]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - fs/btrfs/qgroup.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:


ac9690f36a94b34084f116b352530fdbdd952ef3
[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
25
26 /* TODO XXX FIXME
27 * - subvol delete -> delete when ref goes to 0? delete limits also?
28 * - reorganize keys
29 * - compressed
30 * - sync
31 * - copy also limits on subvol creation
32 * - limit
33 * - caches fuer ulists
34 * - performance benchmarks
35 * - check all ioctl parameters
36 */
37
38 /*
39 * Helpers to access qgroup reservation
40 *
41 * Callers should ensure the lock context and type are valid
42 */
43
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45 {
46 u64 ret = 0;
47 int i;
48
49 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 ret += qgroup->rsv.values[i];
51
52 return ret;
53 }
54
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57 {
58 if (type == BTRFS_QGROUP_RSV_DATA)
59 return "data";
60 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 return "meta_pertrans";
62 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 return "meta_prealloc";
64 return NULL;
65 }
66 #endif
67
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 struct btrfs_qgroup *qgroup, u64 num_bytes,
70 enum btrfs_qgroup_rsv_type type)
71 {
72 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 qgroup->rsv.values[type] += num_bytes;
74 }
75
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 struct btrfs_qgroup *qgroup, u64 num_bytes,
78 enum btrfs_qgroup_rsv_type type)
79 {
80 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 if (qgroup->rsv.values[type] >= num_bytes) {
82 qgroup->rsv.values[type] -= num_bytes;
83 return;
84 }
85 #ifdef CONFIG_BTRFS_DEBUG
86 WARN_RATELIMIT(1,
87 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 qgroup->qgroupid, qgroup_rsv_type_str(type),
89 qgroup->rsv.values[type], num_bytes);
90 #endif
91 qgroup->rsv.values[type] = 0;
92 }
93
94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 struct btrfs_qgroup *dest,
96 struct btrfs_qgroup *src)
97 {
98 int i;
99
100 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
102 }
103
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 struct btrfs_qgroup *dest,
106 struct btrfs_qgroup *src)
107 {
108 int i;
109
110 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
112 }
113
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115 int mod)
116 {
117 if (qg->old_refcnt < seq)
118 qg->old_refcnt = seq;
119 qg->old_refcnt += mod;
120 }
121
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123 int mod)
124 {
125 if (qg->new_refcnt < seq)
126 qg->new_refcnt = seq;
127 qg->new_refcnt += mod;
128 }
129
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131 {
132 if (qg->old_refcnt < seq)
133 return 0;
134 return qg->old_refcnt - seq;
135 }
136
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138 {
139 if (qg->new_refcnt < seq)
140 return 0;
141 return qg->new_refcnt - seq;
142 }
143
144 /*
145 * glue structure to represent the relations between qgroups.
146 */
147 struct btrfs_qgroup_list {
148 struct list_head next_group;
149 struct list_head next_member;
150 struct btrfs_qgroup *group;
151 struct btrfs_qgroup *member;
152 };
153
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 {
156 return (u64)(uintptr_t)qg;
157 }
158
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 {
161 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162 }
163
164 static int
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 int init_flags);
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171 u64 qgroupid)
172 {
173 struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 struct btrfs_qgroup *qgroup;
175
176 while (n) {
177 qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 if (qgroup->qgroupid < qgroupid)
179 n = n->rb_left;
180 else if (qgroup->qgroupid > qgroupid)
181 n = n->rb_right;
182 else
183 return qgroup;
184 }
185 return NULL;
186 }
187
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190 u64 qgroupid)
191 {
192 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 struct rb_node *parent = NULL;
194 struct btrfs_qgroup *qgroup;
195
196 while (*p) {
197 parent = *p;
198 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199
200 if (qgroup->qgroupid < qgroupid)
201 p = &(*p)->rb_left;
202 else if (qgroup->qgroupid > qgroupid)
203 p = &(*p)->rb_right;
204 else
205 return qgroup;
206 }
207
208 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 if (!qgroup)
210 return ERR_PTR(-ENOMEM);
211
212 qgroup->qgroupid = qgroupid;
213 INIT_LIST_HEAD(&qgroup->groups);
214 INIT_LIST_HEAD(&qgroup->members);
215 INIT_LIST_HEAD(&qgroup->dirty);
216
217 rb_link_node(&qgroup->node, parent, p);
218 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219
220 return qgroup;
221 }
222
223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
224 {
225 struct btrfs_qgroup_list *list;
226
227 list_del(&qgroup->dirty);
228 while (!list_empty(&qgroup->groups)) {
229 list = list_first_entry(&qgroup->groups,
230 struct btrfs_qgroup_list, next_group);
231 list_del(&list->next_group);
232 list_del(&list->next_member);
233 kfree(list);
234 }
235
236 while (!list_empty(&qgroup->members)) {
237 list = list_first_entry(&qgroup->members,
238 struct btrfs_qgroup_list, next_member);
239 list_del(&list->next_group);
240 list_del(&list->next_member);
241 kfree(list);
242 }
243 kfree(qgroup);
244 }
245
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248 {
249 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250
251 if (!qgroup)
252 return -ENOENT;
253
254 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 __del_qgroup_rb(qgroup);
256 return 0;
257 }
258
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 u64 memberid, u64 parentid)
262 {
263 struct btrfs_qgroup *member;
264 struct btrfs_qgroup *parent;
265 struct btrfs_qgroup_list *list;
266
267 member = find_qgroup_rb(fs_info, memberid);
268 parent = find_qgroup_rb(fs_info, parentid);
269 if (!member || !parent)
270 return -ENOENT;
271
272 list = kzalloc(sizeof(*list), GFP_ATOMIC);
273 if (!list)
274 return -ENOMEM;
275
276 list->group = parent;
277 list->member = member;
278 list_add_tail(&list->next_group, &member->groups);
279 list_add_tail(&list->next_member, &parent->members);
280
281 return 0;
282 }
283
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 u64 memberid, u64 parentid)
287 {
288 struct btrfs_qgroup *member;
289 struct btrfs_qgroup *parent;
290 struct btrfs_qgroup_list *list;
291
292 member = find_qgroup_rb(fs_info, memberid);
293 parent = find_qgroup_rb(fs_info, parentid);
294 if (!member || !parent)
295 return -ENOENT;
296
297 list_for_each_entry(list, &member->groups, next_group) {
298 if (list->group == parent) {
299 list_del(&list->next_group);
300 list_del(&list->next_member);
301 kfree(list);
302 return 0;
303 }
304 }
305 return -ENOENT;
306 }
307
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310 u64 rfer, u64 excl)
311 {
312 struct btrfs_qgroup *qgroup;
313
314 qgroup = find_qgroup_rb(fs_info, qgroupid);
315 if (!qgroup)
316 return -EINVAL;
317 if (qgroup->rfer != rfer || qgroup->excl != excl)
318 return -EINVAL;
319 return 0;
320 }
321 #endif
322
323 /*
324 * The full config is read in one go, only called from open_ctree()
325 * It doesn't use any locking, as at this point we're still single-threaded
326 */
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328 {
329 struct btrfs_key key;
330 struct btrfs_key found_key;
331 struct btrfs_root *quota_root = fs_info->quota_root;
332 struct btrfs_path *path = NULL;
333 struct extent_buffer *l;
334 int slot;
335 int ret = 0;
336 u64 flags = 0;
337 u64 rescan_progress = 0;
338
339 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
340 return 0;
341
342 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 if (!fs_info->qgroup_ulist) {
344 ret = -ENOMEM;
345 goto out;
346 }
347
348 path = btrfs_alloc_path();
349 if (!path) {
350 ret = -ENOMEM;
351 goto out;
352 }
353
354 /* default this to quota off, in case no status key is found */
355 fs_info->qgroup_flags = 0;
356
357 /*
358 * pass 1: read status, all qgroup infos and limits
359 */
360 key.objectid = 0;
361 key.type = 0;
362 key.offset = 0;
363 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
364 if (ret)
365 goto out;
366
367 while (1) {
368 struct btrfs_qgroup *qgroup;
369
370 slot = path->slots[0];
371 l = path->nodes[0];
372 btrfs_item_key_to_cpu(l, &found_key, slot);
373
374 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 struct btrfs_qgroup_status_item *ptr;
376
377 ptr = btrfs_item_ptr(l, slot,
378 struct btrfs_qgroup_status_item);
379
380 if (btrfs_qgroup_status_version(l, ptr) !=
381 BTRFS_QGROUP_STATUS_VERSION) {
382 btrfs_err(fs_info,
383 "old qgroup version, quota disabled");
384 goto out;
385 }
386 if (btrfs_qgroup_status_generation(l, ptr) !=
387 fs_info->generation) {
388 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 btrfs_err(fs_info,
390 "qgroup generation mismatch, marked as inconsistent");
391 }
392 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 ptr);
394 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
395 goto next1;
396 }
397
398 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
400 goto next1;
401
402 qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 btrfs_err(fs_info, "inconsistent qgroup config");
406 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407 }
408 if (!qgroup) {
409 qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 if (IS_ERR(qgroup)) {
411 ret = PTR_ERR(qgroup);
412 goto out;
413 }
414 }
415 switch (found_key.type) {
416 case BTRFS_QGROUP_INFO_KEY: {
417 struct btrfs_qgroup_info_item *ptr;
418
419 ptr = btrfs_item_ptr(l, slot,
420 struct btrfs_qgroup_info_item);
421 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 /* generation currently unused */
426 break;
427 }
428 case BTRFS_QGROUP_LIMIT_KEY: {
429 struct btrfs_qgroup_limit_item *ptr;
430
431 ptr = btrfs_item_ptr(l, slot,
432 struct btrfs_qgroup_limit_item);
433 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438 break;
439 }
440 }
441 next1:
442 ret = btrfs_next_item(quota_root, path);
443 if (ret < 0)
444 goto out;
445 if (ret)
446 break;
447 }
448 btrfs_release_path(path);
449
450 /*
451 * pass 2: read all qgroup relations
452 */
453 key.objectid = 0;
454 key.type = BTRFS_QGROUP_RELATION_KEY;
455 key.offset = 0;
456 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457 if (ret)
458 goto out;
459 while (1) {
460 slot = path->slots[0];
461 l = path->nodes[0];
462 btrfs_item_key_to_cpu(l, &found_key, slot);
463
464 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465 goto next2;
466
467 if (found_key.objectid > found_key.offset) {
468 /* parent <- member, not needed to build config */
469 /* FIXME should we omit the key completely? */
470 goto next2;
471 }
472
473 ret = add_relation_rb(fs_info, found_key.objectid,
474 found_key.offset);
475 if (ret == -ENOENT) {
476 btrfs_warn(fs_info,
477 "orphan qgroup relation 0x%llx->0x%llx",
478 found_key.objectid, found_key.offset);
479 ret = 0; /* ignore the error */
480 }
481 if (ret)
482 goto out;
483 next2:
484 ret = btrfs_next_item(quota_root, path);
485 if (ret < 0)
486 goto out;
487 if (ret)
488 break;
489 }
490 out:
491 fs_info->qgroup_flags |= flags;
492 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495 ret >= 0)
496 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 btrfs_free_path(path);
498
499 if (ret < 0) {
500 ulist_free(fs_info->qgroup_ulist);
501 fs_info->qgroup_ulist = NULL;
502 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
503 }
504
505 return ret < 0 ? ret : 0;
506 }
507
508 /*
509 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510 * first two are in single-threaded paths.And for the third one, we have set
511 * quota_root to be null with qgroup_lock held before, so it is safe to clean
512 * up the in-memory structures without qgroup_lock held.
513 */
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515 {
516 struct rb_node *n;
517 struct btrfs_qgroup *qgroup;
518
519 while ((n = rb_first(&fs_info->qgroup_tree))) {
520 qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 rb_erase(n, &fs_info->qgroup_tree);
522 __del_qgroup_rb(qgroup);
523 }
524 /*
525 * we call btrfs_free_qgroup_config() when umounting
526 * filesystem and disabling quota, so we set qgroup_ulist
527 * to be null here to avoid double free.
528 */
529 ulist_free(fs_info->qgroup_ulist);
530 fs_info->qgroup_ulist = NULL;
531 }
532
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 u64 dst)
535 {
536 int ret;
537 struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 struct btrfs_path *path;
539 struct btrfs_key key;
540
541 path = btrfs_alloc_path();
542 if (!path)
543 return -ENOMEM;
544
545 key.objectid = src;
546 key.type = BTRFS_QGROUP_RELATION_KEY;
547 key.offset = dst;
548
549 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550
551 btrfs_mark_buffer_dirty(path->nodes[0]);
552
553 btrfs_free_path(path);
554 return ret;
555 }
556
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 u64 dst)
559 {
560 int ret;
561 struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 struct btrfs_path *path;
563 struct btrfs_key key;
564
565 path = btrfs_alloc_path();
566 if (!path)
567 return -ENOMEM;
568
569 key.objectid = src;
570 key.type = BTRFS_QGROUP_RELATION_KEY;
571 key.offset = dst;
572
573 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574 if (ret < 0)
575 goto out;
576
577 if (ret > 0) {
578 ret = -ENOENT;
579 goto out;
580 }
581
582 ret = btrfs_del_item(trans, quota_root, path);
583 out:
584 btrfs_free_path(path);
585 return ret;
586 }
587
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 struct btrfs_root *quota_root, u64 qgroupid)
590 {
591 int ret;
592 struct btrfs_path *path;
593 struct btrfs_qgroup_info_item *qgroup_info;
594 struct btrfs_qgroup_limit_item *qgroup_limit;
595 struct extent_buffer *leaf;
596 struct btrfs_key key;
597
598 if (btrfs_is_testing(quota_root->fs_info))
599 return 0;
600
601 path = btrfs_alloc_path();
602 if (!path)
603 return -ENOMEM;
604
605 key.objectid = 0;
606 key.type = BTRFS_QGROUP_INFO_KEY;
607 key.offset = qgroupid;
608
609 /*
610 * Avoid a transaction abort by catching -EEXIST here. In that
611 * case, we proceed by re-initializing the existing structure
612 * on disk.
