]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - fs/btrfs/qgroup.c
projects / mirror_ubuntu-hirsute-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: qgroup: Introduce function to trace two swaped extents
[mirror_ubuntu-hirsute-kernel.git] / fs / btrfs / qgroup.c
CommitLineData
c1d7c514 1// SPDX-License-Identifier: GPL-2.0
bed92eae
AJ		 2/*
3 * Copyright (C) 2011 STRATO. All rights reserved.
bed92eae
AJ		 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>
55e301fd 13#include <linux/btrfs.h>
a514d638 14#include <linux/sizes.h>
bed92eae
AJ		 15
16#include "ctree.h"
17#include "transaction.h"
18#include "disk-io.h"
19#include "locking.h"
20#include "ulist.h"
bed92eae 21#include "backref.h"
2f232036 22#include "extent_io.h"
fcebe456 23#include "qgroup.h"
bed92eae 24
e69bcee3 25
bed92eae
AJ		 26/* TODO XXX FIXME
27 * - subvol delete -> delete when ref goes to 0? delete limits also?
28 * - reorganize keys
29 * - compressed
30 * - sync
bed92eae
AJ		 31 * - copy also limits on subvol creation
32 * - limit
33 * - caches fuer ulists
34 * - performance benchmarks
35 * - check all ioctl parameters
36 */
37
f59c0347
QW		 38/*
39 * Helpers to access qgroup reservation
40 *
41 * Callers should ensure the lock context and type are valid
42 */
43
44static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45{
46 u64 ret = 0;
47 int i;
48
49 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 ret += qgroup->rsv.values[i];
51
52 return ret;
53}
54
55#ifdef CONFIG_BTRFS_DEBUG
56static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57{
58 if (type == BTRFS_QGROUP_RSV_DATA)
59 return "data";
733e03a0
QW		 60 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 return "meta_pertrans";
62 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 return "meta_prealloc";
f59c0347
QW		 64 return NULL;
65}
66#endif
67
64ee4e75
QW		 68static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 struct btrfs_qgroup *qgroup, u64 num_bytes,
f59c0347
QW		 70 enum btrfs_qgroup_rsv_type type)
71{
64ee4e75 72 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
f59c0347
QW		 73 qgroup->rsv.values[type] += num_bytes;
74}
75
64ee4e75
QW		 76static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 struct btrfs_qgroup *qgroup, u64 num_bytes,
f59c0347
QW		 78 enum btrfs_qgroup_rsv_type type)
79{
64ee4e75 80 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
f59c0347
QW		 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
64ee4e75
QW		 94static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 struct btrfs_qgroup *dest,
96 struct btrfs_qgroup *src)
f59c0347
QW		 97{
98 int i;
99
100 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
64ee4e75 101 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
f59c0347
QW		 102}
103
64ee4e75
QW		 104static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 struct btrfs_qgroup *dest,
f59c0347
QW		 106 struct btrfs_qgroup *src)
107{
108 int i;
109
110 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
64ee4e75 111 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
f59c0347
QW		 112}
113
9c542136
QW		 114static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115 int mod)
116{
117 if (qg->old_refcnt < seq)
118 qg->old_refcnt = seq;
119 qg->old_refcnt += mod;
120}
121
122static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123 int mod)
124{
125 if (qg->new_refcnt < seq)
126 qg->new_refcnt = seq;
127 qg->new_refcnt += mod;
128}
129
130static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131{
132 if (qg->old_refcnt < seq)
133 return 0;
134 return qg->old_refcnt - seq;
135}
136
137static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138{
139 if (qg->new_refcnt < seq)
140 return 0;
141 return qg->new_refcnt - seq;
142}
143
bed92eae
AJ		 144/*
145 * glue structure to represent the relations between qgroups.
146 */
147struct btrfs_qgroup_list {
148 struct list_head next_group;
149 struct list_head next_member;
150 struct btrfs_qgroup *group;
151 struct btrfs_qgroup *member;
152};
153
ef2fff64
DS		 154static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155{
156 return (u64)(uintptr_t)qg;
157}
158
159static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160{
161 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162}
fcebe456 163
b382a324
JS		 164static int
165qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 int init_flags);
167static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
2f232036 168
58400fce 169/* must be called with qgroup_ioctl_lock held */
bed92eae
AJ		 170static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171 u64 qgroupid)
172{
173 struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 struct btrfs_qgroup *qgroup;
175
176 while (n) {
177 qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 if (qgroup->qgroupid < qgroupid)
179 n = n->rb_left;
180 else if (qgroup->qgroupid > qgroupid)
181 n = n->rb_right;
182 else
183 return qgroup;
184 }
185 return NULL;
186}
187
188/* must be called with qgroup_lock held */
189static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190 u64 qgroupid)
191{
192 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 struct rb_node *parent = NULL;
194 struct btrfs_qgroup *qgroup;
195
196 while (*p) {
197 parent = *p;
198 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199
200 if (qgroup->qgroupid < qgroupid)
201 p = &(*p)->rb_left;
202 else if (qgroup->qgroupid > qgroupid)
203 p = &(*p)->rb_right;
204 else
205 return qgroup;
206 }
207
208 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 if (!qgroup)
210 return ERR_PTR(-ENOMEM);
211
212 qgroup->qgroupid = qgroupid;
213 INIT_LIST_HEAD(&qgroup->groups);
214 INIT_LIST_HEAD(&qgroup->members);
215 INIT_LIST_HEAD(&qgroup->dirty);
216
217 rb_link_node(&qgroup->node, parent, p);
218 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219
220 return qgroup;
221}
222
4082bd3d 223static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
bed92eae 224{
bed92eae
AJ		 225 struct btrfs_qgroup_list *list;
226
bed92eae 227 list_del(&qgroup->dirty);
bed92eae
AJ		 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);
4082bd3d 244}
bed92eae 245
4082bd3d
WS		 246/* must be called with qgroup_lock held */
247static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248{
249 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250
251 if (!qgroup)
252 return -ENOENT;
253
254 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 __del_qgroup_rb(qgroup);
bed92eae
AJ		 256 return 0;
257}
258
259/* must be called with qgroup_lock held */
260static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 u64 memberid, u64 parentid)
262{
263 struct btrfs_qgroup *member;
264 struct btrfs_qgroup *parent;
265 struct btrfs_qgroup_list *list;
266
267 member = find_qgroup_rb(fs_info, memberid);
268 parent = find_qgroup_rb(fs_info, parentid);
269 if (!member || !parent)
270 return -ENOENT;
271
272 list = kzalloc(sizeof(*list), GFP_ATOMIC);
273 if (!list)
274 return -ENOMEM;
275
276 list->group = parent;
277 list->member = member;
278 list_add_tail(&list->next_group, &member->groups);
279 list_add_tail(&list->next_member, &parent->members);
280
281 return 0;
282}
283
284/* must be called with qgroup_lock held */
285static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 u64 memberid, u64 parentid)
287{
288 struct btrfs_qgroup *member;
289 struct btrfs_qgroup *parent;
290 struct btrfs_qgroup_list *list;
291
292 member = find_qgroup_rb(fs_info, memberid);
293 parent = find_qgroup_rb(fs_info, parentid);
294 if (!member || !parent)
295 return -ENOENT;
296
297 list_for_each_entry(list, &member->groups, next_group) {
298 if (list->group == parent) {
299 list_del(&list->next_group);
300 list_del(&list->next_member);
301 kfree(list);
302 return 0;
303 }
304 }
305 return -ENOENT;
306}
307
faa2dbf0
JB		 308#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310 u64 rfer, u64 excl)
311{
312 struct btrfs_qgroup *qgroup;
313
314 qgroup = find_qgroup_rb(fs_info, qgroupid);
315 if (!qgroup)
316 return -EINVAL;
317 if (qgroup->rfer != rfer || qgroup->excl != excl)
318 return -EINVAL;
319 return 0;
320}
321#endif
322
bed92eae
AJ		 323/*
324 * The full config is read in one go, only called from open_ctree()
325 * It doesn't use any locking, as at this point we're still single-threaded
326 */
327int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328{
329 struct btrfs_key key;
330 struct btrfs_key found_key;
331 struct btrfs_root *quota_root = fs_info->quota_root;
332 struct btrfs_path *path = NULL;
333 struct extent_buffer *l;
334 int slot;
335 int ret = 0;
336 u64 flags = 0;
b382a324 337 u64 rescan_progress = 0;
bed92eae 338
afcdd129 339 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
bed92eae
AJ		 340 return 0;
341
323b88f4 342 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
1e8f9158
WS		 343 if (!fs_info->qgroup_ulist) {
344 ret = -ENOMEM;
345 goto out;
346 }
347
bed92eae
AJ		 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) {
efe120a0
FH		 382 btrfs_err(fs_info,
383 "old qgroup version, quota disabled");
bed92eae
AJ		 384 goto out;
385 }
386 if (btrfs_qgroup_status_generation(l, ptr) !=
387 fs_info->generation) {
388 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
efe120a0 389 btrfs_err(fs_info,
5d163e0e 390 "qgroup generation mismatch, marked as inconsistent");
bed92eae
AJ		 391 }
392 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 ptr);
b382a324 394 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
bed92eae
AJ		 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)) {
d41e36a0 405 btrfs_err(fs_info, "inconsistent qgroup config");
bed92eae
AJ		 406 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407 }
408 if (!qgroup) {
409 qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 if (IS_ERR(qgroup)) {
411 ret = PTR_ERR(qgroup);
412 goto out;
413 }
414 }
415 switch (found_key.type) {
416 case BTRFS_QGROUP_INFO_KEY: {
417 struct btrfs_qgroup_info_item *ptr;
418
419 ptr = btrfs_item_ptr(l, slot,
420 struct btrfs_qgroup_info_item);
421 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 /* generation currently unused */
426 break;
427 }
428 case BTRFS_QGROUP_LIMIT_KEY: {
429 struct btrfs_qgroup_limit_item *ptr;
430
431 ptr = btrfs_item_ptr(l, slot,
432 struct btrfs_qgroup_limit_item);
433 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438 break;
439 }
440 }
441next1:
442 ret = btrfs_next_item(quota_root, path);
443 if (ret < 0)
444 goto out;
445 if (ret)
446 break;
447 }
448 btrfs_release_path(path);
449
450 /*
451 * pass 2: read all qgroup relations
452 */
453 key.objectid = 0;
454 key.type = BTRFS_QGROUP_RELATION_KEY;
455 key.offset = 0;
456 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457 if (ret)
458 goto out;
459 while (1) {
460 slot = path->slots[0];
461 l = path->nodes[0];
462 btrfs_item_key_to_cpu(l, &found_key, slot);
463
464 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465 goto next2;
466
467 if (found_key.objectid > found_key.offset) {
468 /* parent <- member, not needed to build config */
469 /* FIXME should we omit the key completely? */
470 goto next2;
471 }
472
473 ret = add_relation_rb(fs_info, found_key.objectid,
474 found_key.offset);
ff24858c 475 if (ret == -ENOENT) {
efe120a0
FH		 476 btrfs_warn(fs_info,
477 "orphan qgroup relation 0x%llx->0x%llx",
c1c9ff7c 478 found_key.objectid, found_key.offset);
ff24858c
AJ		 479 ret = 0; /* ignore the error */
480 }
bed92eae
AJ		 481 if (ret)
482 goto out;
483next2:
484 ret = btrfs_next_item(quota_root, path);
485 if (ret < 0)
486 goto out;
487 if (ret)
488 break;
489 }
490out:
491 fs_info->qgroup_flags |= flags;
afcdd129
JB		 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)
b382a324 496 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
bed92eae
AJ		 497 btrfs_free_path(path);
498
eb1716af 499 if (ret < 0) {
1e8f9158 500 ulist_free(fs_info->qgroup_ulist);
eb1716af 501 fs_info->qgroup_ulist = NULL;
b382a324 502 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
eb1716af 503 }
1e8f9158 504
bed92eae
AJ		 505 return ret < 0 ? ret : 0;
506}
507
508/*
e685da14
WS		 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.
bed92eae
AJ		 513 */
514void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515{
516 struct rb_node *n;
517 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 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);
4082bd3d 522 __del_qgroup_rb(qgroup);
bed92eae 523 }
1e7bac1e
WS		 524 /*
525 * we call btrfs_free_qgroup_config() when umounting
01327610 526 * filesystem and disabling quota, so we set qgroup_ulist
1e7bac1e
WS		 527 * to be null here to avoid double free.
