1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2011 STRATO. All rights reserved.
4 */
5
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sched/mm.h>
15
16 #include "ctree.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "locking.h"
20 #include "ulist.h"
21 #include "backref.h"
22 #include "extent_io.h"
23 #include "qgroup.h"
24 #include "block-group.h"
25 #include "sysfs.h"
26 #include "tree-mod-log.h"
27 #include "fs.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
31 #include "tree-checker.h"
32
33 /*
34 * Helpers to access qgroup reservation
35 *
36 * Callers should ensure the lock context and type are valid
37 */
38
qgroup_rsv_total(const struct btrfs_qgroup * qgroup)39 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
40 {
41 u64 ret = 0;
42 int i;
43
44 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
45 ret += qgroup->rsv.values[i];
46
47 return ret;
48 }
49
50 #ifdef CONFIG_BTRFS_DEBUG
qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)51 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
52 {
53 if (type == BTRFS_QGROUP_RSV_DATA)
54 return "data";
55 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
56 return "meta_pertrans";
57 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
58 return "meta_prealloc";
59 return NULL;
60 }
61 #endif
62
qgroup_rsv_add(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup,u64 num_bytes,enum btrfs_qgroup_rsv_type type)63 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
64 struct btrfs_qgroup *qgroup, u64 num_bytes,
65 enum btrfs_qgroup_rsv_type type)
66 {
67 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
68 qgroup->rsv.values[type] += num_bytes;
69 }
70
qgroup_rsv_release(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup,u64 num_bytes,enum btrfs_qgroup_rsv_type type)71 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
72 struct btrfs_qgroup *qgroup, u64 num_bytes,
73 enum btrfs_qgroup_rsv_type type)
74 {
75 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
76 if (qgroup->rsv.values[type] >= num_bytes) {
77 qgroup->rsv.values[type] -= num_bytes;
78 return;
79 }
80 #ifdef CONFIG_BTRFS_DEBUG
81 WARN_RATELIMIT(1,
82 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
83 qgroup->qgroupid, qgroup_rsv_type_str(type),
84 qgroup->rsv.values[type], num_bytes);
85 #endif
86 qgroup->rsv.values[type] = 0;
87 }
88
qgroup_rsv_add_by_qgroup(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * dest,struct btrfs_qgroup * src)89 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
90 struct btrfs_qgroup *dest,
91 struct btrfs_qgroup *src)
92 {
93 int i;
94
95 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
96 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
97 }
98
qgroup_rsv_release_by_qgroup(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * dest,struct btrfs_qgroup * src)99 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
100 struct btrfs_qgroup *dest,
101 struct btrfs_qgroup *src)
102 {
103 int i;
104
105 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
106 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
107 }
108
btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup * qg,u64 seq,int mod)109 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
110 int mod)
111 {
112 if (qg->old_refcnt < seq)
113 qg->old_refcnt = seq;
114 qg->old_refcnt += mod;
115 }
116
btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup * qg,u64 seq,int mod)117 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
118 int mod)
119 {
120 if (qg->new_refcnt < seq)
121 qg->new_refcnt = seq;
122 qg->new_refcnt += mod;
123 }
124
btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup * qg,u64 seq)125 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
126 {
127 if (qg->old_refcnt < seq)
128 return 0;
129 return qg->old_refcnt - seq;
130 }
131
btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup * qg,u64 seq)132 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
133 {
134 if (qg->new_refcnt < seq)
135 return 0;
136 return qg->new_refcnt - seq;
137 }
138
139 /*
140 * glue structure to represent the relations between qgroups.
141 */
142 struct btrfs_qgroup_list {
143 struct list_head next_group;
144 struct list_head next_member;
145 struct btrfs_qgroup *group;
146 struct btrfs_qgroup *member;
147 };
148
qgroup_to_aux(struct btrfs_qgroup * qg)149 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
150 {
151 return (u64)(uintptr_t)qg;
152 }
153
unode_aux_to_qgroup(struct ulist_node * n)154 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
155 {
156 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
157 }
158
159 static int
160 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
161 int init_flags);
162 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
163
164 /* must be called with qgroup_ioctl_lock held */
find_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)165 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
166 u64 qgroupid)
167 {
168 struct rb_node *n = fs_info->qgroup_tree.rb_node;
169 struct btrfs_qgroup *qgroup;
170
171 while (n) {
172 qgroup = rb_entry(n, struct btrfs_qgroup, node);
173 if (qgroup->qgroupid < qgroupid)
174 n = n->rb_left;
175 else if (qgroup->qgroupid > qgroupid)
176 n = n->rb_right;
177 else
178 return qgroup;
179 }
180 return NULL;
181 }
182
183 /* must be called with qgroup_lock held */
add_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)184 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
185 u64 qgroupid)
186 {
187 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
188 struct rb_node *parent = NULL;
189 struct btrfs_qgroup *qgroup;
190
191 while (*p) {
192 parent = *p;
193 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
194
195 if (qgroup->qgroupid < qgroupid)
196 p = &(*p)->rb_left;
197 else if (qgroup->qgroupid > qgroupid)
198 p = &(*p)->rb_right;
199 else
200 return qgroup;
201 }
202
203 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
204 if (!qgroup)
205 return ERR_PTR(-ENOMEM);
206
207 qgroup->qgroupid = qgroupid;
208 INIT_LIST_HEAD(&qgroup->groups);
209 INIT_LIST_HEAD(&qgroup->members);
210 INIT_LIST_HEAD(&qgroup->dirty);
211 INIT_LIST_HEAD(&qgroup->iterator);
212
213 rb_link_node(&qgroup->node, parent, p);
214 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
215
216 return qgroup;
217 }
218
__del_qgroup_rb(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup)219 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
220 struct btrfs_qgroup *qgroup)
221 {
222 struct btrfs_qgroup_list *list;
223
224 list_del(&qgroup->dirty);
225 while (!list_empty(&qgroup->groups)) {
226 list = list_first_entry(&qgroup->groups,
227 struct btrfs_qgroup_list, next_group);
228 list_del(&list->next_group);
229 list_del(&list->next_member);
230 kfree(list);
231 }
232
233 while (!list_empty(&qgroup->members)) {
234 list = list_first_entry(&qgroup->members,
235 struct btrfs_qgroup_list, next_member);
236 list_del(&list->next_group);
237 list_del(&list->next_member);
238 kfree(list);
239 }
240 }
241
242 /* must be called with qgroup_lock held */
del_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)243 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
244 {
245 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
246
247 if (!qgroup)
248 return -ENOENT;
249
250 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
251 __del_qgroup_rb(fs_info, qgroup);
252 return 0;
253 }
254
255 /*
256 * Add relation specified by two qgroups.
257 *
258 * Must be called with qgroup_lock held.
259 *
260 * Return: 0 on success
261 * -ENOENT if one of the qgroups is NULL
262 * <0 other errors
263 */
__add_relation_rb(struct btrfs_qgroup * member,struct btrfs_qgroup * parent)264 static int __add_relation_rb(struct btrfs_qgroup *member, struct btrfs_qgroup *parent)
265 {
266 struct btrfs_qgroup_list *list;
267
268 if (!member || !parent)
269 return -ENOENT;
270
271 list = kzalloc(sizeof(*list), GFP_ATOMIC);
272 if (!list)
273 return -ENOMEM;
274
275 list->group = parent;
276 list->member = member;
277 list_add_tail(&list->next_group, &member->groups);
278 list_add_tail(&list->next_member, &parent->members);
279
280 return 0;
281 }
282
283 /*
284 * Add relation specified by two qgroup ids.
285 *
286 * Must be called with qgroup_lock held.
287 *
288 * Return: 0 on success
289 * -ENOENT if one of the ids does not exist
290 * <0 other errors
291 */
add_relation_rb(struct btrfs_fs_info * fs_info,u64 memberid,u64 parentid)292 static int add_relation_rb(struct btrfs_fs_info *fs_info, u64 memberid, u64 parentid)
293 {
294 struct btrfs_qgroup *member;
295 struct btrfs_qgroup *parent;
296
297 member = find_qgroup_rb(fs_info, memberid);
298 parent = find_qgroup_rb(fs_info, parentid);
299
300 return __add_relation_rb(member, parent);
301 }
302
303 /* Must be called with qgroup_lock held */
del_relation_rb(struct btrfs_fs_info * fs_info,u64 memberid,u64 parentid)304 static int del_relation_rb(struct btrfs_fs_info *fs_info,
305 u64 memberid, u64 parentid)
306 {
307 struct btrfs_qgroup *member;
308 struct btrfs_qgroup *parent;
309 struct btrfs_qgroup_list *list;
310
311 member = find_qgroup_rb(fs_info, memberid);
312 parent = find_qgroup_rb(fs_info, parentid);
313 if (!member || !parent)
314 return -ENOENT;
315
316 list_for_each_entry(list, &member->groups, next_group) {
317 if (list->group == parent) {
318 list_del(&list->next_group);
319 list_del(&list->next_member);
320 kfree(list);
321 return 0;
322 }
323 }
324 return -ENOENT;
325 }
326
327 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_verify_qgroup_counts(struct btrfs_fs_info * fs_info,u64 qgroupid,u64 rfer,u64 excl)328 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
329 u64 rfer, u64 excl)
330 {
331 struct btrfs_qgroup *qgroup;
332
333 qgroup = find_qgroup_rb(fs_info, qgroupid);
334 if (!qgroup)
335 return -EINVAL;
336 if (qgroup->rfer != rfer || qgroup->excl != excl)
337 return -EINVAL;
338 return 0;
339 }
340 #endif
341
qgroup_mark_inconsistent(struct btrfs_fs_info * fs_info)342 static void qgroup_mark_inconsistent(struct btrfs_fs_info *fs_info)
343 {
344 fs_info->qgroup_flags |= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
345 BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
346 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
347 }
348
349 /*
350 * The full config is read in one go, only called from open_ctree()
351 * It doesn't use any locking, as at this point we're still single-threaded
352 */
btrfs_read_qgroup_config(struct btrfs_fs_info * fs_info)353 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
354 {
355 struct btrfs_key key;
356 struct btrfs_key found_key;
357 struct btrfs_root *quota_root = fs_info->quota_root;
358 struct btrfs_path *path = NULL;
359 struct extent_buffer *l;
360 int slot;
361 int ret = 0;
362 u64 flags = 0;
363 u64 rescan_progress = 0;
364
365 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
366 return 0;
367
368 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
369 if (!fs_info->qgroup_ulist) {
370 ret = -ENOMEM;
371 goto out;
372 }
373
374 path = btrfs_alloc_path();
375 if (!path) {
376 ret = -ENOMEM;
377 goto out;
378 }
379
380 ret = btrfs_sysfs_add_qgroups(fs_info);
381 if (ret < 0)
382 goto out;
383 /* default this to quota off, in case no status key is found */
384 fs_info->qgroup_flags = 0;
385
386 /*
387 * pass 1: read status, all qgroup infos and limits
388 */
389 key.objectid = 0;
390 key.type = 0;
391 key.offset = 0;
392 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
393 if (ret)
394 goto out;
395
396 while (1) {
397 struct btrfs_qgroup *qgroup;
398
399 slot = path->slots[0];
400 l = path->nodes[0];
401 btrfs_item_key_to_cpu(l, &found_key, slot);
402
403 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
404 struct btrfs_qgroup_status_item *ptr;
405
406 ptr = btrfs_item_ptr(l, slot,
407 struct btrfs_qgroup_status_item);
408
409 if (btrfs_qgroup_status_version(l, ptr) !=
410 BTRFS_QGROUP_STATUS_VERSION) {
411 btrfs_err(fs_info,
412 "old qgroup version, quota disabled");
413 goto out;
414 }
415 if (btrfs_qgroup_status_generation(l, ptr) !=
416 fs_info->generation) {
417 qgroup_mark_inconsistent(fs_info);
418 btrfs_err(fs_info,
419 "qgroup generation mismatch, marked as inconsistent");
420 }
421 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
422 ptr);
423 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
424 goto next1;
425 }
426
427 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
428 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
429 goto next1;
430
431 qgroup = find_qgroup_rb(fs_info, found_key.offset);
432 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
433 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
434 btrfs_err(fs_info, "inconsistent qgroup config");
435 qgroup_mark_inconsistent(fs_info);
436 }
437 if (!qgroup) {
438 qgroup = add_qgroup_rb(fs_info, found_key.offset);
439 if (IS_ERR(qgroup)) {
440 ret = PTR_ERR(qgroup);
441 goto out;
442 }
443 }
444 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
445 if (ret < 0)
446 goto out;
447
448 switch (found_key.type) {
449 case BTRFS_QGROUP_INFO_KEY: {
450 struct btrfs_qgroup_info_item *ptr;
451
452 ptr = btrfs_item_ptr(l, slot,
453 struct btrfs_qgroup_info_item);
454 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
455 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
456 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
457 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
458 /* generation currently unused */
459 break;
460 }
461 case BTRFS_QGROUP_LIMIT_KEY: {
462 struct btrfs_qgroup_limit_item *ptr;
463
464 ptr = btrfs_item_ptr(l, slot,
465 struct btrfs_qgroup_limit_item);
466 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
467 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
468 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
469 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
470 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
471 break;
472 }
473 }
474 next1:
475 ret = btrfs_next_item(quota_root, path);
476 if (ret < 0)
477 goto out;
478 if (ret)
479 break;
480 }
481 btrfs_release_path(path);
482
483 /*
484 * pass 2: read all qgroup relations
485 */
486 key.objectid = 0;
487 key.type = BTRFS_QGROUP_RELATION_KEY;
488 key.offset = 0;
489 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
490 if (ret)
491 goto out;
492 while (1) {
493 slot = path->slots[0];
494 l = path->nodes[0];
495 btrfs_item_key_to_cpu(l, &found_key, slot);
496
497 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
498 goto next2;
499
500 if (found_key.objectid > found_key.offset) {
501 /* parent <- member, not needed to build config */
502 /* FIXME should we omit the key completely? */
503 goto next2;
504 }
505
506 ret = add_relation_rb(fs_info, found_key.objectid,
507 found_key.offset);
508 if (ret == -ENOENT) {
509 btrfs_warn(fs_info,
510 "orphan qgroup relation 0x%llx->0x%llx",
511 found_key.objectid, found_key.offset);
512 ret = 0; /* ignore the error */
513 }
514 if (ret)
515 goto out;
516 next2:
517 ret = btrfs_next_item(quota_root, path);
518 if (ret < 0)
519 goto out;
520 if (ret)
521 break;
522 }
523 out:
524 btrfs_free_path(path);
525 fs_info->qgroup_flags |= flags;
526 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
527 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
528 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
529 ret >= 0)
530 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
531
532 if (ret < 0) {
533 ulist_free(fs_info->qgroup_ulist);
534 fs_info->qgroup_ulist = NULL;
535 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
536 btrfs_sysfs_del_qgroups(fs_info);
537 }
538
539 return ret < 0 ? ret : 0;
540 }
541
542 /*
543 * Called in close_ctree() when quota is still enabled. This verifies we don't
544 * leak some reserved space.
545 *
546 * Return false if no reserved space is left.
547 * Return true if some reserved space is leaked.
548 */
btrfs_check_quota_leak(struct btrfs_fs_info * fs_info)549 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
550 {
551 struct rb_node *node;
552 bool ret = false;
553
554 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
555 return ret;
556 /*
557 * Since we're unmounting, there is no race and no need to grab qgroup
558 * lock. And here we don't go post-order to provide a more user
559 * friendly sorted result.