613 */
614
615 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 sizeof(*qgroup_info));
617 if (ret && ret != -EEXIST)
618 goto out;
619
620 leaf = path->nodes[0];
621 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 struct btrfs_qgroup_info_item);
623 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628
629 btrfs_mark_buffer_dirty(leaf);
630
631 btrfs_release_path(path);
632
633 key.type = BTRFS_QGROUP_LIMIT_KEY;
634 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 sizeof(*qgroup_limit));
636 if (ret && ret != -EEXIST)
637 goto out;
638
639 leaf = path->nodes[0];
640 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_qgroup_limit_item);
642 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647
648 btrfs_mark_buffer_dirty(leaf);
649
650 ret = 0;
651 out:
652 btrfs_free_path(path);
653 return ret;
654 }
655
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
657 {
658 int ret;
659 struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 struct btrfs_path *path;
661 struct btrfs_key key;
662
663 path = btrfs_alloc_path();
664 if (!path)
665 return -ENOMEM;
666
667 key.objectid = 0;
668 key.type = BTRFS_QGROUP_INFO_KEY;
669 key.offset = qgroupid;
670 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 if (ret < 0)
672 goto out;
673
674 if (ret > 0) {
675 ret = -ENOENT;
676 goto out;
677 }
678
679 ret = btrfs_del_item(trans, quota_root, path);
680 if (ret)
681 goto out;
682
683 btrfs_release_path(path);
684
685 key.type = BTRFS_QGROUP_LIMIT_KEY;
686 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687 if (ret < 0)
688 goto out;
689
690 if (ret > 0) {
691 ret = -ENOENT;
692 goto out;
693 }
694
695 ret = btrfs_del_item(trans, quota_root, path);
696
697 out:
698 btrfs_free_path(path);
699 return ret;
700 }
701
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 struct btrfs_qgroup *qgroup)
704 {
705 struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 struct btrfs_path *path;
707 struct btrfs_key key;
708 struct extent_buffer *l;
709 struct btrfs_qgroup_limit_item *qgroup_limit;
710 int ret;
711 int slot;
712
713 key.objectid = 0;
714 key.type = BTRFS_QGROUP_LIMIT_KEY;
715 key.offset = qgroup->qgroupid;
716
717 path = btrfs_alloc_path();
718 if (!path)
719 return -ENOMEM;
720
721 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
722 if (ret > 0)
723 ret = -ENOENT;
724
725 if (ret)
726 goto out;
727
728 l = path->nodes[0];
729 slot = path->slots[0];
730 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736
737 btrfs_mark_buffer_dirty(l);
738
739 out:
740 btrfs_free_path(path);
741 return ret;
742 }
743
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 struct btrfs_qgroup *qgroup)
746 {
747 struct btrfs_fs_info *fs_info = trans->fs_info;
748 struct btrfs_root *quota_root = fs_info->quota_root;
749 struct btrfs_path *path;
750 struct btrfs_key key;
751 struct extent_buffer *l;
752 struct btrfs_qgroup_info_item *qgroup_info;
753 int ret;
754 int slot;
755
756 if (btrfs_is_testing(fs_info))
757 return 0;
758
759 key.objectid = 0;
760 key.type = BTRFS_QGROUP_INFO_KEY;
761 key.offset = qgroup->qgroupid;
762
763 path = btrfs_alloc_path();
764 if (!path)
765 return -ENOMEM;
766
767 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
768 if (ret > 0)
769 ret = -ENOENT;
770
771 if (ret)
772 goto out;
773
774 l = path->nodes[0];
775 slot = path->slots[0];
776 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782
783 btrfs_mark_buffer_dirty(l);
784
785 out:
786 btrfs_free_path(path);
787 return ret;
788 }
789
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
791 {
792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *quota_root = fs_info->quota_root;
794 struct btrfs_path *path;
795 struct btrfs_key key;
796 struct extent_buffer *l;
797 struct btrfs_qgroup_status_item *ptr;
798 int ret;
799 int slot;
800
801 key.objectid = 0;
802 key.type = BTRFS_QGROUP_STATUS_KEY;
803 key.offset = 0;
804
805 path = btrfs_alloc_path();
806 if (!path)
807 return -ENOMEM;
808
809 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
810 if (ret > 0)
811 ret = -ENOENT;
812
813 if (ret)
814 goto out;
815
816 l = path->nodes[0];
817 slot = path->slots[0];
818 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 btrfs_set_qgroup_status_rescan(l, ptr,
822 fs_info->qgroup_rescan_progress.objectid);
823
824 btrfs_mark_buffer_dirty(l);
825
826 out:
827 btrfs_free_path(path);
828 return ret;
829 }
830
831 /*
832 * called with qgroup_lock held
833 */
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root)
836 {
837 struct btrfs_path *path;
838 struct btrfs_key key;
839 struct extent_buffer *leaf = NULL;
840 int ret;
841 int nr = 0;
842
843 path = btrfs_alloc_path();
844 if (!path)
845 return -ENOMEM;
846
847 path->leave_spinning = 1;
848
849 key.objectid = 0;
850 key.offset = 0;
851 key.type = 0;
852
853 while (1) {
854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855 if (ret < 0)
856 goto out;
857 leaf = path->nodes[0];
858 nr = btrfs_header_nritems(leaf);
859 if (!nr)
860 break;
861 /*
862 * delete the leaf one by one
863 * since the whole tree is going
864 * to be deleted.
865 */
866 path->slots[0] = 0;
867 ret = btrfs_del_items(trans, root, path, 0, nr);
868 if (ret)
869 goto out;
870
871 btrfs_release_path(path);
872 }
873 ret = 0;
874 out:
875 btrfs_free_path(path);
876 return ret;
877 }
878
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
880 {
881 struct btrfs_root *quota_root;
882 struct btrfs_root *tree_root = fs_info->tree_root;
883 struct btrfs_path *path = NULL;
884 struct btrfs_qgroup_status_item *ptr;
885 struct extent_buffer *leaf;
886 struct btrfs_key key;
887 struct btrfs_key found_key;
888 struct btrfs_qgroup *qgroup = NULL;
889 struct btrfs_trans_handle *trans = NULL;
890 int ret = 0;
891 int slot;
892
893 mutex_lock(&fs_info->qgroup_ioctl_lock);
894 if (fs_info->quota_root)
895 goto out;
896
897 /*
898 * 1 for quota root item
899 * 1 for BTRFS_QGROUP_STATUS item
900 *
901 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
902 * per subvolume. However those are not currently reserved since it
903 * would be a lot of overkill.
904 */
905 trans = btrfs_start_transaction(tree_root, 2);
906 if (IS_ERR(trans)) {
907 ret = PTR_ERR(trans);
908 trans = NULL;
909 goto out;
910 }
911
912 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
913 if (!fs_info->qgroup_ulist) {
914 ret = -ENOMEM;
915 btrfs_abort_transaction(trans, ret);
916 goto out;
917 }
918
919 /*
920 * initially create the quota tree
921 */
922 quota_root = btrfs_create_tree(trans, fs_info,
923 BTRFS_QUOTA_TREE_OBJECTID);
924 if (IS_ERR(quota_root)) {
925 ret = PTR_ERR(quota_root);
926 btrfs_abort_transaction(trans, ret);
927 goto out;
928 }
929
930 path = btrfs_alloc_path();
931 if (!path) {
932 ret = -ENOMEM;
933 btrfs_abort_transaction(trans, ret);
934 goto out_free_root;
935 }
936
937 key.objectid = 0;
938 key.type = BTRFS_QGROUP_STATUS_KEY;
939 key.offset = 0;
940
941 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
942 sizeof(*ptr));
943 if (ret) {
944 btrfs_abort_transaction(trans, ret);
945 goto out_free_path;
946 }
947
948 leaf = path->nodes[0];
949 ptr = btrfs_item_ptr(leaf, path->slots[0],
950 struct btrfs_qgroup_status_item);
951 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
952 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
953 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
954 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
955 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
956 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
957
958 btrfs_mark_buffer_dirty(leaf);
959
960 key.objectid = 0;
961 key.type = BTRFS_ROOT_REF_KEY;
962 key.offset = 0;
963
964 btrfs_release_path(path);
965 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
966 if (ret > 0)
967 goto out_add_root;
968 if (ret < 0) {
969 btrfs_abort_transaction(trans, ret);
970 goto out_free_path;
971 }
972
973 while (1) {
974 slot = path->slots[0];
975 leaf = path->nodes[0];
976 btrfs_item_key_to_cpu(leaf, &found_key, slot);
977
978 if (found_key.type == BTRFS_ROOT_REF_KEY) {
979 ret = add_qgroup_item(trans, quota_root,
980 found_key.offset);
981 if (ret) {
982 btrfs_abort_transaction(trans, ret);
983 goto out_free_path;
984 }
985
986 qgroup = add_qgroup_rb(fs_info, found_key.offset);
987 if (IS_ERR(qgroup)) {
988 ret = PTR_ERR(qgroup);
989 btrfs_abort_transaction(trans, ret);
990 goto out_free_path;
991 }
992 }
993 ret = btrfs_next_item(tree_root, path);
994 if (ret < 0) {
995 btrfs_abort_transaction(trans, ret);
996 goto out_free_path;
997 }
998 if (ret)
999 break;
1000 }
1001
1002 out_add_root:
1003 btrfs_release_path(path);
1004 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1005 if (ret) {
1006 btrfs_abort_transaction(trans, ret);
1007 goto out_free_path;
1008 }
1009
1010 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011 if (IS_ERR(qgroup)) {
1012 ret = PTR_ERR(qgroup);
1013 btrfs_abort_transaction(trans, ret);
1014 goto out_free_path;
1015 }
1016 spin_lock(&fs_info->qgroup_lock);
1017 fs_info->quota_root = quota_root;
1018 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1019 spin_unlock(&fs_info->qgroup_lock);
1020
1021 ret = btrfs_commit_transaction(trans);
1022 trans = NULL;
1023 if (ret)
1024 goto out_free_path;
1025
1026 ret = qgroup_rescan_init(fs_info, 0, 1);
1027 if (!ret) {
1028 qgroup_rescan_zero_tracking(fs_info);
1029 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1030 &fs_info->qgroup_rescan_work);
1031 }
1032
1033 out_free_path:
1034 btrfs_free_path(path);
1035 out_free_root:
1036 if (ret) {
1037 free_extent_buffer(quota_root->node);
1038 free_extent_buffer(quota_root->commit_root);
1039 kfree(quota_root);
1040 }
1041 out:
1042 if (ret) {
1043 ulist_free(fs_info->qgroup_ulist);
1044 fs_info->qgroup_ulist = NULL;
1045 if (trans)
1046 btrfs_end_transaction(trans);
1047 }
1048 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1049 return ret;
1050 }
1051
1052 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1053 {
1054 struct btrfs_root *quota_root;
1055 struct btrfs_trans_handle *trans = NULL;
1056 int ret = 0;
1057
1058 mutex_lock(&fs_info->qgroup_ioctl_lock);
1059 if (!fs_info->quota_root)
1060 goto out;
1061
1062 /*
1063 * 1 For the root item
1064 *
1065 * We should also reserve enough items for the quota tree deletion in
1066 * btrfs_clean_quota_tree but this is not done.
1067 */
1068 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1069 if (IS_ERR(trans)) {
1070 ret = PTR_ERR(trans);
1071 goto out;
1072 }
1073
1074 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1075 btrfs_qgroup_wait_for_completion(fs_info, false);
1076 spin_lock(&fs_info->qgroup_lock);
1077 quota_root = fs_info->quota_root;
1078 fs_info->quota_root = NULL;
1079 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1080 spin_unlock(&fs_info->qgroup_lock);
1081
1082 btrfs_free_qgroup_config(fs_info);
1083
1084 ret = btrfs_clean_quota_tree(trans, quota_root);
1085 if (ret) {
1086 btrfs_abort_transaction(trans, ret);
1087 goto end_trans;
1088 }
1089
1090 ret = btrfs_del_root(trans, &quota_root->root_key);
1091 if (ret) {
1092 btrfs_abort_transaction(trans, ret);
1093 goto end_trans;
1094 }
1095
1096 list_del(&quota_root->dirty_list);
1097
1098 btrfs_tree_lock(quota_root->node);
1099 clean_tree_block(fs_info, quota_root->node);
1100 btrfs_tree_unlock(quota_root->node);
1101 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1102
1103 free_extent_buffer(quota_root->node);
1104 free_extent_buffer(quota_root->commit_root);
1105 kfree(quota_root);
1106
1107 end_trans:
1108 ret = btrfs_end_transaction(trans);
1109 out:
1110 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1111 return ret;
1112 }
1113
1114 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1115 struct btrfs_qgroup *qgroup)
1116 {
1117 if (list_empty(&qgroup->dirty))
1118 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1119 }
1120
1121 /*
1122 * The easy accounting, we're updating qgroup relationship whose child qgroup
1123 * only has exclusive extents.
1124 *
1125 * In this case, all exclsuive extents will also be exlusive for parent, so
1126 * excl/rfer just get added/removed.
1127 *
1128 * So is qgroup reservation space, which should also be added/removed to
1129 * parent.