528 */
1e8f9158 529 ulist_free(fs_info->qgroup_ulist);
1e7bac1e 530 fs_info->qgroup_ulist = NULL;
bed92eae
AJ		 531}
532
711169c4
LF		 533static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 u64 dst)
bed92eae
AJ		 535{
536 int ret;
711169c4 537 struct btrfs_root *quota_root = trans->fs_info->quota_root;
bed92eae
AJ		 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
99d7f09a
LF		 557static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 u64 dst)
bed92eae
AJ		 559{
560 int ret;
99d7f09a 561 struct btrfs_root *quota_root = trans->fs_info->quota_root;
bed92eae
AJ		 562 struct btrfs_path *path;
563 struct btrfs_key key;
564
565 path = btrfs_alloc_path();
566 if (!path)
567 return -ENOMEM;
568
569 key.objectid = src;
570 key.type = BTRFS_QGROUP_RELATION_KEY;
571 key.offset = dst;
572
573 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574 if (ret < 0)
575 goto out;
576
577 if (ret > 0) {
578 ret = -ENOENT;
579 goto out;
580 }
581
582 ret = btrfs_del_item(trans, quota_root, path);
583out:
584 btrfs_free_path(path);
585 return ret;
586}
587
588static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 struct btrfs_root *quota_root, u64 qgroupid)
590{
591 int ret;
592 struct btrfs_path *path;
593 struct btrfs_qgroup_info_item *qgroup_info;
594 struct btrfs_qgroup_limit_item *qgroup_limit;
595 struct extent_buffer *leaf;
596 struct btrfs_key key;
597
f5ee5c9a 598 if (btrfs_is_testing(quota_root->fs_info))
faa2dbf0 599 return 0;
fccb84c9 600
bed92eae
AJ		 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
0b4699dc
MF		 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
bed92eae
AJ		 615 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 sizeof(*qgroup_info));
0b4699dc 617 if (ret && ret != -EEXIST)
bed92eae
AJ		 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));
0b4699dc 636 if (ret && ret != -EEXIST)
bed92eae
AJ		 637 goto out;
638
639 leaf = path->nodes[0];
640 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_qgroup_limit_item);
642 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647
648 btrfs_mark_buffer_dirty(leaf);
649
650 ret = 0;
651out:
652 btrfs_free_path(path);
653 return ret;
654}
655
69104618 656static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
bed92eae
AJ		 657{
658 int ret;
69104618 659 struct btrfs_root *quota_root = trans->fs_info->quota_root;
bed92eae
AJ		 660 struct btrfs_path *path;
661 struct btrfs_key key;
662
663 path = btrfs_alloc_path();
664 if (!path)
665 return -ENOMEM;
666
667 key.objectid = 0;
668 key.type = BTRFS_QGROUP_INFO_KEY;
669 key.offset = qgroupid;
670 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 if (ret < 0)
672 goto out;
673
674 if (ret > 0) {
675 ret = -ENOENT;
676 goto out;
677 }
678
679 ret = btrfs_del_item(trans, quota_root, path);
680 if (ret)
681 goto out;
682
683 btrfs_release_path(path);
684
685 key.type = BTRFS_QGROUP_LIMIT_KEY;
686 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687 if (ret < 0)
688 goto out;
689
690 if (ret > 0) {
691 ret = -ENOENT;
692 goto out;
693 }
694
695 ret = btrfs_del_item(trans, quota_root, path);
696
697out:
698 btrfs_free_path(path);
699 return ret;
700}
701
702static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
1510e71c 703 struct btrfs_qgroup *qgroup)
bed92eae 704{
ac8a866a 705 struct btrfs_root *quota_root = trans->fs_info->quota_root;
bed92eae
AJ		 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;
1510e71c 715 key.offset = qgroup->qgroupid;
bed92eae
AJ		 716
717 path = btrfs_alloc_path();
84cbe2f7
WS		 718 if (!path)
719 return -ENOMEM;
720
ac8a866a 721 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
bed92eae
AJ		 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];
a3df41ee 730 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
1510e71c
DY		 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);
bed92eae
AJ		 736
737 btrfs_mark_buffer_dirty(l);
738
739out:
740 btrfs_free_path(path);
741 return ret;
742}
743
744static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
bed92eae
AJ		 745 struct btrfs_qgroup *qgroup)
746{
3e07e9a0
LF		 747 struct btrfs_fs_info *fs_info = trans->fs_info;
748 struct btrfs_root *quota_root = fs_info->quota_root;
bed92eae
AJ		 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
3e07e9a0 756 if (btrfs_is_testing(fs_info))
faa2dbf0 757 return 0;
fccb84c9 758
bed92eae
AJ		 759 key.objectid = 0;
760 key.type = BTRFS_QGROUP_INFO_KEY;
761 key.offset = qgroup->qgroupid;
762
763 path = btrfs_alloc_path();
84cbe2f7
WS		 764 if (!path)
765 return -ENOMEM;
766
3e07e9a0 767 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
bed92eae
AJ		 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];
a3df41ee 776 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
bed92eae
AJ		 777 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782
783 btrfs_mark_buffer_dirty(l);
784
785out:
786 btrfs_free_path(path);
787 return ret;
788}
789
2e980acd 790static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
bed92eae 791{
2e980acd
LF		 792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *quota_root = fs_info->quota_root;
bed92eae
AJ		 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();
84cbe2f7
WS		 806 if (!path)
807 return -ENOMEM;
808
2e980acd 809 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
bed92eae
AJ		 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);
2f232036
JS		 821 btrfs_set_qgroup_status_rescan(l, ptr,
822 fs_info->qgroup_rescan_progress.objectid);
bed92eae
AJ		 823
824 btrfs_mark_buffer_dirty(l);
825
826out:
827 btrfs_free_path(path);
828 return ret;
829}
830
831/*
832 * called with qgroup_lock held
833 */
834static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root)
836{
837 struct btrfs_path *path;
838 struct btrfs_key key;
06b3a860 839 struct extent_buffer *leaf = NULL;
bed92eae 840 int ret;
06b3a860 841 int nr = 0;
bed92eae 842
bed92eae
AJ		 843 path = btrfs_alloc_path();
844 if (!path)
845 return -ENOMEM;
846
06b3a860
WS		 847 path->leave_spinning = 1;
848
849 key.objectid = 0;
850 key.offset = 0;
851 key.type = 0;
bed92eae 852
06b3a860 853 while (1) {
bed92eae 854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
06b3a860
WS		 855 if (ret < 0)
856 goto out;
857 leaf = path->nodes[0];
858 nr = btrfs_header_nritems(leaf);
859 if (!nr)
bed92eae 860 break;
06b3a860
WS		 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);
bed92eae
AJ		 868 if (ret)
869 goto out;
06b3a860 870
bed92eae
AJ		 871 btrfs_release_path(path);
872 }
873 ret = 0;
874out:
bed92eae
AJ		 875 btrfs_free_path(path);
876 return ret;
877}
878
340f1aa2 879int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
bed92eae
AJ		 880{
881 struct btrfs_root *quota_root;
7708f029 882 struct btrfs_root *tree_root = fs_info->tree_root;
bed92eae
AJ		 883 struct btrfs_path *path = NULL;
884 struct btrfs_qgroup_status_item *ptr;
885 struct extent_buffer *leaf;
886 struct btrfs_key key;
7708f029
WS		 887 struct btrfs_key found_key;
888 struct btrfs_qgroup *qgroup = NULL;
340f1aa2 889 struct btrfs_trans_handle *trans = NULL;
bed92eae 890 int ret = 0;
7708f029 891 int slot;
bed92eae 892
f2f6ed3d 893 mutex_lock(&fs_info->qgroup_ioctl_lock);
5d23515b 894 if (fs_info->quota_root)
bed92eae 895 goto out;
bed92eae 896
340f1aa2
NB		 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
52bf8e7a 912 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
1e8f9158
WS		 913 if (!fs_info->qgroup_ulist) {
914 ret = -ENOMEM;
340f1aa2 915 btrfs_abort_transaction(trans, ret);
1e8f9158
WS		 916 goto out;
917 }
918
bed92eae
AJ		 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);
340f1aa2 926 btrfs_abort_transaction(trans, ret);
bed92eae
AJ		 927 goto out;
928 }
929
930 path = btrfs_alloc_path();
5b7ff5b3
TI		 931 if (!path) {
932 ret = -ENOMEM;
340f1aa2 933 btrfs_abort_transaction(trans, ret);
5b7ff5b3
TI		 934 goto out_free_root;
935 }
bed92eae
AJ		 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));
340f1aa2
NB		 943 if (ret) {
944 btrfs_abort_transaction(trans, ret);
5b7ff5b3 945 goto out_free_path;
340f1aa2 946 }
bed92eae
AJ		 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);
2f232036 956 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
bed92eae
AJ		 957
958 btrfs_mark_buffer_dirty(leaf);
959
7708f029
WS		 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;
340f1aa2
NB		 968 if (ret < 0) {
969 btrfs_abort_transaction(trans, ret);
7708f029 970 goto out_free_path;
340f1aa2 971 }
7708f029
WS		 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);
340f1aa2
NB		 981 if (ret) {
982 btrfs_abort_transaction(trans, ret);
7708f029 983 goto out_free_path;
340f1aa2 984 }
7708f029 985
7708f029
WS		 986 qgroup = add_qgroup_rb(fs_info, found_key.offset);
987 if (IS_ERR(qgroup)) {
7708f029 988 ret = PTR_ERR(qgroup);
340f1aa2 989 btrfs_abort_transaction(trans, ret);
7708f029
WS		 990 goto out_free_path;
991 }
7708f029
WS		 992 }
993 ret = btrfs_next_item(tree_root, path);
340f1aa2
NB		 994 if (ret < 0) {
995 btrfs_abort_transaction(trans, ret);
7708f029 996 goto out_free_path;
340f1aa2 997 }
7708f029
WS		 998 if (ret)
999 break;
1000 }
1001
1002out_add_root:
1003 btrfs_release_path(path);
1004 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
340f1aa2
NB		 1005 if (ret) {
1006 btrfs_abort_transaction(trans, ret);
7708f029 1007 goto out_free_path;
340f1aa2 1008 }
7708f029 1009
7708f029
WS		 1010 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011 if (IS_ERR(qgroup)) {
7708f029 1012 ret = PTR_ERR(qgroup);
340f1aa2 1013 btrfs_abort_transaction(trans, ret);
7708f029
WS		 1014 goto out_free_path;
1015 }
58400fce 1016 spin_lock(&fs_info->qgroup_lock);
bed92eae 1017 fs_info->quota_root = quota_root;
5d23515b 1018 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
bed92eae 1019 spin_unlock(&fs_info->qgroup_lock);
340f1aa2
NB		 1020
1021 ret = btrfs_commit_transaction(trans);
b9b8a41a
DC		 1022 trans = NULL;
1023 if (ret)
340f1aa2 1024 goto out_free_path;
340f1aa2 1025
5d23515b
NB		 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
5b7ff5b3 1033out_free_path:
bed92eae 1034 btrfs_free_path(path);
5b7ff5b3
TI		 1035out_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 }
1041out:
eb1716af 1042 if (ret) {
1e8f9158 1043 ulist_free(fs_info->qgroup_ulist);
eb1716af 1044 fs_info->qgroup_ulist = NULL;
340f1aa2
NB		 1045 if (trans)
1046 btrfs_end_transaction(trans);
eb1716af 1047 }
f2f6ed3d 1048 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1049 return ret;
1050}
1051
340f1aa2 1052int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
bed92eae 1053{
bed92eae 1054 struct btrfs_root *quota_root;
340f1aa2 1055 struct btrfs_trans_handle *trans = NULL;
bed92eae
AJ		 1056 int ret = 0;
1057
f2f6ed3d 1058 mutex_lock(&fs_info->qgroup_ioctl_lock);
58400fce 1059 if (!fs_info->quota_root)
f2f6ed3d 1060 goto out;
340f1aa2
NB		 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
afcdd129 1074 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
d06f23d6 1075 btrfs_qgroup_wait_for_completion(fs_info, false);
967ef513 1076 spin_lock(&fs_info->qgroup_lock);
bed92eae
AJ		 1077 quota_root = fs_info->quota_root;
1078 fs_info->quota_root = NULL;
8ea0ec9e 1079 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
bed92eae
AJ		 1080 spin_unlock(&fs_info->qgroup_lock);
1081
e685da14
WS		 1082 btrfs_free_qgroup_config(fs_info);
1083
bed92eae 1084 ret = btrfs_clean_quota_tree(trans, quota_root);
340f1aa2
NB		 1085 if (ret) {
1086 btrfs_abort_transaction(trans, ret);
1087 goto end_trans;
1088 }
bed92eae 1089
ab9ce7d4 1090 ret = btrfs_del_root(trans, &quota_root->root_key);
340f1aa2
NB		 1091 if (ret) {
1092 btrfs_abort_transaction(trans, ret);
1093 goto end_trans;
1094 }
bed92eae
AJ		 1095
1096 list_del(&quota_root->dirty_list);
1097
1098 btrfs_tree_lock(quota_root->node);
7c302b49 1099 clean_tree_block(fs_info, quota_root->node);
bed92eae
AJ		 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);
340f1aa2
NB		 1106
1107end_trans:
1108 ret = btrfs_end_transaction(trans);
bed92eae 1109out:
f2f6ed3d 1110 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1111 return ret;
1112}
1113
2f232036
JS		 1114static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1115 struct btrfs_qgroup *qgroup)
bed92eae 1116{
2f232036
JS		 1117 if (list_empty(&qgroup->dirty))
1118 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
bed92eae
AJ		 1119}
1120
9c8b35b1 1121/*
429d6275
QW		 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).
