152 break;
3153 }
3154 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3155 err = -EINTR;
3156 } else {
3157 err = qgroup_rescan_leaf(trans, path);
3158 }
3159 if (err > 0)
3160 btrfs_commit_transaction(trans);
3161 else
3162 btrfs_end_transaction(trans);
3163 }
3164
3165out:
3166 btrfs_free_path(path);
3167
3168 mutex_lock(&fs_info->qgroup_rescan_lock);
3169 if (err > 0 &&
3170 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3171 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3172 } else if (err < 0) {
3173 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3174 }
3175 mutex_unlock(&fs_info->qgroup_rescan_lock);
3176
3177 /*
3178 * only update status, since the previous part has already updated the
3179 * qgroup info.
3180 */
3181 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3182 if (IS_ERR(trans)) {
3183 err = PTR_ERR(trans);
3184 trans = NULL;
3185 btrfs_err(fs_info,
3186 "fail to start transaction for status update: %d",
3187 err);
3188 }
3189
3190 mutex_lock(&fs_info->qgroup_rescan_lock);
3191 if (!btrfs_fs_closing(fs_info))
3192 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3193 if (trans) {
3194 ret = update_qgroup_status_item(trans);
3195 if (ret < 0) {
3196 err = ret;
3197 btrfs_err(fs_info, "fail to update qgroup status: %d",
3198 err);
3199 }
3200 }
3201 fs_info->qgroup_rescan_running = false;
3202 complete_all(&fs_info->qgroup_rescan_completion);
3203 mutex_unlock(&fs_info->qgroup_rescan_lock);
3204
3205 if (!trans)
3206 return;
3207
3208 btrfs_end_transaction(trans);
3209
3210 if (btrfs_fs_closing(fs_info)) {
3211 btrfs_info(fs_info, "qgroup scan paused");
3212 } else if (err >= 0) {
3213 btrfs_info(fs_info, "qgroup scan completed%s",
3214 err > 0 ? " (inconsistency flag cleared)" : "");
3215 } else {
3216 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3217 }
3218}
3219
3220/*
3221 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3222 * memory required for the rescan context.
3223 */
3224static int
3225qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3226 int init_flags)
3227{
3228 int ret = 0;
3229
3230 if (!init_flags) {
3231 /* we're resuming qgroup rescan at mount time */
3232 if (!(fs_info->qgroup_flags &
3233 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3234 btrfs_warn(fs_info,
3235 "qgroup rescan init failed, qgroup is not enabled");
3236 ret = -EINVAL;
3237 } else if (!(fs_info->qgroup_flags &
3238 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3239 btrfs_warn(fs_info,
3240 "qgroup rescan init failed, qgroup rescan is not queued");
3241 ret = -EINVAL;
3242 }
3243
3244 if (ret)
3245 return ret;
3246 }
3247
3248 mutex_lock(&fs_info->qgroup_rescan_lock);
3249 spin_lock(&fs_info->qgroup_lock);
3250
3251 if (init_flags) {
3252 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3253 btrfs_warn(fs_info,
3254 "qgroup rescan is already in progress");
3255 ret = -EINPROGRESS;
3256 } else if (!(fs_info->qgroup_flags &
3257 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3258 btrfs_warn(fs_info,
3259 "qgroup rescan init failed, qgroup is not enabled");
3260 ret = -EINVAL;
3261 }
3262
3263 if (ret) {
3264 spin_unlock(&fs_info->qgroup_lock);
3265 mutex_unlock(&fs_info->qgroup_rescan_lock);
3266 return ret;
3267 }
3268 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3269 }
3270
3271 memset(&fs_info->qgroup_rescan_progress, 0,
3272 sizeof(fs_info->qgroup_rescan_progress));
3273 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3274 init_completion(&fs_info->qgroup_rescan_completion);
3275 fs_info->qgroup_rescan_running = true;
3276
3277 spin_unlock(&fs_info->qgroup_lock);
3278 mutex_unlock(&fs_info->qgroup_rescan_lock);
3279
3280 memset(&fs_info->qgroup_rescan_work, 0,
3281 sizeof(fs_info->qgroup_rescan_work));
3282 btrfs_init_work(&fs_info->qgroup_rescan_work,
3283 btrfs_qgroup_rescan_helper,
3284 btrfs_qgroup_rescan_worker, NULL, NULL);
3285 return 0;
3286}
3287
3288static void
3289qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3290{
3291 struct rb_node *n;
3292 struct btrfs_qgroup *qgroup;
3293
3294 spin_lock(&fs_info->qgroup_lock);
3295 /* clear all current qgroup tracking information */
3296 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3297 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3298 qgroup->rfer = 0;
3299 qgroup->rfer_cmpr = 0;
3300 qgroup->excl = 0;
3301 qgroup->excl_cmpr = 0;
3302 qgroup_dirty(fs_info, qgroup);
3303 }
3304 spin_unlock(&fs_info->qgroup_lock);
3305}
3306
3307int
3308btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3309{
3310 int ret = 0;
3311 struct btrfs_trans_handle *trans;
3312
3313 ret = qgroup_rescan_init(fs_info, 0, 1);
3314 if (ret)
3315 return ret;
3316
3317 /*
3318 * We have set the rescan_progress to 0, which means no more
3319 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3320 * However, btrfs_qgroup_account_ref may be right after its call
3321 * to btrfs_find_all_roots, in which case it would still do the
3322 * accounting.