560 */
561 for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
562 struct btrfs_qgroup *qgroup;
563 int i;
564
565 qgroup = rb_entry(node, struct btrfs_qgroup, node);
566 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
567 if (qgroup->rsv.values[i]) {
568 ret = true;
569 btrfs_warn(fs_info,
570 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
571 btrfs_qgroup_level(qgroup->qgroupid),
572 btrfs_qgroup_subvolid(qgroup->qgroupid),
573 i, qgroup->rsv.values[i]);
574 }
575 }
576 }
577 return ret;
578 }
579
580 /*
581 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
582 * first two are in single-threaded paths.And for the third one, we have set
583 * quota_root to be null with qgroup_lock held before, so it is safe to clean
584 * up the in-memory structures without qgroup_lock held.
585 */
btrfs_free_qgroup_config(struct btrfs_fs_info * fs_info)586 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
587 {
588 struct rb_node *n;
589 struct btrfs_qgroup *qgroup;
590
591 while ((n = rb_first(&fs_info->qgroup_tree))) {
592 qgroup = rb_entry(n, struct btrfs_qgroup, node);
593 rb_erase(n, &fs_info->qgroup_tree);
594 __del_qgroup_rb(fs_info, qgroup);
595 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
596 kfree(qgroup);
597 }
598 /*
599 * We call btrfs_free_qgroup_config() when unmounting
600 * filesystem and disabling quota, so we set qgroup_ulist
601 * to be null here to avoid double free.
602 */
603 ulist_free(fs_info->qgroup_ulist);
604 fs_info->qgroup_ulist = NULL;
605 btrfs_sysfs_del_qgroups(fs_info);
606 }
607
add_qgroup_relation_item(struct btrfs_trans_handle * trans,u64 src,u64 dst)608 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
609 u64 dst)
610 {
611 int ret;
612 struct btrfs_root *quota_root = trans->fs_info->quota_root;
613 struct btrfs_path *path;
614 struct btrfs_key key;
615
616 path = btrfs_alloc_path();
617 if (!path)
618 return -ENOMEM;
619
620 key.objectid = src;
621 key.type = BTRFS_QGROUP_RELATION_KEY;
622 key.offset = dst;
623
624 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
625
626 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
627
628 btrfs_free_path(path);
629 return ret;
630 }
631
del_qgroup_relation_item(struct btrfs_trans_handle * trans,u64 src,u64 dst)632 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
633 u64 dst)
634 {
635 int ret;
636 struct btrfs_root *quota_root = trans->fs_info->quota_root;
637 struct btrfs_path *path;
638 struct btrfs_key key;
639
640 path = btrfs_alloc_path();
641 if (!path)
642 return -ENOMEM;
643
644 key.objectid = src;
645 key.type = BTRFS_QGROUP_RELATION_KEY;
646 key.offset = dst;
647
648 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
649 if (ret < 0)
650 goto out;
651
652 if (ret > 0) {
653 ret = -ENOENT;
654 goto out;
655 }
656
657 ret = btrfs_del_item(trans, quota_root, path);
658 out:
659 btrfs_free_path(path);
660 return ret;
661 }
662
add_qgroup_item(struct btrfs_trans_handle * trans,struct btrfs_root * quota_root,u64 qgroupid)663 static int add_qgroup_item(struct btrfs_trans_handle *trans,
664 struct btrfs_root *quota_root, u64 qgroupid)
665 {
666 int ret;
667 struct btrfs_path *path;
668 struct btrfs_qgroup_info_item *qgroup_info;
669 struct btrfs_qgroup_limit_item *qgroup_limit;
670 struct extent_buffer *leaf;
671 struct btrfs_key key;
672
673 if (btrfs_is_testing(quota_root->fs_info))
674 return 0;
675
676 path = btrfs_alloc_path();
677 if (!path)
678 return -ENOMEM;
679
680 key.objectid = 0;
681 key.type = BTRFS_QGROUP_INFO_KEY;
682 key.offset = qgroupid;
683
684 /*
685 * Avoid a transaction abort by catching -EEXIST here. In that
686 * case, we proceed by re-initializing the existing structure
687 * on disk.
688 */
689
690 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
691 sizeof(*qgroup_info));
692 if (ret && ret != -EEXIST)
693 goto out;
694
695 leaf = path->nodes[0];
696 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
697 struct btrfs_qgroup_info_item);
698 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
699 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
700 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
701 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
702 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
703
704 btrfs_mark_buffer_dirty(trans, leaf);
705
706 btrfs_release_path(path);
707
708 key.type = BTRFS_QGROUP_LIMIT_KEY;
709 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
710 sizeof(*qgroup_limit));
711 if (ret && ret != -EEXIST)
712 goto out;
713
714 leaf = path->nodes[0];
715 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
716 struct btrfs_qgroup_limit_item);
717 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
718 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
719 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
720 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
721 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
722
723 btrfs_mark_buffer_dirty(trans, leaf);
724
725 ret = 0;
726 out:
727 btrfs_free_path(path);
728 return ret;
729 }
730
del_qgroup_item(struct btrfs_trans_handle * trans,u64 qgroupid)731 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
732 {
733 int ret;
734 struct btrfs_root *quota_root = trans->fs_info->quota_root;
735 struct btrfs_path *path;
736 struct btrfs_key key;
737
738 path = btrfs_alloc_path();
739 if (!path)
740 return -ENOMEM;
741
742 key.objectid = 0;
743 key.type = BTRFS_QGROUP_INFO_KEY;
744 key.offset = qgroupid;
745 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
746 if (ret < 0)
747 goto out;
748
749 if (ret > 0) {
750 ret = -ENOENT;
751 goto out;
752 }
753
754 ret = btrfs_del_item(trans, quota_root, path);
755 if (ret)
756 goto out;
757
758 btrfs_release_path(path);
759
760 key.type = BTRFS_QGROUP_LIMIT_KEY;
761 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
762 if (ret < 0)
763 goto out;
764
765 if (ret > 0) {
766 ret = -ENOENT;
767 goto out;
768 }
769
770 ret = btrfs_del_item(trans, quota_root, path);
771
772 out:
773 btrfs_free_path(path);
774 return ret;
775 }
776
update_qgroup_limit_item(struct btrfs_trans_handle * trans,struct btrfs_qgroup * qgroup)777 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
778 struct btrfs_qgroup *qgroup)
779 {
780 struct btrfs_root *quota_root = trans->fs_info->quota_root;
781 struct btrfs_path *path;
782 struct btrfs_key key;
783 struct extent_buffer *l;
784 struct btrfs_qgroup_limit_item *qgroup_limit;
785 int ret;
786 int slot;
787
788 key.objectid = 0;
789 key.type = BTRFS_QGROUP_LIMIT_KEY;
790 key.offset = qgroup->qgroupid;
791
792 path = btrfs_alloc_path();
793 if (!path)
794 return -ENOMEM;
795
796 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
797 if (ret > 0)
798 ret = -ENOENT;
799
800 if (ret)
801 goto out;
802
803 l = path->nodes[0];
804 slot = path->slots[0];
805 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
806 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
807 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
808 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
809 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
810 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
811
812 btrfs_mark_buffer_dirty(trans, l);
813
814 out:
815 btrfs_free_path(path);
816 return ret;
817 }
818
update_qgroup_info_item(struct btrfs_trans_handle * trans,struct btrfs_qgroup * qgroup)819 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
820 struct btrfs_qgroup *qgroup)
821 {
822 struct btrfs_fs_info *fs_info = trans->fs_info;
823 struct btrfs_root *quota_root = fs_info->quota_root;
824 struct btrfs_path *path;
825 struct btrfs_key key;
826 struct extent_buffer *l;
827 struct btrfs_qgroup_info_item *qgroup_info;
828 int ret;
829 int slot;
830
831 if (btrfs_is_testing(fs_info))
832 return 0;
833
834 key.objectid = 0;
835 key.type = BTRFS_QGROUP_INFO_KEY;
836 key.offset = qgroup->qgroupid;
837
838 path = btrfs_alloc_path();
839 if (!path)
840 return -ENOMEM;
841
842 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
843 if (ret > 0)
844 ret = -ENOENT;
845
846 if (ret)
847 goto out;
848
849 l = path->nodes[0];
850 slot = path->slots[0];
851 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
852 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
853 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
854 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
855 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
856 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
857
858 btrfs_mark_buffer_dirty(trans, l);
859
860 out:
861 btrfs_free_path(path);
862 return ret;
863 }
864
update_qgroup_status_item(struct btrfs_trans_handle * trans)865 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
866 {
867 struct btrfs_fs_info *fs_info = trans->fs_info;
868 struct btrfs_root *quota_root = fs_info->quota_root;
869 struct btrfs_path *path;
870 struct btrfs_key key;
871 struct extent_buffer *l;
872 struct btrfs_qgroup_status_item *ptr;
873 int ret;
874 int slot;
875
876 key.objectid = 0;
877 key.type = BTRFS_QGROUP_STATUS_KEY;
878 key.offset = 0;
879
880 path = btrfs_alloc_path();
881 if (!path)
882 return -ENOMEM;
883
884 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
885 if (ret > 0)
886 ret = -ENOENT;
887
888 if (ret)
889 goto out;
890
891 l = path->nodes[0];
892 slot = path->slots[0];
893 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
894 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags &
895 BTRFS_QGROUP_STATUS_FLAGS_MASK);
896 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
897 btrfs_set_qgroup_status_rescan(l, ptr,
898 fs_info->qgroup_rescan_progress.objectid);
899
900 btrfs_mark_buffer_dirty(trans, l);
901
902 out:
903 btrfs_free_path(path);
904 return ret;
905 }
906
907 /*
908 * called with qgroup_lock held
909 */
btrfs_clean_quota_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)910 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
911 struct btrfs_root *root)
912 {
913 struct btrfs_path *path;
914 struct btrfs_key key;
915 struct extent_buffer *leaf = NULL;
916 int ret;
917 int nr = 0;
918
919 path = btrfs_alloc_path();
920 if (!path)
921 return -ENOMEM;
922
923 key.objectid = 0;
924 key.offset = 0;
925 key.type = 0;
926
927 while (1) {
928 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
929 if (ret < 0)
930 goto out;
931 leaf = path->nodes[0];
932 nr = btrfs_header_nritems(leaf);
933 if (!nr)
934 break;
935 /*
936 * delete the leaf one by one
937 * since the whole tree is going
938 * to be deleted.
939 */
940 path->slots[0] = 0;
941 ret = btrfs_del_items(trans, root, path, 0, nr);
942 if (ret)
943 goto out;
944
945 btrfs_release_path(path);
946 }
947 ret = 0;
948 out:
949 btrfs_free_path(path);
950 return ret;
951 }
952
btrfs_quota_enable(struct btrfs_fs_info * fs_info)953 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
954 {
955 struct btrfs_root *quota_root;
956 struct btrfs_root *tree_root = fs_info->tree_root;
957 struct btrfs_path *path = NULL;
958 struct btrfs_qgroup_status_item *ptr;
959 struct extent_buffer *leaf;
960 struct btrfs_key key;
961 struct btrfs_key found_key;
962 struct btrfs_qgroup *qgroup = NULL;
963 struct btrfs_trans_handle *trans = NULL;
964 struct ulist *ulist = NULL;
965 int ret = 0;
966 int slot;
967
968 /*
969 * We need to have subvol_sem write locked, to prevent races between
970 * concurrent tasks trying to enable quotas, because we will unlock
971 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
972 * and before setting BTRFS_FS_QUOTA_ENABLED.
973 */
974 lockdep_assert_held_write(&fs_info->subvol_sem);
975
976 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
977 btrfs_err(fs_info,
978 "qgroups are currently unsupported in extent tree v2");
979 return -EINVAL;
980 }
981
982 mutex_lock(&fs_info->qgroup_ioctl_lock);
983 if (fs_info->quota_root)
984 goto out;
985
986 ulist = ulist_alloc(GFP_KERNEL);
987 if (!ulist) {
988 ret = -ENOMEM;
989 goto out;
990 }
991
992 ret = btrfs_sysfs_add_qgroups(fs_info);
993 if (ret < 0)
994 goto out;
995
996 /*
997 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
998 * avoid lock acquisition inversion problems (reported by lockdep) between
999 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
1000 * start a transaction.
1001 * After we started the transaction lock qgroup_ioctl_lock again and
1002 * check if someone else created the quota root in the meanwhile. If so,
1003 * just return success and release the transaction handle.
1004 *
1005 * Also we don't need to worry about someone else calling
1006 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
1007 * that function returns 0 (success) when the sysfs entries already exist.
1008 */
1009 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1010
1011 /*
1012 * 1 for quota root item
1013 * 1 for BTRFS_QGROUP_STATUS item
1014 *
1015 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1016 * per subvolume. However those are not currently reserved since it
1017 * would be a lot of overkill.
1018 */
1019 trans = btrfs_start_transaction(tree_root, 2);
1020
1021 mutex_lock(&fs_info->qgroup_ioctl_lock);
1022 if (IS_ERR(trans)) {
1023 ret = PTR_ERR(trans);
1024 trans = NULL;
1025 goto out;
1026 }
1027
1028 if (fs_info->quota_root)
1029 goto out;
1030
1031 fs_info->qgroup_ulist = ulist;
1032 ulist = NULL;
1033
1034 /*
1035 * initially create the quota tree
1036 */
1037 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1038 if (IS_ERR(quota_root)) {
1039 ret = PTR_ERR(quota_root);
1040 btrfs_abort_transaction(trans, ret);
1041 goto out;
1042 }
1043
1044 path = btrfs_alloc_path();
1045 if (!path) {
1046 ret = -ENOMEM;
1047 btrfs_abort_transaction(trans, ret);
1048 goto out_free_root;
1049 }
1050
1051 key.objectid = 0;
1052 key.type = BTRFS_QGROUP_STATUS_KEY;
1053 key.offset = 0;
1054
1055 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1056 sizeof(*ptr));
1057 if (ret) {
1058 btrfs_abort_transaction(trans, ret);
1059 goto out_free_path;
1060 }
1061
1062 leaf = path->nodes[0];
1063 ptr = btrfs_item_ptr(leaf, path->slots[0],
1064 struct btrfs_qgroup_status_item);
1065 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1066 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1067 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1068 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1069 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
1070 BTRFS_QGROUP_STATUS_FLAGS_MASK);
1071 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1072
1073 btrfs_mark_buffer_dirty(trans, leaf);
1074
1075 key.objectid = 0;
1076 key.type = BTRFS_ROOT_REF_KEY;
1077 key.offset = 0;
1078
1079 btrfs_release_path(path);
1080 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1081 if (ret > 0)
1082 goto out_add_root;
1083 if (ret < 0) {
1084 btrfs_abort_transaction(trans, ret);
1085 goto out_free_path;
1086 }
1087
1088 while (1) {
1089 slot = path->slots[0];
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1092
1093 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1094
1095 /* Release locks on tree_root before we access quota_root */
1096 btrfs_release_path(path);
1097
1098 ret = add_qgroup_item(trans, quota_root,
1099 found_key.offset);
1100 if (ret) {
1101 btrfs_abort_transaction(trans, ret);
1102 goto out_free_path;
1103 }
1104
1105 qgroup = add_qgroup_rb(fs_info, found_key.offset);
1106 if (IS_ERR(qgroup)) {
1107 ret = PTR_ERR(qgroup);
1108 btrfs_abort_transaction(trans, ret);
1109 goto out_free_path;
1110 }
1111 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1112 if (ret < 0) {
1113 btrfs_abort_transaction(trans, ret);
1114 goto out_free_path;
1115 }
1116 ret = btrfs_search_slot_for_read(tree_root, &found_key,
1117 path, 1, 0);
1118 if (ret < 0) {
1119 btrfs_abort_transaction(trans, ret);
1120 goto out_free_path;
1121 }
1122 if (ret > 0) {
1123 /*
1124 * Shouldn't happen, but in case it does we
1125 * don't need to do the btrfs_next_item, just
1126 * continue.