9c8b35b1
QW		 1132 *
1133 * Caller should hold fs_info->qgroup_lock.
1134 */
1135static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1136 struct ulist *tmp, u64 ref_root,
429d6275 1137 struct btrfs_qgroup *src, int sign)
9c8b35b1
QW		 1138{
1139 struct btrfs_qgroup *qgroup;
1140 struct btrfs_qgroup_list *glist;
1141 struct ulist_node *unode;
1142 struct ulist_iterator uiter;
429d6275 1143 u64 num_bytes = src->excl;
9c8b35b1
QW		 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;
429d6275
QW		 1156
1157 if (sign > 0)
64ee4e75 1158 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
429d6275 1159 else
64ee4e75 1160 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
9c8b35b1
QW		 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,
ef2fff64 1167 qgroup_to_aux(glist->group), GFP_ATOMIC);
9c8b35b1
QW		 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))) {
ef2fff64 1175 qgroup = unode_aux_to_qgroup(unode);
9c8b35b1
QW		 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;
429d6275 1180 if (sign > 0)
64ee4e75 1181 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
429d6275 1182 else
64ee4e75 1183 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
9c8b35b1
QW		 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,
ef2fff64 1190 qgroup_to_aux(glist->group), GFP_ATOMIC);
9c8b35b1
QW		 1191 if (ret < 0)
1192 goto out;
1193 }
1194 }
1195 ret = 0;
1196out:
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 */
1212static 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,
429d6275 1226 qgroup, sign);
9c8b35b1
QW		 1227 if (err < 0) {
1228 ret = err;
1229 goto out;
1230 }
1231 }
1232out:
1233 if (ret)
1234 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1235 return ret;
1236}
1237
9f8a6ce6
LF		 1238int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1239 u64 dst)
bed92eae 1240{
9f8a6ce6 1241 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae 1242 struct btrfs_root *quota_root;
b7fef4f5
WS		 1243 struct btrfs_qgroup *parent;
1244 struct btrfs_qgroup *member;
534e6623 1245 struct btrfs_qgroup_list *list;
9c8b35b1 1246 struct ulist *tmp;
bed92eae
AJ		 1247 int ret = 0;
1248
8465ecec
QW		 1249 /* Check the level of src and dst first */
1250 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1251 return -EINVAL;
1252
6602caf1 1253 tmp = ulist_alloc(GFP_KERNEL);
ab3680dd
CE		 1254 if (!tmp)
1255 return -ENOMEM;
1256
f2f6ed3d 1257 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1258 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1259 if (!quota_root) {
1260 ret = -EINVAL;
1261 goto out;
1262 }
b7fef4f5
WS		 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 }
bed92eae 1269
534e6623
WS		 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
711169c4 1278 ret = add_qgroup_relation_item(trans, src, dst);
bed92eae 1279 if (ret)
f2f6ed3d 1280 goto out;
bed92eae 1281
711169c4 1282 ret = add_qgroup_relation_item(trans, dst, src);
bed92eae 1283 if (ret) {
99d7f09a 1284 del_qgroup_relation_item(trans, src, dst);
f2f6ed3d 1285 goto out;
bed92eae
AJ		 1286 }
1287
1288 spin_lock(&fs_info->qgroup_lock);
0b246afa 1289 ret = add_relation_rb(fs_info, src, dst);
9c8b35b1
QW		 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);
bed92eae 1295 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1296out:
1297 mutex_unlock(&fs_info->qgroup_ioctl_lock);
9c8b35b1 1298 ulist_free(tmp);
bed92eae
AJ		 1299 return ret;
1300}
1301
6b36f1aa
LF		 1302static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1303 u64 dst)
bed92eae 1304{
6b36f1aa 1305 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae 1306 struct btrfs_root *quota_root;
534e6623
WS		 1307 struct btrfs_qgroup *parent;
1308 struct btrfs_qgroup *member;
1309 struct btrfs_qgroup_list *list;
9c8b35b1 1310 struct ulist *tmp;
bed92eae
AJ		 1311 int ret = 0;
1312 int err;
1313
6602caf1 1314 tmp = ulist_alloc(GFP_KERNEL);
9c8b35b1
QW		 1315 if (!tmp)
1316 return -ENOMEM;
1317
bed92eae 1318 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1319 if (!quota_root) {
1320 ret = -EINVAL;
1321 goto out;
1322 }
bed92eae 1323
534e6623
WS		 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;
1338exist:
99d7f09a
LF		 1339 ret = del_qgroup_relation_item(trans, src, dst);
1340 err = del_qgroup_relation_item(trans, dst, src);
bed92eae
AJ		 1341 if (err && !ret)
1342 ret = err;
1343
1344 spin_lock(&fs_info->qgroup_lock);
1345 del_relation_rb(fs_info, src, dst);
9c8b35b1 1346 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
bed92eae 1347 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d 1348out:
9c8b35b1 1349 ulist_free(tmp);
f5a6b1c5
DY		 1350 return ret;
1351}
1352
39616c27
LF		 1353int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1354 u64 dst)
f5a6b1c5 1355{
39616c27 1356 struct btrfs_fs_info *fs_info = trans->fs_info;
f5a6b1c5
DY		 1357 int ret = 0;
1358
1359 mutex_lock(&fs_info->qgroup_ioctl_lock);
6b36f1aa 1360 ret = __del_qgroup_relation(trans, src, dst);
f2f6ed3d 1361 mutex_unlock(&fs_info->qgroup_ioctl_lock);
f5a6b1c5 1362
bed92eae
AJ		 1363 return ret;
1364}
1365
49a05ecd 1366int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
bed92eae 1367{
49a05ecd 1368 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae
AJ		 1369 struct btrfs_root *quota_root;
1370 struct btrfs_qgroup *qgroup;
1371 int ret = 0;
1372
f2f6ed3d 1373 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1374 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1375 if (!quota_root) {
1376 ret = -EINVAL;
1377 goto out;
1378 }
534e6623
WS		 1379 qgroup = find_qgroup_rb(fs_info, qgroupid);
1380 if (qgroup) {
1381 ret = -EEXIST;
1382 goto out;
1383 }
bed92eae
AJ		 1384
1385 ret = add_qgroup_item(trans, quota_root, qgroupid);
534e6623
WS		 1386 if (ret)
1387 goto out;
bed92eae
AJ		 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);
f2f6ed3d
WS		 1395out:
1396 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1397 return ret;
1398}
1399
3efbee1d 1400int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
bed92eae 1401{
3efbee1d 1402 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae 1403 struct btrfs_root *quota_root;
2cf68703 1404 struct btrfs_qgroup *qgroup;
f5a6b1c5 1405 struct btrfs_qgroup_list *list;
bed92eae
AJ		 1406 int ret = 0;
1407
f2f6ed3d 1408 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1409 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1410 if (!quota_root) {
1411 ret = -EINVAL;
1412 goto out;
1413 }
bed92eae 1414
2cf68703 1415 qgroup = find_qgroup_rb(fs_info, qgroupid);
534e6623
WS		 1416 if (!qgroup) {
1417 ret = -ENOENT;
1418 goto out;
1419 } else {
f5a6b1c5
DY		 1420 /* check if there are no children of this qgroup */
1421 if (!list_empty(&qgroup->members)) {
f2f6ed3d
WS		 1422 ret = -EBUSY;
1423 goto out;
2cf68703
AJ		 1424 }
1425 }
69104618 1426 ret = del_qgroup_item(trans, qgroupid);
36b96fdc
SD		 1427 if (ret && ret != -ENOENT)
1428 goto out;
bed92eae 1429
f5a6b1c5
DY		 1430 while (!list_empty(&qgroup->groups)) {
1431 list = list_first_entry(&qgroup->groups,
1432 struct btrfs_qgroup_list, next_group);
6b36f1aa
LF		 1433 ret = __del_qgroup_relation(trans, qgroupid,
1434 list->group->qgroupid);
f5a6b1c5
DY		 1435 if (ret)
1436 goto out;
1437 }
1438
bed92eae 1439 spin_lock(&fs_info->qgroup_lock);
0b246afa 1440 del_qgroup_rb(fs_info, qgroupid);
bed92eae 1441 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1442out:
1443 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1444 return ret;
1445}
1446
f0042d5e 1447int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
bed92eae
AJ		 1448 struct btrfs_qgroup_limit *limit)
1449{
f0042d5e 1450 struct btrfs_fs_info *fs_info = trans->fs_info;
f2f6ed3d 1451 struct btrfs_root *quota_root;
bed92eae
AJ		 1452 struct btrfs_qgroup *qgroup;
1453 int ret = 0;
fe759907
YD		 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;
bed92eae 1459
f2f6ed3d
WS		 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 }
bed92eae 1466
ddb47afa
WS		 1467 qgroup = find_qgroup_rb(fs_info, qgroupid);
1468 if (!qgroup) {
1469 ret = -ENOENT;
1470 goto out;
1471 }
bed92eae 1472
58400fce 1473 spin_lock(&fs_info->qgroup_lock);
fe759907
YD		 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 }
03477d94
DY		 1510 qgroup->lim_flags |= limit->flags;
1511
bed92eae 1512 spin_unlock(&fs_info->qgroup_lock);
1510e71c 1513
ac8a866a 1514 ret = update_qgroup_limit_item(trans, qgroup);
1510e71c
DY		 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
f2f6ed3d
WS		 1521out:
1522 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1523 return ret;
1524}
1152651a 1525
50b3e040 1526int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
cb93b52c
QW		 1527 struct btrfs_delayed_ref_root *delayed_refs,
1528 struct btrfs_qgroup_extent_record *record)
3368d001
QW		 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
a4666e68 1535 lockdep_assert_held(&delayed_refs->lock);
50b3e040 1536 trace_btrfs_qgroup_trace_extent(fs_info, record);
82bd101b 1537
3368d001
QW		 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
cb93b52c 1547 return 1;
3368d001
QW		 1548 }
1549
1550 rb_link_node(&record->node, parent_node, p);
1551 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
cb93b52c
QW		 1552 return 0;
1553}
1554
fb235dc0
QW		 1555int 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
c995ab3c 1562 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
952bd3db
NB		 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 }
fb235dc0
QW		 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
a95f3aaf
LF		 1582int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1583 u64 num_bytes, gfp_t gfp_flag)
cb93b52c 1584{
a95f3aaf 1585 struct btrfs_fs_info *fs_info = trans->fs_info;
cb93b52c
QW		 1586 struct btrfs_qgroup_extent_record *record;
1587 struct btrfs_delayed_ref_root *delayed_refs;
1588 int ret;
1589
afcdd129
JB		 1590 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1591 || bytenr == 0 || num_bytes == 0)
cb93b52c 1592 return 0;
cb93b52c
QW		 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);
2ff7e61e 1603 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
cb93b52c 1604 spin_unlock(&delayed_refs->lock);
fb235dc0 1605 if (ret > 0) {
cb93b52c 1606 kfree(record);
fb235dc0
QW		 1607 return 0;
1608 }
1609 return btrfs_qgroup_trace_extent_post(fs_info, record);
3368d001
QW		 1610}
1611
33d1f05c 1612int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
33d1f05c
QW		 1613 struct extent_buffer *eb)
1614{
8d38d7eb 1615 struct btrfs_fs_info *fs_info = trans->fs_info;
33d1f05c
QW		 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() */
0b246afa 1623 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
33d1f05c
QW		 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
a95f3aaf
LF		 1645 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1646 GFP_NOFS);
33d1f05c
QW		 1647 if (ret)
1648 return ret;
1649 }
cddf3b2c 1650 cond_resched();
33d1f05c
QW		 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 */
15b34517 1668static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
33d1f05c
QW		 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
25982561
QW		 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 */
1761static 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{
1766 struct btrfs_key key;
1767 struct btrfs_path *src_path;
1768 struct btrfs_fs_info *fs_info = trans->fs_info;
1769 u32 nodesize = fs_info->nodesize;
1770 int cur_level = root_level;
1771 int ret;
1772
1773 BUG_ON(dst_level > root_level);
1774 /* Level mismatch */
1775 if (btrfs_header_level(src_eb) != root_level)
1776 return -EINVAL;
1777
1778 src_path = btrfs_alloc_path();
1779 if (!src_path) {
1780 ret = -ENOMEM;
1781 goto out;
1782 }
1783
1784 if (dst_level)
1785 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1786 else
1787 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1788
1789 /* For src_path */
1790 extent_buffer_get(src_eb);
1791 src_path->nodes[root_level] = src_eb;
1792 src_path->slots[root_level] = dst_path->slots[root_level];
1793 src_path->locks[root_level] = 0;
1794
1795 /* A simplified version of btrfs_search_slot() */
1796 while (cur_level >= dst_level) {
1797 struct btrfs_key src_key;
1798 struct btrfs_key dst_key;
1799
1800 if (src_path->nodes[cur_level] == NULL) {
1801 struct btrfs_key first_key;
1802 struct extent_buffer *eb;
1803 int parent_slot;
1804 u64 child_gen;
1805 u64 child_bytenr;
1806
1807 eb = src_path->nodes[cur_level + 1];
1808 parent_slot = src_path->slots[cur_level + 1];
1809 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1810 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1811 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1812
1813 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1814 cur_level, &first_key);
1815 if (IS_ERR(eb)) {
1816 ret = PTR_ERR(eb);
1817 goto out;
1818 } else if (!extent_buffer_uptodate(eb)) {
1819 free_extent_buffer(eb);
1820 ret = -EIO;
1821 goto out;
1822 }
1823
1824 src_path->nodes[cur_level] = eb;
1825
1826 btrfs_tree_read_lock(eb);
1827 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1828 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1829 }
1830
1831 src_path->slots[cur_level] = dst_path->slots[cur_level];
1832 if (cur_level) {
1833 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1834 &dst_key, dst_path->slots[cur_level]);
1835 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1836 &src_key, src_path->slots[cur_level]);
1837 } else {
1838 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1839 &dst_key, dst_path->slots[cur_level]);
1840 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1841 &src_key, src_path->slots[cur_level]);
1842 }
1843 /* Content mismatch, something went wrong */
1844 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1845 ret = -ENOENT;
1846 goto out;
1847 }
1848 cur_level--;
1849 }
1850
1851 /*
1852 * Now both @dst_path and @src_path have been populated, record the tree
1853 * blocks for qgroup accounting.