3323 * To solve this, we're committing the transaction, which will
3324 * ensure we run all delayed refs and only after that, we are
3325 * going to clear all tracking information for a clean start.
3326 */
3327
3328 trans = btrfs_join_transaction(fs_info->fs_root);
3329 if (IS_ERR(trans)) {
3330 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3331 return PTR_ERR(trans);
3332 }
3333 ret = btrfs_commit_transaction(trans);
3334 if (ret) {
3335 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3336 return ret;
3337 }
3338
3339 qgroup_rescan_zero_tracking(fs_info);
3340
3341 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3342 &fs_info->qgroup_rescan_work);
3343
3344 return 0;
3345}
3346
3347int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3348 bool interruptible)
3349{
3350 int running;
3351 int ret = 0;
3352
3353 mutex_lock(&fs_info->qgroup_rescan_lock);
3354 spin_lock(&fs_info->qgroup_lock);
3355 running = fs_info->qgroup_rescan_running;
3356 spin_unlock(&fs_info->qgroup_lock);
3357 mutex_unlock(&fs_info->qgroup_rescan_lock);
3358
3359 if (!running)
3360 return 0;
3361
3362 if (interruptible)
3363 ret = wait_for_completion_interruptible(
3364 &fs_info->qgroup_rescan_completion);
3365 else
3366 wait_for_completion(&fs_info->qgroup_rescan_completion);
3367
3368 return ret;
3369}
3370
3371/*
3372 * this is only called from open_ctree where we're still single threaded, thus
3373 * locking is omitted here.
3374 */
3375void
3376btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3377{
3378 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3379 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3380 &fs_info->qgroup_rescan_work);
3381}
3382
3383/*
3384 * Reserve qgroup space for range [start, start + len).
3385 *
3386 * This function will either reserve space from related qgroups or doing
3387 * nothing if the range is already reserved.
3388 *
3389 * Return 0 for successful reserve
3390 * Return <0 for error (including -EQUOT)
3391 *
3392 * NOTE: this function may sleep for memory allocation.
3393 * if btrfs_qgroup_reserve_data() is called multiple times with
3394 * same @reserved, caller must ensure when error happens it's OK
3395 * to free *ALL* reserved space.
3396 */
3397int btrfs_qgroup_reserve_data(struct inode *inode,
3398 struct extent_changeset **reserved_ret, u64 start,
3399 u64 len)
3400{
3401 struct btrfs_root *root = BTRFS_I(inode)->root;
3402 struct ulist_node *unode;
3403 struct ulist_iterator uiter;
3404 struct extent_changeset *reserved;
3405 u64 orig_reserved;
3406 u64 to_reserve;
3407 int ret;
3408
3409 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3410 !is_fstree(root->root_key.objectid) || len == 0)
3411 return 0;
3412
3413 /* @reserved parameter is mandatory for qgroup */
3414 if (WARN_ON(!reserved_ret))
3415 return -EINVAL;
3416 if (!*reserved_ret) {
3417 *reserved_ret = extent_changeset_alloc();
3418 if (!*reserved_ret)
3419 return -ENOMEM;
3420 }
3421 reserved = *reserved_ret;
3422 /* Record already reserved space */
3423 orig_reserved = reserved->bytes_changed;
3424 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3425 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3426
3427 /* Newly reserved space */
3428 to_reserve = reserved->bytes_changed - orig_reserved;
3429 trace_btrfs_qgroup_reserve_data(inode, start, len,
3430 to_reserve, QGROUP_RESERVE);
3431 if (ret < 0)
3432 goto cleanup;
3433 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3434 if (ret < 0)
3435 goto cleanup;
3436
3437 return ret;
3438
3439cleanup:
3440 /* cleanup *ALL* already reserved ranges */
3441 ULIST_ITER_INIT(&uiter);
3442 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3443 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3444 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3445 /* Also free data bytes of already reserved one */
3446 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3447 orig_reserved, BTRFS_QGROUP_RSV_DATA);
3448 extent_changeset_release(reserved);
3449 return ret;
3450}