1127 */
1128 continue;
1129 }
1130 }
1131 ret = btrfs_next_item(tree_root, path);
1132 if (ret < 0) {
1133 btrfs_abort_transaction(trans, ret);
1134 goto out_free_path;
1135 }
1136 if (ret)
1137 break;
1138 }
1139
1140 out_add_root:
1141 btrfs_release_path(path);
1142 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1143 if (ret) {
1144 btrfs_abort_transaction(trans, ret);
1145 goto out_free_path;
1146 }
1147
1148 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1149 if (IS_ERR(qgroup)) {
1150 ret = PTR_ERR(qgroup);
1151 btrfs_abort_transaction(trans, ret);
1152 goto out_free_path;
1153 }
1154 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1155 if (ret < 0) {
1156 btrfs_abort_transaction(trans, ret);
1157 goto out_free_path;
1158 }
1159
1160 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1161 /*
1162 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1163 * a deadlock with tasks concurrently doing other qgroup operations, such
1164 * adding/removing qgroups or adding/deleting qgroup relations for example,
1165 * because all qgroup operations first start or join a transaction and then
1166 * lock the qgroup_ioctl_lock mutex.
1167 * We are safe from a concurrent task trying to enable quotas, by calling
1168 * this function, since we are serialized by fs_info->subvol_sem.
1169 */
1170 ret = btrfs_commit_transaction(trans);
1171 trans = NULL;
1172 mutex_lock(&fs_info->qgroup_ioctl_lock);
1173 if (ret)
1174 goto out_free_path;
1175
1176 /*
1177 * Set quota enabled flag after committing the transaction, to avoid
1178 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1179 * creation.
1180 */
1181 spin_lock(&fs_info->qgroup_lock);
1182 fs_info->quota_root = quota_root;
1183 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1184 spin_unlock(&fs_info->qgroup_lock);
1185
1186 ret = qgroup_rescan_init(fs_info, 0, 1);
1187 if (!ret) {
1188 qgroup_rescan_zero_tracking(fs_info);
1189 fs_info->qgroup_rescan_running = true;
1190 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1191 &fs_info->qgroup_rescan_work);
1192 } else {
1193 /*
1194 * We have set both BTRFS_FS_QUOTA_ENABLED and
1195 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1196 * -EINPROGRESS. That can happen because someone started the
1197 * rescan worker by calling quota rescan ioctl before we
1198 * attempted to initialize the rescan worker. Failure due to
1199 * quotas disabled in the meanwhile is not possible, because
1200 * we are holding a write lock on fs_info->subvol_sem, which
1201 * is also acquired when disabling quotas.
1202 * Ignore such error, and any other error would need to undo
1203 * everything we did in the transaction we just committed.
1204 */
1205 ASSERT(ret == -EINPROGRESS);
1206 ret = 0;
1207 }
1208
1209 out_free_path:
1210 btrfs_free_path(path);
1211 out_free_root:
1212 if (ret)
1213 btrfs_put_root(quota_root);
1214 out:
1215 if (ret) {
1216 ulist_free(fs_info->qgroup_ulist);
1217 fs_info->qgroup_ulist = NULL;
1218 btrfs_sysfs_del_qgroups(fs_info);
1219 }
1220 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1221 if (ret && trans)
1222 btrfs_end_transaction(trans);
1223 else if (trans)
1224 ret = btrfs_end_transaction(trans);
1225 ulist_free(ulist);
1226 return ret;
1227 }
1228
btrfs_quota_disable(struct btrfs_fs_info * fs_info)1229 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1230 {
1231 struct btrfs_root *quota_root;
1232 struct btrfs_trans_handle *trans = NULL;
1233 int ret = 0;
1234
1235 /*
1236 * We need to have subvol_sem write locked to prevent races with
1237 * snapshot creation.
1238 */
1239 lockdep_assert_held_write(&fs_info->subvol_sem);
1240
1241 /*
1242 * Lock the cleaner mutex to prevent races with concurrent relocation,
1243 * because relocation may be building backrefs for blocks of the quota
1244 * root while we are deleting the root. This is like dropping fs roots
1245 * of deleted snapshots/subvolumes, we need the same protection.
1246 *
1247 * This also prevents races between concurrent tasks trying to disable
1248 * quotas, because we will unlock and relock qgroup_ioctl_lock across
1249 * BTRFS_FS_QUOTA_ENABLED changes.
1250 */
1251 mutex_lock(&fs_info->cleaner_mutex);
1252
1253 mutex_lock(&fs_info->qgroup_ioctl_lock);
1254 if (!fs_info->quota_root)
1255 goto out;
1256
1257 /*
1258 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1259 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1260 * to lock that mutex while holding a transaction handle and the rescan
1261 * worker needs to commit a transaction.
1262 */
1263 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1264
1265 /*
1266 * Request qgroup rescan worker to complete and wait for it. This wait
1267 * must be done before transaction start for quota disable since it may
1268 * deadlock with transaction by the qgroup rescan worker.
1269 */
1270 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1271 btrfs_qgroup_wait_for_completion(fs_info, false);
1272
1273 /*
1274 * 1 For the root item
1275 *
1276 * We should also reserve enough items for the quota tree deletion in
1277 * btrfs_clean_quota_tree but this is not done.
1278 *
1279 * Also, we must always start a transaction without holding the mutex
1280 * qgroup_ioctl_lock, see btrfs_quota_enable().
1281 */
1282 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1283
1284 mutex_lock(&fs_info->qgroup_ioctl_lock);
1285 if (IS_ERR(trans)) {
1286 ret = PTR_ERR(trans);
1287 trans = NULL;
1288 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1289 goto out;
1290 }
1291
1292 if (!fs_info->quota_root)
1293 goto out;
1294
1295 spin_lock(&fs_info->qgroup_lock);
1296 quota_root = fs_info->quota_root;
1297 fs_info->quota_root = NULL;
1298 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1299 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1300 spin_unlock(&fs_info->qgroup_lock);
1301
1302 btrfs_free_qgroup_config(fs_info);
1303
1304 ret = btrfs_clean_quota_tree(trans, quota_root);
1305 if (ret) {
1306 btrfs_abort_transaction(trans, ret);
1307 goto out;
1308 }
1309
1310 ret = btrfs_del_root(trans, "a_root->root_key);
1311 if (ret) {
1312 btrfs_abort_transaction(trans, ret);
1313 goto out;
1314 }
1315
1316 spin_lock(&fs_info->trans_lock);
1317 list_del("a_root->dirty_list);
1318 spin_unlock(&fs_info->trans_lock);
1319
1320 btrfs_tree_lock(quota_root->node);
1321 btrfs_clear_buffer_dirty(trans, quota_root->node);
1322 btrfs_tree_unlock(quota_root->node);
1323 btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1324 quota_root->node, 0, 1);
1325
1326 btrfs_put_root(quota_root);
1327
1328 out:
1329 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1330 if (ret && trans)
1331 btrfs_end_transaction(trans);
1332 else if (trans)
1333 ret = btrfs_end_transaction(trans);
1334 mutex_unlock(&fs_info->cleaner_mutex);
1335
1336 return ret;
1337 }
1338
qgroup_dirty(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup)1339 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1340 struct btrfs_qgroup *qgroup)
1341 {
1342 if (list_empty(&qgroup->dirty))
1343 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1344 }
1345
qgroup_iterator_add(struct list_head * head,struct btrfs_qgroup * qgroup)1346 static void qgroup_iterator_add(struct list_head *head, struct btrfs_qgroup *qgroup)
1347 {
1348 if (!list_empty(&qgroup->iterator))
1349 return;
1350
1351 list_add_tail(&qgroup->iterator, head);
1352 }
1353
qgroup_iterator_clean(struct list_head * head)1354 static void qgroup_iterator_clean(struct list_head *head)
1355 {
1356 while (!list_empty(head)) {
1357 struct btrfs_qgroup *qgroup;
1358
1359 qgroup = list_first_entry(head, struct btrfs_qgroup, iterator);
1360 list_del_init(&qgroup->iterator);
1361 }
1362 }
1363
1364 /*
1365 * The easy accounting, we're updating qgroup relationship whose child qgroup
1366 * only has exclusive extents.
1367 *
1368 * In this case, all exclusive extents will also be exclusive for parent, so
1369 * excl/rfer just get added/removed.
1370 *
1371 * So is qgroup reservation space, which should also be added/removed to
1372 * parent.
1373 * Or when child tries to release reservation space, parent will underflow its
1374 * reservation (for relationship adding case).
1375 *
1376 * Caller should hold fs_info->qgroup_lock.
1377 */
__qgroup_excl_accounting(struct btrfs_fs_info * fs_info,struct ulist * tmp,u64 ref_root,struct btrfs_qgroup * src,int sign)1378 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1379 struct ulist *tmp, u64 ref_root,
1380 struct btrfs_qgroup *src, int sign)
1381 {
1382 struct btrfs_qgroup *qgroup;
1383 struct btrfs_qgroup_list *glist;
1384 struct ulist_node *unode;
1385 struct ulist_iterator uiter;
1386 u64 num_bytes = src->excl;
1387 int ret = 0;
1388
1389 qgroup = find_qgroup_rb(fs_info, ref_root);
1390 if (!qgroup)
1391 goto out;
1392
1393 qgroup->rfer += sign * num_bytes;
1394 qgroup->rfer_cmpr += sign * num_bytes;
1395
1396 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1397 qgroup->excl += sign * num_bytes;
1398 qgroup->excl_cmpr += sign * num_bytes;
1399
1400 if (sign > 0)
1401 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1402 else
1403 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1404
1405 qgroup_dirty(fs_info, qgroup);
1406
1407 /* Get all of the parent groups that contain this qgroup */
1408 list_for_each_entry(glist, &qgroup->groups, next_group) {
1409 ret = ulist_add(tmp, glist->group->qgroupid,
1410 qgroup_to_aux(glist->group), GFP_ATOMIC);
1411 if (ret < 0)
1412 goto out;
1413 }
1414
1415 /* Iterate all of the parents and adjust their reference counts */
1416 ULIST_ITER_INIT(&uiter);
1417 while ((unode = ulist_next(tmp, &uiter))) {
1418 qgroup = unode_aux_to_qgroup(unode);
1419 qgroup->rfer += sign * num_bytes;
1420 qgroup->rfer_cmpr += sign * num_bytes;
1421 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1422 qgroup->excl += sign * num_bytes;
1423 if (sign > 0)
1424 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1425 else
1426 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1427 qgroup->excl_cmpr += sign * num_bytes;
1428 qgroup_dirty(fs_info, qgroup);
1429
1430 /* Add any parents of the parents */
1431 list_for_each_entry(glist, &qgroup->groups, next_group) {
1432 ret = ulist_add(tmp, glist->group->qgroupid,
1433 qgroup_to_aux(glist->group), GFP_ATOMIC);
1434 if (ret < 0)
1435 goto out;
1436 }
1437 }
1438 ret = 0;
1439 out:
1440 return ret;
1441 }
1442
1443
1444 /*
1445 * Quick path for updating qgroup with only excl refs.
1446 *
1447 * In that case, just update all parent will be enough.
1448 * Or we needs to do a full rescan.
1449 * Caller should also hold fs_info->qgroup_lock.
1450 *
1451 * Return 0 for quick update, return >0 for need to full rescan
1452 * and mark INCONSISTENT flag.
1453 * Return < 0 for other error.
1454 */
quick_update_accounting(struct btrfs_fs_info * fs_info,struct ulist * tmp,u64 src,u64 dst,int sign)1455 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1456 struct ulist *tmp, u64 src, u64 dst,
1457 int sign)
1458 {
1459 struct btrfs_qgroup *qgroup;
1460 int ret = 1;
1461 int err = 0;
1462
1463 qgroup = find_qgroup_rb(fs_info, src);
1464 if (!qgroup)
1465 goto out;
1466 if (qgroup->excl == qgroup->rfer) {
1467 ret = 0;
1468 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1469 qgroup, sign);
1470 if (err < 0) {
1471 ret = err;
1472 goto out;
1473 }
1474 }
1475 out:
1476 if (ret)
1477 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1478 return ret;
1479 }
1480
btrfs_add_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1481 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1482 u64 dst)
1483 {
1484 struct btrfs_fs_info *fs_info = trans->fs_info;
1485 struct btrfs_qgroup *parent;
1486 struct btrfs_qgroup *member;
1487 struct btrfs_qgroup_list *list;
1488 struct ulist *tmp;
1489 unsigned int nofs_flag;
1490 int ret = 0;
1491
1492 /* Check the level of src and dst first */
1493 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1494 return -EINVAL;
1495
1496 /* We hold a transaction handle open, must do a NOFS allocation. */
1497 nofs_flag = memalloc_nofs_save();
1498 tmp = ulist_alloc(GFP_KERNEL);
1499 memalloc_nofs_restore(nofs_flag);
1500 if (!tmp)
1501 return -ENOMEM;
1502
1503 mutex_lock(&fs_info->qgroup_ioctl_lock);
1504 if (!fs_info->quota_root) {
1505 ret = -ENOTCONN;
1506 goto out;
1507 }
1508 member = find_qgroup_rb(fs_info, src);
1509 parent = find_qgroup_rb(fs_info, dst);
1510 if (!member || !parent) {
1511 ret = -EINVAL;
1512 goto out;
1513 }
1514
1515 /* check if such qgroup relation exist firstly */
1516 list_for_each_entry(list, &member->groups, next_group) {
1517 if (list->group == parent) {
1518 ret = -EEXIST;
1519 goto out;
1520 }
1521 }
1522
1523 ret = add_qgroup_relation_item(trans, src, dst);
1524 if (ret)
1525 goto out;
1526
1527 ret = add_qgroup_relation_item(trans, dst, src);
1528 if (ret) {
1529 del_qgroup_relation_item(trans, src, dst);
1530 goto out;
1531 }
1532
1533 spin_lock(&fs_info->qgroup_lock);
1534 ret = __add_relation_rb(member, parent);
1535 if (ret < 0) {
1536 spin_unlock(&fs_info->qgroup_lock);
1537 goto out;
1538 }
1539 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1540 spin_unlock(&fs_info->qgroup_lock);
1541 out:
1542 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1543 ulist_free(tmp);
1544 return ret;
1545 }
1546
__del_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1547 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1548 u64 dst)
1549 {
1550 struct btrfs_fs_info *fs_info = trans->fs_info;
1551 struct btrfs_qgroup *parent;
1552 struct btrfs_qgroup *member;
1553 struct btrfs_qgroup_list *list;
1554 struct ulist *tmp;
1555 bool found = false;
1556 unsigned int nofs_flag;
1557 int ret = 0;
1558 int ret2;
1559
1560 /* We hold a transaction handle open, must do a NOFS allocation. */
1561 nofs_flag = memalloc_nofs_save();
1562 tmp = ulist_alloc(GFP_KERNEL);
1563 memalloc_nofs_restore(nofs_flag);
1564 if (!tmp)
1565 return -ENOMEM;
1566
1567 if (!fs_info->quota_root) {
1568 ret = -ENOTCONN;
1569 goto out;
1570 }
1571
1572 member = find_qgroup_rb(fs_info, src);
1573 parent = find_qgroup_rb(fs_info, dst);
1574 /*
1575 * The parent/member pair doesn't exist, then try to delete the dead
1576 * relation items only.