1854 */
1855 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1856 nodesize, GFP_NOFS);
1857 if (ret < 0)
1858 goto out;
1859 ret = btrfs_qgroup_trace_extent(trans,
1860 dst_path->nodes[dst_level]->start,
1861 nodesize, GFP_NOFS);
1862 if (ret < 0)
1863 goto out;
1864
1865 /* Record leaf file extents */
1866 if (dst_level == 0) {
1867 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1868 if (ret < 0)
1869 goto out;
1870 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1871 }
1872out:
1873 btrfs_free_path(src_path);
1874 return ret;
1875}
1876
33d1f05c 1877int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
33d1f05c
QW		 1878 struct extent_buffer *root_eb,
1879 u64 root_gen, int root_level)
1880{
deb40627 1881 struct btrfs_fs_info *fs_info = trans->fs_info;
33d1f05c
QW		 1882 int ret = 0;
1883 int level;
1884 struct extent_buffer *eb = root_eb;
1885 struct btrfs_path *path = NULL;
1886
b6e6bca5 1887 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
33d1f05c
QW		 1888 BUG_ON(root_eb == NULL);
1889
0b246afa 1890 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
33d1f05c
QW		 1891 return 0;
1892
1893 if (!extent_buffer_uptodate(root_eb)) {
581c1760 1894 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
33d1f05c
QW		 1895 if (ret)
1896 goto out;
1897 }
1898
1899 if (root_level == 0) {
8d38d7eb 1900 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
33d1f05c
QW		 1901 goto out;
1902 }
1903
1904 path = btrfs_alloc_path();
1905 if (!path)
1906 return -ENOMEM;
1907
1908 /*
1909 * Walk down the tree. Missing extent blocks are filled in as
1910 * we go. Metadata is accounted every time we read a new
1911 * extent block.
1912 *
1913 * When we reach a leaf, we account for file extent items in it,
1914 * walk back up the tree (adjusting slot pointers as we go)
1915 * and restart the search process.
1916 */
1917 extent_buffer_get(root_eb); /* For path */
1918 path->nodes[root_level] = root_eb;
1919 path->slots[root_level] = 0;
1920 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1921walk_down:
1922 level = root_level;
1923 while (level >= 0) {
1924 if (path->nodes[level] == NULL) {
581c1760 1925 struct btrfs_key first_key;
33d1f05c
QW		 1926 int parent_slot;
1927 u64 child_gen;
1928 u64 child_bytenr;
1929
1930 /*
1931 * We need to get child blockptr/gen from parent before
1932 * we can read it.
1933 */
1934 eb = path->nodes[level + 1];
1935 parent_slot = path->slots[level + 1];
1936 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1937 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
581c1760 1938 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
33d1f05c 1939
581c1760
QW		 1940 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1941 level, &first_key);
33d1f05c
QW		 1942 if (IS_ERR(eb)) {
1943 ret = PTR_ERR(eb);
1944 goto out;
1945 } else if (!extent_buffer_uptodate(eb)) {
1946 free_extent_buffer(eb);
1947 ret = -EIO;
1948 goto out;
1949 }
1950
1951 path->nodes[level] = eb;
1952 path->slots[level] = 0;
1953
1954 btrfs_tree_read_lock(eb);
1955 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1956 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1957
a95f3aaf 1958 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
0b246afa
JM		 1959 fs_info->nodesize,
1960 GFP_NOFS);
33d1f05c
QW		 1961 if (ret)
1962 goto out;
1963 }
1964
1965 if (level == 0) {
8d38d7eb
LF		 1966 ret = btrfs_qgroup_trace_leaf_items(trans,
1967 path->nodes[level]);
33d1f05c
QW		 1968 if (ret)
1969 goto out;
1970
1971 /* Nonzero return here means we completed our search */
15b34517 1972 ret = adjust_slots_upwards(path, root_level);
33d1f05c
QW		 1973 if (ret)
1974 break;
1975
1976 /* Restart search with new slots */
1977 goto walk_down;
1978 }
1979
1980 level--;
1981 }
1982
1983 ret = 0;
1984out:
1985 btrfs_free_path(path);
1986
1987 return ret;
1988}
1989
d810ef2b
QW		 1990#define UPDATE_NEW 0
1991#define UPDATE_OLD 1
1992/*
1993 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1994 */
1995static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1996 struct ulist *roots, struct ulist *tmp,
1997 struct ulist *qgroups, u64 seq, int update_old)
1998{
1999 struct ulist_node *unode;
2000 struct ulist_iterator uiter;
2001 struct ulist_node *tmp_unode;
2002 struct ulist_iterator tmp_uiter;
2003 struct btrfs_qgroup *qg;
2004 int ret = 0;
2005
2006 if (!roots)
2007 return 0;
2008 ULIST_ITER_INIT(&uiter);
2009 while ((unode = ulist_next(roots, &uiter))) {
2010 qg = find_qgroup_rb(fs_info, unode->val);
2011 if (!qg)
2012 continue;
2013
2014 ulist_reinit(tmp);
ef2fff64 2015 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
d810ef2b
QW		 2016 GFP_ATOMIC);
2017 if (ret < 0)
2018 return ret;
ef2fff64 2019 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
d810ef2b
QW		 2020 if (ret < 0)
2021 return ret;
2022 ULIST_ITER_INIT(&tmp_uiter);
2023 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2024 struct btrfs_qgroup_list *glist;
2025
ef2fff64 2026 qg = unode_aux_to_qgroup(tmp_unode);
d810ef2b
QW		 2027 if (update_old)
2028 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2029 else
2030 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2031 list_for_each_entry(glist, &qg->groups, next_group) {
2032 ret = ulist_add(qgroups, glist->group->qgroupid,
ef2fff64 2033 qgroup_to_aux(glist->group),
d810ef2b
QW		 2034 GFP_ATOMIC);
2035 if (ret < 0)
2036 return ret;
2037 ret = ulist_add(tmp, glist->group->qgroupid,
ef2fff64 2038 qgroup_to_aux(glist->group),
d810ef2b
QW		 2039 GFP_ATOMIC);
2040 if (ret < 0)
2041 return ret;
2042 }
2043 }
2044 }
2045 return 0;
2046}
2047
823ae5b8
QW		 2048/*
2049 * Update qgroup rfer/excl counters.
2050 * Rfer update is easy, codes can explain themselves.
e69bcee3 2051 *
823ae5b8
QW		 2052 * Excl update is tricky, the update is split into 2 part.
2053 * Part 1: Possible exclusive <-> sharing detect:
2054 * | A | !A |
2055 * -------------------------------------
2056 * B | * | - |
2057 * -------------------------------------
2058 * !B | + | ** |
2059 * -------------------------------------
2060 *
2061 * Conditions:
2062 * A: cur_old_roots < nr_old_roots (not exclusive before)
2063 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2064 * B: cur_new_roots < nr_new_roots (not exclusive now)
01327610 2065 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
823ae5b8
QW		 2066 *
2067 * Results:
2068 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2069 * *: Definitely not changed. **: Possible unchanged.
2070 *
2071 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2072 *
2073 * To make the logic clear, we first use condition A and B to split
2074 * combination into 4 results.
2075 *
2076 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2077 * only on variant maybe 0.
2078 *
2079 * Lastly, check result **, since there are 2 variants maybe 0, split them
2080 * again(2x2).
2081 * But this time we don't need to consider other things, the codes and logic
2082 * is easy to understand now.
2083 */
2084static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2085 struct ulist *qgroups,
2086 u64 nr_old_roots,
2087 u64 nr_new_roots,
2088 u64 num_bytes, u64 seq)
2089{
2090 struct ulist_node *unode;
2091 struct ulist_iterator uiter;
2092 struct btrfs_qgroup *qg;
2093 u64 cur_new_count, cur_old_count;
2094
2095 ULIST_ITER_INIT(&uiter);
2096 while ((unode = ulist_next(qgroups, &uiter))) {
2097 bool dirty = false;
2098
ef2fff64 2099 qg = unode_aux_to_qgroup(unode);
823ae5b8
QW		 2100 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2101 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2102
8b317901
QW		 2103 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2104 cur_new_count);
0f5dcf8d 2105
823ae5b8
QW		 2106 /* Rfer update part */
2107 if (cur_old_count == 0 && cur_new_count > 0) {
2108 qg->rfer += num_bytes;
2109 qg->rfer_cmpr += num_bytes;
2110 dirty = true;
2111 }
2112 if (cur_old_count > 0 && cur_new_count == 0) {
2113 qg->rfer -= num_bytes;
2114 qg->rfer_cmpr -= num_bytes;
2115 dirty = true;
2116 }
2117
2118 /* Excl update part */
2119 /* Exclusive/none -> shared case */
2120 if (cur_old_count == nr_old_roots &&
2121 cur_new_count < nr_new_roots) {
2122 /* Exclusive -> shared */
2123 if (cur_old_count != 0) {
2124 qg->excl -= num_bytes;
2125 qg->excl_cmpr -= num_bytes;
2126 dirty = true;
2127 }
2128 }
2129
2130 /* Shared -> exclusive/none case */
2131 if (cur_old_count < nr_old_roots &&
2132 cur_new_count == nr_new_roots) {
2133 /* Shared->exclusive */
2134 if (cur_new_count != 0) {
2135 qg->excl += num_bytes;
2136 qg->excl_cmpr += num_bytes;
2137 dirty = true;
2138 }
2139 }
2140
2141 /* Exclusive/none -> exclusive/none case */
2142 if (cur_old_count == nr_old_roots &&
2143 cur_new_count == nr_new_roots) {
2144 if (cur_old_count == 0) {
2145 /* None -> exclusive/none */
2146
2147 if (cur_new_count != 0) {
2148 /* None -> exclusive */
2149 qg->excl += num_bytes;
2150 qg->excl_cmpr += num_bytes;
2151 dirty = true;
2152 }
2153 /* None -> none, nothing changed */
2154 } else {
2155 /* Exclusive -> exclusive/none */
2156
2157 if (cur_new_count == 0) {
2158 /* Exclusive -> none */
2159 qg->excl -= num_bytes;
2160 qg->excl_cmpr -= num_bytes;
2161 dirty = true;
2162 }
2163 /* Exclusive -> exclusive, nothing changed */
2164 }
2165 }
c05f9429 2166
823ae5b8
QW		 2167 if (dirty)
2168 qgroup_dirty(fs_info, qg);
2169 }
2170 return 0;
2171}
2172
5edfd9fd
QW		 2173/*
2174 * Check if the @roots potentially is a list of fs tree roots
2175 *
2176 * Return 0 for definitely not a fs/subvol tree roots ulist
2177 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2178 * one as well)
2179 */
2180static int maybe_fs_roots(struct ulist *roots)
2181{
2182 struct ulist_node *unode;
2183 struct ulist_iterator uiter;
2184
2185 /* Empty one, still possible for fs roots */
2186 if (!roots || roots->nnodes == 0)
2187 return 1;
2188
2189 ULIST_ITER_INIT(&uiter);
2190 unode = ulist_next(roots, &uiter);
2191 if (!unode)
2192 return 1;
2193
2194 /*
2195 * If it contains fs tree roots, then it must belong to fs/subvol
2196 * trees.
