projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: correctly validate compression type
[linux-block.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
52042d8e 33 * - caches for ulists
bed92eae
AJ		 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 524 /*
52042d8e 525 * We call btrfs_free_qgroup_config() when unmounting
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
7503b83d
DS		 897 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 if (!fs_info->qgroup_ulist) {
899 ret = -ENOMEM;
900 goto out;
901 }
902
340f1aa2
NB		 903 /*
904 * 1 for quota root item
905 * 1 for BTRFS_QGROUP_STATUS item
906 *
907 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
908 * per subvolume. However those are not currently reserved since it
909 * would be a lot of overkill.
910 */
911 trans = btrfs_start_transaction(tree_root, 2);
912 if (IS_ERR(trans)) {
913 ret = PTR_ERR(trans);
914 trans = NULL;
915 goto out;
916 }
917
bed92eae
AJ		 918 /*
919 * initially create the quota tree
920 */
9b7a2440 921 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
bed92eae
AJ		 922 if (IS_ERR(quota_root)) {
923 ret = PTR_ERR(quota_root);
340f1aa2 924 btrfs_abort_transaction(trans, ret);
bed92eae
AJ		 925 goto out;
926 }
927
928 path = btrfs_alloc_path();
5b7ff5b3
TI		 929 if (!path) {
930 ret = -ENOMEM;
340f1aa2 931 btrfs_abort_transaction(trans, ret);
5b7ff5b3
TI		 932 goto out_free_root;
933 }
bed92eae
AJ		 934
935 key.objectid = 0;
936 key.type = BTRFS_QGROUP_STATUS_KEY;
937 key.offset = 0;
938
939 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
940 sizeof(*ptr));
340f1aa2
NB		 941 if (ret) {
942 btrfs_abort_transaction(trans, ret);
5b7ff5b3 943 goto out_free_path;
340f1aa2 944 }
bed92eae
AJ		 945
946 leaf = path->nodes[0];
947 ptr = btrfs_item_ptr(leaf, path->slots[0],
948 struct btrfs_qgroup_status_item);
949 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
950 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
951 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
952 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
953 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
2f232036 954 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
bed92eae
AJ		 955
956 btrfs_mark_buffer_dirty(leaf);
957
7708f029
WS		 958 key.objectid = 0;
959 key.type = BTRFS_ROOT_REF_KEY;
960 key.offset = 0;
961
962 btrfs_release_path(path);
963 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
964 if (ret > 0)
965 goto out_add_root;
340f1aa2
NB		 966 if (ret < 0) {
967 btrfs_abort_transaction(trans, ret);
7708f029 968 goto out_free_path;
340f1aa2 969 }
7708f029
WS		 970
971 while (1) {
972 slot = path->slots[0];
973 leaf = path->nodes[0];
974 btrfs_item_key_to_cpu(leaf, &found_key, slot);
975
976 if (found_key.type == BTRFS_ROOT_REF_KEY) {
977 ret = add_qgroup_item(trans, quota_root,
978 found_key.offset);
340f1aa2
NB		 979 if (ret) {
980 btrfs_abort_transaction(trans, ret);
7708f029 981 goto out_free_path;
340f1aa2 982 }
7708f029 983
7708f029
WS		 984 qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 if (IS_ERR(qgroup)) {
7708f029 986 ret = PTR_ERR(qgroup);
340f1aa2 987 btrfs_abort_transaction(trans, ret);
7708f029
WS		 988 goto out_free_path;
989 }
7708f029
WS		 990 }
991 ret = btrfs_next_item(tree_root, path);
340f1aa2
NB		 992 if (ret < 0) {
993 btrfs_abort_transaction(trans, ret);
7708f029 994 goto out_free_path;
340f1aa2 995 }
7708f029
WS		 996 if (ret)
997 break;
998 }
999
1000out_add_root:
1001 btrfs_release_path(path);
1002 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
340f1aa2
NB		 1003 if (ret) {
1004 btrfs_abort_transaction(trans, ret);
7708f029 1005 goto out_free_path;
340f1aa2 1006 }
7708f029 1007
7708f029
WS		 1008 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 if (IS_ERR(qgroup)) {
7708f029 1010 ret = PTR_ERR(qgroup);
340f1aa2 1011 btrfs_abort_transaction(trans, ret);
7708f029
WS		 1012 goto out_free_path;
1013 }
340f1aa2
NB		 1014
1015 ret = btrfs_commit_transaction(trans);
b9b8a41a
DC		 1016 trans = NULL;
1017 if (ret)
340f1aa2 1018 goto out_free_path;
340f1aa2 1019
9a6f209e
FM		 1020 /*
1021 * Set quota enabled flag after committing the transaction, to avoid
1022 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1023 * creation.
1024 */
1025 spin_lock(&fs_info->qgroup_lock);
1026 fs_info->quota_root = quota_root;
1027 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1028 spin_unlock(&fs_info->qgroup_lock);
1029
5d23515b
NB		 1030 ret = qgroup_rescan_init(fs_info, 0, 1);
1031 if (!ret) {
1032 qgroup_rescan_zero_tracking(fs_info);
1033 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1034 &fs_info->qgroup_rescan_work);
1035 }
1036
5b7ff5b3 1037out_free_path:
bed92eae 1038 btrfs_free_path(path);
5b7ff5b3
TI		 1039out_free_root:
1040 if (ret) {
1041 free_extent_buffer(quota_root->node);
1042 free_extent_buffer(quota_root->commit_root);
1043 kfree(quota_root);
1044 }
1045out:
eb1716af 1046 if (ret) {
1e8f9158 1047 ulist_free(fs_info->qgroup_ulist);
eb1716af 1048 fs_info->qgroup_ulist = NULL;
340f1aa2
NB		 1049 if (trans)
1050 btrfs_end_transaction(trans);
eb1716af 1051 }
f2f6ed3d 1052 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1053 return ret;
1054}
1055
340f1aa2 1056int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
bed92eae 1057{
bed92eae 1058 struct btrfs_root *quota_root;
340f1aa2 1059 struct btrfs_trans_handle *trans = NULL;
bed92eae
AJ		 1060 int ret = 0;
1061
f2f6ed3d 1062 mutex_lock(&fs_info->qgroup_ioctl_lock);
58400fce 1063 if (!fs_info->quota_root)
f2f6ed3d 1064 goto out;
340f1aa2
NB		 1065
1066 /*
1067 * 1 For the root item
1068 *
1069 * We should also reserve enough items for the quota tree deletion in
1070 * btrfs_clean_quota_tree but this is not done.
1071 */
1072 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1073 if (IS_ERR(trans)) {
1074 ret = PTR_ERR(trans);
1075 goto out;
1076 }
1077
afcdd129 1078 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
d06f23d6 1079 btrfs_qgroup_wait_for_completion(fs_info, false);
967ef513 1080 spin_lock(&fs_info->qgroup_lock);
bed92eae
AJ		 1081 quota_root = fs_info->quota_root;
1082 fs_info->quota_root = NULL;
8ea0ec9e 1083 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
bed92eae
AJ		 1084 spin_unlock(&fs_info->qgroup_lock);
1085
e685da14
WS		 1086 btrfs_free_qgroup_config(fs_info);
1087
bed92eae 1088 ret = btrfs_clean_quota_tree(trans, quota_root);
340f1aa2
NB		 1089 if (ret) {
1090 btrfs_abort_transaction(trans, ret);
1091 goto end_trans;
1092 }
bed92eae 1093
ab9ce7d4 1094 ret = btrfs_del_root(trans, &quota_root->root_key);
340f1aa2
NB		 1095 if (ret) {
1096 btrfs_abort_transaction(trans, ret);
1097 goto end_trans;
1098 }
bed92eae
AJ		 1099
1100 list_del(&quota_root->dirty_list);
1101
1102 btrfs_tree_lock(quota_root->node);
6a884d7d 1103 btrfs_clean_tree_block(quota_root->node);
bed92eae
AJ		 1104 btrfs_tree_unlock(quota_root->node);
1105 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1106
1107 free_extent_buffer(quota_root->node);
1108 free_extent_buffer(quota_root->commit_root);
1109 kfree(quota_root);
340f1aa2
NB		 1110
1111end_trans:
1112 ret = btrfs_end_transaction(trans);
bed92eae 1113out:
f2f6ed3d 1114 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1115 return ret;
1116}
1117
2f232036
JS		 1118static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1119 struct btrfs_qgroup *qgroup)
bed92eae 1120{
2f232036
JS		 1121 if (list_empty(&qgroup->dirty))
1122 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
bed92eae
AJ		 1123}
1124
9c8b35b1 1125/*
429d6275
QW		 1126 * The easy accounting, we're updating qgroup relationship whose child qgroup
1127 * only has exclusive extents.
1128 *
52042d8e 1129 * In this case, all exclusive extents will also be exclusive for parent, so
429d6275
QW		 1130 * excl/rfer just get added/removed.
1131 *
1132 * So is qgroup reservation space, which should also be added/removed to
1133 * parent.
1134 * Or when child tries to release reservation space, parent will underflow its
1135 * reservation (for relationship adding case).
9c8b35b1
QW		 1136 *
1137 * Caller should hold fs_info->qgroup_lock.
1138 */
1139static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1140 struct ulist *tmp, u64 ref_root,
429d6275 1141 struct btrfs_qgroup *src, int sign)
9c8b35b1
QW		 1142{
1143 struct btrfs_qgroup *qgroup;
1144 struct btrfs_qgroup_list *glist;
1145 struct ulist_node *unode;
1146 struct ulist_iterator uiter;
429d6275 1147 u64 num_bytes = src->excl;
9c8b35b1
QW		 1148 int ret = 0;
1149
1150 qgroup = find_qgroup_rb(fs_info, ref_root);
1151 if (!qgroup)
1152 goto out;
1153
1154 qgroup->rfer += sign * num_bytes;
1155 qgroup->rfer_cmpr += sign * num_bytes;
1156
1157 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1158 qgroup->excl += sign * num_bytes;
1159 qgroup->excl_cmpr += sign * num_bytes;
429d6275
QW		 1160
1161 if (sign > 0)
64ee4e75 1162 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
429d6275 1163 else
64ee4e75 1164 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
9c8b35b1
QW		 1165
1166 qgroup_dirty(fs_info, qgroup);
1167
1168 /* Get all of the parent groups that contain this qgroup */
1169 list_for_each_entry(glist, &qgroup->groups, next_group) {
1170 ret = ulist_add(tmp, glist->group->qgroupid,
ef2fff64 1171 qgroup_to_aux(glist->group), GFP_ATOMIC);
9c8b35b1
QW		 1172 if (ret < 0)
1173 goto out;
1174 }
1175
1176 /* Iterate all of the parents and adjust their reference counts */
1177 ULIST_ITER_INIT(&uiter);
1178 while ((unode = ulist_next(tmp, &uiter))) {
ef2fff64 1179 qgroup = unode_aux_to_qgroup(unode);
9c8b35b1
QW		 1180 qgroup->rfer += sign * num_bytes;
1181 qgroup->rfer_cmpr += sign * num_bytes;
1182 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1183 qgroup->excl += sign * num_bytes;
429d6275 1184 if (sign > 0)
64ee4e75 1185 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
429d6275 1186 else
64ee4e75 1187 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
9c8b35b1
QW		 1188 qgroup->excl_cmpr += sign * num_bytes;
1189 qgroup_dirty(fs_info, qgroup);
1190
1191 /* Add any parents of the parents */
1192 list_for_each_entry(glist, &qgroup->groups, next_group) {
1193 ret = ulist_add(tmp, glist->group->qgroupid,
ef2fff64 1194 qgroup_to_aux(glist->group), GFP_ATOMIC);
9c8b35b1
QW		 1195 if (ret < 0)
1196 goto out;
1197 }
1198 }
1199 ret = 0;
1200out:
1201 return ret;
1202}
1203
1204
1205/*
1206 * Quick path for updating qgroup with only excl refs.
1207 *
1208 * In that case, just update all parent will be enough.
1209 * Or we needs to do a full rescan.
1210 * Caller should also hold fs_info->qgroup_lock.
1211 *
1212 * Return 0 for quick update, return >0 for need to full rescan
1213 * and mark INCONSISTENT flag.
1214 * Return < 0 for other error.