1577 */
1578 if (!member || !parent)
1579 goto delete_item;
1580
1581 /* check if such qgroup relation exist firstly */
1582 list_for_each_entry(list, &member->groups, next_group) {
1583 if (list->group == parent) {
1584 found = true;
1585 break;
1586 }
1587 }
1588
1589 delete_item:
1590 ret = del_qgroup_relation_item(trans, src, dst);
1591 if (ret < 0 && ret != -ENOENT)
1592 goto out;
1593 ret2 = del_qgroup_relation_item(trans, dst, src);
1594 if (ret2 < 0 && ret2 != -ENOENT)
1595 goto out;
1596
1597 /* At least one deletion succeeded, return 0 */
1598 if (!ret || !ret2)
1599 ret = 0;
1600
1601 if (found) {
1602 spin_lock(&fs_info->qgroup_lock);
1603 del_relation_rb(fs_info, src, dst);
1604 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1605 spin_unlock(&fs_info->qgroup_lock);
1606 }
1607 out:
1608 ulist_free(tmp);
1609 return ret;
1610 }
1611
btrfs_del_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1612 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1613 u64 dst)
1614 {
1615 struct btrfs_fs_info *fs_info = trans->fs_info;
1616 int ret = 0;
1617
1618 mutex_lock(&fs_info->qgroup_ioctl_lock);
1619 ret = __del_qgroup_relation(trans, src, dst);
1620 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1621
1622 return ret;
1623 }
1624
btrfs_create_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)1625 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1626 {
1627 struct btrfs_fs_info *fs_info = trans->fs_info;
1628 struct btrfs_root *quota_root;
1629 struct btrfs_qgroup *qgroup;
1630 int ret = 0;
1631
1632 mutex_lock(&fs_info->qgroup_ioctl_lock);
1633 if (!fs_info->quota_root) {
1634 ret = -ENOTCONN;
1635 goto out;
1636 }
1637 quota_root = fs_info->quota_root;
1638 qgroup = find_qgroup_rb(fs_info, qgroupid);
1639 if (qgroup) {
1640 ret = -EEXIST;
1641 goto out;
1642 }
1643
1644 ret = add_qgroup_item(trans, quota_root, qgroupid);
1645 if (ret)
1646 goto out;
1647
1648 spin_lock(&fs_info->qgroup_lock);
1649 qgroup = add_qgroup_rb(fs_info, qgroupid);
1650 spin_unlock(&fs_info->qgroup_lock);
1651
1652 if (IS_ERR(qgroup)) {
1653 ret = PTR_ERR(qgroup);
1654 goto out;
1655 }
1656 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1657 out:
1658 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1659 return ret;
1660 }
1661
qgroup_has_usage(struct btrfs_qgroup * qgroup)1662 static bool qgroup_has_usage(struct btrfs_qgroup *qgroup)
1663 {
1664 return (qgroup->rfer > 0 || qgroup->rfer_cmpr > 0 ||
1665 qgroup->excl > 0 || qgroup->excl_cmpr > 0 ||
1666 qgroup->rsv.values[BTRFS_QGROUP_RSV_DATA] > 0 ||
1667 qgroup->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC] > 0 ||
1668 qgroup->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > 0);
1669 }
1670
btrfs_remove_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)1671 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1672 {
1673 struct btrfs_fs_info *fs_info = trans->fs_info;
1674 struct btrfs_qgroup *qgroup;
1675 struct btrfs_qgroup_list *list;
1676 int ret = 0;
1677
1678 mutex_lock(&fs_info->qgroup_ioctl_lock);
1679 if (!fs_info->quota_root) {
1680 ret = -ENOTCONN;
1681 goto out;
1682 }
1683
1684 qgroup = find_qgroup_rb(fs_info, qgroupid);
1685 if (!qgroup) {
1686 ret = -ENOENT;
1687 goto out;
1688 }
1689
1690 if (is_fstree(qgroupid) && qgroup_has_usage(qgroup)) {
1691 ret = -EBUSY;
1692 goto out;
1693 }
1694
1695 /* Check if there are no children of this qgroup */
1696 if (!list_empty(&qgroup->members)) {
1697 ret = -EBUSY;
1698 goto out;
1699 }
1700
1701 ret = del_qgroup_item(trans, qgroupid);
1702 if (ret && ret != -ENOENT)
1703 goto out;
1704
1705 while (!list_empty(&qgroup->groups)) {
1706 list = list_first_entry(&qgroup->groups,
1707 struct btrfs_qgroup_list, next_group);
1708 ret = __del_qgroup_relation(trans, qgroupid,
1709 list->group->qgroupid);
1710 if (ret)
1711 goto out;
1712 }
1713
1714 spin_lock(&fs_info->qgroup_lock);
1715 del_qgroup_rb(fs_info, qgroupid);
1716 spin_unlock(&fs_info->qgroup_lock);
1717
1718 /*
1719 * Remove the qgroup from sysfs now without holding the qgroup_lock
1720 * spinlock, since the sysfs_remove_group() function needs to take
1721 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1722 */
1723 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1724 kfree(qgroup);
1725 out:
1726 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1727 return ret;
1728 }
1729
btrfs_limit_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid,struct btrfs_qgroup_limit * limit)1730 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1731 struct btrfs_qgroup_limit *limit)
1732 {
1733 struct btrfs_fs_info *fs_info = trans->fs_info;
1734 struct btrfs_qgroup *qgroup;
1735 int ret = 0;
1736 /* Sometimes we would want to clear the limit on this qgroup.
1737 * To meet this requirement, we treat the -1 as a special value
1738 * which tell kernel to clear the limit on this qgroup.
1739 */
1740 const u64 CLEAR_VALUE = -1;
1741
1742 mutex_lock(&fs_info->qgroup_ioctl_lock);
1743 if (!fs_info->quota_root) {
1744 ret = -ENOTCONN;
1745 goto out;
1746 }
1747
1748 qgroup = find_qgroup_rb(fs_info, qgroupid);
1749 if (!qgroup) {
1750 ret = -ENOENT;
1751 goto out;
1752 }
1753
1754 spin_lock(&fs_info->qgroup_lock);
1755 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1756 if (limit->max_rfer == CLEAR_VALUE) {
1757 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1758 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1759 qgroup->max_rfer = 0;
1760 } else {
1761 qgroup->max_rfer = limit->max_rfer;
1762 }
1763 }
1764 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1765 if (limit->max_excl == CLEAR_VALUE) {
1766 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1767 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1768 qgroup->max_excl = 0;
1769 } else {
1770 qgroup->max_excl = limit->max_excl;
1771 }
1772 }
1773 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1774 if (limit->rsv_rfer == CLEAR_VALUE) {
1775 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1776 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1777 qgroup->rsv_rfer = 0;
1778 } else {
1779 qgroup->rsv_rfer = limit->rsv_rfer;
1780 }
1781 }
1782 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1783 if (limit->rsv_excl == CLEAR_VALUE) {
1784 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1785 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1786 qgroup->rsv_excl = 0;
1787 } else {
1788 qgroup->rsv_excl = limit->rsv_excl;
1789 }
1790 }
1791 qgroup->lim_flags |= limit->flags;
1792
1793 spin_unlock(&fs_info->qgroup_lock);
1794
1795 ret = update_qgroup_limit_item(trans, qgroup);
1796 if (ret) {
1797 qgroup_mark_inconsistent(fs_info);
1798 btrfs_info(fs_info, "unable to update quota limit for %llu",
1799 qgroupid);
1800 }
1801
1802 out:
1803 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1804 return ret;
1805 }
1806
btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info * fs_info,struct btrfs_delayed_ref_root * delayed_refs,struct btrfs_qgroup_extent_record * record)1807 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1808 struct btrfs_delayed_ref_root *delayed_refs,
1809 struct btrfs_qgroup_extent_record *record)
1810 {
1811 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1812 struct rb_node *parent_node = NULL;
1813 struct btrfs_qgroup_extent_record *entry;
1814 u64 bytenr = record->bytenr;
1815
1816 lockdep_assert_held(&delayed_refs->lock);
1817 trace_btrfs_qgroup_trace_extent(fs_info, record);
1818
1819 while (*p) {
1820 parent_node = *p;
1821 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1822 node);
1823 if (bytenr < entry->bytenr) {
1824 p = &(*p)->rb_left;
1825 } else if (bytenr > entry->bytenr) {
1826 p = &(*p)->rb_right;
1827 } else {
1828 if (record->data_rsv && !entry->data_rsv) {
1829 entry->data_rsv = record->data_rsv;
1830 entry->data_rsv_refroot =
1831 record->data_rsv_refroot;
1832 }
1833 return 1;
1834 }
1835 }
1836
1837 rb_link_node(&record->node, parent_node, p);
1838 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1839 return 0;
1840 }
1841
btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle * trans,struct btrfs_qgroup_extent_record * qrecord)1842 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1843 struct btrfs_qgroup_extent_record *qrecord)
1844 {
1845 struct btrfs_backref_walk_ctx ctx = { 0 };
1846 int ret;
1847
1848 /*
1849 * We are always called in a context where we are already holding a
1850 * transaction handle. Often we are called when adding a data delayed
1851 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1852 * in which case we will be holding a write lock on extent buffer from a
1853 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1854 * acquire fs_info->commit_root_sem, because that is a higher level lock
1855 * that must be acquired before locking any extent buffers.
1856 *
1857 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1858 * but we can't pass it a non-NULL transaction handle, because otherwise
1859 * it would not use commit roots and would lock extent buffers, causing
1860 * a deadlock if it ends up trying to read lock the same extent buffer
1861 * that was previously write locked at btrfs_truncate_inode_items().
1862 *
1863 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1864 * explicitly tell it to not acquire the commit_root_sem - if we are
1865 * holding a transaction handle we don't need its protection.
1866 */
1867 ASSERT(trans != NULL);
1868
1869 if (trans->fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
1870 return 0;
1871
1872 ctx.bytenr = qrecord->bytenr;
1873 ctx.fs_info = trans->fs_info;
1874
1875 ret = btrfs_find_all_roots(&ctx, true);
1876 if (ret < 0) {
1877 qgroup_mark_inconsistent(trans->fs_info);
1878 btrfs_warn(trans->fs_info,
1879 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1880 ret);
1881 return 0;
1882 }
1883
1884 /*
1885 * Here we don't need to get the lock of
1886 * trans->transaction->delayed_refs, since inserted qrecord won't
1887 * be deleted, only qrecord->node may be modified (new qrecord insert)
1888 *
1889 * So modifying qrecord->old_roots is safe here
1890 */
1891 qrecord->old_roots = ctx.roots;
1892 return 0;
1893 }
1894
btrfs_qgroup_trace_extent(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes)1895 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1896 u64 num_bytes)
1897 {
1898 struct btrfs_fs_info *fs_info = trans->fs_info;
1899 struct btrfs_qgroup_extent_record *record;
1900 struct btrfs_delayed_ref_root *delayed_refs;
1901 int ret;
1902
1903 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1904 || bytenr == 0 || num_bytes == 0)
1905 return 0;
1906 record = kzalloc(sizeof(*record), GFP_NOFS);
1907 if (!record)
1908 return -ENOMEM;
1909
1910 delayed_refs = &trans->transaction->delayed_refs;
1911 record->bytenr = bytenr;
1912 record->num_bytes = num_bytes;
1913 record->old_roots = NULL;
1914
1915 spin_lock(&delayed_refs->lock);
1916 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1917 spin_unlock(&delayed_refs->lock);
1918 if (ret > 0) {
1919 kfree(record);
1920 return 0;
1921 }
1922 return btrfs_qgroup_trace_extent_post(trans, record);
1923 }
1924
btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle * trans,struct extent_buffer * eb)1925 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1926 struct extent_buffer *eb)
1927 {
1928 struct btrfs_fs_info *fs_info = trans->fs_info;
1929 int nr = btrfs_header_nritems(eb);
1930 int i, extent_type, ret;
1931 struct btrfs_key key;
1932 struct btrfs_file_extent_item *fi;
1933 u64 bytenr, num_bytes;
1934
1935 /* We can be called directly from walk_up_proc() */
1936 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1937 return 0;
1938
1939 for (i = 0; i < nr; i++) {
1940 btrfs_item_key_to_cpu(eb, &key, i);
1941
1942 if (key.type != BTRFS_EXTENT_DATA_KEY)
1943 continue;
1944
1945 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1946 /* filter out non qgroup-accountable extents */
1947 extent_type = btrfs_file_extent_type(eb, fi);
1948
1949 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1950 continue;
1951
1952 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1953 if (!bytenr)
1954 continue;
1955
1956 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1957
1958 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes);
1959 if (ret)
1960 return ret;
1961 }
1962 cond_resched();
1963 return 0;
1964 }
1965
1966 /*
1967 * Walk up the tree from the bottom, freeing leaves and any interior
1968 * nodes which have had all slots visited. If a node (leaf or
1969 * interior) is freed, the node above it will have it's slot
1970 * incremented. The root node will never be freed.
1971 *
1972 * At the end of this function, we should have a path which has all
1973 * slots incremented to the next position for a search. If we need to
1974 * read a new node it will be NULL and the node above it will have the
1975 * correct slot selected for a later read.
1976 *
1977 * If we increment the root nodes slot counter past the number of
1978 * elements, 1 is returned to signal completion of the search.
1979 */
adjust_slots_upwards(struct btrfs_path * path,int root_level)1980 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1981 {
1982 int level = 0;
1983 int nr, slot;
1984 struct extent_buffer *eb;
1985
1986 if (root_level == 0)
1987 return 1;
1988
1989 while (level <= root_level) {
1990 eb = path->nodes[level];
1991 nr = btrfs_header_nritems(eb);
1992 path->slots[level]++;
1993 slot = path->slots[level];
1994 if (slot >= nr || level == 0) {
1995 /*
1996 * Don't free the root - we will detect this
1997 * condition after our loop and return a
1998 * positive value for caller to stop walking the tree.
1999 */
2000 if (level != root_level) {
2001 btrfs_tree_unlock_rw(eb, path->locks[level]);
2002 path->locks[level] = 0;
2003
2004 free_extent_buffer(eb);
2005 path->nodes[level] = NULL;
2006 path->slots[level] = 0;
2007 }
2008 } else {
2009 /*
2010 * We have a valid slot to walk back down
2011 * from. Stop here so caller can process these
2012 * new nodes.
2013 */
2014 break;
2015 }
2016
2017 level++;
2018 }
2019
2020 eb = path->nodes[root_level];
2021 if (path->slots[root_level] >= btrfs_header_nritems(eb))
2022 return 1;
2023
2024 return 0;
2025 }
2026
2027 /*
2028 * Helper function to trace a subtree tree block swap.
2029 *
2030 * The swap will happen in highest tree block, but there may be a lot of
2031 * tree blocks involved.
2032 *
2033 * For example:
2034 * OO = Old tree blocks
2035 * NN = New tree blocks allocated during balance
2036 *
2037 * File tree (257) Reloc tree for 257
2038 * L2 OO NN
2039 * / \ / \
2040 * L1 OO OO (a) OO NN (a)
2041 * / \ / \ / \ / \
2042 * L0 OO OO OO OO OO OO NN NN
2043 * (b) (c) (b) (c)
2044 *
2045 * When calling qgroup_trace_extent_swap(), we will pass:
2046 * @src_eb = OO(a)
2047 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
2048 * @dst_level = 0
2049 * @root_level = 1
2050 *
2051 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
2052 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
2053 *
2054 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
2055 *
2056 * 1) Tree search from @src_eb
2057 * It should acts as a simplified btrfs_search_slot().
2058 * The key for search can be extracted from @dst_path->nodes[dst_level]
2059 * (first key).
2060 *
2061 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
2062 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
2063 * They should be marked during previous (@dst_level = 1) iteration.
2064 *
2065 * 3) Mark file extents in leaves dirty
2066 * We don't have good way to pick out new file extents only.
2067 * So we still follow the old method by scanning all file extents in
2068 * the leave.
2069 *
2070 * This function can free us from keeping two paths, thus later we only need
2071 * to care about how to iterate all new tree blocks in reloc tree.