2197 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2198 */
2199 return is_fstree(unode->val);
2200}
2201
8696d760
LF		 2202int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2203 u64 num_bytes, struct ulist *old_roots,
2204 struct ulist *new_roots)
550d7a2e 2205{
8696d760 2206 struct btrfs_fs_info *fs_info = trans->fs_info;
550d7a2e
QW		 2207 struct ulist *qgroups = NULL;
2208 struct ulist *tmp = NULL;
2209 u64 seq;
2210 u64 nr_new_roots = 0;
2211 u64 nr_old_roots = 0;
2212 int ret = 0;
2213
81353d50
DS		 2214 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2215 return 0;
2216
5edfd9fd
QW		 2217 if (new_roots) {
2218 if (!maybe_fs_roots(new_roots))
2219 goto out_free;
550d7a2e 2220 nr_new_roots = new_roots->nnodes;
5edfd9fd
QW		 2221 }
2222 if (old_roots) {
2223 if (!maybe_fs_roots(old_roots))
2224 goto out_free;
550d7a2e 2225 nr_old_roots = old_roots->nnodes;
5edfd9fd
QW		 2226 }
2227
2228 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2229 if (nr_old_roots == 0 && nr_new_roots == 0)
2230 goto out_free;
550d7a2e 2231
550d7a2e
QW		 2232 BUG_ON(!fs_info->quota_root);
2233
c9f6f3cd
QW		 2234 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2235 num_bytes, nr_old_roots, nr_new_roots);
0f5dcf8d 2236
550d7a2e
QW		 2237 qgroups = ulist_alloc(GFP_NOFS);
2238 if (!qgroups) {
2239 ret = -ENOMEM;
2240 goto out_free;
2241 }
2242 tmp = ulist_alloc(GFP_NOFS);
2243 if (!tmp) {
2244 ret = -ENOMEM;
2245 goto out_free;
2246 }
2247
2248 mutex_lock(&fs_info->qgroup_rescan_lock);
2249 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2250 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2251 mutex_unlock(&fs_info->qgroup_rescan_lock);
2252 ret = 0;
2253 goto out_free;
2254 }
2255 }
2256 mutex_unlock(&fs_info->qgroup_rescan_lock);
2257
2258 spin_lock(&fs_info->qgroup_lock);
2259 seq = fs_info->qgroup_seq;
2260
2261 /* Update old refcnts using old_roots */
2262 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2263 UPDATE_OLD);
2264 if (ret < 0)
2265 goto out;
2266
2267 /* Update new refcnts using new_roots */
2268 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2269 UPDATE_NEW);
2270 if (ret < 0)
2271 goto out;
2272
2273 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2274 num_bytes, seq);
2275
2276 /*
2277 * Bump qgroup_seq to avoid seq overlap
2278 */
2279 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2280out:
2281 spin_unlock(&fs_info->qgroup_lock);
2282out_free:
2283 ulist_free(tmp);
2284 ulist_free(qgroups);
2285 ulist_free(old_roots);
2286 ulist_free(new_roots);
2287 return ret;
2288}
2289
460fb20a 2290int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
550d7a2e 2291{
460fb20a 2292 struct btrfs_fs_info *fs_info = trans->fs_info;
550d7a2e
QW		 2293 struct btrfs_qgroup_extent_record *record;
2294 struct btrfs_delayed_ref_root *delayed_refs;
2295 struct ulist *new_roots = NULL;
2296 struct rb_node *node;
c337e7b0 2297 u64 num_dirty_extents = 0;
9086db86 2298 u64 qgroup_to_skip;
550d7a2e
QW		 2299 int ret = 0;
2300
2301 delayed_refs = &trans->transaction->delayed_refs;
9086db86 2302 qgroup_to_skip = delayed_refs->qgroup_to_skip;
550d7a2e
QW		 2303 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2304 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2305 node);
2306
c337e7b0 2307 num_dirty_extents++;
bc074524 2308 trace_btrfs_qgroup_account_extents(fs_info, record);
0f5dcf8d 2309
550d7a2e 2310 if (!ret) {
d1b8b94a
QW		 2311 /*
2312 * Old roots should be searched when inserting qgroup
2313 * extent record
2314 */
2315 if (WARN_ON(!record->old_roots)) {
2316 /* Search commit root to find old_roots */
2317 ret = btrfs_find_all_roots(NULL, fs_info,
2318 record->bytenr, 0,
c995ab3c 2319 &record->old_roots, false);
d1b8b94a
QW		 2320 if (ret < 0)
2321 goto cleanup;
2322 }
2323
550d7a2e 2324 /*
de47c9d3 2325 * Use SEQ_LAST as time_seq to do special search, which
550d7a2e
QW		 2326 * doesn't lock tree or delayed_refs and search current
2327 * root. It's safe inside commit_transaction().
2328 */
2329 ret = btrfs_find_all_roots(trans, fs_info,
c995ab3c 2330 record->bytenr, SEQ_LAST, &new_roots, false);
550d7a2e
QW		 2331 if (ret < 0)
2332 goto cleanup;
d1b8b94a 2333 if (qgroup_to_skip) {
9086db86 2334 ulist_del(new_roots, qgroup_to_skip, 0);
d1b8b94a
QW		 2335 ulist_del(record->old_roots, qgroup_to_skip,
2336 0);
2337 }
8696d760
LF		 2338 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2339 record->num_bytes,
2340 record->old_roots,
2341 new_roots);
550d7a2e
QW		 2342 record->old_roots = NULL;
2343 new_roots = NULL;
2344 }
2345cleanup:
2346 ulist_free(record->old_roots);
2347 ulist_free(new_roots);
2348 new_roots = NULL;
2349 rb_erase(node, &delayed_refs->dirty_extent_root);
2350 kfree(record);
2351
2352 }
c337e7b0
QW		 2353 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2354 num_dirty_extents);
550d7a2e
QW		 2355 return ret;
2356}
2357
bed92eae
AJ		 2358/*
2359 * called from commit_transaction. Writes all changed qgroups to disk.
2360 */
280f8bd2 2361int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
bed92eae 2362{
280f8bd2 2363 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae
AJ		 2364 struct btrfs_root *quota_root = fs_info->quota_root;
2365 int ret = 0;
2366
2367 if (!quota_root)
5d23515b 2368 return ret;
bed92eae
AJ		 2369
2370 spin_lock(&fs_info->qgroup_lock);
2371 while (!list_empty(&fs_info->dirty_qgroups)) {
2372 struct btrfs_qgroup *qgroup;
2373 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2374 struct btrfs_qgroup, dirty);
2375 list_del_init(&qgroup->dirty);
2376 spin_unlock(&fs_info->qgroup_lock);
3e07e9a0 2377 ret = update_qgroup_info_item(trans, qgroup);
d3001ed3
DY		 2378 if (ret)
2379 fs_info->qgroup_flags |=
2380 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
ac8a866a 2381 ret = update_qgroup_limit_item(trans, qgroup);
bed92eae
AJ		 2382 if (ret)
2383 fs_info->qgroup_flags |=
2384 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2385 spin_lock(&fs_info->qgroup_lock);
2386 }
afcdd129 2387 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
bed92eae
AJ		 2388 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2389 else
2390 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2391 spin_unlock(&fs_info->qgroup_lock);
2392
2e980acd 2393 ret = update_qgroup_status_item(trans);
bed92eae
AJ		 2394 if (ret)
2395 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2396
bed92eae
AJ		 2397 return ret;
2398}
2399
2400/*
01327610 2401 * Copy the accounting information between qgroups. This is necessary
918c2ee1
MF		 2402 * when a snapshot or a subvolume is created. Throwing an error will
2403 * cause a transaction abort so we take extra care here to only error
2404 * when a readonly fs is a reasonable outcome.
bed92eae 2405 */
a9377422
LF		 2406int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2407 u64 objectid, struct btrfs_qgroup_inherit *inherit)
bed92eae
AJ		 2408{
2409 int ret = 0;
2410 int i;
2411 u64 *i_qgroups;
a9377422 2412 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae
AJ		 2413 struct btrfs_root *quota_root = fs_info->quota_root;
2414 struct btrfs_qgroup *srcgroup;
2415 struct btrfs_qgroup *dstgroup;
2416 u32 level_size = 0;
3f5e2d3b 2417 u64 nums;
bed92eae 2418
f2f6ed3d 2419 mutex_lock(&fs_info->qgroup_ioctl_lock);
afcdd129 2420 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
f2f6ed3d 2421 goto out;
bed92eae 2422
f2f6ed3d
WS		 2423 if (!quota_root) {
2424 ret = -EINVAL;
2425 goto out;
2426 }
bed92eae 2427
3f5e2d3b
WS		 2428 if (inherit) {
2429 i_qgroups = (u64 *)(inherit + 1);
2430 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2431 2 * inherit->num_excl_copies;
2432 for (i = 0; i < nums; ++i) {
2433 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
09870d27 2434
918c2ee1
MF		 2435 /*
2436 * Zero out invalid groups so we can ignore
2437 * them later.
2438 */
2439 if (!srcgroup ||
2440 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2441 *i_qgroups = 0ULL;
2442
3f5e2d3b
WS		 2443 ++i_qgroups;
2444 }
2445 }
2446
bed92eae
AJ		 2447 /*
2448 * create a tracking group for the subvol itself
2449 */
2450 ret = add_qgroup_item(trans, quota_root, objectid);
2451 if (ret)
2452 goto out;
2453
bed92eae
AJ		 2454 /*
2455 * add qgroup to all inherited groups
2456 */
2457 if (inherit) {
2458 i_qgroups = (u64 *)(inherit + 1);
918c2ee1
MF		 2459 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2460 if (*i_qgroups == 0)
2461 continue;
711169c4
LF		 2462 ret = add_qgroup_relation_item(trans, objectid,
2463 *i_qgroups);
918c2ee1 2464 if (ret && ret != -EEXIST)
bed92eae 2465 goto out;
711169c4
LF		 2466 ret = add_qgroup_relation_item(trans, *i_qgroups,
2467 objectid);
918c2ee1 2468 if (ret && ret != -EEXIST)
bed92eae 2469 goto out;
bed92eae 2470 }
918c2ee1 2471 ret = 0;
bed92eae
AJ		 2472 }
2473
2474
2475 spin_lock(&fs_info->qgroup_lock);
2476
2477 dstgroup = add_qgroup_rb(fs_info, objectid);
57a5a882
DC		 2478 if (IS_ERR(dstgroup)) {
2479 ret = PTR_ERR(dstgroup);
bed92eae 2480 goto unlock;
57a5a882 2481 }
bed92eae 2482
e8c8541a 2483 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
e8c8541a
DY		 2484 dstgroup->lim_flags = inherit->lim.flags;
2485 dstgroup->max_rfer = inherit->lim.max_rfer;
2486 dstgroup->max_excl = inherit->lim.max_excl;
2487 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2488 dstgroup->rsv_excl = inherit->lim.rsv_excl;
1510e71c 2489
ac8a866a 2490 ret = update_qgroup_limit_item(trans, dstgroup);
1510e71c
DY		 2491 if (ret) {
2492 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
5d163e0e
JM		 2493 btrfs_info(fs_info,
2494 "unable to update quota limit for %llu",
2495 dstgroup->qgroupid);
1510e71c
DY		 2496 goto unlock;
2497 }
e8c8541a
DY		 2498 }
2499
bed92eae
AJ		 2500 if (srcid) {
2501 srcgroup = find_qgroup_rb(fs_info, srcid);
f3a87f1b 2502 if (!srcgroup)
bed92eae 2503 goto unlock;
fcebe456
JB		 2504
2505 /*
2506 * We call inherit after we clone the root in order to make sure
2507 * our counts don't go crazy, so at this point the only
2508 * difference between the two roots should be the root node.