3451
3452/* Free ranges specified by @reserved, normally in error path */
3453static int qgroup_free_reserved_data(struct inode *inode,
3454 struct extent_changeset *reserved, u64 start, u64 len)
3455{
3456 struct btrfs_root *root = BTRFS_I(inode)->root;
3457 struct ulist_node *unode;
3458 struct ulist_iterator uiter;
3459 struct extent_changeset changeset;
3460 int freed = 0;
3461 int ret;
3462
3463 extent_changeset_init(&changeset);
3464 len = round_up(start + len, root->fs_info->sectorsize);
3465 start = round_down(start, root->fs_info->sectorsize);
3466
3467 ULIST_ITER_INIT(&uiter);
3468 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3469 u64 range_start = unode->val;
3470 /* unode->aux is the inclusive end */
3471 u64 range_len = unode->aux - range_start + 1;
3472 u64 free_start;
3473 u64 free_len;
3474
3475 extent_changeset_release(&changeset);
3476
3477 /* Only free range in range [start, start + len) */
3478 if (range_start >= start + len ||
3479 range_start + range_len <= start)
3480 continue;
3481 free_start = max(range_start, start);
3482 free_len = min(start + len, range_start + range_len) -
3483 free_start;
3484 /*
3485 * TODO: To also modify reserved->ranges_reserved to reflect
3486 * the modification.
3487 *
3488 * However as long as we free qgroup reserved according to
3489 * EXTENT_QGROUP_RESERVED, we won't double free.
3490 * So not need to rush.
3491 */
3492 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree,
3493 free_start, free_start + free_len - 1,
3494 EXTENT_QGROUP_RESERVED, &changeset);
3495 if (ret < 0)
3496 goto out;
3497 freed += changeset.bytes_changed;
3498 }
3499 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3500 BTRFS_QGROUP_RSV_DATA);
3501 ret = freed;
3502out:
3503 extent_changeset_release(&changeset);
3504 return ret;
3505}
3506
3507static int __btrfs_qgroup_release_data(struct inode *inode,
3508 struct extent_changeset *reserved, u64 start, u64 len,
3509 int free)
3510{
3511 struct extent_changeset changeset;
3512 int trace_op = QGROUP_RELEASE;
3513 int ret;
3514
3515 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3516 &BTRFS_I(inode)->root->fs_info->flags))
3517 return 0;
3518
3519 /* In release case, we shouldn't have @reserved */
3520 WARN_ON(!free && reserved);
3521 if (free && reserved)
3522 return qgroup_free_reserved_data(inode, reserved, start, len);
3523 extent_changeset_init(&changeset);
3524 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3525 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3526 if (ret < 0)
3527 goto out;
3528
3529 if (free)
3530 trace_op = QGROUP_FREE;
3531 trace_btrfs_qgroup_release_data(inode, start, len,
3532 changeset.bytes_changed, trace_op);
3533 if (free)
3534 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3535 BTRFS_I(inode)->root->root_key.objectid,
3536 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3537 ret = changeset.bytes_changed;
3538out:
3539 extent_changeset_release(&changeset);
3540 return ret;
3541}
3542
3543/*
3544 * Free a reserved space range from io_tree and related qgroups
3545 *
3546 * Should be called when a range of pages get invalidated before reaching disk.
3547 * Or for error cleanup case.
3548 * if @reserved is given, only reserved range in [@start, @start + @len) will
3549 * be freed.
3550 *
3551 * For data written to disk, use btrfs_qgroup_release_data().
3552 *
3553 * NOTE: This function may sleep for memory allocation.
3554 */
3555int btrfs_qgroup_free_data(struct inode *inode,
3556 struct extent_changeset *reserved, u64 start, u64 len)
3557{
3558 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3559}
3560
3561/*
3562 * Release a reserved space range from io_tree only.
3563 *
3564 * Should be called when a range of pages get written to disk and corresponding
3565 * FILE_EXTENT is inserted into corresponding root.
3566 *
3567 * Since new qgroup accounting framework will only update qgroup numbers at
3568 * commit_transaction() time, its reserved space shouldn't be freed from
3569 * related qgroups.