2072 */
qgroup_trace_extent_swap(struct btrfs_trans_handle * trans,struct extent_buffer * src_eb,struct btrfs_path * dst_path,int dst_level,int root_level,bool trace_leaf)2073 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
2074 struct extent_buffer *src_eb,
2075 struct btrfs_path *dst_path,
2076 int dst_level, int root_level,
2077 bool trace_leaf)
2078 {
2079 struct btrfs_key key;
2080 struct btrfs_path *src_path;
2081 struct btrfs_fs_info *fs_info = trans->fs_info;
2082 u32 nodesize = fs_info->nodesize;
2083 int cur_level = root_level;
2084 int ret;
2085
2086 BUG_ON(dst_level > root_level);
2087 /* Level mismatch */
2088 if (btrfs_header_level(src_eb) != root_level)
2089 return -EINVAL;
2090
2091 src_path = btrfs_alloc_path();
2092 if (!src_path) {
2093 ret = -ENOMEM;
2094 goto out;
2095 }
2096
2097 if (dst_level)
2098 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2099 else
2100 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2101
2102 /* For src_path */
2103 atomic_inc(&src_eb->refs);
2104 src_path->nodes[root_level] = src_eb;
2105 src_path->slots[root_level] = dst_path->slots[root_level];
2106 src_path->locks[root_level] = 0;
2107
2108 /* A simplified version of btrfs_search_slot() */
2109 while (cur_level >= dst_level) {
2110 struct btrfs_key src_key;
2111 struct btrfs_key dst_key;
2112
2113 if (src_path->nodes[cur_level] == NULL) {
2114 struct extent_buffer *eb;
2115 int parent_slot;
2116
2117 eb = src_path->nodes[cur_level + 1];
2118 parent_slot = src_path->slots[cur_level + 1];
2119
2120 eb = btrfs_read_node_slot(eb, parent_slot);
2121 if (IS_ERR(eb)) {
2122 ret = PTR_ERR(eb);
2123 goto out;
2124 }
2125
2126 src_path->nodes[cur_level] = eb;
2127
2128 btrfs_tree_read_lock(eb);
2129 src_path->locks[cur_level] = BTRFS_READ_LOCK;
2130 }
2131
2132 src_path->slots[cur_level] = dst_path->slots[cur_level];
2133 if (cur_level) {
2134 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2135 &dst_key, dst_path->slots[cur_level]);
2136 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2137 &src_key, src_path->slots[cur_level]);
2138 } else {
2139 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2140 &dst_key, dst_path->slots[cur_level]);
2141 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2142 &src_key, src_path->slots[cur_level]);
2143 }
2144 /* Content mismatch, something went wrong */
2145 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2146 ret = -ENOENT;
2147 goto out;
2148 }
2149 cur_level--;
2150 }
2151
2152 /*
2153 * Now both @dst_path and @src_path have been populated, record the tree
2154 * blocks for qgroup accounting.
2155 */
2156 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2157 nodesize);
2158 if (ret < 0)
2159 goto out;
2160 ret = btrfs_qgroup_trace_extent(trans, dst_path->nodes[dst_level]->start,
2161 nodesize);
2162 if (ret < 0)
2163 goto out;
2164
2165 /* Record leaf file extents */
2166 if (dst_level == 0 && trace_leaf) {
2167 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2168 if (ret < 0)
2169 goto out;
2170 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2171 }
2172 out:
2173 btrfs_free_path(src_path);
2174 return ret;
2175 }
2176
2177 /*
2178 * Helper function to do recursive generation-aware depth-first search, to
2179 * locate all new tree blocks in a subtree of reloc tree.
2180 *
2181 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2182 * reloc tree
2183 * L2 NN (a)
2184 * / \
2185 * L1 OO NN (b)
2186 * / \ / \
2187 * L0 OO OO OO NN
2188 * (c) (d)
2189 * If we pass:
2190 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2191 * @cur_level = 1
2192 * @root_level = 1
2193 *
2194 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2195 * above tree blocks along with their counter parts in file tree.
2196 * While during search, old tree blocks OO(c) will be skipped as tree block swap
2197 * won't affect OO(c).
2198 */
qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle * trans,struct extent_buffer * src_eb,struct btrfs_path * dst_path,int cur_level,int root_level,u64 last_snapshot,bool trace_leaf)2199 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2200 struct extent_buffer *src_eb,
2201 struct btrfs_path *dst_path,
2202 int cur_level, int root_level,
2203 u64 last_snapshot, bool trace_leaf)
2204 {
2205 struct btrfs_fs_info *fs_info = trans->fs_info;
2206 struct extent_buffer *eb;
2207 bool need_cleanup = false;
2208 int ret = 0;
2209 int i;
2210
2211 /* Level sanity check */
2212 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2213 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2214 root_level < cur_level) {
2215 btrfs_err_rl(fs_info,
2216 "%s: bad levels, cur_level=%d root_level=%d",
2217 __func__, cur_level, root_level);
2218 return -EUCLEAN;
2219 }
2220
2221 /* Read the tree block if needed */
2222 if (dst_path->nodes[cur_level] == NULL) {
2223 int parent_slot;
2224 u64 child_gen;
2225
2226 /*
2227 * dst_path->nodes[root_level] must be initialized before
2228 * calling this function.
2229 */
2230 if (cur_level == root_level) {
2231 btrfs_err_rl(fs_info,
2232 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2233 __func__, root_level, root_level, cur_level);
2234 return -EUCLEAN;
2235 }
2236
2237 /*
2238 * We need to get child blockptr/gen from parent before we can
2239 * read it.
2240 */
2241 eb = dst_path->nodes[cur_level + 1];
2242 parent_slot = dst_path->slots[cur_level + 1];
2243 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2244
2245 /* This node is old, no need to trace */
2246 if (child_gen < last_snapshot)
2247 goto out;
2248
2249 eb = btrfs_read_node_slot(eb, parent_slot);
2250 if (IS_ERR(eb)) {
2251 ret = PTR_ERR(eb);
2252 goto out;
2253 }
2254
2255 dst_path->nodes[cur_level] = eb;
2256 dst_path->slots[cur_level] = 0;
2257
2258 btrfs_tree_read_lock(eb);
2259 dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2260 need_cleanup = true;
2261 }
2262
2263 /* Now record this tree block and its counter part for qgroups */
2264 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2265 root_level, trace_leaf);
2266 if (ret < 0)
2267 goto cleanup;
2268
2269 eb = dst_path->nodes[cur_level];
2270
2271 if (cur_level > 0) {
2272 /* Iterate all child tree blocks */
2273 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2274 /* Skip old tree blocks as they won't be swapped */
2275 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2276 continue;
2277 dst_path->slots[cur_level] = i;
2278
2279 /* Recursive call (at most 7 times) */
2280 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2281 dst_path, cur_level - 1, root_level,
2282 last_snapshot, trace_leaf);
2283 if (ret < 0)
2284 goto cleanup;
2285 }
2286 }
2287
2288 cleanup:
2289 if (need_cleanup) {
2290 /* Clean up */
2291 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2292 dst_path->locks[cur_level]);
2293 free_extent_buffer(dst_path->nodes[cur_level]);
2294 dst_path->nodes[cur_level] = NULL;
2295 dst_path->slots[cur_level] = 0;
2296 dst_path->locks[cur_level] = 0;
2297 }
2298 out:
2299 return ret;
2300 }
2301
qgroup_trace_subtree_swap(struct btrfs_trans_handle * trans,struct extent_buffer * src_eb,struct extent_buffer * dst_eb,u64 last_snapshot,bool trace_leaf)2302 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2303 struct extent_buffer *src_eb,
2304 struct extent_buffer *dst_eb,
2305 u64 last_snapshot, bool trace_leaf)
2306 {
2307 struct btrfs_fs_info *fs_info = trans->fs_info;
2308 struct btrfs_path *dst_path = NULL;
2309 int level;
2310 int ret;
2311
2312 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2313 return 0;
2314
2315 /* Wrong parameter order */
2316 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2317 btrfs_err_rl(fs_info,
2318 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2319 btrfs_header_generation(src_eb),
2320 btrfs_header_generation(dst_eb));
2321 return -EUCLEAN;
2322 }
2323
2324 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2325 ret = -EIO;
2326 goto out;
2327 }
2328
2329 level = btrfs_header_level(dst_eb);
2330 dst_path = btrfs_alloc_path();
2331 if (!dst_path) {
2332 ret = -ENOMEM;
2333 goto out;
2334 }
2335 /* For dst_path */
2336 atomic_inc(&dst_eb->refs);
2337 dst_path->nodes[level] = dst_eb;
2338 dst_path->slots[level] = 0;
2339 dst_path->locks[level] = 0;
2340
2341 /* Do the generation aware breadth-first search */
2342 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2343 level, last_snapshot, trace_leaf);
2344 if (ret < 0)
2345 goto out;
2346 ret = 0;
2347
2348 out:
2349 btrfs_free_path(dst_path);
2350 if (ret < 0)
2351 qgroup_mark_inconsistent(fs_info);
2352 return ret;
2353 }
2354
btrfs_qgroup_trace_subtree(struct btrfs_trans_handle * trans,struct extent_buffer * root_eb,u64 root_gen,int root_level)2355 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2356 struct extent_buffer *root_eb,
2357 u64 root_gen, int root_level)
2358 {
2359 struct btrfs_fs_info *fs_info = trans->fs_info;
2360 int ret = 0;
2361 int level;
2362 u8 drop_subptree_thres;
2363 struct extent_buffer *eb = root_eb;
2364 struct btrfs_path *path = NULL;
2365
2366 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2367 BUG_ON(root_eb == NULL);
2368
2369 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2370 return 0;
2371
2372 spin_lock(&fs_info->qgroup_lock);
2373 drop_subptree_thres = fs_info->qgroup_drop_subtree_thres;
2374 spin_unlock(&fs_info->qgroup_lock);
2375
2376 /*
2377 * This function only gets called for snapshot drop, if we hit a high
2378 * node here, it means we are going to change ownership for quite a lot
2379 * of extents, which will greatly slow down btrfs_commit_transaction().
2380 *
2381 * So here if we find a high tree here, we just skip the accounting and
2382 * mark qgroup inconsistent.
2383 */
2384 if (root_level >= drop_subptree_thres) {
2385 qgroup_mark_inconsistent(fs_info);
2386 return 0;
2387 }
2388
2389 if (!extent_buffer_uptodate(root_eb)) {
2390 struct btrfs_tree_parent_check check = {
2391 .has_first_key = false,
2392 .transid = root_gen,
2393 .level = root_level
2394 };
2395
2396 ret = btrfs_read_extent_buffer(root_eb, &check);
2397 if (ret)
2398 goto out;
2399 }
2400
2401 if (root_level == 0) {
2402 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2403 goto out;
2404 }
2405
2406 path = btrfs_alloc_path();
2407 if (!path)
2408 return -ENOMEM;
2409
2410 /*
2411 * Walk down the tree. Missing extent blocks are filled in as
2412 * we go. Metadata is accounted every time we read a new
2413 * extent block.
2414 *
2415 * When we reach a leaf, we account for file extent items in it,
2416 * walk back up the tree (adjusting slot pointers as we go)
2417 * and restart the search process.
2418 */
2419 atomic_inc(&root_eb->refs); /* For path */
2420 path->nodes[root_level] = root_eb;
2421 path->slots[root_level] = 0;
2422 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2423 walk_down:
2424 level = root_level;
2425 while (level >= 0) {
2426 if (path->nodes[level] == NULL) {
2427 int parent_slot;
2428 u64 child_bytenr;
2429
2430 /*
2431 * We need to get child blockptr from parent before we
2432 * can read it.
2433 */
2434 eb = path->nodes[level + 1];
2435 parent_slot = path->slots[level + 1];
2436 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2437
2438 eb = btrfs_read_node_slot(eb, parent_slot);
2439 if (IS_ERR(eb)) {
2440 ret = PTR_ERR(eb);
2441 goto out;
2442 }
2443
2444 path->nodes[level] = eb;
2445 path->slots[level] = 0;
2446
2447 btrfs_tree_read_lock(eb);
2448 path->locks[level] = BTRFS_READ_LOCK;
2449
2450 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2451 fs_info->nodesize);
2452 if (ret)
2453 goto out;
2454 }
2455
2456 if (level == 0) {
2457 ret = btrfs_qgroup_trace_leaf_items(trans,
2458 path->nodes[level]);
2459 if (ret)
2460 goto out;
2461
2462 /* Nonzero return here means we completed our search */
2463 ret = adjust_slots_upwards(path, root_level);
2464 if (ret)
2465 break;
2466
2467 /* Restart search with new slots */
2468 goto walk_down;
2469 }
2470
2471 level--;
2472 }
2473
2474 ret = 0;
2475 out:
2476 btrfs_free_path(path);
2477
2478 return ret;
2479 }
2480
2481 #define UPDATE_NEW 0
2482 #define UPDATE_OLD 1
2483 /*
2484 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2485 */
qgroup_update_refcnt(struct btrfs_fs_info * fs_info,struct ulist * roots,struct ulist * tmp,struct ulist * qgroups,u64 seq,int update_old)2486 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2487 struct ulist *roots, struct ulist *tmp,
2488 struct ulist *qgroups, u64 seq, int update_old)
2489 {
2490 struct ulist_node *unode;
2491 struct ulist_iterator uiter;
2492 struct ulist_node *tmp_unode;
2493 struct ulist_iterator tmp_uiter;
2494 struct btrfs_qgroup *qg;
2495 int ret = 0;
2496
2497 if (!roots)
2498 return 0;
2499 ULIST_ITER_INIT(&uiter);
2500 while ((unode = ulist_next(roots, &uiter))) {
2501 qg = find_qgroup_rb(fs_info, unode->val);
2502 if (!qg)
2503 continue;
2504
2505 ulist_reinit(tmp);
2506 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2507 GFP_ATOMIC);
2508 if (ret < 0)
2509 return ret;
2510 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2511 if (ret < 0)
2512 return ret;
2513 ULIST_ITER_INIT(&tmp_uiter);
2514 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2515 struct btrfs_qgroup_list *glist;
2516
2517 qg = unode_aux_to_qgroup(tmp_unode);
2518 if (update_old)
2519 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2520 else
2521 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2522 list_for_each_entry(glist, &qg->groups, next_group) {
2523 ret = ulist_add(qgroups, glist->group->qgroupid,
2524 qgroup_to_aux(glist->group),
2525 GFP_ATOMIC);
2526 if (ret < 0)
2527 return ret;
2528 ret = ulist_add(tmp, glist->group->qgroupid,
2529 qgroup_to_aux(glist->group),
2530 GFP_ATOMIC);
2531 if (ret < 0)
2532 return ret;
2533 }
2534 }
2535 }
2536 return 0;
2537 }
2538
2539 /*
2540 * Update qgroup rfer/excl counters.
2541 * Rfer update is easy, codes can explain themselves.
2542 *
2543 * Excl update is tricky, the update is split into 2 parts.
2544 * Part 1: Possible exclusive <-> sharing detect:
2545 * | A | !A |
2546 * -------------------------------------
2547 * B | * | - |
2548 * -------------------------------------
2549 * !B | + | ** |
2550 * -------------------------------------
2551 *
2552 * Conditions:
2553 * A: cur_old_roots < nr_old_roots (not exclusive before)
2554 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2555 * B: cur_new_roots < nr_new_roots (not exclusive now)
2556 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2557 *
2558 * Results:
2559 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2560 * *: Definitely not changed. **: Possible unchanged.
2561 *
2562 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2563 *
2564 * To make the logic clear, we first use condition A and B to split
2565 * combination into 4 results.
2566 *
2567 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2568 * only on variant maybe 0.
2569 *
2570 * Lastly, check result **, since there are 2 variants maybe 0, split them
2571 * again(2x2).
2572 * But this time we don't need to consider other things, the codes and logic
2573 * is easy to understand now.