2509 */
c8389d4c 2510 level_size = fs_info->nodesize;
fcebe456
JB		 2511 dstgroup->rfer = srcgroup->rfer;
2512 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2513 dstgroup->excl = level_size;
2514 dstgroup->excl_cmpr = level_size;
bed92eae
AJ		 2515 srcgroup->excl = level_size;
2516 srcgroup->excl_cmpr = level_size;
3eeb4d59
DY		 2517
2518 /* inherit the limit info */
2519 dstgroup->lim_flags = srcgroup->lim_flags;
2520 dstgroup->max_rfer = srcgroup->max_rfer;
2521 dstgroup->max_excl = srcgroup->max_excl;
2522 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2523 dstgroup->rsv_excl = srcgroup->rsv_excl;
2524
bed92eae
AJ		 2525 qgroup_dirty(fs_info, dstgroup);
2526 qgroup_dirty(fs_info, srcgroup);
2527 }
2528
f3a87f1b 2529 if (!inherit)
bed92eae
AJ		 2530 goto unlock;
2531
2532 i_qgroups = (u64 *)(inherit + 1);
2533 for (i = 0; i < inherit->num_qgroups; ++i) {
918c2ee1 2534 if (*i_qgroups) {
0b246afa 2535 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
918c2ee1
MF		 2536 if (ret)
2537 goto unlock;
2538 }
bed92eae
AJ		 2539 ++i_qgroups;
2540 }
2541
918c2ee1 2542 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
bed92eae
AJ		 2543 struct btrfs_qgroup *src;
2544 struct btrfs_qgroup *dst;
2545
918c2ee1
MF		 2546 if (!i_qgroups[0] || !i_qgroups[1])
2547 continue;
2548
bed92eae
AJ		 2549 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2550 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2551
2552 if (!src || !dst) {
2553 ret = -EINVAL;
2554 goto unlock;
2555 }
2556
2557 dst->rfer = src->rfer - level_size;
2558 dst->rfer_cmpr = src->rfer_cmpr - level_size;
bed92eae 2559 }
918c2ee1 2560 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
bed92eae
AJ		 2561 struct btrfs_qgroup *src;
2562 struct btrfs_qgroup *dst;
2563
918c2ee1
MF		 2564 if (!i_qgroups[0] || !i_qgroups[1])
2565 continue;
2566
bed92eae
AJ		 2567 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2568 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2569
2570 if (!src || !dst) {
2571 ret = -EINVAL;
2572 goto unlock;
2573 }
2574
2575 dst->excl = src->excl + level_size;
2576 dst->excl_cmpr = src->excl_cmpr + level_size;
bed92eae
AJ		 2577 }
2578
2579unlock:
2580 spin_unlock(&fs_info->qgroup_lock);
2581out:
f2f6ed3d 2582 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 2583 return ret;
2584}
2585
a514d638
QW		 2586/*
2587 * Two limits to commit transaction in advance.
2588 *
2589 * For RATIO, it will be 1/RATIO of the remaining limit
2590 * (excluding data and prealloc meta) as threshold.
2591 * For SIZE, it will be in byte unit as threshold.
2592 */
2593#define QGROUP_PERTRANS_RATIO 32
2594#define QGROUP_PERTRANS_SIZE SZ_32M
2595static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2596 const struct btrfs_qgroup *qg, u64 num_bytes)
003d7c59 2597{
a514d638
QW		 2598 u64 limit;
2599 u64 threshold;
2600
003d7c59 2601 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
dba21324 2602 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
003d7c59
JM		 2603 return false;
2604
2605 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
dba21324 2606 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
003d7c59
JM		 2607 return false;
2608
a514d638
QW		 2609 /*
2610 * Even if we passed the check, it's better to check if reservation
2611 * for meta_pertrans is pushing us near limit.
2612 * If there is too much pertrans reservation or it's near the limit,
2613 * let's try commit transaction to free some, using transaction_kthread
2614 */
2615 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2616 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2617 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2618 limit = qg->max_excl;
2619 else
2620 limit = qg->max_rfer;
2621 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2622 qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2623 QGROUP_PERTRANS_RATIO;
2624 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2625
2626 /*
2627 * Use transaction_kthread to commit transaction, so we no
2628 * longer need to bother nested transaction nor lock context.
2629 */
2630 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2631 btrfs_commit_transaction_locksafe(fs_info);
2632 }
2633
003d7c59
JM		 2634 return true;
2635}
2636
dba21324
QW		 2637static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2638 enum btrfs_qgroup_rsv_type type)
bed92eae
AJ		 2639{
2640 struct btrfs_root *quota_root;
2641 struct btrfs_qgroup *qgroup;
2642 struct btrfs_fs_info *fs_info = root->fs_info;
2643 u64 ref_root = root->root_key.objectid;
2644 int ret = 0;
bed92eae
AJ		 2645 struct ulist_node *unode;
2646 struct ulist_iterator uiter;
2647
2648 if (!is_fstree(ref_root))
2649 return 0;
2650
2651 if (num_bytes == 0)
2652 return 0;
f29efe29
SD		 2653
2654 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2655 capable(CAP_SYS_RESOURCE))
2656 enforce = false;
2657
bed92eae
AJ		 2658 spin_lock(&fs_info->qgroup_lock);
2659 quota_root = fs_info->quota_root;
2660 if (!quota_root)
2661 goto out;
2662
2663 qgroup = find_qgroup_rb(fs_info, ref_root);
2664 if (!qgroup)
2665 goto out;
2666
2667 /*
2668 * in a first step, we check all affected qgroups if any limits would
2669 * be exceeded
2670 */
1e8f9158
WS		 2671 ulist_reinit(fs_info->qgroup_ulist);
2672 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
a1840b50 2673 qgroup_to_aux(qgroup), GFP_ATOMIC);
3c97185c
WS		 2674 if (ret < 0)
2675 goto out;
bed92eae 2676 ULIST_ITER_INIT(&uiter);
1e8f9158 2677 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 2678 struct btrfs_qgroup *qg;
2679 struct btrfs_qgroup_list *glist;
2680
ef2fff64 2681 qg = unode_aux_to_qgroup(unode);
bed92eae 2682
a514d638 2683 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
bed92eae 2684 ret = -EDQUOT;
720f1e20
WS		 2685 goto out;
2686 }
bed92eae
AJ		 2687
2688 list_for_each_entry(glist, &qg->groups, next_group) {
1e8f9158
WS		 2689 ret = ulist_add(fs_info->qgroup_ulist,
2690 glist->group->qgroupid,
a1840b50 2691 qgroup_to_aux(glist->group), GFP_ATOMIC);
3c97185c
WS		 2692 if (ret < 0)
2693 goto out;
bed92eae
AJ		 2694 }
2695 }
3c97185c 2696 ret = 0;
bed92eae
AJ		 2697 /*
2698 * no limits exceeded, now record the reservation into all qgroups
2699 */
2700 ULIST_ITER_INIT(&uiter);
1e8f9158 2701 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 2702 struct btrfs_qgroup *qg;
2703
ef2fff64 2704 qg = unode_aux_to_qgroup(unode);
bed92eae 2705
64ee4e75
QW		 2706 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2707 qgroup_rsv_add(fs_info, qg, num_bytes, type);
bed92eae
AJ		 2708 }
2709
2710out:
2711 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 2712 return ret;
2713}
2714
e1211d0e
QW		 2715/*
2716 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2717 * qgroup).
2718 *
2719 * Will handle all higher level qgroup too.
2720 *
2721 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2722 * This special case is only used for META_PERTRANS type.
2723 */
297d750b 2724void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
d4e5c920
QW		 2725 u64 ref_root, u64 num_bytes,
2726 enum btrfs_qgroup_rsv_type type)
bed92eae
AJ		 2727{
2728 struct btrfs_root *quota_root;
2729 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 2730 struct ulist_node *unode;
2731 struct ulist_iterator uiter;
3c97185c 2732 int ret = 0;
bed92eae
AJ		 2733
2734 if (!is_fstree(ref_root))
2735 return;
2736
2737 if (num_bytes == 0)
2738 return;
2739
e1211d0e
QW		 2740 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2741 WARN(1, "%s: Invalid type to free", __func__);
2742 return;
2743 }
bed92eae
AJ		 2744 spin_lock(&fs_info->qgroup_lock);
2745
2746 quota_root = fs_info->quota_root;
2747 if (!quota_root)
2748 goto out;
2749
2750 qgroup = find_qgroup_rb(fs_info, ref_root);
2751 if (!qgroup)
2752 goto out;
2753
e1211d0e 2754 if (num_bytes == (u64)-1)
8287475a
QW		 2755 /*
2756 * We're freeing all pertrans rsv, get reserved value from
2757 * level 0 qgroup as real num_bytes to free.
2758 */
e1211d0e
QW		 2759 num_bytes = qgroup->rsv.values[type];
2760
1e8f9158
WS		 2761 ulist_reinit(fs_info->qgroup_ulist);
2762 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
a1840b50 2763 qgroup_to_aux(qgroup), GFP_ATOMIC);
3c97185c
WS		 2764 if (ret < 0)
2765 goto out;
bed92eae 2766 ULIST_ITER_INIT(&uiter);
1e8f9158 2767 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 2768 struct btrfs_qgroup *qg;
2769 struct btrfs_qgroup_list *glist;
2770
ef2fff64 2771 qg = unode_aux_to_qgroup(unode);
bed92eae 2772
64ee4e75
QW		 2773 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2774 qgroup_rsv_release(fs_info, qg, num_bytes, type);
bed92eae
AJ		 2775
2776 list_for_each_entry(glist, &qg->groups, next_group) {
1e8f9158
WS		 2777 ret = ulist_add(fs_info->qgroup_ulist,
2778 glist->group->qgroupid,
a1840b50 2779 qgroup_to_aux(glist->group), GFP_ATOMIC);
3c97185c
WS		 2780 if (ret < 0)
2781 goto out;
bed92eae
AJ		 2782 }
2783 }
2784
2785out:
2786 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 2787}
2788
ff3d27a0
QW		 2789/*
2790 * Check if the leaf is the last leaf. Which means all node pointers
2791 * are at their last position.
2792 */
2793static bool is_last_leaf(struct btrfs_path *path)
2794{
2795 int i;
2796
2797 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2798 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2799 return false;
2800 }
2801 return true;
2802}
2803
2f232036
JS		 2804/*
2805 * returns < 0 on error, 0 when more leafs are to be scanned.
3393168d 2806 * returns 1 when done.
2f232036 2807 */
62088ca7
LF		 2808static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
2809 struct btrfs_path *path)
2f232036 2810{
62088ca7 2811 struct btrfs_fs_info *fs_info = trans->fs_info;
2f232036 2812 struct btrfs_key found;
0a0e8b89 2813 struct extent_buffer *scratch_leaf = NULL;
2f232036 2814 struct ulist *roots = NULL;
fcebe456 2815 u64 num_bytes;
ff3d27a0 2816 bool done;
2f232036
JS		 2817 int slot;
2818 int ret;
2819
2f232036
JS		 2820 mutex_lock(&fs_info->qgroup_rescan_lock);
2821 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2822 &fs_info->qgroup_rescan_progress,
2823 path, 1, 0);
2824
ab8d0fc4
JM		 2825 btrfs_debug(fs_info,
2826 "current progress key (%llu %u %llu), search_slot ret %d",
2827 fs_info->qgroup_rescan_progress.objectid,
2828 fs_info->qgroup_rescan_progress.type,
2829 fs_info->qgroup_rescan_progress.offset, ret);
2f232036
JS		 2830
2831 if (ret) {
2832 /*
2833 * The rescan is about to end, we will not be scanning any
2834 * further blocks. We cannot unset the RESCAN flag here, because
2835 * we want to commit the transaction if everything went well.
2836 * To make the live accounting work in this phase, we set our
2837 * scan progress pointer such that every real extent objectid
2838 * will be smaller.