1215 */
1216static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1217 struct ulist *tmp, u64 src, u64 dst,
1218 int sign)
1219{
1220 struct btrfs_qgroup *qgroup;
1221 int ret = 1;
1222 int err = 0;
1223
1224 qgroup = find_qgroup_rb(fs_info, src);
1225 if (!qgroup)
1226 goto out;
1227 if (qgroup->excl == qgroup->rfer) {
1228 ret = 0;
1229 err = __qgroup_excl_accounting(fs_info, tmp, dst,
429d6275 1230 qgroup, sign);
9c8b35b1
QW		 1231 if (err < 0) {
1232 ret = err;
1233 goto out;
1234 }
1235 }
1236out:
1237 if (ret)
1238 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1239 return ret;
1240}
1241
9f8a6ce6
LF		 1242int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1243 u64 dst)
bed92eae 1244{
9f8a6ce6 1245 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae 1246 struct btrfs_root *quota_root;
b7fef4f5
WS		 1247 struct btrfs_qgroup *parent;
1248 struct btrfs_qgroup *member;
534e6623 1249 struct btrfs_qgroup_list *list;
9c8b35b1 1250 struct ulist *tmp;
bed92eae
AJ		 1251 int ret = 0;
1252
8465ecec
QW		 1253 /* Check the level of src and dst first */
1254 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1255 return -EINVAL;
1256
6602caf1 1257 tmp = ulist_alloc(GFP_KERNEL);
ab3680dd
CE		 1258 if (!tmp)
1259 return -ENOMEM;
1260
f2f6ed3d 1261 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1262 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1263 if (!quota_root) {
1264 ret = -EINVAL;
1265 goto out;
1266 }
b7fef4f5
WS		 1267 member = find_qgroup_rb(fs_info, src);
1268 parent = find_qgroup_rb(fs_info, dst);
1269 if (!member || !parent) {
1270 ret = -EINVAL;
1271 goto out;
1272 }
bed92eae 1273
534e6623
WS		 1274 /* check if such qgroup relation exist firstly */
1275 list_for_each_entry(list, &member->groups, next_group) {
1276 if (list->group == parent) {
1277 ret = -EEXIST;
1278 goto out;
1279 }
1280 }
1281
711169c4 1282 ret = add_qgroup_relation_item(trans, src, dst);
bed92eae 1283 if (ret)
f2f6ed3d 1284 goto out;
bed92eae 1285
711169c4 1286 ret = add_qgroup_relation_item(trans, dst, src);
bed92eae 1287 if (ret) {
99d7f09a 1288 del_qgroup_relation_item(trans, src, dst);
f2f6ed3d 1289 goto out;
bed92eae
AJ		 1290 }
1291
1292 spin_lock(&fs_info->qgroup_lock);
0b246afa 1293 ret = add_relation_rb(fs_info, src, dst);
9c8b35b1
QW		 1294 if (ret < 0) {
1295 spin_unlock(&fs_info->qgroup_lock);
1296 goto out;
1297 }
1298 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
bed92eae 1299 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1300out:
1301 mutex_unlock(&fs_info->qgroup_ioctl_lock);
9c8b35b1 1302 ulist_free(tmp);
bed92eae
AJ		 1303 return ret;
1304}
1305
6b36f1aa
LF		 1306static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1307 u64 dst)
bed92eae 1308{
6b36f1aa 1309 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae 1310 struct btrfs_root *quota_root;
534e6623
WS		 1311 struct btrfs_qgroup *parent;
1312 struct btrfs_qgroup *member;
1313 struct btrfs_qgroup_list *list;
9c8b35b1 1314 struct ulist *tmp;
bed92eae
AJ		 1315 int ret = 0;
1316 int err;
1317
6602caf1 1318 tmp = ulist_alloc(GFP_KERNEL);
9c8b35b1
QW		 1319 if (!tmp)
1320 return -ENOMEM;
1321
bed92eae 1322 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1323 if (!quota_root) {
1324 ret = -EINVAL;
1325 goto out;
1326 }
bed92eae 1327
534e6623
WS		 1328 member = find_qgroup_rb(fs_info, src);
1329 parent = find_qgroup_rb(fs_info, dst);
1330 if (!member || !parent) {
1331 ret = -EINVAL;
1332 goto out;
1333 }
1334
1335 /* check if such qgroup relation exist firstly */
1336 list_for_each_entry(list, &member->groups, next_group) {
1337 if (list->group == parent)
1338 goto exist;
1339 }
1340 ret = -ENOENT;
1341 goto out;
1342exist:
99d7f09a
LF		 1343 ret = del_qgroup_relation_item(trans, src, dst);
1344 err = del_qgroup_relation_item(trans, dst, src);
bed92eae
AJ		 1345 if (err && !ret)
1346 ret = err;
1347
1348 spin_lock(&fs_info->qgroup_lock);
1349 del_relation_rb(fs_info, src, dst);
9c8b35b1 1350 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
bed92eae 1351 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d 1352out:
9c8b35b1 1353 ulist_free(tmp);
f5a6b1c5
DY		 1354 return ret;
1355}
1356
39616c27
LF		 1357int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1358 u64 dst)
f5a6b1c5 1359{
39616c27 1360 struct btrfs_fs_info *fs_info = trans->fs_info;
f5a6b1c5
DY		 1361 int ret = 0;
1362
1363 mutex_lock(&fs_info->qgroup_ioctl_lock);
6b36f1aa 1364 ret = __del_qgroup_relation(trans, src, dst);
f2f6ed3d 1365 mutex_unlock(&fs_info->qgroup_ioctl_lock);
f5a6b1c5 1366
bed92eae
AJ		 1367 return ret;
1368}
1369
49a05ecd 1370int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
bed92eae 1371{
49a05ecd 1372 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae
AJ		 1373 struct btrfs_root *quota_root;
1374 struct btrfs_qgroup *qgroup;
1375 int ret = 0;
1376
f2f6ed3d 1377 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1378 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1379 if (!quota_root) {
1380 ret = -EINVAL;
1381 goto out;
1382 }
534e6623
WS		 1383 qgroup = find_qgroup_rb(fs_info, qgroupid);
1384 if (qgroup) {
1385 ret = -EEXIST;
1386 goto out;
1387 }
bed92eae
AJ		 1388
1389 ret = add_qgroup_item(trans, quota_root, qgroupid);
534e6623
WS		 1390 if (ret)
1391 goto out;
bed92eae
AJ		 1392
1393 spin_lock(&fs_info->qgroup_lock);
1394 qgroup = add_qgroup_rb(fs_info, qgroupid);
1395 spin_unlock(&fs_info->qgroup_lock);
1396
1397 if (IS_ERR(qgroup))
1398 ret = PTR_ERR(qgroup);
f2f6ed3d
WS		 1399out:
1400 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1401 return ret;
1402}
1403
3efbee1d 1404int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
bed92eae 1405{
3efbee1d 1406 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae 1407 struct btrfs_root *quota_root;
2cf68703 1408 struct btrfs_qgroup *qgroup;
f5a6b1c5 1409 struct btrfs_qgroup_list *list;
bed92eae
AJ		 1410 int ret = 0;
1411
f2f6ed3d 1412 mutex_lock(&fs_info->qgroup_ioctl_lock);
bed92eae 1413 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 1414 if (!quota_root) {
1415 ret = -EINVAL;
1416 goto out;
1417 }
bed92eae 1418
2cf68703 1419 qgroup = find_qgroup_rb(fs_info, qgroupid);
534e6623
WS		 1420 if (!qgroup) {
1421 ret = -ENOENT;
1422 goto out;
2cf68703 1423 }
b90e22ba
LF		 1424
1425 /* Check if there are no children of this qgroup */
1426 if (!list_empty(&qgroup->members)) {
1427 ret = -EBUSY;
1428 goto out;
1429 }
1430
69104618 1431 ret = del_qgroup_item(trans, qgroupid);
36b96fdc
SD		 1432 if (ret && ret != -ENOENT)
1433 goto out;
bed92eae 1434
f5a6b1c5
DY		 1435 while (!list_empty(&qgroup->groups)) {
1436 list = list_first_entry(&qgroup->groups,
1437 struct btrfs_qgroup_list, next_group);
6b36f1aa
LF		 1438 ret = __del_qgroup_relation(trans, qgroupid,
1439 list->group->qgroupid);
f5a6b1c5
DY		 1440 if (ret)
1441 goto out;
1442 }
1443
bed92eae 1444 spin_lock(&fs_info->qgroup_lock);
0b246afa 1445 del_qgroup_rb(fs_info, qgroupid);
bed92eae 1446 spin_unlock(&fs_info->qgroup_lock);
f2f6ed3d
WS		 1447out:
1448 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1449 return ret;
1450}
1451
f0042d5e 1452int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
bed92eae
AJ		 1453 struct btrfs_qgroup_limit *limit)
1454{
f0042d5e 1455 struct btrfs_fs_info *fs_info = trans->fs_info;
f2f6ed3d 1456 struct btrfs_root *quota_root;
bed92eae
AJ		 1457 struct btrfs_qgroup *qgroup;
1458 int ret = 0;
fe759907
YD		 1459 /* Sometimes we would want to clear the limit on this qgroup.
1460 * To meet this requirement, we treat the -1 as a special value
1461 * which tell kernel to clear the limit on this qgroup.
1462 */
1463 const u64 CLEAR_VALUE = -1;
bed92eae 1464
f2f6ed3d
WS		 1465 mutex_lock(&fs_info->qgroup_ioctl_lock);
1466 quota_root = fs_info->quota_root;
1467 if (!quota_root) {
1468 ret = -EINVAL;
1469 goto out;
1470 }
bed92eae 1471
ddb47afa
WS		 1472 qgroup = find_qgroup_rb(fs_info, qgroupid);
1473 if (!qgroup) {
1474 ret = -ENOENT;
1475 goto out;
1476 }
bed92eae 1477
58400fce 1478 spin_lock(&fs_info->qgroup_lock);
fe759907
YD		 1479 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1480 if (limit->max_rfer == CLEAR_VALUE) {
1481 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1482 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1483 qgroup->max_rfer = 0;
1484 } else {
1485 qgroup->max_rfer = limit->max_rfer;
1486 }
1487 }
1488 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1489 if (limit->max_excl == CLEAR_VALUE) {
1490 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1491 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1492 qgroup->max_excl = 0;
1493 } else {
1494 qgroup->max_excl = limit->max_excl;
1495 }
1496 }
1497 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1498 if (limit->rsv_rfer == CLEAR_VALUE) {
1499 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1500 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1501 qgroup->rsv_rfer = 0;
1502 } else {
1503 qgroup->rsv_rfer = limit->rsv_rfer;
1504 }
1505 }
1506 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1507 if (limit->rsv_excl == CLEAR_VALUE) {
1508 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1509 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1510 qgroup->rsv_excl = 0;
1511 } else {
1512 qgroup->rsv_excl = limit->rsv_excl;
1513 }
1514 }
03477d94
DY		 1515 qgroup->lim_flags |= limit->flags;
1516
bed92eae 1517 spin_unlock(&fs_info->qgroup_lock);
1510e71c 1518
ac8a866a 1519 ret = update_qgroup_limit_item(trans, qgroup);
1510e71c
DY		 1520 if (ret) {
1521 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1522 btrfs_info(fs_info, "unable to update quota limit for %llu",
1523 qgroupid);
1524 }
1525
f2f6ed3d
WS		 1526out:
1527 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 1528 return ret;
1529}
1152651a 1530
50b3e040 1531int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
cb93b52c
QW		 1532 struct btrfs_delayed_ref_root *delayed_refs,
1533 struct btrfs_qgroup_extent_record *record)
3368d001
QW		 1534{
1535 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1536 struct rb_node *parent_node = NULL;
1537 struct btrfs_qgroup_extent_record *entry;
1538 u64 bytenr = record->bytenr;
1539
a4666e68 1540 lockdep_assert_held(&delayed_refs->lock);
50b3e040 1541 trace_btrfs_qgroup_trace_extent(fs_info, record);
82bd101b 1542
3368d001
QW		 1543 while (*p) {
1544 parent_node = *p;
1545 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1546 node);
1418bae1 1547 if (bytenr < entry->bytenr) {
3368d001 1548 p = &(*p)->rb_left;
1418bae1 1549 } else if (bytenr > entry->bytenr) {
3368d001 1550 p = &(*p)->rb_right;
1418bae1
QW		 1551 } else {
1552 if (record->data_rsv && !entry->data_rsv) {
1553 entry->data_rsv = record->data_rsv;
1554 entry->data_rsv_refroot =
1555 record->data_rsv_refroot;
1556 }
cb93b52c 1557 return 1;
1418bae1 1558 }
3368d001
QW		 1559 }
1560
1561 rb_link_node(&record->node, parent_node, p);
1562 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
cb93b52c
QW		 1563 return 0;
1564}
1565
fb235dc0
QW		 1566int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1567 struct btrfs_qgroup_extent_record *qrecord)
1568{
1569 struct ulist *old_root;
1570 u64 bytenr = qrecord->bytenr;
1571 int ret;
1572
c995ab3c 1573 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
952bd3db
NB		 1574 if (ret < 0) {
1575 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1576 btrfs_warn(fs_info,
1577"error accounting new delayed refs extent (err code: %d), quota inconsistent",
1578 ret);
1579 return 0;
1580 }
fb235dc0
QW		 1581
1582 /*
1583 * Here we don't need to get the lock of
1584 * trans->transaction->delayed_refs, since inserted qrecord won't
1585 * be deleted, only qrecord->node may be modified (new qrecord insert)
1586 *
1587 * So modifying qrecord->old_roots is safe here
1588 */
1589 qrecord->old_roots = old_root;
1590 return 0;
1591}
1592
a95f3aaf
LF		 1593int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1594 u64 num_bytes, gfp_t gfp_flag)
cb93b52c 1595{
a95f3aaf 1596 struct btrfs_fs_info *fs_info = trans->fs_info;
cb93b52c
QW		 1597 struct btrfs_qgroup_extent_record *record;
1598 struct btrfs_delayed_ref_root *delayed_refs;
1599 int ret;
1600
afcdd129
JB		 1601 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1602 || bytenr == 0 || num_bytes == 0)
cb93b52c 1603 return 0;
1418bae1 1604 record = kzalloc(sizeof(*record), gfp_flag);
cb93b52c
QW		 1605 if (!record)
1606 return -ENOMEM;
1607
1608 delayed_refs = &trans->transaction->delayed_refs;
1609 record->bytenr = bytenr;
1610 record->num_bytes = num_bytes;
1611 record->old_roots = NULL;
1612
1613 spin_lock(&delayed_refs->lock);
2ff7e61e 1614 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
cb93b52c 1615 spin_unlock(&delayed_refs->lock);
fb235dc0 1616 if (ret > 0) {
cb93b52c 1617 kfree(record);
fb235dc0
QW		 1618 return 0;
1619 }
1620 return btrfs_qgroup_trace_extent_post(fs_info, record);
3368d001
QW		 1621}
1622
33d1f05c 1623int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
33d1f05c
QW		 1624 struct extent_buffer *eb)
1625{
8d38d7eb 1626 struct btrfs_fs_info *fs_info = trans->fs_info;
33d1f05c
QW		 1627 int nr = btrfs_header_nritems(eb);
1628 int i, extent_type, ret;
1629 struct btrfs_key key;
1630 struct btrfs_file_extent_item *fi;
1631 u64 bytenr, num_bytes;
1632
1633 /* We can be called directly from walk_up_proc() */
0b246afa 1634 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
33d1f05c
QW		 1635 return 0;
1636
1637 for (i = 0; i < nr; i++) {
1638 btrfs_item_key_to_cpu(eb, &key, i);
1639
1640 if (key.type != BTRFS_EXTENT_DATA_KEY)
1641 continue;
1642
1643 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1644 /* filter out non qgroup-accountable extents */
1645 extent_type = btrfs_file_extent_type(eb, fi);
1646
1647 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1648 continue;
1649
1650 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1651 if (!bytenr)
1652 continue;
1653
1654 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1655
a95f3aaf
LF		 1656 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1657 GFP_NOFS);
33d1f05c
QW		 1658 if (ret)
1659 return ret;
1660 }
cddf3b2c 1661 cond_resched();
33d1f05c
QW		 1662 return 0;
1663}
1664
1665/*
1666 * Walk up the tree from the bottom, freeing leaves and any interior
1667 * nodes which have had all slots visited. If a node (leaf or
1668 * interior) is freed, the node above it will have it's slot
1669 * incremented. The root node will never be freed.