2574 */
qgroup_update_counters(struct btrfs_fs_info * fs_info,struct ulist * qgroups,u64 nr_old_roots,u64 nr_new_roots,u64 num_bytes,u64 seq)2575 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2576 struct ulist *qgroups,
2577 u64 nr_old_roots,
2578 u64 nr_new_roots,
2579 u64 num_bytes, u64 seq)
2580 {
2581 struct ulist_node *unode;
2582 struct ulist_iterator uiter;
2583 struct btrfs_qgroup *qg;
2584 u64 cur_new_count, cur_old_count;
2585
2586 ULIST_ITER_INIT(&uiter);
2587 while ((unode = ulist_next(qgroups, &uiter))) {
2588 bool dirty = false;
2589
2590 qg = unode_aux_to_qgroup(unode);
2591 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2592 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2593
2594 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2595 cur_new_count);
2596
2597 /* Rfer update part */
2598 if (cur_old_count == 0 && cur_new_count > 0) {
2599 qg->rfer += num_bytes;
2600 qg->rfer_cmpr += num_bytes;
2601 dirty = true;
2602 }
2603 if (cur_old_count > 0 && cur_new_count == 0) {
2604 qg->rfer -= num_bytes;
2605 qg->rfer_cmpr -= num_bytes;
2606 dirty = true;
2607 }
2608
2609 /* Excl update part */
2610 /* Exclusive/none -> shared case */
2611 if (cur_old_count == nr_old_roots &&
2612 cur_new_count < nr_new_roots) {
2613 /* Exclusive -> shared */
2614 if (cur_old_count != 0) {
2615 qg->excl -= num_bytes;
2616 qg->excl_cmpr -= num_bytes;
2617 dirty = true;
2618 }
2619 }
2620
2621 /* Shared -> exclusive/none case */
2622 if (cur_old_count < nr_old_roots &&
2623 cur_new_count == nr_new_roots) {
2624 /* Shared->exclusive */
2625 if (cur_new_count != 0) {
2626 qg->excl += num_bytes;
2627 qg->excl_cmpr += num_bytes;
2628 dirty = true;
2629 }
2630 }
2631
2632 /* Exclusive/none -> exclusive/none case */
2633 if (cur_old_count == nr_old_roots &&
2634 cur_new_count == nr_new_roots) {
2635 if (cur_old_count == 0) {
2636 /* None -> exclusive/none */
2637
2638 if (cur_new_count != 0) {
2639 /* None -> exclusive */
2640 qg->excl += num_bytes;
2641 qg->excl_cmpr += num_bytes;
2642 dirty = true;
2643 }
2644 /* None -> none, nothing changed */
2645 } else {
2646 /* Exclusive -> exclusive/none */
2647
2648 if (cur_new_count == 0) {
2649 /* Exclusive -> none */
2650 qg->excl -= num_bytes;
2651 qg->excl_cmpr -= num_bytes;
2652 dirty = true;
2653 }
2654 /* Exclusive -> exclusive, nothing changed */
2655 }
2656 }
2657
2658 if (dirty)
2659 qgroup_dirty(fs_info, qg);
2660 }
2661 return 0;
2662 }
2663
2664 /*
2665 * Check if the @roots potentially is a list of fs tree roots
2666 *
2667 * Return 0 for definitely not a fs/subvol tree roots ulist
2668 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2669 * one as well)
2670 */
maybe_fs_roots(struct ulist * roots)2671 static int maybe_fs_roots(struct ulist *roots)
2672 {
2673 struct ulist_node *unode;
2674 struct ulist_iterator uiter;
2675
2676 /* Empty one, still possible for fs roots */
2677 if (!roots || roots->nnodes == 0)
2678 return 1;
2679
2680 ULIST_ITER_INIT(&uiter);
2681 unode = ulist_next(roots, &uiter);
2682 if (!unode)
2683 return 1;
2684
2685 /*
2686 * If it contains fs tree roots, then it must belong to fs/subvol
2687 * trees.
2688 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2689 */
2690 return is_fstree(unode->val);
2691 }
2692
btrfs_qgroup_account_extent(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes,struct ulist * old_roots,struct ulist * new_roots)2693 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2694 u64 num_bytes, struct ulist *old_roots,
2695 struct ulist *new_roots)
2696 {
2697 struct btrfs_fs_info *fs_info = trans->fs_info;
2698 struct ulist *qgroups = NULL;
2699 struct ulist *tmp = NULL;
2700 u64 seq;
2701 u64 nr_new_roots = 0;
2702 u64 nr_old_roots = 0;
2703 int ret = 0;
2704
2705 /*
2706 * If quotas get disabled meanwhile, the resources need to be freed and
2707 * we can't just exit here.
2708 */
2709 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2710 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
2711 goto out_free;
2712
2713 if (new_roots) {
2714 if (!maybe_fs_roots(new_roots))
2715 goto out_free;
2716 nr_new_roots = new_roots->nnodes;
2717 }
2718 if (old_roots) {
2719 if (!maybe_fs_roots(old_roots))
2720 goto out_free;
2721 nr_old_roots = old_roots->nnodes;
2722 }
2723
2724 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2725 if (nr_old_roots == 0 && nr_new_roots == 0)
2726 goto out_free;
2727
2728 BUG_ON(!fs_info->quota_root);
2729
2730 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2731 num_bytes, nr_old_roots, nr_new_roots);
2732
2733 qgroups = ulist_alloc(GFP_NOFS);
2734 if (!qgroups) {
2735 ret = -ENOMEM;
2736 goto out_free;
2737 }
2738 tmp = ulist_alloc(GFP_NOFS);
2739 if (!tmp) {
2740 ret = -ENOMEM;
2741 goto out_free;
2742 }
2743
2744 mutex_lock(&fs_info->qgroup_rescan_lock);
2745 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2746 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2747 mutex_unlock(&fs_info->qgroup_rescan_lock);
2748 ret = 0;
2749 goto out_free;
2750 }
2751 }
2752 mutex_unlock(&fs_info->qgroup_rescan_lock);
2753
2754 spin_lock(&fs_info->qgroup_lock);
2755 seq = fs_info->qgroup_seq;
2756
2757 /* Update old refcnts using old_roots */
2758 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2759 UPDATE_OLD);
2760 if (ret < 0)
2761 goto out;
2762
2763 /* Update new refcnts using new_roots */
2764 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2765 UPDATE_NEW);
2766 if (ret < 0)
2767 goto out;
2768
2769 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2770 num_bytes, seq);
2771
2772 /*
2773 * Bump qgroup_seq to avoid seq overlap
2774 */
2775 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2776 out:
2777 spin_unlock(&fs_info->qgroup_lock);
2778 out_free:
2779 ulist_free(tmp);
2780 ulist_free(qgroups);
2781 ulist_free(old_roots);
2782 ulist_free(new_roots);
2783 return ret;
2784 }
2785
btrfs_qgroup_account_extents(struct btrfs_trans_handle * trans)2786 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2787 {
2788 struct btrfs_fs_info *fs_info = trans->fs_info;
2789 struct btrfs_qgroup_extent_record *record;
2790 struct btrfs_delayed_ref_root *delayed_refs;
2791 struct ulist *new_roots = NULL;
2792 struct rb_node *node;
2793 u64 num_dirty_extents = 0;
2794 u64 qgroup_to_skip;
2795 int ret = 0;
2796
2797 delayed_refs = &trans->transaction->delayed_refs;
2798 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2799 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2800 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2801 node);
2802
2803 num_dirty_extents++;
2804 trace_btrfs_qgroup_account_extents(fs_info, record);
2805
2806 if (!ret && !(fs_info->qgroup_flags &
2807 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)) {
2808 struct btrfs_backref_walk_ctx ctx = { 0 };
2809
2810 ctx.bytenr = record->bytenr;
2811 ctx.fs_info = fs_info;
2812
2813 /*
2814 * Old roots should be searched when inserting qgroup
2815 * extent record.
2816 *
2817 * But for INCONSISTENT (NO_ACCOUNTING) -> rescan case,
2818 * we may have some record inserted during
2819 * NO_ACCOUNTING (thus no old_roots populated), but
2820 * later we start rescan, which clears NO_ACCOUNTING,
2821 * leaving some inserted records without old_roots
2822 * populated.
2823 *
2824 * Those cases are rare and should not cause too much
2825 * time spent during commit_transaction().
2826 */
2827 if (!record->old_roots) {
2828 /* Search commit root to find old_roots */
2829 ret = btrfs_find_all_roots(&ctx, false);
2830 if (ret < 0)
2831 goto cleanup;
2832 record->old_roots = ctx.roots;
2833 ctx.roots = NULL;
2834 }
2835
2836 /* Free the reserved data space */
2837 btrfs_qgroup_free_refroot(fs_info,
2838 record->data_rsv_refroot,
2839 record->data_rsv,
2840 BTRFS_QGROUP_RSV_DATA);
2841 /*
2842 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2843 * which doesn't lock tree or delayed_refs and search
2844 * current root. It's safe inside commit_transaction().
2845 */
2846 ctx.trans = trans;
2847 ctx.time_seq = BTRFS_SEQ_LAST;
2848 ret = btrfs_find_all_roots(&ctx, false);
2849 if (ret < 0)
2850 goto cleanup;
2851 new_roots = ctx.roots;
2852 if (qgroup_to_skip) {
2853 ulist_del(new_roots, qgroup_to_skip, 0);
2854 ulist_del(record->old_roots, qgroup_to_skip,
2855 0);
2856 }
2857 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2858 record->num_bytes,
2859 record->old_roots,
2860 new_roots);
2861 record->old_roots = NULL;
2862 new_roots = NULL;
2863 }
2864 cleanup:
2865 ulist_free(record->old_roots);
2866 ulist_free(new_roots);
2867 new_roots = NULL;
2868 rb_erase(node, &delayed_refs->dirty_extent_root);
2869 kfree(record);
2870
2871 }
2872 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2873 num_dirty_extents);
2874 return ret;
2875 }
2876
2877 /*
2878 * Writes all changed qgroups to disk.
2879 * Called by the transaction commit path and the qgroup assign ioctl.
2880 */
btrfs_run_qgroups(struct btrfs_trans_handle * trans)2881 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2882 {
2883 struct btrfs_fs_info *fs_info = trans->fs_info;
2884 int ret = 0;
2885
2886 /*
2887 * In case we are called from the qgroup assign ioctl, assert that we
2888 * are holding the qgroup_ioctl_lock, otherwise we can race with a quota
2889 * disable operation (ioctl) and access a freed quota root.
2890 */
2891 if (trans->transaction->state != TRANS_STATE_COMMIT_DOING)
2892 lockdep_assert_held(&fs_info->qgroup_ioctl_lock);
2893
2894 if (!fs_info->quota_root)
2895 return ret;
2896
2897 spin_lock(&fs_info->qgroup_lock);
2898 while (!list_empty(&fs_info->dirty_qgroups)) {
2899 struct btrfs_qgroup *qgroup;
2900 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2901 struct btrfs_qgroup, dirty);
2902 list_del_init(&qgroup->dirty);
2903 spin_unlock(&fs_info->qgroup_lock);
2904 ret = update_qgroup_info_item(trans, qgroup);
2905 if (ret)
2906 qgroup_mark_inconsistent(fs_info);
2907 ret = update_qgroup_limit_item(trans, qgroup);
2908 if (ret)
2909 qgroup_mark_inconsistent(fs_info);
2910 spin_lock(&fs_info->qgroup_lock);
2911 }
2912 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2913 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2914 else
2915 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2916 spin_unlock(&fs_info->qgroup_lock);
2917
2918 ret = update_qgroup_status_item(trans);
2919 if (ret)
2920 qgroup_mark_inconsistent(fs_info);
2921
2922 return ret;
2923 }
2924
2925 /*
2926 * Copy the accounting information between qgroups. This is necessary
2927 * when a snapshot or a subvolume is created. Throwing an error will
2928 * cause a transaction abort so we take extra care here to only error
2929 * when a readonly fs is a reasonable outcome.
2930 */
btrfs_qgroup_inherit(struct btrfs_trans_handle * trans,u64 srcid,u64 objectid,struct btrfs_qgroup_inherit * inherit)2931 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2932 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2933 {
2934 int ret = 0;
2935 int i;
2936 u64 *i_qgroups;
2937 bool committing = false;
2938 struct btrfs_fs_info *fs_info = trans->fs_info;
2939 struct btrfs_root *quota_root;
2940 struct btrfs_qgroup *srcgroup;
2941 struct btrfs_qgroup *dstgroup;
2942 bool need_rescan = false;
2943 u32 level_size = 0;
2944 u64 nums;
2945
2946 /*
2947 * There are only two callers of this function.
2948 *
2949 * One in create_subvol() in the ioctl context, which needs to hold
2950 * the qgroup_ioctl_lock.
2951 *
2952 * The other one in create_pending_snapshot() where no other qgroup
2953 * code can modify the fs as they all need to either start a new trans
2954 * or hold a trans handler, thus we don't need to hold
2955 * qgroup_ioctl_lock.
2956 * This would avoid long and complex lock chain and make lockdep happy.
2957 */
2958 spin_lock(&fs_info->trans_lock);
2959 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2960 committing = true;
2961 spin_unlock(&fs_info->trans_lock);
2962
2963 if (!committing)
2964 mutex_lock(&fs_info->qgroup_ioctl_lock);
2965 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2966 goto out;
2967
2968 quota_root = fs_info->quota_root;
2969 if (!quota_root) {
2970 ret = -EINVAL;
2971 goto out;
2972 }
2973
2974 if (inherit) {
2975 i_qgroups = (u64 *)(inherit + 1);
2976 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2977 2 * inherit->num_excl_copies;
2978 for (i = 0; i < nums; ++i) {
2979 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2980
2981 /*
2982 * Zero out invalid groups so we can ignore
2983 * them later.
2984 */
2985 if (!srcgroup ||
2986 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2987 *i_qgroups = 0ULL;
2988
2989 ++i_qgroups;
2990 }
2991 }
2992
2993 /*
2994 * create a tracking group for the subvol itself
2995 */
2996 ret = add_qgroup_item(trans, quota_root, objectid);
2997 if (ret)
2998 goto out;
2999
3000 /*
3001 * add qgroup to all inherited groups
3002 */
3003 if (inherit) {
3004 i_qgroups = (u64 *)(inherit + 1);
3005 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
3006 if (*i_qgroups == 0)
3007 continue;
3008 ret = add_qgroup_relation_item(trans, objectid,
3009 *i_qgroups);
3010 if (ret && ret != -EEXIST)
3011 goto out;
3012 ret = add_qgroup_relation_item(trans, *i_qgroups,
3013 objectid);
3014 if (ret && ret != -EEXIST)
3015 goto out;
3016 }
3017 ret = 0;
3018 }
3019
3020
3021 spin_lock(&fs_info->qgroup_lock);
3022
3023 dstgroup = add_qgroup_rb(fs_info, objectid);
3024 if (IS_ERR(dstgroup)) {
3025 ret = PTR_ERR(dstgroup);
3026 goto unlock;
3027 }
3028
3029 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
3030 dstgroup->lim_flags = inherit->lim.flags;
3031 dstgroup->max_rfer = inherit->lim.max_rfer;
3032 dstgroup->max_excl = inherit->lim.max_excl;
3033 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
3034 dstgroup->rsv_excl = inherit->lim.rsv_excl;
3035
3036 qgroup_dirty(fs_info, dstgroup);
3037 }
3038
3039 if (srcid) {
3040 srcgroup = find_qgroup_rb(fs_info, srcid);
3041 if (!srcgroup)
3042 goto unlock;
3043
3044 /*
3045 * We call inherit after we clone the root in order to make sure
3046 * our counts don't go crazy, so at this point the only
3047 * difference between the two roots should be the root node.