1130 * Or when child tries to release reservation space, parent will underflow its
1131 * reservation (for relationship adding case).
1132 *
1133 * Caller should hold fs_info->qgroup_lock.
1134 */
1135 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1136 struct ulist *tmp, u64 ref_root,
1137 struct btrfs_qgroup *src, int sign)
1138 {
1139 struct btrfs_qgroup *qgroup;
1140 struct btrfs_qgroup_list *glist;
1141 struct ulist_node *unode;
1142 struct ulist_iterator uiter;
1143 u64 num_bytes = src->excl;
1144 int ret = 0;
1145
1146 qgroup = find_qgroup_rb(fs_info, ref_root);
1147 if (!qgroup)
1148 goto out;
1149
1150 qgroup->rfer += sign * num_bytes;
1151 qgroup->rfer_cmpr += sign * num_bytes;
1152
1153 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1154 qgroup->excl += sign * num_bytes;
1155 qgroup->excl_cmpr += sign * num_bytes;
1156
1157 if (sign > 0)
1158 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1159 else
1160 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1161
1162 qgroup_dirty(fs_info, qgroup);
1163
1164 /* Get all of the parent groups that contain this qgroup */
1165 list_for_each_entry(glist, &qgroup->groups, next_group) {
1166 ret = ulist_add(tmp, glist->group->qgroupid,
1167 qgroup_to_aux(glist->group), GFP_ATOMIC);
1168 if (ret < 0)
1169 goto out;
1170 }
1171
1172 /* Iterate all of the parents and adjust their reference counts */
1173 ULIST_ITER_INIT(&uiter);
1174 while ((unode = ulist_next(tmp, &uiter))) {
1175 qgroup = unode_aux_to_qgroup(unode);
1176 qgroup->rfer += sign * num_bytes;
1177 qgroup->rfer_cmpr += sign * num_bytes;
1178 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1179 qgroup->excl += sign * num_bytes;
1180 if (sign > 0)
1181 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1182 else
1183 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1184 qgroup->excl_cmpr += sign * num_bytes;
1185 qgroup_dirty(fs_info, qgroup);
1186
1187 /* Add any parents of the parents */
1188 list_for_each_entry(glist, &qgroup->groups, next_group) {
1189 ret = ulist_add(tmp, glist->group->qgroupid,
1190 qgroup_to_aux(glist->group), GFP_ATOMIC);
1191 if (ret < 0)
1192 goto out;
1193 }
1194 }
1195 ret = 0;
1196 out:
1197 return ret;
1198 }
1199
1200
1201 /*
1202 * Quick path for updating qgroup with only excl refs.
1203 *
1204 * In that case, just update all parent will be enough.
1205 * Or we needs to do a full rescan.
1206 * Caller should also hold fs_info->qgroup_lock.
1207 *
1208 * Return 0 for quick update, return >0 for need to full rescan
1209 * and mark INCONSISTENT flag.
1210 * Return < 0 for other error.
1211 */
1212 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1213 struct ulist *tmp, u64 src, u64 dst,
1214 int sign)
1215 {
1216 struct btrfs_qgroup *qgroup;
1217 int ret = 1;
1218 int err = 0;
1219
1220 qgroup = find_qgroup_rb(fs_info, src);
1221 if (!qgroup)
1222 goto out;
1223 if (qgroup->excl == qgroup->rfer) {
1224 ret = 0;
1225 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1226 qgroup, sign);
1227 if (err < 0) {
1228 ret = err;
1229 goto out;
1230 }
1231 }
1232 out:
1233 if (ret)
1234 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1235 return ret;
1236 }
1237
1238 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1239 u64 dst)
1240 {
1241 struct btrfs_fs_info *fs_info = trans->fs_info;
1242 struct btrfs_root *quota_root;
1243 struct btrfs_qgroup *parent;
1244 struct btrfs_qgroup *member;
1245 struct btrfs_qgroup_list *list;
1246 struct ulist *tmp;
1247 int ret = 0;
1248
1249 /* Check the level of src and dst first */
1250 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1251 return -EINVAL;
1252
1253 tmp = ulist_alloc(GFP_KERNEL);
1254 if (!tmp)
1255 return -ENOMEM;
1256
1257 mutex_lock(&fs_info->qgroup_ioctl_lock);
1258 quota_root = fs_info->quota_root;
1259 if (!quota_root) {
1260 ret = -EINVAL;
1261 goto out;
1262 }
1263 member = find_qgroup_rb(fs_info, src);
1264 parent = find_qgroup_rb(fs_info, dst);
1265 if (!member || !parent) {
1266 ret = -EINVAL;
1267 goto out;
1268 }
1269
1270 /* check if such qgroup relation exist firstly */
1271 list_for_each_entry(list, &member->groups, next_group) {
1272 if (list->group == parent) {
1273 ret = -EEXIST;
1274 goto out;
1275 }
1276 }
1277
1278 ret = add_qgroup_relation_item(trans, src, dst);
1279 if (ret)
1280 goto out;
1281
1282 ret = add_qgroup_relation_item(trans, dst, src);
1283 if (ret) {
1284 del_qgroup_relation_item(trans, src, dst);
1285 goto out;
1286 }
1287
1288 spin_lock(&fs_info->qgroup_lock);
1289 ret = add_relation_rb(fs_info, src, dst);
1290 if (ret < 0) {
1291 spin_unlock(&fs_info->qgroup_lock);
1292 goto out;
1293 }
1294 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1295 spin_unlock(&fs_info->qgroup_lock);
1296 out:
1297 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1298 ulist_free(tmp);
1299 return ret;
1300 }
1301
1302 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1303 u64 dst)
1304 {
1305 struct btrfs_fs_info *fs_info = trans->fs_info;
1306 struct btrfs_root *quota_root;
1307 struct btrfs_qgroup *parent;
1308 struct btrfs_qgroup *member;
1309 struct btrfs_qgroup_list *list;
1310 struct ulist *tmp;
1311 int ret = 0;
1312 int err;
1313
1314 tmp = ulist_alloc(GFP_KERNEL);
1315 if (!tmp)
1316 return -ENOMEM;
1317
1318 quota_root = fs_info->quota_root;
1319 if (!quota_root) {
1320 ret = -EINVAL;
1321 goto out;
1322 }
1323
1324 member = find_qgroup_rb(fs_info, src);
1325 parent = find_qgroup_rb(fs_info, dst);
1326 if (!member || !parent) {
1327 ret = -EINVAL;
1328 goto out;
1329 }
1330
1331 /* check if such qgroup relation exist firstly */
1332 list_for_each_entry(list, &member->groups, next_group) {
1333 if (list->group == parent)
1334 goto exist;
1335 }
1336 ret = -ENOENT;
1337 goto out;
1338 exist:
1339 ret = del_qgroup_relation_item(trans, src, dst);
1340 err = del_qgroup_relation_item(trans, dst, src);
1341 if (err && !ret)
1342 ret = err;
1343
1344 spin_lock(&fs_info->qgroup_lock);
1345 del_relation_rb(fs_info, src, dst);
1346 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1347 spin_unlock(&fs_info->qgroup_lock);
1348 out:
1349 ulist_free(tmp);
1350 return ret;
1351 }
1352
1353 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1354 u64 dst)
1355 {
1356 struct btrfs_fs_info *fs_info = trans->fs_info;
1357 int ret = 0;
1358
1359 mutex_lock(&fs_info->qgroup_ioctl_lock);
1360 ret = __del_qgroup_relation(trans, src, dst);
1361 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1362
1363 return ret;
1364 }
1365
1366 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1367 {
1368 struct btrfs_fs_info *fs_info = trans->fs_info;
1369 struct btrfs_root *quota_root;
1370 struct btrfs_qgroup *qgroup;
1371 int ret = 0;
1372
1373 mutex_lock(&fs_info->qgroup_ioctl_lock);
1374 quota_root = fs_info->quota_root;
1375 if (!quota_root) {
1376 ret = -EINVAL;
1377 goto out;
1378 }
1379 qgroup = find_qgroup_rb(fs_info, qgroupid);
1380 if (qgroup) {
1381 ret = -EEXIST;
1382 goto out;
1383 }
1384
1385 ret = add_qgroup_item(trans, quota_root, qgroupid);
1386 if (ret)
1387 goto out;
1388
1389 spin_lock(&fs_info->qgroup_lock);
1390 qgroup = add_qgroup_rb(fs_info, qgroupid);
1391 spin_unlock(&fs_info->qgroup_lock);
1392
1393 if (IS_ERR(qgroup))
1394 ret = PTR_ERR(qgroup);
1395 out:
1396 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1397 return ret;
1398 }
1399
1400 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1401 {
1402 struct btrfs_fs_info *fs_info = trans->fs_info;
1403 struct btrfs_root *quota_root;
1404 struct btrfs_qgroup *qgroup;
1405 struct btrfs_qgroup_list *list;
1406 int ret = 0;
1407
1408 mutex_lock(&fs_info->qgroup_ioctl_lock);
1409 quota_root = fs_info->quota_root;
1410 if (!quota_root) {
1411 ret = -EINVAL;
1412 goto out;
1413 }
1414
1415 qgroup = find_qgroup_rb(fs_info, qgroupid);
1416 if (!qgroup) {
1417 ret = -ENOENT;
1418 goto out;
1419 } else {
1420 /* check if there are no children of this qgroup */
1421 if (!list_empty(&qgroup->members)) {
1422 ret = -EBUSY;
1423 goto out;
1424 }
1425 }
1426 ret = del_qgroup_item(trans, qgroupid);
1427 if (ret && ret != -ENOENT)
1428 goto out;
1429
1430 while (!list_empty(&qgroup->groups)) {
1431 list = list_first_entry(&qgroup->groups,
1432 struct btrfs_qgroup_list, next_group);
1433 ret = __del_qgroup_relation(trans, qgroupid,
1434 list->group->qgroupid);
1435 if (ret)
1436 goto out;
1437 }
1438
1439 spin_lock(&fs_info->qgroup_lock);
1440 del_qgroup_rb(fs_info, qgroupid);
1441 spin_unlock(&fs_info->qgroup_lock);
1442 out:
1443 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1444 return ret;
1445 }
1446
1447 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1448 struct btrfs_qgroup_limit *limit)
1449 {
1450 struct btrfs_fs_info *fs_info = trans->fs_info;
1451 struct btrfs_root *quota_root;
1452 struct btrfs_qgroup *qgroup;
1453 int ret = 0;
1454 /* Sometimes we would want to clear the limit on this qgroup.
1455 * To meet this requirement, we treat the -1 as a special value
1456 * which tell kernel to clear the limit on this qgroup.
1457 */
1458 const u64 CLEAR_VALUE = -1;
1459
1460 mutex_lock(&fs_info->qgroup_ioctl_lock);
1461 quota_root = fs_info->quota_root;
1462 if (!quota_root) {
1463 ret = -EINVAL;
1464 goto out;
1465 }
1466
1467 qgroup = find_qgroup_rb(fs_info, qgroupid);
1468 if (!qgroup) {
1469 ret = -ENOENT;
1470 goto out;
1471 }
1472
1473 spin_lock(&fs_info->qgroup_lock);
1474 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1475 if (limit->max_rfer == CLEAR_VALUE) {
1476 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1477 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1478 qgroup->max_rfer = 0;
1479 } else {
1480 qgroup->max_rfer = limit->max_rfer;
1481 }
1482 }
1483 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1484 if (limit->max_excl == CLEAR_VALUE) {
1485 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1486 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1487 qgroup->max_excl = 0;
1488 } else {
1489 qgroup->max_excl = limit->max_excl;
1490 }
1491 }
1492 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1493 if (limit->rsv_rfer == CLEAR_VALUE) {
1494 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1495 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1496 qgroup->rsv_rfer = 0;
1497 } else {
1498 qgroup->rsv_rfer = limit->rsv_rfer;
1499 }
1500 }
1501 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1502 if (limit->rsv_excl == CLEAR_VALUE) {
1503 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1504 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1505 qgroup->rsv_excl = 0;
1506 } else {
1507 qgroup->rsv_excl = limit->rsv_excl;
1508 }
1509 }
1510 qgroup->lim_flags |= limit->flags;
1511
1512 spin_unlock(&fs_info->qgroup_lock);
1513
1514 ret = update_qgroup_limit_item(trans, qgroup);
1515 if (ret) {
1516 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1517 btrfs_info(fs_info, "unable to update quota limit for %llu",
1518 qgroupid);
1519 }
1520
1521 out:
1522 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1523 return ret;
1524 }
1525
1526 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1527 struct btrfs_delayed_ref_root *delayed_refs,
1528 struct btrfs_qgroup_extent_record *record)
1529 {
1530 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1531 struct rb_node *parent_node = NULL;
1532 struct btrfs_qgroup_extent_record *entry;
1533 u64 bytenr = record->bytenr;
1534
1535 lockdep_assert_held(&delayed_refs->lock);
1536 trace_btrfs_qgroup_trace_extent(fs_info, record);
1537
1538 while (*p) {
1539 parent_node = *p;
1540 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1541 node);
1542 if (bytenr < entry->bytenr)
1543 p = &(*p)->rb_left;
1544 else if (bytenr > entry->bytenr)
1545 p = &(*p)->rb_right;
1546 else
1547 return 1;
1548 }
1549
1550 rb_link_node(&record->node, parent_node, p);
1551 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1552 return 0;
1553 }
1554
1555 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1556 struct btrfs_qgroup_extent_record *qrecord)
1557 {
1558 struct ulist *old_root;
1559 u64 bytenr = qrecord->bytenr;
1560 int ret;
1561
1562 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1563 if (ret < 0) {
1564 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1565 btrfs_warn(fs_info,
1566 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1567 ret);
1568 return 0;
1569 }
1570
1571 /*
1572 * Here we don't need to get the lock of
1573 * trans->transaction->delayed_refs, since inserted qrecord won't
1574 * be deleted, only qrecord->node may be modified (new qrecord insert)
1575 *
1576 * So modifying qrecord->old_roots is safe here
1577 */
1578 qrecord->old_roots = old_root;
1579 return 0;
1580 }
1581
1582 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1583 u64 num_bytes, gfp_t gfp_flag)
1584 {
1585 struct btrfs_fs_info *fs_info = trans->fs_info;
1586 struct btrfs_qgroup_extent_record *record;
1587 struct btrfs_delayed_ref_root *delayed_refs;
1588 int ret;
1589
1590 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1591 || bytenr == 0 || num_bytes == 0)
1592 return 0;
1593 record = kmalloc(sizeof(*record), gfp_flag);
1594 if (!record)
1595 return -ENOMEM;
1596
1597 delayed_refs = &trans->transaction->delayed_refs;
1598 record->bytenr = bytenr;
1599 record->num_bytes = num_bytes;
1600 record->old_roots = NULL;
1601
1602 spin_lock(&delayed_refs->lock);
1603 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1604 spin_unlock(&delayed_refs->lock);
1605 if (ret > 0) {
1606 kfree(record);
1607 return 0;
1608 }
1609 return btrfs_qgroup_trace_extent_post(fs_info, record);
1610 }
1611
1612 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1613 struct extent_buffer *eb)
1614 {
1615 struct btrfs_fs_info *fs_info = trans->fs_info;
1616 int nr = btrfs_header_nritems(eb);
1617 int i, extent_type, ret;
1618 struct btrfs_key key;
1619 struct btrfs_file_extent_item *fi;
1620 u64 bytenr, num_bytes;
1621
1622 /* We can be called directly from walk_up_proc() */
1623 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1624 return 0;
1625
1626 for (i = 0; i < nr; i++) {
1627 btrfs_item_key_to_cpu(eb, &key, i);
1628
1629 if (key.type != BTRFS_EXTENT_DATA_KEY)
1630 continue;
1631
1632 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1633 /* filter out non qgroup-accountable extents */
1634 extent_type = btrfs_file_extent_type(eb, fi);
1635
1636 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1637 continue;
1638
1639 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1640 if (!bytenr)
1641 continue;
1642
1643 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1644
1645 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1646 GFP_NOFS);
1647 if (ret)
1648 return ret;
1649 }
1650 cond_resched();
1651 return 0;
1652 }
1653
1654 /*
1655 * Walk up the tree from the bottom, freeing leaves and any interior
1656 * nodes which have had all slots visited. If a node (leaf or
1657 * interior) is freed, the node above it will have it's slot
1658 * incremented. The root node will never be freed.