1670 *
1671 * At the end of this function, we should have a path which has all
1672 * slots incremented to the next position for a search. If we need to
1673 * read a new node it will be NULL and the node above it will have the
1674 * correct slot selected for a later read.
1675 *
1676 * If we increment the root nodes slot counter past the number of
1677 * elements, 1 is returned to signal completion of the search.
1678 */
15b34517 1679static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
33d1f05c
QW		 1680{
1681 int level = 0;
1682 int nr, slot;
1683 struct extent_buffer *eb;
1684
1685 if (root_level == 0)
1686 return 1;
1687
1688 while (level <= root_level) {
1689 eb = path->nodes[level];
1690 nr = btrfs_header_nritems(eb);
1691 path->slots[level]++;
1692 slot = path->slots[level];
1693 if (slot >= nr || level == 0) {
1694 /*
1695 * Don't free the root - we will detect this
1696 * condition after our loop and return a
1697 * positive value for caller to stop walking the tree.
1698 */
1699 if (level != root_level) {
1700 btrfs_tree_unlock_rw(eb, path->locks[level]);
1701 path->locks[level] = 0;
1702
1703 free_extent_buffer(eb);
1704 path->nodes[level] = NULL;
1705 path->slots[level] = 0;
1706 }
1707 } else {
1708 /*
1709 * We have a valid slot to walk back down
1710 * from. Stop here so caller can process these
1711 * new nodes.
1712 */
1713 break;
1714 }
1715
1716 level++;
1717 }
1718
1719 eb = path->nodes[root_level];
1720 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1721 return 1;
1722
1723 return 0;
1724}
1725
25982561
QW		 1726/*
1727 * Helper function to trace a subtree tree block swap.
1728 *
1729 * The swap will happen in highest tree block, but there may be a lot of
1730 * tree blocks involved.
1731 *
1732 * For example:
1733 * OO = Old tree blocks
1734 * NN = New tree blocks allocated during balance
1735 *
1736 * File tree (257) Reloc tree for 257
1737 * L2 OO NN
1738 * / \ / \
1739 * L1 OO OO (a) OO NN (a)
1740 * / \ / \ / \ / \
1741 * L0 OO OO OO OO OO OO NN NN
1742 * (b) (c) (b) (c)
1743 *
1744 * When calling qgroup_trace_extent_swap(), we will pass:
1745 * @src_eb = OO(a)
1746 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1747 * @dst_level = 0
1748 * @root_level = 1
1749 *
1750 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1751 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1752 *
1753 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1754 *
1755 * 1) Tree search from @src_eb
1756 * It should acts as a simplified btrfs_search_slot().
1757 * The key for search can be extracted from @dst_path->nodes[dst_level]
1758 * (first key).
1759 *
1760 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1761 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
52042d8e 1762 * They should be marked during previous (@dst_level = 1) iteration.
25982561
QW		 1763 *
1764 * 3) Mark file extents in leaves dirty
1765 * We don't have good way to pick out new file extents only.
1766 * So we still follow the old method by scanning all file extents in
1767 * the leave.
1768 *
52042d8e 1769 * This function can free us from keeping two paths, thus later we only need
25982561
QW		 1770 * to care about how to iterate all new tree blocks in reloc tree.
1771 */
1772static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1773 struct extent_buffer *src_eb,
1774 struct btrfs_path *dst_path,
3d0174f7
QW		 1775 int dst_level, int root_level,
1776 bool trace_leaf)
25982561
QW		 1777{
1778 struct btrfs_key key;
1779 struct btrfs_path *src_path;
1780 struct btrfs_fs_info *fs_info = trans->fs_info;
1781 u32 nodesize = fs_info->nodesize;
1782 int cur_level = root_level;
1783 int ret;
1784
1785 BUG_ON(dst_level > root_level);
1786 /* Level mismatch */
1787 if (btrfs_header_level(src_eb) != root_level)
1788 return -EINVAL;
1789
1790 src_path = btrfs_alloc_path();
1791 if (!src_path) {
1792 ret = -ENOMEM;
1793 goto out;
1794 }
1795
1796 if (dst_level)
1797 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1798 else
1799 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1800
1801 /* For src_path */
1802 extent_buffer_get(src_eb);
1803 src_path->nodes[root_level] = src_eb;
1804 src_path->slots[root_level] = dst_path->slots[root_level];
1805 src_path->locks[root_level] = 0;
1806
1807 /* A simplified version of btrfs_search_slot() */
1808 while (cur_level >= dst_level) {
1809 struct btrfs_key src_key;
1810 struct btrfs_key dst_key;
1811
1812 if (src_path->nodes[cur_level] == NULL) {
1813 struct btrfs_key first_key;
1814 struct extent_buffer *eb;
1815 int parent_slot;
1816 u64 child_gen;
1817 u64 child_bytenr;
1818
1819 eb = src_path->nodes[cur_level + 1];
1820 parent_slot = src_path->slots[cur_level + 1];
1821 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1822 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1823 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1824
1825 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1826 cur_level, &first_key);
1827 if (IS_ERR(eb)) {
1828 ret = PTR_ERR(eb);
1829 goto out;
1830 } else if (!extent_buffer_uptodate(eb)) {
1831 free_extent_buffer(eb);
1832 ret = -EIO;
1833 goto out;
1834 }
1835
1836 src_path->nodes[cur_level] = eb;
1837
1838 btrfs_tree_read_lock(eb);
300aa896 1839 btrfs_set_lock_blocking_read(eb);
25982561
QW		 1840 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1841 }
1842
1843 src_path->slots[cur_level] = dst_path->slots[cur_level];
1844 if (cur_level) {
1845 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1846 &dst_key, dst_path->slots[cur_level]);
1847 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1848 &src_key, src_path->slots[cur_level]);
1849 } else {
1850 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1851 &dst_key, dst_path->slots[cur_level]);
1852 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1853 &src_key, src_path->slots[cur_level]);
1854 }
1855 /* Content mismatch, something went wrong */
1856 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1857 ret = -ENOENT;
1858 goto out;
1859 }
1860 cur_level--;
1861 }
1862
1863 /*
1864 * Now both @dst_path and @src_path have been populated, record the tree
1865 * blocks for qgroup accounting.
1866 */
1867 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1868 nodesize, GFP_NOFS);
1869 if (ret < 0)
1870 goto out;
1871 ret = btrfs_qgroup_trace_extent(trans,
1872 dst_path->nodes[dst_level]->start,
1873 nodesize, GFP_NOFS);
1874 if (ret < 0)
1875 goto out;
1876
1877 /* Record leaf file extents */
3d0174f7 1878 if (dst_level == 0 && trace_leaf) {
25982561
QW		 1879 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1880 if (ret < 0)
1881 goto out;
1882 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1883 }
1884out:
1885 btrfs_free_path(src_path);
1886 return ret;
1887}
1888
ea49f3e7
QW		 1889/*
1890 * Helper function to do recursive generation-aware depth-first search, to
1891 * locate all new tree blocks in a subtree of reloc tree.
1892 *
1893 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1894 * reloc tree
1895 * L2 NN (a)
1896 * / \
1897 * L1 OO NN (b)
1898 * / \ / \
1899 * L0 OO OO OO NN
1900 * (c) (d)
1901 * If we pass:
1902 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1903 * @cur_level = 1
1904 * @root_level = 1
1905 *
1906 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1907 * above tree blocks along with their counter parts in file tree.
52042d8e 1908 * While during search, old tree blocks OO(c) will be skipped as tree block swap
ea49f3e7
QW		 1909 * won't affect OO(c).
1910 */
1911static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1912 struct extent_buffer *src_eb,
1913 struct btrfs_path *dst_path,
1914 int cur_level, int root_level,
3d0174f7 1915 u64 last_snapshot, bool trace_leaf)
ea49f3e7
QW		 1916{
1917 struct btrfs_fs_info *fs_info = trans->fs_info;
1918 struct extent_buffer *eb;
1919 bool need_cleanup = false;
1920 int ret = 0;
1921 int i;
1922
1923 /* Level sanity check */
7ff2c2a1
NB		 1924 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1925 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
ea49f3e7
QW		 1926 root_level < cur_level) {
1927 btrfs_err_rl(fs_info,
1928 "%s: bad levels, cur_level=%d root_level=%d",
1929 __func__, cur_level, root_level);
1930 return -EUCLEAN;
1931 }
1932
1933 /* Read the tree block if needed */
1934 if (dst_path->nodes[cur_level] == NULL) {
1935 struct btrfs_key first_key;
1936 int parent_slot;
1937 u64 child_gen;
1938 u64 child_bytenr;
1939
1940 /*
1941 * dst_path->nodes[root_level] must be initialized before
1942 * calling this function.
1943 */
1944 if (cur_level == root_level) {
1945 btrfs_err_rl(fs_info,
1946 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1947 __func__, root_level, root_level, cur_level);
1948 return -EUCLEAN;
1949 }
1950
1951 /*
1952 * We need to get child blockptr/gen from parent before we can
1953 * read it.
1954 */
1955 eb = dst_path->nodes[cur_level + 1];
1956 parent_slot = dst_path->slots[cur_level + 1];
1957 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1958 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1959 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1960
1961 /* This node is old, no need to trace */
1962 if (child_gen < last_snapshot)
1963 goto out;
1964
1965 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1966 cur_level, &first_key);
1967 if (IS_ERR(eb)) {
1968 ret = PTR_ERR(eb);
1969 goto out;
1970 } else if (!extent_buffer_uptodate(eb)) {
1971 free_extent_buffer(eb);
1972 ret = -EIO;
1973 goto out;
1974 }
1975
1976 dst_path->nodes[cur_level] = eb;
1977 dst_path->slots[cur_level] = 0;
1978
1979 btrfs_tree_read_lock(eb);
300aa896 1980 btrfs_set_lock_blocking_read(eb);
ea49f3e7
QW		 1981 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1982 need_cleanup = true;
1983 }
1984
1985 /* Now record this tree block and its counter part for qgroups */
1986 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
3d0174f7 1987 root_level, trace_leaf);
ea49f3e7
QW		 1988 if (ret < 0)
1989 goto cleanup;
1990
1991 eb = dst_path->nodes[cur_level];
1992
1993 if (cur_level > 0) {
1994 /* Iterate all child tree blocks */
1995 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1996 /* Skip old tree blocks as they won't be swapped */
1997 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1998 continue;
1999 dst_path->slots[cur_level] = i;
2000
2001 /* Recursive call (at most 7 times) */
2002 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2003 dst_path, cur_level - 1, root_level,
3d0174f7 2004 last_snapshot, trace_leaf);
ea49f3e7
QW		 2005 if (ret < 0)
2006 goto cleanup;
2007 }
2008 }
2009
2010cleanup:
2011 if (need_cleanup) {
2012 /* Clean up */
2013 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2014 dst_path->locks[cur_level]);
2015 free_extent_buffer(dst_path->nodes[cur_level]);
2016 dst_path->nodes[cur_level] = NULL;
2017 dst_path->slots[cur_level] = 0;
2018 dst_path->locks[cur_level] = 0;
2019 }
2020out:
2021 return ret;
2022}
2023
5aea1a4f
QW		 2024static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2025 struct extent_buffer *src_eb,
2026 struct extent_buffer *dst_eb,
2027 u64 last_snapshot, bool trace_leaf)
2028{
2029 struct btrfs_fs_info *fs_info = trans->fs_info;
2030 struct btrfs_path *dst_path = NULL;
2031 int level;
2032 int ret;
2033
2034 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2035 return 0;
2036
2037 /* Wrong parameter order */
2038 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2039 btrfs_err_rl(fs_info,
2040 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2041 btrfs_header_generation(src_eb),
2042 btrfs_header_generation(dst_eb));
2043 return -EUCLEAN;
2044 }
2045
2046 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2047 ret = -EIO;
2048 goto out;
2049 }
2050
2051 level = btrfs_header_level(dst_eb);
2052 dst_path = btrfs_alloc_path();
2053 if (!dst_path) {
2054 ret = -ENOMEM;
2055 goto out;
2056 }
2057 /* For dst_path */
2058 extent_buffer_get(dst_eb);
2059 dst_path->nodes[level] = dst_eb;
2060 dst_path->slots[level] = 0;
2061 dst_path->locks[level] = 0;
2062
2063 /* Do the generation aware breadth-first search */
2064 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2065 level, last_snapshot, trace_leaf);
2066 if (ret < 0)
2067 goto out;
2068 ret = 0;
2069
2070out:
2071 btrfs_free_path(dst_path);
2072 if (ret < 0)
2073 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2074 return ret;
2075}
2076
33d1f05c 2077int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
33d1f05c
QW		 2078 struct extent_buffer *root_eb,
2079 u64 root_gen, int root_level)
2080{
deb40627 2081 struct btrfs_fs_info *fs_info = trans->fs_info;
33d1f05c
QW		 2082 int ret = 0;
2083 int level;
2084 struct extent_buffer *eb = root_eb;
2085 struct btrfs_path *path = NULL;
2086
b6e6bca5 2087 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
33d1f05c
QW		 2088 BUG_ON(root_eb == NULL);
2089
0b246afa 2090 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
33d1f05c
QW		 2091 return 0;
2092
2093 if (!extent_buffer_uptodate(root_eb)) {
581c1760 2094 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
33d1f05c
QW		 2095 if (ret)
2096 goto out;
2097 }
2098
2099 if (root_level == 0) {
8d38d7eb 2100 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
33d1f05c
QW		 2101 goto out;
2102 }
2103
2104 path = btrfs_alloc_path();
2105 if (!path)
2106 return -ENOMEM;
2107
2108 /*
2109 * Walk down the tree. Missing extent blocks are filled in as
2110 * we go. Metadata is accounted every time we read a new
2111 * extent block.
