?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs: qgroup: Drop root parameter from update_qgroup_info_item
[linux/fpc-iii.git] / fs / btrfs / qgroup.c
blobf4b25e0a32f8b04829fe885b169f2f8812c611ab
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
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 fuer 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
44 static 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
56 static 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
68 static 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
76 static 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
94 static 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
104 static 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
114 static 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
122 static 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
130 static 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
137 static 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 */
147 struct 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
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 {
156 return (u64)(uintptr_t)qg;
157 }
158
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 {
161 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162 }
163
164 static int
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 int init_flags);
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168
169 /* must be called with qgroup_ioctl_lock held */
170 static 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 */
189 static 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
223 static 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 */
247 static 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 */
260 static 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 */
285 static 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
309 int 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 */
327 int 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 }
441 next1:
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;
483 next2:
484 ret = btrfs_next_item(quota_root, path);
485 if (ret < 0)
486 goto out;
487 if (ret)
488 break;
489 }
490 out:
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 */
514 void 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 umounting
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
533 static 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
557 static 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);
583 out:
584 btrfs_free_path(path);
585 return ret;
586 }
587
588 static 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;
651 out:
652 btrfs_free_path(path);
653 return ret;
654 }
655
656 static 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
697 out:
698 btrfs_free_path(path);
699 return ret;
700 }
701
702 static 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
739 out:
740 btrfs_free_path(path);
741 return ret;
742 }
743
744 static 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
785 out:
786 btrfs_free_path(path);
787 return ret;
788 }
789
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
791 struct btrfs_fs_info *fs_info,
792 struct btrfs_root *root)
793 {
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, 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
826 out:
827 btrfs_free_path(path);
828 return ret;
829 }
830
831 /*
832 * called with qgroup_lock held
833 */
834 static 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;
874 out:
875 btrfs_free_path(path);
876 return ret;
877 }
878
879 int 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 /*
898 * 1 for quota root item
899 * 1 for BTRFS_QGROUP_STATUS item
900 *
901 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
902 * per subvolume. However those are not currently reserved since it
903 * would be a lot of overkill.
904 */
905 trans = btrfs_start_transaction(tree_root, 2);
906 if (IS_ERR(trans)) {
907 ret = PTR_ERR(trans);
908 trans = NULL;
909 goto out;
910 }
911
912 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
913 if (!fs_info->qgroup_ulist) {
914 ret = -ENOMEM;
915 btrfs_abort_transaction(trans, ret);
916 goto out;
917 }
918
919 /*
920 * initially create the quota tree
921 */
922 quota_root = btrfs_create_tree(trans, fs_info,
923 BTRFS_QUOTA_TREE_OBJECTID);
924 if (IS_ERR(quota_root)) {
925 ret = PTR_ERR(quota_root);
926 btrfs_abort_transaction(trans, ret);
927 goto out;
928 }
929
930 path = btrfs_alloc_path();
931 if (!path) {
932 ret = -ENOMEM;
933 btrfs_abort_transaction(trans, ret);
934 goto out_free_root;
935 }
936
937 key.objectid = 0;
938 key.type = BTRFS_QGROUP_STATUS_KEY;
939 key.offset = 0;
940
941 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
942 sizeof(*ptr));
943 if (ret) {
944 btrfs_abort_transaction(trans, ret);
945 goto out_free_path;
946 }
947
948 leaf = path->nodes[0];
949 ptr = btrfs_item_ptr(leaf, path->slots[0],
950 struct btrfs_qgroup_status_item);
951 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
952 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
953 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
954 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
955 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
956 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
957
958 btrfs_mark_buffer_dirty(leaf);
959
960 key.objectid = 0;
961 key.type = BTRFS_ROOT_REF_KEY;
962 key.offset = 0;
963
964 btrfs_release_path(path);
965 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
966 if (ret > 0)
967 goto out_add_root;
968 if (ret < 0) {
969 btrfs_abort_transaction(trans, ret);
970 goto out_free_path;
971 }
972
973 while (1) {
974 slot = path->slots[0];
975 leaf = path->nodes[0];
976 btrfs_item_key_to_cpu(leaf, &found_key, slot);
977
978 if (found_key.type == BTRFS_ROOT_REF_KEY) {
979 ret = add_qgroup_item(trans, quota_root,
980 found_key.offset);
981 if (ret) {
982 btrfs_abort_transaction(trans, ret);
983 goto out_free_path;
984 }
985
986 qgroup = add_qgroup_rb(fs_info, found_key.offset);
987 if (IS_ERR(qgroup)) {
988 ret = PTR_ERR(qgroup);
989 btrfs_abort_transaction(trans, ret);
990 goto out_free_path;
991 }
992 }
993 ret = btrfs_next_item(tree_root, path);
994 if (ret < 0) {
995 btrfs_abort_transaction(trans, ret);
996 goto out_free_path;
997 }
998 if (ret)
999 break;
1000 }
1001
1002 out_add_root:
1003 btrfs_release_path(path);
1004 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1005 if (ret) {
1006 btrfs_abort_transaction(trans, ret);
1007 goto out_free_path;
1008 }
1009
1010 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011 if (IS_ERR(qgroup)) {
1012 ret = PTR_ERR(qgroup);
1013 btrfs_abort_transaction(trans, ret);
1014 goto out_free_path;
1015 }
1016 spin_lock(&fs_info->qgroup_lock);
1017 fs_info->quota_root = quota_root;
1018 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1019 spin_unlock(&fs_info->qgroup_lock);
1020
1021 ret = btrfs_commit_transaction(trans);
1022 if (ret) {
1023 trans = NULL;
1024 goto out_free_path;
1025 }
1026
1027 ret = qgroup_rescan_init(fs_info, 0, 1);
1028 if (!ret) {
1029 qgroup_rescan_zero_tracking(fs_info);
1030 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1031 &fs_info->qgroup_rescan_work);
1032 }
1033
1034 out_free_path:
1035 btrfs_free_path(path);
1036 out_free_root:
1037 if (ret) {
1038 free_extent_buffer(quota_root->node);
1039 free_extent_buffer(quota_root->commit_root);
1040 kfree(quota_root);
1041 }
1042 out:
1043 if (ret) {
1044 ulist_free(fs_info->qgroup_ulist);
1045 fs_info->qgroup_ulist = NULL;
1046 if (trans)
1047 btrfs_end_transaction(trans);
1048 }
1049 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1050 return ret;
1051 }
1052
1053 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1054 {
1055 struct btrfs_root *quota_root;
1056 struct btrfs_trans_handle *trans = NULL;
1057 int ret = 0;
1058
1059 mutex_lock(&fs_info->qgroup_ioctl_lock);
1060 if (!fs_info->quota_root)
1061 goto out;
1062
1063 /*
1064 * 1 For the root item
1065 *
1066 * We should also reserve enough items for the quota tree deletion in
1067 * btrfs_clean_quota_tree but this is not done.
1068 */
1069 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1070 if (IS_ERR(trans)) {
1071 ret = PTR_ERR(trans);
1072 goto out;
1073 }
1074
1075 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1076 btrfs_qgroup_wait_for_completion(fs_info, false);
1077 spin_lock(&fs_info->qgroup_lock);
1078 quota_root = fs_info->quota_root;
1079 fs_info->quota_root = NULL;
1080 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1081 spin_unlock(&fs_info->qgroup_lock);
1082
1083 btrfs_free_qgroup_config(fs_info);
1084
1085 ret = btrfs_clean_quota_tree(trans, quota_root);
1086 if (ret) {
1087 btrfs_abort_transaction(trans, ret);
1088 goto end_trans;
1089 }
1090
1091 ret = btrfs_del_root(trans, fs_info, &quota_root->root_key);
1092 if (ret) {
1093 btrfs_abort_transaction(trans, ret);
1094 goto end_trans;
1095 }
1096
1097 list_del(&quota_root->dirty_list);
1098
1099 btrfs_tree_lock(quota_root->node);
1100 clean_tree_block(fs_info, quota_root->node);
1101 btrfs_tree_unlock(quota_root->node);
1102 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1103
1104 free_extent_buffer(quota_root->node);
1105 free_extent_buffer(quota_root->commit_root);
1106 kfree(quota_root);
1107
1108 end_trans:
1109 ret = btrfs_end_transaction(trans);
1110 out:
1111 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1112 return ret;
1113 }
1114
1115 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1116 struct btrfs_qgroup *qgroup)
1117 {
1118 if (list_empty(&qgroup->dirty))
1119 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1120 }
1121
1122 /*
1123 * The easy accounting, we're updating qgroup relationship whose child qgroup
1124 * only has exclusive extents.
1125 *
1126 * In this case, all exclsuive extents will also be exlusive for parent, so
1127 * excl/rfer just get added/removed.
1128 *
1129 * So is qgroup reservation space, which should also be added/removed to
1130 * parent.
1131 * Or when child tries to release reservation space, parent will underflow its
1132 * reservation (for relationship adding case).
1133 *
1134 * Caller should hold fs_info->qgroup_lock.