3570 *
3571 * But we should release the range from io_tree, to allow further write to be
3572 * COWed.
3573 *
3574 * NOTE: This function may sleep for memory allocation.
3575 */
3576int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3577{
3578 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3579}
3580
3581static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3582 enum btrfs_qgroup_rsv_type type)
3583{
3584 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3585 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3586 return;
3587 if (num_bytes == 0)
3588 return;
3589
3590 spin_lock(&root->qgroup_meta_rsv_lock);
3591 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3592 root->qgroup_meta_rsv_prealloc += num_bytes;
3593 else
3594 root->qgroup_meta_rsv_pertrans += num_bytes;
3595 spin_unlock(&root->qgroup_meta_rsv_lock);
3596}
3597
3598static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3599 enum btrfs_qgroup_rsv_type type)
3600{
3601 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3602 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3603 return 0;
3604 if (num_bytes == 0)
3605 return 0;
3606
3607 spin_lock(&root->qgroup_meta_rsv_lock);
3608 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3609 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3610 num_bytes);
3611 root->qgroup_meta_rsv_prealloc -= num_bytes;
3612 } else {
3613 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3614 num_bytes);
3615 root->qgroup_meta_rsv_pertrans -= num_bytes;
3616 }
3617 spin_unlock(&root->qgroup_meta_rsv_lock);
3618 return num_bytes;
3619}
3620
3621int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3622 enum btrfs_qgroup_rsv_type type, bool enforce)
3623{
3624 struct btrfs_fs_info *fs_info = root->fs_info;
3625 int ret;
3626
3627 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3628 !is_fstree(root->root_key.objectid) || num_bytes == 0)
3629 return 0;
3630
3631 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3632 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3633 ret = qgroup_reserve(root, num_bytes, enforce, type);
3634 if (ret < 0)
3635 return ret;
3636 /*
3637 * Record what we have reserved into root.
3638 *
3639 * To avoid quota disabled->enabled underflow.
3640 * In that case, we may try to free space we haven't reserved
3641 * (since quota was disabled), so record what we reserved into root.
3642 * And ensure later release won't underflow this number.
3643 */
3644 add_root_meta_rsv(root, num_bytes, type);
3645 return ret;
3646}
3647
3648void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3649{
3650 struct btrfs_fs_info *fs_info = root->fs_info;
3651
3652 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3653 !is_fstree(root->root_key.objectid))
3654 return;
3655
3656 /* TODO: Update trace point to handle such free */
3657 trace_qgroup_meta_free_all_pertrans(root);
3658 /* Special value -1 means to free all reserved space */
3659 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3660 BTRFS_QGROUP_RSV_META_PERTRANS);
3661}
3662
3663void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3664 enum btrfs_qgroup_rsv_type type)
3665{
3666 struct btrfs_fs_info *fs_info = root->fs_info;
3667
3668 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3669 !is_fstree(root->root_key.objectid))
3670 return;
3671
3672 /*
3673 * reservation for META_PREALLOC can happen before quota is enabled,
3674 * which can lead to underflow.
3675 * Here ensure we will only free what we really have reserved.