2112 *
2113 * When we reach a leaf, we account for file extent items in it,
2114 * walk back up the tree (adjusting slot pointers as we go)
2115 * and restart the search process.
2116 */
2117 extent_buffer_get(root_eb); /* For path */
2118 path->nodes[root_level] = root_eb;
2119 path->slots[root_level] = 0;
2120 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2121walk_down:
2122 level = root_level;
2123 while (level >= 0) {
2124 if (path->nodes[level] == NULL) {
581c1760 2125 struct btrfs_key first_key;
33d1f05c
QW		 2126 int parent_slot;
2127 u64 child_gen;
2128 u64 child_bytenr;
2129
2130 /*
2131 * We need to get child blockptr/gen from parent before
2132 * we can read it.
2133 */
2134 eb = path->nodes[level + 1];
2135 parent_slot = path->slots[level + 1];
2136 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2137 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
581c1760 2138 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
33d1f05c 2139
581c1760
QW		 2140 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2141 level, &first_key);
33d1f05c
QW		 2142 if (IS_ERR(eb)) {
2143 ret = PTR_ERR(eb);
2144 goto out;
2145 } else if (!extent_buffer_uptodate(eb)) {
2146 free_extent_buffer(eb);
2147 ret = -EIO;
2148 goto out;
2149 }
2150
2151 path->nodes[level] = eb;
2152 path->slots[level] = 0;
2153
2154 btrfs_tree_read_lock(eb);
300aa896 2155 btrfs_set_lock_blocking_read(eb);
33d1f05c
QW		 2156 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2157
a95f3aaf 2158 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
0b246afa
JM		 2159 fs_info->nodesize,
2160 GFP_NOFS);
33d1f05c
QW		 2161 if (ret)
2162 goto out;
2163 }
2164
2165 if (level == 0) {
8d38d7eb
LF		 2166 ret = btrfs_qgroup_trace_leaf_items(trans,
2167 path->nodes[level]);
33d1f05c
QW		 2168 if (ret)
2169 goto out;
2170
2171 /* Nonzero return here means we completed our search */
15b34517 2172 ret = adjust_slots_upwards(path, root_level);
33d1f05c
QW		 2173 if (ret)
2174 break;
2175
2176 /* Restart search with new slots */
2177 goto walk_down;
2178 }
2179
2180 level--;
2181 }
2182
2183 ret = 0;
2184out:
2185 btrfs_free_path(path);
2186
2187 return ret;
2188}
2189
d810ef2b
QW		 2190#define UPDATE_NEW 0
2191#define UPDATE_OLD 1
2192/*
2193 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2194 */
2195static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2196 struct ulist *roots, struct ulist *tmp,
2197 struct ulist *qgroups, u64 seq, int update_old)
2198{
2199 struct ulist_node *unode;
2200 struct ulist_iterator uiter;
2201 struct ulist_node *tmp_unode;
2202 struct ulist_iterator tmp_uiter;
2203 struct btrfs_qgroup *qg;
2204 int ret = 0;
2205
2206 if (!roots)
2207 return 0;
2208 ULIST_ITER_INIT(&uiter);
2209 while ((unode = ulist_next(roots, &uiter))) {
2210 qg = find_qgroup_rb(fs_info, unode->val);
2211 if (!qg)
2212 continue;
2213
2214 ulist_reinit(tmp);
ef2fff64 2215 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
d810ef2b
QW		 2216 GFP_ATOMIC);
2217 if (ret < 0)
2218 return ret;
ef2fff64 2219 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
d810ef2b
QW		 2220 if (ret < 0)
2221 return ret;
2222 ULIST_ITER_INIT(&tmp_uiter);
2223 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2224 struct btrfs_qgroup_list *glist;
2225
ef2fff64 2226 qg = unode_aux_to_qgroup(tmp_unode);
d810ef2b
QW		 2227 if (update_old)
2228 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2229 else
2230 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2231 list_for_each_entry(glist, &qg->groups, next_group) {
2232 ret = ulist_add(qgroups, glist->group->qgroupid,
ef2fff64 2233 qgroup_to_aux(glist->group),
d810ef2b
QW		 2234 GFP_ATOMIC);
2235 if (ret < 0)
2236 return ret;
2237 ret = ulist_add(tmp, glist->group->qgroupid,
ef2fff64 2238 qgroup_to_aux(glist->group),
d810ef2b
QW		 2239 GFP_ATOMIC);
2240 if (ret < 0)
2241 return ret;
2242 }
2243 }
2244 }
2245 return 0;
2246}
2247
823ae5b8
QW		 2248/*
2249 * Update qgroup rfer/excl counters.
2250 * Rfer update is easy, codes can explain themselves.
e69bcee3 2251 *
823ae5b8
QW		 2252 * Excl update is tricky, the update is split into 2 part.
2253 * Part 1: Possible exclusive <-> sharing detect:
2254 * | A | !A |
2255 * -------------------------------------
2256 * B | * | - |
2257 * -------------------------------------
2258 * !B | + | ** |
2259 * -------------------------------------
2260 *
2261 * Conditions:
2262 * A: cur_old_roots < nr_old_roots (not exclusive before)
2263 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2264 * B: cur_new_roots < nr_new_roots (not exclusive now)
01327610 2265 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
823ae5b8
QW		 2266 *
2267 * Results:
2268 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2269 * *: Definitely not changed. **: Possible unchanged.
2270 *
2271 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2272 *
2273 * To make the logic clear, we first use condition A and B to split
2274 * combination into 4 results.
2275 *
2276 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2277 * only on variant maybe 0.
2278 *
2279 * Lastly, check result **, since there are 2 variants maybe 0, split them
2280 * again(2x2).
2281 * But this time we don't need to consider other things, the codes and logic
2282 * is easy to understand now.
2283 */
2284static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2285 struct ulist *qgroups,
2286 u64 nr_old_roots,
2287 u64 nr_new_roots,
2288 u64 num_bytes, u64 seq)
2289{
2290 struct ulist_node *unode;
2291 struct ulist_iterator uiter;
2292 struct btrfs_qgroup *qg;
2293 u64 cur_new_count, cur_old_count;
2294
2295 ULIST_ITER_INIT(&uiter);
2296 while ((unode = ulist_next(qgroups, &uiter))) {
2297 bool dirty = false;
2298
ef2fff64 2299 qg = unode_aux_to_qgroup(unode);
823ae5b8
QW		 2300 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2301 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2302
8b317901
QW		 2303 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2304 cur_new_count);
0f5dcf8d 2305
823ae5b8
QW		 2306 /* Rfer update part */
2307 if (cur_old_count == 0 && cur_new_count > 0) {
2308 qg->rfer += num_bytes;
2309 qg->rfer_cmpr += num_bytes;
2310 dirty = true;
2311 }
2312 if (cur_old_count > 0 && cur_new_count == 0) {
2313 qg->rfer -= num_bytes;
2314 qg->rfer_cmpr -= num_bytes;
2315 dirty = true;
2316 }
2317
2318 /* Excl update part */
2319 /* Exclusive/none -> shared case */
2320 if (cur_old_count == nr_old_roots &&
2321 cur_new_count < nr_new_roots) {
2322 /* Exclusive -> shared */
2323 if (cur_old_count != 0) {
2324 qg->excl -= num_bytes;
2325 qg->excl_cmpr -= num_bytes;
2326 dirty = true;
2327 }
2328 }
2329
2330 /* Shared -> exclusive/none case */
2331 if (cur_old_count < nr_old_roots &&
2332 cur_new_count == nr_new_roots) {
2333 /* Shared->exclusive */
2334 if (cur_new_count != 0) {
2335 qg->excl += num_bytes;
2336 qg->excl_cmpr += num_bytes;
2337 dirty = true;
2338 }
2339 }
2340
2341 /* Exclusive/none -> exclusive/none case */
2342 if (cur_old_count == nr_old_roots &&
2343 cur_new_count == nr_new_roots) {
2344 if (cur_old_count == 0) {
2345 /* None -> exclusive/none */
2346
2347 if (cur_new_count != 0) {
2348 /* None -> exclusive */
2349 qg->excl += num_bytes;
2350 qg->excl_cmpr += num_bytes;
2351 dirty = true;
2352 }
2353 /* None -> none, nothing changed */
2354 } else {
2355 /* Exclusive -> exclusive/none */
2356
2357 if (cur_new_count == 0) {
2358 /* Exclusive -> none */
2359 qg->excl -= num_bytes;
2360 qg->excl_cmpr -= num_bytes;
2361 dirty = true;
2362 }
2363 /* Exclusive -> exclusive, nothing changed */
2364 }
2365 }
c05f9429 2366
823ae5b8
QW		 2367 if (dirty)
2368 qgroup_dirty(fs_info, qg);
2369 }
2370 return 0;
2371}
2372
5edfd9fd
QW		 2373/*
2374 * Check if the @roots potentially is a list of fs tree roots
2375 *
2376 * Return 0 for definitely not a fs/subvol tree roots ulist
2377 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2378 * one as well)
2379 */
2380static int maybe_fs_roots(struct ulist *roots)
2381{
2382 struct ulist_node *unode;
2383 struct ulist_iterator uiter;
2384
2385 /* Empty one, still possible for fs roots */
2386 if (!roots || roots->nnodes == 0)
2387 return 1;
2388
2389 ULIST_ITER_INIT(&uiter);
2390 unode = ulist_next(roots, &uiter);
2391 if (!unode)
2392 return 1;
2393
2394 /*
2395 * If it contains fs tree roots, then it must belong to fs/subvol
2396 * trees.
2397 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2398 */
2399 return is_fstree(unode->val);
2400}
2401
8696d760
LF		 2402int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2403 u64 num_bytes, struct ulist *old_roots,
2404 struct ulist *new_roots)
550d7a2e 2405{
8696d760 2406 struct btrfs_fs_info *fs_info = trans->fs_info;
550d7a2e
QW		 2407 struct ulist *qgroups = NULL;
2408 struct ulist *tmp = NULL;
2409 u64 seq;
2410 u64 nr_new_roots = 0;
2411 u64 nr_old_roots = 0;
2412 int ret = 0;
2413
81353d50
DS		 2414 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2415 return 0;
2416
5edfd9fd
QW		 2417 if (new_roots) {
2418 if (!maybe_fs_roots(new_roots))
2419 goto out_free;
550d7a2e 2420 nr_new_roots = new_roots->nnodes;
5edfd9fd
QW		 2421 }
2422 if (old_roots) {
2423 if (!maybe_fs_roots(old_roots))
2424 goto out_free;
550d7a2e 2425 nr_old_roots = old_roots->nnodes;
5edfd9fd
QW		 2426 }
2427
2428 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2429 if (nr_old_roots == 0 && nr_new_roots == 0)
2430 goto out_free;
550d7a2e 2431
550d7a2e
QW		 2432 BUG_ON(!fs_info->quota_root);
2433
c9f6f3cd
QW		 2434 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2435 num_bytes, nr_old_roots, nr_new_roots);
0f5dcf8d 2436
550d7a2e
QW		 2437 qgroups = ulist_alloc(GFP_NOFS);
2438 if (!qgroups) {
2439 ret = -ENOMEM;
2440 goto out_free;
2441 }
2442 tmp = ulist_alloc(GFP_NOFS);
2443 if (!tmp) {
2444 ret = -ENOMEM;
2445 goto out_free;
2446 }
2447
2448 mutex_lock(&fs_info->qgroup_rescan_lock);
2449 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2450 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2451 mutex_unlock(&fs_info->qgroup_rescan_lock);
2452 ret = 0;
2453 goto out_free;
2454 }
2455 }
2456 mutex_unlock(&fs_info->qgroup_rescan_lock);
2457
2458 spin_lock(&fs_info->qgroup_lock);
2459 seq = fs_info->qgroup_seq;
2460
2461 /* Update old refcnts using old_roots */
2462 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2463 UPDATE_OLD);
2464 if (ret < 0)
2465 goto out;
2466
2467 /* Update new refcnts using new_roots */
2468 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2469 UPDATE_NEW);
2470 if (ret < 0)
2471 goto out;
2472
2473 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2474 num_bytes, seq);
2475
2476 /*
2477 * Bump qgroup_seq to avoid seq overlap
2478 */
2479 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2480out:
2481 spin_unlock(&fs_info->qgroup_lock);
2482out_free:
2483 ulist_free(tmp);
2484 ulist_free(qgroups);
2485 ulist_free(old_roots);
2486 ulist_free(new_roots);
2487 return ret;
2488}
2489
460fb20a 2490int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
550d7a2e 2491{
460fb20a 2492 struct btrfs_fs_info *fs_info = trans->fs_info;
550d7a2e
QW		 2493 struct btrfs_qgroup_extent_record *record;
2494 struct btrfs_delayed_ref_root *delayed_refs;
2495 struct ulist *new_roots = NULL;
2496 struct rb_node *node;
c337e7b0 2497 u64 num_dirty_extents = 0;
9086db86 2498 u64 qgroup_to_skip;
550d7a2e
QW		 2499 int ret = 0;
2500
2501 delayed_refs = &trans->transaction->delayed_refs;
9086db86 2502 qgroup_to_skip = delayed_refs->qgroup_to_skip;
550d7a2e
QW		 2503 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2504 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2505 node);
2506
c337e7b0 2507 num_dirty_extents++;
bc074524 2508 trace_btrfs_qgroup_account_extents(fs_info, record);
0f5dcf8d 2509
550d7a2e 2510 if (!ret) {
d1b8b94a
QW		 2511 /*
2512 * Old roots should be searched when inserting qgroup
2513 * extent record
2514 */
2515 if (WARN_ON(!record->old_roots)) {
2516 /* Search commit root to find old_roots */
2517 ret = btrfs_find_all_roots(NULL, fs_info,
2518 record->bytenr, 0,
c995ab3c 2519 &record->old_roots, false);
d1b8b94a
QW		 2520 if (ret < 0)
2521 goto cleanup;
2522 }
2523
1418bae1
QW		 2524 /* Free the reserved data space */
2525 btrfs_qgroup_free_refroot(fs_info,
2526 record->data_rsv_refroot,
2527 record->data_rsv,
2528 BTRFS_QGROUP_RSV_DATA);
550d7a2e 2529 /*
de47c9d3 2530 * Use SEQ_LAST as time_seq to do special search, which
550d7a2e
QW		 2531 * doesn't lock tree or delayed_refs and search current
2532 * root. It's safe inside commit_transaction().