1135 */
1136 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1137 struct ulist *tmp, u64 ref_root,
1138 struct btrfs_qgroup *src, int sign)
1139 {
1140 struct btrfs_qgroup *qgroup;
1141 struct btrfs_qgroup_list *glist;
1142 struct ulist_node *unode;
1143 struct ulist_iterator uiter;
1144 u64 num_bytes = src->excl;
1145 int ret = 0;
1146
1147 qgroup = find_qgroup_rb(fs_info, ref_root);
1148 if (!qgroup)
1149 goto out;
1150
1151 qgroup->rfer += sign * num_bytes;
1152 qgroup->rfer_cmpr += sign * num_bytes;
1153
1154 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1155 qgroup->excl += sign * num_bytes;
1156 qgroup->excl_cmpr += sign * num_bytes;
1157
1158 if (sign > 0)
1159 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1160 else
1161 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1162
1163 qgroup_dirty(fs_info, qgroup);
1164
1165 /* Get all of the parent groups that contain this qgroup */
1166 list_for_each_entry(glist, &qgroup->groups, next_group) {
1167 ret = ulist_add(tmp, glist->group->qgroupid,
1168 qgroup_to_aux(glist->group), GFP_ATOMIC);
1169 if (ret < 0)
1170 goto out;
1171 }
1172
1173 /* Iterate all of the parents and adjust their reference counts */
1174 ULIST_ITER_INIT(&uiter);
1175 while ((unode = ulist_next(tmp, &uiter))) {
1176 qgroup = unode_aux_to_qgroup(unode);
1177 qgroup->rfer += sign * num_bytes;
1178 qgroup->rfer_cmpr += sign * num_bytes;
1179 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1180 qgroup->excl += sign * num_bytes;
1181 if (sign > 0)
1182 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1183 else
1184 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1185 qgroup->excl_cmpr += sign * num_bytes;
1186 qgroup_dirty(fs_info, qgroup);
1187
1188 /* Add any parents of the parents */
1189 list_for_each_entry(glist, &qgroup->groups, next_group) {
1190 ret = ulist_add(tmp, glist->group->qgroupid,
1191 qgroup_to_aux(glist->group), GFP_ATOMIC);
1192 if (ret < 0)
1193 goto out;
1194 }
1195 }
1196 ret = 0;
1197 out:
1198 return ret;
1199 }
1200
1201
1202 /*
1203 * Quick path for updating qgroup with only excl refs.
1204 *
1205 * In that case, just update all parent will be enough.
1206 * Or we needs to do a full rescan.
1207 * Caller should also hold fs_info->qgroup_lock.
1208 *
1209 * Return 0 for quick update, return >0 for need to full rescan
1210 * and mark INCONSISTENT flag.
1211 * Return < 0 for other error.
1212 */
1213 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1214 struct ulist *tmp, u64 src, u64 dst,
1215 int sign)
1216 {
1217 struct btrfs_qgroup *qgroup;
1218 int ret = 1;
1219 int err = 0;
1220
1221 qgroup = find_qgroup_rb(fs_info, src);
1222 if (!qgroup)
1223 goto out;
1224 if (qgroup->excl == qgroup->rfer) {
1225 ret = 0;
1226 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1227 qgroup, sign);
1228 if (err < 0) {
1229 ret = err;
1230 goto out;
1231 }
1232 }
1233 out:
1234 if (ret)
1235 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1236 return ret;
1237 }
1238
1239 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1240 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1241 {
1242 struct btrfs_root *quota_root;
1243 struct btrfs_qgroup *parent;
1244 struct btrfs_qgroup *member;
1245 struct btrfs_qgroup_list *list;
1246 struct ulist *tmp;
1247 int ret = 0;
1248
1249 /* Check the level of src and dst first */
1250 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1251 return -EINVAL;
1252
1253 tmp = ulist_alloc(GFP_KERNEL);
1254 if (!tmp)
1255 return -ENOMEM;
1256
1257 mutex_lock(&fs_info->qgroup_ioctl_lock);
1258 quota_root = fs_info->quota_root;
1259 if (!quota_root) {
1260 ret = -EINVAL;
1261 goto out;
1262 }
1263 member = find_qgroup_rb(fs_info, src);
1264 parent = find_qgroup_rb(fs_info, dst);
1265 if (!member || !parent) {
1266 ret = -EINVAL;
1267 goto out;
1268 }
1269
1270 /* check if such qgroup relation exist firstly */
1271 list_for_each_entry(list, &member->groups, next_group) {
1272 if (list->group == parent) {
1273 ret = -EEXIST;
1274 goto out;
1275 }
1276 }
1277
1278 ret = add_qgroup_relation_item(trans, src, dst);
1279 if (ret)
1280 goto out;
1281
1282 ret = add_qgroup_relation_item(trans, dst, src);
1283 if (ret) {
1284 del_qgroup_relation_item(trans, src, dst);
1285 goto out;
1286 }
1287
1288 spin_lock(&fs_info->qgroup_lock);
1289 ret = add_relation_rb(fs_info, src, dst);
1290 if (ret < 0) {
1291 spin_unlock(&fs_info->qgroup_lock);
1292 goto out;
1293 }
1294 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1295 spin_unlock(&fs_info->qgroup_lock);
1296 out:
1297 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1298 ulist_free(tmp);
1299 return ret;
1300 }
1301
1302 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1303 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1304 {
1305 struct btrfs_root *quota_root;
1306 struct btrfs_qgroup *parent;
1307 struct btrfs_qgroup *member;
1308 struct btrfs_qgroup_list *list;
1309 struct ulist *tmp;
1310 int ret = 0;
1311 int err;
1312
1313 tmp = ulist_alloc(GFP_KERNEL);
1314 if (!tmp)
1315 return -ENOMEM;
1316
1317 quota_root = fs_info->quota_root;
1318 if (!quota_root) {
1319 ret = -EINVAL;
1320 goto out;
1321 }
1322
1323 member = find_qgroup_rb(fs_info, src);
1324 parent = find_qgroup_rb(fs_info, dst);
1325 if (!member || !parent) {
1326 ret = -EINVAL;
1327 goto out;
1328 }
1329
1330 /* check if such qgroup relation exist firstly */
1331 list_for_each_entry(list, &member->groups, next_group) {
1332 if (list->group == parent)
1333 goto exist;
1334 }
1335 ret = -ENOENT;
1336 goto out;
1337 exist:
1338 ret = del_qgroup_relation_item(trans, src, dst);
1339 err = del_qgroup_relation_item(trans, dst, src);
1340 if (err && !ret)
1341 ret = err;
1342
1343 spin_lock(&fs_info->qgroup_lock);
1344 del_relation_rb(fs_info, src, dst);
1345 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1346 spin_unlock(&fs_info->qgroup_lock);
1347 out:
1348 ulist_free(tmp);
1349 return ret;
1350 }
1351
1352 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1353 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1354 {
1355 int ret = 0;
1356
1357 mutex_lock(&fs_info->qgroup_ioctl_lock);
1358 ret = __del_qgroup_relation(trans, fs_info, src, dst);
1359 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1360
1361 return ret;
1362 }
1363
1364 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1365 struct btrfs_fs_info *fs_info, u64 qgroupid)
1366 {
1367 struct btrfs_root *quota_root;
1368 struct btrfs_qgroup *qgroup;
1369 int ret = 0;
1370
1371 mutex_lock(&fs_info->qgroup_ioctl_lock);
1372 quota_root = fs_info->quota_root;
1373 if (!quota_root) {
1374 ret = -EINVAL;
1375 goto out;
1376 }
1377 qgroup = find_qgroup_rb(fs_info, qgroupid);
1378 if (qgroup) {
1379 ret = -EEXIST;
1380 goto out;
1381 }
1382
1383 ret = add_qgroup_item(trans, quota_root, qgroupid);
1384 if (ret)
1385 goto out;
1386
1387 spin_lock(&fs_info->qgroup_lock);
1388 qgroup = add_qgroup_rb(fs_info, qgroupid);
1389 spin_unlock(&fs_info->qgroup_lock);
1390
1391 if (IS_ERR(qgroup))
1392 ret = PTR_ERR(qgroup);
1393 out:
1394 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1395 return ret;
1396 }
1397
1398 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1399 struct btrfs_fs_info *fs_info, u64 qgroupid)
1400 {
1401 struct btrfs_root *quota_root;
1402 struct btrfs_qgroup *qgroup;
1403 struct btrfs_qgroup_list *list;
1404 int ret = 0;
1405
1406 mutex_lock(&fs_info->qgroup_ioctl_lock);
1407 quota_root = fs_info->quota_root;
1408 if (!quota_root) {
1409 ret = -EINVAL;
1410 goto out;
1411 }
1412
1413 qgroup = find_qgroup_rb(fs_info, qgroupid);
1414 if (!qgroup) {
1415 ret = -ENOENT;
1416 goto out;
1417 } else {
1418 /* check if there are no children of this qgroup */
1419 if (!list_empty(&qgroup->members)) {
1420 ret = -EBUSY;
1421 goto out;
1422 }
1423 }
1424 ret = del_qgroup_item(trans, qgroupid);
1425 if (ret && ret != -ENOENT)
1426 goto out;
1427
1428 while (!list_empty(&qgroup->groups)) {
1429 list = list_first_entry(&qgroup->groups,
1430 struct btrfs_qgroup_list, next_group);
1431 ret = __del_qgroup_relation(trans, fs_info,
1432 qgroupid,
1433 list->group->qgroupid);
1434 if (ret)
1435 goto out;
1436 }
1437
1438 spin_lock(&fs_info->qgroup_lock);
1439 del_qgroup_rb(fs_info, qgroupid);
1440 spin_unlock(&fs_info->qgroup_lock);
1441 out:
1442 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1443 return ret;
1444 }
1445
1446 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1447 struct btrfs_fs_info *fs_info, u64 qgroupid,
1448 struct btrfs_qgroup_limit *limit)
1449 {
1450 struct btrfs_root *quota_root;
1451 struct btrfs_qgroup *qgroup;
1452 int ret = 0;
1453 /* Sometimes we would want to clear the limit on this qgroup.
1454 * To meet this requirement, we treat the -1 as a special value
1455 * which tell kernel to clear the limit on this qgroup.