1659 *
1660 * At the end of this function, we should have a path which has all
1661 * slots incremented to the next position for a search. If we need to
1662 * read a new node it will be NULL and the node above it will have the
1663 * correct slot selected for a later read.
1664 *
1665 * If we increment the root nodes slot counter past the number of
1666 * elements, 1 is returned to signal completion of the search.
1667 */
1668 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1669 {
1670 int level = 0;
1671 int nr, slot;
1672 struct extent_buffer *eb;
1673
1674 if (root_level == 0)
1675 return 1;
1676
1677 while (level <= root_level) {
1678 eb = path->nodes[level];
1679 nr = btrfs_header_nritems(eb);
1680 path->slots[level]++;
1681 slot = path->slots[level];
1682 if (slot >= nr || level == 0) {
1683 /*
1684 * Don't free the root - we will detect this
1685 * condition after our loop and return a
1686 * positive value for caller to stop walking the tree.
1687 */
1688 if (level != root_level) {
1689 btrfs_tree_unlock_rw(eb, path->locks[level]);
1690 path->locks[level] = 0;
1691
1692 free_extent_buffer(eb);
1693 path->nodes[level] = NULL;
1694 path->slots[level] = 0;
1695 }
1696 } else {
1697 /*
1698 * We have a valid slot to walk back down
1699 * from. Stop here so caller can process these
1700 * new nodes.
1701 */
1702 break;
1703 }
1704
1705 level++;
1706 }
1707
1708 eb = path->nodes[root_level];
1709 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1710 return 1;
1711
1712 return 0;
1713 }
1714
1715 /*
1716 * Helper function to trace a subtree tree block swap.
1717 *
1718 * The swap will happen in highest tree block, but there may be a lot of
1719 * tree blocks involved.
1720 *
1721 * For example:
1722 * OO = Old tree blocks
1723 * NN = New tree blocks allocated during balance
1724 *
1725 * File tree (257) Reloc tree for 257
1726 * L2 OO NN
1727 * / \ / \
1728 * L1 OO OO (a) OO NN (a)
1729 * / \ / \ / \ / \
1730 * L0 OO OO OO OO OO OO NN NN
1731 * (b) (c) (b) (c)
1732 *
1733 * When calling qgroup_trace_extent_swap(), we will pass:
1734 * @src_eb = OO(a)
1735 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1736 * @dst_level = 0
1737 * @root_level = 1
1738 *
1739 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1740 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1741 *
1742 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1743 *
1744 * 1) Tree search from @src_eb
1745 * It should acts as a simplified btrfs_search_slot().
1746 * The key for search can be extracted from @dst_path->nodes[dst_level]
1747 * (first key).
1748 *
1749 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1750 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1751 * They should be marked during preivous (@dst_level = 1) iteration.
1752 *
1753 * 3) Mark file extents in leaves dirty
1754 * We don't have good way to pick out new file extents only.
1755 * So we still follow the old method by scanning all file extents in
1756 * the leave.
1757 *
1758 * This function can free us from keeping two pathes, thus later we only need
1759 * to care about how to iterate all new tree blocks in reloc tree.
1760 */
1761 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1762 struct extent_buffer *src_eb,
1763 struct btrfs_path *dst_path,
1764 int dst_level, int root_level,
1765 bool trace_leaf)
1766 {
1767 struct btrfs_key key;
1768 struct btrfs_path *src_path;
1769 struct btrfs_fs_info *fs_info = trans->fs_info;
1770 u32 nodesize = fs_info->nodesize;
1771 int cur_level = root_level;
1772 int ret;
1773
1774 BUG_ON(dst_level > root_level);
1775 /* Level mismatch */
1776 if (btrfs_header_level(src_eb) != root_level)
1777 return -EINVAL;
1778
1779 src_path = btrfs_alloc_path();
1780 if (!src_path) {
1781 ret = -ENOMEM;
1782 goto out;
1783 }
1784
1785 if (dst_level)
1786 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1787 else
1788 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1789
1790 /* For src_path */
1791 extent_buffer_get(src_eb);
1792 src_path->nodes[root_level] = src_eb;
1793 src_path->slots[root_level] = dst_path->slots[root_level];
1794 src_path->locks[root_level] = 0;
1795
1796 /* A simplified version of btrfs_search_slot() */
1797 while (cur_level >= dst_level) {
1798 struct btrfs_key src_key;
1799 struct btrfs_key dst_key;
1800
1801 if (src_path->nodes[cur_level] == NULL) {
1802 struct btrfs_key first_key;
1803 struct extent_buffer *eb;
1804 int parent_slot;
1805 u64 child_gen;
1806 u64 child_bytenr;
1807
1808 eb = src_path->nodes[cur_level + 1];
1809 parent_slot = src_path->slots[cur_level + 1];
1810 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1811 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1812 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1813
1814 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1815 cur_level, &first_key);
1816 if (IS_ERR(eb)) {
1817 ret = PTR_ERR(eb);
1818 goto out;
1819 } else if (!extent_buffer_uptodate(eb)) {
1820 free_extent_buffer(eb);
1821 ret = -EIO;
1822 goto out;
1823 }
1824
1825 src_path->nodes[cur_level] = eb;
1826
1827 btrfs_tree_read_lock(eb);
1828 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1829 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1830 }
1831
1832 src_path->slots[cur_level] = dst_path->slots[cur_level];
1833 if (cur_level) {
1834 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1835 &dst_key, dst_path->slots[cur_level]);
1836 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1837 &src_key, src_path->slots[cur_level]);
1838 } else {
1839 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1840 &dst_key, dst_path->slots[cur_level]);
1841 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1842 &src_key, src_path->slots[cur_level]);
1843 }
1844 /* Content mismatch, something went wrong */
1845 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1846 ret = -ENOENT;
1847 goto out;
1848 }
1849 cur_level--;
1850 }
1851
1852 /*
1853 * Now both @dst_path and @src_path have been populated, record the tree
1854 * blocks for qgroup accounting.
1855 */
1856 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1857 nodesize, GFP_NOFS);
1858 if (ret < 0)
1859 goto out;
1860 ret = btrfs_qgroup_trace_extent(trans,
1861 dst_path->nodes[dst_level]->start,
1862 nodesize, GFP_NOFS);
1863 if (ret < 0)
1864 goto out;
1865
1866 /* Record leaf file extents */
1867 if (dst_level == 0 && trace_leaf) {
1868 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1869 if (ret < 0)
1870 goto out;
1871 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1872 }
1873 out:
1874 btrfs_free_path(src_path);
1875 return ret;
1876 }
1877
1878 /*
1879 * Helper function to do recursive generation-aware depth-first search, to
1880 * locate all new tree blocks in a subtree of reloc tree.
1881 *
1882 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1883 * reloc tree
1884 * L2 NN (a)
1885 * / \
1886 * L1 OO NN (b)
1887 * / \ / \
1888 * L0 OO OO OO NN
1889 * (c) (d)
1890 * If we pass:
1891 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1892 * @cur_level = 1
1893 * @root_level = 1
1894 *
1895 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1896 * above tree blocks along with their counter parts in file tree.
1897 * While during search, old tree blocsk OO(c) will be skiped as tree block swap
1898 * won't affect OO(c).
1899 */
1900 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1901 struct extent_buffer *src_eb,
1902 struct btrfs_path *dst_path,
1903 int cur_level, int root_level,
1904 u64 last_snapshot, bool trace_leaf)
1905 {
1906 struct btrfs_fs_info *fs_info = trans->fs_info;
1907 struct extent_buffer *eb;
1908 bool need_cleanup = false;
1909 int ret = 0;
1910 int i;
1911
1912 /* Level sanity check */
1913 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL ||
1914 root_level < 0 || root_level >= BTRFS_MAX_LEVEL ||
1915 root_level < cur_level) {
1916 btrfs_err_rl(fs_info,
1917 "%s: bad levels, cur_level=%d root_level=%d",
1918 __func__, cur_level, root_level);
1919 return -EUCLEAN;
1920 }
1921
1922 /* Read the tree block if needed */
1923 if (dst_path->nodes[cur_level] == NULL) {
1924 struct btrfs_key first_key;
1925 int parent_slot;
1926 u64 child_gen;
1927 u64 child_bytenr;
1928
1929 /*
1930 * dst_path->nodes[root_level] must be initialized before
1931 * calling this function.
1932 */
1933 if (cur_level == root_level) {
1934 btrfs_err_rl(fs_info,
1935 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1936 __func__, root_level, root_level, cur_level);
1937 return -EUCLEAN;
1938 }
1939
1940 /*
1941 * We need to get child blockptr/gen from parent before we can
1942 * read it.
1943 */
1944 eb = dst_path->nodes[cur_level + 1];
1945 parent_slot = dst_path->slots[cur_level + 1];
1946 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1947 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1948 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1949
1950 /* This node is old, no need to trace */
1951 if (child_gen < last_snapshot)
1952 goto out;
1953
1954 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1955 cur_level, &first_key);
1956 if (IS_ERR(eb)) {
1957 ret = PTR_ERR(eb);
1958 goto out;
1959 } else if (!extent_buffer_uptodate(eb)) {
1960 free_extent_buffer(eb);
1961 ret = -EIO;
1962 goto out;
1963 }
1964
1965 dst_path->nodes[cur_level] = eb;
1966 dst_path->slots[cur_level] = 0;
1967
1968 btrfs_tree_read_lock(eb);
1969 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1970 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1971 need_cleanup = true;
1972 }
1973
1974 /* Now record this tree block and its counter part for qgroups */
1975 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1976 root_level, trace_leaf);
1977 if (ret < 0)
1978 goto cleanup;
1979
1980 eb = dst_path->nodes[cur_level];
1981
1982 if (cur_level > 0) {
1983 /* Iterate all child tree blocks */
1984 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1985 /* Skip old tree blocks as they won't be swapped */
1986 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1987 continue;
1988 dst_path->slots[cur_level] = i;
1989
1990 /* Recursive call (at most 7 times) */
1991 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
1992 dst_path, cur_level - 1, root_level,
1993 last_snapshot, trace_leaf);
1994 if (ret < 0)
1995 goto cleanup;
1996 }
1997 }
1998
1999 cleanup:
2000 if (need_cleanup) {
2001 /* Clean up */
2002 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2003 dst_path->locks[cur_level]);
2004 free_extent_buffer(dst_path->nodes[cur_level]);
2005 dst_path->nodes[cur_level] = NULL;
2006 dst_path->slots[cur_level] = 0;
2007 dst_path->locks[cur_level] = 0;
2008 }
2009 out:
2010 return ret;
2011 }
2012
2013 /*
2014 * Inform qgroup to trace subtree swap used in balance.
2015 *
2016 * Unlike btrfs_qgroup_trace_subtree(), this function will only trace
2017 * new tree blocks whose generation is equal to (or larger than) @last_snapshot.