2533 */
2534 ret = btrfs_find_all_roots(trans, fs_info,
c995ab3c 2535 record->bytenr, SEQ_LAST, &new_roots, false);
550d7a2e
QW		 2536 if (ret < 0)
2537 goto cleanup;
d1b8b94a 2538 if (qgroup_to_skip) {
9086db86 2539 ulist_del(new_roots, qgroup_to_skip, 0);
d1b8b94a
QW		 2540 ulist_del(record->old_roots, qgroup_to_skip,
2541 0);
2542 }
8696d760
LF		 2543 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2544 record->num_bytes,
2545 record->old_roots,
2546 new_roots);
550d7a2e
QW		 2547 record->old_roots = NULL;
2548 new_roots = NULL;
2549 }
2550cleanup:
2551 ulist_free(record->old_roots);
2552 ulist_free(new_roots);
2553 new_roots = NULL;
2554 rb_erase(node, &delayed_refs->dirty_extent_root);
2555 kfree(record);
2556
2557 }
c337e7b0
QW		 2558 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2559 num_dirty_extents);
550d7a2e
QW		 2560 return ret;
2561}
2562
bed92eae
AJ		 2563/*
2564 * called from commit_transaction. Writes all changed qgroups to disk.
2565 */
280f8bd2 2566int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
bed92eae 2567{
280f8bd2 2568 struct btrfs_fs_info *fs_info = trans->fs_info;
bed92eae
AJ		 2569 struct btrfs_root *quota_root = fs_info->quota_root;
2570 int ret = 0;
2571
2572 if (!quota_root)
5d23515b 2573 return ret;
bed92eae
AJ		 2574
2575 spin_lock(&fs_info->qgroup_lock);
2576 while (!list_empty(&fs_info->dirty_qgroups)) {
2577 struct btrfs_qgroup *qgroup;
2578 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2579 struct btrfs_qgroup, dirty);
2580 list_del_init(&qgroup->dirty);
2581 spin_unlock(&fs_info->qgroup_lock);
3e07e9a0 2582 ret = update_qgroup_info_item(trans, qgroup);
d3001ed3
DY		 2583 if (ret)
2584 fs_info->qgroup_flags |=
2585 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
ac8a866a 2586 ret = update_qgroup_limit_item(trans, qgroup);
bed92eae
AJ		 2587 if (ret)
2588 fs_info->qgroup_flags |=
2589 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2590 spin_lock(&fs_info->qgroup_lock);
2591 }
afcdd129 2592 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
bed92eae
AJ		 2593 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2594 else
2595 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2596 spin_unlock(&fs_info->qgroup_lock);
2597
2e980acd 2598 ret = update_qgroup_status_item(trans);
bed92eae
AJ		 2599 if (ret)
2600 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2601
bed92eae
AJ		 2602 return ret;
2603}
2604
2605/*
01327610 2606 * Copy the accounting information between qgroups. This is necessary
918c2ee1
MF		 2607 * when a snapshot or a subvolume is created. Throwing an error will
2608 * cause a transaction abort so we take extra care here to only error
2609 * when a readonly fs is a reasonable outcome.
bed92eae 2610 */
a9377422
LF		 2611int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2612 u64 objectid, struct btrfs_qgroup_inherit *inherit)
bed92eae
AJ		 2613{
2614 int ret = 0;
2615 int i;
2616 u64 *i_qgroups;
a9377422 2617 struct btrfs_fs_info *fs_info = trans->fs_info;
552f0329 2618 struct btrfs_root *quota_root;
bed92eae
AJ		 2619 struct btrfs_qgroup *srcgroup;
2620 struct btrfs_qgroup *dstgroup;
2621 u32 level_size = 0;
3f5e2d3b 2622 u64 nums;
bed92eae 2623
f2f6ed3d 2624 mutex_lock(&fs_info->qgroup_ioctl_lock);
afcdd129 2625 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
f2f6ed3d 2626 goto out;
bed92eae 2627
552f0329 2628 quota_root = fs_info->quota_root;
f2f6ed3d
WS		 2629 if (!quota_root) {
2630 ret = -EINVAL;
2631 goto out;
2632 }
bed92eae 2633
3f5e2d3b
WS		 2634 if (inherit) {
2635 i_qgroups = (u64 *)(inherit + 1);
2636 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2637 2 * inherit->num_excl_copies;
2638 for (i = 0; i < nums; ++i) {
2639 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
09870d27 2640
918c2ee1
MF		 2641 /*
2642 * Zero out invalid groups so we can ignore
2643 * them later.
2644 */
2645 if (!srcgroup ||
2646 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2647 *i_qgroups = 0ULL;
2648
3f5e2d3b
WS		 2649 ++i_qgroups;
2650 }
2651 }
2652
bed92eae
AJ		 2653 /*
2654 * create a tracking group for the subvol itself
2655 */
2656 ret = add_qgroup_item(trans, quota_root, objectid);
2657 if (ret)
2658 goto out;
2659
bed92eae
AJ		 2660 /*
2661 * add qgroup to all inherited groups
2662 */
2663 if (inherit) {
2664 i_qgroups = (u64 *)(inherit + 1);
918c2ee1
MF		 2665 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2666 if (*i_qgroups == 0)
2667 continue;
711169c4
LF		 2668 ret = add_qgroup_relation_item(trans, objectid,
2669 *i_qgroups);
918c2ee1 2670 if (ret && ret != -EEXIST)
bed92eae 2671 goto out;
711169c4
LF		 2672 ret = add_qgroup_relation_item(trans, *i_qgroups,
2673 objectid);
918c2ee1 2674 if (ret && ret != -EEXIST)
bed92eae 2675 goto out;
bed92eae 2676 }
918c2ee1 2677 ret = 0;
bed92eae
AJ		 2678 }
2679
2680
2681 spin_lock(&fs_info->qgroup_lock);
2682
2683 dstgroup = add_qgroup_rb(fs_info, objectid);
57a5a882
DC		 2684 if (IS_ERR(dstgroup)) {
2685 ret = PTR_ERR(dstgroup);
bed92eae 2686 goto unlock;
57a5a882 2687 }
bed92eae 2688
e8c8541a 2689 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
e8c8541a
DY		 2690 dstgroup->lim_flags = inherit->lim.flags;
2691 dstgroup->max_rfer = inherit->lim.max_rfer;
2692 dstgroup->max_excl = inherit->lim.max_excl;
2693 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2694 dstgroup->rsv_excl = inherit->lim.rsv_excl;
1510e71c 2695
ac8a866a 2696 ret = update_qgroup_limit_item(trans, dstgroup);
1510e71c
DY		 2697 if (ret) {
2698 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
5d163e0e
JM		 2699 btrfs_info(fs_info,
2700 "unable to update quota limit for %llu",
2701 dstgroup->qgroupid);
1510e71c
DY		 2702 goto unlock;
2703 }
e8c8541a
DY		 2704 }
2705
bed92eae
AJ		 2706 if (srcid) {
2707 srcgroup = find_qgroup_rb(fs_info, srcid);
f3a87f1b 2708 if (!srcgroup)
bed92eae 2709 goto unlock;
fcebe456
JB		 2710
2711 /*
2712 * We call inherit after we clone the root in order to make sure
2713 * our counts don't go crazy, so at this point the only
2714 * difference between the two roots should be the root node.
2715 */
c8389d4c 2716 level_size = fs_info->nodesize;
fcebe456
JB		 2717 dstgroup->rfer = srcgroup->rfer;
2718 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2719 dstgroup->excl = level_size;
2720 dstgroup->excl_cmpr = level_size;
bed92eae
AJ		 2721 srcgroup->excl = level_size;
2722 srcgroup->excl_cmpr = level_size;
3eeb4d59
DY		 2723
2724 /* inherit the limit info */
2725 dstgroup->lim_flags = srcgroup->lim_flags;
2726 dstgroup->max_rfer = srcgroup->max_rfer;
2727 dstgroup->max_excl = srcgroup->max_excl;
2728 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2729 dstgroup->rsv_excl = srcgroup->rsv_excl;
2730
bed92eae
AJ		 2731 qgroup_dirty(fs_info, dstgroup);
2732 qgroup_dirty(fs_info, srcgroup);
2733 }
2734
f3a87f1b 2735 if (!inherit)
bed92eae
AJ		 2736 goto unlock;
2737
2738 i_qgroups = (u64 *)(inherit + 1);
2739 for (i = 0; i < inherit->num_qgroups; ++i) {
918c2ee1 2740 if (*i_qgroups) {
0b246afa 2741 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
918c2ee1
MF		 2742 if (ret)
2743 goto unlock;
2744 }
bed92eae
AJ		 2745 ++i_qgroups;
2746 }
2747
918c2ee1 2748 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
bed92eae
AJ		 2749 struct btrfs_qgroup *src;
2750 struct btrfs_qgroup *dst;
2751
918c2ee1
MF		 2752 if (!i_qgroups[0] || !i_qgroups[1])
2753 continue;
2754
bed92eae
AJ		 2755 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2756 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2757
2758 if (!src || !dst) {
2759 ret = -EINVAL;
2760 goto unlock;
2761 }
2762
2763 dst->rfer = src->rfer - level_size;
2764 dst->rfer_cmpr = src->rfer_cmpr - level_size;
bed92eae 2765 }
918c2ee1 2766 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
bed92eae
AJ		 2767 struct btrfs_qgroup *src;
2768 struct btrfs_qgroup *dst;
2769
918c2ee1
MF		 2770 if (!i_qgroups[0] || !i_qgroups[1])
2771 continue;
2772
bed92eae
AJ		 2773 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2774 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2775
2776 if (!src || !dst) {
2777 ret = -EINVAL;
2778 goto unlock;
2779 }
2780
2781 dst->excl = src->excl + level_size;
2782 dst->excl_cmpr = src->excl_cmpr + level_size;
bed92eae
AJ		 2783 }
2784
2785unlock:
2786 spin_unlock(&fs_info->qgroup_lock);
2787out:
f2f6ed3d 2788 mutex_unlock(&fs_info->qgroup_ioctl_lock);
bed92eae
AJ		 2789 return ret;
2790}
2791
a514d638
QW		 2792/*
2793 * Two limits to commit transaction in advance.
2794 *
f5fef459 2795 * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
a514d638
QW		 2796 * For SIZE, it will be in byte unit as threshold.
2797 */
f5fef459
QW		 2798#define QGROUP_FREE_RATIO 32
2799#define QGROUP_FREE_SIZE SZ_32M
a514d638
QW		 2800static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2801 const struct btrfs_qgroup *qg, u64 num_bytes)
003d7c59 2802{
f5fef459 2803 u64 free;
a514d638
QW		 2804 u64 threshold;
2805
003d7c59 2806 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
dba21324 2807 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
003d7c59
JM		 2808 return false;
2809
2810 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
dba21324 2811 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
003d7c59
JM		 2812 return false;
2813
a514d638
QW		 2814 /*
2815 * Even if we passed the check, it's better to check if reservation
2816 * for meta_pertrans is pushing us near limit.
2817 * If there is too much pertrans reservation or it's near the limit,
2818 * let's try commit transaction to free some, using transaction_kthread
2819 */
2820 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2821 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
f5fef459
QW		 2822 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2823 free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2824 threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2825 QGROUP_FREE_SIZE);
2826 } else {
2827 free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2828 threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2829 QGROUP_FREE_SIZE);
2830 }
a514d638
QW		 2831
2832 /*
2833 * Use transaction_kthread to commit transaction, so we no
2834 * longer need to bother nested transaction nor lock context.