3676 */
3677 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3678 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3679 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3680 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3681 num_bytes, type);
3682}
3683
3684static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3685 int num_bytes)
3686{
3687 struct btrfs_root *quota_root = fs_info->quota_root;
3688 struct btrfs_qgroup *qgroup;
3689 struct ulist_node *unode;
3690 struct ulist_iterator uiter;
3691 int ret = 0;
3692
3693 if (num_bytes == 0)
3694 return;
3695 if (!quota_root)
3696 return;
3697
3698 spin_lock(&fs_info->qgroup_lock);
3699 qgroup = find_qgroup_rb(fs_info, ref_root);
3700 if (!qgroup)
3701 goto out;
3702 ulist_reinit(fs_info->qgroup_ulist);
3703 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3704 qgroup_to_aux(qgroup), GFP_ATOMIC);
3705 if (ret < 0)
3706 goto out;
3707 ULIST_ITER_INIT(&uiter);
3708 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3709 struct btrfs_qgroup *qg;
3710 struct btrfs_qgroup_list *glist;
3711
3712 qg = unode_aux_to_qgroup(unode);
3713
3714 qgroup_rsv_release(fs_info, qg, num_bytes,
3715 BTRFS_QGROUP_RSV_META_PREALLOC);
3716 qgroup_rsv_add(fs_info, qg, num_bytes,
3717 BTRFS_QGROUP_RSV_META_PERTRANS);
3718 list_for_each_entry(glist, &qg->groups, next_group) {
3719 ret = ulist_add(fs_info->qgroup_ulist,
3720 glist->group->qgroupid,
3721 qgroup_to_aux(glist->group), GFP_ATOMIC);
3722 if (ret < 0)
3723 goto out;
3724 }
3725 }
3726out:
3727 spin_unlock(&fs_info->qgroup_lock);
3728}
3729
3730void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3731{
3732 struct btrfs_fs_info *fs_info = root->fs_info;
3733
3734 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3735 !is_fstree(root->root_key.objectid))
3736 return;
3737 /* Same as btrfs_qgroup_free_meta_prealloc() */
3738 num_bytes = sub_root_meta_rsv(root, num_bytes,
3739 BTRFS_QGROUP_RSV_META_PREALLOC);
3740 trace_qgroup_meta_convert(root, num_bytes);
3741 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3742}
3743
3744/*
3745 * Check qgroup reserved space leaking, normally at destroy inode
3746 * time
3747 */
3748void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3749{
3750 struct extent_changeset changeset;
3751 struct ulist_node *unode;
3752 struct ulist_iterator iter;
3753 int ret;
3754
3755 extent_changeset_init(&changeset);
3756 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3757 EXTENT_QGROUP_RESERVED, &changeset);
3758
3759 WARN_ON(ret < 0);
3760 if (WARN_ON(changeset.bytes_changed)) {
3761 ULIST_ITER_INIT(&iter);
3762 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3763 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3764 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3765 inode->i_ino, unode->val, unode->aux);
3766 }
3767 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3768 BTRFS_I(inode)->root->root_key.objectid,
3769 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3770
3771 }
3772 extent_changeset_release(&changeset);
3773}
3774
3775void btrfs_qgroup_init_swapped_blocks(
3776 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3777{
3778 int i;
3779
3780 spin_lock_init(&swapped_blocks->lock);
3781 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3782 swapped_blocks->blocks[i] = RB_ROOT;
3783 swapped_blocks->swapped = false;
3784}
3785
3786/*
3787 * Delete all swapped blocks record of @root.
3788 * Every record here means we skipped a full subtree scan for qgroup.
3789 *
3790 * Gets called when committing one transaction.
3791 */
3792void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3793{
3794 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3795 int i;
3796
3797 swapped_blocks = &root->swapped_blocks;
3798
3799 spin_lock(&swapped_blocks->lock);
3800 if (!swapped_blocks->swapped)
3801 goto out;
3802 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3803 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3804 struct btrfs_qgroup_swapped_block *entry;
3805 struct btrfs_qgroup_swapped_block *next;
3806
3807 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3808 node)
3809 kfree(entry);
3810 swapped_blocks->blocks[i] = RB_ROOT;
3811 }
3812 swapped_blocks->swapped = false;
3813out:
3814 spin_unlock(&swapped_blocks->lock);
3815}
3816
3817/*
3818 * Add subtree roots record into @subvol_root.
3819 *
3820 * @subvol_root: tree root of the subvolume tree get swapped
3821 * @bg: block group under balance
3822 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3823 * @reloc_parent/slot: pointer to the subtree root in reloc tree
3824 * BOTH POINTERS ARE BEFORE TREE SWAP
3825 * @last_snapshot: last snapshot generation of the subvolume tree
3826 */
3827int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3828 struct btrfs_root *subvol_root,
3829 struct btrfs_block_group_cache *bg,
3830 struct extent_buffer *subvol_parent, int subvol_slot,
3831 struct extent_buffer *reloc_parent, int reloc_slot,
3832 u64 last_snapshot)
3833{
3834 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3835 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3836 struct btrfs_qgroup_swapped_block *block;
3837 struct rb_node **cur;
3838 struct rb_node *parent = NULL;
3839 int level = btrfs_header_level(subvol_parent) - 1;
3840 int ret = 0;
3841
3842 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3843 return 0;
3844
3845 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3846 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3847 btrfs_err_rl(fs_info,
3848 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3849 __func__,
3850 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3851 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3852 return -EUCLEAN;
3853 }
3854
3855 block = kmalloc(sizeof(*block), GFP_NOFS);
3856 if (!block) {
3857 ret = -ENOMEM;
3858 goto out;
3859 }
3860
3861 /*
3862 * @reloc_parent/slot is still before swap, while @block is going to
3863 * record the bytenr after swap, so we do the swap here.