1456 */
1457 const u64 CLEAR_VALUE = -1;
1458
1459 mutex_lock(&fs_info->qgroup_ioctl_lock);
1460 quota_root = fs_info->quota_root;
1461 if (!quota_root) {
1462 ret = -EINVAL;
1463 goto out;
1464 }
1465
1466 qgroup = find_qgroup_rb(fs_info, qgroupid);
1467 if (!qgroup) {
1468 ret = -ENOENT;
1469 goto out;
1470 }
1471
1472 spin_lock(&fs_info->qgroup_lock);
1473 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1474 if (limit->max_rfer == CLEAR_VALUE) {
1475 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1476 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1477 qgroup->max_rfer = 0;
1478 } else {
1479 qgroup->max_rfer = limit->max_rfer;
1480 }
1481 }
1482 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1483 if (limit->max_excl == CLEAR_VALUE) {
1484 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1485 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1486 qgroup->max_excl = 0;
1487 } else {
1488 qgroup->max_excl = limit->max_excl;
1489 }
1490 }
1491 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1492 if (limit->rsv_rfer == CLEAR_VALUE) {
1493 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1494 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1495 qgroup->rsv_rfer = 0;
1496 } else {
1497 qgroup->rsv_rfer = limit->rsv_rfer;
1498 }
1499 }
1500 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1501 if (limit->rsv_excl == CLEAR_VALUE) {
1502 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1503 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1504 qgroup->rsv_excl = 0;
1505 } else {
1506 qgroup->rsv_excl = limit->rsv_excl;
1507 }
1508 }
1509 qgroup->lim_flags |= limit->flags;
1510
1511 spin_unlock(&fs_info->qgroup_lock);
1512
1513 ret = update_qgroup_limit_item(trans, qgroup);
1514 if (ret) {
1515 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1516 btrfs_info(fs_info, "unable to update quota limit for %llu",
1517 qgroupid);
1518 }
1519
1520 out:
1521 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1522 return ret;
1523 }
1524
1525 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1526 struct btrfs_delayed_ref_root *delayed_refs,
1527 struct btrfs_qgroup_extent_record *record)
1528 {
1529 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1530 struct rb_node *parent_node = NULL;
1531 struct btrfs_qgroup_extent_record *entry;
1532 u64 bytenr = record->bytenr;
1533
1534 lockdep_assert_held(&delayed_refs->lock);
1535 trace_btrfs_qgroup_trace_extent(fs_info, record);
1536
1537 while (*p) {
1538 parent_node = *p;
1539 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1540 node);
1541 if (bytenr < entry->bytenr)
1542 p = &(*p)->rb_left;
1543 else if (bytenr > entry->bytenr)
1544 p = &(*p)->rb_right;
1545 else
1546 return 1;
1547 }
1548
1549 rb_link_node(&record->node, parent_node, p);
1550 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1551 return 0;
1552 }
1553
1554 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1555 struct btrfs_qgroup_extent_record *qrecord)
1556 {
1557 struct ulist *old_root;
1558 u64 bytenr = qrecord->bytenr;
1559 int ret;
1560
1561 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1562 if (ret < 0) {
1563 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1564 btrfs_warn(fs_info,
1565 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1566 ret);
1567 return 0;
1568 }
1569
1570 /*
1571 * Here we don't need to get the lock of
1572 * trans->transaction->delayed_refs, since inserted qrecord won't
1573 * be deleted, only qrecord->node may be modified (new qrecord insert)
1574 *
1575 * So modifying qrecord->old_roots is safe here
1576 */
1577 qrecord->old_roots = old_root;
1578 return 0;
1579 }
1580
1581 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1582 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1583 gfp_t gfp_flag)
1584 {
1585 struct btrfs_qgroup_extent_record *record;
1586 struct btrfs_delayed_ref_root *delayed_refs;
1587 int ret;
1588
1589 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1590 || bytenr == 0 || num_bytes == 0)
1591 return 0;
1592 if (WARN_ON(trans == NULL))
1593 return -EINVAL;
1594 record = kmalloc(sizeof(*record), gfp_flag);
1595 if (!record)
1596 return -ENOMEM;
1597
1598 delayed_refs = &trans->transaction->delayed_refs;
1599 record->bytenr = bytenr;
1600 record->num_bytes = num_bytes;
1601 record->old_roots = NULL;
1602
1603 spin_lock(&delayed_refs->lock);
1604 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1605 spin_unlock(&delayed_refs->lock);
1606 if (ret > 0) {
1607 kfree(record);
1608 return 0;
1609 }
1610 return btrfs_qgroup_trace_extent_post(fs_info, record);
1611 }
1612
1613 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1614 struct btrfs_fs_info *fs_info,
1615 struct extent_buffer *eb)
1616 {
1617 int nr = btrfs_header_nritems(eb);
1618 int i, extent_type, ret;
1619 struct btrfs_key key;
1620 struct btrfs_file_extent_item *fi;
1621 u64 bytenr, num_bytes;
1622
1623 /* We can be called directly from walk_up_proc() */
1624 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1625 return 0;
1626
1627 for (i = 0; i < nr; i++) {
1628 btrfs_item_key_to_cpu(eb, &key, i);
1629
1630 if (key.type != BTRFS_EXTENT_DATA_KEY)
1631 continue;
1632
1633 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1634 /* filter out non qgroup-accountable extents */
1635 extent_type = btrfs_file_extent_type(eb, fi);
1636
1637 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1638 continue;
1639
1640 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1641 if (!bytenr)
1642 continue;
1643
1644 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1645
1646 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1647 num_bytes, GFP_NOFS);
1648 if (ret)
1649 return ret;
1650 }
1651 cond_resched();
1652 return 0;
1653 }
1654
1655 /*
1656 * Walk up the tree from the bottom, freeing leaves and any interior
1657 * nodes which have had all slots visited. If a node (leaf or
1658 * interior) is freed, the node above it will have it's slot
1659 * incremented. The root node will never be freed.
1660 *
1661 * At the end of this function, we should have a path which has all
1662 * slots incremented to the next position for a search. If we need to
1663 * read a new node it will be NULL and the node above it will have the
1664 * correct slot selected for a later read.
1665 *
1666 * If we increment the root nodes slot counter past the number of
1667 * elements, 1 is returned to signal completion of the search.
1668 */
1669 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1670 {
1671 int level = 0;
1672 int nr, slot;
1673 struct extent_buffer *eb;
1674
1675 if (root_level == 0)
1676 return 1;
1677
1678 while (level <= root_level) {
1679 eb = path->nodes[level];
1680 nr = btrfs_header_nritems(eb);
1681 path->slots[level]++;
1682 slot = path->slots[level];
1683 if (slot >= nr || level == 0) {
1684 /*
1685 * Don't free the root - we will detect this
1686 * condition after our loop and return a
1687 * positive value for caller to stop walking the tree.
1688 */
1689 if (level != root_level) {
1690 btrfs_tree_unlock_rw(eb, path->locks[level]);
1691 path->locks[level] = 0;
1692
1693 free_extent_buffer(eb);
1694 path->nodes[level] = NULL;
1695 path->slots[level] = 0;
1696 }
1697 } else {
1698 /*
1699 * We have a valid slot to walk back down
1700 * from. Stop here so caller can process these
1701 * new nodes.
1702 */
1703 break;
1704 }
1705
1706 level++;
1707 }
1708
1709 eb = path->nodes[root_level];
1710 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1711 return 1;
1712
1713 return 0;
1714 }
1715
1716 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1717 struct btrfs_root *root,
1718 struct extent_buffer *root_eb,
1719 u64 root_gen, int root_level)
1720 {
1721 struct btrfs_fs_info *fs_info = root->fs_info;
1722 int ret = 0;
1723 int level;
1724 struct extent_buffer *eb = root_eb;
1725 struct btrfs_path *path = NULL;
1726
1727 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1728 BUG_ON(root_eb == NULL);
1729
1730 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1731 return 0;
1732
1733 if (!extent_buffer_uptodate(root_eb)) {
1734 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1735 if (ret)
1736 goto out;
1737 }
1738
1739 if (root_level == 0) {
1740 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1741 goto out;
1742 }
1743
1744 path = btrfs_alloc_path();
1745 if (!path)
1746 return -ENOMEM;
1747
1748 /*
1749 * Walk down the tree. Missing extent blocks are filled in as
1750 * we go. Metadata is accounted every time we read a new
1751 * extent block.
1752 *
1753 * When we reach a leaf, we account for file extent items in it,
1754 * walk back up the tree (adjusting slot pointers as we go)
1755 * and restart the search process.
1756 */
1757 extent_buffer_get(root_eb); /* For path */
1758 path->nodes[root_level] = root_eb;
1759 path->slots[root_level] = 0;
1760 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1761 walk_down:
1762 level = root_level;
1763 while (level >= 0) {
1764 if (path->nodes[level] == NULL) {
1765 struct btrfs_key first_key;
1766 int parent_slot;
1767 u64 child_gen;
1768 u64 child_bytenr;
1769
1770 /*
1771 * We need to get child blockptr/gen from parent before
1772 * we can read it.
