blob: 08c8ab199b6e79c6f99fa24f15c489322c5bef7d [file] [log] [blame]
David Sterbac1d7c512018-04-03 19:23:33 +02001// SPDX-License-Identifier: GPL-2.0
Arne Jansenbed92ea2012-06-28 18:03:02 +02002/*
3 * Copyright (C) 2011 STRATO. All rights reserved.
Arne Jansenbed92ea2012-06-28 18:03:02 +02004 */
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>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000013#include <linux/btrfs.h>
Qu Wenruoa514d632017-12-22 16:06:39 +080014#include <linux/sizes.h>
Arne Jansenbed92ea2012-06-28 18:03:02 +020015
16#include "ctree.h"
17#include "transaction.h"
18#include "disk-io.h"
19#include "locking.h"
20#include "ulist.h"
Arne Jansenbed92ea2012-06-28 18:03:02 +020021#include "backref.h"
Jan Schmidt2f232032013-04-25 16:04:51 +000022#include "extent_io.h"
Josef Bacikfcebe452014-05-13 17:30:47 -070023#include "qgroup.h"
Josef Bacikaac00232019-06-20 15:37:44 -040024#include "block-group.h"
Qu Wenruoe69bcee2015-04-17 10:23:16 +080025
Arne Jansenbed92ea2012-06-28 18:03:02 +020026/* TODO XXX FIXME
27 * - subvol delete -> delete when ref goes to 0? delete limits also?
28 * - reorganize keys
29 * - compressed
30 * - sync
Arne Jansenbed92ea2012-06-28 18:03:02 +020031 * - copy also limits on subvol creation
32 * - limit
Andrea Gelmini52042d82018-11-28 12:05:13 +010033 * - caches for ulists
Arne Jansenbed92ea2012-06-28 18:03:02 +020034 * - performance benchmarks
35 * - check all ioctl parameters
36 */
37
Qu Wenruof59c0342017-12-12 15:34:24 +080038/*
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";
Qu Wenruo733e03a2017-12-12 15:34:29 +080060 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 return "meta_pertrans";
62 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 return "meta_prealloc";
Qu Wenruof59c0342017-12-12 15:34:24 +080064 return NULL;
65}
66#endif
67
Qu Wenruo64ee4e72017-12-12 15:34:27 +080068static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 struct btrfs_qgroup *qgroup, u64 num_bytes,
Qu Wenruof59c0342017-12-12 15:34:24 +080070 enum btrfs_qgroup_rsv_type type)
71{
Qu Wenruo64ee4e72017-12-12 15:34:27 +080072 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
Qu Wenruof59c0342017-12-12 15:34:24 +080073 qgroup->rsv.values[type] += num_bytes;
74}
75
Qu Wenruo64ee4e72017-12-12 15:34:27 +080076static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 struct btrfs_qgroup *qgroup, u64 num_bytes,
Qu Wenruof59c0342017-12-12 15:34:24 +080078 enum btrfs_qgroup_rsv_type type)
79{
Qu Wenruo64ee4e72017-12-12 15:34:27 +080080 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
Qu Wenruof59c0342017-12-12 15:34:24 +080081 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
Qu Wenruo64ee4e72017-12-12 15:34:27 +080094static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 struct btrfs_qgroup *dest,
96 struct btrfs_qgroup *src)
Qu Wenruof59c0342017-12-12 15:34:24 +080097{
98 int i;
99
100 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
Qu Wenruo64ee4e72017-12-12 15:34:27 +0800101 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
Qu Wenruof59c0342017-12-12 15:34:24 +0800102}
103
Qu Wenruo64ee4e72017-12-12 15:34:27 +0800104static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 struct btrfs_qgroup *dest,
Qu Wenruof59c0342017-12-12 15:34:24 +0800106 struct btrfs_qgroup *src)
107{
108 int i;
109
110 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
Qu Wenruo64ee4e72017-12-12 15:34:27 +0800111 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
Qu Wenruof59c0342017-12-12 15:34:24 +0800112}
113
Qu Wenruo9c542132015-03-12 16:10:13 +0800114static 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
Arne Jansenbed92ea2012-06-28 18:03:02 +0200144/*
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
David Sterbaef2fff62016-10-26 16:23:50 +0200154static 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}
Josef Bacikfcebe452014-05-13 17:30:47 -0700163
Jan Schmidtb382a322013-05-28 15:47:24 +0000164static 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);
Jan Schmidt2f232032013-04-25 16:04:51 +0000168
Wang Shilong58400fc2013-04-07 10:50:17 +0000169/* must be called with qgroup_ioctl_lock held */
Arne Jansenbed92ea2012-06-28 18:03:02 +0200170static 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
Wang Shilong4082bd32013-08-14 09:13:36 +0800223static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200224{
Arne Jansenbed92ea2012-06-28 18:03:02 +0200225 struct btrfs_qgroup_list *list;
226
Arne Jansenbed92ea2012-06-28 18:03:02 +0200227 list_del(&qgroup->dirty);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200228 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);
Wang Shilong4082bd32013-08-14 09:13:36 +0800244}
Arne Jansenbed92ea2012-06-28 18:03:02 +0200245
Wang Shilong4082bd32013-08-14 09:13:36 +0800246/* 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);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200256 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
Josef Bacikfaa2dbf2014-05-07 17:06:09 -0400308#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
Arne Jansenbed92ea2012-06-28 18:03:02 +0200323/*
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;
Jan Schmidtb382a322013-05-28 15:47:24 +0000337 u64 rescan_progress = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200338
Josef Bacikafcdd122016-09-02 15:40:02 -0400339 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Arne Jansenbed92ea2012-06-28 18:03:02 +0200340 return 0;
341
David Sterba323b88f2017-02-13 12:10:20 +0100342 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
Wang Shilong1e8f9152013-05-06 11:03:27 +0000343 if (!fs_info->qgroup_ulist) {
344 ret = -ENOMEM;
345 goto out;
346 }
347
Arne Jansenbed92ea2012-06-28 18:03:02 +0200348 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) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500382 btrfs_err(fs_info,
383 "old qgroup version, quota disabled");
Arne Jansenbed92ea2012-06-28 18:03:02 +0200384 goto out;
385 }
386 if (btrfs_qgroup_status_generation(l, ptr) !=
387 fs_info->generation) {
388 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Frank Holtonefe120a2013-12-20 11:37:06 -0500389 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -0400390 "qgroup generation mismatch, marked as inconsistent");
Arne Jansenbed92ea2012-06-28 18:03:02 +0200391 }
392 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 ptr);
Jan Schmidtb382a322013-05-28 15:47:24 +0000394 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200395 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)) {
Geert Uytterhoevend41e36a2015-07-06 15:38:11 +0200405 btrfs_err(fs_info, "inconsistent qgroup config");
Arne Jansenbed92ea2012-06-28 18:03:02 +0200406 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);
Arne Jansenff248582013-01-17 01:22:08 -0700475 if (ret == -ENOENT) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500476 btrfs_warn(fs_info,
477 "orphan qgroup relation 0x%llx->0x%llx",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200478 found_key.objectid, found_key.offset);
Arne Jansenff248582013-01-17 01:22:08 -0700479 ret = 0; /* ignore the error */
480 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200481 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;
Josef Bacikafcdd122016-09-02 15:40:02 -0400492 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)
Jan Schmidtb382a322013-05-28 15:47:24 +0000496 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200497 btrfs_free_path(path);
498
Jan Schmidteb1716a2013-05-28 15:47:23 +0000499 if (ret < 0) {
Wang Shilong1e8f9152013-05-06 11:03:27 +0000500 ulist_free(fs_info->qgroup_ulist);
Jan Schmidteb1716a2013-05-28 15:47:23 +0000501 fs_info->qgroup_ulist = NULL;
Jan Schmidtb382a322013-05-28 15:47:24 +0000502 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
Jan Schmidteb1716a2013-05-28 15:47:23 +0000503 }
Wang Shilong1e8f9152013-05-06 11:03:27 +0000504
Arne Jansenbed92ea2012-06-28 18:03:02 +0200505 return ret < 0 ? ret : 0;
506}
507
508/*
Wang Shilonge685da12013-08-14 09:13:37 +0800509 * 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.
Arne Jansenbed92ea2012-06-28 18:03:02 +0200513 */
514void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515{
516 struct rb_node *n;
517 struct btrfs_qgroup *qgroup;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200518
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);
Wang Shilong4082bd32013-08-14 09:13:36 +0800522 __del_qgroup_rb(qgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200523 }
Wang Shilong1e7bac12013-07-13 21:02:54 +0800524 /*
Andrea Gelmini52042d82018-11-28 12:05:13 +0100525 * We call btrfs_free_qgroup_config() when unmounting
Nicholas D Steeves01327612016-05-19 21:18:45 -0400526 * filesystem and disabling quota, so we set qgroup_ulist
Wang Shilong1e7bac12013-07-13 21:02:54 +0800527 * to be null here to avoid double free.
528 */
Wang Shilong1e8f9152013-05-06 11:03:27 +0000529 ulist_free(fs_info->qgroup_ulist);
Wang Shilong1e7bac12013-07-13 21:02:54 +0800530 fs_info->qgroup_ulist = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200531}
532
Lu Fengqi711169c2018-07-18 14:45:24 +0800533static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 u64 dst)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200535{
536 int ret;
Lu Fengqi711169c2018-07-18 14:45:24 +0800537 struct btrfs_root *quota_root = trans->fs_info->quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200538 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
Lu Fengqi99d7f092018-07-18 14:45:25 +0800557static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 u64 dst)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200559{
560 int ret;
Lu Fengqi99d7f092018-07-18 14:45:25 +0800561 struct btrfs_root *quota_root = trans->fs_info->quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200562 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
Jeff Mahoneyf5ee5c92016-06-21 09:52:41 -0400598 if (btrfs_is_testing(quota_root->fs_info))
Josef Bacikfaa2dbf2014-05-07 17:06:09 -0400599 return 0;
David Sterbafccb84c2014-09-29 23:53:21 +0200600
Arne Jansenbed92ea2012-06-28 18:03:02 +0200601 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
Mark Fasheh0b4699d2014-08-18 14:01:17 -0700609 /*
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
Arne Jansenbed92ea2012-06-28 18:03:02 +0200615 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 sizeof(*qgroup_info));
Mark Fasheh0b4699d2014-08-18 14:01:17 -0700617 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200618 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));
Mark Fasheh0b4699d2014-08-18 14:01:17 -0700636 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200637 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
Lu Fengqi69104612018-07-18 14:45:26 +0800656static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200657{
658 int ret;
Lu Fengqi69104612018-07-18 14:45:26 +0800659 struct btrfs_root *quota_root = trans->fs_info->quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200660 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,
Dongsheng Yang1510e712014-11-20 21:01:41 -0500703 struct btrfs_qgroup *qgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200704{
Lu Fengqiac8a8662018-07-18 14:45:27 +0800705 struct btrfs_root *quota_root = trans->fs_info->quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200706 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;
Dongsheng Yang1510e712014-11-20 21:01:41 -0500715 key.offset = qgroup->qgroupid;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200716
717 path = btrfs_alloc_path();
Wang Shilong84cbe2f2013-02-27 11:20:56 +0000718 if (!path)
719 return -ENOMEM;
720
Lu Fengqiac8a8662018-07-18 14:45:27 +0800721 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200722 if (ret > 0)
723 ret = -ENOENT;
724
725 if (ret)
726 goto out;
727
728 l = path->nodes[0];
729 slot = path->slots[0];
Valentina Giustia3df41e2013-11-04 22:34:29 +0100730 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
Dongsheng Yang1510e712014-11-20 21:01:41 -0500731 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);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200736
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,
Arne Jansenbed92ea2012-06-28 18:03:02 +0200745 struct btrfs_qgroup *qgroup)
746{
Lu Fengqi3e07e9a2018-07-18 14:45:28 +0800747 struct btrfs_fs_info *fs_info = trans->fs_info;
748 struct btrfs_root *quota_root = fs_info->quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200749 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
Lu Fengqi3e07e9a2018-07-18 14:45:28 +0800756 if (btrfs_is_testing(fs_info))
Josef Bacikfaa2dbf2014-05-07 17:06:09 -0400757 return 0;
David Sterbafccb84c2014-09-29 23:53:21 +0200758
Arne Jansenbed92ea2012-06-28 18:03:02 +0200759 key.objectid = 0;
760 key.type = BTRFS_QGROUP_INFO_KEY;
761 key.offset = qgroup->qgroupid;
762
763 path = btrfs_alloc_path();
Wang Shilong84cbe2f2013-02-27 11:20:56 +0000764 if (!path)
765 return -ENOMEM;
766
Lu Fengqi3e07e9a2018-07-18 14:45:28 +0800767 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200768 if (ret > 0)
769 ret = -ENOENT;
770
771 if (ret)
772 goto out;
773
774 l = path->nodes[0];
775 slot = path->slots[0];
Valentina Giustia3df41e2013-11-04 22:34:29 +0100776 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200777 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
Lu Fengqi2e980ac2018-07-18 14:45:29 +0800790static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200791{
Lu Fengqi2e980ac2018-07-18 14:45:29 +0800792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *quota_root = fs_info->quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200794 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();
Wang Shilong84cbe2f2013-02-27 11:20:56 +0000806 if (!path)
807 return -ENOMEM;
808
Lu Fengqi2e980ac2018-07-18 14:45:29 +0800809 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200810 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);
Jan Schmidt2f232032013-04-25 16:04:51 +0000821 btrfs_set_qgroup_status_rescan(l, ptr,
822 fs_info->qgroup_rescan_progress.objectid);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200823
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;
Wang Shilong06b3a862013-02-27 11:16:57 +0000839 struct extent_buffer *leaf = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200840 int ret;
Wang Shilong06b3a862013-02-27 11:16:57 +0000841 int nr = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200842
Arne Jansenbed92ea2012-06-28 18:03:02 +0200843 path = btrfs_alloc_path();
844 if (!path)
845 return -ENOMEM;
846
Wang Shilong06b3a862013-02-27 11:16:57 +0000847 path->leave_spinning = 1;
848
849 key.objectid = 0;
850 key.offset = 0;
851 key.type = 0;
852
Arne Jansenbed92ea2012-06-28 18:03:02 +0200853 while (1) {
Arne Jansenbed92ea2012-06-28 18:03:02 +0200854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Wang Shilong06b3a862013-02-27 11:16:57 +0000855 if (ret < 0)
856 goto out;
857 leaf = path->nodes[0];
858 nr = btrfs_header_nritems(leaf);
859 if (!nr)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200860 break;
Wang Shilong06b3a862013-02-27 11:16:57 +0000861 /*
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);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200868 if (ret)
869 goto out;
Wang Shilong06b3a862013-02-27 11:16:57 +0000870
Arne Jansenbed92ea2012-06-28 18:03:02 +0200871 btrfs_release_path(path);
872 }
873 ret = 0;
874out:
Arne Jansenbed92ea2012-06-28 18:03:02 +0200875 btrfs_free_path(path);
876 return ret;
877}
878
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300879int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200880{
881 struct btrfs_root *quota_root;
Wang Shilong7708f022013-04-07 10:24:57 +0000882 struct btrfs_root *tree_root = fs_info->tree_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200883 struct btrfs_path *path = NULL;
884 struct btrfs_qgroup_status_item *ptr;
885 struct extent_buffer *leaf;
886 struct btrfs_key key;
Wang Shilong7708f022013-04-07 10:24:57 +0000887 struct btrfs_key found_key;
888 struct btrfs_qgroup *qgroup = NULL;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300889 struct btrfs_trans_handle *trans = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200890 int ret = 0;
Wang Shilong7708f022013-04-07 10:24:57 +0000891 int slot;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200892
Wang Shilongf2f6ed32013-04-07 10:50:16 +0000893 mutex_lock(&fs_info->qgroup_ioctl_lock);
Nikolay Borisov5d235152018-01-31 10:52:04 +0200894 if (fs_info->quota_root)
Arne Jansenbed92ea2012-06-28 18:03:02 +0200895 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200896
David Sterba7503b832018-12-19 19:47:37 +0100897 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 if (!fs_info->qgroup_ulist) {
899 ret = -ENOMEM;
900 goto out;
901 }
902
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300903 /*
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
Arne Jansenbed92ea2012-06-28 18:03:02 +0200918 /*
919 * initially create the quota tree
920 */
David Sterba9b7a2442019-03-20 13:20:49 +0100921 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200922 if (IS_ERR(quota_root)) {
923 ret = PTR_ERR(quota_root);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300924 btrfs_abort_transaction(trans, ret);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200925 goto out;
926 }
927
928 path = btrfs_alloc_path();
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000929 if (!path) {
930 ret = -ENOMEM;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300931 btrfs_abort_transaction(trans, ret);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000932 goto out_free_root;
933 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200934
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));
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300941 if (ret) {
942 btrfs_abort_transaction(trans, ret);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000943 goto out_free_path;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300944 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200945
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);
Jan Schmidt2f232032013-04-25 16:04:51 +0000954 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
Arne Jansenbed92ea2012-06-28 18:03:02 +0200955
956 btrfs_mark_buffer_dirty(leaf);
957
Wang Shilong7708f022013-04-07 10:24:57 +0000958 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;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300966 if (ret < 0) {
967 btrfs_abort_transaction(trans, ret);
Wang Shilong7708f022013-04-07 10:24:57 +0000968 goto out_free_path;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300969 }
Wang Shilong7708f022013-04-07 10:24:57 +0000970
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);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300979 if (ret) {
980 btrfs_abort_transaction(trans, ret);
Wang Shilong7708f022013-04-07 10:24:57 +0000981 goto out_free_path;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300982 }
Wang Shilong7708f022013-04-07 10:24:57 +0000983
Wang Shilong7708f022013-04-07 10:24:57 +0000984 qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 if (IS_ERR(qgroup)) {
Wang Shilong7708f022013-04-07 10:24:57 +0000986 ret = PTR_ERR(qgroup);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300987 btrfs_abort_transaction(trans, ret);
Wang Shilong7708f022013-04-07 10:24:57 +0000988 goto out_free_path;
989 }
Wang Shilong7708f022013-04-07 10:24:57 +0000990 }
991 ret = btrfs_next_item(tree_root, path);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300992 if (ret < 0) {
993 btrfs_abort_transaction(trans, ret);
Wang Shilong7708f022013-04-07 10:24:57 +0000994 goto out_free_path;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +0300995 }
Wang Shilong7708f022013-04-07 10:24:57 +0000996 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);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001003 if (ret) {
1004 btrfs_abort_transaction(trans, ret);
Wang Shilong7708f022013-04-07 10:24:57 +00001005 goto out_free_path;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001006 }
Wang Shilong7708f022013-04-07 10:24:57 +00001007
Wang Shilong7708f022013-04-07 10:24:57 +00001008 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 if (IS_ERR(qgroup)) {
Wang Shilong7708f022013-04-07 10:24:57 +00001010 ret = PTR_ERR(qgroup);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001011 btrfs_abort_transaction(trans, ret);
Wang Shilong7708f022013-04-07 10:24:57 +00001012 goto out_free_path;
1013 }
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001014
1015 ret = btrfs_commit_transaction(trans);
Dan Carpenterb9b8a412018-08-20 11:25:33 +03001016 trans = NULL;
1017 if (ret)
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001018 goto out_free_path;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001019
Filipe Manana9a6f2092018-11-19 14:15:36 +00001020 /*
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
Nikolay Borisov5d235152018-01-31 10:52:04 +02001030 ret = qgroup_rescan_init(fs_info, 0, 1);
1031 if (!ret) {
1032 qgroup_rescan_zero_tracking(fs_info);
Qu Wenruod61acbb2020-02-07 13:38:20 +08001033 fs_info->qgroup_rescan_running = true;
Nikolay Borisov5d235152018-01-31 10:52:04 +02001034 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1035 &fs_info->qgroup_rescan_work);
1036 }
1037
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +00001038out_free_path:
Arne Jansenbed92ea2012-06-28 18:03:02 +02001039 btrfs_free_path(path);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +00001040out_free_root:
1041 if (ret) {
1042 free_extent_buffer(quota_root->node);
1043 free_extent_buffer(quota_root->commit_root);
Josef Bacik00246522020-01-24 09:33:01 -05001044 btrfs_put_root(quota_root);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +00001045 }
1046out:
Jan Schmidteb1716a2013-05-28 15:47:23 +00001047 if (ret) {
Wang Shilong1e8f9152013-05-06 11:03:27 +00001048 ulist_free(fs_info->qgroup_ulist);
Jan Schmidteb1716a2013-05-28 15:47:23 +00001049 fs_info->qgroup_ulist = NULL;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001050 if (trans)
1051 btrfs_end_transaction(trans);
Jan Schmidteb1716a2013-05-28 15:47:23 +00001052 }
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001053 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001054 return ret;
1055}
1056
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001057int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001058{
Arne Jansenbed92ea2012-06-28 18:03:02 +02001059 struct btrfs_root *quota_root;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001060 struct btrfs_trans_handle *trans = NULL;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001061 int ret = 0;
1062
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001063 mutex_lock(&fs_info->qgroup_ioctl_lock);
Wang Shilong58400fc2013-04-07 10:50:17 +00001064 if (!fs_info->quota_root)
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001065 goto out;
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001066
1067 /*
1068 * 1 For the root item
1069 *
1070 * We should also reserve enough items for the quota tree deletion in
1071 * btrfs_clean_quota_tree but this is not done.