3048 */
3049 level_size = fs_info->nodesize;
3050 dstgroup->rfer = srcgroup->rfer;
3051 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
3052 dstgroup->excl = level_size;
3053 dstgroup->excl_cmpr = level_size;
3054 srcgroup->excl = level_size;
3055 srcgroup->excl_cmpr = level_size;
3056
3057 /* inherit the limit info */
3058 dstgroup->lim_flags = srcgroup->lim_flags;
3059 dstgroup->max_rfer = srcgroup->max_rfer;
3060 dstgroup->max_excl = srcgroup->max_excl;
3061 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
3062 dstgroup->rsv_excl = srcgroup->rsv_excl;
3063
3064 qgroup_dirty(fs_info, dstgroup);
3065 qgroup_dirty(fs_info, srcgroup);
3066 }
3067
3068 if (!inherit)
3069 goto unlock;
3070
3071 i_qgroups = (u64 *)(inherit + 1);
3072 for (i = 0; i < inherit->num_qgroups; ++i) {
3073 if (*i_qgroups) {
3074 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
3075 if (ret)
3076 goto unlock;
3077 }
3078 ++i_qgroups;
3079
3080 /*
3081 * If we're doing a snapshot, and adding the snapshot to a new
3082 * qgroup, the numbers are guaranteed to be incorrect.
3083 */
3084 if (srcid)
3085 need_rescan = true;
3086 }
3087
3088 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
3089 struct btrfs_qgroup *src;
3090 struct btrfs_qgroup *dst;
3091
3092 if (!i_qgroups[0] || !i_qgroups[1])
3093 continue;
3094
3095 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3096 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3097
3098 if (!src || !dst) {
3099 ret = -EINVAL;
3100 goto unlock;
3101 }
3102
3103 dst->rfer = src->rfer - level_size;
3104 dst->rfer_cmpr = src->rfer_cmpr - level_size;
3105
3106 /* Manually tweaking numbers certainly needs a rescan */
3107 need_rescan = true;
3108 }
3109 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
3110 struct btrfs_qgroup *src;
3111 struct btrfs_qgroup *dst;
3112
3113 if (!i_qgroups[0] || !i_qgroups[1])
3114 continue;
3115
3116 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3117 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3118
3119 if (!src || !dst) {
3120 ret = -EINVAL;
3121 goto unlock;
3122 }
3123
3124 dst->excl = src->excl + level_size;
3125 dst->excl_cmpr = src->excl_cmpr + level_size;
3126 need_rescan = true;
3127 }
3128
3129 unlock:
3130 spin_unlock(&fs_info->qgroup_lock);
3131 if (!ret)
3132 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3133 out:
3134 if (!committing)
3135 mutex_unlock(&fs_info->qgroup_ioctl_lock);
3136 if (need_rescan)
3137 qgroup_mark_inconsistent(fs_info);
3138 return ret;
3139 }
3140
qgroup_check_limits(const struct btrfs_qgroup * qg,u64 num_bytes)3141 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3142 {
3143 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3144 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3145 return false;
3146
3147 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3148 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3149 return false;
3150
3151 return true;
3152 }
3153
qgroup_reserve(struct btrfs_root * root,u64 num_bytes,bool enforce,enum btrfs_qgroup_rsv_type type)3154 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3155 enum btrfs_qgroup_rsv_type type)
3156 {
3157 struct btrfs_qgroup *qgroup;
3158 struct btrfs_fs_info *fs_info = root->fs_info;
3159 u64 ref_root = root->root_key.objectid;
3160 int ret = 0;
3161 LIST_HEAD(qgroup_list);
3162
3163 if (!is_fstree(ref_root))
3164 return 0;
3165
3166 if (num_bytes == 0)
3167 return 0;
3168
3169 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3170 capable(CAP_SYS_RESOURCE))
3171 enforce = false;
3172
3173 spin_lock(&fs_info->qgroup_lock);
3174 if (!fs_info->quota_root)
3175 goto out;
3176
3177 qgroup = find_qgroup_rb(fs_info, ref_root);
3178 if (!qgroup)
3179 goto out;
3180
3181 qgroup_iterator_add(&qgroup_list, qgroup);
3182 list_for_each_entry(qgroup, &qgroup_list, iterator) {
3183 struct btrfs_qgroup_list *glist;
3184
3185 if (enforce && !qgroup_check_limits(qgroup, num_bytes)) {
3186 ret = -EDQUOT;
3187 goto out;
3188 }
3189
3190 list_for_each_entry(glist, &qgroup->groups, next_group)
3191 qgroup_iterator_add(&qgroup_list, glist->group);
3192 }
3193
3194 ret = 0;
3195 /*
3196 * no limits exceeded, now record the reservation into all qgroups
3197 */
3198 list_for_each_entry(qgroup, &qgroup_list, iterator)
3199 qgroup_rsv_add(fs_info, qgroup, num_bytes, type);
3200
3201 out:
3202 qgroup_iterator_clean(&qgroup_list);
3203 spin_unlock(&fs_info->qgroup_lock);
3204 return ret;
3205 }
3206
3207 /*
3208 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
3209 * qgroup).
3210 *
3211 * Will handle all higher level qgroup too.
3212 *
3213 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3214 * This special case is only used for META_PERTRANS type.
3215 */
btrfs_qgroup_free_refroot(struct btrfs_fs_info * fs_info,u64 ref_root,u64 num_bytes,enum btrfs_qgroup_rsv_type type)3216 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3217 u64 ref_root, u64 num_bytes,
3218 enum btrfs_qgroup_rsv_type type)
3219 {
3220 struct btrfs_qgroup *qgroup;
3221 struct ulist_node *unode;
3222 struct ulist_iterator uiter;
3223 int ret = 0;
3224
3225 if (!is_fstree(ref_root))
3226 return;
3227
3228 if (num_bytes == 0)
3229 return;
3230
3231 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3232 WARN(1, "%s: Invalid type to free", __func__);
3233 return;
3234 }
3235 spin_lock(&fs_info->qgroup_lock);
3236
3237 if (!fs_info->quota_root)
3238 goto out;
3239
3240 qgroup = find_qgroup_rb(fs_info, ref_root);
3241 if (!qgroup)
3242 goto out;
3243
3244 if (num_bytes == (u64)-1)
3245 /*
3246 * We're freeing all pertrans rsv, get reserved value from
3247 * level 0 qgroup as real num_bytes to free.
3248 */
3249 num_bytes = qgroup->rsv.values[type];
3250
3251 ulist_reinit(fs_info->qgroup_ulist);
3252 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3253 qgroup_to_aux(qgroup), GFP_ATOMIC);
3254 if (ret < 0)
3255 goto out;
3256 ULIST_ITER_INIT(&uiter);
3257 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3258 struct btrfs_qgroup *qg;
3259 struct btrfs_qgroup_list *glist;
3260
3261 qg = unode_aux_to_qgroup(unode);
3262
3263 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3264
3265 list_for_each_entry(glist, &qg->groups, next_group) {
3266 ret = ulist_add(fs_info->qgroup_ulist,
3267 glist->group->qgroupid,
3268 qgroup_to_aux(glist->group), GFP_ATOMIC);
3269 if (ret < 0)
3270 goto out;
3271 }
3272 }
3273
3274 out:
3275 spin_unlock(&fs_info->qgroup_lock);
3276 }
3277
3278 /*
3279 * Check if the leaf is the last leaf. Which means all node pointers
3280 * are at their last position.
3281 */
is_last_leaf(struct btrfs_path * path)3282 static bool is_last_leaf(struct btrfs_path *path)
3283 {
3284 int i;
3285
3286 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3287 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3288 return false;
3289 }
3290 return true;
3291 }
3292
3293 /*
3294 * returns < 0 on error, 0 when more leafs are to be scanned.
3295 * returns 1 when done.
3296 */
qgroup_rescan_leaf(struct btrfs_trans_handle * trans,struct btrfs_path * path)3297 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3298 struct btrfs_path *path)
3299 {
3300 struct btrfs_fs_info *fs_info = trans->fs_info;
3301 struct btrfs_root *extent_root;
3302 struct btrfs_key found;
3303 struct extent_buffer *scratch_leaf = NULL;
3304 u64 num_bytes;
3305 bool done;
3306 int slot;
3307 int ret;
3308
3309 mutex_lock(&fs_info->qgroup_rescan_lock);
3310 extent_root = btrfs_extent_root(fs_info,
3311 fs_info->qgroup_rescan_progress.objectid);
3312 ret = btrfs_search_slot_for_read(extent_root,
3313 &fs_info->qgroup_rescan_progress,
3314 path, 1, 0);
3315
3316 btrfs_debug(fs_info,
3317 "current progress key (%llu %u %llu), search_slot ret %d",
3318 fs_info->qgroup_rescan_progress.objectid,
3319 fs_info->qgroup_rescan_progress.type,
3320 fs_info->qgroup_rescan_progress.offset, ret);
3321
3322 if (ret) {
3323 /*
3324 * The rescan is about to end, we will not be scanning any
3325 * further blocks. We cannot unset the RESCAN flag here, because
3326 * we want to commit the transaction if everything went well.
3327 * To make the live accounting work in this phase, we set our
3328 * scan progress pointer such that every real extent objectid
3329 * will be smaller.
3330 */
3331 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3332 btrfs_release_path(path);
3333 mutex_unlock(&fs_info->qgroup_rescan_lock);
3334 return ret;
3335 }
3336 done = is_last_leaf(path);
3337
3338 btrfs_item_key_to_cpu(path->nodes[0], &found,
3339 btrfs_header_nritems(path->nodes[0]) - 1);
3340 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3341
3342 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3343 if (!scratch_leaf) {
3344 ret = -ENOMEM;
3345 mutex_unlock(&fs_info->qgroup_rescan_lock);
3346 goto out;
3347 }
3348 slot = path->slots[0];
3349 btrfs_release_path(path);
3350 mutex_unlock(&fs_info->qgroup_rescan_lock);
3351
3352 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3353 struct btrfs_backref_walk_ctx ctx = { 0 };
3354
3355 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3356 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3357 found.type != BTRFS_METADATA_ITEM_KEY)
3358 continue;
3359 if (found.type == BTRFS_METADATA_ITEM_KEY)
3360 num_bytes = fs_info->nodesize;
3361 else
3362 num_bytes = found.offset;
3363
3364 ctx.bytenr = found.objectid;
3365 ctx.fs_info = fs_info;
3366
3367 ret = btrfs_find_all_roots(&ctx, false);
3368 if (ret < 0)
3369 goto out;
3370 /* For rescan, just pass old_roots as NULL */
3371 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3372 num_bytes, NULL, ctx.roots);
3373 if (ret < 0)
3374 goto out;
3375 }
3376 out:
3377 if (scratch_leaf)
3378 free_extent_buffer(scratch_leaf);
3379
3380 if (done && !ret) {
3381 ret = 1;
3382 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3383 }
3384 return ret;
3385 }
3386
rescan_should_stop(struct btrfs_fs_info * fs_info)3387 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3388 {
3389 return btrfs_fs_closing(fs_info) ||
3390 test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3391 !test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3392 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3393 }
3394
btrfs_qgroup_rescan_worker(struct btrfs_work * work)3395 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3396 {
3397 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3398 qgroup_rescan_work);
3399 struct btrfs_path *path;
3400 struct btrfs_trans_handle *trans = NULL;
3401 int err = -ENOMEM;
3402 int ret = 0;
3403 bool stopped = false;
3404 bool did_leaf_rescans = false;
3405
3406 path = btrfs_alloc_path();
3407 if (!path)
3408 goto out;
3409 /*
3410 * Rescan should only search for commit root, and any later difference
3411 * should be recorded by qgroup
3412 */
3413 path->search_commit_root = 1;
3414 path->skip_locking = 1;
3415
3416 err = 0;
3417 while (!err && !(stopped = rescan_should_stop(fs_info))) {
3418 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3419 if (IS_ERR(trans)) {
3420 err = PTR_ERR(trans);
3421 break;
3422 }
3423
3424 err = qgroup_rescan_leaf(trans, path);
3425 did_leaf_rescans = true;
3426
3427 if (err > 0)
3428 btrfs_commit_transaction(trans);
3429 else
3430 btrfs_end_transaction(trans);
3431 }
3432
3433 out:
3434 btrfs_free_path(path);
3435
3436 mutex_lock(&fs_info->qgroup_rescan_lock);
3437 if (err > 0 &&
3438 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3439 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3440 } else if (err < 0 || stopped) {
3441 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3442 }
3443 mutex_unlock(&fs_info->qgroup_rescan_lock);
3444
3445 /*
3446 * Only update status, since the previous part has already updated the
3447 * qgroup info, and only if we did any actual work. This also prevents
3448 * race with a concurrent quota disable, which has already set
3449 * fs_info->quota_root to NULL and cleared BTRFS_FS_QUOTA_ENABLED at
3450 * btrfs_quota_disable().
3451 */
3452 if (did_leaf_rescans) {
3453 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3454 if (IS_ERR(trans)) {
3455 err = PTR_ERR(trans);
3456 trans = NULL;
3457 btrfs_err(fs_info,
3458 "fail to start transaction for status update: %d",
3459 err);
3460 }
3461 } else {
3462 trans = NULL;
3463 }
3464
3465 mutex_lock(&fs_info->qgroup_rescan_lock);
3466 if (!stopped ||
3467 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN)
3468 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3469 if (trans) {
3470 ret = update_qgroup_status_item(trans);
3471 if (ret < 0) {
3472 err = ret;
3473 btrfs_err(fs_info, "fail to update qgroup status: %d",
3474 err);
3475 }
3476 }
3477 fs_info->qgroup_rescan_running = false;
3478 fs_info->qgroup_flags &= ~BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3479 complete_all(&fs_info->qgroup_rescan_completion);
3480 mutex_unlock(&fs_info->qgroup_rescan_lock);
3481
3482 if (!trans)
3483 return;
3484
3485 btrfs_end_transaction(trans);
3486
3487 if (stopped) {
3488 btrfs_info(fs_info, "qgroup scan paused");
3489 } else if (fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN) {
3490 btrfs_info(fs_info, "qgroup scan cancelled");
3491 } else if (err >= 0) {
3492 btrfs_info(fs_info, "qgroup scan completed%s",
3493 err > 0 ? " (inconsistency flag cleared)" : "");
3494 } else {
3495 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3496 }
3497 }
3498
3499 /*
3500 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3501 * memory required for the rescan context.