1773 */
1774 eb = path->nodes[level + 1];
1775 parent_slot = path->slots[level + 1];
1776 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1777 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1778 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1779
1780 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1781 level, &first_key);
1782 if (IS_ERR(eb)) {
1783 ret = PTR_ERR(eb);
1784 goto out;
1785 } else if (!extent_buffer_uptodate(eb)) {
1786 free_extent_buffer(eb);
1787 ret = -EIO;
1788 goto out;
1789 }
1790
1791 path->nodes[level] = eb;
1792 path->slots[level] = 0;
1793
1794 btrfs_tree_read_lock(eb);
1795 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1796 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1797
1798 ret = btrfs_qgroup_trace_extent(trans, fs_info,
1799 child_bytenr,
1800 fs_info->nodesize,
1801 GFP_NOFS);
1802 if (ret)
1803 goto out;
1804 }
1805
1806 if (level == 0) {
1807 ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1808 path->nodes[level]);
1809 if (ret)
1810 goto out;
1811
1812 /* Nonzero return here means we completed our search */
1813 ret = adjust_slots_upwards(path, root_level);
1814 if (ret)
1815 break;
1816
1817 /* Restart search with new slots */
1818 goto walk_down;
1819 }
1820
1821 level--;
1822 }
1823
1824 ret = 0;
1825 out:
1826 btrfs_free_path(path);
1827
1828 return ret;
1829 }
1830
1831 #define UPDATE_NEW 0
1832 #define UPDATE_OLD 1
1833 /*
1834 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1835 */
1836 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1837 struct ulist *roots, struct ulist *tmp,
1838 struct ulist *qgroups, u64 seq, int update_old)
1839 {
1840 struct ulist_node *unode;
1841 struct ulist_iterator uiter;
1842 struct ulist_node *tmp_unode;
1843 struct ulist_iterator tmp_uiter;
1844 struct btrfs_qgroup *qg;
1845 int ret = 0;
1846
1847 if (!roots)
1848 return 0;
1849 ULIST_ITER_INIT(&uiter);
1850 while ((unode = ulist_next(roots, &uiter))) {
1851 qg = find_qgroup_rb(fs_info, unode->val);
1852 if (!qg)
1853 continue;
1854
1855 ulist_reinit(tmp);
1856 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1857 GFP_ATOMIC);
1858 if (ret < 0)
1859 return ret;
1860 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1861 if (ret < 0)
1862 return ret;
1863 ULIST_ITER_INIT(&tmp_uiter);
1864 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1865 struct btrfs_qgroup_list *glist;
1866
1867 qg = unode_aux_to_qgroup(tmp_unode);
1868 if (update_old)
1869 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1870 else
1871 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1872 list_for_each_entry(glist, &qg->groups, next_group) {
1873 ret = ulist_add(qgroups, glist->group->qgroupid,
1874 qgroup_to_aux(glist->group),
1875 GFP_ATOMIC);
1876 if (ret < 0)
1877 return ret;
1878 ret = ulist_add(tmp, glist->group->qgroupid,
1879 qgroup_to_aux(glist->group),
1880 GFP_ATOMIC);
1881 if (ret < 0)
1882 return ret;
1883 }
1884 }
1885 }
1886 return 0;
1887 }
1888
1889 /*
1890 * Update qgroup rfer/excl counters.
1891 * Rfer update is easy, codes can explain themselves.
1892 *
1893 * Excl update is tricky, the update is split into 2 part.
1894 * Part 1: Possible exclusive <-> sharing detect:
1895 * | A | !A |
1896 * -------------------------------------
1897 * B | * | - |
1898 * -------------------------------------
1899 * !B | + | ** |
1900 * -------------------------------------
1901 *
1902 * Conditions:
1903 * A: cur_old_roots < nr_old_roots (not exclusive before)
1904 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1905 * B: cur_new_roots < nr_new_roots (not exclusive now)
1906 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1907 *
1908 * Results:
1909 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1910 * *: Definitely not changed. **: Possible unchanged.
1911 *
1912 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1913 *
1914 * To make the logic clear, we first use condition A and B to split
1915 * combination into 4 results.
1916 *
1917 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1918 * only on variant maybe 0.
1919 *
1920 * Lastly, check result **, since there are 2 variants maybe 0, split them
1921 * again(2x2).
1922 * But this time we don't need to consider other things, the codes and logic
1923 * is easy to understand now.
1924 */
1925 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1926 struct ulist *qgroups,
1927 u64 nr_old_roots,
1928 u64 nr_new_roots,
1929 u64 num_bytes, u64 seq)
1930 {
1931 struct ulist_node *unode;
1932 struct ulist_iterator uiter;
1933 struct btrfs_qgroup *qg;
1934 u64 cur_new_count, cur_old_count;
1935
1936 ULIST_ITER_INIT(&uiter);
1937 while ((unode = ulist_next(qgroups, &uiter))) {
1938 bool dirty = false;
1939
1940 qg = unode_aux_to_qgroup(unode);
1941 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1942 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1943
1944 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
1945 cur_new_count);
1946
1947 /* Rfer update part */
1948 if (cur_old_count == 0 && cur_new_count > 0) {
1949 qg->rfer += num_bytes;
1950 qg->rfer_cmpr += num_bytes;
1951 dirty = true;
1952 }
1953 if (cur_old_count > 0 && cur_new_count == 0) {
1954 qg->rfer -= num_bytes;
1955 qg->rfer_cmpr -= num_bytes;
1956 dirty = true;
1957 }
1958
1959 /* Excl update part */
1960 /* Exclusive/none -> shared case */
1961 if (cur_old_count == nr_old_roots &&
1962 cur_new_count < nr_new_roots) {
1963 /* Exclusive -> shared */
1964 if (cur_old_count != 0) {
1965 qg->excl -= num_bytes;
1966 qg->excl_cmpr -= num_bytes;
1967 dirty = true;
1968 }
1969 }
1970
1971 /* Shared -> exclusive/none case */
1972 if (cur_old_count < nr_old_roots &&
1973 cur_new_count == nr_new_roots) {
1974 /* Shared->exclusive */
1975 if (cur_new_count != 0) {
1976 qg->excl += num_bytes;
1977 qg->excl_cmpr += num_bytes;
1978 dirty = true;
1979 }
1980 }
1981
1982 /* Exclusive/none -> exclusive/none case */
1983 if (cur_old_count == nr_old_roots &&
1984 cur_new_count == nr_new_roots) {
1985 if (cur_old_count == 0) {
1986 /* None -> exclusive/none */
1987
1988 if (cur_new_count != 0) {
1989 /* None -> exclusive */
1990 qg->excl += num_bytes;
1991 qg->excl_cmpr += num_bytes;
1992 dirty = true;
1993 }
1994 /* None -> none, nothing changed */
1995 } else {
1996 /* Exclusive -> exclusive/none */
1997
1998 if (cur_new_count == 0) {
1999 /* Exclusive -> none */
2000 qg->excl -= num_bytes;
2001 qg->excl_cmpr -= num_bytes;
2002 dirty = true;
2003 }
2004 /* Exclusive -> exclusive, nothing changed */
2005 }
2006 }
2007
2008 if (dirty)
2009 qgroup_dirty(fs_info, qg);
2010 }
2011 return 0;
2012 }
2013
2014 /*
2015 * Check if the @roots potentially is a list of fs tree roots
2016 *
2017 * Return 0 for definitely not a fs/subvol tree roots ulist
2018 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2019 * one as well)
2020 */
2021 static int maybe_fs_roots(struct ulist *roots)
2022 {
2023 struct ulist_node *unode;
2024 struct ulist_iterator uiter;
2025
2026 /* Empty one, still possible for fs roots */
2027 if (!roots || roots->nnodes == 0)
2028 return 1;
2029
2030 ULIST_ITER_INIT(&uiter);
2031 unode = ulist_next(roots, &uiter);
2032 if (!unode)
2033 return 1;
2034
2035 /*
2036 * If it contains fs tree roots, then it must belong to fs/subvol
2037 * trees.
2038 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2039 */
2040 return is_fstree(unode->val);
2041 }
2042
2043 int
2044 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
2045 struct btrfs_fs_info *fs_info,
2046 u64 bytenr, u64 num_bytes,
2047 struct ulist *old_roots, struct ulist *new_roots)
2048 {
2049 struct ulist *qgroups = NULL;
2050 struct ulist *tmp = NULL;
2051 u64 seq;
2052 u64 nr_new_roots = 0;
2053 u64 nr_old_roots = 0;
2054 int ret = 0;
2055
2056 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2057 return 0;
2058
2059 if (new_roots) {
2060 if (!maybe_fs_roots(new_roots))
2061 goto out_free;
2062 nr_new_roots = new_roots->nnodes;
2063 }
2064 if (old_roots) {
2065 if (!maybe_fs_roots(old_roots))
2066 goto out_free;
2067 nr_old_roots = old_roots->nnodes;
2068 }
2069
2070 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2071 if (nr_old_roots == 0 && nr_new_roots == 0)
2072 goto out_free;
2073
2074 BUG_ON(!fs_info->quota_root);
2075
2076 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2077 num_bytes, nr_old_roots, nr_new_roots);
2078
2079 qgroups = ulist_alloc(GFP_NOFS);
2080 if (!qgroups) {
2081 ret = -ENOMEM;
2082 goto out_free;
2083 }
2084 tmp = ulist_alloc(GFP_NOFS);
2085 if (!tmp) {
2086 ret = -ENOMEM;
2087 goto out_free;
2088 }
2089
2090 mutex_lock(&fs_info->qgroup_rescan_lock);
2091 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2092 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2093 mutex_unlock(&fs_info->qgroup_rescan_lock);
2094 ret = 0;
2095 goto out_free;
2096 }
2097 }
2098 mutex_unlock(&fs_info->qgroup_rescan_lock);
2099
2100 spin_lock(&fs_info->qgroup_lock);
2101 seq = fs_info->qgroup_seq;
2102
2103 /* Update old refcnts using old_roots */
2104 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2105 UPDATE_OLD);
2106 if (ret < 0)
2107 goto out;
2108
2109 /* Update new refcnts using new_roots */
2110 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2111 UPDATE_NEW);
2112 if (ret < 0)
2113 goto out;
2114
2115 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2116 num_bytes, seq);
2117
2118 /*
2119 * Bump qgroup_seq to avoid seq overlap
2120 */
2121 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2122 out:
2123 spin_unlock(&fs_info->qgroup_lock);
2124 out_free:
2125 ulist_free(tmp);
2126 ulist_free(qgroups);
2127 ulist_free(old_roots);
2128 ulist_free(new_roots);
2129 return ret;
2130 }
2131
2132 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2133 {
2134 struct btrfs_fs_info *fs_info = trans->fs_info;
2135 struct btrfs_qgroup_extent_record *record;
2136 struct btrfs_delayed_ref_root *delayed_refs;
2137 struct ulist *new_roots = NULL;
2138 struct rb_node *node;
2139 u64 qgroup_to_skip;
2140 int ret = 0;
2141
2142 delayed_refs = &trans->transaction->delayed_refs;
2143 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2144 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2145 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2146 node);
2147
2148 trace_btrfs_qgroup_account_extents(fs_info, record);
2149
2150 if (!ret) {
2151 /*
2152 * Old roots should be searched when inserting qgroup
2153 * extent record
2154 */
2155 if (WARN_ON(!record->old_roots)) {
2156 /* Search commit root to find old_roots */
2157 ret = btrfs_find_all_roots(NULL, fs_info,
2158 record->bytenr, 0,
2159 &record->old_roots, false);
2160 if (ret < 0)
2161 goto cleanup;
2162 }
2163
2164 /*
2165 * Use SEQ_LAST as time_seq to do special search, which
2166 * doesn't lock tree or delayed_refs and search current
2167 * root. It's safe inside commit_transaction().