1072 */
1073 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1074 if (IS_ERR(trans)) {
1075 ret = PTR_ERR(trans);
1076 goto out;
1077 }
1078
Josef Bacikafcdd122016-09-02 15:40:02 -04001079 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04001080 btrfs_qgroup_wait_for_completion(fs_info, false);
Justin Maggard967ef512015-11-06 10:36:42 -08001081 spin_lock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001082 quota_root = fs_info->quota_root;
1083 fs_info->quota_root = NULL;
Dongsheng Yang8ea0ec92015-02-27 16:24:26 +08001084 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001085 spin_unlock(&fs_info->qgroup_lock);
1086
Wang Shilonge685da12013-08-14 09:13:37 +08001087 btrfs_free_qgroup_config(fs_info);
1088
Arne Jansenbed92ea2012-06-28 18:03:02 +02001089 ret = btrfs_clean_quota_tree(trans, quota_root);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001090 if (ret) {
1091 btrfs_abort_transaction(trans, ret);
1092 goto end_trans;
1093 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001094
Lu Fengqiab9ce7d2018-08-01 11:32:27 +08001095 ret = btrfs_del_root(trans, &quota_root->root_key);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001096 if (ret) {
1097 btrfs_abort_transaction(trans, ret);
1098 goto end_trans;
1099 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001100
1101 list_del(&quota_root->dirty_list);
1102
1103 btrfs_tree_lock(quota_root->node);
David Sterba6a884d7d2019-03-20 14:30:02 +01001104 btrfs_clean_tree_block(quota_root->node);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001105 btrfs_tree_unlock(quota_root->node);
1106 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1107
1108 free_extent_buffer(quota_root->node);
1109 free_extent_buffer(quota_root->commit_root);
Josef Bacik00246522020-01-24 09:33:01 -05001110 btrfs_put_root(quota_root);
Nikolay Borisov340f1aa2018-07-05 14:50:48 +03001111
1112end_trans:
1113 ret = btrfs_end_transaction(trans);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001114out:
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001115 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001116 return ret;
1117}
1118
Jan Schmidt2f232032013-04-25 16:04:51 +00001119static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1120 struct btrfs_qgroup *qgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001121{
Jan Schmidt2f232032013-04-25 16:04:51 +00001122 if (list_empty(&qgroup->dirty))
1123 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001124}
1125
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001126/*
Qu Wenruo429d6272017-12-12 15:34:26 +08001127 * The easy accounting, we're updating qgroup relationship whose child qgroup
1128 * only has exclusive extents.
1129 *
Andrea Gelmini52042d82018-11-28 12:05:13 +01001130 * In this case, all exclusive extents will also be exclusive for parent, so
Qu Wenruo429d6272017-12-12 15:34:26 +08001131 * excl/rfer just get added/removed.
1132 *
1133 * So is qgroup reservation space, which should also be added/removed to
1134 * parent.
1135 * Or when child tries to release reservation space, parent will underflow its
1136 * reservation (for relationship adding case).
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001137 *
1138 * Caller should hold fs_info->qgroup_lock.
1139 */
1140static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1141 struct ulist *tmp, u64 ref_root,
Qu Wenruo429d6272017-12-12 15:34:26 +08001142 struct btrfs_qgroup *src, int sign)
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001143{
1144 struct btrfs_qgroup *qgroup;
1145 struct btrfs_qgroup_list *glist;
1146 struct ulist_node *unode;
1147 struct ulist_iterator uiter;
Qu Wenruo429d6272017-12-12 15:34:26 +08001148 u64 num_bytes = src->excl;
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001149 int ret = 0;
1150
1151 qgroup = find_qgroup_rb(fs_info, ref_root);
1152 if (!qgroup)
1153 goto out;
1154
1155 qgroup->rfer += sign * num_bytes;
1156 qgroup->rfer_cmpr += sign * num_bytes;
1157
1158 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1159 qgroup->excl += sign * num_bytes;
1160 qgroup->excl_cmpr += sign * num_bytes;
Qu Wenruo429d6272017-12-12 15:34:26 +08001161
1162 if (sign > 0)
Qu Wenruo64ee4e72017-12-12 15:34:27 +08001163 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
Qu Wenruo429d6272017-12-12 15:34:26 +08001164 else
Qu Wenruo64ee4e72017-12-12 15:34:27 +08001165 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001166
1167 qgroup_dirty(fs_info, qgroup);
1168
1169 /* Get all of the parent groups that contain this qgroup */
1170 list_for_each_entry(glist, &qgroup->groups, next_group) {
1171 ret = ulist_add(tmp, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02001172 qgroup_to_aux(glist->group), GFP_ATOMIC);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001173 if (ret < 0)
1174 goto out;
1175 }
1176
1177 /* Iterate all of the parents and adjust their reference counts */
1178 ULIST_ITER_INIT(&uiter);
1179 while ((unode = ulist_next(tmp, &uiter))) {
David Sterbaef2fff62016-10-26 16:23:50 +02001180 qgroup = unode_aux_to_qgroup(unode);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001181 qgroup->rfer += sign * num_bytes;
1182 qgroup->rfer_cmpr += sign * num_bytes;
1183 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1184 qgroup->excl += sign * num_bytes;
Qu Wenruo429d6272017-12-12 15:34:26 +08001185 if (sign > 0)
Qu Wenruo64ee4e72017-12-12 15:34:27 +08001186 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
Qu Wenruo429d6272017-12-12 15:34:26 +08001187 else
Qu Wenruo64ee4e72017-12-12 15:34:27 +08001188 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001189 qgroup->excl_cmpr += sign * num_bytes;
1190 qgroup_dirty(fs_info, qgroup);
1191
1192 /* Add any parents of the parents */
1193 list_for_each_entry(glist, &qgroup->groups, next_group) {
1194 ret = ulist_add(tmp, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02001195 qgroup_to_aux(glist->group), GFP_ATOMIC);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001196 if (ret < 0)
1197 goto out;
1198 }
1199 }
1200 ret = 0;
1201out:
1202 return ret;
1203}
1204
1205
1206/*
1207 * Quick path for updating qgroup with only excl refs.
1208 *
1209 * In that case, just update all parent will be enough.
1210 * Or we needs to do a full rescan.
1211 * Caller should also hold fs_info->qgroup_lock.
1212 *
1213 * Return 0 for quick update, return >0 for need to full rescan
1214 * and mark INCONSISTENT flag.
1215 * Return < 0 for other error.
1216 */
1217static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1218 struct ulist *tmp, u64 src, u64 dst,
1219 int sign)
1220{
1221 struct btrfs_qgroup *qgroup;
1222 int ret = 1;
1223 int err = 0;
1224
1225 qgroup = find_qgroup_rb(fs_info, src);
1226 if (!qgroup)
1227 goto out;
1228 if (qgroup->excl == qgroup->rfer) {
1229 ret = 0;
1230 err = __qgroup_excl_accounting(fs_info, tmp, dst,
Qu Wenruo429d6272017-12-12 15:34:26 +08001231 qgroup, sign);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001232 if (err < 0) {
1233 ret = err;
1234 goto out;
1235 }
1236 }
1237out:
1238 if (ret)
1239 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1240 return ret;
1241}
1242
Lu Fengqi9f8a6ce2018-07-18 14:45:30 +08001243int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1244 u64 dst)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001245{
Lu Fengqi9f8a6ce2018-07-18 14:45:30 +08001246 struct btrfs_fs_info *fs_info = trans->fs_info;
Wang Shilongb7fef4f2013-04-07 10:50:18 +00001247 struct btrfs_qgroup *parent;
1248 struct btrfs_qgroup *member;
Wang Shilong534e6622013-04-17 14:49:51 +00001249 struct btrfs_qgroup_list *list;
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001250 struct ulist *tmp;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001251 int ret = 0;
1252
Qu Wenruo8465ece2015-02-27 16:24:22 +08001253 /* Check the level of src and dst first */
1254 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1255 return -EINVAL;
1256
David Sterba6602caf2017-02-13 12:41:02 +01001257 tmp = ulist_alloc(GFP_KERNEL);
Christian Engelmayerab3680d2015-05-02 17:19:55 +02001258 if (!tmp)
1259 return -ENOMEM;
1260
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001261 mutex_lock(&fs_info->qgroup_ioctl_lock);
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03001262 if (!fs_info->quota_root) {
Marcos Paulo de Souza8a36e402019-11-25 21:58:51 -03001263 ret = -ENOTCONN;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001264 goto out;
1265 }
Wang Shilongb7fef4f2013-04-07 10:50:18 +00001266 member = find_qgroup_rb(fs_info, src);
1267 parent = find_qgroup_rb(fs_info, dst);
1268 if (!member || !parent) {
1269 ret = -EINVAL;
1270 goto out;
1271 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001272
Wang Shilong534e6622013-04-17 14:49:51 +00001273 /* check if such qgroup relation exist firstly */
1274 list_for_each_entry(list, &member->groups, next_group) {
1275 if (list->group == parent) {
1276 ret = -EEXIST;
1277 goto out;
1278 }
1279 }
1280
Lu Fengqi711169c2018-07-18 14:45:24 +08001281 ret = add_qgroup_relation_item(trans, src, dst);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001282 if (ret)
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001283 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001284
Lu Fengqi711169c2018-07-18 14:45:24 +08001285 ret = add_qgroup_relation_item(trans, dst, src);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001286 if (ret) {
Lu Fengqi99d7f092018-07-18 14:45:25 +08001287 del_qgroup_relation_item(trans, src, dst);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001288 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001289 }
1290
1291 spin_lock(&fs_info->qgroup_lock);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001292 ret = add_relation_rb(fs_info, src, dst);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001293 if (ret < 0) {
1294 spin_unlock(&fs_info->qgroup_lock);
1295 goto out;
1296 }
1297 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001298 spin_unlock(&fs_info->qgroup_lock);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001299out:
1300 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001301 ulist_free(tmp);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001302 return ret;
1303}
1304
Lu Fengqi6b36f1a2018-07-18 14:45:31 +08001305static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1306 u64 dst)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001307{
Lu Fengqi6b36f1a2018-07-18 14:45:31 +08001308 struct btrfs_fs_info *fs_info = trans->fs_info;
Wang Shilong534e6622013-04-17 14:49:51 +00001309 struct btrfs_qgroup *parent;
1310 struct btrfs_qgroup *member;
1311 struct btrfs_qgroup_list *list;
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001312 struct ulist *tmp;
Qu Wenruo73798c42019-08-06 22:05:07 +08001313 bool found = false;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001314 int ret = 0;
Qu Wenruo73798c42019-08-06 22:05:07 +08001315 int ret2;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001316
David Sterba6602caf2017-02-13 12:41:02 +01001317 tmp = ulist_alloc(GFP_KERNEL);
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001318 if (!tmp)
1319 return -ENOMEM;
1320
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03001321 if (!fs_info->quota_root) {
Marcos Paulo de Souza8a36e402019-11-25 21:58:51 -03001322 ret = -ENOTCONN;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001323 goto out;
1324 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001325
Wang Shilong534e6622013-04-17 14:49:51 +00001326 member = find_qgroup_rb(fs_info, src);
1327 parent = find_qgroup_rb(fs_info, dst);
Qu Wenruo73798c42019-08-06 22:05:07 +08001328 /*
1329 * The parent/member pair doesn't exist, then try to delete the dead
1330 * relation items only.
1331 */
1332 if (!member || !parent)
1333 goto delete_item;
Wang Shilong534e6622013-04-17 14:49:51 +00001334
1335 /* check if such qgroup relation exist firstly */
1336 list_for_each_entry(list, &member->groups, next_group) {
Qu Wenruo73798c42019-08-06 22:05:07 +08001337 if (list->group == parent) {
1338 found = true;
1339 break;
1340 }
Wang Shilong534e6622013-04-17 14:49:51 +00001341 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001342
Qu Wenruo73798c42019-08-06 22:05:07 +08001343delete_item:
1344 ret = del_qgroup_relation_item(trans, src, dst);
1345 if (ret < 0 && ret != -ENOENT)
1346 goto out;
1347 ret2 = del_qgroup_relation_item(trans, dst, src);
1348 if (ret2 < 0 && ret2 != -ENOENT)
1349 goto out;
1350
1351 /* At least one deletion succeeded, return 0 */
1352 if (!ret || !ret2)
1353 ret = 0;
1354
1355 if (found) {
1356 spin_lock(&fs_info->qgroup_lock);
1357 del_relation_rb(fs_info, src, dst);
1358 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1359 spin_unlock(&fs_info->qgroup_lock);
1360 }
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001361out:
Qu Wenruo9c8b35b2015-02-27 16:24:27 +08001362 ulist_free(tmp);
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001363 return ret;
1364}
1365
Lu Fengqi39616c22018-07-18 14:45:32 +08001366int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1367 u64 dst)
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001368{
Lu Fengqi39616c22018-07-18 14:45:32 +08001369 struct btrfs_fs_info *fs_info = trans->fs_info;
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001370 int ret = 0;
1371
1372 mutex_lock(&fs_info->qgroup_ioctl_lock);
Lu Fengqi6b36f1a2018-07-18 14:45:31 +08001373 ret = __del_qgroup_relation(trans, src, dst);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001374 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001375
Arne Jansenbed92ea2012-06-28 18:03:02 +02001376 return ret;
1377}
1378
Lu Fengqi49a05ec2018-07-18 14:45:33 +08001379int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001380{
Lu Fengqi49a05ec2018-07-18 14:45:33 +08001381 struct btrfs_fs_info *fs_info = trans->fs_info;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001382 struct btrfs_root *quota_root;
1383 struct btrfs_qgroup *qgroup;
1384 int ret = 0;
1385
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001386 mutex_lock(&fs_info->qgroup_ioctl_lock);
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03001387 if (!fs_info->quota_root) {
Marcos Paulo de Souza8a36e402019-11-25 21:58:51 -03001388 ret = -ENOTCONN;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001389 goto out;
1390 }
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03001391 quota_root = fs_info->quota_root;
Wang Shilong534e6622013-04-17 14:49:51 +00001392 qgroup = find_qgroup_rb(fs_info, qgroupid);
1393 if (qgroup) {
1394 ret = -EEXIST;
1395 goto out;
1396 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001397
1398 ret = add_qgroup_item(trans, quota_root, qgroupid);
Wang Shilong534e6622013-04-17 14:49:51 +00001399 if (ret)
1400 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001401
1402 spin_lock(&fs_info->qgroup_lock);
1403 qgroup = add_qgroup_rb(fs_info, qgroupid);
1404 spin_unlock(&fs_info->qgroup_lock);
1405
1406 if (IS_ERR(qgroup))
1407 ret = PTR_ERR(qgroup);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001408out:
1409 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001410 return ret;
1411}
1412
Lu Fengqi3efbee12018-07-18 14:45:34 +08001413int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001414{
Lu Fengqi3efbee12018-07-18 14:45:34 +08001415 struct btrfs_fs_info *fs_info = trans->fs_info;
Arne Jansen2cf68702013-01-17 01:22:09 -07001416 struct btrfs_qgroup *qgroup;
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001417 struct btrfs_qgroup_list *list;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001418 int ret = 0;
1419
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001420 mutex_lock(&fs_info->qgroup_ioctl_lock);
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03001421 if (!fs_info->quota_root) {
Marcos Paulo de Souza8a36e402019-11-25 21:58:51 -03001422 ret = -ENOTCONN;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001423 goto out;
1424 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001425
Arne Jansen2cf68702013-01-17 01:22:09 -07001426 qgroup = find_qgroup_rb(fs_info, qgroupid);
Wang Shilong534e6622013-04-17 14:49:51 +00001427 if (!qgroup) {
1428 ret = -ENOENT;
1429 goto out;
Arne Jansen2cf68702013-01-17 01:22:09 -07001430 }
Lu Fengqib90e22b2018-10-11 13:42:56 +08001431
1432 /* Check if there are no children of this qgroup */
1433 if (!list_empty(&qgroup->members)) {
1434 ret = -EBUSY;
1435 goto out;
1436 }
1437
Lu Fengqi69104612018-07-18 14:45:26 +08001438 ret = del_qgroup_item(trans, qgroupid);
Sargun Dhillon36b96fd2017-09-17 09:02:29 +00001439 if (ret && ret != -ENOENT)
1440 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001441
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001442 while (!list_empty(&qgroup->groups)) {
1443 list = list_first_entry(&qgroup->groups,
1444 struct btrfs_qgroup_list, next_group);
Lu Fengqi6b36f1a2018-07-18 14:45:31 +08001445 ret = __del_qgroup_relation(trans, qgroupid,
1446 list->group->qgroupid);
Dongsheng Yangf5a6b1c2014-11-24 10:27:09 -05001447 if (ret)
1448 goto out;
1449 }
1450
Arne Jansenbed92ea2012-06-28 18:03:02 +02001451 spin_lock(&fs_info->qgroup_lock);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001452 del_qgroup_rb(fs_info, qgroupid);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001453 spin_unlock(&fs_info->qgroup_lock);
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001454out:
1455 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001456 return ret;
1457}
1458
Lu Fengqif0042d52018-07-18 14:45:35 +08001459int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
Arne Jansenbed92ea2012-06-28 18:03:02 +02001460 struct btrfs_qgroup_limit *limit)
1461{
Lu Fengqif0042d52018-07-18 14:45:35 +08001462 struct btrfs_fs_info *fs_info = trans->fs_info;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001463 struct btrfs_qgroup *qgroup;
1464 int ret = 0;
Yang Dongshengfe759902015-06-03 14:57:32 +08001465 /* Sometimes we would want to clear the limit on this qgroup.