2018 *
2019 * Will go down the tree block pointed by @dst_eb (pointed by @dst_parent and
2020 * @dst_slot), and find any tree blocks whose generation is at @last_snapshot,
2021 * and then go down @src_eb (pointed by @src_parent and @src_slot) to find
2022 * the conterpart of the tree block, then mark both tree blocks as qgroup dirty,
2023 * and skip all tree blocks whose generation is smaller than last_snapshot.
2024 *
2025 * This would skip tons of tree blocks of original btrfs_qgroup_trace_subtree(),
2026 * which could be the cause of very slow balance if the file tree is large.
2027 *
2028 * @src_parent, @src_slot: pointer to src (file tree) eb.
2029 * @dst_parent, @dst_slot: pointer to dst (reloc tree) eb.
2030 */
2031 int btrfs_qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2032 struct btrfs_block_group_cache *bg_cache,
2033 struct extent_buffer *src_parent, int src_slot,
2034 struct extent_buffer *dst_parent, int dst_slot,
2035 u64 last_snapshot)
2036 {
2037 struct btrfs_fs_info *fs_info = trans->fs_info;
2038 struct btrfs_path *dst_path = NULL;
2039 struct btrfs_key first_key;
2040 struct extent_buffer *src_eb = NULL;
2041 struct extent_buffer *dst_eb = NULL;
2042 bool trace_leaf = false;
2043 u64 child_gen;
2044 u64 child_bytenr;
2045 int level;
2046 int ret;
2047
2048 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2049 return 0;
2050
2051 /* Check parameter order */
2052 if (btrfs_node_ptr_generation(src_parent, src_slot) >
2053 btrfs_node_ptr_generation(dst_parent, dst_slot)) {
2054 btrfs_err_rl(fs_info,
2055 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2056 btrfs_node_ptr_generation(src_parent, src_slot),
2057 btrfs_node_ptr_generation(dst_parent, dst_slot));
2058 return -EUCLEAN;
2059 }
2060
2061 /*
2062 * Only trace leaf if we're relocating data block groups, this could
2063 * reduce tons of data extents tracing for meta/sys bg relocation.
2064 */
2065 if (bg_cache->flags & BTRFS_BLOCK_GROUP_DATA)
2066 trace_leaf = true;
2067 /* Read out real @src_eb, pointed by @src_parent and @src_slot */
2068 child_bytenr = btrfs_node_blockptr(src_parent, src_slot);
2069 child_gen = btrfs_node_ptr_generation(src_parent, src_slot);
2070 btrfs_node_key_to_cpu(src_parent, &first_key, src_slot);
2071
2072 src_eb = read_tree_block(fs_info, child_bytenr, child_gen,
2073 btrfs_header_level(src_parent) - 1, &first_key);
2074 if (IS_ERR(src_eb)) {
2075 ret = PTR_ERR(src_eb);
2076 goto out;
2077 }
2078
2079 /* Read out real @dst_eb, pointed by @src_parent and @src_slot */
2080 child_bytenr = btrfs_node_blockptr(dst_parent, dst_slot);
2081 child_gen = btrfs_node_ptr_generation(dst_parent, dst_slot);
2082 btrfs_node_key_to_cpu(dst_parent, &first_key, dst_slot);
2083
2084 dst_eb = read_tree_block(fs_info, child_bytenr, child_gen,
2085 btrfs_header_level(dst_parent) - 1, &first_key);
2086 if (IS_ERR(dst_eb)) {
2087 ret = PTR_ERR(dst_eb);
2088 goto out;
2089 }
2090
2091 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2092 ret = -EINVAL;
2093 goto out;
2094 }
2095
2096 level = btrfs_header_level(dst_eb);
2097 dst_path = btrfs_alloc_path();
2098 if (!dst_path) {
2099 ret = -ENOMEM;
2100 goto out;
2101 }
2102
2103 /* For dst_path */
2104 extent_buffer_get(dst_eb);
2105 dst_path->nodes[level] = dst_eb;
2106 dst_path->slots[level] = 0;
2107 dst_path->locks[level] = 0;
2108
2109 /* Do the generation-aware breadth-first search */
2110 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2111 level, last_snapshot, trace_leaf);
2112 if (ret < 0)
2113 goto out;
2114 ret = 0;
2115
2116 out:
2117 free_extent_buffer(src_eb);
2118 free_extent_buffer(dst_eb);
2119 btrfs_free_path(dst_path);
2120 if (ret < 0)
2121 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2122 return ret;
2123 }
2124
2125 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2126 struct extent_buffer *root_eb,
2127 u64 root_gen, int root_level)
2128 {
2129 struct btrfs_fs_info *fs_info = trans->fs_info;
2130 int ret = 0;
2131 int level;
2132 struct extent_buffer *eb = root_eb;
2133 struct btrfs_path *path = NULL;
2134
2135 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2136 BUG_ON(root_eb == NULL);
2137
2138 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2139 return 0;
2140
2141 if (!extent_buffer_uptodate(root_eb)) {
2142 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2143 if (ret)
2144 goto out;
2145 }
2146
2147 if (root_level == 0) {
2148 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2149 goto out;
2150 }
2151
2152 path = btrfs_alloc_path();
2153 if (!path)
2154 return -ENOMEM;
2155
2156 /*
2157 * Walk down the tree. Missing extent blocks are filled in as
2158 * we go. Metadata is accounted every time we read a new
2159 * extent block.
2160 *
2161 * When we reach a leaf, we account for file extent items in it,
2162 * walk back up the tree (adjusting slot pointers as we go)
2163 * and restart the search process.
2164 */
2165 extent_buffer_get(root_eb); /* For path */
2166 path->nodes[root_level] = root_eb;
2167 path->slots[root_level] = 0;
2168 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2169 walk_down:
2170 level = root_level;
2171 while (level >= 0) {
2172 if (path->nodes[level] == NULL) {
2173 struct btrfs_key first_key;
2174 int parent_slot;
2175 u64 child_gen;
2176 u64 child_bytenr;
2177
2178 /*
2179 * We need to get child blockptr/gen from parent before
2180 * we can read it.
2181 */
2182 eb = path->nodes[level + 1];
2183 parent_slot = path->slots[level + 1];
2184 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2185 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2186 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2187
2188 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2189 level, &first_key);
2190 if (IS_ERR(eb)) {
2191 ret = PTR_ERR(eb);
2192 goto out;
2193 } else if (!extent_buffer_uptodate(eb)) {
2194 free_extent_buffer(eb);
2195 ret = -EIO;
2196 goto out;
2197 }
2198
2199 path->nodes[level] = eb;
2200 path->slots[level] = 0;
2201
2202 btrfs_tree_read_lock(eb);
2203 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
2204 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2205
2206 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2207 fs_info->nodesize,
2208 GFP_NOFS);
2209 if (ret)
2210 goto out;
2211 }
2212
2213 if (level == 0) {
2214 ret = btrfs_qgroup_trace_leaf_items(trans,
2215 path->nodes[level]);
2216 if (ret)
2217 goto out;
2218
2219 /* Nonzero return here means we completed our search */
2220 ret = adjust_slots_upwards(path, root_level);
2221 if (ret)
2222 break;
2223
2224 /* Restart search with new slots */
2225 goto walk_down;
2226 }
2227
2228 level--;
2229 }
2230
2231 ret = 0;
2232 out:
2233 btrfs_free_path(path);
2234
2235 return ret;
2236 }
2237
2238 #define UPDATE_NEW 0
2239 #define UPDATE_OLD 1
2240 /*
2241 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2242 */
2243 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2244 struct ulist *roots, struct ulist *tmp,
2245 struct ulist *qgroups, u64 seq, int update_old)
2246 {
2247 struct ulist_node *unode;
2248 struct ulist_iterator uiter;
2249 struct ulist_node *tmp_unode;
2250 struct ulist_iterator tmp_uiter;
2251 struct btrfs_qgroup *qg;
2252 int ret = 0;
2253
2254 if (!roots)
2255 return 0;
2256 ULIST_ITER_INIT(&uiter);
2257 while ((unode = ulist_next(roots, &uiter))) {
2258 qg = find_qgroup_rb(fs_info, unode->val);
2259 if (!qg)
2260 continue;
2261
2262 ulist_reinit(tmp);
2263 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2264 GFP_ATOMIC);
2265 if (ret < 0)
2266 return ret;
2267 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2268 if (ret < 0)
2269 return ret;
2270 ULIST_ITER_INIT(&tmp_uiter);
2271 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2272 struct btrfs_qgroup_list *glist;
2273
2274 qg = unode_aux_to_qgroup(tmp_unode);
2275 if (update_old)
2276 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2277 else
2278 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2279 list_for_each_entry(glist, &qg->groups, next_group) {
2280 ret = ulist_add(qgroups, glist->group->qgroupid,
2281 qgroup_to_aux(glist->group),
2282 GFP_ATOMIC);
2283 if (ret < 0)
2284 return ret;
2285 ret = ulist_add(tmp, glist->group->qgroupid,
2286 qgroup_to_aux(glist->group),
2287 GFP_ATOMIC);
2288 if (ret < 0)
2289 return ret;
2290 }
2291 }
2292 }
2293 return 0;
2294 }
2295
2296 /*
2297 * Update qgroup rfer/excl counters.
2298 * Rfer update is easy, codes can explain themselves.
2299 *
2300 * Excl update is tricky, the update is split into 2 part.
2301 * Part 1: Possible exclusive <-> sharing detect:
2302 * | A | !A |
2303 * -------------------------------------
2304 * B | * | - |
2305 * -------------------------------------
2306 * !B | + | ** |
2307 * -------------------------------------
2308 *
2309 * Conditions:
2310 * A: cur_old_roots < nr_old_roots (not exclusive before)
2311 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2312 * B: cur_new_roots < nr_new_roots (not exclusive now)
2313 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2314 *
2315 * Results:
2316 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2317 * *: Definitely not changed. **: Possible unchanged.
2318 *
2319 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2320 *
2321 * To make the logic clear, we first use condition A and B to split
2322 * combination into 4 results.
2323 *
2324 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2325 * only on variant maybe 0.
2326 *
2327 * Lastly, check result **, since there are 2 variants maybe 0, split them
2328 * again(2x2).
2329 * But this time we don't need to consider other things, the codes and logic
2330 * is easy to understand now.
2331 */
2332 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2333 struct ulist *qgroups,
2334 u64 nr_old_roots,
2335 u64 nr_new_roots,
2336 u64 num_bytes, u64 seq)
2337 {
2338 struct ulist_node *unode;
2339 struct ulist_iterator uiter;
2340 struct btrfs_qgroup *qg;
2341 u64 cur_new_count, cur_old_count;
2342
2343 ULIST_ITER_INIT(&uiter);
2344 while ((unode = ulist_next(qgroups, &uiter))) {
2345 bool dirty = false;
2346
2347 qg = unode_aux_to_qgroup(unode);
2348 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2349 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2350
2351 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2352 cur_new_count);
2353
2354 /* Rfer update part */
2355 if (cur_old_count == 0 && cur_new_count > 0) {
2356 qg->rfer += num_bytes;
2357 qg->rfer_cmpr += num_bytes;
2358 dirty = true;
2359 }
2360 if (cur_old_count > 0 && cur_new_count == 0) {
2361 qg->rfer -= num_bytes;
2362 qg->rfer_cmpr -= num_bytes;
2363 dirty = true;
2364 }
2365
2366 /* Excl update part */
2367 /* Exclusive/none -> shared case */
2368 if (cur_old_count == nr_old_roots &&
2369 cur_new_count < nr_new_roots) {
2370 /* Exclusive -> shared */
2371 if (cur_old_count != 0) {
2372 qg->excl -= num_bytes;
2373 qg->excl_cmpr -= num_bytes;
2374 dirty = true;
2375 }
2376 }
2377
2378 /* Shared -> exclusive/none case */
2379 if (cur_old_count < nr_old_roots &&
2380 cur_new_count == nr_new_roots) {
2381 /* Shared->exclusive */
2382 if (cur_new_count != 0) {
2383 qg->excl += num_bytes;
2384 qg->excl_cmpr += num_bytes;
2385 dirty = true;
2386 }
2387 }
2388
2389 /* Exclusive/none -> exclusive/none case */
2390 if (cur_old_count == nr_old_roots &&
2391 cur_new_count == nr_new_roots) {
2392 if (cur_old_count == 0) {
2393 /* None -> exclusive/none */
2394
2395 if (cur_new_count != 0) {
2396 /* None -> exclusive */
2397 qg->excl += num_bytes;
2398 qg->excl_cmpr += num_bytes;
2399 dirty = true;
2400 }
2401 /* None -> none, nothing changed */
2402 } else {
2403 /* Exclusive -> exclusive/none */
2404
2405 if (cur_new_count == 0) {
2406 /* Exclusive -> none */
2407 qg->excl -= num_bytes;
2408 qg->excl_cmpr -= num_bytes;
2409 dirty = true;
2410 }
2411 /* Exclusive -> exclusive, nothing changed */
2412 }
2413 }
2414
2415 if (dirty)
2416 qgroup_dirty(fs_info, qg);
2417 }
2418 return 0;
2419 }
2420
2421 /*
2422 * Check if the @roots potentially is a list of fs tree roots
2423 *
2424 * Return 0 for definitely not a fs/subvol tree roots ulist
2425 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2426 * one as well)
2427 */
2428 static int maybe_fs_roots(struct ulist *roots)
2429 {
2430 struct ulist_node *unode;
2431 struct ulist_iterator uiter;
2432
2433 /* Empty one, still possible for fs roots */
2434 if (!roots || roots->nnodes == 0)
2435 return 1;
2436
2437 ULIST_ITER_INIT(&uiter);
2438 unode = ulist_next(roots, &uiter);
2439 if (!unode)
2440 return 1;
2441
2442 /*
2443 * If it contains fs tree roots, then it must belong to fs/subvol
2444 * trees.