2835 */
f5fef459 2836 if (free < threshold)
a514d638
QW		 2837 btrfs_commit_transaction_locksafe(fs_info);
2838 }
2839
003d7c59
JM		 2840 return true;
2841}
2842
dba21324
QW		 2843static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2844 enum btrfs_qgroup_rsv_type type)
bed92eae
AJ		 2845{
2846 struct btrfs_root *quota_root;
2847 struct btrfs_qgroup *qgroup;
2848 struct btrfs_fs_info *fs_info = root->fs_info;
2849 u64 ref_root = root->root_key.objectid;
2850 int ret = 0;
bed92eae
AJ		 2851 struct ulist_node *unode;
2852 struct ulist_iterator uiter;
2853
2854 if (!is_fstree(ref_root))
2855 return 0;
2856
2857 if (num_bytes == 0)
2858 return 0;
f29efe29
SD		 2859
2860 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2861 capable(CAP_SYS_RESOURCE))
2862 enforce = false;
2863
bed92eae
AJ		 2864 spin_lock(&fs_info->qgroup_lock);
2865 quota_root = fs_info->quota_root;
2866 if (!quota_root)
2867 goto out;
2868
2869 qgroup = find_qgroup_rb(fs_info, ref_root);
2870 if (!qgroup)
2871 goto out;
2872
2873 /*
2874 * in a first step, we check all affected qgroups if any limits would
2875 * be exceeded
2876 */
1e8f9158
WS		 2877 ulist_reinit(fs_info->qgroup_ulist);
2878 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
a1840b50 2879 qgroup_to_aux(qgroup), GFP_ATOMIC);
3c97185c
WS		 2880 if (ret < 0)
2881 goto out;
bed92eae 2882 ULIST_ITER_INIT(&uiter);
1e8f9158 2883 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 2884 struct btrfs_qgroup *qg;
2885 struct btrfs_qgroup_list *glist;
2886
ef2fff64 2887 qg = unode_aux_to_qgroup(unode);
bed92eae 2888
a514d638 2889 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
bed92eae 2890 ret = -EDQUOT;
720f1e20
WS		 2891 goto out;
2892 }
bed92eae
AJ		 2893
2894 list_for_each_entry(glist, &qg->groups, next_group) {
1e8f9158
WS		 2895 ret = ulist_add(fs_info->qgroup_ulist,
2896 glist->group->qgroupid,
a1840b50 2897 qgroup_to_aux(glist->group), GFP_ATOMIC);
3c97185c
WS		 2898 if (ret < 0)
2899 goto out;
bed92eae
AJ		 2900 }
2901 }
3c97185c 2902 ret = 0;
bed92eae
AJ		 2903 /*
2904 * no limits exceeded, now record the reservation into all qgroups
2905 */
2906 ULIST_ITER_INIT(&uiter);
1e8f9158 2907 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 2908 struct btrfs_qgroup *qg;
2909
ef2fff64 2910 qg = unode_aux_to_qgroup(unode);
bed92eae 2911
64ee4e75 2912 qgroup_rsv_add(fs_info, qg, num_bytes, type);
bed92eae
AJ		 2913 }
2914
2915out:
2916 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 2917 return ret;
2918}
2919
e1211d0e
QW		 2920/*
2921 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2922 * qgroup).
2923 *
2924 * Will handle all higher level qgroup too.
2925 *
2926 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2927 * This special case is only used for META_PERTRANS type.
2928 */
297d750b 2929void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
d4e5c920
QW		 2930 u64 ref_root, u64 num_bytes,
2931 enum btrfs_qgroup_rsv_type type)
bed92eae
AJ		 2932{
2933 struct btrfs_root *quota_root;
2934 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 2935 struct ulist_node *unode;
2936 struct ulist_iterator uiter;
3c97185c 2937 int ret = 0;
bed92eae
AJ		 2938
2939 if (!is_fstree(ref_root))
2940 return;
2941
2942 if (num_bytes == 0)
2943 return;
2944
e1211d0e
QW		 2945 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2946 WARN(1, "%s: Invalid type to free", __func__);
2947 return;
2948 }
bed92eae
AJ		 2949 spin_lock(&fs_info->qgroup_lock);
2950
2951 quota_root = fs_info->quota_root;
2952 if (!quota_root)
2953 goto out;
2954
2955 qgroup = find_qgroup_rb(fs_info, ref_root);
2956 if (!qgroup)
2957 goto out;
2958
e1211d0e 2959 if (num_bytes == (u64)-1)
8287475a
QW		 2960 /*
2961 * We're freeing all pertrans rsv, get reserved value from
2962 * level 0 qgroup as real num_bytes to free.
2963 */
e1211d0e
QW		 2964 num_bytes = qgroup->rsv.values[type];
2965
1e8f9158
WS		 2966 ulist_reinit(fs_info->qgroup_ulist);
2967 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
a1840b50 2968 qgroup_to_aux(qgroup), GFP_ATOMIC);
3c97185c
WS		 2969 if (ret < 0)
2970 goto out;
bed92eae 2971 ULIST_ITER_INIT(&uiter);
1e8f9158 2972 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
bed92eae
AJ		 2973 struct btrfs_qgroup *qg;
2974 struct btrfs_qgroup_list *glist;
2975
ef2fff64 2976 qg = unode_aux_to_qgroup(unode);
bed92eae 2977
64ee4e75 2978 qgroup_rsv_release(fs_info, qg, num_bytes, type);
bed92eae
AJ		 2979
2980 list_for_each_entry(glist, &qg->groups, next_group) {
1e8f9158
WS		 2981 ret = ulist_add(fs_info->qgroup_ulist,
2982 glist->group->qgroupid,
a1840b50 2983 qgroup_to_aux(glist->group), GFP_ATOMIC);
3c97185c
WS		 2984 if (ret < 0)
2985 goto out;
bed92eae
AJ		 2986 }
2987 }
2988
2989out:
2990 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 2991}
2992
ff3d27a0
QW		 2993/*
2994 * Check if the leaf is the last leaf. Which means all node pointers
2995 * are at their last position.
2996 */
2997static bool is_last_leaf(struct btrfs_path *path)
2998{
2999 int i;
3000
3001 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3002 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3003 return false;
3004 }
3005 return true;
3006}
3007
2f232036
JS		 3008/*
3009 * returns < 0 on error, 0 when more leafs are to be scanned.
3393168d 3010 * returns 1 when done.
2f232036 3011 */
62088ca7
LF		 3012static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3013 struct btrfs_path *path)
2f232036 3014{
62088ca7 3015 struct btrfs_fs_info *fs_info = trans->fs_info;
2f232036 3016 struct btrfs_key found;
0a0e8b89 3017 struct extent_buffer *scratch_leaf = NULL;
2f232036 3018 struct ulist *roots = NULL;
fcebe456 3019 u64 num_bytes;
ff3d27a0 3020 bool done;
2f232036
JS		 3021 int slot;
3022 int ret;
3023
2f232036
JS		 3024 mutex_lock(&fs_info->qgroup_rescan_lock);
3025 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3026 &fs_info->qgroup_rescan_progress,
3027 path, 1, 0);
3028
ab8d0fc4
JM		 3029 btrfs_debug(fs_info,
3030 "current progress key (%llu %u %llu), search_slot ret %d",
3031 fs_info->qgroup_rescan_progress.objectid,
3032 fs_info->qgroup_rescan_progress.type,
3033 fs_info->qgroup_rescan_progress.offset, ret);
2f232036
JS		 3034
3035 if (ret) {
3036 /*
3037 * The rescan is about to end, we will not be scanning any
3038 * further blocks. We cannot unset the RESCAN flag here, because
3039 * we want to commit the transaction if everything went well.
3040 * To make the live accounting work in this phase, we set our
3041 * scan progress pointer such that every real extent objectid
3042 * will be smaller.
3043 */
3044 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3045 btrfs_release_path(path);
3046 mutex_unlock(&fs_info->qgroup_rescan_lock);
3047 return ret;
3048 }
ff3d27a0 3049 done = is_last_leaf(path);
2f232036
JS		 3050
3051 btrfs_item_key_to_cpu(path->nodes[0], &found,
3052 btrfs_header_nritems(path->nodes[0]) - 1);
3053 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3054
0a0e8b89
QW		 3055 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3056 if (!scratch_leaf) {
3057 ret = -ENOMEM;
3058 mutex_unlock(&fs_info->qgroup_rescan_lock);
3059 goto out;
3060 }
2f232036
JS		 3061 slot = path->slots[0];
3062 btrfs_release_path(path);
3063 mutex_unlock(&fs_info->qgroup_rescan_lock);
3064
3065 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3066 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3a6d75e8
JB		 3067 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3068 found.type != BTRFS_METADATA_ITEM_KEY)
2f232036 3069 continue;
3a6d75e8 3070 if (found.type == BTRFS_METADATA_ITEM_KEY)
da17066c 3071 num_bytes = fs_info->nodesize;
3a6d75e8
JB		 3072 else
3073 num_bytes = found.offset;
3074
fcebe456 3075 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
c995ab3c 3076 &roots, false);
2f232036
JS		 3077 if (ret < 0)
3078 goto out;
9d220c95 3079 /* For rescan, just pass old_roots as NULL */
8696d760
LF		 3080 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3081 num_bytes, NULL, roots);
9d220c95 3082 if (ret < 0)
fcebe456 3083 goto out;
2f232036 3084 }
2f232036 3085out:
df449714 3086 if (scratch_leaf)
0a0e8b89 3087 free_extent_buffer(scratch_leaf);
2f232036 3088
6f7de19e 3089 if (done && !ret) {
ff3d27a0 3090 ret = 1;
6f7de19e
QW		 3091 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3092 }
2f232036
JS		 3093 return ret;
3094}
3095
d458b054 3096static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2f232036 3097{
b382a324
JS		 3098 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3099 qgroup_rescan_work);
2f232036
JS		 3100 struct btrfs_path *path;
3101 struct btrfs_trans_handle *trans = NULL;
2f232036 3102 int err = -ENOMEM;
53b7cde9 3103 int ret = 0;
2f232036
JS		 3104
3105 path = btrfs_alloc_path();
3106 if (!path)
3107 goto out;
b6debf15
QW		 3108 /*
3109 * Rescan should only search for commit root, and any later difference
3110 * should be recorded by qgroup
3111 */
3112 path->search_commit_root = 1;
3113 path->skip_locking = 1;
2f232036
JS		 3114
3115 err = 0;
7343dd61 3116 while (!err && !btrfs_fs_closing(fs_info)) {
2f232036
JS		 3117 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3118 if (IS_ERR(trans)) {
3119 err = PTR_ERR(trans);
3120 break;
3121 }
afcdd129 3122 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2f232036
JS		 3123 err = -EINTR;
3124 } else {
62088ca7 3125 err = qgroup_rescan_leaf(trans, path);
2f232036
JS		 3126 }
3127 if (err > 0)
3a45bb20 3128 btrfs_commit_transaction(trans);
2f232036 3129 else
3a45bb20 3130 btrfs_end_transaction(trans);
2f232036
JS		 3131 }
3132
3133out:
2f232036 3134 btrfs_free_path(path);
2f232036
JS		 3135
3136 mutex_lock(&fs_info->qgroup_rescan_lock);
7343dd61
JM		 3137 if (!btrfs_fs_closing(fs_info))
3138 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2f232036 3139
3393168d 3140 if (err > 0 &&
2f232036
JS		 3141 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3142 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3143 } else if (err < 0) {
3144 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3145 }
3146 mutex_unlock(&fs_info->qgroup_rescan_lock);
3147
53b7cde9 3148 /*
01327610 3149 * only update status, since the previous part has already updated the
53b7cde9
QW		 3150 * qgroup info.
3151 */
3152 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3153 if (IS_ERR(trans)) {
3154 err = PTR_ERR(trans);
3155 btrfs_err(fs_info,
913e1535 3156 "fail to start transaction for status update: %d",
53b7cde9
QW		 3157 err);
3158 goto done;
3159 }
2e980acd 3160 ret = update_qgroup_status_item(trans);
53b7cde9
QW		 3161 if (ret < 0) {
3162 err = ret;
ab8d0fc4 3163 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
53b7cde9 3164 }
3a45bb20 3165 btrfs_end_transaction(trans);
53b7cde9 3166
7343dd61
JM		 3167 if (btrfs_fs_closing(fs_info)) {
3168 btrfs_info(fs_info, "qgroup scan paused");
3169 } else if (err >= 0) {
efe120a0 3170 btrfs_info(fs_info, "qgroup scan completed%s",
3393168d 3171 err > 0 ? " (inconsistency flag cleared)" : "");
2f232036 3172 } else {
efe120a0 3173 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2f232036 3174 }
57254b6e 3175
53b7cde9 3176done:
d2c609b8
JM		 3177 mutex_lock(&fs_info->qgroup_rescan_lock);
3178 fs_info->qgroup_rescan_running = false;
3179 mutex_unlock(&fs_info->qgroup_rescan_lock);
57254b6e 3180 complete_all(&fs_info->qgroup_rescan_completion);
2f232036
JS		 3181}
3182
b382a324
JS		 3183/*
3184 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3185 * memory required for the rescan context.