1466 * To meet this requirement, we treat the -1 as a special value
1467 * which tell kernel to clear the limit on this qgroup.
1468 */
1469 const u64 CLEAR_VALUE = -1;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001470
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001471 mutex_lock(&fs_info->qgroup_ioctl_lock);
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03001472 if (!fs_info->quota_root) {
Marcos Paulo de Souza8a36e402019-11-25 21:58:51 -03001473 ret = -ENOTCONN;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001474 goto out;
1475 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001476
Wang Shilongddb47af2013-04-07 10:50:20 +00001477 qgroup = find_qgroup_rb(fs_info, qgroupid);
1478 if (!qgroup) {
1479 ret = -ENOENT;
1480 goto out;
1481 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001482
Wang Shilong58400fc2013-04-07 10:50:17 +00001483 spin_lock(&fs_info->qgroup_lock);
Yang Dongshengfe759902015-06-03 14:57:32 +08001484 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1485 if (limit->max_rfer == CLEAR_VALUE) {
1486 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1487 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1488 qgroup->max_rfer = 0;
1489 } else {
1490 qgroup->max_rfer = limit->max_rfer;
1491 }
1492 }
1493 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1494 if (limit->max_excl == CLEAR_VALUE) {
1495 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1496 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1497 qgroup->max_excl = 0;
1498 } else {
1499 qgroup->max_excl = limit->max_excl;
1500 }
1501 }
1502 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1503 if (limit->rsv_rfer == CLEAR_VALUE) {
1504 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1505 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1506 qgroup->rsv_rfer = 0;
1507 } else {
1508 qgroup->rsv_rfer = limit->rsv_rfer;
1509 }
1510 }
1511 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1512 if (limit->rsv_excl == CLEAR_VALUE) {
1513 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1514 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1515 qgroup->rsv_excl = 0;
1516 } else {
1517 qgroup->rsv_excl = limit->rsv_excl;
1518 }
1519 }
Dongsheng Yang03477d92015-02-06 11:06:25 -05001520 qgroup->lim_flags |= limit->flags;
1521
Arne Jansenbed92ea2012-06-28 18:03:02 +02001522 spin_unlock(&fs_info->qgroup_lock);
Dongsheng Yang1510e712014-11-20 21:01:41 -05001523
Lu Fengqiac8a8662018-07-18 14:45:27 +08001524 ret = update_qgroup_limit_item(trans, qgroup);
Dongsheng Yang1510e712014-11-20 21:01:41 -05001525 if (ret) {
1526 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1527 btrfs_info(fs_info, "unable to update quota limit for %llu",
1528 qgroupid);
1529 }
1530
Wang Shilongf2f6ed32013-04-07 10:50:16 +00001531out:
1532 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001533 return ret;
1534}
Mark Fasheh11526512014-07-17 12:39:01 -07001535
Qu Wenruo50b3e042016-10-18 09:31:27 +08001536int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
Qu Wenruocb93b522016-08-15 10:36:50 +08001537 struct btrfs_delayed_ref_root *delayed_refs,
1538 struct btrfs_qgroup_extent_record *record)
Qu Wenruo3368d002015-04-16 14:34:17 +08001539{
1540 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1541 struct rb_node *parent_node = NULL;
1542 struct btrfs_qgroup_extent_record *entry;
1543 u64 bytenr = record->bytenr;
1544
David Sterbaa4666e62018-03-16 02:21:22 +01001545 lockdep_assert_held(&delayed_refs->lock);
Qu Wenruo50b3e042016-10-18 09:31:27 +08001546 trace_btrfs_qgroup_trace_extent(fs_info, record);
Mark Fasheh82bd1012015-11-05 14:38:00 -08001547
Qu Wenruo3368d002015-04-16 14:34:17 +08001548 while (*p) {
1549 parent_node = *p;
1550 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1551 node);
Qu Wenruo1418bae2019-01-23 15:15:12 +08001552 if (bytenr < entry->bytenr) {
Qu Wenruo3368d002015-04-16 14:34:17 +08001553 p = &(*p)->rb_left;
Qu Wenruo1418bae2019-01-23 15:15:12 +08001554 } else if (bytenr > entry->bytenr) {
Qu Wenruo3368d002015-04-16 14:34:17 +08001555 p = &(*p)->rb_right;
Qu Wenruo1418bae2019-01-23 15:15:12 +08001556 } else {
1557 if (record->data_rsv && !entry->data_rsv) {
1558 entry->data_rsv = record->data_rsv;
1559 entry->data_rsv_refroot =
1560 record->data_rsv_refroot;
1561 }
Qu Wenruocb93b522016-08-15 10:36:50 +08001562 return 1;
Qu Wenruo1418bae2019-01-23 15:15:12 +08001563 }
Qu Wenruo3368d002015-04-16 14:34:17 +08001564 }
1565
1566 rb_link_node(&record->node, parent_node, p);
1567 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
Qu Wenruocb93b522016-08-15 10:36:50 +08001568 return 0;
1569}
1570
Qu Wenruofb235dc2017-02-15 10:43:03 +08001571int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1572 struct btrfs_qgroup_extent_record *qrecord)
1573{
1574 struct ulist *old_root;
1575 u64 bytenr = qrecord->bytenr;
1576 int ret;
1577
Zygo Blaxellc995ab32017-09-22 13:58:45 -04001578 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
Nikolay Borisov952bd3db2018-01-29 15:53:01 +02001579 if (ret < 0) {
1580 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1581 btrfs_warn(fs_info,
1582"error accounting new delayed refs extent (err code: %d), quota inconsistent",
1583 ret);
1584 return 0;
1585 }
Qu Wenruofb235dc2017-02-15 10:43:03 +08001586
1587 /*
1588 * Here we don't need to get the lock of
1589 * trans->transaction->delayed_refs, since inserted qrecord won't
1590 * be deleted, only qrecord->node may be modified (new qrecord insert)
1591 *
1592 * So modifying qrecord->old_roots is safe here
1593 */
1594 qrecord->old_roots = old_root;
1595 return 0;
1596}
1597
Lu Fengqia95f3aa2018-07-18 16:28:03 +08001598int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1599 u64 num_bytes, gfp_t gfp_flag)
Qu Wenruocb93b522016-08-15 10:36:50 +08001600{
Lu Fengqia95f3aa2018-07-18 16:28:03 +08001601 struct btrfs_fs_info *fs_info = trans->fs_info;
Qu Wenruocb93b522016-08-15 10:36:50 +08001602 struct btrfs_qgroup_extent_record *record;
1603 struct btrfs_delayed_ref_root *delayed_refs;
1604 int ret;
1605
Josef Bacikafcdd122016-09-02 15:40:02 -04001606 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1607 || bytenr == 0 || num_bytes == 0)
Qu Wenruocb93b522016-08-15 10:36:50 +08001608 return 0;
Qu Wenruo1418bae2019-01-23 15:15:12 +08001609 record = kzalloc(sizeof(*record), gfp_flag);
Qu Wenruocb93b522016-08-15 10:36:50 +08001610 if (!record)
1611 return -ENOMEM;
1612
1613 delayed_refs = &trans->transaction->delayed_refs;
1614 record->bytenr = bytenr;
1615 record->num_bytes = num_bytes;
1616 record->old_roots = NULL;
1617
1618 spin_lock(&delayed_refs->lock);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001619 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
Qu Wenruocb93b522016-08-15 10:36:50 +08001620 spin_unlock(&delayed_refs->lock);
Qu Wenruofb235dc2017-02-15 10:43:03 +08001621 if (ret > 0) {
Qu Wenruocb93b522016-08-15 10:36:50 +08001622 kfree(record);
Qu Wenruofb235dc2017-02-15 10:43:03 +08001623 return 0;
1624 }
1625 return btrfs_qgroup_trace_extent_post(fs_info, record);
Qu Wenruo3368d002015-04-16 14:34:17 +08001626}
1627
Qu Wenruo33d1f052016-10-18 09:31:28 +08001628int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
Qu Wenruo33d1f052016-10-18 09:31:28 +08001629 struct extent_buffer *eb)
1630{
Lu Fengqi8d38d7e2018-07-18 14:45:37 +08001631 struct btrfs_fs_info *fs_info = trans->fs_info;
Qu Wenruo33d1f052016-10-18 09:31:28 +08001632 int nr = btrfs_header_nritems(eb);
1633 int i, extent_type, ret;
1634 struct btrfs_key key;
1635 struct btrfs_file_extent_item *fi;
1636 u64 bytenr, num_bytes;
1637
1638 /* We can be called directly from walk_up_proc() */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001639 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Qu Wenruo33d1f052016-10-18 09:31:28 +08001640 return 0;
1641
1642 for (i = 0; i < nr; i++) {
1643 btrfs_item_key_to_cpu(eb, &key, i);
1644
1645 if (key.type != BTRFS_EXTENT_DATA_KEY)
1646 continue;
1647
1648 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1649 /* filter out non qgroup-accountable extents */
1650 extent_type = btrfs_file_extent_type(eb, fi);
1651
1652 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1653 continue;
1654
1655 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1656 if (!bytenr)
1657 continue;
1658
1659 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1660
Lu Fengqia95f3aa2018-07-18 16:28:03 +08001661 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1662 GFP_NOFS);
Qu Wenruo33d1f052016-10-18 09:31:28 +08001663 if (ret)
1664 return ret;
1665 }
Jeff Mahoneycddf3b22017-06-20 08:15:26 -04001666 cond_resched();
Qu Wenruo33d1f052016-10-18 09:31:28 +08001667 return 0;
1668}
1669
1670/*
1671 * Walk up the tree from the bottom, freeing leaves and any interior
1672 * nodes which have had all slots visited. If a node (leaf or
1673 * interior) is freed, the node above it will have it's slot
1674 * incremented. The root node will never be freed.
1675 *
1676 * At the end of this function, we should have a path which has all
1677 * slots incremented to the next position for a search. If we need to
1678 * read a new node it will be NULL and the node above it will have the
1679 * correct slot selected for a later read.
1680 *
1681 * If we increment the root nodes slot counter past the number of
1682 * elements, 1 is returned to signal completion of the search.
1683 */
David Sterba15b34512017-02-10 20:30:23 +01001684static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
Qu Wenruo33d1f052016-10-18 09:31:28 +08001685{
1686 int level = 0;
1687 int nr, slot;
1688 struct extent_buffer *eb;
1689
1690 if (root_level == 0)
1691 return 1;
1692
1693 while (level <= root_level) {
1694 eb = path->nodes[level];
1695 nr = btrfs_header_nritems(eb);
1696 path->slots[level]++;
1697 slot = path->slots[level];
1698 if (slot >= nr || level == 0) {
1699 /*
1700 * Don't free the root - we will detect this
1701 * condition after our loop and return a
1702 * positive value for caller to stop walking the tree.
1703 */
1704 if (level != root_level) {
1705 btrfs_tree_unlock_rw(eb, path->locks[level]);
1706 path->locks[level] = 0;
1707
1708 free_extent_buffer(eb);
1709 path->nodes[level] = NULL;
1710 path->slots[level] = 0;
1711 }
1712 } else {
1713 /*
1714 * We have a valid slot to walk back down
1715 * from. Stop here so caller can process these
1716 * new nodes.
1717 */
1718 break;
1719 }
1720
1721 level++;
1722 }
1723
1724 eb = path->nodes[root_level];
1725 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1726 return 1;
1727
1728 return 0;
1729}
1730
Qu Wenruo25982562018-09-27 14:42:30 +08001731/*
1732 * Helper function to trace a subtree tree block swap.
1733 *
1734 * The swap will happen in highest tree block, but there may be a lot of
1735 * tree blocks involved.
1736 *
1737 * For example:
1738 * OO = Old tree blocks
1739 * NN = New tree blocks allocated during balance
1740 *
1741 * File tree (257) Reloc tree for 257
1742 * L2 OO NN
1743 * / \ / \
1744 * L1 OO OO (a) OO NN (a)
1745 * / \ / \ / \ / \
1746 * L0 OO OO OO OO OO OO NN NN
1747 * (b) (c) (b) (c)
1748 *
1749 * When calling qgroup_trace_extent_swap(), we will pass:
1750 * @src_eb = OO(a)
1751 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1752 * @dst_level = 0
1753 * @root_level = 1
1754 *
1755 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1756 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1757 *
1758 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1759 *
1760 * 1) Tree search from @src_eb
1761 * It should acts as a simplified btrfs_search_slot().
1762 * The key for search can be extracted from @dst_path->nodes[dst_level]
1763 * (first key).
1764 *
1765 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1766 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
Andrea Gelmini52042d82018-11-28 12:05:13 +01001767 * They should be marked during previous (@dst_level = 1) iteration.
Qu Wenruo25982562018-09-27 14:42:30 +08001768 *
1769 * 3) Mark file extents in leaves dirty
1770 * We don't have good way to pick out new file extents only.
1771 * So we still follow the old method by scanning all file extents in
1772 * the leave.
1773 *
Andrea Gelmini52042d82018-11-28 12:05:13 +01001774 * This function can free us from keeping two paths, thus later we only need
Qu Wenruo25982562018-09-27 14:42:30 +08001775 * to care about how to iterate all new tree blocks in reloc tree.
1776 */
1777static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1778 struct extent_buffer *src_eb,
1779 struct btrfs_path *dst_path,
Qu Wenruo3d0174f2018-09-27 14:42:35 +08001780 int dst_level, int root_level,
1781 bool trace_leaf)
Qu Wenruo25982562018-09-27 14:42:30 +08001782{
1783 struct btrfs_key key;
1784 struct btrfs_path *src_path;
1785 struct btrfs_fs_info *fs_info = trans->fs_info;
1786 u32 nodesize = fs_info->nodesize;
1787 int cur_level = root_level;
1788 int ret;
1789
1790 BUG_ON(dst_level > root_level);
1791 /* Level mismatch */
1792 if (btrfs_header_level(src_eb) != root_level)
1793 return -EINVAL;
1794
1795 src_path = btrfs_alloc_path();
1796 if (!src_path) {
1797 ret = -ENOMEM;
1798 goto out;
1799 }
1800
1801 if (dst_level)
1802 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1803 else
1804 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1805
1806 /* For src_path */
David Sterba67439da2019-10-08 13:28:47 +02001807 atomic_inc(&src_eb->refs);
Qu Wenruo25982562018-09-27 14:42:30 +08001808 src_path->nodes[root_level] = src_eb;
1809 src_path->slots[root_level] = dst_path->slots[root_level];
1810 src_path->locks[root_level] = 0;
1811
1812 /* A simplified version of btrfs_search_slot() */
1813 while (cur_level >= dst_level) {
1814 struct btrfs_key src_key;
1815 struct btrfs_key dst_key;
1816
1817 if (src_path->nodes[cur_level] == NULL) {
1818 struct btrfs_key first_key;
1819 struct extent_buffer *eb;
1820 int parent_slot;
1821 u64 child_gen;
1822 u64 child_bytenr;
1823
1824 eb = src_path->nodes[cur_level + 1];
1825 parent_slot = src_path->slots[cur_level + 1];
1826 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1827 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1828 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1829
1830 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1831 cur_level, &first_key);
1832 if (IS_ERR(eb)) {
1833 ret = PTR_ERR(eb);
1834 goto out;
1835 } else if (!extent_buffer_uptodate(eb)) {
1836 free_extent_buffer(eb);
1837 ret = -EIO;
1838 goto out;
1839 }
1840
1841 src_path->nodes[cur_level] = eb;
1842
1843 btrfs_tree_read_lock(eb);
David Sterba300aa892018-04-04 02:00:17 +02001844 btrfs_set_lock_blocking_read(eb);
Qu Wenruo25982562018-09-27 14:42:30 +08001845 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1846 }
1847
1848 src_path->slots[cur_level] = dst_path->slots[cur_level];
1849 if (cur_level) {
1850 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1851 &dst_key, dst_path->slots[cur_level]);
1852 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1853 &src_key, src_path->slots[cur_level]);
1854 } else {
1855 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1856 &dst_key, dst_path->slots[cur_level]);
1857 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1858 &src_key, src_path->slots[cur_level]);
1859 }
1860 /* Content mismatch, something went wrong */
1861 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1862 ret = -ENOENT;
1863 goto out;
1864 }
1865 cur_level--;
1866 }
1867
1868 /*
1869 * Now both @dst_path and @src_path have been populated, record the tree
1870 * blocks for qgroup accounting.
