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