2839 */
2840 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2841 btrfs_release_path(path);
2842 mutex_unlock(&fs_info->qgroup_rescan_lock);
2843 return ret;
2844 }
ff3d27a0 2845 done = is_last_leaf(path);
2f232036
JS		 2846
2847 btrfs_item_key_to_cpu(path->nodes[0], &found,
2848 btrfs_header_nritems(path->nodes[0]) - 1);
2849 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2850
0a0e8b89
QW		 2851 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2852 if (!scratch_leaf) {
2853 ret = -ENOMEM;
2854 mutex_unlock(&fs_info->qgroup_rescan_lock);
2855 goto out;
2856 }
2857 extent_buffer_get(scratch_leaf);
2858 btrfs_tree_read_lock(scratch_leaf);
2859 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2f232036
JS		 2860 slot = path->slots[0];
2861 btrfs_release_path(path);
2862 mutex_unlock(&fs_info->qgroup_rescan_lock);
2863
2864 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2865 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3a6d75e8
JB		 2866 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2867 found.type != BTRFS_METADATA_ITEM_KEY)
2f232036 2868 continue;
3a6d75e8 2869 if (found.type == BTRFS_METADATA_ITEM_KEY)
da17066c 2870 num_bytes = fs_info->nodesize;
3a6d75e8
JB		 2871 else
2872 num_bytes = found.offset;
2873
fcebe456 2874 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
c995ab3c 2875 &roots, false);
2f232036
JS		 2876 if (ret < 0)
2877 goto out;
9d220c95 2878 /* For rescan, just pass old_roots as NULL */
8696d760
LF		 2879 ret = btrfs_qgroup_account_extent(trans, found.objectid,
2880 num_bytes, NULL, roots);
9d220c95 2881 if (ret < 0)
fcebe456 2882 goto out;
2f232036 2883 }
2f232036 2884out:
0a0e8b89
QW		 2885 if (scratch_leaf) {
2886 btrfs_tree_read_unlock_blocking(scratch_leaf);
2887 free_extent_buffer(scratch_leaf);
2888 }
2f232036 2889
6f7de19e 2890 if (done && !ret) {
ff3d27a0 2891 ret = 1;
6f7de19e
QW		 2892 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2893 }
2f232036
JS		 2894 return ret;
2895}
2896
d458b054 2897static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2f232036 2898{
b382a324
JS		 2899 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2900 qgroup_rescan_work);
2f232036
JS		 2901 struct btrfs_path *path;
2902 struct btrfs_trans_handle *trans = NULL;
2f232036 2903 int err = -ENOMEM;
53b7cde9 2904 int ret = 0;
2f232036
JS		 2905
2906 path = btrfs_alloc_path();
2907 if (!path)
2908 goto out;
b6debf15
QW		 2909 /*
2910 * Rescan should only search for commit root, and any later difference
2911 * should be recorded by qgroup
2912 */
2913 path->search_commit_root = 1;
2914 path->skip_locking = 1;
2f232036
JS		 2915
2916 err = 0;
7343dd61 2917 while (!err && !btrfs_fs_closing(fs_info)) {
2f232036
JS		 2918 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2919 if (IS_ERR(trans)) {
2920 err = PTR_ERR(trans);
2921 break;
2922 }
afcdd129 2923 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2f232036
JS		 2924 err = -EINTR;
2925 } else {
62088ca7 2926 err = qgroup_rescan_leaf(trans, path);
2f232036
JS		 2927 }
2928 if (err > 0)
3a45bb20 2929 btrfs_commit_transaction(trans);
2f232036 2930 else
3a45bb20 2931 btrfs_end_transaction(trans);
2f232036
JS		 2932 }
2933
2934out:
2f232036 2935 btrfs_free_path(path);
2f232036
JS		 2936
2937 mutex_lock(&fs_info->qgroup_rescan_lock);
7343dd61
JM		 2938 if (!btrfs_fs_closing(fs_info))
2939 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2f232036 2940
3393168d 2941 if (err > 0 &&
2f232036
JS		 2942 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2943 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2944 } else if (err < 0) {
2945 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2946 }
2947 mutex_unlock(&fs_info->qgroup_rescan_lock);
2948
53b7cde9 2949 /*
01327610 2950 * only update status, since the previous part has already updated the
53b7cde9
QW		 2951 * qgroup info.
2952 */
2953 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2954 if (IS_ERR(trans)) {
2955 err = PTR_ERR(trans);
2956 btrfs_err(fs_info,
913e1535 2957 "fail to start transaction for status update: %d",
53b7cde9
QW		 2958 err);
2959 goto done;
2960 }
2e980acd 2961 ret = update_qgroup_status_item(trans);
53b7cde9
QW		 2962 if (ret < 0) {
2963 err = ret;
ab8d0fc4 2964 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
53b7cde9 2965 }
3a45bb20 2966 btrfs_end_transaction(trans);
53b7cde9 2967
7343dd61
JM		 2968 if (btrfs_fs_closing(fs_info)) {
2969 btrfs_info(fs_info, "qgroup scan paused");
2970 } else if (err >= 0) {
efe120a0 2971 btrfs_info(fs_info, "qgroup scan completed%s",
3393168d 2972 err > 0 ? " (inconsistency flag cleared)" : "");
2f232036 2973 } else {
efe120a0 2974 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2f232036 2975 }
57254b6e 2976
53b7cde9 2977done:
d2c609b8
JM		 2978 mutex_lock(&fs_info->qgroup_rescan_lock);
2979 fs_info->qgroup_rescan_running = false;
2980 mutex_unlock(&fs_info->qgroup_rescan_lock);
57254b6e 2981 complete_all(&fs_info->qgroup_rescan_completion);
2f232036
JS		 2982}
2983
b382a324
JS		 2984/*
2985 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2986 * memory required for the rescan context.
2987 */
2988static int
2989qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2990 int init_flags)
2f232036
JS		 2991{
2992 int ret = 0;
2f232036 2993
9593bf49
QW		 2994 if (!init_flags) {
2995 /* we're resuming qgroup rescan at mount time */
e4e7ede7
FM		 2996 if (!(fs_info->qgroup_flags &
2997 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
9593bf49
QW		 2998 btrfs_warn(fs_info,
2999 "qgroup rescan init failed, qgroup is not enabled");
e4e7ede7
FM		 3000 ret = -EINVAL;
3001 } else if (!(fs_info->qgroup_flags &
3002 BTRFS_QGROUP_STATUS_FLAG_ON)) {
9593bf49
QW		 3003 btrfs_warn(fs_info,
3004 "qgroup rescan init failed, qgroup rescan is not queued");
e4e7ede7
FM		 3005 ret = -EINVAL;
3006 }
3007
3008 if (ret)
3009 return ret;
b382a324 3010 }
2f232036
JS		 3011
3012 mutex_lock(&fs_info->qgroup_rescan_lock);
3013 spin_lock(&fs_info->qgroup_lock);
b382a324
JS		 3014
3015 if (init_flags) {
9593bf49
QW		 3016 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3017 btrfs_warn(fs_info,
3018 "qgroup rescan is already in progress");
b382a324 3019 ret = -EINPROGRESS;
9593bf49
QW		 3020 } else if (!(fs_info->qgroup_flags &
3021 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3022 btrfs_warn(fs_info,
3023 "qgroup rescan init failed, qgroup is not enabled");
b382a324 3024 ret = -EINVAL;
9593bf49 3025 }
b382a324
JS		 3026
3027 if (ret) {
3028 spin_unlock(&fs_info->qgroup_lock);
3029 mutex_unlock(&fs_info->qgroup_rescan_lock);
9593bf49 3030 return ret;
b382a324 3031 }
b382a324 3032 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2f232036
JS		 3033 }
3034
2f232036
JS		 3035 memset(&fs_info->qgroup_rescan_progress, 0,
3036 sizeof(fs_info->qgroup_rescan_progress));
b382a324 3037 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
190631f1 3038 init_completion(&fs_info->qgroup_rescan_completion);
8d9eddad 3039 fs_info->qgroup_rescan_running = true;
b382a324
JS		 3040
3041 spin_unlock(&fs_info->qgroup_lock);
3042 mutex_unlock(&fs_info->qgroup_rescan_lock);
3043
b382a324
JS		 3044 memset(&fs_info->qgroup_rescan_work, 0,
3045 sizeof(fs_info->qgroup_rescan_work));
fc97fab0 3046 btrfs_init_work(&fs_info->qgroup_rescan_work,
9e0af237 3047 btrfs_qgroup_rescan_helper,
fc97fab0 3048 btrfs_qgroup_rescan_worker, NULL, NULL);
b382a324
JS		 3049 return 0;
3050}
3051
3052static void
3053qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3054{
3055 struct rb_node *n;
3056 struct btrfs_qgroup *qgroup;
3057
3058 spin_lock(&fs_info->qgroup_lock);
2f232036
JS		 3059 /* clear all current qgroup tracking information */
3060 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3061 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3062 qgroup->rfer = 0;
3063 qgroup->rfer_cmpr = 0;
3064 qgroup->excl = 0;
3065 qgroup->excl_cmpr = 0;
9c7b0c2e 3066 qgroup_dirty(fs_info, qgroup);
2f232036
JS		 3067 }
3068 spin_unlock(&fs_info->qgroup_lock);
b382a324 3069}
2f232036 3070
b382a324
JS		 3071int
3072btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3073{
3074 int ret = 0;
3075 struct btrfs_trans_handle *trans;
3076
3077 ret = qgroup_rescan_init(fs_info, 0, 1);
3078 if (ret)
3079 return ret;
3080
3081 /*
3082 * We have set the rescan_progress to 0, which means no more
3083 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3084 * However, btrfs_qgroup_account_ref may be right after its call
3085 * to btrfs_find_all_roots, in which case it would still do the
3086 * accounting.
3087 * To solve this, we're committing the transaction, which will
3088 * ensure we run all delayed refs and only after that, we are
3089 * going to clear all tracking information for a clean start.
3090 */
3091
3092 trans = btrfs_join_transaction(fs_info->fs_root);
3093 if (IS_ERR(trans)) {
3094 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3095 return PTR_ERR(trans);
3096 }
3a45bb20 3097 ret = btrfs_commit_transaction(trans);
b382a324
JS		 3098 if (ret) {
3099 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3100 return ret;
3101 }
3102
3103 qgroup_rescan_zero_tracking(fs_info);
3104
fc97fab0
QW		 3105 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3106 &fs_info->qgroup_rescan_work);
2f232036
JS		 3107
3108 return 0;
3109}
57254b6e 3110
d06f23d6
JM		 3111int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3112 bool interruptible)
57254b6e
JS		 3113{
3114 int running;
3115 int ret = 0;
3116
3117 mutex_lock(&fs_info->qgroup_rescan_lock);
3118 spin_lock(&fs_info->qgroup_lock);
d2c609b8 3119 running = fs_info->qgroup_rescan_running;
57254b6e
JS		 3120 spin_unlock(&fs_info->qgroup_lock);
3121 mutex_unlock(&fs_info->qgroup_rescan_lock);
3122
d06f23d6
JM		 3123 if (!running)
3124 return 0;
3125
3126 if (interruptible)
57254b6e
JS		 3127 ret = wait_for_completion_interruptible(
3128 &fs_info->qgroup_rescan_completion);
d06f23d6
JM		 3129 else
3130 wait_for_completion(&fs_info->qgroup_rescan_completion);
57254b6e
JS		 3131
3132 return ret;
3133}
b382a324
JS		 3134
3135/*
3136 * this is only called from open_ctree where we're still single threaded, thus
3137 * locking is omitted here.
3138 */
3139void
3140btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3141{
3142 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
fc97fab0
QW		 3143 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3144 &fs_info->qgroup_rescan_work);
b382a324 3145}
52472553
QW		 3146
3147/*
3148 * Reserve qgroup space for range [start, start + len).
3149 *
3150 * This function will either reserve space from related qgroups or doing
3151 * nothing if the range is already reserved.
3152 *
3153 * Return 0 for successful reserve
3154 * Return <0 for error (including -EQUOT)
3155 *
3156 * NOTE: this function may sleep for memory allocation.
364ecf36
QW		 3157 * if btrfs_qgroup_reserve_data() is called multiple times with
3158 * same @reserved, caller must ensure when error happens it's OK
3159 * to free *ALL* reserved space.