1871 */
1872 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1873 nodesize, GFP_NOFS);
1874 if (ret < 0)
1875 goto out;
1876 ret = btrfs_qgroup_trace_extent(trans,
1877 dst_path->nodes[dst_level]->start,
1878 nodesize, GFP_NOFS);
1879 if (ret < 0)
1880 goto out;
1881
1882 /* Record leaf file extents */
Qu Wenruo3d0174f2018-09-27 14:42:35 +08001883 if (dst_level == 0 && trace_leaf) {
Qu Wenruo25982562018-09-27 14:42:30 +08001884 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1885 if (ret < 0)
1886 goto out;
1887 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1888 }
1889out:
1890 btrfs_free_path(src_path);
1891 return ret;
1892}
1893
Qu Wenruoea49f3e2018-09-27 14:42:31 +08001894/*
1895 * Helper function to do recursive generation-aware depth-first search, to
1896 * locate all new tree blocks in a subtree of reloc tree.
1897 *
1898 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1899 * reloc tree
1900 * L2 NN (a)
1901 * / \
1902 * L1 OO NN (b)
1903 * / \ / \
1904 * L0 OO OO OO NN
1905 * (c) (d)
1906 * If we pass:
1907 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1908 * @cur_level = 1
1909 * @root_level = 1
1910 *
1911 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1912 * above tree blocks along with their counter parts in file tree.
Andrea Gelmini52042d82018-11-28 12:05:13 +01001913 * While during search, old tree blocks OO(c) will be skipped as tree block swap
Qu Wenruoea49f3e2018-09-27 14:42:31 +08001914 * won't affect OO(c).
1915 */
1916static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1917 struct extent_buffer *src_eb,
1918 struct btrfs_path *dst_path,
1919 int cur_level, int root_level,
Qu Wenruo3d0174f2018-09-27 14:42:35 +08001920 u64 last_snapshot, bool trace_leaf)
Qu Wenruoea49f3e2018-09-27 14:42:31 +08001921{
1922 struct btrfs_fs_info *fs_info = trans->fs_info;
1923 struct extent_buffer *eb;
1924 bool need_cleanup = false;
1925 int ret = 0;
1926 int i;
1927
1928 /* Level sanity check */
Nikolay Borisov7ff2c2a2019-03-18 17:45:19 +02001929 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1930 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
Qu Wenruoea49f3e2018-09-27 14:42:31 +08001931 root_level < cur_level) {
1932 btrfs_err_rl(fs_info,
1933 "%s: bad levels, cur_level=%d root_level=%d",
1934 __func__, cur_level, root_level);
1935 return -EUCLEAN;
1936 }
1937
1938 /* Read the tree block if needed */
1939 if (dst_path->nodes[cur_level] == NULL) {
1940 struct btrfs_key first_key;
1941 int parent_slot;
1942 u64 child_gen;
1943 u64 child_bytenr;
1944
1945 /*
1946 * dst_path->nodes[root_level] must be initialized before
1947 * calling this function.
1948 */
1949 if (cur_level == root_level) {
1950 btrfs_err_rl(fs_info,
1951 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1952 __func__, root_level, root_level, cur_level);
1953 return -EUCLEAN;
1954 }
1955
1956 /*
1957 * We need to get child blockptr/gen from parent before we can
1958 * read it.
1959 */
1960 eb = dst_path->nodes[cur_level + 1];
1961 parent_slot = dst_path->slots[cur_level + 1];
1962 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1963 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1964 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1965
1966 /* This node is old, no need to trace */
1967 if (child_gen < last_snapshot)
1968 goto out;
1969
1970 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1971 cur_level, &first_key);
1972 if (IS_ERR(eb)) {
1973 ret = PTR_ERR(eb);
1974 goto out;
1975 } else if (!extent_buffer_uptodate(eb)) {
1976 free_extent_buffer(eb);
1977 ret = -EIO;
1978 goto out;
1979 }
1980
1981 dst_path->nodes[cur_level] = eb;
1982 dst_path->slots[cur_level] = 0;
1983
1984 btrfs_tree_read_lock(eb);
David Sterba300aa892018-04-04 02:00:17 +02001985 btrfs_set_lock_blocking_read(eb);
Qu Wenruoea49f3e2018-09-27 14:42:31 +08001986 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1987 need_cleanup = true;
1988 }
1989
1990 /* Now record this tree block and its counter part for qgroups */
1991 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
Qu Wenruo3d0174f2018-09-27 14:42:35 +08001992 root_level, trace_leaf);
Qu Wenruoea49f3e2018-09-27 14:42:31 +08001993 if (ret < 0)
1994 goto cleanup;
1995
1996 eb = dst_path->nodes[cur_level];
1997
1998 if (cur_level > 0) {
1999 /* Iterate all child tree blocks */
2000 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2001 /* Skip old tree blocks as they won't be swapped */
2002 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2003 continue;
2004 dst_path->slots[cur_level] = i;
2005
2006 /* Recursive call (at most 7 times) */
2007 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2008 dst_path, cur_level - 1, root_level,
Qu Wenruo3d0174f2018-09-27 14:42:35 +08002009 last_snapshot, trace_leaf);
Qu Wenruoea49f3e2018-09-27 14:42:31 +08002010 if (ret < 0)
2011 goto cleanup;
2012 }
2013 }
2014
2015cleanup:
2016 if (need_cleanup) {
2017 /* Clean up */
2018 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2019 dst_path->locks[cur_level]);
2020 free_extent_buffer(dst_path->nodes[cur_level]);
2021 dst_path->nodes[cur_level] = NULL;
2022 dst_path->slots[cur_level] = 0;
2023 dst_path->locks[cur_level] = 0;
2024 }
2025out:
2026 return ret;
2027}
2028
Qu Wenruo5aea1a42019-01-23 15:15:15 +08002029static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2030 struct extent_buffer *src_eb,
2031 struct extent_buffer *dst_eb,
2032 u64 last_snapshot, bool trace_leaf)
2033{
2034 struct btrfs_fs_info *fs_info = trans->fs_info;
2035 struct btrfs_path *dst_path = NULL;
2036 int level;
2037 int ret;
2038
2039 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2040 return 0;
2041
2042 /* Wrong parameter order */
2043 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2044 btrfs_err_rl(fs_info,
2045 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2046 btrfs_header_generation(src_eb),
2047 btrfs_header_generation(dst_eb));
2048 return -EUCLEAN;
2049 }
2050
2051 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2052 ret = -EIO;
2053 goto out;
2054 }
2055
2056 level = btrfs_header_level(dst_eb);
2057 dst_path = btrfs_alloc_path();
2058 if (!dst_path) {
2059 ret = -ENOMEM;
2060 goto out;
2061 }
2062 /* For dst_path */
David Sterba67439da2019-10-08 13:28:47 +02002063 atomic_inc(&dst_eb->refs);
Qu Wenruo5aea1a42019-01-23 15:15:15 +08002064 dst_path->nodes[level] = dst_eb;
2065 dst_path->slots[level] = 0;
2066 dst_path->locks[level] = 0;
2067
2068 /* Do the generation aware breadth-first search */
2069 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2070 level, last_snapshot, trace_leaf);
2071 if (ret < 0)
2072 goto out;
2073 ret = 0;
2074
2075out:
2076 btrfs_free_path(dst_path);
2077 if (ret < 0)
2078 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2079 return ret;
2080}
2081
Qu Wenruo33d1f052016-10-18 09:31:28 +08002082int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
Qu Wenruo33d1f052016-10-18 09:31:28 +08002083 struct extent_buffer *root_eb,
2084 u64 root_gen, int root_level)
2085{
Lu Fengqideb40622018-07-18 14:45:38 +08002086 struct btrfs_fs_info *fs_info = trans->fs_info;
Qu Wenruo33d1f052016-10-18 09:31:28 +08002087 int ret = 0;
2088 int level;
2089 struct extent_buffer *eb = root_eb;
2090 struct btrfs_path *path = NULL;
2091
Nikolay Borisovb6e6bca2017-07-12 09:42:19 +03002092 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002093 BUG_ON(root_eb == NULL);
2094
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002095 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Qu Wenruo33d1f052016-10-18 09:31:28 +08002096 return 0;
2097
2098 if (!extent_buffer_uptodate(root_eb)) {
Qu Wenruo581c1762018-03-29 09:08:11 +08002099 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002100 if (ret)
2101 goto out;
2102 }
2103
2104 if (root_level == 0) {
Lu Fengqi8d38d7e2018-07-18 14:45:37 +08002105 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002106 goto out;
2107 }
2108
2109 path = btrfs_alloc_path();
2110 if (!path)
2111 return -ENOMEM;
2112
2113 /*
2114 * Walk down the tree. Missing extent blocks are filled in as
2115 * we go. Metadata is accounted every time we read a new
2116 * extent block.
2117 *
2118 * When we reach a leaf, we account for file extent items in it,
2119 * walk back up the tree (adjusting slot pointers as we go)
2120 * and restart the search process.
2121 */
David Sterba67439da2019-10-08 13:28:47 +02002122 atomic_inc(&root_eb->refs); /* For path */
Qu Wenruo33d1f052016-10-18 09:31:28 +08002123 path->nodes[root_level] = root_eb;
2124 path->slots[root_level] = 0;
2125 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2126walk_down:
2127 level = root_level;
2128 while (level >= 0) {
2129 if (path->nodes[level] == NULL) {
Qu Wenruo581c1762018-03-29 09:08:11 +08002130 struct btrfs_key first_key;
Qu Wenruo33d1f052016-10-18 09:31:28 +08002131 int parent_slot;
2132 u64 child_gen;
2133 u64 child_bytenr;
2134
2135 /*
2136 * We need to get child blockptr/gen from parent before
2137 * we can read it.
2138 */
2139 eb = path->nodes[level + 1];
2140 parent_slot = path->slots[level + 1];
2141 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2142 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
Qu Wenruo581c1762018-03-29 09:08:11 +08002143 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002144
Qu Wenruo581c1762018-03-29 09:08:11 +08002145 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2146 level, &first_key);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002147 if (IS_ERR(eb)) {
2148 ret = PTR_ERR(eb);
2149 goto out;
2150 } else if (!extent_buffer_uptodate(eb)) {
2151 free_extent_buffer(eb);
2152 ret = -EIO;
2153 goto out;
2154 }
2155
2156 path->nodes[level] = eb;
2157 path->slots[level] = 0;
2158
2159 btrfs_tree_read_lock(eb);
David Sterba300aa892018-04-04 02:00:17 +02002160 btrfs_set_lock_blocking_read(eb);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002161 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2162
Lu Fengqia95f3aa2018-07-18 16:28:03 +08002163 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002164 fs_info->nodesize,
2165 GFP_NOFS);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002166 if (ret)
2167 goto out;
2168 }
2169
2170 if (level == 0) {
Lu Fengqi8d38d7e2018-07-18 14:45:37 +08002171 ret = btrfs_qgroup_trace_leaf_items(trans,
2172 path->nodes[level]);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002173 if (ret)
2174 goto out;
2175
2176 /* Nonzero return here means we completed our search */
David Sterba15b34512017-02-10 20:30:23 +01002177 ret = adjust_slots_upwards(path, root_level);
Qu Wenruo33d1f052016-10-18 09:31:28 +08002178 if (ret)
2179 break;
2180
2181 /* Restart search with new slots */
2182 goto walk_down;
2183 }
2184
2185 level--;
2186 }
2187
2188 ret = 0;
2189out:
2190 btrfs_free_path(path);
2191
2192 return ret;
2193}
2194
Qu Wenruod810ef22015-04-12 16:52:34 +08002195#define UPDATE_NEW 0
2196#define UPDATE_OLD 1
2197/*
2198 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2199 */
2200static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2201 struct ulist *roots, struct ulist *tmp,
2202 struct ulist *qgroups, u64 seq, int update_old)
2203{
2204 struct ulist_node *unode;
2205 struct ulist_iterator uiter;
2206 struct ulist_node *tmp_unode;
2207 struct ulist_iterator tmp_uiter;
2208 struct btrfs_qgroup *qg;
2209 int ret = 0;
2210
2211 if (!roots)
2212 return 0;
2213 ULIST_ITER_INIT(&uiter);
2214 while ((unode = ulist_next(roots, &uiter))) {
2215 qg = find_qgroup_rb(fs_info, unode->val);
2216 if (!qg)
2217 continue;
2218
2219 ulist_reinit(tmp);
David Sterbaef2fff62016-10-26 16:23:50 +02002220 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
Qu Wenruod810ef22015-04-12 16:52:34 +08002221 GFP_ATOMIC);
2222 if (ret < 0)
2223 return ret;
David Sterbaef2fff62016-10-26 16:23:50 +02002224 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
Qu Wenruod810ef22015-04-12 16:52:34 +08002225 if (ret < 0)
2226 return ret;
2227 ULIST_ITER_INIT(&tmp_uiter);
2228 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2229 struct btrfs_qgroup_list *glist;
2230
David Sterbaef2fff62016-10-26 16:23:50 +02002231 qg = unode_aux_to_qgroup(tmp_unode);
Qu Wenruod810ef22015-04-12 16:52:34 +08002232 if (update_old)
2233 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2234 else
2235 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2236 list_for_each_entry(glist, &qg->groups, next_group) {
2237 ret = ulist_add(qgroups, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02002238 qgroup_to_aux(glist->group),
Qu Wenruod810ef22015-04-12 16:52:34 +08002239 GFP_ATOMIC);
2240 if (ret < 0)
2241 return ret;
2242 ret = ulist_add(tmp, glist->group->qgroupid,
David Sterbaef2fff62016-10-26 16:23:50 +02002243 qgroup_to_aux(glist->group),
Qu Wenruod810ef22015-04-12 16:52:34 +08002244 GFP_ATOMIC);
2245 if (ret < 0)
2246 return ret;
2247 }
2248 }
2249 }
2250 return 0;
2251}
2252
Josef Bacikfcebe452014-05-13 17:30:47 -07002253/*
Qu Wenruo823ae5b2015-04-12 16:59:57 +08002254 * Update qgroup rfer/excl counters.
2255 * Rfer update is easy, codes can explain themselves.
Qu Wenruoe69bcee2015-04-17 10:23:16 +08002256 *
Qu Wenruo823ae5b2015-04-12 16:59:57 +08002257 * Excl update is tricky, the update is split into 2 part.
2258 * Part 1: Possible exclusive <-> sharing detect:
2259 * | A | !A |
2260 * -------------------------------------
2261 * B | * | - |
2262 * -------------------------------------
2263 * !B | + | ** |
2264 * -------------------------------------
2265 *
2266 * Conditions:
2267 * A: cur_old_roots < nr_old_roots (not exclusive before)
2268 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2269 * B: cur_new_roots < nr_new_roots (not exclusive now)
Nicholas D Steeves01327612016-05-19 21:18:45 -04002270 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
Qu Wenruo823ae5b2015-04-12 16:59:57 +08002271 *
2272 * Results:
2273 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2274 * *: Definitely not changed. **: Possible unchanged.
2275 *
2276 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2277 *
2278 * To make the logic clear, we first use condition A and B to split
2279 * combination into 4 results.
2280 *
2281 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2282 * only on variant maybe 0.
2283 *
2284 * Lastly, check result **, since there are 2 variants maybe 0, split them
2285 * again(2x2).
2286 * But this time we don't need to consider other things, the codes and logic
2287 * is easy to understand now.
2288 */
2289static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2290 struct ulist *qgroups,
2291 u64 nr_old_roots,
2292 u64 nr_new_roots,
2293 u64 num_bytes, u64 seq)
2294{
2295 struct ulist_node *unode;
2296 struct ulist_iterator uiter;
2297 struct btrfs_qgroup *qg;
2298 u64 cur_new_count, cur_old_count;
2299
2300 ULIST_ITER_INIT(&uiter);
2301 while ((unode = ulist_next(qgroups, &uiter))) {
2302 bool dirty = false;
2303
David Sterbaef2fff62016-10-26 16:23:50 +02002304 qg = unode_aux_to_qgroup(unode);
Qu Wenruo823ae5b2015-04-12 16:59:57 +08002305 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2306 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2307
Qu Wenruo8b317902018-04-30 15:04:44 +08002308 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2309 cur_new_count);
Mark Fasheh0f5dcf82016-03-29 17:19:55 -07002310
Qu Wenruo823ae5b2015-04-12 16:59:57 +08002311 /* Rfer update part */
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 if (cur_old_count > 0 && cur_new_count == 0) {
2318 qg->rfer -= num_bytes;
2319 qg->rfer_cmpr -= num_bytes;
2320 dirty = true;
2321 }
2322
2323 /* Excl update part */
2324 /* Exclusive/none -> shared case */
2325 if (cur_old_count == nr_old_roots &&
2326 cur_new_count < nr_new_roots) {
2327 /* Exclusive -> shared */
2328 if (cur_old_count != 0) {
2329 qg->excl -= num_bytes;
2330 qg->excl_cmpr -= num_bytes;
2331 dirty = true;
2332 }
2333 }
2334
2335 /* Shared -> exclusive/none case */
2336 if (cur_old_count < nr_old_roots &&
2337 cur_new_count == nr_new_roots) {
2338 /* Shared->exclusive */
2339 if (cur_new_count != 0) {
2340 qg->excl += num_bytes;
2341 qg->excl_cmpr += num_bytes;
2342 dirty = true;
2343 }
2344 }
2345
2346 /* Exclusive/none -> exclusive/none case */
2347 if (cur_old_count == nr_old_roots &&
2348 cur_new_count == nr_new_roots) {
2349 if (cur_old_count == 0) {
2350 /* None -> exclusive/none */
2351
2352 if (cur_new_count != 0) {
2353 /* None -> exclusive */
2354 qg->excl += num_bytes;
2355 qg->excl_cmpr += num_bytes;
2356 dirty = true;
2357 }
2358 /* None -> none, nothing changed */
2359 } else {
2360 /* Exclusive -> exclusive/none */
2361
2362 if (cur_new_count == 0) {
2363 /* Exclusive -> none */
2364 qg->excl -= num_bytes;
2365 qg->excl_cmpr -= num_bytes;
2366 dirty = true;
2367 }
2368 /* Exclusive -> exclusive, nothing changed */
2369 }
2370 }
Qu Wenruoc05f9422015-08-03 14:44:29 +08002371
Qu Wenruo823ae5b2015-04-12 16:59:57 +08002372 if (dirty)
2373 qgroup_dirty(fs_info, qg);
2374 }
2375 return 0;
2376}
2377
Qu Wenruo5edfd9f2017-02-27 15:10:34 +08002378/*
2379 * Check if the @roots potentially is a list of fs tree roots
2380 *
2381 * Return 0 for definitely not a fs/subvol tree roots ulist
2382 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2383 * one as well)
2384 */
2385static int maybe_fs_roots(struct ulist *roots)
2386{
2387 struct ulist_node *unode;
2388 struct ulist_iterator uiter;
2389
2390 /* Empty one, still possible for fs roots */
2391 if (!roots || roots->nnodes == 0)
2392 return 1;
2393
2394 ULIST_ITER_INIT(&uiter);
2395 unode = ulist_next(roots, &uiter);
2396 if (!unode)
2397 return 1;
2398
2399 /*
2400 * If it contains fs tree roots, then it must belong to fs/subvol
2401 * trees.