3864 */
3865 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3866 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3867 reloc_slot);
3868 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3869 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3870 subvol_slot);
3871 block->last_snapshot = last_snapshot;
3872 block->level = level;
3873
3874 /*
3875 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3876 * no one else can modify tree blocks thus we qgroup will not change
3877 * no matter the value of trace_leaf.
3878 */
3879 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
3880 block->trace_leaf = true;
3881 else
3882 block->trace_leaf = false;
3883 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3884
3885 /* Insert @block into @blocks */
3886 spin_lock(&blocks->lock);
3887 cur = &blocks->blocks[level].rb_node;
3888 while (*cur) {
3889 struct btrfs_qgroup_swapped_block *entry;
3890
3891 parent = *cur;
3892 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3893 node);
3894
3895 if (entry->subvol_bytenr < block->subvol_bytenr) {
3896 cur = &(*cur)->rb_left;
3897 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3898 cur = &(*cur)->rb_right;
3899 } else {
3900 if (entry->subvol_generation !=
3901 block->subvol_generation ||
3902 entry->reloc_bytenr != block->reloc_bytenr ||
3903 entry->reloc_generation !=
3904 block->reloc_generation) {
3905 /*
3906 * Duplicated but mismatch entry found.
3907 * Shouldn't happen.
3908 *
3909 * Marking qgroup inconsistent should be enough
3910 * for end users.
3911 */
3912 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3913 ret = -EEXIST;
3914 }
3915 kfree(block);
3916 goto out_unlock;
3917 }
3918 }
3919 rb_link_node(&block->node, parent, cur);
3920 rb_insert_color(&block->node, &blocks->blocks[level]);
3921 blocks->swapped = true;
3922out_unlock:
3923 spin_unlock(&blocks->lock);
3924out:
3925 if (ret < 0)
3926 fs_info->qgroup_flags |=
3927 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3928 return ret;
3929}
3930
3931/*
3932 * Check if the tree block is a subtree root, and if so do the needed
3933 * delayed subtree trace for qgroup.
3934 *
3935 * This is called during btrfs_cow_block().
3936 */
3937int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3938 struct btrfs_root *root,
3939 struct extent_buffer *subvol_eb)
3940{
3941 struct btrfs_fs_info *fs_info = root->fs_info;
3942 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3943 struct btrfs_qgroup_swapped_block *block;
3944 struct extent_buffer *reloc_eb = NULL;
3945 struct rb_node *node;
3946 bool found = false;
3947 bool swapped = false;
3948 int level = btrfs_header_level(subvol_eb);
3949 int ret = 0;
3950 int i;
3951
3952 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3953 return 0;
3954 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3955 return 0;
3956
3957 spin_lock(&blocks->lock);
3958 if (!blocks->swapped) {
3959 spin_unlock(&blocks->lock);
3960 return 0;
3961 }
3962 node = blocks->blocks[level].rb_node;
3963
3964 while (node) {
3965 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3966 if (block->subvol_bytenr < subvol_eb->start) {
3967 node = node->rb_left;
3968 } else if (block->subvol_bytenr > subvol_eb->start) {
3969 node = node->rb_right;
3970 } else {
3971 found = true;
3972 break;
3973 }
3974 }
3975 if (!found) {
3976 spin_unlock(&blocks->lock);
3977 goto out;
3978 }
3979 /* Found one, remove it from @blocks first and update blocks->swapped */
3980 rb_erase(&block->node, &blocks->blocks[level]);
3981 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3982 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3983 swapped = true;
3984 break;
3985 }
3986 }
3987 blocks->swapped = swapped;
3988 spin_unlock(&blocks->lock);
3989
3990 /* Read out reloc subtree root */
3991 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3992 block->reloc_generation, block->level,
3993 &block->first_key);
3994 if (IS_ERR(reloc_eb)) {
3995 ret = PTR_ERR(reloc_eb);
3996 reloc_eb = NULL;
3997 goto free_out;
3998 }
3999 if (!extent_buffer_uptodate(reloc_eb)) {
4000 ret = -EIO;
4001 goto free_out;
4002 }
4003
4004 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4005 block->last_snapshot, block->trace_leaf);
4006free_out:
4007 kfree(block);
4008 free_extent_buffer(reloc_eb);
4009out:
4010 if (ret < 0) {
4011 btrfs_err_rl(fs_info,
4012 "failed to account subtree at bytenr %llu: %d",
4013 subvol_eb->start, ret);
4014 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4015 }
4016 return ret;
4017}