2445 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2446 */
2447 return is_fstree(unode->val);
2448 }
2449
2450 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2451 u64 num_bytes, struct ulist *old_roots,
2452 struct ulist *new_roots)
2453 {
2454 struct btrfs_fs_info *fs_info = trans->fs_info;
2455 struct ulist *qgroups = NULL;
2456 struct ulist *tmp = NULL;
2457 u64 seq;
2458 u64 nr_new_roots = 0;
2459 u64 nr_old_roots = 0;
2460 int ret = 0;
2461
2462 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2463 return 0;
2464
2465 if (new_roots) {
2466 if (!maybe_fs_roots(new_roots))
2467 goto out_free;
2468 nr_new_roots = new_roots->nnodes;
2469 }
2470 if (old_roots) {
2471 if (!maybe_fs_roots(old_roots))
2472 goto out_free;
2473 nr_old_roots = old_roots->nnodes;
2474 }
2475
2476 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2477 if (nr_old_roots == 0 && nr_new_roots == 0)
2478 goto out_free;
2479
2480 BUG_ON(!fs_info->quota_root);
2481
2482 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2483 num_bytes, nr_old_roots, nr_new_roots);
2484
2485 qgroups = ulist_alloc(GFP_NOFS);
2486 if (!qgroups) {
2487 ret = -ENOMEM;
2488 goto out_free;
2489 }
2490 tmp = ulist_alloc(GFP_NOFS);
2491 if (!tmp) {
2492 ret = -ENOMEM;
2493 goto out_free;
2494 }
2495
2496 mutex_lock(&fs_info->qgroup_rescan_lock);
2497 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2498 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2499 mutex_unlock(&fs_info->qgroup_rescan_lock);
2500 ret = 0;
2501 goto out_free;
2502 }
2503 }
2504 mutex_unlock(&fs_info->qgroup_rescan_lock);
2505
2506 spin_lock(&fs_info->qgroup_lock);
2507 seq = fs_info->qgroup_seq;
2508
2509 /* Update old refcnts using old_roots */
2510 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2511 UPDATE_OLD);
2512 if (ret < 0)
2513 goto out;
2514
2515 /* Update new refcnts using new_roots */
2516 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2517 UPDATE_NEW);
2518 if (ret < 0)
2519 goto out;
2520
2521 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2522 num_bytes, seq);
2523
2524 /*
2525 * Bump qgroup_seq to avoid seq overlap
2526 */
2527 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2528 out:
2529 spin_unlock(&fs_info->qgroup_lock);
2530 out_free:
2531 ulist_free(tmp);
2532 ulist_free(qgroups);
2533 ulist_free(old_roots);
2534 ulist_free(new_roots);
2535 return ret;
2536 }
2537
2538 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2539 {
2540 struct btrfs_fs_info *fs_info = trans->fs_info;
2541 struct btrfs_qgroup_extent_record *record;
2542 struct btrfs_delayed_ref_root *delayed_refs;
2543 struct ulist *new_roots = NULL;
2544 struct rb_node *node;
2545 u64 num_dirty_extents = 0;
2546 u64 qgroup_to_skip;
2547 int ret = 0;
2548
2549 delayed_refs = &trans->transaction->delayed_refs;
2550 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2551 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2552 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2553 node);
2554
2555 num_dirty_extents++;
2556 trace_btrfs_qgroup_account_extents(fs_info, record);
2557
2558 if (!ret) {
2559 /*
2560 * Old roots should be searched when inserting qgroup
2561 * extent record
2562 */
2563 if (WARN_ON(!record->old_roots)) {
2564 /* Search commit root to find old_roots */
2565 ret = btrfs_find_all_roots(NULL, fs_info,
2566 record->bytenr, 0,
2567 &record->old_roots, false);
2568 if (ret < 0)
2569 goto cleanup;
2570 }
2571
2572 /*
2573 * Use SEQ_LAST as time_seq to do special search, which
2574 * doesn't lock tree or delayed_refs and search current
2575 * root. It's safe inside commit_transaction().
2576 */
2577 ret = btrfs_find_all_roots(trans, fs_info,
2578 record->bytenr, SEQ_LAST, &new_roots, false);
2579 if (ret < 0)
2580 goto cleanup;
2581 if (qgroup_to_skip) {
2582 ulist_del(new_roots, qgroup_to_skip, 0);
2583 ulist_del(record->old_roots, qgroup_to_skip,
2584 0);
2585 }
2586 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2587 record->num_bytes,
2588 record->old_roots,
2589 new_roots);
2590 record->old_roots = NULL;
2591 new_roots = NULL;
2592 }
2593 cleanup:
2594 ulist_free(record->old_roots);
2595 ulist_free(new_roots);
2596 new_roots = NULL;
2597 rb_erase(node, &delayed_refs->dirty_extent_root);
2598 kfree(record);
2599
2600 }
2601 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2602 num_dirty_extents);
2603 return ret;
2604 }
2605
2606 /*
2607 * called from commit_transaction. Writes all changed qgroups to disk.
2608 */
2609 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2610 {
2611 struct btrfs_fs_info *fs_info = trans->fs_info;
2612 struct btrfs_root *quota_root = fs_info->quota_root;
2613 int ret = 0;
2614
2615 if (!quota_root)
2616 return ret;
2617
2618 spin_lock(&fs_info->qgroup_lock);
2619 while (!list_empty(&fs_info->dirty_qgroups)) {
2620 struct btrfs_qgroup *qgroup;
2621 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2622 struct btrfs_qgroup, dirty);
2623 list_del_init(&qgroup->dirty);
2624 spin_unlock(&fs_info->qgroup_lock);
2625 ret = update_qgroup_info_item(trans, qgroup);
2626 if (ret)
2627 fs_info->qgroup_flags |=
2628 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2629 ret = update_qgroup_limit_item(trans, qgroup);
2630 if (ret)
2631 fs_info->qgroup_flags |=
2632 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2633 spin_lock(&fs_info->qgroup_lock);
2634 }
2635 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2636 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2637 else
2638 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2639 spin_unlock(&fs_info->qgroup_lock);
2640
2641 ret = update_qgroup_status_item(trans);
2642 if (ret)
2643 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2644
2645 return ret;
2646 }
2647
2648 /*
2649 * Copy the accounting information between qgroups. This is necessary
2650 * when a snapshot or a subvolume is created. Throwing an error will
2651 * cause a transaction abort so we take extra care here to only error
2652 * when a readonly fs is a reasonable outcome.
2653 */
2654 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2655 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2656 {
2657 int ret = 0;
2658 int i;
2659 u64 *i_qgroups;
2660 struct btrfs_fs_info *fs_info = trans->fs_info;
2661 struct btrfs_root *quota_root = fs_info->quota_root;
2662 struct btrfs_qgroup *srcgroup;
2663 struct btrfs_qgroup *dstgroup;
2664 u32 level_size = 0;
2665 u64 nums;
2666
2667 mutex_lock(&fs_info->qgroup_ioctl_lock);
2668 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2669 goto out;
2670
2671 if (!quota_root) {
2672 ret = -EINVAL;
2673 goto out;
2674 }
2675
2676 if (inherit) {
2677 i_qgroups = (u64 *)(inherit + 1);
2678 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2679 2 * inherit->num_excl_copies;
2680 for (i = 0; i < nums; ++i) {
2681 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2682
2683 /*
2684 * Zero out invalid groups so we can ignore
2685 * them later.
2686 */
2687 if (!srcgroup ||
2688 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2689 *i_qgroups = 0ULL;
2690
2691 ++i_qgroups;
2692 }
2693 }
2694
2695 /*
2696 * create a tracking group for the subvol itself
2697 */
2698 ret = add_qgroup_item(trans, quota_root, objectid);
2699 if (ret)
2700 goto out;
2701
2702 /*
2703 * add qgroup to all inherited groups
2704 */
2705 if (inherit) {
2706 i_qgroups = (u64 *)(inherit + 1);
2707 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2708 if (*i_qgroups == 0)
2709 continue;
2710 ret = add_qgroup_relation_item(trans, objectid,
2711 *i_qgroups);
2712 if (ret && ret != -EEXIST)
2713 goto out;
2714 ret = add_qgroup_relation_item(trans, *i_qgroups,
2715 objectid);
2716 if (ret && ret != -EEXIST)
2717 goto out;
2718 }
2719 ret = 0;
2720 }
2721
2722
2723 spin_lock(&fs_info->qgroup_lock);
2724
2725 dstgroup = add_qgroup_rb(fs_info, objectid);
2726 if (IS_ERR(dstgroup)) {
2727 ret = PTR_ERR(dstgroup);
2728 goto unlock;
2729 }
2730
2731 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2732 dstgroup->lim_flags = inherit->lim.flags;
2733 dstgroup->max_rfer = inherit->lim.max_rfer;
2734 dstgroup->max_excl = inherit->lim.max_excl;
2735 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2736 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2737
2738 ret = update_qgroup_limit_item(trans, dstgroup);
2739 if (ret) {
2740 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2741 btrfs_info(fs_info,
2742 "unable to update quota limit for %llu",
2743 dstgroup->qgroupid);
2744 goto unlock;
2745 }
2746 }
2747
2748 if (srcid) {
2749 srcgroup = find_qgroup_rb(fs_info, srcid);
2750 if (!srcgroup)
2751 goto unlock;
2752
2753 /*
2754 * We call inherit after we clone the root in order to make sure
2755 * our counts don't go crazy, so at this point the only
2756 * difference between the two roots should be the root node.
2757 */
2758 level_size = fs_info->nodesize;
2759 dstgroup->rfer = srcgroup->rfer;
2760 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2761 dstgroup->excl = level_size;
2762 dstgroup->excl_cmpr = level_size;
2763 srcgroup->excl = level_size;
2764 srcgroup->excl_cmpr = level_size;
2765
2766 /* inherit the limit info */
2767 dstgroup->lim_flags = srcgroup->lim_flags;
2768 dstgroup->max_rfer = srcgroup->max_rfer;
2769 dstgroup->max_excl = srcgroup->max_excl;
2770 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2771 dstgroup->rsv_excl = srcgroup->rsv_excl;
2772
2773 qgroup_dirty(fs_info, dstgroup);
2774 qgroup_dirty(fs_info, srcgroup);
2775 }
2776
2777 if (!inherit)
2778 goto unlock;
2779
2780 i_qgroups = (u64 *)(inherit + 1);
2781 for (i = 0; i < inherit->num_qgroups; ++i) {
2782 if (*i_qgroups) {
2783 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2784 if (ret)
2785 goto unlock;
2786 }
2787 ++i_qgroups;
2788 }
2789
2790 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2791 struct btrfs_qgroup *src;
2792 struct btrfs_qgroup *dst;
2793
2794 if (!i_qgroups[0] || !i_qgroups[1])
2795 continue;
2796
2797 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2798 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2799
2800 if (!src || !dst) {
2801 ret = -EINVAL;
2802 goto unlock;
2803 }
2804
2805 dst->rfer = src->rfer - level_size;
2806 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2807 }
2808 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2809 struct btrfs_qgroup *src;
2810 struct btrfs_qgroup *dst;
2811
2812 if (!i_qgroups[0] || !i_qgroups[1])
2813 continue;
2814
2815 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2816 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2817
2818 if (!src || !dst) {
2819 ret = -EINVAL;
2820 goto unlock;
2821 }
2822
2823 dst->excl = src->excl + level_size;
2824 dst->excl_cmpr = src->excl_cmpr + level_size;
2825 }
2826
2827 unlock:
2828 spin_unlock(&fs_info->qgroup_lock);
2829 out:
2830 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2831 return ret;
2832 }
2833
2834 /*
2835 * Two limits to commit transaction in advance.
2836 *
2837 * For RATIO, it will be 1/RATIO of the remaining limit
2838 * (excluding data and prealloc meta) as threshold.
2839 * For SIZE, it will be in byte unit as threshold.
2840 */
2841 #define QGROUP_PERTRANS_RATIO 32
2842 #define QGROUP_PERTRANS_SIZE SZ_32M
2843 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2844 const struct btrfs_qgroup *qg, u64 num_bytes)
2845 {
2846 u64 limit;
2847 u64 threshold;
2848
2849 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2850 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2851 return false;
2852
2853 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2854 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2855 return false;
2856
2857 /*
2858 * Even if we passed the check, it's better to check if reservation
2859 * for meta_pertrans is pushing us near limit.
2860 * If there is too much pertrans reservation or it's near the limit,
2861 * let's try commit transaction to free some, using transaction_kthread
2862 */
2863 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2864 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2865 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2866 limit = qg->max_excl;
2867 else
2868 limit = qg->max_rfer;
2869 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2870 qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2871 QGROUP_PERTRANS_RATIO;
2872 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2873
2874 /*
2875 * Use transaction_kthread to commit transaction, so we no
2876 * longer need to bother nested transaction nor lock context.