3186 */
3187static int
3188qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3189 int init_flags)
2f232036
JS		 3190{
3191 int ret = 0;
2f232036 3192
9593bf49
QW		 3193 if (!init_flags) {
3194 /* we're resuming qgroup rescan at mount time */
e4e7ede7
FM		 3195 if (!(fs_info->qgroup_flags &
3196 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
9593bf49
QW		 3197 btrfs_warn(fs_info,
3198 "qgroup rescan init failed, qgroup is not enabled");
e4e7ede7
FM		 3199 ret = -EINVAL;
3200 } else if (!(fs_info->qgroup_flags &
3201 BTRFS_QGROUP_STATUS_FLAG_ON)) {
9593bf49
QW		 3202 btrfs_warn(fs_info,
3203 "qgroup rescan init failed, qgroup rescan is not queued");
e4e7ede7
FM		 3204 ret = -EINVAL;
3205 }
3206
3207 if (ret)
3208 return ret;
b382a324 3209 }
2f232036
JS		 3210
3211 mutex_lock(&fs_info->qgroup_rescan_lock);
3212 spin_lock(&fs_info->qgroup_lock);
b382a324
JS		 3213
3214 if (init_flags) {
9593bf49
QW		 3215 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3216 btrfs_warn(fs_info,
3217 "qgroup rescan is already in progress");
b382a324 3218 ret = -EINPROGRESS;
9593bf49
QW		 3219 } else if (!(fs_info->qgroup_flags &
3220 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3221 btrfs_warn(fs_info,
3222 "qgroup rescan init failed, qgroup is not enabled");
b382a324 3223 ret = -EINVAL;
9593bf49 3224 }
b382a324
JS		 3225
3226 if (ret) {
3227 spin_unlock(&fs_info->qgroup_lock);
3228 mutex_unlock(&fs_info->qgroup_rescan_lock);
9593bf49 3229 return ret;
b382a324 3230 }
b382a324 3231 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2f232036
JS		 3232 }
3233
2f232036
JS		 3234 memset(&fs_info->qgroup_rescan_progress, 0,
3235 sizeof(fs_info->qgroup_rescan_progress));
b382a324 3236 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
190631f1 3237 init_completion(&fs_info->qgroup_rescan_completion);
8d9eddad 3238 fs_info->qgroup_rescan_running = true;
b382a324
JS		 3239
3240 spin_unlock(&fs_info->qgroup_lock);
3241 mutex_unlock(&fs_info->qgroup_rescan_lock);
3242
b382a324
JS		 3243 memset(&fs_info->qgroup_rescan_work, 0,
3244 sizeof(fs_info->qgroup_rescan_work));
fc97fab0 3245 btrfs_init_work(&fs_info->qgroup_rescan_work,
9e0af237 3246 btrfs_qgroup_rescan_helper,
fc97fab0 3247 btrfs_qgroup_rescan_worker, NULL, NULL);
b382a324
JS		 3248 return 0;
3249}
3250
3251static void
3252qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3253{
3254 struct rb_node *n;
3255 struct btrfs_qgroup *qgroup;
3256
3257 spin_lock(&fs_info->qgroup_lock);
2f232036
JS		 3258 /* clear all current qgroup tracking information */
3259 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3260 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3261 qgroup->rfer = 0;
3262 qgroup->rfer_cmpr = 0;
3263 qgroup->excl = 0;
3264 qgroup->excl_cmpr = 0;
9c7b0c2e 3265 qgroup_dirty(fs_info, qgroup);
2f232036
JS		 3266 }
3267 spin_unlock(&fs_info->qgroup_lock);
b382a324 3268}
2f232036 3269
b382a324
JS		 3270int
3271btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3272{
3273 int ret = 0;
3274 struct btrfs_trans_handle *trans;
3275
3276 ret = qgroup_rescan_init(fs_info, 0, 1);
3277 if (ret)
3278 return ret;
3279
3280 /*
3281 * We have set the rescan_progress to 0, which means no more
3282 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3283 * However, btrfs_qgroup_account_ref may be right after its call
3284 * to btrfs_find_all_roots, in which case it would still do the
3285 * accounting.
3286 * To solve this, we're committing the transaction, which will
3287 * ensure we run all delayed refs and only after that, we are
3288 * going to clear all tracking information for a clean start.
3289 */
3290
3291 trans = btrfs_join_transaction(fs_info->fs_root);
3292 if (IS_ERR(trans)) {
3293 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3294 return PTR_ERR(trans);
3295 }
3a45bb20 3296 ret = btrfs_commit_transaction(trans);
b382a324
JS		 3297 if (ret) {
3298 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3299 return ret;
3300 }
3301
3302 qgroup_rescan_zero_tracking(fs_info);
3303
fc97fab0
QW		 3304 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3305 &fs_info->qgroup_rescan_work);
2f232036
JS		 3306
3307 return 0;
3308}
57254b6e 3309
d06f23d6
JM		 3310int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3311 bool interruptible)
57254b6e
JS		 3312{
3313 int running;
3314 int ret = 0;
3315
3316 mutex_lock(&fs_info->qgroup_rescan_lock);
3317 spin_lock(&fs_info->qgroup_lock);
d2c609b8 3318 running = fs_info->qgroup_rescan_running;
57254b6e
JS		 3319 spin_unlock(&fs_info->qgroup_lock);
3320 mutex_unlock(&fs_info->qgroup_rescan_lock);
3321
d06f23d6
JM		 3322 if (!running)
3323 return 0;
3324
3325 if (interruptible)
57254b6e
JS		 3326 ret = wait_for_completion_interruptible(
3327 &fs_info->qgroup_rescan_completion);
d06f23d6
JM		 3328 else
3329 wait_for_completion(&fs_info->qgroup_rescan_completion);
57254b6e
JS		 3330
3331 return ret;
3332}
b382a324
JS		 3333
3334/*
3335 * this is only called from open_ctree where we're still single threaded, thus
3336 * locking is omitted here.
3337 */
3338void
3339btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3340{
3341 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
fc97fab0
QW		 3342 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3343 &fs_info->qgroup_rescan_work);
b382a324 3344}
52472553
QW		 3345
3346/*
3347 * Reserve qgroup space for range [start, start + len).
3348 *
3349 * This function will either reserve space from related qgroups or doing
3350 * nothing if the range is already reserved.
3351 *
3352 * Return 0 for successful reserve
3353 * Return <0 for error (including -EQUOT)
3354 *
3355 * NOTE: this function may sleep for memory allocation.
364ecf36
QW		 3356 * if btrfs_qgroup_reserve_data() is called multiple times with
3357 * same @reserved, caller must ensure when error happens it's OK
3358 * to free *ALL* reserved space.
52472553 3359 */
364ecf36
QW		 3360int btrfs_qgroup_reserve_data(struct inode *inode,
3361 struct extent_changeset **reserved_ret, u64 start,
3362 u64 len)
52472553
QW		 3363{
3364 struct btrfs_root *root = BTRFS_I(inode)->root;
52472553
QW		 3365 struct ulist_node *unode;
3366 struct ulist_iterator uiter;
364ecf36
QW		 3367 struct extent_changeset *reserved;
3368 u64 orig_reserved;
3369 u64 to_reserve;
52472553
QW		 3370 int ret;
3371
afcdd129 3372 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
4fd786e6 3373 !is_fstree(root->root_key.objectid) || len == 0)
52472553
QW		 3374 return 0;
3375
364ecf36
QW		 3376 /* @reserved parameter is mandatory for qgroup */
3377 if (WARN_ON(!reserved_ret))
3378 return -EINVAL;
3379 if (!*reserved_ret) {
3380 *reserved_ret = extent_changeset_alloc();
3381 if (!*reserved_ret)
3382 return -ENOMEM;
3383 }
3384 reserved = *reserved_ret;
3385 /* Record already reserved space */
3386 orig_reserved = reserved->bytes_changed;
52472553 3387 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
364ecf36
QW		 3388 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3389
3390 /* Newly reserved space */
3391 to_reserve = reserved->bytes_changed - orig_reserved;
81fb6f77 3392 trace_btrfs_qgroup_reserve_data(inode, start, len,
364ecf36 3393 to_reserve, QGROUP_RESERVE);
52472553
QW		 3394 if (ret < 0)
3395 goto cleanup;
dba21324 3396 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
52472553
QW		 3397 if (ret < 0)
3398 goto cleanup;
3399
52472553
QW		 3400 return ret;
3401
3402cleanup:
364ecf36 3403 /* cleanup *ALL* already reserved ranges */
52472553 3404 ULIST_ITER_INIT(&uiter);
364ecf36 3405 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
52472553 3406 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
ae0f1625 3407 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
364ecf36 3408 extent_changeset_release(reserved);
52472553
QW		 3409 return ret;
3410}
f695fdce 3411
bc42bda2
QW		 3412/* Free ranges specified by @reserved, normally in error path */
3413static int qgroup_free_reserved_data(struct inode *inode,
3414 struct extent_changeset *reserved, u64 start, u64 len)
3415{
3416 struct btrfs_root *root = BTRFS_I(inode)->root;
3417 struct ulist_node *unode;
3418 struct ulist_iterator uiter;
3419 struct extent_changeset changeset;
3420 int freed = 0;
3421 int ret;
3422
3423 extent_changeset_init(&changeset);
3424 len = round_up(start + len, root->fs_info->sectorsize);
3425 start = round_down(start, root->fs_info->sectorsize);
3426
3427 ULIST_ITER_INIT(&uiter);
3428 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3429 u64 range_start = unode->val;
3430 /* unode->aux is the inclusive end */
3431 u64 range_len = unode->aux - range_start + 1;
3432 u64 free_start;
3433 u64 free_len;
3434
3435 extent_changeset_release(&changeset);
3436
3437 /* Only free range in range [start, start + len) */
3438 if (range_start >= start + len ||
3439 range_start + range_len <= start)
3440 continue;
3441 free_start = max(range_start, start);
3442 free_len = min(start + len, range_start + range_len) -
3443 free_start;
3444 /*
3445 * TODO: To also modify reserved->ranges_reserved to reflect
3446 * the modification.
3447 *
3448 * However as long as we free qgroup reserved according to
3449 * EXTENT_QGROUP_RESERVED, we won't double free.
3450 * So not need to rush.
3451 */
3452 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3453 free_start, free_start + free_len - 1,
3454 EXTENT_QGROUP_RESERVED, &changeset);
3455 if (ret < 0)
3456 goto out;
3457 freed += changeset.bytes_changed;
3458 }
4fd786e6 3459 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
d4e5c920 3460 BTRFS_QGROUP_RSV_DATA);
bc42bda2
QW		 3461 ret = freed;
3462out:
3463 extent_changeset_release(&changeset);
3464 return ret;
3465}
3466
3467static int __btrfs_qgroup_release_data(struct inode *inode,
3468 struct extent_changeset *reserved, u64 start, u64 len,
3469 int free)
f695fdce
QW		 3470{
3471 struct extent_changeset changeset;
81fb6f77 3472 int trace_op = QGROUP_RELEASE;
f695fdce
QW		 3473 int ret;
3474
3628b4ca
QW		 3475 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3476 &BTRFS_I(inode)->root->fs_info->flags))
3477 return 0;
3478
bc42bda2
QW		 3479 /* In release case, we shouldn't have @reserved */
3480 WARN_ON(!free && reserved);
3481 if (free && reserved)
3482 return qgroup_free_reserved_data(inode, reserved, start, len);
364ecf36 3483 extent_changeset_init(&changeset);
f695fdce 3484 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
f734c44a 3485 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
f695fdce
QW		 3486 if (ret < 0)
3487 goto out;
3488
d51ea5dd 3489 if (free)
81fb6f77 3490 trace_op = QGROUP_FREE;
81fb6f77
QW		 3491 trace_btrfs_qgroup_release_data(inode, start, len,
3492 changeset.bytes_changed, trace_op);
d51ea5dd
QW		 3493 if (free)
3494 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
4fd786e6 3495 BTRFS_I(inode)->root->root_key.objectid,
d4e5c920 3496 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
7bc329c1 3497 ret = changeset.bytes_changed;
f695fdce 3498out:
364ecf36 3499 extent_changeset_release(&changeset);
f695fdce
QW		 3500 return ret;
3501}
3502
3503/*
3504 * Free a reserved space range from io_tree and related qgroups
3505 *
3506 * Should be called when a range of pages get invalidated before reaching disk.
3507 * Or for error cleanup case.
bc42bda2
QW		 3508 * if @reserved is given, only reserved range in [@start, @start + @len) will
3509 * be freed.
f695fdce
QW		 3510 *
3511 * For data written to disk, use btrfs_qgroup_release_data().
3512 *
3513 * NOTE: This function may sleep for memory allocation.
3514 */
bc42bda2
QW		 3515int btrfs_qgroup_free_data(struct inode *inode,
3516 struct extent_changeset *reserved, u64 start, u64 len)
f695fdce 3517{
bc42bda2 3518 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
f695fdce
QW		 3519}
3520
3521/*
3522 * Release a reserved space range from io_tree only.
3523 *
3524 * Should be called when a range of pages get written to disk and corresponding
3525 * FILE_EXTENT is inserted into corresponding root.
3526 *
3527 * Since new qgroup accounting framework will only update qgroup numbers at
3528 * commit_transaction() time, its reserved space shouldn't be freed from
3529 * related qgroups.
3530 *
3531 * But we should release the range from io_tree, to allow further write to be
3532 * COWed.
3533 *
3534 * NOTE: This function may sleep for memory allocation.