2168 */
2169 ret = btrfs_find_all_roots(trans, fs_info,
2170 record->bytenr, SEQ_LAST, &new_roots, false);
2171 if (ret < 0)
2172 goto cleanup;
2173 if (qgroup_to_skip) {
2174 ulist_del(new_roots, qgroup_to_skip, 0);
2175 ulist_del(record->old_roots, qgroup_to_skip,
2176 0);
2177 }
2178 ret = btrfs_qgroup_account_extent(trans, fs_info,
2179 record->bytenr, record->num_bytes,
2180 record->old_roots, new_roots);
2181 record->old_roots = NULL;
2182 new_roots = NULL;
2183 }
2184 cleanup:
2185 ulist_free(record->old_roots);
2186 ulist_free(new_roots);
2187 new_roots = NULL;
2188 rb_erase(node, &delayed_refs->dirty_extent_root);
2189 kfree(record);
2190
2191 }
2192 return ret;
2193 }
2194
2195 /*
2196 * called from commit_transaction. Writes all changed qgroups to disk.
2197 */
2198 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2199 struct btrfs_fs_info *fs_info)
2200 {
2201 struct btrfs_root *quota_root = fs_info->quota_root;
2202 int ret = 0;
2203
2204 if (!quota_root)
2205 return ret;
2206
2207 spin_lock(&fs_info->qgroup_lock);
2208 while (!list_empty(&fs_info->dirty_qgroups)) {
2209 struct btrfs_qgroup *qgroup;
2210 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2211 struct btrfs_qgroup, dirty);
2212 list_del_init(&qgroup->dirty);
2213 spin_unlock(&fs_info->qgroup_lock);
2214 ret = update_qgroup_info_item(trans, qgroup);
2215 if (ret)
2216 fs_info->qgroup_flags |=
2217 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2218 ret = update_qgroup_limit_item(trans, qgroup);
2219 if (ret)
2220 fs_info->qgroup_flags |=
2221 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2222 spin_lock(&fs_info->qgroup_lock);
2223 }
2224 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2225 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2226 else
2227 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2228 spin_unlock(&fs_info->qgroup_lock);
2229
2230 ret = update_qgroup_status_item(trans, fs_info, quota_root);
2231 if (ret)
2232 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2233
2234 return ret;
2235 }
2236
2237 /*
2238 * Copy the accounting information between qgroups. This is necessary
2239 * when a snapshot or a subvolume is created. Throwing an error will
2240 * cause a transaction abort so we take extra care here to only error
2241 * when a readonly fs is a reasonable outcome.
2242 */
2243 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2244 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2245 struct btrfs_qgroup_inherit *inherit)
2246 {
2247 int ret = 0;
2248 int i;
2249 u64 *i_qgroups;
2250 struct btrfs_root *quota_root = fs_info->quota_root;
2251 struct btrfs_qgroup *srcgroup;
2252 struct btrfs_qgroup *dstgroup;
2253 u32 level_size = 0;
2254 u64 nums;
2255
2256 mutex_lock(&fs_info->qgroup_ioctl_lock);
2257 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2258 goto out;
2259
2260 if (!quota_root) {
2261 ret = -EINVAL;
2262 goto out;
2263 }
2264
2265 if (inherit) {
2266 i_qgroups = (u64 *)(inherit + 1);
2267 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2268 2 * inherit->num_excl_copies;
2269 for (i = 0; i < nums; ++i) {
2270 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2271
2272 /*
2273 * Zero out invalid groups so we can ignore
2274 * them later.
2275 */
2276 if (!srcgroup ||
2277 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2278 *i_qgroups = 0ULL;
2279
2280 ++i_qgroups;
2281 }
2282 }
2283
2284 /*
2285 * create a tracking group for the subvol itself
2286 */
2287 ret = add_qgroup_item(trans, quota_root, objectid);
2288 if (ret)
2289 goto out;
2290
2291 /*
2292 * add qgroup to all inherited groups
2293 */
2294 if (inherit) {
2295 i_qgroups = (u64 *)(inherit + 1);
2296 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2297 if (*i_qgroups == 0)
2298 continue;
2299 ret = add_qgroup_relation_item(trans, objectid,
2300 *i_qgroups);
2301 if (ret && ret != -EEXIST)
2302 goto out;
2303 ret = add_qgroup_relation_item(trans, *i_qgroups,
2304 objectid);
2305 if (ret && ret != -EEXIST)
2306 goto out;
2307 }
2308 ret = 0;
2309 }
2310
2311
2312 spin_lock(&fs_info->qgroup_lock);
2313
2314 dstgroup = add_qgroup_rb(fs_info, objectid);
2315 if (IS_ERR(dstgroup)) {
2316 ret = PTR_ERR(dstgroup);
2317 goto unlock;
2318 }
2319
2320 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2321 dstgroup->lim_flags = inherit->lim.flags;
2322 dstgroup->max_rfer = inherit->lim.max_rfer;
2323 dstgroup->max_excl = inherit->lim.max_excl;
2324 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2325 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2326
2327 ret = update_qgroup_limit_item(trans, dstgroup);
2328 if (ret) {
2329 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2330 btrfs_info(fs_info,
2331 "unable to update quota limit for %llu",
2332 dstgroup->qgroupid);
2333 goto unlock;
2334 }
2335 }
2336
2337 if (srcid) {
2338 srcgroup = find_qgroup_rb(fs_info, srcid);
2339 if (!srcgroup)
2340 goto unlock;
2341
2342 /*
2343 * We call inherit after we clone the root in order to make sure
2344 * our counts don't go crazy, so at this point the only
2345 * difference between the two roots should be the root node.
2346 */
2347 level_size = fs_info->nodesize;
2348 dstgroup->rfer = srcgroup->rfer;
2349 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2350 dstgroup->excl = level_size;
2351 dstgroup->excl_cmpr = level_size;
2352 srcgroup->excl = level_size;
2353 srcgroup->excl_cmpr = level_size;
2354
2355 /* inherit the limit info */
2356 dstgroup->lim_flags = srcgroup->lim_flags;
2357 dstgroup->max_rfer = srcgroup->max_rfer;
2358 dstgroup->max_excl = srcgroup->max_excl;
2359 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2360 dstgroup->rsv_excl = srcgroup->rsv_excl;
2361
2362 qgroup_dirty(fs_info, dstgroup);
2363 qgroup_dirty(fs_info, srcgroup);
2364 }
2365
2366 if (!inherit)
2367 goto unlock;
2368
2369 i_qgroups = (u64 *)(inherit + 1);
2370 for (i = 0; i < inherit->num_qgroups; ++i) {
2371 if (*i_qgroups) {
2372 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2373 if (ret)
2374 goto unlock;
2375 }
2376 ++i_qgroups;
2377 }
2378
2379 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2380 struct btrfs_qgroup *src;
2381 struct btrfs_qgroup *dst;
2382
2383 if (!i_qgroups[0] || !i_qgroups[1])
2384 continue;
2385
2386 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2387 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2388
2389 if (!src || !dst) {
2390 ret = -EINVAL;
2391 goto unlock;
2392 }
2393
2394 dst->rfer = src->rfer - level_size;
2395 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2396 }
2397 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2398 struct btrfs_qgroup *src;
2399 struct btrfs_qgroup *dst;
2400
2401 if (!i_qgroups[0] || !i_qgroups[1])
2402 continue;
2403
2404 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2405 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2406
2407 if (!src || !dst) {
2408 ret = -EINVAL;
2409 goto unlock;
2410 }
2411
2412 dst->excl = src->excl + level_size;
2413 dst->excl_cmpr = src->excl_cmpr + level_size;
2414 }
2415
2416 unlock:
2417 spin_unlock(&fs_info->qgroup_lock);
2418 out:
2419 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2420 return ret;
2421 }
2422
2423 /*
2424 * Two limits to commit transaction in advance.
2425 *
2426 * For RATIO, it will be 1/RATIO of the remaining limit
2427 * (excluding data and prealloc meta) as threshold.
2428 * For SIZE, it will be in byte unit as threshold.