2402 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2403 */
2404 return is_fstree(unode->val);
2405}
2406
Lu Fengqi8696d762018-07-18 14:45:39 +08002407int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2408 u64 num_bytes, struct ulist *old_roots,
2409 struct ulist *new_roots)
Qu Wenruo550d7a22015-04-16 15:37:33 +08002410{
Lu Fengqi8696d762018-07-18 14:45:39 +08002411 struct btrfs_fs_info *fs_info = trans->fs_info;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002412 struct ulist *qgroups = NULL;
2413 struct ulist *tmp = NULL;
2414 u64 seq;
2415 u64 nr_new_roots = 0;
2416 u64 nr_old_roots = 0;
2417 int ret = 0;
2418
Johannes Thumshirn26ef8492020-01-08 21:07:32 +09002419 /*
2420 * If quotas get disabled meanwhile, the resouces need to be freed and
2421 * we can't just exit here.
2422 */
David Sterba81353d52017-02-13 14:05:24 +01002423 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Johannes Thumshirn26ef8492020-01-08 21:07:32 +09002424 goto out_free;
David Sterba81353d52017-02-13 14:05:24 +01002425
Qu Wenruo5edfd9f2017-02-27 15:10:34 +08002426 if (new_roots) {
2427 if (!maybe_fs_roots(new_roots))
2428 goto out_free;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002429 nr_new_roots = new_roots->nnodes;
Qu Wenruo5edfd9f2017-02-27 15:10:34 +08002430 }
2431 if (old_roots) {
2432 if (!maybe_fs_roots(old_roots))
2433 goto out_free;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002434 nr_old_roots = old_roots->nnodes;
Qu Wenruo5edfd9f2017-02-27 15:10:34 +08002435 }
2436
2437 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2438 if (nr_old_roots == 0 && nr_new_roots == 0)
2439 goto out_free;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002440
Qu Wenruo550d7a22015-04-16 15:37:33 +08002441 BUG_ON(!fs_info->quota_root);
2442
Qu Wenruoc9f6f3c2018-05-03 09:59:02 +08002443 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2444 num_bytes, nr_old_roots, nr_new_roots);
Mark Fasheh0f5dcf82016-03-29 17:19:55 -07002445
Qu Wenruo550d7a22015-04-16 15:37:33 +08002446 qgroups = ulist_alloc(GFP_NOFS);
2447 if (!qgroups) {
2448 ret = -ENOMEM;
2449 goto out_free;
2450 }
2451 tmp = ulist_alloc(GFP_NOFS);
2452 if (!tmp) {
2453 ret = -ENOMEM;
2454 goto out_free;
2455 }
2456
2457 mutex_lock(&fs_info->qgroup_rescan_lock);
2458 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2459 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2460 mutex_unlock(&fs_info->qgroup_rescan_lock);
2461 ret = 0;
2462 goto out_free;
2463 }
2464 }
2465 mutex_unlock(&fs_info->qgroup_rescan_lock);
2466
2467 spin_lock(&fs_info->qgroup_lock);
2468 seq = fs_info->qgroup_seq;
2469
2470 /* Update old refcnts using old_roots */
2471 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2472 UPDATE_OLD);
2473 if (ret < 0)
2474 goto out;
2475
2476 /* Update new refcnts using new_roots */
2477 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2478 UPDATE_NEW);
2479 if (ret < 0)
2480 goto out;
2481
2482 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2483 num_bytes, seq);
2484
2485 /*
2486 * Bump qgroup_seq to avoid seq overlap
2487 */
2488 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2489out:
2490 spin_unlock(&fs_info->qgroup_lock);
2491out_free:
2492 ulist_free(tmp);
2493 ulist_free(qgroups);
2494 ulist_free(old_roots);
2495 ulist_free(new_roots);
2496 return ret;
2497}
2498
Nikolay Borisov460fb202018-03-15 16:00:25 +02002499int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
Qu Wenruo550d7a22015-04-16 15:37:33 +08002500{
Nikolay Borisov460fb202018-03-15 16:00:25 +02002501 struct btrfs_fs_info *fs_info = trans->fs_info;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002502 struct btrfs_qgroup_extent_record *record;
2503 struct btrfs_delayed_ref_root *delayed_refs;
2504 struct ulist *new_roots = NULL;
2505 struct rb_node *node;
Qu Wenruoc337e7b2018-09-27 14:42:29 +08002506 u64 num_dirty_extents = 0;
Qu Wenruo9086db82015-04-20 09:53:50 +08002507 u64 qgroup_to_skip;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002508 int ret = 0;
2509
2510 delayed_refs = &trans->transaction->delayed_refs;
Qu Wenruo9086db82015-04-20 09:53:50 +08002511 qgroup_to_skip = delayed_refs->qgroup_to_skip;
Qu Wenruo550d7a22015-04-16 15:37:33 +08002512 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2513 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2514 node);
2515
Qu Wenruoc337e7b2018-09-27 14:42:29 +08002516 num_dirty_extents++;
Jeff Mahoneybc074522016-06-09 17:27:55 -04002517 trace_btrfs_qgroup_account_extents(fs_info, record);
Mark Fasheh0f5dcf82016-03-29 17:19:55 -07002518
Qu Wenruo550d7a22015-04-16 15:37:33 +08002519 if (!ret) {
2520 /*
Qu Wenruod1b8b942017-02-27 15:10:35 +08002521 * Old roots should be searched when inserting qgroup
2522 * extent record
2523 */
2524 if (WARN_ON(!record->old_roots)) {
2525 /* Search commit root to find old_roots */
2526 ret = btrfs_find_all_roots(NULL, fs_info,
2527 record->bytenr, 0,
Zygo Blaxellc995ab32017-09-22 13:58:45 -04002528 &record->old_roots, false);
Qu Wenruod1b8b942017-02-27 15:10:35 +08002529 if (ret < 0)
2530 goto cleanup;
2531 }
2532
Qu Wenruo1418bae2019-01-23 15:15:12 +08002533 /* Free the reserved data space */
2534 btrfs_qgroup_free_refroot(fs_info,
2535 record->data_rsv_refroot,
2536 record->data_rsv,
2537 BTRFS_QGROUP_RSV_DATA);
Qu Wenruod1b8b942017-02-27 15:10:35 +08002538 /*
Edmund Nadolskide47c9d2017-03-16 10:04:34 -06002539 * Use SEQ_LAST as time_seq to do special search, which
Qu Wenruo550d7a22015-04-16 15:37:33 +08002540 * doesn't lock tree or delayed_refs and search current
2541 * root. It's safe inside commit_transaction().
2542 */
2543 ret = btrfs_find_all_roots(trans, fs_info,
Zygo Blaxellc995ab32017-09-22 13:58:45 -04002544 record->bytenr, SEQ_LAST, &new_roots, false);
Qu Wenruo550d7a22015-04-16 15:37:33 +08002545 if (ret < 0)
2546 goto cleanup;
Qu Wenruod1b8b942017-02-27 15:10:35 +08002547 if (qgroup_to_skip) {
Qu Wenruo9086db82015-04-20 09:53:50 +08002548 ulist_del(new_roots, qgroup_to_skip, 0);
Qu Wenruod1b8b942017-02-27 15:10:35 +08002549 ulist_del(record->old_roots, qgroup_to_skip,
2550 0);
2551 }
Lu Fengqi8696d762018-07-18 14:45:39 +08002552 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2553 record->num_bytes,
2554 record->old_roots,
2555 new_roots);
Qu Wenruo550d7a22015-04-16 15:37:33 +08002556 record->old_roots = NULL;
2557 new_roots = NULL;
2558 }
2559cleanup:
2560 ulist_free(record->old_roots);
2561 ulist_free(new_roots);
2562 new_roots = NULL;
2563 rb_erase(node, &delayed_refs->dirty_extent_root);
2564 kfree(record);
2565
2566 }
Qu Wenruoc337e7b2018-09-27 14:42:29 +08002567 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2568 num_dirty_extents);
Qu Wenruo550d7a22015-04-16 15:37:33 +08002569 return ret;
2570}
2571
Josef Bacikfcebe452014-05-13 17:30:47 -07002572/*
Arne Jansenbed92ea2012-06-28 18:03:02 +02002573 * called from commit_transaction. Writes all changed qgroups to disk.
2574 */
Lu Fengqi280f8bd2018-07-18 14:45:40 +08002575int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002576{
Lu Fengqi280f8bd2018-07-18 14:45:40 +08002577 struct btrfs_fs_info *fs_info = trans->fs_info;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002578 int ret = 0;
2579
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03002580 if (!fs_info->quota_root)
Nikolay Borisov5d235152018-01-31 10:52:04 +02002581 return ret;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002582
2583 spin_lock(&fs_info->qgroup_lock);
2584 while (!list_empty(&fs_info->dirty_qgroups)) {
2585 struct btrfs_qgroup *qgroup;
2586 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2587 struct btrfs_qgroup, dirty);
2588 list_del_init(&qgroup->dirty);
2589 spin_unlock(&fs_info->qgroup_lock);
Lu Fengqi3e07e9a2018-07-18 14:45:28 +08002590 ret = update_qgroup_info_item(trans, qgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002591 if (ret)
2592 fs_info->qgroup_flags |=
2593 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Lu Fengqiac8a8662018-07-18 14:45:27 +08002594 ret = update_qgroup_limit_item(trans, qgroup);
Dongsheng Yangd3001ed2014-11-20 21:04:56 -05002595 if (ret)
2596 fs_info->qgroup_flags |=
2597 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002598 spin_lock(&fs_info->qgroup_lock);
2599 }
Josef Bacikafcdd122016-09-02 15:40:02 -04002600 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Arne Jansenbed92ea2012-06-28 18:03:02 +02002601 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2602 else
2603 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2604 spin_unlock(&fs_info->qgroup_lock);
2605
Lu Fengqi2e980ac2018-07-18 14:45:29 +08002606 ret = update_qgroup_status_item(trans);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002607 if (ret)
2608 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2609
Arne Jansenbed92ea2012-06-28 18:03:02 +02002610 return ret;
2611}
2612
2613/*
Nicholas D Steeves01327612016-05-19 21:18:45 -04002614 * Copy the accounting information between qgroups. This is necessary
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002615 * when a snapshot or a subvolume is created. Throwing an error will
2616 * cause a transaction abort so we take extra care here to only error
2617 * when a readonly fs is a reasonable outcome.
Arne Jansenbed92ea2012-06-28 18:03:02 +02002618 */
Lu Fengqia93774222018-07-18 14:45:41 +08002619int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2620 u64 objectid, struct btrfs_qgroup_inherit *inherit)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002621{
2622 int ret = 0;
2623 int i;
2624 u64 *i_qgroups;
Qu Wenruoe88439d2019-06-13 17:31:24 +08002625 bool committing = false;
Lu Fengqia93774222018-07-18 14:45:41 +08002626 struct btrfs_fs_info *fs_info = trans->fs_info;
Filipe Manana552f0322018-11-19 16:20:34 +00002627 struct btrfs_root *quota_root;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002628 struct btrfs_qgroup *srcgroup;
2629 struct btrfs_qgroup *dstgroup;
2630 u32 level_size = 0;
Wang Shilong3f5e2d32013-04-07 10:50:19 +00002631 u64 nums;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002632
Qu Wenruoe88439d2019-06-13 17:31:24 +08002633 /*
2634 * There are only two callers of this function.
2635 *
2636 * One in create_subvol() in the ioctl context, which needs to hold
2637 * the qgroup_ioctl_lock.
2638 *
2639 * The other one in create_pending_snapshot() where no other qgroup
2640 * code can modify the fs as they all need to either start a new trans
2641 * or hold a trans handler, thus we don't need to hold
2642 * qgroup_ioctl_lock.
2643 * This would avoid long and complex lock chain and make lockdep happy.
2644 */
2645 spin_lock(&fs_info->trans_lock);
2646 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2647 committing = true;
2648 spin_unlock(&fs_info->trans_lock);
2649
2650 if (!committing)
2651 mutex_lock(&fs_info->qgroup_ioctl_lock);
Josef Bacikafcdd122016-09-02 15:40:02 -04002652 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
Wang Shilongf2f6ed32013-04-07 10:50:16 +00002653 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002654
Filipe Manana552f0322018-11-19 16:20:34 +00002655 quota_root = fs_info->quota_root;
Wang Shilongf2f6ed32013-04-07 10:50:16 +00002656 if (!quota_root) {
2657 ret = -EINVAL;
2658 goto out;
2659 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002660
Wang Shilong3f5e2d32013-04-07 10:50:19 +00002661 if (inherit) {
2662 i_qgroups = (u64 *)(inherit + 1);
2663 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2664 2 * inherit->num_excl_copies;
2665 for (i = 0; i < nums; ++i) {
2666 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
Dongsheng Yang09870d22014-11-11 07:18:22 -05002667
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002668 /*
2669 * Zero out invalid groups so we can ignore
2670 * them later.
2671 */
2672 if (!srcgroup ||
2673 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2674 *i_qgroups = 0ULL;
2675
Wang Shilong3f5e2d32013-04-07 10:50:19 +00002676 ++i_qgroups;
2677 }
2678 }
2679
Arne Jansenbed92ea2012-06-28 18:03:02 +02002680 /*
2681 * create a tracking group for the subvol itself
2682 */
2683 ret = add_qgroup_item(trans, quota_root, objectid);
2684 if (ret)
2685 goto out;
2686
Arne Jansenbed92ea2012-06-28 18:03:02 +02002687 /*
2688 * add qgroup to all inherited groups
2689 */
2690 if (inherit) {
2691 i_qgroups = (u64 *)(inherit + 1);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002692 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2693 if (*i_qgroups == 0)
2694 continue;
Lu Fengqi711169c2018-07-18 14:45:24 +08002695 ret = add_qgroup_relation_item(trans, objectid,
2696 *i_qgroups);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002697 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002698 goto out;
Lu Fengqi711169c2018-07-18 14:45:24 +08002699 ret = add_qgroup_relation_item(trans, *i_qgroups,
2700 objectid);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002701 if (ret && ret != -EEXIST)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002702 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002703 }
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002704 ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002705 }
2706
2707
2708 spin_lock(&fs_info->qgroup_lock);
2709
2710 dstgroup = add_qgroup_rb(fs_info, objectid);
Dan Carpenter57a5a882012-07-30 02:15:43 -06002711 if (IS_ERR(dstgroup)) {
2712 ret = PTR_ERR(dstgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002713 goto unlock;
Dan Carpenter57a5a882012-07-30 02:15:43 -06002714 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002715
Dongsheng Yange8c85412014-11-20 20:58:34 -05002716 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
Dongsheng Yange8c85412014-11-20 20:58:34 -05002717 dstgroup->lim_flags = inherit->lim.flags;
2718 dstgroup->max_rfer = inherit->lim.max_rfer;
2719 dstgroup->max_excl = inherit->lim.max_excl;
2720 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2721 dstgroup->rsv_excl = inherit->lim.rsv_excl;
Dongsheng Yang1510e712014-11-20 21:01:41 -05002722
Lu Fengqiac8a8662018-07-18 14:45:27 +08002723 ret = update_qgroup_limit_item(trans, dstgroup);
Dongsheng Yang1510e712014-11-20 21:01:41 -05002724 if (ret) {
2725 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
Jeff Mahoney5d163e02016-09-20 10:05:00 -04002726 btrfs_info(fs_info,
2727 "unable to update quota limit for %llu",
2728 dstgroup->qgroupid);
Dongsheng Yang1510e712014-11-20 21:01:41 -05002729 goto unlock;
2730 }
Dongsheng Yange8c85412014-11-20 20:58:34 -05002731 }
2732
Arne Jansenbed92ea2012-06-28 18:03:02 +02002733 if (srcid) {
2734 srcgroup = find_qgroup_rb(fs_info, srcid);
Chris Masonf3a87f12012-09-14 20:06:30 -04002735 if (!srcgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002736 goto unlock;
Josef Bacikfcebe452014-05-13 17:30:47 -07002737
2738 /*
2739 * We call inherit after we clone the root in order to make sure
2740 * our counts don't go crazy, so at this point the only
2741 * difference between the two roots should be the root node.
2742 */
Lu Fengqic8389d42018-07-17 16:58:22 +08002743 level_size = fs_info->nodesize;
Josef Bacikfcebe452014-05-13 17:30:47 -07002744 dstgroup->rfer = srcgroup->rfer;
2745 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2746 dstgroup->excl = level_size;
2747 dstgroup->excl_cmpr = level_size;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002748 srcgroup->excl = level_size;
2749 srcgroup->excl_cmpr = level_size;
Dongsheng Yang3eeb4d52014-11-20 20:14:38 -05002750
2751 /* inherit the limit info */
2752 dstgroup->lim_flags = srcgroup->lim_flags;
2753 dstgroup->max_rfer = srcgroup->max_rfer;
2754 dstgroup->max_excl = srcgroup->max_excl;
2755 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2756 dstgroup->rsv_excl = srcgroup->rsv_excl;
2757
Arne Jansenbed92ea2012-06-28 18:03:02 +02002758 qgroup_dirty(fs_info, dstgroup);
2759 qgroup_dirty(fs_info, srcgroup);
2760 }
2761
Chris Masonf3a87f12012-09-14 20:06:30 -04002762 if (!inherit)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002763 goto unlock;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002764
2765 i_qgroups = (u64 *)(inherit + 1);
2766 for (i = 0; i < inherit->num_qgroups; ++i) {
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002767 if (*i_qgroups) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002768 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002769 if (ret)
2770 goto unlock;
2771 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002772 ++i_qgroups;
2773 }
2774
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002775 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002776 struct btrfs_qgroup *src;
2777 struct btrfs_qgroup *dst;
2778
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002779 if (!i_qgroups[0] || !i_qgroups[1])
2780 continue;
2781
Arne Jansenbed92ea2012-06-28 18:03:02 +02002782 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2783 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2784
2785 if (!src || !dst) {
2786 ret = -EINVAL;
2787 goto unlock;
2788 }
2789
2790 dst->rfer = src->rfer - level_size;
2791 dst->rfer_cmpr = src->rfer_cmpr - level_size;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002792 }
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002793 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002794 struct btrfs_qgroup *src;
2795 struct btrfs_qgroup *dst;
2796
Mark Fasheh918c2ee2016-03-30 17:57:48 -07002797 if (!i_qgroups[0] || !i_qgroups[1])
2798 continue;
2799
Arne Jansenbed92ea2012-06-28 18:03:02 +02002800 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2801 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2802
2803 if (!src || !dst) {
2804 ret = -EINVAL;
2805 goto unlock;
2806 }
2807
2808 dst->excl = src->excl + level_size;
2809 dst->excl_cmpr = src->excl_cmpr + level_size;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002810 }
2811
2812unlock:
2813 spin_unlock(&fs_info->qgroup_lock);
2814out:
Qu Wenruoe88439d2019-06-13 17:31:24 +08002815 if (!committing)
2816 mutex_unlock(&fs_info->qgroup_ioctl_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002817 return ret;
2818}
2819
Qu Wenruoa514d632017-12-22 16:06:39 +08002820/*
2821 * Two limits to commit transaction in advance.