52472553 3160 */
364ecf36
QW		 3161int btrfs_qgroup_reserve_data(struct inode *inode,
3162 struct extent_changeset **reserved_ret, u64 start,
3163 u64 len)
52472553
QW		 3164{
3165 struct btrfs_root *root = BTRFS_I(inode)->root;
52472553
QW		 3166 struct ulist_node *unode;
3167 struct ulist_iterator uiter;
364ecf36
QW		 3168 struct extent_changeset *reserved;
3169 u64 orig_reserved;
3170 u64 to_reserve;
52472553
QW		 3171 int ret;
3172
afcdd129 3173 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
4fd786e6 3174 !is_fstree(root->root_key.objectid) || len == 0)
52472553
QW		 3175 return 0;
3176
364ecf36
QW		 3177 /* @reserved parameter is mandatory for qgroup */
3178 if (WARN_ON(!reserved_ret))
3179 return -EINVAL;
3180 if (!*reserved_ret) {
3181 *reserved_ret = extent_changeset_alloc();
3182 if (!*reserved_ret)
3183 return -ENOMEM;
3184 }
3185 reserved = *reserved_ret;
3186 /* Record already reserved space */
3187 orig_reserved = reserved->bytes_changed;
52472553 3188 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
364ecf36
QW		 3189 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3190
3191 /* Newly reserved space */
3192 to_reserve = reserved->bytes_changed - orig_reserved;
81fb6f77 3193 trace_btrfs_qgroup_reserve_data(inode, start, len,
364ecf36 3194 to_reserve, QGROUP_RESERVE);
52472553
QW		 3195 if (ret < 0)
3196 goto cleanup;
dba21324 3197 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
52472553
QW		 3198 if (ret < 0)
3199 goto cleanup;
3200
52472553
QW		 3201 return ret;
3202
3203cleanup:
364ecf36 3204 /* cleanup *ALL* already reserved ranges */
52472553 3205 ULIST_ITER_INIT(&uiter);
364ecf36 3206 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
52472553 3207 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
ae0f1625 3208 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
364ecf36 3209 extent_changeset_release(reserved);
52472553
QW		 3210 return ret;
3211}
f695fdce 3212
bc42bda2
QW		 3213/* Free ranges specified by @reserved, normally in error path */
3214static int qgroup_free_reserved_data(struct inode *inode,
3215 struct extent_changeset *reserved, u64 start, u64 len)
3216{
3217 struct btrfs_root *root = BTRFS_I(inode)->root;
3218 struct ulist_node *unode;
3219 struct ulist_iterator uiter;
3220 struct extent_changeset changeset;
3221 int freed = 0;
3222 int ret;
3223
3224 extent_changeset_init(&changeset);
3225 len = round_up(start + len, root->fs_info->sectorsize);
3226 start = round_down(start, root->fs_info->sectorsize);
3227
3228 ULIST_ITER_INIT(&uiter);
3229 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3230 u64 range_start = unode->val;
3231 /* unode->aux is the inclusive end */
3232 u64 range_len = unode->aux - range_start + 1;
3233 u64 free_start;
3234 u64 free_len;
3235
3236 extent_changeset_release(&changeset);
3237
3238 /* Only free range in range [start, start + len) */
3239 if (range_start >= start + len ||
3240 range_start + range_len <= start)
3241 continue;
3242 free_start = max(range_start, start);
3243 free_len = min(start + len, range_start + range_len) -
3244 free_start;
3245 /*
3246 * TODO: To also modify reserved->ranges_reserved to reflect
3247 * the modification.
3248 *
3249 * However as long as we free qgroup reserved according to
3250 * EXTENT_QGROUP_RESERVED, we won't double free.
3251 * So not need to rush.
3252 */
3253 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3254 free_start, free_start + free_len - 1,
3255 EXTENT_QGROUP_RESERVED, &changeset);
3256 if (ret < 0)
3257 goto out;
3258 freed += changeset.bytes_changed;
3259 }
4fd786e6 3260 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
d4e5c920 3261 BTRFS_QGROUP_RSV_DATA);
bc42bda2
QW		 3262 ret = freed;
3263out:
3264 extent_changeset_release(&changeset);
3265 return ret;
3266}
3267
3268static int __btrfs_qgroup_release_data(struct inode *inode,
3269 struct extent_changeset *reserved, u64 start, u64 len,
3270 int free)
f695fdce
QW		 3271{
3272 struct extent_changeset changeset;
81fb6f77 3273 int trace_op = QGROUP_RELEASE;
f695fdce
QW		 3274 int ret;
3275
bc42bda2
QW		 3276 /* In release case, we shouldn't have @reserved */
3277 WARN_ON(!free && reserved);
3278 if (free && reserved)
3279 return qgroup_free_reserved_data(inode, reserved, start, len);
364ecf36 3280 extent_changeset_init(&changeset);
f695fdce 3281 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
f734c44a 3282 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
f695fdce
QW		 3283 if (ret < 0)
3284 goto out;
3285
d51ea5dd 3286 if (free)
81fb6f77 3287 trace_op = QGROUP_FREE;
81fb6f77
QW		 3288 trace_btrfs_qgroup_release_data(inode, start, len,
3289 changeset.bytes_changed, trace_op);
d51ea5dd
QW		 3290 if (free)
3291 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
4fd786e6 3292 BTRFS_I(inode)->root->root_key.objectid,
d4e5c920 3293 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
7bc329c1 3294 ret = changeset.bytes_changed;
f695fdce 3295out:
364ecf36 3296 extent_changeset_release(&changeset);
f695fdce
QW		 3297 return ret;
3298}
3299
3300/*
3301 * Free a reserved space range from io_tree and related qgroups
3302 *
3303 * Should be called when a range of pages get invalidated before reaching disk.
3304 * Or for error cleanup case.
bc42bda2
QW		 3305 * if @reserved is given, only reserved range in [@start, @start + @len) will
3306 * be freed.
f695fdce
QW		 3307 *
3308 * For data written to disk, use btrfs_qgroup_release_data().
3309 *
3310 * NOTE: This function may sleep for memory allocation.
3311 */
bc42bda2
QW		 3312int btrfs_qgroup_free_data(struct inode *inode,
3313 struct extent_changeset *reserved, u64 start, u64 len)
f695fdce 3314{
bc42bda2 3315 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
f695fdce
QW		 3316}
3317
3318/*
3319 * Release a reserved space range from io_tree only.
3320 *
3321 * Should be called when a range of pages get written to disk and corresponding
3322 * FILE_EXTENT is inserted into corresponding root.
3323 *
3324 * Since new qgroup accounting framework will only update qgroup numbers at
3325 * commit_transaction() time, its reserved space shouldn't be freed from
3326 * related qgroups.
3327 *
3328 * But we should release the range from io_tree, to allow further write to be
3329 * COWed.
3330 *
3331 * NOTE: This function may sleep for memory allocation.
3332 */
3333int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3334{
bc42bda2 3335 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
f695fdce 3336}
55eeaf05 3337
8287475a
QW		 3338static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3339 enum btrfs_qgroup_rsv_type type)
3340{
3341 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3342 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3343 return;
3344 if (num_bytes == 0)
3345 return;
3346
3347 spin_lock(&root->qgroup_meta_rsv_lock);
3348 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3349 root->qgroup_meta_rsv_prealloc += num_bytes;
3350 else
3351 root->qgroup_meta_rsv_pertrans += num_bytes;
3352 spin_unlock(&root->qgroup_meta_rsv_lock);
3353}
3354
3355static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3356 enum btrfs_qgroup_rsv_type type)
3357{
3358 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3359 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3360 return 0;
3361 if (num_bytes == 0)
3362 return 0;
3363
3364 spin_lock(&root->qgroup_meta_rsv_lock);
3365 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3366 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3367 num_bytes);
3368 root->qgroup_meta_rsv_prealloc -= num_bytes;
3369 } else {
3370 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3371 num_bytes);
3372 root->qgroup_meta_rsv_pertrans -= num_bytes;
3373 }
3374 spin_unlock(&root->qgroup_meta_rsv_lock);
3375 return num_bytes;
3376}
3377
733e03a0
QW		 3378int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3379 enum btrfs_qgroup_rsv_type type, bool enforce)
55eeaf05 3380{
0b246afa 3381 struct btrfs_fs_info *fs_info = root->fs_info;
55eeaf05
QW		 3382 int ret;
3383
0b246afa 3384 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3385 !is_fstree(root->root_key.objectid) || num_bytes == 0)
55eeaf05
QW		 3386 return 0;
3387
0b246afa 3388 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4ee0d883 3389 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
733e03a0 3390 ret = qgroup_reserve(root, num_bytes, enforce, type);
55eeaf05
QW		 3391 if (ret < 0)
3392 return ret;
8287475a
QW		 3393 /*
3394 * Record what we have reserved into root.
3395 *
3396 * To avoid quota disabled->enabled underflow.
3397 * In that case, we may try to free space we haven't reserved
3398 * (since quota was disabled), so record what we reserved into root.
3399 * And ensure later release won't underflow this number.
3400 */
3401 add_root_meta_rsv(root, num_bytes, type);
55eeaf05
QW		 3402 return ret;
3403}
3404
733e03a0 3405void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
55eeaf05 3406{
0b246afa 3407 struct btrfs_fs_info *fs_info = root->fs_info;
55eeaf05 3408
0b246afa 3409 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3410 !is_fstree(root->root_key.objectid))
55eeaf05
QW		 3411 return;
3412
e1211d0e 3413 /* TODO: Update trace point to handle such free */
4ee0d883 3414 trace_qgroup_meta_free_all_pertrans(root);
e1211d0e 3415 /* Special value -1 means to free all reserved space */
4fd786e6 3416 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
733e03a0 3417 BTRFS_QGROUP_RSV_META_PERTRANS);
55eeaf05
QW		 3418}
3419
733e03a0
QW		 3420void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3421 enum btrfs_qgroup_rsv_type type)
55eeaf05 3422{
0b246afa
JM		 3423 struct btrfs_fs_info *fs_info = root->fs_info;
3424
3425 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3426 !is_fstree(root->root_key.objectid))
55eeaf05
QW		 3427 return;
3428
8287475a
QW		 3429 /*
3430 * reservation for META_PREALLOC can happen before quota is enabled,
3431 * which can lead to underflow.
3432 * Here ensure we will only free what we really have reserved.
3433 */
3434 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
0b246afa 3435 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4ee0d883 3436 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
4fd786e6
MT		 3437 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3438 num_bytes, type);
55eeaf05 3439}
56fa9d07 3440
64cfaef6
QW		 3441static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3442 int num_bytes)
3443{
3444 struct btrfs_root *quota_root = fs_info->quota_root;
3445 struct btrfs_qgroup *qgroup;
3446 struct ulist_node *unode;
3447 struct ulist_iterator uiter;
3448 int ret = 0;
3449
3450 if (num_bytes == 0)
3451 return;
3452 if (!quota_root)
3453 return;
3454
3455 spin_lock(&fs_info->qgroup_lock);
3456 qgroup = find_qgroup_rb(fs_info, ref_root);
3457 if (!qgroup)
3458 goto out;
3459 ulist_reinit(fs_info->qgroup_ulist);
3460 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
a1840b50 3461 qgroup_to_aux(qgroup), GFP_ATOMIC);
64cfaef6
QW		 3462 if (ret < 0)
3463 goto out;
3464 ULIST_ITER_INIT(&uiter);
3465 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3466 struct btrfs_qgroup *qg;
3467 struct btrfs_qgroup_list *glist;
3468
3469 qg = unode_aux_to_qgroup(unode);
3470
3471 qgroup_rsv_release(fs_info, qg, num_bytes,
3472 BTRFS_QGROUP_RSV_META_PREALLOC);
3473 qgroup_rsv_add(fs_info, qg, num_bytes,
3474 BTRFS_QGROUP_RSV_META_PERTRANS);
3475 list_for_each_entry(glist, &qg->groups, next_group) {
3476 ret = ulist_add(fs_info->qgroup_ulist,
3477 glist->group->qgroupid,
a1840b50 3478 qgroup_to_aux(glist->group), GFP_ATOMIC);
64cfaef6
QW		 3479 if (ret < 0)
3480 goto out;
3481 }
3482 }
3483out:
3484 spin_unlock(&fs_info->qgroup_lock);
3485}
3486
3487void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3488{
3489 struct btrfs_fs_info *fs_info = root->fs_info;
3490
3491 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3492 !is_fstree(root->root_key.objectid))
64cfaef6 3493 return;
8287475a
QW		 3494 /* Same as btrfs_qgroup_free_meta_prealloc() */
3495 num_bytes = sub_root_meta_rsv(root, num_bytes,
3496 BTRFS_QGROUP_RSV_META_PREALLOC);
4ee0d883 3497 trace_qgroup_meta_convert(root, num_bytes);
4fd786e6 3498 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
64cfaef6
QW		 3499}
3500
56fa9d07 3501/*
01327610 3502 * Check qgroup reserved space leaking, normally at destroy inode
56fa9d07
QW		 3503 * time
3504 */
3505void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3506{
3507 struct extent_changeset changeset;
3508 struct ulist_node *unode;
3509 struct ulist_iterator iter;
3510 int ret;
3511
364ecf36 3512 extent_changeset_init(&changeset);
56fa9d07 3513 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
f734c44a 3514 EXTENT_QGROUP_RESERVED, &changeset);
56fa9d07
QW		 3515
3516 WARN_ON(ret < 0);
3517 if (WARN_ON(changeset.bytes_changed)) {
3518 ULIST_ITER_INIT(&iter);
53d32359 3519 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
56fa9d07
QW		 3520 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3521 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3522 inode->i_ino, unode->val, unode->aux);
3523 }
0b08e1f4 3524 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
4fd786e6 3525 BTRFS_I(inode)->root->root_key.objectid,
d4e5c920 3526 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
0b08e1f4 3527
56fa9d07 3528 }
364ecf36 3529 extent_changeset_release(&changeset);
56fa9d07 3530}
Ubuntu Hirsute Linux kernel mirror