2429 */
2430 #define QGROUP_PERTRANS_RATIO 32
2431 #define QGROUP_PERTRANS_SIZE SZ_32M
2432 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2433 const struct btrfs_qgroup *qg, u64 num_bytes)
2434 {
2435 u64 limit;
2436 u64 threshold;
2437
2438 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2439 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2440 return false;
2441
2442 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2443 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2444 return false;
2445
2446 /*
2447 * Even if we passed the check, it's better to check if reservation
2448 * for meta_pertrans is pushing us near limit.
2449 * If there is too much pertrans reservation or it's near the limit,
2450 * let's try commit transaction to free some, using transaction_kthread
2451 */
2452 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2453 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2454 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2455 limit = qg->max_excl;
2456 else
2457 limit = qg->max_rfer;
2458 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2459 qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2460 QGROUP_PERTRANS_RATIO;
2461 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2462
2463 /*
2464 * Use transaction_kthread to commit transaction, so we no
2465 * longer need to bother nested transaction nor lock context.
2466 */
2467 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2468 btrfs_commit_transaction_locksafe(fs_info);
2469 }
2470
2471 return true;
2472 }
2473
2474 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2475 enum btrfs_qgroup_rsv_type type)
2476 {
2477 struct btrfs_root *quota_root;
2478 struct btrfs_qgroup *qgroup;
2479 struct btrfs_fs_info *fs_info = root->fs_info;
2480 u64 ref_root = root->root_key.objectid;
2481 int ret = 0;
2482 struct ulist_node *unode;
2483 struct ulist_iterator uiter;
2484
2485 if (!is_fstree(ref_root))
2486 return 0;
2487
2488 if (num_bytes == 0)
2489 return 0;
2490
2491 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2492 capable(CAP_SYS_RESOURCE))
2493 enforce = false;
2494
2495 spin_lock(&fs_info->qgroup_lock);
2496 quota_root = fs_info->quota_root;
2497 if (!quota_root)
2498 goto out;
2499
2500 qgroup = find_qgroup_rb(fs_info, ref_root);
2501 if (!qgroup)
2502 goto out;
2503
2504 /*
2505 * in a first step, we check all affected qgroups if any limits would
2506 * be exceeded
2507 */
2508 ulist_reinit(fs_info->qgroup_ulist);
2509 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2510 qgroup_to_aux(qgroup), GFP_ATOMIC);
2511 if (ret < 0)
2512 goto out;
2513 ULIST_ITER_INIT(&uiter);
2514 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2515 struct btrfs_qgroup *qg;
2516 struct btrfs_qgroup_list *glist;
2517
2518 qg = unode_aux_to_qgroup(unode);
2519
2520 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2521 ret = -EDQUOT;
2522 goto out;
2523 }
2524
2525 list_for_each_entry(glist, &qg->groups, next_group) {
2526 ret = ulist_add(fs_info->qgroup_ulist,
2527 glist->group->qgroupid,
2528 qgroup_to_aux(glist->group), GFP_ATOMIC);
2529 if (ret < 0)
2530 goto out;
2531 }
2532 }
2533 ret = 0;
2534 /*
2535 * no limits exceeded, now record the reservation into all qgroups
2536 */
2537 ULIST_ITER_INIT(&uiter);
2538 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2539 struct btrfs_qgroup *qg;
2540
2541 qg = unode_aux_to_qgroup(unode);
2542
2543 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2544 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2545 }
2546
2547 out:
2548 spin_unlock(&fs_info->qgroup_lock);
2549 return ret;
2550 }
2551
2552 /*
2553 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2554 * qgroup).
2555 *
2556 * Will handle all higher level qgroup too.
2557 *
2558 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2559 * This special case is only used for META_PERTRANS type.
2560 */
2561 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2562 u64 ref_root, u64 num_bytes,
2563 enum btrfs_qgroup_rsv_type type)
2564 {
2565 struct btrfs_root *quota_root;
2566 struct btrfs_qgroup *qgroup;
2567 struct ulist_node *unode;
2568 struct ulist_iterator uiter;
2569 int ret = 0;
2570
2571 if (!is_fstree(ref_root))
2572 return;
2573
2574 if (num_bytes == 0)
2575 return;
2576
2577 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2578 WARN(1, "%s: Invalid type to free", __func__);
2579 return;
2580 }
2581 spin_lock(&fs_info->qgroup_lock);
2582
2583 quota_root = fs_info->quota_root;
2584 if (!quota_root)
2585 goto out;
2586
2587 qgroup = find_qgroup_rb(fs_info, ref_root);
2588 if (!qgroup)
2589 goto out;
2590
2591 if (num_bytes == (u64)-1)
2592 /*
2593 * We're freeing all pertrans rsv, get reserved value from
2594 * level 0 qgroup as real num_bytes to free.
2595 */
2596 num_bytes = qgroup->rsv.values[type];
2597
2598 ulist_reinit(fs_info->qgroup_ulist);
2599 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2600 qgroup_to_aux(qgroup), GFP_ATOMIC);
2601 if (ret < 0)
2602 goto out;
2603 ULIST_ITER_INIT(&uiter);
2604 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2605 struct btrfs_qgroup *qg;
2606 struct btrfs_qgroup_list *glist;
2607
2608 qg = unode_aux_to_qgroup(unode);
2609
2610 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2611 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2612
2613 list_for_each_entry(glist, &qg->groups, next_group) {
2614 ret = ulist_add(fs_info->qgroup_ulist,
2615 glist->group->qgroupid,
2616 qgroup_to_aux(glist->group), GFP_ATOMIC);
2617 if (ret < 0)
2618 goto out;
2619 }
2620 }
2621
2622 out:
2623 spin_unlock(&fs_info->qgroup_lock);
2624 }
2625
2626 /*
2627 * Check if the leaf is the last leaf. Which means all node pointers
2628 * are at their last position.
2629 */
2630 static bool is_last_leaf(struct btrfs_path *path)
2631 {
2632 int i;
2633
2634 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2635 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2636 return false;
2637 }
2638 return true;
2639 }
2640
2641 /*
2642 * returns < 0 on error, 0 when more leafs are to be scanned.
2643 * returns 1 when done.
2644 */
2645 static int
2646 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2647 struct btrfs_trans_handle *trans)
2648 {
2649 struct btrfs_key found;
2650 struct extent_buffer *scratch_leaf = NULL;
2651 struct ulist *roots = NULL;
2652 u64 num_bytes;
2653 bool done;
2654 int slot;
2655 int ret;
2656
2657 mutex_lock(&fs_info->qgroup_rescan_lock);
2658 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2659 &fs_info->qgroup_rescan_progress,
2660 path, 1, 0);
2661
2662 btrfs_debug(fs_info,
2663 "current progress key (%llu %u %llu), search_slot ret %d",
2664 fs_info->qgroup_rescan_progress.objectid,
2665 fs_info->qgroup_rescan_progress.type,
2666 fs_info->qgroup_rescan_progress.offset, ret);
2667
2668 if (ret) {
2669 /*
2670 * The rescan is about to end, we will not be scanning any
2671 * further blocks. We cannot unset the RESCAN flag here, because
2672 * we want to commit the transaction if everything went well.
2673 * To make the live accounting work in this phase, we set our
2674 * scan progress pointer such that every real extent objectid
2675 * will be smaller.
2676 */
2677 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2678 btrfs_release_path(path);
2679 mutex_unlock(&fs_info->qgroup_rescan_lock);
2680 return ret;
2681 }
2682 done = is_last_leaf(path);
2683
2684 btrfs_item_key_to_cpu(path->nodes[0], &found,
2685 btrfs_header_nritems(path->nodes[0]) - 1);
2686 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2687
2688 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2689 if (!scratch_leaf) {
2690 ret = -ENOMEM;
2691 mutex_unlock(&fs_info->qgroup_rescan_lock);
2692 goto out;
2693 }
2694 extent_buffer_get(scratch_leaf);
2695 btrfs_tree_read_lock(scratch_leaf);
2696 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2697 slot = path->slots[0];
2698 btrfs_release_path(path);
2699 mutex_unlock(&fs_info->qgroup_rescan_lock);
2700
2701 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2702 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2703 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2704 found.type != BTRFS_METADATA_ITEM_KEY)
2705 continue;
2706 if (found.type == BTRFS_METADATA_ITEM_KEY)
2707 num_bytes = fs_info->nodesize;
2708 else
2709 num_bytes = found.offset;
2710
2711 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2712 &roots, false);
2713 if (ret < 0)
2714 goto out;
2715 /* For rescan, just pass old_roots as NULL */
2716 ret = btrfs_qgroup_account_extent(trans, fs_info,
2717 found.objectid, num_bytes, NULL, roots);
2718 if (ret < 0)
2719 goto out;
2720 }
2721 out:
2722 if (scratch_leaf) {
2723 btrfs_tree_read_unlock_blocking(scratch_leaf);
2724 free_extent_buffer(scratch_leaf);
2725 }
2726
2727 if (done && !ret) {
2728 ret = 1;
2729 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2730 }
2731 return ret;
2732 }
2733
2734 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2735 {
2736 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2737 qgroup_rescan_work);
2738 struct btrfs_path *path;
2739 struct btrfs_trans_handle *trans = NULL;
2740 int err = -ENOMEM;
2741 int ret = 0;
2742
2743 path = btrfs_alloc_path();
2744 if (!path)
2745 goto out;
2746 /*
2747 * Rescan should only search for commit root, and any later difference
2748 * should be recorded by qgroup
2749 */
2750 path->search_commit_root = 1;
2751 path->skip_locking = 1;
2752
2753 err = 0;
2754 while (!err && !btrfs_fs_closing(fs_info)) {
2755 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2756 if (IS_ERR(trans)) {
2757 err = PTR_ERR(trans);
2758 break;
2759 }
2760 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2761 err = -EINTR;
2762 } else {
2763 err = qgroup_rescan_leaf(fs_info, path, trans);
2764 }
2765 if (err > 0)
2766 btrfs_commit_transaction(trans);
2767 else
2768 btrfs_end_transaction(trans);
2769 }
2770
2771 out:
2772 btrfs_free_path(path);
2773
2774 mutex_lock(&fs_info->qgroup_rescan_lock);
2775 if (!btrfs_fs_closing(fs_info))
2776 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2777
2778 if (err > 0 &&
2779 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2780 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2781 } else if (err < 0) {
2782 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2783 }
2784 mutex_unlock(&fs_info->qgroup_rescan_lock);
2785
2786 /*
2787 * only update status, since the previous part has already updated the
2788 * qgroup info.