2822 *
Qu Wenruof5fef452019-01-25 07:55:27 +08002823 * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
Qu Wenruoa514d632017-12-22 16:06:39 +08002824 * For SIZE, it will be in byte unit as threshold.
2825 */
Qu Wenruof5fef452019-01-25 07:55:27 +08002826#define QGROUP_FREE_RATIO 32
2827#define QGROUP_FREE_SIZE SZ_32M
Qu Wenruoa514d632017-12-22 16:06:39 +08002828static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2829 const struct btrfs_qgroup *qg, u64 num_bytes)
Jeff Mahoney003d7c52017-01-25 09:50:33 -05002830{
Qu Wenruof5fef452019-01-25 07:55:27 +08002831 u64 free;
Qu Wenruoa514d632017-12-22 16:06:39 +08002832 u64 threshold;
2833
Jeff Mahoney003d7c52017-01-25 09:50:33 -05002834 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
Qu Wenruodba21322017-12-12 15:34:25 +08002835 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
Jeff Mahoney003d7c52017-01-25 09:50:33 -05002836 return false;
2837
2838 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
Qu Wenruodba21322017-12-12 15:34:25 +08002839 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
Jeff Mahoney003d7c52017-01-25 09:50:33 -05002840 return false;
2841
Qu Wenruoa514d632017-12-22 16:06:39 +08002842 /*
2843 * Even if we passed the check, it's better to check if reservation
2844 * for meta_pertrans is pushing us near limit.
2845 * If there is too much pertrans reservation or it's near the limit,
2846 * let's try commit transaction to free some, using transaction_kthread
2847 */
2848 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2849 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
Qu Wenruof5fef452019-01-25 07:55:27 +08002850 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2851 free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2852 threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2853 QGROUP_FREE_SIZE);
2854 } else {
2855 free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2856 threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2857 QGROUP_FREE_SIZE);
2858 }
Qu Wenruoa514d632017-12-22 16:06:39 +08002859
2860 /*
2861 * Use transaction_kthread to commit transaction, so we no
2862 * longer need to bother nested transaction nor lock context.
2863 */
Qu Wenruof5fef452019-01-25 07:55:27 +08002864 if (free < threshold)
Qu Wenruoa514d632017-12-22 16:06:39 +08002865 btrfs_commit_transaction_locksafe(fs_info);
2866 }
2867
Jeff Mahoney003d7c52017-01-25 09:50:33 -05002868 return true;
2869}
2870
Qu Wenruodba21322017-12-12 15:34:25 +08002871static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2872 enum btrfs_qgroup_rsv_type type)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002873{
Arne Jansenbed92ea2012-06-28 18:03:02 +02002874 struct btrfs_qgroup *qgroup;
2875 struct btrfs_fs_info *fs_info = root->fs_info;
2876 u64 ref_root = root->root_key.objectid;
2877 int ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002878 struct ulist_node *unode;
2879 struct ulist_iterator uiter;
2880
2881 if (!is_fstree(ref_root))
2882 return 0;
2883
2884 if (num_bytes == 0)
2885 return 0;
Sargun Dhillonf29efe22017-05-11 21:17:33 +00002886
2887 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2888 capable(CAP_SYS_RESOURCE))
2889 enforce = false;
2890
Arne Jansenbed92ea2012-06-28 18:03:02 +02002891 spin_lock(&fs_info->qgroup_lock);
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03002892 if (!fs_info->quota_root)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002893 goto out;
2894
2895 qgroup = find_qgroup_rb(fs_info, ref_root);
2896 if (!qgroup)
2897 goto out;
2898
2899 /*
2900 * in a first step, we check all affected qgroups if any limits would
2901 * be exceeded
2902 */
Wang Shilong1e8f9152013-05-06 11:03:27 +00002903 ulist_reinit(fs_info->qgroup_ulist);
2904 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
David Sterbaa1840b52018-03-27 19:04:50 +02002905 qgroup_to_aux(qgroup), GFP_ATOMIC);
Wang Shilong3c971852013-04-17 14:00:36 +00002906 if (ret < 0)
2907 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002908 ULIST_ITER_INIT(&uiter);
Wang Shilong1e8f9152013-05-06 11:03:27 +00002909 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002910 struct btrfs_qgroup *qg;
2911 struct btrfs_qgroup_list *glist;
2912
David Sterbaef2fff62016-10-26 16:23:50 +02002913 qg = unode_aux_to_qgroup(unode);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002914
Qu Wenruoa514d632017-12-22 16:06:39 +08002915 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002916 ret = -EDQUOT;
Wang Shilong720f1e22013-03-06 11:51:47 +00002917 goto out;
2918 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002919
2920 list_for_each_entry(glist, &qg->groups, next_group) {
Wang Shilong1e8f9152013-05-06 11:03:27 +00002921 ret = ulist_add(fs_info->qgroup_ulist,
2922 glist->group->qgroupid,
David Sterbaa1840b52018-03-27 19:04:50 +02002923 qgroup_to_aux(glist->group), GFP_ATOMIC);
Wang Shilong3c971852013-04-17 14:00:36 +00002924 if (ret < 0)
2925 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002926 }
2927 }
Wang Shilong3c971852013-04-17 14:00:36 +00002928 ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002929 /*
2930 * no limits exceeded, now record the reservation into all qgroups
2931 */
2932 ULIST_ITER_INIT(&uiter);
Wang Shilong1e8f9152013-05-06 11:03:27 +00002933 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002934 struct btrfs_qgroup *qg;
2935
David Sterbaef2fff62016-10-26 16:23:50 +02002936 qg = unode_aux_to_qgroup(unode);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002937
Qu Wenruo64ee4e72017-12-12 15:34:27 +08002938 qgroup_rsv_add(fs_info, qg, num_bytes, type);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002939 }
2940
2941out:
2942 spin_unlock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002943 return ret;
2944}
2945
Qu Wenruoe1211d02017-12-12 15:34:30 +08002946/*
2947 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2948 * qgroup).
2949 *
2950 * Will handle all higher level qgroup too.
2951 *
2952 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2953 * This special case is only used for META_PERTRANS type.
2954 */
Qu Wenruo297d7502015-09-08 17:08:37 +08002955void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
Qu Wenruod4e5c922017-12-12 15:34:23 +08002956 u64 ref_root, u64 num_bytes,
2957 enum btrfs_qgroup_rsv_type type)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002958{
Arne Jansenbed92ea2012-06-28 18:03:02 +02002959 struct btrfs_qgroup *qgroup;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002960 struct ulist_node *unode;
2961 struct ulist_iterator uiter;
Wang Shilong3c971852013-04-17 14:00:36 +00002962 int ret = 0;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002963
2964 if (!is_fstree(ref_root))
2965 return;
2966
2967 if (num_bytes == 0)
2968 return;
2969
Qu Wenruoe1211d02017-12-12 15:34:30 +08002970 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2971 WARN(1, "%s: Invalid type to free", __func__);
2972 return;
2973 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02002974 spin_lock(&fs_info->qgroup_lock);
2975
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03002976 if (!fs_info->quota_root)
Arne Jansenbed92ea2012-06-28 18:03:02 +02002977 goto out;
2978
2979 qgroup = find_qgroup_rb(fs_info, ref_root);
2980 if (!qgroup)
2981 goto out;
2982
Qu Wenruoe1211d02017-12-12 15:34:30 +08002983 if (num_bytes == (u64)-1)
Qu Wenruo82874752017-12-12 15:34:34 +08002984 /*
2985 * We're freeing all pertrans rsv, get reserved value from
2986 * level 0 qgroup as real num_bytes to free.
2987 */
Qu Wenruoe1211d02017-12-12 15:34:30 +08002988 num_bytes = qgroup->rsv.values[type];
2989
Wang Shilong1e8f9152013-05-06 11:03:27 +00002990 ulist_reinit(fs_info->qgroup_ulist);
2991 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
David Sterbaa1840b52018-03-27 19:04:50 +02002992 qgroup_to_aux(qgroup), GFP_ATOMIC);
Wang Shilong3c971852013-04-17 14:00:36 +00002993 if (ret < 0)
2994 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02002995 ULIST_ITER_INIT(&uiter);
Wang Shilong1e8f9152013-05-06 11:03:27 +00002996 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
Arne Jansenbed92ea2012-06-28 18:03:02 +02002997 struct btrfs_qgroup *qg;
2998 struct btrfs_qgroup_list *glist;
2999
David Sterbaef2fff62016-10-26 16:23:50 +02003000 qg = unode_aux_to_qgroup(unode);
Arne Jansenbed92ea2012-06-28 18:03:02 +02003001
Qu Wenruo64ee4e72017-12-12 15:34:27 +08003002 qgroup_rsv_release(fs_info, qg, num_bytes, type);
Arne Jansenbed92ea2012-06-28 18:03:02 +02003003
3004 list_for_each_entry(glist, &qg->groups, next_group) {
Wang Shilong1e8f9152013-05-06 11:03:27 +00003005 ret = ulist_add(fs_info->qgroup_ulist,
3006 glist->group->qgroupid,
David Sterbaa1840b52018-03-27 19:04:50 +02003007 qgroup_to_aux(glist->group), GFP_ATOMIC);
Wang Shilong3c971852013-04-17 14:00:36 +00003008 if (ret < 0)
3009 goto out;
Arne Jansenbed92ea2012-06-28 18:03:02 +02003010 }
3011 }
3012
3013out:
3014 spin_unlock(&fs_info->qgroup_lock);
Arne Jansenbed92ea2012-06-28 18:03:02 +02003015}
3016
Jan Schmidt2f232032013-04-25 16:04:51 +00003017/*
Qu Wenruoff3d27a02018-05-14 09:38:13 +08003018 * Check if the leaf is the last leaf. Which means all node pointers
3019 * are at their last position.
3020 */
3021static bool is_last_leaf(struct btrfs_path *path)
3022{
3023 int i;
3024
3025 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3026 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3027 return false;
3028 }
3029 return true;
3030}
3031
3032/*
Jan Schmidt2f232032013-04-25 16:04:51 +00003033 * returns < 0 on error, 0 when more leafs are to be scanned.
Qu Wenruo33931682015-02-27 16:24:24 +08003034 * returns 1 when done.
Jan Schmidt2f232032013-04-25 16:04:51 +00003035 */
Lu Fengqi62088ca2018-07-18 14:45:42 +08003036static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3037 struct btrfs_path *path)
Jan Schmidt2f232032013-04-25 16:04:51 +00003038{
Lu Fengqi62088ca2018-07-18 14:45:42 +08003039 struct btrfs_fs_info *fs_info = trans->fs_info;
Jan Schmidt2f232032013-04-25 16:04:51 +00003040 struct btrfs_key found;
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08003041 struct extent_buffer *scratch_leaf = NULL;
Jan Schmidt2f232032013-04-25 16:04:51 +00003042 struct ulist *roots = NULL;
Josef Bacikfcebe452014-05-13 17:30:47 -07003043 u64 num_bytes;
Qu Wenruoff3d27a02018-05-14 09:38:13 +08003044 bool done;
Jan Schmidt2f232032013-04-25 16:04:51 +00003045 int slot;
3046 int ret;
3047
Jan Schmidt2f232032013-04-25 16:04:51 +00003048 mutex_lock(&fs_info->qgroup_rescan_lock);
3049 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3050 &fs_info->qgroup_rescan_progress,
3051 path, 1, 0);
3052
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04003053 btrfs_debug(fs_info,
3054 "current progress key (%llu %u %llu), search_slot ret %d",
3055 fs_info->qgroup_rescan_progress.objectid,
3056 fs_info->qgroup_rescan_progress.type,
3057 fs_info->qgroup_rescan_progress.offset, ret);
Jan Schmidt2f232032013-04-25 16:04:51 +00003058
3059 if (ret) {
3060 /*
3061 * The rescan is about to end, we will not be scanning any
3062 * further blocks. We cannot unset the RESCAN flag here, because
3063 * we want to commit the transaction if everything went well.
3064 * To make the live accounting work in this phase, we set our
3065 * scan progress pointer such that every real extent objectid
3066 * will be smaller.
3067 */
3068 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3069 btrfs_release_path(path);
3070 mutex_unlock(&fs_info->qgroup_rescan_lock);
3071 return ret;
3072 }
Qu Wenruoff3d27a02018-05-14 09:38:13 +08003073 done = is_last_leaf(path);
Jan Schmidt2f232032013-04-25 16:04:51 +00003074
3075 btrfs_item_key_to_cpu(path->nodes[0], &found,
3076 btrfs_header_nritems(path->nodes[0]) - 1);
3077 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3078
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08003079 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3080 if (!scratch_leaf) {
3081 ret = -ENOMEM;
3082 mutex_unlock(&fs_info->qgroup_rescan_lock);
3083 goto out;
3084 }
Jan Schmidt2f232032013-04-25 16:04:51 +00003085 slot = path->slots[0];
3086 btrfs_release_path(path);
3087 mutex_unlock(&fs_info->qgroup_rescan_lock);
3088
3089 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3090 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
Josef Bacik3a6d75e2014-01-23 16:45:10 -05003091 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3092 found.type != BTRFS_METADATA_ITEM_KEY)
Jan Schmidt2f232032013-04-25 16:04:51 +00003093 continue;
Josef Bacik3a6d75e2014-01-23 16:45:10 -05003094 if (found.type == BTRFS_METADATA_ITEM_KEY)
Jeff Mahoneyda170662016-06-15 09:22:56 -04003095 num_bytes = fs_info->nodesize;
Josef Bacik3a6d75e2014-01-23 16:45:10 -05003096 else
3097 num_bytes = found.offset;
3098
Josef Bacikfcebe452014-05-13 17:30:47 -07003099 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
Zygo Blaxellc995ab32017-09-22 13:58:45 -04003100 &roots, false);
Jan Schmidt2f232032013-04-25 16:04:51 +00003101 if (ret < 0)
3102 goto out;
Qu Wenruo9d220c92015-04-13 11:02:16 +08003103 /* For rescan, just pass old_roots as NULL */
Lu Fengqi8696d762018-07-18 14:45:39 +08003104 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3105 num_bytes, NULL, roots);
Qu Wenruo9d220c92015-04-13 11:02:16 +08003106 if (ret < 0)
Jan Schmidt2f232032013-04-25 16:04:51 +00003107 goto out;
Jan Schmidt2f232032013-04-25 16:04:51 +00003108 }
Jan Schmidt2f232032013-04-25 16:04:51 +00003109out:
Nikolay Borisovdf449712018-08-15 18:26:56 +03003110 if (scratch_leaf)
Qu Wenruo0a0e8b892015-10-26 09:19:43 +08003111 free_extent_buffer(scratch_leaf);
Jan Schmidt2f232032013-04-25 16:04:51 +00003112
Qu Wenruo6f7de192018-06-27 18:19:55 +08003113 if (done && !ret) {
Qu Wenruoff3d27a02018-05-14 09:38:13 +08003114 ret = 1;
Qu Wenruo6f7de192018-06-27 18:19:55 +08003115 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3116 }
Jan Schmidt2f232032013-04-25 16:04:51 +00003117 return ret;
3118}
3119
Qu Wenruod458b052014-02-28 10:46:19 +08003120static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
Jan Schmidt2f232032013-04-25 16:04:51 +00003121{
Jan Schmidtb382a322013-05-28 15:47:24 +00003122 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3123 qgroup_rescan_work);
Jan Schmidt2f232032013-04-25 16:04:51 +00003124 struct btrfs_path *path;
3125 struct btrfs_trans_handle *trans = NULL;
Jan Schmidt2f232032013-04-25 16:04:51 +00003126 int err = -ENOMEM;
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003127 int ret = 0;
Jan Schmidt2f232032013-04-25 16:04:51 +00003128
3129 path = btrfs_alloc_path();
3130 if (!path)
3131 goto out;
Qu Wenruob6debf12018-05-14 09:38:12 +08003132 /*
3133 * Rescan should only search for commit root, and any later difference
3134 * should be recorded by qgroup
3135 */
3136 path->search_commit_root = 1;
3137 path->skip_locking = 1;
Jan Schmidt2f232032013-04-25 16:04:51 +00003138
3139 err = 0;
Justin Maggard7343dd62015-11-04 15:56:16 -08003140 while (!err && !btrfs_fs_closing(fs_info)) {
Jan Schmidt2f232032013-04-25 16:04:51 +00003141 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3142 if (IS_ERR(trans)) {
3143 err = PTR_ERR(trans);
3144 break;
3145 }
Josef Bacikafcdd122016-09-02 15:40:02 -04003146 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
Jan Schmidt2f232032013-04-25 16:04:51 +00003147 err = -EINTR;
3148 } else {
Lu Fengqi62088ca2018-07-18 14:45:42 +08003149 err = qgroup_rescan_leaf(trans, path);
Jan Schmidt2f232032013-04-25 16:04:51 +00003150 }
3151 if (err > 0)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003152 btrfs_commit_transaction(trans);
Jan Schmidt2f232032013-04-25 16:04:51 +00003153 else
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003154 btrfs_end_transaction(trans);
Jan Schmidt2f232032013-04-25 16:04:51 +00003155 }
3156
3157out:
Jan Schmidt2f232032013-04-25 16:04:51 +00003158 btrfs_free_path(path);
Jan Schmidt2f232032013-04-25 16:04:51 +00003159
3160 mutex_lock(&fs_info->qgroup_rescan_lock);
Qu Wenruo33931682015-02-27 16:24:24 +08003161 if (err > 0 &&
Jan Schmidt2f232032013-04-25 16:04:51 +00003162 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3163 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3164 } else if (err < 0) {
3165 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3166 }
3167 mutex_unlock(&fs_info->qgroup_rescan_lock);
3168
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003169 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04003170 * only update status, since the previous part has already updated the
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003171 * qgroup info.