3535 */
3536int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3537{
bc42bda2 3538 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
f695fdce 3539}
55eeaf05 3540
8287475a
QW		 3541static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3542 enum btrfs_qgroup_rsv_type type)
3543{
3544 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3545 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3546 return;
3547 if (num_bytes == 0)
3548 return;
3549
3550 spin_lock(&root->qgroup_meta_rsv_lock);
3551 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3552 root->qgroup_meta_rsv_prealloc += num_bytes;
3553 else
3554 root->qgroup_meta_rsv_pertrans += num_bytes;
3555 spin_unlock(&root->qgroup_meta_rsv_lock);
3556}
3557
3558static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3559 enum btrfs_qgroup_rsv_type type)
3560{
3561 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3562 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3563 return 0;
3564 if (num_bytes == 0)
3565 return 0;
3566
3567 spin_lock(&root->qgroup_meta_rsv_lock);
3568 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3569 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3570 num_bytes);
3571 root->qgroup_meta_rsv_prealloc -= num_bytes;
3572 } else {
3573 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3574 num_bytes);
3575 root->qgroup_meta_rsv_pertrans -= num_bytes;
3576 }
3577 spin_unlock(&root->qgroup_meta_rsv_lock);
3578 return num_bytes;
3579}
3580
733e03a0
QW		 3581int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3582 enum btrfs_qgroup_rsv_type type, bool enforce)
55eeaf05 3583{
0b246afa 3584 struct btrfs_fs_info *fs_info = root->fs_info;
55eeaf05
QW		 3585 int ret;
3586
0b246afa 3587 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3588 !is_fstree(root->root_key.objectid) || num_bytes == 0)
55eeaf05
QW		 3589 return 0;
3590
0b246afa 3591 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4ee0d883 3592 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
733e03a0 3593 ret = qgroup_reserve(root, num_bytes, enforce, type);
55eeaf05
QW		 3594 if (ret < 0)
3595 return ret;
8287475a
QW		 3596 /*
3597 * Record what we have reserved into root.
3598 *
3599 * To avoid quota disabled->enabled underflow.
3600 * In that case, we may try to free space we haven't reserved
3601 * (since quota was disabled), so record what we reserved into root.
3602 * And ensure later release won't underflow this number.
3603 */
3604 add_root_meta_rsv(root, num_bytes, type);
55eeaf05
QW		 3605 return ret;
3606}
3607
733e03a0 3608void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
55eeaf05 3609{
0b246afa 3610 struct btrfs_fs_info *fs_info = root->fs_info;
55eeaf05 3611
0b246afa 3612 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3613 !is_fstree(root->root_key.objectid))
55eeaf05
QW		 3614 return;
3615
e1211d0e 3616 /* TODO: Update trace point to handle such free */
4ee0d883 3617 trace_qgroup_meta_free_all_pertrans(root);
e1211d0e 3618 /* Special value -1 means to free all reserved space */
4fd786e6 3619 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
733e03a0 3620 BTRFS_QGROUP_RSV_META_PERTRANS);
55eeaf05
QW		 3621}
3622
733e03a0
QW		 3623void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3624 enum btrfs_qgroup_rsv_type type)
55eeaf05 3625{
0b246afa
JM		 3626 struct btrfs_fs_info *fs_info = root->fs_info;
3627
3628 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3629 !is_fstree(root->root_key.objectid))
55eeaf05
QW		 3630 return;
3631
8287475a
QW		 3632 /*
3633 * reservation for META_PREALLOC can happen before quota is enabled,
3634 * which can lead to underflow.
3635 * Here ensure we will only free what we really have reserved.
3636 */
3637 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
0b246afa 3638 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4ee0d883 3639 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
4fd786e6
MT		 3640 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3641 num_bytes, type);
55eeaf05 3642}
56fa9d07 3643
64cfaef6
QW		 3644static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3645 int num_bytes)
3646{
3647 struct btrfs_root *quota_root = fs_info->quota_root;
3648 struct btrfs_qgroup *qgroup;
3649 struct ulist_node *unode;
3650 struct ulist_iterator uiter;
3651 int ret = 0;
3652
3653 if (num_bytes == 0)
3654 return;
3655 if (!quota_root)
3656 return;
3657
3658 spin_lock(&fs_info->qgroup_lock);
3659 qgroup = find_qgroup_rb(fs_info, ref_root);
3660 if (!qgroup)
3661 goto out;
3662 ulist_reinit(fs_info->qgroup_ulist);
3663 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
a1840b50 3664 qgroup_to_aux(qgroup), GFP_ATOMIC);
64cfaef6
QW		 3665 if (ret < 0)
3666 goto out;
3667 ULIST_ITER_INIT(&uiter);
3668 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3669 struct btrfs_qgroup *qg;
3670 struct btrfs_qgroup_list *glist;
3671
3672 qg = unode_aux_to_qgroup(unode);
3673
3674 qgroup_rsv_release(fs_info, qg, num_bytes,
3675 BTRFS_QGROUP_RSV_META_PREALLOC);
3676 qgroup_rsv_add(fs_info, qg, num_bytes,
3677 BTRFS_QGROUP_RSV_META_PERTRANS);
3678 list_for_each_entry(glist, &qg->groups, next_group) {
3679 ret = ulist_add(fs_info->qgroup_ulist,
3680 glist->group->qgroupid,
a1840b50 3681 qgroup_to_aux(glist->group), GFP_ATOMIC);
64cfaef6
QW		 3682 if (ret < 0)
3683 goto out;
3684 }
3685 }
3686out:
3687 spin_unlock(&fs_info->qgroup_lock);
3688}
3689
3690void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3691{
3692 struct btrfs_fs_info *fs_info = root->fs_info;
3693
3694 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4fd786e6 3695 !is_fstree(root->root_key.objectid))
64cfaef6 3696 return;
8287475a
QW		 3697 /* Same as btrfs_qgroup_free_meta_prealloc() */
3698 num_bytes = sub_root_meta_rsv(root, num_bytes,
3699 BTRFS_QGROUP_RSV_META_PREALLOC);
4ee0d883 3700 trace_qgroup_meta_convert(root, num_bytes);
4fd786e6 3701 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
64cfaef6
QW		 3702}
3703
56fa9d07 3704/*
01327610 3705 * Check qgroup reserved space leaking, normally at destroy inode
56fa9d07
QW		 3706 * time
3707 */
3708void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3709{
3710 struct extent_changeset changeset;
3711 struct ulist_node *unode;
3712 struct ulist_iterator iter;
3713 int ret;
3714
364ecf36 3715 extent_changeset_init(&changeset);
56fa9d07 3716 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
f734c44a 3717 EXTENT_QGROUP_RESERVED, &changeset);
56fa9d07
QW		 3718
3719 WARN_ON(ret < 0);
3720 if (WARN_ON(changeset.bytes_changed)) {
3721 ULIST_ITER_INIT(&iter);
53d32359 3722 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
56fa9d07
QW		 3723 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3724 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3725 inode->i_ino, unode->val, unode->aux);
3726 }
0b08e1f4 3727 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
4fd786e6 3728 BTRFS_I(inode)->root->root_key.objectid,
d4e5c920 3729 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
0b08e1f4 3730
56fa9d07 3731 }
364ecf36 3732 extent_changeset_release(&changeset);
56fa9d07 3733}
370a11b8
QW		 3734
3735void btrfs_qgroup_init_swapped_blocks(
3736 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3737{
3738 int i;
3739
3740 spin_lock_init(&swapped_blocks->lock);
3741 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3742 swapped_blocks->blocks[i] = RB_ROOT;
3743 swapped_blocks->swapped = false;
3744}
3745
3746/*
3747 * Delete all swapped blocks record of @root.
3748 * Every record here means we skipped a full subtree scan for qgroup.
3749 *
3750 * Gets called when committing one transaction.
3751 */
3752void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3753{
3754 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3755 int i;
3756
3757 swapped_blocks = &root->swapped_blocks;
3758
3759 spin_lock(&swapped_blocks->lock);
3760 if (!swapped_blocks->swapped)
3761 goto out;
3762 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3763 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3764 struct btrfs_qgroup_swapped_block *entry;
3765 struct btrfs_qgroup_swapped_block *next;
3766
3767 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3768 node)
3769 kfree(entry);
3770 swapped_blocks->blocks[i] = RB_ROOT;
3771 }
3772 swapped_blocks->swapped = false;
3773out:
3774 spin_unlock(&swapped_blocks->lock);
3775}
3776
3777/*
3778 * Add subtree roots record into @subvol_root.
3779 *
3780 * @subvol_root: tree root of the subvolume tree get swapped
3781 * @bg: block group under balance
3782 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3783 * @reloc_parent/slot: pointer to the subtree root in reloc tree
3784 * BOTH POINTERS ARE BEFORE TREE SWAP
3785 * @last_snapshot: last snapshot generation of the subvolume tree
3786 */
3787int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3788 struct btrfs_root *subvol_root,
3789 struct btrfs_block_group_cache *bg,
3790 struct extent_buffer *subvol_parent, int subvol_slot,
3791 struct extent_buffer *reloc_parent, int reloc_slot,
3792 u64 last_snapshot)
3793{
3794 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3795 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3796 struct btrfs_qgroup_swapped_block *block;
3797 struct rb_node **cur;
3798 struct rb_node *parent = NULL;
3799 int level = btrfs_header_level(subvol_parent) - 1;
3800 int ret = 0;
3801
3802 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3803 return 0;
3804
3805 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3806 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3807 btrfs_err_rl(fs_info,
3808 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3809 __func__,
3810 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3811 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3812 return -EUCLEAN;
3813 }
3814
3815 block = kmalloc(sizeof(*block), GFP_NOFS);
3816 if (!block) {
3817 ret = -ENOMEM;
3818 goto out;
3819 }
3820
3821 /*
3822 * @reloc_parent/slot is still before swap, while @block is going to
3823 * record the bytenr after swap, so we do the swap here.
3824 */
3825 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3826 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3827 reloc_slot);
3828 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3829 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3830 subvol_slot);
3831 block->last_snapshot = last_snapshot;
3832 block->level = level;
57949d03
QW		 3833
3834 /*
3835 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3836 * no one else can modify tree blocks thus we qgroup will not change
3837 * no matter the value of trace_leaf.
3838 */
3839 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
370a11b8
QW		 3840 block->trace_leaf = true;
3841 else
3842 block->trace_leaf = false;
3843 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3844
3845 /* Insert @block into @blocks */
3846 spin_lock(&blocks->lock);
3847 cur = &blocks->blocks[level].rb_node;
3848 while (*cur) {
3849 struct btrfs_qgroup_swapped_block *entry;
3850
3851 parent = *cur;
3852 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3853 node);
3854
3855 if (entry->subvol_bytenr < block->subvol_bytenr) {
3856 cur = &(*cur)->rb_left;
3857 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3858 cur = &(*cur)->rb_right;
3859 } else {
3860 if (entry->subvol_generation !=
3861 block->subvol_generation ||
3862 entry->reloc_bytenr != block->reloc_bytenr ||
3863 entry->reloc_generation !=
3864 block->reloc_generation) {
3865 /*
3866 * Duplicated but mismatch entry found.
3867 * Shouldn't happen.
3868 *
3869 * Marking qgroup inconsistent should be enough
3870 * for end users.
3871 */
3872 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3873 ret = -EEXIST;
3874 }
3875 kfree(block);
3876 goto out_unlock;
3877 }
3878 }
3879 rb_link_node(&block->node, parent, cur);
3880 rb_insert_color(&block->node, &blocks->blocks[level]);
3881 blocks->swapped = true;
3882out_unlock:
3883 spin_unlock(&blocks->lock);
3884out:
3885 if (ret < 0)
3886 fs_info->qgroup_flags |=
3887 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3888 return ret;
3889}
f616f5cd
QW		 3890
3891/*
3892 * Check if the tree block is a subtree root, and if so do the needed
3893 * delayed subtree trace for qgroup.
3894 *
3895 * This is called during btrfs_cow_block().
3896 */
3897int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3898 struct btrfs_root *root,
3899 struct extent_buffer *subvol_eb)
3900{
3901 struct btrfs_fs_info *fs_info = root->fs_info;
3902 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3903 struct btrfs_qgroup_swapped_block *block;
3904 struct extent_buffer *reloc_eb = NULL;
3905 struct rb_node *node;
3906 bool found = false;
3907 bool swapped = false;
3908 int level = btrfs_header_level(subvol_eb);
3909 int ret = 0;
3910 int i;
3911
3912 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3913 return 0;
3914 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3915 return 0;
3916
3917 spin_lock(&blocks->lock);
3918 if (!blocks->swapped) {
3919 spin_unlock(&blocks->lock);
3920 return 0;
3921 }
3922 node = blocks->blocks[level].rb_node;
3923
3924 while (node) {
3925 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3926 if (block->subvol_bytenr < subvol_eb->start) {
3927 node = node->rb_left;
3928 } else if (block->subvol_bytenr > subvol_eb->start) {
3929 node = node->rb_right;
3930 } else {
3931 found = true;
3932 break;
3933 }
3934 }
3935 if (!found) {
3936 spin_unlock(&blocks->lock);
3937 goto out;
3938 }
3939 /* Found one, remove it from @blocks first and update blocks->swapped */
3940 rb_erase(&block->node, &blocks->blocks[level]);
3941 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3942 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3943 swapped = true;
3944 break;
3945 }
3946 }
3947 blocks->swapped = swapped;
3948 spin_unlock(&blocks->lock);
3949
3950 /* Read out reloc subtree root */
3951 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3952 block->reloc_generation, block->level,
3953 &block->first_key);
3954 if (IS_ERR(reloc_eb)) {
3955 ret = PTR_ERR(reloc_eb);
3956 reloc_eb = NULL;
3957 goto free_out;
3958 }
3959 if (!extent_buffer_uptodate(reloc_eb)) {
3960 ret = -EIO;
3961 goto free_out;
3962 }
3963
3964 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3965 block->last_snapshot, block->trace_leaf);
3966free_out:
3967 kfree(block);
3968 free_extent_buffer(reloc_eb);
3969out:
3970 if (ret < 0) {
3971 btrfs_err_rl(fs_info,
3972 "failed to account subtree at bytenr %llu: %d",
3973 subvol_eb->start, ret);
3974 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3975 }
3976 return ret;
3977}
Linux 6.x block layer and io_uring tree(s)