2789 */
2790 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2791 if (IS_ERR(trans)) {
2792 err = PTR_ERR(trans);
2793 btrfs_err(fs_info,
2794 "fail to start transaction for status update: %d",
2795 err);
2796 goto done;
2797 }
2798 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2799 if (ret < 0) {
2800 err = ret;
2801 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2802 }
2803 btrfs_end_transaction(trans);
2804
2805 if (btrfs_fs_closing(fs_info)) {
2806 btrfs_info(fs_info, "qgroup scan paused");
2807 } else if (err >= 0) {
2808 btrfs_info(fs_info, "qgroup scan completed%s",
2809 err > 0 ? " (inconsistency flag cleared)" : "");
2810 } else {
2811 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2812 }
2813
2814 done:
2815 mutex_lock(&fs_info->qgroup_rescan_lock);
2816 fs_info->qgroup_rescan_running = false;
2817 mutex_unlock(&fs_info->qgroup_rescan_lock);
2818 complete_all(&fs_info->qgroup_rescan_completion);
2819 }
2820
2821 /*
2822 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2823 * memory required for the rescan context.
2824 */
2825 static int
2826 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2827 int init_flags)
2828 {
2829 int ret = 0;
2830
2831 if (!init_flags) {
2832 /* we're resuming qgroup rescan at mount time */
2833 if (!(fs_info->qgroup_flags &
2834 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
2835 btrfs_warn(fs_info,
2836 "qgroup rescan init failed, qgroup is not enabled");
2837 ret = -EINVAL;
2838 } else if (!(fs_info->qgroup_flags &
2839 BTRFS_QGROUP_STATUS_FLAG_ON)) {
2840 btrfs_warn(fs_info,
2841 "qgroup rescan init failed, qgroup rescan is not queued");
2842 ret = -EINVAL;
2843 }
2844
2845 if (ret)
2846 return ret;
2847 }
2848
2849 mutex_lock(&fs_info->qgroup_rescan_lock);
2850 spin_lock(&fs_info->qgroup_lock);
2851
2852 if (init_flags) {
2853 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2854 btrfs_warn(fs_info,
2855 "qgroup rescan is already in progress");
2856 ret = -EINPROGRESS;
2857 } else if (!(fs_info->qgroup_flags &
2858 BTRFS_QGROUP_STATUS_FLAG_ON)) {
2859 btrfs_warn(fs_info,
2860 "qgroup rescan init failed, qgroup is not enabled");
2861 ret = -EINVAL;
2862 }
2863
2864 if (ret) {
2865 spin_unlock(&fs_info->qgroup_lock);
2866 mutex_unlock(&fs_info->qgroup_rescan_lock);
2867 return ret;
2868 }
2869 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2870 }
2871
2872 memset(&fs_info->qgroup_rescan_progress, 0,
2873 sizeof(fs_info->qgroup_rescan_progress));
2874 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2875 init_completion(&fs_info->qgroup_rescan_completion);
2876 fs_info->qgroup_rescan_running = true;
2877
2878 spin_unlock(&fs_info->qgroup_lock);
2879 mutex_unlock(&fs_info->qgroup_rescan_lock);
2880
2881 memset(&fs_info->qgroup_rescan_work, 0,
2882 sizeof(fs_info->qgroup_rescan_work));
2883 btrfs_init_work(&fs_info->qgroup_rescan_work,
2884 btrfs_qgroup_rescan_helper,
2885 btrfs_qgroup_rescan_worker, NULL, NULL);
2886 return 0;
2887 }
2888
2889 static void
2890 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2891 {
2892 struct rb_node *n;
2893 struct btrfs_qgroup *qgroup;
2894
2895 spin_lock(&fs_info->qgroup_lock);
2896 /* clear all current qgroup tracking information */
2897 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2898 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2899 qgroup->rfer = 0;
2900 qgroup->rfer_cmpr = 0;
2901 qgroup->excl = 0;
2902 qgroup->excl_cmpr = 0;
2903 }
2904 spin_unlock(&fs_info->qgroup_lock);
2905 }
2906
2907 int
2908 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2909 {
2910 int ret = 0;
2911 struct btrfs_trans_handle *trans;
2912
2913 ret = qgroup_rescan_init(fs_info, 0, 1);
2914 if (ret)
2915 return ret;
2916
2917 /*
2918 * We have set the rescan_progress to 0, which means no more
2919 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2920 * However, btrfs_qgroup_account_ref may be right after its call
2921 * to btrfs_find_all_roots, in which case it would still do the
2922 * accounting.
2923 * To solve this, we're committing the transaction, which will
2924 * ensure we run all delayed refs and only after that, we are
2925 * going to clear all tracking information for a clean start.
2926 */
2927
2928 trans = btrfs_join_transaction(fs_info->fs_root);
2929 if (IS_ERR(trans)) {
2930 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2931 return PTR_ERR(trans);
2932 }
2933 ret = btrfs_commit_transaction(trans);
2934 if (ret) {
2935 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2936 return ret;
2937 }
2938
2939 qgroup_rescan_zero_tracking(fs_info);
2940
2941 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2942 &fs_info->qgroup_rescan_work);
2943
2944 return 0;
2945 }
2946
2947 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2948 bool interruptible)
2949 {
2950 int running;
2951 int ret = 0;
2952
2953 mutex_lock(&fs_info->qgroup_rescan_lock);
2954 spin_lock(&fs_info->qgroup_lock);
2955 running = fs_info->qgroup_rescan_running;
2956 spin_unlock(&fs_info->qgroup_lock);
2957 mutex_unlock(&fs_info->qgroup_rescan_lock);
2958
2959 if (!running)
2960 return 0;
2961
2962 if (interruptible)
2963 ret = wait_for_completion_interruptible(
2964 &fs_info->qgroup_rescan_completion);
2965 else
2966 wait_for_completion(&fs_info->qgroup_rescan_completion);
2967
2968 return ret;
2969 }
2970
2971 /*
2972 * this is only called from open_ctree where we're still single threaded, thus
2973 * locking is omitted here.
2974 */
2975 void
2976 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2977 {
2978 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2979 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2980 &fs_info->qgroup_rescan_work);
2981 }
2982
2983 /*
2984 * Reserve qgroup space for range [start, start + len).
2985 *
2986 * This function will either reserve space from related qgroups or doing
2987 * nothing if the range is already reserved.
2988 *
2989 * Return 0 for successful reserve
2990 * Return <0 for error (including -EQUOT)
2991 *
2992 * NOTE: this function may sleep for memory allocation.
2993 * if btrfs_qgroup_reserve_data() is called multiple times with
2994 * same @reserved, caller must ensure when error happens it's OK
2995 * to free *ALL* reserved space.
2996 */
2997 int btrfs_qgroup_reserve_data(struct inode *inode,
2998 struct extent_changeset **reserved_ret, u64 start,
2999 u64 len)
3000 {
3001 struct btrfs_root *root = BTRFS_I(inode)->root;
3002 struct ulist_node *unode;
3003 struct ulist_iterator uiter;
3004 struct extent_changeset *reserved;
3005 u64 orig_reserved;
3006 u64 to_reserve;
3007 int ret;
3008
3009 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3010 !is_fstree(root->objectid) || len == 0)
3011 return 0;
3012
3013 /* @reserved parameter is mandatory for qgroup */
3014 if (WARN_ON(!reserved_ret))
3015 return -EINVAL;
3016 if (!*reserved_ret) {
3017 *reserved_ret = extent_changeset_alloc();
3018 if (!*reserved_ret)
3019 return -ENOMEM;
3020 }
3021 reserved = *reserved_ret;
3022 /* Record already reserved space */
3023 orig_reserved = reserved->bytes_changed;
3024 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3025 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3026
3027 /* Newly reserved space */
3028 to_reserve = reserved->bytes_changed - orig_reserved;
3029 trace_btrfs_qgroup_reserve_data(inode, start, len,
3030 to_reserve, QGROUP_RESERVE);
3031 if (ret < 0)
3032 goto cleanup;
3033 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3034 if (ret < 0)
3035 goto cleanup;
3036
3037 return ret;
3038
3039 cleanup:
3040 /* cleanup *ALL* already reserved ranges */
3041 ULIST_ITER_INIT(&uiter);
3042 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3043 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3044 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3045 extent_changeset_release(reserved);
3046 return ret;
3047 }
3048
3049 /* Free ranges specified by @reserved, normally in error path */
3050 static int qgroup_free_reserved_data(struct inode *inode,
3051 struct extent_changeset *reserved, u64 start, u64 len)
3052 {
3053 struct btrfs_root *root = BTRFS_I(inode)->root;
3054 struct ulist_node *unode;
3055 struct ulist_iterator uiter;
3056 struct extent_changeset changeset;
3057 int freed = 0;
3058 int ret;
3059
3060 extent_changeset_init(&changeset);
3061 len = round_up(start + len, root->fs_info->sectorsize);
3062 start = round_down(start, root->fs_info->sectorsize);
3063
3064 ULIST_ITER_INIT(&uiter);
3065 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3066 u64 range_start = unode->val;
3067 /* unode->aux is the inclusive end */
3068 u64 range_len = unode->aux - range_start + 1;
3069 u64 free_start;
3070 u64 free_len;
3071
3072 extent_changeset_release(&changeset);
3073
3074 /* Only free range in range [start, start + len) */
3075 if (range_start >= start + len ||
3076 range_start + range_len <= start)
3077 continue;
3078 free_start = max(range_start, start);
3079 free_len = min(start + len, range_start + range_len) -
3080 free_start;
3081 /*
3082 * TODO: To also modify reserved->ranges_reserved to reflect
3083 * the modification.