3172 */
3173 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3174 if (IS_ERR(trans)) {
3175 err = PTR_ERR(trans);
Filipe Manana13fc1d22019-09-24 10:49:54 +01003176 trans = NULL;
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003177 btrfs_err(fs_info,
David Sterba913e1532017-07-13 15:32:18 +02003178 "fail to start transaction for status update: %d",
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003179 err);
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003180 }
Filipe Manana13fc1d22019-09-24 10:49:54 +01003181
3182 mutex_lock(&fs_info->qgroup_rescan_lock);
3183 if (!btrfs_fs_closing(fs_info))
3184 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3185 if (trans) {
3186 ret = update_qgroup_status_item(trans);
3187 if (ret < 0) {
3188 err = ret;
3189 btrfs_err(fs_info, "fail to update qgroup status: %d",
3190 err);
3191 }
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003192 }
Filipe Manana13fc1d22019-09-24 10:49:54 +01003193 fs_info->qgroup_rescan_running = false;
3194 complete_all(&fs_info->qgroup_rescan_completion);
3195 mutex_unlock(&fs_info->qgroup_rescan_lock);
3196
3197 if (!trans)
3198 return;
3199
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003200 btrfs_end_transaction(trans);
Qu Wenruo53b7cde2015-02-27 16:24:25 +08003201
Justin Maggard7343dd62015-11-04 15:56:16 -08003202 if (btrfs_fs_closing(fs_info)) {
3203 btrfs_info(fs_info, "qgroup scan paused");
3204 } else if (err >= 0) {
Frank Holtonefe120a2013-12-20 11:37:06 -05003205 btrfs_info(fs_info, "qgroup scan completed%s",
Qu Wenruo33931682015-02-27 16:24:24 +08003206 err > 0 ? " (inconsistency flag cleared)" : "");
Jan Schmidt2f232032013-04-25 16:04:51 +00003207 } else {
Frank Holtonefe120a2013-12-20 11:37:06 -05003208 btrfs_err(fs_info, "qgroup scan failed with %d", err);
Jan Schmidt2f232032013-04-25 16:04:51 +00003209 }
3210}
3211
Jan Schmidtb382a322013-05-28 15:47:24 +00003212/*
3213 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3214 * memory required for the rescan context.
3215 */
3216static int
3217qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3218 int init_flags)
Jan Schmidt2f232032013-04-25 16:04:51 +00003219{
3220 int ret = 0;
Jan Schmidt2f232032013-04-25 16:04:51 +00003221
Qu Wenruo9593bf492018-05-02 13:28:03 +08003222 if (!init_flags) {
3223 /* we're resuming qgroup rescan at mount time */
Filipe Mananae4e7ede2018-06-27 00:43:15 +01003224 if (!(fs_info->qgroup_flags &
3225 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
Qu Wenruo9593bf492018-05-02 13:28:03 +08003226 btrfs_warn(fs_info,
Nikolay Borisov37d02592019-11-18 14:16:44 +02003227 "qgroup rescan init failed, qgroup rescan is not queued");
Filipe Mananae4e7ede2018-06-27 00:43:15 +01003228 ret = -EINVAL;
3229 } else if (!(fs_info->qgroup_flags &
3230 BTRFS_QGROUP_STATUS_FLAG_ON)) {
Qu Wenruo9593bf492018-05-02 13:28:03 +08003231 btrfs_warn(fs_info,
Nikolay Borisov37d02592019-11-18 14:16:44 +02003232 "qgroup rescan init failed, qgroup is not enabled");
Filipe Mananae4e7ede2018-06-27 00:43:15 +01003233 ret = -EINVAL;
3234 }
3235
3236 if (ret)
3237 return ret;
Jan Schmidtb382a322013-05-28 15:47:24 +00003238 }
Jan Schmidt2f232032013-04-25 16:04:51 +00003239
3240 mutex_lock(&fs_info->qgroup_rescan_lock);
3241 spin_lock(&fs_info->qgroup_lock);
Jan Schmidtb382a322013-05-28 15:47:24 +00003242
3243 if (init_flags) {
Qu Wenruo9593bf492018-05-02 13:28:03 +08003244 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3245 btrfs_warn(fs_info,
3246 "qgroup rescan is already in progress");
Jan Schmidtb382a322013-05-28 15:47:24 +00003247 ret = -EINPROGRESS;
Qu Wenruo9593bf492018-05-02 13:28:03 +08003248 } else if (!(fs_info->qgroup_flags &
3249 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3250 btrfs_warn(fs_info,
3251 "qgroup rescan init failed, qgroup is not enabled");
Jan Schmidtb382a322013-05-28 15:47:24 +00003252 ret = -EINVAL;
Qu Wenruo9593bf492018-05-02 13:28:03 +08003253 }
Jan Schmidtb382a322013-05-28 15:47:24 +00003254
3255 if (ret) {
3256 spin_unlock(&fs_info->qgroup_lock);
3257 mutex_unlock(&fs_info->qgroup_rescan_lock);
Qu Wenruo9593bf492018-05-02 13:28:03 +08003258 return ret;
Jan Schmidtb382a322013-05-28 15:47:24 +00003259 }
Jan Schmidtb382a322013-05-28 15:47:24 +00003260 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3261 }
3262
3263 memset(&fs_info->qgroup_rescan_progress, 0,
3264 sizeof(fs_info->qgroup_rescan_progress));
3265 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
Filipe Manana190631f2015-11-05 10:06:23 +00003266 init_completion(&fs_info->qgroup_rescan_completion);
Jan Schmidtb382a322013-05-28 15:47:24 +00003267
3268 spin_unlock(&fs_info->qgroup_lock);
3269 mutex_unlock(&fs_info->qgroup_rescan_lock);
3270
Qu Wenruofc97fab2014-02-28 10:46:16 +08003271 btrfs_init_work(&fs_info->qgroup_rescan_work,
3272 btrfs_qgroup_rescan_worker, NULL, NULL);
Jan Schmidtb382a322013-05-28 15:47:24 +00003273 return 0;
3274}
Jan Schmidt2f232032013-04-25 16:04:51 +00003275
Jan Schmidtb382a322013-05-28 15:47:24 +00003276static void
3277qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3278{
3279 struct rb_node *n;
3280 struct btrfs_qgroup *qgroup;
3281
3282 spin_lock(&fs_info->qgroup_lock);
Jan Schmidt2f232032013-04-25 16:04:51 +00003283 /* clear all current qgroup tracking information */
3284 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3285 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3286 qgroup->rfer = 0;
3287 qgroup->rfer_cmpr = 0;
3288 qgroup->excl = 0;
3289 qgroup->excl_cmpr = 0;
Qu Wenruo9c7b0c22018-08-10 10:20:26 +08003290 qgroup_dirty(fs_info, qgroup);
Jan Schmidt2f232032013-04-25 16:04:51 +00003291 }
3292 spin_unlock(&fs_info->qgroup_lock);
Jan Schmidtb382a322013-05-28 15:47:24 +00003293}
Jan Schmidt2f232032013-04-25 16:04:51 +00003294
Jan Schmidtb382a322013-05-28 15:47:24 +00003295int
3296btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3297{
3298 int ret = 0;
3299 struct btrfs_trans_handle *trans;
3300
3301 ret = qgroup_rescan_init(fs_info, 0, 1);
3302 if (ret)
3303 return ret;
3304
3305 /*
3306 * We have set the rescan_progress to 0, which means no more
3307 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3308 * However, btrfs_qgroup_account_ref may be right after its call
3309 * to btrfs_find_all_roots, in which case it would still do the
3310 * accounting.
3311 * To solve this, we're committing the transaction, which will
3312 * ensure we run all delayed refs and only after that, we are
3313 * going to clear all tracking information for a clean start.
3314 */
3315
3316 trans = btrfs_join_transaction(fs_info->fs_root);
3317 if (IS_ERR(trans)) {
3318 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3319 return PTR_ERR(trans);
3320 }
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003321 ret = btrfs_commit_transaction(trans);
Jan Schmidtb382a322013-05-28 15:47:24 +00003322 if (ret) {
3323 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3324 return ret;
3325 }
3326
3327 qgroup_rescan_zero_tracking(fs_info);
3328
Qu Wenruod61acbb2020-02-07 13:38:20 +08003329 mutex_lock(&fs_info->qgroup_rescan_lock);
3330 fs_info->qgroup_rescan_running = true;
Qu Wenruofc97fab2014-02-28 10:46:16 +08003331 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3332 &fs_info->qgroup_rescan_work);
Qu Wenruod61acbb2020-02-07 13:38:20 +08003333 mutex_unlock(&fs_info->qgroup_rescan_lock);
Jan Schmidt2f232032013-04-25 16:04:51 +00003334
3335 return 0;
3336}
Jan Schmidt57254b6e2013-05-06 19:14:17 +00003337
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04003338int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3339 bool interruptible)
Jan Schmidt57254b6e2013-05-06 19:14:17 +00003340{
3341 int running;
3342 int ret = 0;
3343
3344 mutex_lock(&fs_info->qgroup_rescan_lock);
3345 spin_lock(&fs_info->qgroup_lock);
Jeff Mahoneyd2c609b2016-08-15 12:10:33 -04003346 running = fs_info->qgroup_rescan_running;
Jan Schmidt57254b6e2013-05-06 19:14:17 +00003347 spin_unlock(&fs_info->qgroup_lock);
3348 mutex_unlock(&fs_info->qgroup_rescan_lock);
3349
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04003350 if (!running)
3351 return 0;
3352
3353 if (interruptible)
Jan Schmidt57254b6e2013-05-06 19:14:17 +00003354 ret = wait_for_completion_interruptible(
3355 &fs_info->qgroup_rescan_completion);
Jeff Mahoneyd06f23d2016-08-08 22:08:06 -04003356 else
3357 wait_for_completion(&fs_info->qgroup_rescan_completion);
Jan Schmidt57254b6e2013-05-06 19:14:17 +00003358
3359 return ret;
3360}
Jan Schmidtb382a322013-05-28 15:47:24 +00003361
3362/*
3363 * this is only called from open_ctree where we're still single threaded, thus
3364 * locking is omitted here.
3365 */
3366void
3367btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3368{
Qu Wenruod61acbb2020-02-07 13:38:20 +08003369 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3370 mutex_lock(&fs_info->qgroup_rescan_lock);
3371 fs_info->qgroup_rescan_running = true;
Qu Wenruofc97fab2014-02-28 10:46:16 +08003372 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3373 &fs_info->qgroup_rescan_work);
Qu Wenruod61acbb2020-02-07 13:38:20 +08003374 mutex_unlock(&fs_info->qgroup_rescan_lock);
3375 }
Jan Schmidtb382a322013-05-28 15:47:24 +00003376}
Qu Wenruo52472552015-10-12 16:05:40 +08003377
3378/*
3379 * Reserve qgroup space for range [start, start + len).
3380 *
3381 * This function will either reserve space from related qgroups or doing
3382 * nothing if the range is already reserved.
3383 *
3384 * Return 0 for successful reserve
3385 * Return <0 for error (including -EQUOT)
3386 *
3387 * NOTE: this function may sleep for memory allocation.
Qu Wenruo364ecf32017-02-27 15:10:38 +08003388 * if btrfs_qgroup_reserve_data() is called multiple times with
3389 * same @reserved, caller must ensure when error happens it's OK
3390 * to free *ALL* reserved space.
Qu Wenruo52472552015-10-12 16:05:40 +08003391 */
Qu Wenruo364ecf32017-02-27 15:10:38 +08003392int btrfs_qgroup_reserve_data(struct inode *inode,
3393 struct extent_changeset **reserved_ret, u64 start,
3394 u64 len)
Qu Wenruo52472552015-10-12 16:05:40 +08003395{
3396 struct btrfs_root *root = BTRFS_I(inode)->root;
Qu Wenruo52472552015-10-12 16:05:40 +08003397 struct ulist_node *unode;
3398 struct ulist_iterator uiter;
Qu Wenruo364ecf32017-02-27 15:10:38 +08003399 struct extent_changeset *reserved;
3400 u64 orig_reserved;
3401 u64 to_reserve;
Qu Wenruo52472552015-10-12 16:05:40 +08003402 int ret;
3403
Josef Bacikafcdd122016-09-02 15:40:02 -04003404 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003405 !is_fstree(root->root_key.objectid) || len == 0)
Qu Wenruo52472552015-10-12 16:05:40 +08003406 return 0;
3407
Qu Wenruo364ecf32017-02-27 15:10:38 +08003408 /* @reserved parameter is mandatory for qgroup */
3409 if (WARN_ON(!reserved_ret))
3410 return -EINVAL;
3411 if (!*reserved_ret) {
3412 *reserved_ret = extent_changeset_alloc();
3413 if (!*reserved_ret)
3414 return -ENOMEM;
3415 }
3416 reserved = *reserved_ret;
3417 /* Record already reserved space */
3418 orig_reserved = reserved->bytes_changed;
Qu Wenruo52472552015-10-12 16:05:40 +08003419 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
Qu Wenruo364ecf32017-02-27 15:10:38 +08003420 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3421
3422 /* Newly reserved space */
3423 to_reserve = reserved->bytes_changed - orig_reserved;
Qu Wenruo81fb6f72015-09-28 16:57:53 +08003424 trace_btrfs_qgroup_reserve_data(inode, start, len,
Qu Wenruo364ecf32017-02-27 15:10:38 +08003425 to_reserve, QGROUP_RESERVE);
Qu Wenruo52472552015-10-12 16:05:40 +08003426 if (ret < 0)
3427 goto cleanup;
Qu Wenruodba21322017-12-12 15:34:25 +08003428 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
Qu Wenruo52472552015-10-12 16:05:40 +08003429 if (ret < 0)
3430 goto cleanup;
3431
Qu Wenruo52472552015-10-12 16:05:40 +08003432 return ret;
3433
3434cleanup:
Qu Wenruo364ecf32017-02-27 15:10:38 +08003435 /* cleanup *ALL* already reserved ranges */
Qu Wenruo52472552015-10-12 16:05:40 +08003436 ULIST_ITER_INIT(&uiter);
Qu Wenruo364ecf32017-02-27 15:10:38 +08003437 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
Qu Wenruo52472552015-10-12 16:05:40 +08003438 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
David Sterbaae0f1622017-10-31 16:37:52 +01003439 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
Qu Wenruod4e20492019-09-16 20:02:39 +08003440 /* Also free data bytes of already reserved one */
3441 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3442 orig_reserved, BTRFS_QGROUP_RSV_DATA);
Qu Wenruo364ecf32017-02-27 15:10:38 +08003443 extent_changeset_release(reserved);
Qu Wenruo52472552015-10-12 16:05:40 +08003444 return ret;
3445}
Qu Wenruof695fdc2015-10-12 16:28:06 +08003446
Qu Wenruobc42bda2017-02-27 15:10:39 +08003447/* Free ranges specified by @reserved, normally in error path */
3448static int qgroup_free_reserved_data(struct inode *inode,
3449 struct extent_changeset *reserved, u64 start, u64 len)
3450{
3451 struct btrfs_root *root = BTRFS_I(inode)->root;
3452 struct ulist_node *unode;
3453 struct ulist_iterator uiter;
3454 struct extent_changeset changeset;
3455 int freed = 0;
3456 int ret;
3457
3458 extent_changeset_init(&changeset);
3459 len = round_up(start + len, root->fs_info->sectorsize);
3460 start = round_down(start, root->fs_info->sectorsize);
3461
3462 ULIST_ITER_INIT(&uiter);
3463 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3464 u64 range_start = unode->val;
3465 /* unode->aux is the inclusive end */
3466 u64 range_len = unode->aux - range_start + 1;
3467 u64 free_start;
3468 u64 free_len;
3469
3470 extent_changeset_release(&changeset);
3471
3472 /* Only free range in range [start, start + len) */
3473 if (range_start >= start + len ||
3474 range_start + range_len <= start)
3475 continue;
3476 free_start = max(range_start, start);
3477 free_len = min(start + len, range_start + range_len) -
3478 free_start;
3479 /*
3480 * TODO: To also modify reserved->ranges_reserved to reflect
3481 * the modification.
3482 *
3483 * However as long as we free qgroup reserved according to
3484 * EXTENT_QGROUP_RESERVED, we won't double free.
3485 * So not need to rush.
3486 */
Qu Wenruobab32fc2019-09-16 20:02:38 +08003487 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree,
Qu Wenruobc42bda2017-02-27 15:10:39 +08003488 free_start, free_start + free_len - 1,
3489 EXTENT_QGROUP_RESERVED, &changeset);
3490 if (ret < 0)
3491 goto out;
3492 freed += changeset.bytes_changed;
3493 }
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003494 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
Qu Wenruod4e5c922017-12-12 15:34:23 +08003495 BTRFS_QGROUP_RSV_DATA);
Qu Wenruobc42bda2017-02-27 15:10:39 +08003496 ret = freed;
3497out:
3498 extent_changeset_release(&changeset);
3499 return ret;
3500}
3501
3502static int __btrfs_qgroup_release_data(struct inode *inode,
3503 struct extent_changeset *reserved, u64 start, u64 len,
3504 int free)
Qu Wenruof695fdc2015-10-12 16:28:06 +08003505{
3506 struct extent_changeset changeset;
Qu Wenruo81fb6f72015-09-28 16:57:53 +08003507 int trace_op = QGROUP_RELEASE;
Qu Wenruof695fdc2015-10-12 16:28:06 +08003508 int ret;
3509
Qu Wenruo3628b4c2018-10-09 14:36:45 +08003510 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3511 &BTRFS_I(inode)->root->fs_info->flags))
3512 return 0;
3513
Qu Wenruobc42bda2017-02-27 15:10:39 +08003514 /* In release case, we shouldn't have @reserved */
3515 WARN_ON(!free && reserved);
3516 if (free && reserved)
3517 return qgroup_free_reserved_data(inode, reserved, start, len);
Qu Wenruo364ecf32017-02-27 15:10:38 +08003518 extent_changeset_init(&changeset);
Qu Wenruof695fdc2015-10-12 16:28:06 +08003519 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
David Sterbaf734c442016-04-26 23:54:39 +02003520 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
Qu Wenruof695fdc2015-10-12 16:28:06 +08003521 if (ret < 0)
3522 goto out;
3523
Qu Wenruod51ea5d2017-03-13 15:52:09 +08003524 if (free)
3525 trace_op = QGROUP_FREE;
3526 trace_btrfs_qgroup_release_data(inode, start, len,
3527 changeset.bytes_changed, trace_op);
3528 if (free)
David Sterba0b08e1f2017-02-13 14:24:35 +01003529 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003530 BTRFS_I(inode)->root->root_key.objectid,
Qu Wenruod4e5c922017-12-12 15:34:23 +08003531 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
Qu Wenruo7bc329c2017-02-27 15:10:36 +08003532 ret = changeset.bytes_changed;
Qu Wenruof695fdc2015-10-12 16:28:06 +08003533out:
Qu Wenruo364ecf32017-02-27 15:10:38 +08003534 extent_changeset_release(&changeset);
Qu Wenruof695fdc2015-10-12 16:28:06 +08003535 return ret;
3536}
3537
3538/*
3539 * Free a reserved space range from io_tree and related qgroups
3540 *
3541 * Should be called when a range of pages get invalidated before reaching disk.