2877 */
2878 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2879 btrfs_commit_transaction_locksafe(fs_info);
2880 }
2881
2882 return true;
2883 }
2884
2885 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2886 enum btrfs_qgroup_rsv_type type)
2887 {
2888 struct btrfs_root *quota_root;
2889 struct btrfs_qgroup *qgroup;
2890 struct btrfs_fs_info *fs_info = root->fs_info;
2891 u64 ref_root = root->root_key.objectid;
2892 int ret = 0;
2893 struct ulist_node *unode;
2894 struct ulist_iterator uiter;
2895
2896 if (!is_fstree(ref_root))
2897 return 0;
2898
2899 if (num_bytes == 0)
2900 return 0;
2901
2902 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2903 capable(CAP_SYS_RESOURCE))
2904 enforce = false;
2905
2906 spin_lock(&fs_info->qgroup_lock);
2907 quota_root = fs_info->quota_root;
2908 if (!quota_root)
2909 goto out;
2910
2911 qgroup = find_qgroup_rb(fs_info, ref_root);
2912 if (!qgroup)
2913 goto out;
2914
2915 /*
2916 * in a first step, we check all affected qgroups if any limits would
2917 * be exceeded
2918 */
2919 ulist_reinit(fs_info->qgroup_ulist);
2920 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2921 qgroup_to_aux(qgroup), GFP_ATOMIC);
2922 if (ret < 0)
2923 goto out;
2924 ULIST_ITER_INIT(&uiter);
2925 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2926 struct btrfs_qgroup *qg;
2927 struct btrfs_qgroup_list *glist;
2928
2929 qg = unode_aux_to_qgroup(unode);
2930
2931 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2932 ret = -EDQUOT;
2933 goto out;
2934 }
2935
2936 list_for_each_entry(glist, &qg->groups, next_group) {
2937 ret = ulist_add(fs_info->qgroup_ulist,
2938 glist->group->qgroupid,
2939 qgroup_to_aux(glist->group), GFP_ATOMIC);
2940 if (ret < 0)
2941 goto out;
2942 }
2943 }
2944 ret = 0;
2945 /*
2946 * no limits exceeded, now record the reservation into all qgroups
2947 */
2948 ULIST_ITER_INIT(&uiter);
2949 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2950 struct btrfs_qgroup *qg;
2951
2952 qg = unode_aux_to_qgroup(unode);
2953
2954 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2955 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2956 }
2957
2958 out:
2959 spin_unlock(&fs_info->qgroup_lock);
2960 return ret;
2961 }
2962
2963 /*
2964 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2965 * qgroup).
2966 *
2967 * Will handle all higher level qgroup too.
2968 *
2969 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2970 * This special case is only used for META_PERTRANS type.
2971 */
2972 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2973 u64 ref_root, u64 num_bytes,
2974 enum btrfs_qgroup_rsv_type type)
2975 {
2976 struct btrfs_root *quota_root;
2977 struct btrfs_qgroup *qgroup;
2978 struct ulist_node *unode;
2979 struct ulist_iterator uiter;
2980 int ret = 0;
2981
2982 if (!is_fstree(ref_root))
2983 return;
2984
2985 if (num_bytes == 0)
2986 return;
2987
2988 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2989 WARN(1, "%s: Invalid type to free", __func__);
2990 return;
2991 }
2992 spin_lock(&fs_info->qgroup_lock);
2993
2994 quota_root = fs_info->quota_root;
2995 if (!quota_root)
2996 goto out;
2997
2998 qgroup = find_qgroup_rb(fs_info, ref_root);
2999 if (!qgroup)
3000 goto out;
3001
3002 if (num_bytes == (u64)-1)
3003 /*
3004 * We're freeing all pertrans rsv, get reserved value from
3005 * level 0 qgroup as real num_bytes to free.
3006 */
3007 num_bytes = qgroup->rsv.values[type];
3008
3009 ulist_reinit(fs_info->qgroup_ulist);
3010 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3011 qgroup_to_aux(qgroup), GFP_ATOMIC);
3012 if (ret < 0)
3013 goto out;
3014 ULIST_ITER_INIT(&uiter);
3015 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3016 struct btrfs_qgroup *qg;
3017 struct btrfs_qgroup_list *glist;
3018
3019 qg = unode_aux_to_qgroup(unode);
3020
3021 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
3022 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3023
3024 list_for_each_entry(glist, &qg->groups, next_group) {
3025 ret = ulist_add(fs_info->qgroup_ulist,
3026 glist->group->qgroupid,
3027 qgroup_to_aux(glist->group), GFP_ATOMIC);
3028 if (ret < 0)
3029 goto out;
3030 }
3031 }
3032
3033 out:
3034 spin_unlock(&fs_info->qgroup_lock);
3035 }
3036
3037 /*
3038 * Check if the leaf is the last leaf. Which means all node pointers
3039 * are at their last position.
3040 */
3041 static bool is_last_leaf(struct btrfs_path *path)
3042 {
3043 int i;
3044
3045 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3046 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3047 return false;
3048 }
3049 return true;
3050 }
3051
3052 /*
3053 * returns < 0 on error, 0 when more leafs are to be scanned.
3054 * returns 1 when done.
3055 */
3056 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3057 struct btrfs_path *path)
3058 {
3059 struct btrfs_fs_info *fs_info = trans->fs_info;
3060 struct btrfs_key found;
3061 struct extent_buffer *scratch_leaf = NULL;
3062 struct ulist *roots = NULL;
3063 u64 num_bytes;
3064 bool done;
3065 int slot;
3066 int ret;
3067
3068 mutex_lock(&fs_info->qgroup_rescan_lock);
3069 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3070 &fs_info->qgroup_rescan_progress,
3071 path, 1, 0);
3072
3073 btrfs_debug(fs_info,
3074 "current progress key (%llu %u %llu), search_slot ret %d",
3075 fs_info->qgroup_rescan_progress.objectid,
3076 fs_info->qgroup_rescan_progress.type,
3077 fs_info->qgroup_rescan_progress.offset, ret);
3078
3079 if (ret) {
3080 /*
3081 * The rescan is about to end, we will not be scanning any
3082 * further blocks. We cannot unset the RESCAN flag here, because
3083 * we want to commit the transaction if everything went well.
3084 * To make the live accounting work in this phase, we set our
3085 * scan progress pointer such that every real extent objectid
3086 * will be smaller.
3087 */
3088 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3089 btrfs_release_path(path);
3090 mutex_unlock(&fs_info->qgroup_rescan_lock);
3091 return ret;
3092 }
3093 done = is_last_leaf(path);
3094
3095 btrfs_item_key_to_cpu(path->nodes[0], &found,
3096 btrfs_header_nritems(path->nodes[0]) - 1);
3097 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3098
3099 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3100 if (!scratch_leaf) {
3101 ret = -ENOMEM;
3102 mutex_unlock(&fs_info->qgroup_rescan_lock);
3103 goto out;
3104 }
3105 extent_buffer_get(scratch_leaf);
3106 btrfs_tree_read_lock(scratch_leaf);
3107 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
3108 slot = path->slots[0];
3109 btrfs_release_path(path);
3110 mutex_unlock(&fs_info->qgroup_rescan_lock);
3111
3112 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3113 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3114 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3115 found.type != BTRFS_METADATA_ITEM_KEY)
3116 continue;
3117 if (found.type == BTRFS_METADATA_ITEM_KEY)
3118 num_bytes = fs_info->nodesize;
3119 else
3120 num_bytes = found.offset;
3121
3122 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3123 &roots, false);
3124 if (ret < 0)
3125 goto out;
3126 /* For rescan, just pass old_roots as NULL */
3127 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3128 num_bytes, NULL, roots);
3129 if (ret < 0)
3130 goto out;
3131 }
3132 out:
3133 if (scratch_leaf) {
3134 btrfs_tree_read_unlock_blocking(scratch_leaf);
3135 free_extent_buffer(scratch_leaf);
3136 }
3137
3138 if (done && !ret) {
3139 ret = 1;
3140 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3141 }
3142 return ret;
3143 }
3144
3145 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3146 {
3147 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3148 qgroup_rescan_work);
3149 struct btrfs_path *path;
3150 struct btrfs_trans_handle *trans = NULL;
3151 int err = -ENOMEM;
3152 int ret = 0;
3153
3154 path = btrfs_alloc_path();
3155 if (!path)
3156 goto out;
3157 /*
3158 * Rescan should only search for commit root, and any later difference
3159 * should be recorded by qgroup
3160 */
3161 path->search_commit_root = 1;
3162 path->skip_locking = 1;
3163
3164 err = 0;
3165 while (!err && !btrfs_fs_closing(fs_info)) {
3166 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3167 if (IS_ERR(trans)) {
3168 err = PTR_ERR(trans);
3169 break;
3170 }
3171 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3172 err = -EINTR;
3173 } else {
3174 err = qgroup_rescan_leaf(trans, path);
3175 }
3176 if (err > 0)
3177 btrfs_commit_transaction(trans);
3178 else
3179 btrfs_end_transaction(trans);
3180 }
3181
3182 out:
3183 btrfs_free_path(path);
3184
3185 mutex_lock(&fs_info->qgroup_rescan_lock);
3186 if (!btrfs_fs_closing(fs_info))
3187 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3188
3189 if (err > 0 &&
3190 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3191 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3192 } else if (err < 0) {
3193 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3194 }
3195 mutex_unlock(&fs_info->qgroup_rescan_lock);
3196
3197 /*
3198 * only update status, since the previous part has already updated the
3199 * qgroup info.
3200 */
3201 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3202 if (IS_ERR(trans)) {
3203 err = PTR_ERR(trans);
3204 btrfs_err(fs_info,
3205 "fail to start transaction for status update: %d",
3206 err);
3207 goto done;
3208 }
3209 ret = update_qgroup_status_item(trans);
3210 if (ret < 0) {
3211 err = ret;
3212 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3213 }
3214 btrfs_end_transaction(trans);
3215
3216 if (btrfs_fs_closing(fs_info)) {
3217 btrfs_info(fs_info, "qgroup scan paused");
3218 } else if (err >= 0) {
3219 btrfs_info(fs_info, "qgroup scan completed%s",
3220 err > 0 ? " (inconsistency flag cleared)" : "");
3221 } else {
3222 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3223 }
3224
3225 done:
3226 mutex_lock(&fs_info->qgroup_rescan_lock);
3227 fs_info->qgroup_rescan_running = false;
3228 mutex_unlock(&fs_info->qgroup_rescan_lock);
3229 complete_all(&fs_info->qgroup_rescan_completion);
3230 }
3231
3232 /*
3233 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3234 * memory required for the rescan context.
3235 */
3236 static int
3237 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3238 int init_flags)
3239 {
3240 int ret = 0;
3241
3242 if (!init_flags) {
3243 /* we're resuming qgroup rescan at mount time */
3244 if (!(fs_info->qgroup_flags &
3245 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3246 btrfs_warn(fs_info,
3247 "qgroup rescan init failed, qgroup is not enabled");
3248 ret = -EINVAL;
3249 } else if (!(fs_info->qgroup_flags &
3250 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3251 btrfs_warn(fs_info,
3252 "qgroup rescan init failed, qgroup rescan is not queued");
3253 ret = -EINVAL;
3254 }
3255
3256 if (ret)
3257 return ret;
3258 }
3259
3260 mutex_lock(&fs_info->qgroup_rescan_lock);
3261 spin_lock(&fs_info->qgroup_lock);
3262
3263 if (init_flags) {
3264 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3265 btrfs_warn(fs_info,
3266 "qgroup rescan is already in progress");
3267 ret = -EINPROGRESS;
3268 } else if (!(fs_info->qgroup_flags &
3269 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3270 btrfs_warn(fs_info,
3271 "qgroup rescan init failed, qgroup is not enabled");
3272 ret = -EINVAL;
3273 }
3274
3275 if (ret) {
3276 spin_unlock(&fs_info->qgroup_lock);
3277 mutex_unlock(&fs_info->qgroup_rescan_lock);
3278 return ret;
3279 }
3280 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3281 }
3282
3283 memset(&fs_info->qgroup_rescan_progress, 0,
3284 sizeof(fs_info->qgroup_rescan_progress));
3285 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3286 init_completion(&fs_info->qgroup_rescan_completion);
3287 fs_info->qgroup_rescan_running = true;
3288
3289 spin_unlock(&fs_info->qgroup_lock);
3290 mutex_unlock(&fs_info->qgroup_rescan_lock);
3291
3292 memset(&fs_info->qgroup_rescan_work, 0,
3293 sizeof(fs_info->qgroup_rescan_work));
3294 btrfs_init_work(&fs_info->qgroup_rescan_work,
3295 btrfs_qgroup_rescan_helper,
3296 btrfs_qgroup_rescan_worker, NULL, NULL);
3297 return 0;
3298 }
3299
3300 static void
3301 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3302 {
3303 struct rb_node *n;
3304 struct btrfs_qgroup *qgroup;
3305
3306 spin_lock(&fs_info->qgroup_lock);
3307 /* clear all current qgroup tracking information */
3308 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3309 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3310 qgroup->rfer = 0;
3311 qgroup->rfer_cmpr = 0;
3312 qgroup->excl = 0;
3313 qgroup->excl_cmpr = 0;
3314 qgroup_dirty(fs_info, qgroup);
3315 }
3316 spin_unlock(&fs_info->qgroup_lock);
3317 }
3318
3319 int
3320 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3321 {
3322 int ret = 0;
3323 struct btrfs_trans_handle *trans;
3324
3325 ret = qgroup_rescan_init(fs_info, 0, 1);
3326 if (ret)
3327 return ret;
3328
3329 /*
3330 * We have set the rescan_progress to 0, which means no more
3331 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3332 * However, btrfs_qgroup_account_ref may be right after its call
3333 * to btrfs_find_all_roots, in which case it would still do the
3334 * accounting.
3335 * To solve this, we're committing the transaction, which will
3336 * ensure we run all delayed refs and only after that, we are
3337 * going to clear all tracking information for a clean start.