3084 *
3085 * However as long as we free qgroup reserved according to
3086 * EXTENT_QGROUP_RESERVED, we won't double free.
3087 * So not need to rush.
3088 */
3089 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3090 free_start, free_start + free_len - 1,
3091 EXTENT_QGROUP_RESERVED, &changeset);
3092 if (ret < 0)
3093 goto out;
3094 freed += changeset.bytes_changed;
3095 }
3096 btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
3097 BTRFS_QGROUP_RSV_DATA);
3098 ret = freed;
3099 out:
3100 extent_changeset_release(&changeset);
3101 return ret;
3102 }
3103
3104 static int __btrfs_qgroup_release_data(struct inode *inode,
3105 struct extent_changeset *reserved, u64 start, u64 len,
3106 int free)
3107 {
3108 struct extent_changeset changeset;
3109 int trace_op = QGROUP_RELEASE;
3110 int ret;
3111
3112 /* In release case, we shouldn't have @reserved */
3113 WARN_ON(!free && reserved);
3114 if (free && reserved)
3115 return qgroup_free_reserved_data(inode, reserved, start, len);
3116 extent_changeset_init(&changeset);
3117 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3118 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3119 if (ret < 0)
3120 goto out;
3121
3122 if (free)
3123 trace_op = QGROUP_FREE;
3124 trace_btrfs_qgroup_release_data(inode, start, len,
3125 changeset.bytes_changed, trace_op);
3126 if (free)
3127 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3128 BTRFS_I(inode)->root->objectid,
3129 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3130 ret = changeset.bytes_changed;
3131 out:
3132 extent_changeset_release(&changeset);
3133 return ret;
3134 }
3135
3136 /*
3137 * Free a reserved space range from io_tree and related qgroups
3138 *
3139 * Should be called when a range of pages get invalidated before reaching disk.
3140 * Or for error cleanup case.
3141 * if @reserved is given, only reserved range in [@start, @start + @len) will
3142 * be freed.
3143 *
3144 * For data written to disk, use btrfs_qgroup_release_data().
3145 *
3146 * NOTE: This function may sleep for memory allocation.
3147 */
3148 int btrfs_qgroup_free_data(struct inode *inode,
3149 struct extent_changeset *reserved, u64 start, u64 len)
3150 {
3151 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3152 }
3153
3154 /*
3155 * Release a reserved space range from io_tree only.
3156 *
3157 * Should be called when a range of pages get written to disk and corresponding
3158 * FILE_EXTENT is inserted into corresponding root.
3159 *
3160 * Since new qgroup accounting framework will only update qgroup numbers at
3161 * commit_transaction() time, its reserved space shouldn't be freed from
3162 * related qgroups.
3163 *
3164 * But we should release the range from io_tree, to allow further write to be
3165 * COWed.
3166 *
3167 * NOTE: This function may sleep for memory allocation.
3168 */
3169 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3170 {
3171 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3172 }
3173
3174 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3175 enum btrfs_qgroup_rsv_type type)
3176 {
3177 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3178 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3179 return;
3180 if (num_bytes == 0)
3181 return;
3182
3183 spin_lock(&root->qgroup_meta_rsv_lock);
3184 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3185 root->qgroup_meta_rsv_prealloc += num_bytes;
3186 else
3187 root->qgroup_meta_rsv_pertrans += num_bytes;
3188 spin_unlock(&root->qgroup_meta_rsv_lock);
3189 }
3190
3191 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3192 enum btrfs_qgroup_rsv_type type)
3193 {
3194 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3195 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3196 return 0;
3197 if (num_bytes == 0)
3198 return 0;
3199
3200 spin_lock(&root->qgroup_meta_rsv_lock);
3201 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3202 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3203 num_bytes);
3204 root->qgroup_meta_rsv_prealloc -= num_bytes;
3205 } else {
3206 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3207 num_bytes);
3208 root->qgroup_meta_rsv_pertrans -= num_bytes;
3209 }
3210 spin_unlock(&root->qgroup_meta_rsv_lock);
3211 return num_bytes;
3212 }
3213
3214 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3215 enum btrfs_qgroup_rsv_type type, bool enforce)
3216 {
3217 struct btrfs_fs_info *fs_info = root->fs_info;
3218 int ret;
3219
3220 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3221 !is_fstree(root->objectid) || num_bytes == 0)
3222 return 0;
3223
3224 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3225 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3226 ret = qgroup_reserve(root, num_bytes, enforce, type);
3227 if (ret < 0)
3228 return ret;
3229 /*
3230 * Record what we have reserved into root.
3231 *
3232 * To avoid quota disabled->enabled underflow.
3233 * In that case, we may try to free space we haven't reserved
3234 * (since quota was disabled), so record what we reserved into root.
3235 * And ensure later release won't underflow this number.
3236 */
3237 add_root_meta_rsv(root, num_bytes, type);
3238 return ret;
3239 }
3240
3241 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3242 {
3243 struct btrfs_fs_info *fs_info = root->fs_info;
3244
3245 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3246 !is_fstree(root->objectid))
3247 return;
3248
3249 /* TODO: Update trace point to handle such free */
3250 trace_qgroup_meta_free_all_pertrans(root);
3251 /* Special value -1 means to free all reserved space */
3252 btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3253 BTRFS_QGROUP_RSV_META_PERTRANS);
3254 }
3255
3256 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3257 enum btrfs_qgroup_rsv_type type)
3258 {
3259 struct btrfs_fs_info *fs_info = root->fs_info;
3260
3261 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3262 !is_fstree(root->objectid))
3263 return;
3264
3265 /*
3266 * reservation for META_PREALLOC can happen before quota is enabled,
3267 * which can lead to underflow.
3268 * Here ensure we will only free what we really have reserved.
3269 */
3270 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3271 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3272 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3273 btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3274 }
3275
3276 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3277 int num_bytes)
3278 {
3279 struct btrfs_root *quota_root = fs_info->quota_root;
3280 struct btrfs_qgroup *qgroup;
3281 struct ulist_node *unode;
3282 struct ulist_iterator uiter;
3283 int ret = 0;
3284
3285 if (num_bytes == 0)
3286 return;
3287 if (!quota_root)
3288 return;
3289
3290 spin_lock(&fs_info->qgroup_lock);
3291 qgroup = find_qgroup_rb(fs_info, ref_root);
3292 if (!qgroup)
3293 goto out;
3294 ulist_reinit(fs_info->qgroup_ulist);
3295 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3296 qgroup_to_aux(qgroup), GFP_ATOMIC);
3297 if (ret < 0)
3298 goto out;
3299 ULIST_ITER_INIT(&uiter);
3300 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3301 struct btrfs_qgroup *qg;
3302 struct btrfs_qgroup_list *glist;
3303
3304 qg = unode_aux_to_qgroup(unode);
3305
3306 qgroup_rsv_release(fs_info, qg, num_bytes,
3307 BTRFS_QGROUP_RSV_META_PREALLOC);
3308 qgroup_rsv_add(fs_info, qg, num_bytes,
3309 BTRFS_QGROUP_RSV_META_PERTRANS);
3310 list_for_each_entry(glist, &qg->groups, next_group) {
3311 ret = ulist_add(fs_info->qgroup_ulist,
3312 glist->group->qgroupid,
3313 qgroup_to_aux(glist->group), GFP_ATOMIC);
3314 if (ret < 0)
3315 goto out;
3316 }
3317 }
3318 out:
3319 spin_unlock(&fs_info->qgroup_lock);
3320 }
3321
3322 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3323 {
3324 struct btrfs_fs_info *fs_info = root->fs_info;
3325
3326 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3327 !is_fstree(root->objectid))
3328 return;
3329 /* Same as btrfs_qgroup_free_meta_prealloc() */
3330 num_bytes = sub_root_meta_rsv(root, num_bytes,
3331 BTRFS_QGROUP_RSV_META_PREALLOC);
3332 trace_qgroup_meta_convert(root, num_bytes);
3333 qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3334 }
3335
3336 /*
3337 * Check qgroup reserved space leaking, normally at destroy inode
3338 * time
3339 */
3340 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3341 {
3342 struct extent_changeset changeset;
3343 struct ulist_node *unode;
3344 struct ulist_iterator iter;
3345 int ret;
3346
3347 extent_changeset_init(&changeset);
3348 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3349 EXTENT_QGROUP_RESERVED, &changeset);
3350
3351 WARN_ON(ret < 0);
3352 if (WARN_ON(changeset.bytes_changed)) {
3353 ULIST_ITER_INIT(&iter);
3354 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3355 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3356 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3357 inode->i_ino, unode->val, unode->aux);
3358 }
3359 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3360 BTRFS_I(inode)->root->objectid,
3361 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3362
3363 }
3364 extent_changeset_release(&changeset);
3365 }
Linux with added FPC-III support