3542 * Or for error cleanup case.
Qu Wenruobc42bda2017-02-27 15:10:39 +08003543 * if @reserved is given, only reserved range in [@start, @start + @len) will
3544 * be freed.
Qu Wenruof695fdc2015-10-12 16:28:06 +08003545 *
3546 * For data written to disk, use btrfs_qgroup_release_data().
3547 *
3548 * NOTE: This function may sleep for memory allocation.
3549 */
Qu Wenruobc42bda2017-02-27 15:10:39 +08003550int btrfs_qgroup_free_data(struct inode *inode,
3551 struct extent_changeset *reserved, u64 start, u64 len)
Qu Wenruof695fdc2015-10-12 16:28:06 +08003552{
Qu Wenruobc42bda2017-02-27 15:10:39 +08003553 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
Qu Wenruof695fdc2015-10-12 16:28:06 +08003554}
3555
3556/*
3557 * Release a reserved space range from io_tree only.
3558 *
3559 * Should be called when a range of pages get written to disk and corresponding
3560 * FILE_EXTENT is inserted into corresponding root.
3561 *
3562 * Since new qgroup accounting framework will only update qgroup numbers at
3563 * commit_transaction() time, its reserved space shouldn't be freed from
3564 * related qgroups.
3565 *
3566 * But we should release the range from io_tree, to allow further write to be
3567 * COWed.
3568 *
3569 * NOTE: This function may sleep for memory allocation.
3570 */
3571int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3572{
Qu Wenruobc42bda2017-02-27 15:10:39 +08003573 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
Qu Wenruof695fdc2015-10-12 16:28:06 +08003574}
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003575
Qu Wenruo82874752017-12-12 15:34:34 +08003576static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3577 enum btrfs_qgroup_rsv_type type)
3578{
3579 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3580 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3581 return;
3582 if (num_bytes == 0)
3583 return;
3584
3585 spin_lock(&root->qgroup_meta_rsv_lock);
3586 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3587 root->qgroup_meta_rsv_prealloc += num_bytes;
3588 else
3589 root->qgroup_meta_rsv_pertrans += num_bytes;
3590 spin_unlock(&root->qgroup_meta_rsv_lock);
3591}
3592
3593static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3594 enum btrfs_qgroup_rsv_type type)
3595{
3596 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3597 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3598 return 0;
3599 if (num_bytes == 0)
3600 return 0;
3601
3602 spin_lock(&root->qgroup_meta_rsv_lock);
3603 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3604 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3605 num_bytes);
3606 root->qgroup_meta_rsv_prealloc -= num_bytes;
3607 } else {
3608 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3609 num_bytes);
3610 root->qgroup_meta_rsv_pertrans -= num_bytes;
3611 }
3612 spin_unlock(&root->qgroup_meta_rsv_lock);
3613 return num_bytes;
3614}
3615
Qu Wenruo733e03a2017-12-12 15:34:29 +08003616int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3617 enum btrfs_qgroup_rsv_type type, bool enforce)
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003618{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003619 struct btrfs_fs_info *fs_info = root->fs_info;
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003620 int ret;
3621
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003622 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003623 !is_fstree(root->root_key.objectid) || num_bytes == 0)
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003624 return 0;
3625
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003626 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
Qu Wenruofd2b0072019-10-17 10:38:36 +08003627 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
Qu Wenruo733e03a2017-12-12 15:34:29 +08003628 ret = qgroup_reserve(root, num_bytes, enforce, type);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003629 if (ret < 0)
3630 return ret;
Qu Wenruo82874752017-12-12 15:34:34 +08003631 /*
3632 * Record what we have reserved into root.
3633 *
3634 * To avoid quota disabled->enabled underflow.
3635 * In that case, we may try to free space we haven't reserved
3636 * (since quota was disabled), so record what we reserved into root.
3637 * And ensure later release won't underflow this number.
3638 */
3639 add_root_meta_rsv(root, num_bytes, type);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003640 return ret;
3641}
3642
Qu Wenruo733e03a2017-12-12 15:34:29 +08003643void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003644{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003645 struct btrfs_fs_info *fs_info = root->fs_info;
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003646
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003647 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003648 !is_fstree(root->root_key.objectid))
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003649 return;
3650
Qu Wenruoe1211d02017-12-12 15:34:30 +08003651 /* TODO: Update trace point to handle such free */
Qu Wenruo4ee0d882017-12-12 15:34:35 +08003652 trace_qgroup_meta_free_all_pertrans(root);
Qu Wenruoe1211d02017-12-12 15:34:30 +08003653 /* Special value -1 means to free all reserved space */
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003654 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
Qu Wenruo733e03a2017-12-12 15:34:29 +08003655 BTRFS_QGROUP_RSV_META_PERTRANS);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003656}
3657
Qu Wenruo733e03a2017-12-12 15:34:29 +08003658void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3659 enum btrfs_qgroup_rsv_type type)
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003660{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003661 struct btrfs_fs_info *fs_info = root->fs_info;
3662
3663 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003664 !is_fstree(root->root_key.objectid))
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003665 return;
3666
Qu Wenruo82874752017-12-12 15:34:34 +08003667 /*
3668 * reservation for META_PREALLOC can happen before quota is enabled,
3669 * which can lead to underflow.
3670 * Here ensure we will only free what we really have reserved.
3671 */
3672 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003673 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
Qu Wenruofd2b0072019-10-17 10:38:36 +08003674 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003675 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3676 num_bytes, type);
Qu Wenruo55eeaf02015-09-08 17:08:38 +08003677}
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003678
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003679static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3680 int num_bytes)
3681{
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003682 struct btrfs_qgroup *qgroup;
3683 struct ulist_node *unode;
3684 struct ulist_iterator uiter;
3685 int ret = 0;
3686
3687 if (num_bytes == 0)
3688 return;
Marcos Paulo de Souzae3b0edd2019-11-25 21:58:50 -03003689 if (!fs_info->quota_root)
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003690 return;
3691
3692 spin_lock(&fs_info->qgroup_lock);
3693 qgroup = find_qgroup_rb(fs_info, ref_root);
3694 if (!qgroup)
3695 goto out;
3696 ulist_reinit(fs_info->qgroup_ulist);
3697 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
David Sterbaa1840b52018-03-27 19:04:50 +02003698 qgroup_to_aux(qgroup), GFP_ATOMIC);
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003699 if (ret < 0)
3700 goto out;
3701 ULIST_ITER_INIT(&uiter);
3702 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3703 struct btrfs_qgroup *qg;
3704 struct btrfs_qgroup_list *glist;
3705
3706 qg = unode_aux_to_qgroup(unode);
3707
3708 qgroup_rsv_release(fs_info, qg, num_bytes,
3709 BTRFS_QGROUP_RSV_META_PREALLOC);
3710 qgroup_rsv_add(fs_info, qg, num_bytes,
3711 BTRFS_QGROUP_RSV_META_PERTRANS);
3712 list_for_each_entry(glist, &qg->groups, next_group) {
3713 ret = ulist_add(fs_info->qgroup_ulist,
3714 glist->group->qgroupid,
David Sterbaa1840b52018-03-27 19:04:50 +02003715 qgroup_to_aux(glist->group), GFP_ATOMIC);
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003716 if (ret < 0)
3717 goto out;
3718 }
3719 }
3720out:
3721 spin_unlock(&fs_info->qgroup_lock);
3722}
3723
3724void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3725{
3726 struct btrfs_fs_info *fs_info = root->fs_info;
3727
3728 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003729 !is_fstree(root->root_key.objectid))
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003730 return;
Qu Wenruo82874752017-12-12 15:34:34 +08003731 /* Same as btrfs_qgroup_free_meta_prealloc() */
3732 num_bytes = sub_root_meta_rsv(root, num_bytes,
3733 BTRFS_QGROUP_RSV_META_PREALLOC);
Qu Wenruo4ee0d882017-12-12 15:34:35 +08003734 trace_qgroup_meta_convert(root, num_bytes);
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003735 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
Qu Wenruo64cfaef2017-12-12 15:34:31 +08003736}
3737
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003738/*
Nicholas D Steeves01327612016-05-19 21:18:45 -04003739 * Check qgroup reserved space leaking, normally at destroy inode
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003740 * time
3741 */
3742void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3743{
3744 struct extent_changeset changeset;
3745 struct ulist_node *unode;
3746 struct ulist_iterator iter;
3747 int ret;
3748
Qu Wenruo364ecf32017-02-27 15:10:38 +08003749 extent_changeset_init(&changeset);
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003750 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
David Sterbaf734c442016-04-26 23:54:39 +02003751 EXTENT_QGROUP_RESERVED, &changeset);
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003752
3753 WARN_ON(ret < 0);
3754 if (WARN_ON(changeset.bytes_changed)) {
3755 ULIST_ITER_INIT(&iter);
David Sterba53d32352017-02-13 13:42:29 +01003756 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003757 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3758 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3759 inode->i_ino, unode->val, unode->aux);
3760 }
David Sterba0b08e1f2017-02-13 14:24:35 +01003761 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09003762 BTRFS_I(inode)->root->root_key.objectid,
Qu Wenruod4e5c922017-12-12 15:34:23 +08003763 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
David Sterba0b08e1f2017-02-13 14:24:35 +01003764
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003765 }
Qu Wenruo364ecf32017-02-27 15:10:38 +08003766 extent_changeset_release(&changeset);
Qu Wenruo56fa9d02015-10-13 09:53:10 +08003767}
Qu Wenruo370a11b2019-01-23 15:15:16 +08003768
3769void btrfs_qgroup_init_swapped_blocks(
3770 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3771{
3772 int i;
3773
3774 spin_lock_init(&swapped_blocks->lock);
3775 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3776 swapped_blocks->blocks[i] = RB_ROOT;
3777 swapped_blocks->swapped = false;
3778}
3779
3780/*
3781 * Delete all swapped blocks record of @root.
3782 * Every record here means we skipped a full subtree scan for qgroup.
3783 *
3784 * Gets called when committing one transaction.
3785 */
3786void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3787{
3788 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3789 int i;
3790
3791 swapped_blocks = &root->swapped_blocks;
3792
3793 spin_lock(&swapped_blocks->lock);
3794 if (!swapped_blocks->swapped)
3795 goto out;
3796 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3797 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3798 struct btrfs_qgroup_swapped_block *entry;
3799 struct btrfs_qgroup_swapped_block *next;
3800
3801 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3802 node)
3803 kfree(entry);
3804 swapped_blocks->blocks[i] = RB_ROOT;
3805 }
3806 swapped_blocks->swapped = false;
3807out:
3808 spin_unlock(&swapped_blocks->lock);
3809}
3810
3811/*
3812 * Add subtree roots record into @subvol_root.
3813 *
3814 * @subvol_root: tree root of the subvolume tree get swapped
3815 * @bg: block group under balance
3816 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3817 * @reloc_parent/slot: pointer to the subtree root in reloc tree
3818 * BOTH POINTERS ARE BEFORE TREE SWAP
3819 * @last_snapshot: last snapshot generation of the subvolume tree
3820 */
3821int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3822 struct btrfs_root *subvol_root,
David Sterba32da53862019-10-29 19:20:18 +01003823 struct btrfs_block_group *bg,
Qu Wenruo370a11b2019-01-23 15:15:16 +08003824 struct extent_buffer *subvol_parent, int subvol_slot,
3825 struct extent_buffer *reloc_parent, int reloc_slot,
3826 u64 last_snapshot)
3827{
3828 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3829 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3830 struct btrfs_qgroup_swapped_block *block;
3831 struct rb_node **cur;
3832 struct rb_node *parent = NULL;
3833 int level = btrfs_header_level(subvol_parent) - 1;
3834 int ret = 0;
3835
3836 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3837 return 0;
3838
3839 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3840 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3841 btrfs_err_rl(fs_info,
3842 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3843 __func__,
3844 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3845 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3846 return -EUCLEAN;
3847 }
3848
3849 block = kmalloc(sizeof(*block), GFP_NOFS);
3850 if (!block) {
3851 ret = -ENOMEM;
3852 goto out;
3853 }
3854
3855 /*
3856 * @reloc_parent/slot is still before swap, while @block is going to
3857 * record the bytenr after swap, so we do the swap here.
3858 */
3859 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3860 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3861 reloc_slot);
3862 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3863 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3864 subvol_slot);
3865 block->last_snapshot = last_snapshot;
3866 block->level = level;
Qu Wenruo57949d02019-05-21 19:28:08 +08003867
3868 /*
3869 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3870 * no one else can modify tree blocks thus we qgroup will not change
3871 * no matter the value of trace_leaf.
3872 */
3873 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
Qu Wenruo370a11b2019-01-23 15:15:16 +08003874 block->trace_leaf = true;
3875 else
3876 block->trace_leaf = false;
3877 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3878
3879 /* Insert @block into @blocks */
3880 spin_lock(&blocks->lock);
3881 cur = &blocks->blocks[level].rb_node;
3882 while (*cur) {
3883 struct btrfs_qgroup_swapped_block *entry;
3884
3885 parent = *cur;
3886 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3887 node);
3888
3889 if (entry->subvol_bytenr < block->subvol_bytenr) {
3890 cur = &(*cur)->rb_left;
3891 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3892 cur = &(*cur)->rb_right;
3893 } else {
3894 if (entry->subvol_generation !=
3895 block->subvol_generation ||
3896 entry->reloc_bytenr != block->reloc_bytenr ||
3897 entry->reloc_generation !=
3898 block->reloc_generation) {
3899 /*
3900 * Duplicated but mismatch entry found.
3901 * Shouldn't happen.
3902 *
3903 * Marking qgroup inconsistent should be enough
3904 * for end users.
3905 */
3906 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3907 ret = -EEXIST;
3908 }
3909 kfree(block);
3910 goto out_unlock;
3911 }
3912 }
3913 rb_link_node(&block->node, parent, cur);
3914 rb_insert_color(&block->node, &blocks->blocks[level]);
3915 blocks->swapped = true;
3916out_unlock:
3917 spin_unlock(&blocks->lock);
3918out:
3919 if (ret < 0)
3920 fs_info->qgroup_flags |=
3921 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3922 return ret;
3923}
Qu Wenruof616f5c2019-01-23 15:15:17 +08003924
3925/*
3926 * Check if the tree block is a subtree root, and if so do the needed
3927 * delayed subtree trace for qgroup.
3928 *
3929 * This is called during btrfs_cow_block().
3930 */
3931int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3932 struct btrfs_root *root,
3933 struct extent_buffer *subvol_eb)
3934{
3935 struct btrfs_fs_info *fs_info = root->fs_info;
3936 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3937 struct btrfs_qgroup_swapped_block *block;
3938 struct extent_buffer *reloc_eb = NULL;
3939 struct rb_node *node;
3940 bool found = false;
3941 bool swapped = false;
3942 int level = btrfs_header_level(subvol_eb);
3943 int ret = 0;
3944 int i;
3945
3946 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3947 return 0;
3948 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3949 return 0;
3950
3951 spin_lock(&blocks->lock);
3952 if (!blocks->swapped) {
3953 spin_unlock(&blocks->lock);
3954 return 0;
3955 }
3956 node = blocks->blocks[level].rb_node;
3957
3958 while (node) {
3959 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3960 if (block->subvol_bytenr < subvol_eb->start) {
3961 node = node->rb_left;
3962 } else if (block->subvol_bytenr > subvol_eb->start) {
3963 node = node->rb_right;
3964 } else {
3965 found = true;
3966 break;
3967 }
3968 }
3969 if (!found) {
3970 spin_unlock(&blocks->lock);
3971 goto out;
3972 }
3973 /* Found one, remove it from @blocks first and update blocks->swapped */
3974 rb_erase(&block->node, &blocks->blocks[level]);
3975 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3976 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3977 swapped = true;
3978 break;
3979 }
3980 }
3981 blocks->swapped = swapped;
3982 spin_unlock(&blocks->lock);
3983
3984 /* Read out reloc subtree root */
3985 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3986 block->reloc_generation, block->level,
3987 &block->first_key);
3988 if (IS_ERR(reloc_eb)) {
3989 ret = PTR_ERR(reloc_eb);
3990 reloc_eb = NULL;
3991 goto free_out;
3992 }
3993 if (!extent_buffer_uptodate(reloc_eb)) {
3994 ret = -EIO;
3995 goto free_out;
3996 }
3997
3998 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3999 block->last_snapshot, block->trace_leaf);
4000free_out:
4001 kfree(block);
4002 free_extent_buffer(reloc_eb);
4003out:
4004 if (ret < 0) {
4005 btrfs_err_rl(fs_info,
4006 "failed to account subtree at bytenr %llu: %d",
4007 subvol_eb->start, ret);
4008 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4009 }
4010 return ret;
4011}
Jeff Mahoney81f7eb02020-02-11 15:25:37 +08004012
4013void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4014{
4015 struct btrfs_qgroup_extent_record *entry;
4016 struct btrfs_qgroup_extent_record *next;
4017 struct rb_root *root;
4018
4019 root = &trans->delayed_refs.dirty_extent_root;
4020 rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4021 ulist_free(entry->old_roots);
4022 kfree(entry);
4023 }
4024}