3338 */
3339
3340 trans = btrfs_join_transaction(fs_info->fs_root);
3341 if (IS_ERR(trans)) {
3342 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3343 return PTR_ERR(trans);
3344 }
3345 ret = btrfs_commit_transaction(trans);
3346 if (ret) {
3347 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3348 return ret;
3349 }
3350
3351 qgroup_rescan_zero_tracking(fs_info);
3352
3353 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3354 &fs_info->qgroup_rescan_work);
3355
3356 return 0;
3357 }
3358
3359 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3360 bool interruptible)
3361 {
3362 int running;
3363 int ret = 0;
3364
3365 mutex_lock(&fs_info->qgroup_rescan_lock);
3366 spin_lock(&fs_info->qgroup_lock);
3367 running = fs_info->qgroup_rescan_running;
3368 spin_unlock(&fs_info->qgroup_lock);
3369 mutex_unlock(&fs_info->qgroup_rescan_lock);
3370
3371 if (!running)
3372 return 0;
3373
3374 if (interruptible)
3375 ret = wait_for_completion_interruptible(
3376 &fs_info->qgroup_rescan_completion);
3377 else
3378 wait_for_completion(&fs_info->qgroup_rescan_completion);
3379
3380 return ret;
3381 }
3382
3383 /*
3384 * this is only called from open_ctree where we're still single threaded, thus
3385 * locking is omitted here.
3386 */
3387 void
3388 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3389 {
3390 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3391 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3392 &fs_info->qgroup_rescan_work);
3393 }
3394
3395 /*
3396 * Reserve qgroup space for range [start, start + len).
3397 *
3398 * This function will either reserve space from related qgroups or doing
3399 * nothing if the range is already reserved.
3400 *
3401 * Return 0 for successful reserve
3402 * Return <0 for error (including -EQUOT)
3403 *
3404 * NOTE: this function may sleep for memory allocation.
3405 * if btrfs_qgroup_reserve_data() is called multiple times with
3406 * same @reserved, caller must ensure when error happens it's OK
3407 * to free *ALL* reserved space.
3408 */
3409 int btrfs_qgroup_reserve_data(struct inode *inode,
3410 struct extent_changeset **reserved_ret, u64 start,
3411 u64 len)
3412 {
3413 struct btrfs_root *root = BTRFS_I(inode)->root;
3414 struct ulist_node *unode;
3415 struct ulist_iterator uiter;
3416 struct extent_changeset *reserved;
3417 u64 orig_reserved;
3418 u64 to_reserve;
3419 int ret;
3420
3421 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3422 !is_fstree(root->root_key.objectid) || len == 0)
3423 return 0;
3424
3425 /* @reserved parameter is mandatory for qgroup */
3426 if (WARN_ON(!reserved_ret))
3427 return -EINVAL;
3428 if (!*reserved_ret) {
3429 *reserved_ret = extent_changeset_alloc();
3430 if (!*reserved_ret)
3431 return -ENOMEM;
3432 }
3433 reserved = *reserved_ret;
3434 /* Record already reserved space */
3435 orig_reserved = reserved->bytes_changed;
3436 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3437 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3438
3439 /* Newly reserved space */
3440 to_reserve = reserved->bytes_changed - orig_reserved;
3441 trace_btrfs_qgroup_reserve_data(inode, start, len,
3442 to_reserve, QGROUP_RESERVE);
3443 if (ret < 0)
3444 goto cleanup;
3445 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3446 if (ret < 0)
3447 goto cleanup;
3448
3449 return ret;
3450
3451 cleanup:
3452 /* cleanup *ALL* already reserved ranges */
3453 ULIST_ITER_INIT(&uiter);
3454 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3455 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3456 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3457 extent_changeset_release(reserved);
3458 return ret;
3459 }
3460
3461 /* Free ranges specified by @reserved, normally in error path */
3462 static int qgroup_free_reserved_data(struct inode *inode,
3463 struct extent_changeset *reserved, u64 start, u64 len)
3464 {
3465 struct btrfs_root *root = BTRFS_I(inode)->root;
3466 struct ulist_node *unode;
3467 struct ulist_iterator uiter;
3468 struct extent_changeset changeset;
3469 int freed = 0;
3470 int ret;
3471
3472 extent_changeset_init(&changeset);
3473 len = round_up(start + len, root->fs_info->sectorsize);
3474 start = round_down(start, root->fs_info->sectorsize);
3475
3476 ULIST_ITER_INIT(&uiter);
3477 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3478 u64 range_start = unode->val;
3479 /* unode->aux is the inclusive end */
3480 u64 range_len = unode->aux - range_start + 1;
3481 u64 free_start;
3482 u64 free_len;
3483
3484 extent_changeset_release(&changeset);
3485
3486 /* Only free range in range [start, start + len) */
3487 if (range_start >= start + len ||
3488 range_start + range_len <= start)
3489 continue;
3490 free_start = max(range_start, start);
3491 free_len = min(start + len, range_start + range_len) -
3492 free_start;
3493 /*
3494 * TODO: To also modify reserved->ranges_reserved to reflect
3495 * the modification.
3496 *
3497 * However as long as we free qgroup reserved according to
3498 * EXTENT_QGROUP_RESERVED, we won't double free.
3499 * So not need to rush.
3500 */
3501 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3502 free_start, free_start + free_len - 1,
3503 EXTENT_QGROUP_RESERVED, &changeset);
3504 if (ret < 0)
3505 goto out;
3506 freed += changeset.bytes_changed;
3507 }
3508 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3509 BTRFS_QGROUP_RSV_DATA);
3510 ret = freed;
3511 out:
3512 extent_changeset_release(&changeset);
3513 return ret;
3514 }
3515
3516 static int __btrfs_qgroup_release_data(struct inode *inode,
3517 struct extent_changeset *reserved, u64 start, u64 len,
3518 int free)
3519 {
3520 struct extent_changeset changeset;
3521 int trace_op = QGROUP_RELEASE;
3522 int ret;
3523
3524 /* In release case, we shouldn't have @reserved */
3525 WARN_ON(!free && reserved);
3526 if (free && reserved)
3527 return qgroup_free_reserved_data(inode, reserved, start, len);
3528 extent_changeset_init(&changeset);
3529 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3530 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3531 if (ret < 0)
3532 goto out;
3533
3534 if (free)
3535 trace_op = QGROUP_FREE;
3536 trace_btrfs_qgroup_release_data(inode, start, len,
3537 changeset.bytes_changed, trace_op);
3538 if (free)
3539 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3540 BTRFS_I(inode)->root->root_key.objectid,
3541 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3542 ret = changeset.bytes_changed;
3543 out:
3544 extent_changeset_release(&changeset);
3545 return ret;
3546 }
3547
3548 /*
3549 * Free a reserved space range from io_tree and related qgroups
3550 *
3551 * Should be called when a range of pages get invalidated before reaching disk.
3552 * Or for error cleanup case.
3553 * if @reserved is given, only reserved range in [@start, @start + @len) will
3554 * be freed.
3555 *
3556 * For data written to disk, use btrfs_qgroup_release_data().
3557 *
3558 * NOTE: This function may sleep for memory allocation.
3559 */
3560 int btrfs_qgroup_free_data(struct inode *inode,
3561 struct extent_changeset *reserved, u64 start, u64 len)
3562 {
3563 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3564 }
3565
3566 /*
3567 * Release a reserved space range from io_tree only.
3568 *
3569 * Should be called when a range of pages get written to disk and corresponding
3570 * FILE_EXTENT is inserted into corresponding root.
3571 *
3572 * Since new qgroup accounting framework will only update qgroup numbers at
3573 * commit_transaction() time, its reserved space shouldn't be freed from
3574 * related qgroups.
3575 *
3576 * But we should release the range from io_tree, to allow further write to be
3577 * COWed.
3578 *
3579 * NOTE: This function may sleep for memory allocation.
3580 */
3581 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3582 {
3583 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3584 }
3585
3586 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3587 enum btrfs_qgroup_rsv_type type)
3588 {
3589 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3590 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3591 return;
3592 if (num_bytes == 0)
3593 return;
3594
3595 spin_lock(&root->qgroup_meta_rsv_lock);
3596 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3597 root->qgroup_meta_rsv_prealloc += num_bytes;
3598 else
3599 root->qgroup_meta_rsv_pertrans += num_bytes;
3600 spin_unlock(&root->qgroup_meta_rsv_lock);
3601 }
3602
3603 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3604 enum btrfs_qgroup_rsv_type type)
3605 {
3606 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3607 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3608 return 0;
3609 if (num_bytes == 0)
3610 return 0;
3611
3612 spin_lock(&root->qgroup_meta_rsv_lock);
3613 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3614 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3615 num_bytes);
3616 root->qgroup_meta_rsv_prealloc -= num_bytes;
3617 } else {
3618 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3619 num_bytes);
3620 root->qgroup_meta_rsv_pertrans -= num_bytes;
3621 }
3622 spin_unlock(&root->qgroup_meta_rsv_lock);
3623 return num_bytes;
3624 }
3625
3626 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3627 enum btrfs_qgroup_rsv_type type, bool enforce)
3628 {
3629 struct btrfs_fs_info *fs_info = root->fs_info;
3630 int ret;
3631
3632 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3633 !is_fstree(root->root_key.objectid) || num_bytes == 0)
3634 return 0;
3635
3636 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3637 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3638 ret = qgroup_reserve(root, num_bytes, enforce, type);
3639 if (ret < 0)
3640 return ret;
3641 /*
3642 * Record what we have reserved into root.
3643 *
3644 * To avoid quota disabled->enabled underflow.
3645 * In that case, we may try to free space we haven't reserved
3646 * (since quota was disabled), so record what we reserved into root.
3647 * And ensure later release won't underflow this number.
3648 */
3649 add_root_meta_rsv(root, num_bytes, type);
3650 return ret;
3651 }
3652
3653 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3654 {
3655 struct btrfs_fs_info *fs_info = root->fs_info;
3656
3657 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3658 !is_fstree(root->root_key.objectid))
3659 return;
3660
3661 /* TODO: Update trace point to handle such free */
3662 trace_qgroup_meta_free_all_pertrans(root);
3663 /* Special value -1 means to free all reserved space */
3664 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3665 BTRFS_QGROUP_RSV_META_PERTRANS);
3666 }
3667
3668 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3669 enum btrfs_qgroup_rsv_type type)
3670 {
3671 struct btrfs_fs_info *fs_info = root->fs_info;
3672
3673 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3674 !is_fstree(root->root_key.objectid))
3675 return;
3676
3677 /*
3678 * reservation for META_PREALLOC can happen before quota is enabled,
3679 * which can lead to underflow.
3680 * Here ensure we will only free what we really have reserved.
3681 */
3682 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3683 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3684 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3685 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3686 num_bytes, type);
3687 }
3688
3689 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3690 int num_bytes)
3691 {
3692 struct btrfs_root *quota_root = fs_info->quota_root;
3693 struct btrfs_qgroup *qgroup;
3694 struct ulist_node *unode;
3695 struct ulist_iterator uiter;
3696 int ret = 0;
3697
3698 if (num_bytes == 0)
3699 return;
3700 if (!quota_root)
3701 return;
3702
3703 spin_lock(&fs_info->qgroup_lock);
3704 qgroup = find_qgroup_rb(fs_info, ref_root);
3705 if (!qgroup)
3706 goto out;
3707 ulist_reinit(fs_info->qgroup_ulist);
3708 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3709 qgroup_to_aux(qgroup), GFP_ATOMIC);
3710 if (ret < 0)
3711 goto out;
3712 ULIST_ITER_INIT(&uiter);
3713 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3714 struct btrfs_qgroup *qg;
3715 struct btrfs_qgroup_list *glist;
3716
3717 qg = unode_aux_to_qgroup(unode);
3718
3719 qgroup_rsv_release(fs_info, qg, num_bytes,
3720 BTRFS_QGROUP_RSV_META_PREALLOC);
3721 qgroup_rsv_add(fs_info, qg, num_bytes,
3722 BTRFS_QGROUP_RSV_META_PERTRANS);
3723 list_for_each_entry(glist, &qg->groups, next_group) {
3724 ret = ulist_add(fs_info->qgroup_ulist,
3725 glist->group->qgroupid,
3726 qgroup_to_aux(glist->group), GFP_ATOMIC);
3727 if (ret < 0)
3728 goto out;
3729 }
3730 }
3731 out:
3732 spin_unlock(&fs_info->qgroup_lock);
3733 }
3734
3735 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3736 {
3737 struct btrfs_fs_info *fs_info = root->fs_info;
3738
3739 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3740 !is_fstree(root->root_key.objectid))
3741 return;
3742 /* Same as btrfs_qgroup_free_meta_prealloc() */
3743 num_bytes = sub_root_meta_rsv(root, num_bytes,
3744 BTRFS_QGROUP_RSV_META_PREALLOC);
3745 trace_qgroup_meta_convert(root, num_bytes);
3746 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3747 }
3748
3749 /*
3750 * Check qgroup reserved space leaking, normally at destroy inode
3751 * time
3752 */
3753 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3754 {
3755 struct extent_changeset changeset;
3756 struct ulist_node *unode;
3757 struct ulist_iterator iter;
3758 int ret;
3759
3760 extent_changeset_init(&changeset);
3761 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3762 EXTENT_QGROUP_RESERVED, &changeset);
3763
3764 WARN_ON(ret < 0);
3765 if (WARN_ON(changeset.bytes_changed)) {
3766 ULIST_ITER_INIT(&iter);
3767 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3768 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3769 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3770 inode->i_ino, unode->val, unode->aux);
3771 }
3772 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3773 BTRFS_I(inode)->root->root_key.objectid,
3774 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3775
3776 }
3777 extent_changeset_release(&changeset);
3778 }
Ubuntu Hirsute Linux kernel mirror