3502 */
3503 static int
qgroup_rescan_init(struct btrfs_fs_info * fs_info,u64 progress_objectid,int init_flags)3504 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3505 int init_flags)
3506 {
3507 int ret = 0;
3508
3509 if (!init_flags) {
3510 /* we're resuming qgroup rescan at mount time */
3511 if (!(fs_info->qgroup_flags &
3512 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3513 btrfs_warn(fs_info,
3514 "qgroup rescan init failed, qgroup rescan is not queued");
3515 ret = -EINVAL;
3516 } else if (!(fs_info->qgroup_flags &
3517 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3518 btrfs_warn(fs_info,
3519 "qgroup rescan init failed, qgroup is not enabled");
3520 ret = -EINVAL;
3521 }
3522
3523 if (ret)
3524 return ret;
3525 }
3526
3527 mutex_lock(&fs_info->qgroup_rescan_lock);
3528
3529 if (init_flags) {
3530 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3531 btrfs_warn(fs_info,
3532 "qgroup rescan is already in progress");
3533 ret = -EINPROGRESS;
3534 } else if (!(fs_info->qgroup_flags &
3535 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3536 btrfs_warn(fs_info,
3537 "qgroup rescan init failed, qgroup is not enabled");
3538 ret = -EINVAL;
3539 } else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3540 /* Quota disable is in progress */
3541 ret = -EBUSY;
3542 }
3543
3544 if (ret) {
3545 mutex_unlock(&fs_info->qgroup_rescan_lock);
3546 return ret;
3547 }
3548 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3549 }
3550
3551 memset(&fs_info->qgroup_rescan_progress, 0,
3552 sizeof(fs_info->qgroup_rescan_progress));
3553 fs_info->qgroup_flags &= ~(BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
3554 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
3555 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3556 init_completion(&fs_info->qgroup_rescan_completion);
3557 mutex_unlock(&fs_info->qgroup_rescan_lock);
3558
3559 btrfs_init_work(&fs_info->qgroup_rescan_work,
3560 btrfs_qgroup_rescan_worker, NULL, NULL);
3561 return 0;
3562 }
3563
3564 static void
qgroup_rescan_zero_tracking(struct btrfs_fs_info * fs_info)3565 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3566 {
3567 struct rb_node *n;
3568 struct btrfs_qgroup *qgroup;
3569
3570 spin_lock(&fs_info->qgroup_lock);
3571 /* clear all current qgroup tracking information */
3572 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3573 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3574 qgroup->rfer = 0;
3575 qgroup->rfer_cmpr = 0;
3576 qgroup->excl = 0;
3577 qgroup->excl_cmpr = 0;
3578 qgroup_dirty(fs_info, qgroup);
3579 }
3580 spin_unlock(&fs_info->qgroup_lock);
3581 }
3582
3583 int
btrfs_qgroup_rescan(struct btrfs_fs_info * fs_info)3584 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3585 {
3586 int ret = 0;
3587 struct btrfs_trans_handle *trans;
3588
3589 ret = qgroup_rescan_init(fs_info, 0, 1);
3590 if (ret)
3591 return ret;
3592
3593 /*
3594 * We have set the rescan_progress to 0, which means no more
3595 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3596 * However, btrfs_qgroup_account_ref may be right after its call
3597 * to btrfs_find_all_roots, in which case it would still do the
3598 * accounting.
3599 * To solve this, we're committing the transaction, which will
3600 * ensure we run all delayed refs and only after that, we are
3601 * going to clear all tracking information for a clean start.
3602 */
3603
3604 trans = btrfs_attach_transaction_barrier(fs_info->fs_root);
3605 if (IS_ERR(trans) && trans != ERR_PTR(-ENOENT)) {
3606 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3607 return PTR_ERR(trans);
3608 } else if (trans != ERR_PTR(-ENOENT)) {
3609 ret = btrfs_commit_transaction(trans);
3610 if (ret) {
3611 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3612 return ret;
3613 }
3614 }
3615
3616 qgroup_rescan_zero_tracking(fs_info);
3617
3618 mutex_lock(&fs_info->qgroup_rescan_lock);
3619 fs_info->qgroup_rescan_running = true;
3620 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3621 &fs_info->qgroup_rescan_work);
3622 mutex_unlock(&fs_info->qgroup_rescan_lock);
3623
3624 return 0;
3625 }
3626
btrfs_qgroup_wait_for_completion(struct btrfs_fs_info * fs_info,bool interruptible)3627 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3628 bool interruptible)
3629 {
3630 int running;
3631 int ret = 0;
3632
3633 mutex_lock(&fs_info->qgroup_rescan_lock);
3634 running = fs_info->qgroup_rescan_running;
3635 mutex_unlock(&fs_info->qgroup_rescan_lock);
3636
3637 if (!running)
3638 return 0;
3639
3640 if (interruptible)
3641 ret = wait_for_completion_interruptible(
3642 &fs_info->qgroup_rescan_completion);
3643 else
3644 wait_for_completion(&fs_info->qgroup_rescan_completion);
3645
3646 return ret;
3647 }
3648
3649 /*
3650 * this is only called from open_ctree where we're still single threaded, thus
3651 * locking is omitted here.
3652 */
3653 void
btrfs_qgroup_rescan_resume(struct btrfs_fs_info * fs_info)3654 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3655 {
3656 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3657 mutex_lock(&fs_info->qgroup_rescan_lock);
3658 fs_info->qgroup_rescan_running = true;
3659 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3660 &fs_info->qgroup_rescan_work);
3661 mutex_unlock(&fs_info->qgroup_rescan_lock);
3662 }
3663 }
3664
3665 #define rbtree_iterate_from_safe(node, next, start) \
3666 for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3667
qgroup_unreserve_range(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len)3668 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3669 struct extent_changeset *reserved, u64 start,
3670 u64 len)
3671 {
3672 struct rb_node *node;
3673 struct rb_node *next;
3674 struct ulist_node *entry;
3675 int ret = 0;
3676
3677 node = reserved->range_changed.root.rb_node;
3678 if (!node)
3679 return 0;
3680 while (node) {
3681 entry = rb_entry(node, struct ulist_node, rb_node);
3682 if (entry->val < start)
3683 node = node->rb_right;
3684 else
3685 node = node->rb_left;
3686 }
3687
3688 if (entry->val > start && rb_prev(&entry->rb_node))
3689 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3690 rb_node);
3691
3692 rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3693 u64 entry_start;
3694 u64 entry_end;
3695 u64 entry_len;
3696 int clear_ret;
3697
3698 entry = rb_entry(node, struct ulist_node, rb_node);
3699 entry_start = entry->val;
3700 entry_end = entry->aux;
3701 entry_len = entry_end - entry_start + 1;
3702
3703 if (entry_start >= start + len)
3704 break;
3705 if (entry_start + entry_len <= start)
3706 continue;
3707 /*
3708 * Now the entry is in [start, start + len), revert the
3709 * EXTENT_QGROUP_RESERVED bit.
3710 */
3711 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3712 entry_end, EXTENT_QGROUP_RESERVED);
3713 if (!ret && clear_ret < 0)
3714 ret = clear_ret;
3715
3716 ulist_del(&reserved->range_changed, entry->val, entry->aux);
3717 if (likely(reserved->bytes_changed >= entry_len)) {
3718 reserved->bytes_changed -= entry_len;
3719 } else {
3720 WARN_ON(1);
3721 reserved->bytes_changed = 0;
3722 }
3723 }
3724
3725 return ret;
3726 }
3727
3728 /*
3729 * Try to free some space for qgroup.
3730 *
3731 * For qgroup, there are only 3 ways to free qgroup space:
3732 * - Flush nodatacow write
3733 * Any nodatacow write will free its reserved data space at run_delalloc_range().
3734 * In theory, we should only flush nodatacow inodes, but it's not yet
3735 * possible, so we need to flush the whole root.
3736 *
3737 * - Wait for ordered extents
3738 * When ordered extents are finished, their reserved metadata is finally
3739 * converted to per_trans status, which can be freed by later commit
3740 * transaction.
3741 *
3742 * - Commit transaction
3743 * This would free the meta_per_trans space.
3744 * In theory this shouldn't provide much space, but any more qgroup space
3745 * is needed.
3746 */
try_flush_qgroup(struct btrfs_root * root)3747 static int try_flush_qgroup(struct btrfs_root *root)
3748 {
3749 struct btrfs_trans_handle *trans;
3750 int ret;
3751
3752 /* Can't hold an open transaction or we run the risk of deadlocking. */
3753 ASSERT(current->journal_info == NULL);
3754 if (WARN_ON(current->journal_info))
3755 return 0;
3756
3757 /*
3758 * We don't want to run flush again and again, so if there is a running
3759 * one, we won't try to start a new flush, but exit directly.
3760 */
3761 if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3762 wait_event(root->qgroup_flush_wait,
3763 !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3764 return 0;
3765 }
3766
3767 ret = btrfs_start_delalloc_snapshot(root, true);
3768 if (ret < 0)
3769 goto out;
3770 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3771
3772 trans = btrfs_attach_transaction_barrier(root);
3773 if (IS_ERR(trans)) {
3774 ret = PTR_ERR(trans);
3775 if (ret == -ENOENT)
3776 ret = 0;
3777 goto out;
3778 }
3779
3780 ret = btrfs_commit_transaction(trans);
3781 out:
3782 clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3783 wake_up(&root->qgroup_flush_wait);
3784 return ret;
3785 }
3786
qgroup_reserve_data(struct btrfs_inode * inode,struct extent_changeset ** reserved_ret,u64 start,u64 len)3787 static int qgroup_reserve_data(struct btrfs_inode *inode,
3788 struct extent_changeset **reserved_ret, u64 start,
3789 u64 len)
3790 {
3791 struct btrfs_root *root = inode->root;
3792 struct extent_changeset *reserved;
3793 bool new_reserved = false;
3794 u64 orig_reserved;
3795 u64 to_reserve;
3796 int ret;
3797
3798 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3799 !is_fstree(root->root_key.objectid) || len == 0)
3800 return 0;
3801
3802 /* @reserved parameter is mandatory for qgroup */
3803 if (WARN_ON(!reserved_ret))
3804 return -EINVAL;
3805 if (!*reserved_ret) {
3806 new_reserved = true;
3807 *reserved_ret = extent_changeset_alloc();
3808 if (!*reserved_ret)
3809 return -ENOMEM;
3810 }
3811 reserved = *reserved_ret;
3812 /* Record already reserved space */
3813 orig_reserved = reserved->bytes_changed;
3814 ret = set_record_extent_bits(&inode->io_tree, start,
3815 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3816
3817 /* Newly reserved space */
3818 to_reserve = reserved->bytes_changed - orig_reserved;
3819 trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3820 to_reserve, QGROUP_RESERVE);
3821 if (ret < 0)
3822 goto out;
3823 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3824 if (ret < 0)
3825 goto cleanup;
3826
3827 return ret;
3828
3829 cleanup:
3830 qgroup_unreserve_range(inode, reserved, start, len);
3831 out:
3832 if (new_reserved) {
3833 extent_changeset_free(reserved);
3834 *reserved_ret = NULL;
3835 }
3836 return ret;
3837 }
3838
3839 /*
3840 * Reserve qgroup space for range [start, start + len).
3841 *
3842 * This function will either reserve space from related qgroups or do nothing
3843 * if the range is already reserved.
3844 *
3845 * Return 0 for successful reservation
3846 * Return <0 for error (including -EQUOT)
3847 *
3848 * NOTE: This function may sleep for memory allocation, dirty page flushing and
3849 * commit transaction. So caller should not hold any dirty page locked.
3850 */
btrfs_qgroup_reserve_data(struct btrfs_inode * inode,struct extent_changeset ** reserved_ret,u64 start,u64 len)3851 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3852 struct extent_changeset **reserved_ret, u64 start,
3853 u64 len)
3854 {
3855 int ret;
3856
3857 ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3858 if (ret <= 0 && ret != -EDQUOT)
3859 return ret;
3860
3861 ret = try_flush_qgroup(inode->root);
3862 if (ret < 0)
3863 return ret;
3864 return qgroup_reserve_data(inode, reserved_ret, start, len);
3865 }
3866
3867 /* Free ranges specified by @reserved, normally in error path */
qgroup_free_reserved_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,u64 * freed_ret)3868 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3869 struct extent_changeset *reserved,
3870 u64 start, u64 len, u64 *freed_ret)
3871 {
3872 struct btrfs_root *root = inode->root;
3873 struct ulist_node *unode;
3874 struct ulist_iterator uiter;
3875 struct extent_changeset changeset;
3876 u64 freed = 0;
3877 int ret;
3878
3879 extent_changeset_init(&changeset);
3880 len = round_up(start + len, root->fs_info->sectorsize);
3881 start = round_down(start, root->fs_info->sectorsize);
3882
3883 ULIST_ITER_INIT(&uiter);
3884 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3885 u64 range_start = unode->val;
3886 /* unode->aux is the inclusive end */
3887 u64 range_len = unode->aux - range_start + 1;
3888 u64 free_start;
3889 u64 free_len;
3890
3891 extent_changeset_release(&changeset);
3892
3893 /* Only free range in range [start, start + len) */
3894 if (range_start >= start + len ||
3895 range_start + range_len <= start)
3896 continue;
3897 free_start = max(range_start, start);
3898 free_len = min(start + len, range_start + range_len) -
3899 free_start;
3900 /*
3901 * TODO: To also modify reserved->ranges_reserved to reflect
3902 * the modification.
3903 *
3904 * However as long as we free qgroup reserved according to
3905 * EXTENT_QGROUP_RESERVED, we won't double free.
3906 * So not need to rush.
3907 */
3908 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3909 free_start + free_len - 1,
3910 EXTENT_QGROUP_RESERVED, &changeset);
3911 if (ret < 0)
3912 goto out;
3913 freed += changeset.bytes_changed;
3914 }
3915 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3916 BTRFS_QGROUP_RSV_DATA);
3917 if (freed_ret)
3918 *freed_ret = freed;
3919 ret = 0;
3920 out:
3921 extent_changeset_release(&changeset);
3922 return ret;
3923 }
3924
__btrfs_qgroup_release_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,u64 * released,int free)3925 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3926 struct extent_changeset *reserved, u64 start, u64 len,
3927 u64 *released, int free)
3928 {
3929 struct extent_changeset changeset;
3930 int trace_op = QGROUP_RELEASE;
3931 int ret;
3932
3933 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3934 return 0;
3935
3936 /* In release case, we shouldn't have @reserved */
3937 WARN_ON(!free && reserved);
3938 if (free && reserved)
3939 return qgroup_free_reserved_data(inode, reserved, start, len, released);
3940 extent_changeset_init(&changeset);
3941 ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3942 EXTENT_QGROUP_RESERVED, &changeset);
3943 if (ret < 0)
3944 goto out;
3945
3946 if (free)
3947 trace_op = QGROUP_FREE;
3948 trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3949 changeset.bytes_changed, trace_op);
3950 if (free)
3951 btrfs_qgroup_free_refroot(inode->root->fs_info,
3952 inode->root->root_key.objectid,
3953 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3954 if (released)
3955 *released = changeset.bytes_changed;
3956 out:
3957 extent_changeset_release(&changeset);
3958 return ret;
3959 }
3960
3961 /*
3962 * Free a reserved space range from io_tree and related qgroups
3963 *
3964 * Should be called when a range of pages get invalidated before reaching disk.
3965 * Or for error cleanup case.
3966 * if @reserved is given, only reserved range in [@start, @start + @len) will
3967 * be freed.
3968 *
3969 * For data written to disk, use btrfs_qgroup_release_data().
3970 *
3971 * NOTE: This function may sleep for memory allocation.
3972 */
btrfs_qgroup_free_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,u64 * freed)3973 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3974 struct extent_changeset *reserved,
3975 u64 start, u64 len, u64 *freed)
3976 {
3977 return __btrfs_qgroup_release_data(inode, reserved, start, len, freed, 1);
3978 }
3979
3980 /*
3981 * Release a reserved space range from io_tree only.
3982 *
3983 * Should be called when a range of pages get written to disk and corresponding
3984 * FILE_EXTENT is inserted into corresponding root.
3985 *
3986 * Since new qgroup accounting framework will only update qgroup numbers at
3987 * commit_transaction() time, its reserved space shouldn't be freed from
3988 * related qgroups.
3989 *
3990 * But we should release the range from io_tree, to allow further write to be
3991 * COWed.
3992 *
3993 * NOTE: This function may sleep for memory allocation.
3994 */
btrfs_qgroup_release_data(struct btrfs_inode * inode,u64 start,u64 len,u64 * released)3995 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released)
3996 {
3997 return __btrfs_qgroup_release_data(inode, NULL, start, len, released, 0);
3998 }
3999
add_root_meta_rsv(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)4000 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
4001 enum btrfs_qgroup_rsv_type type)
4002 {
4003 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
4004 type != BTRFS_QGROUP_RSV_META_PERTRANS)
4005 return;
4006 if (num_bytes == 0)
4007 return;
4008
4009 spin_lock(&root->qgroup_meta_rsv_lock);
4010 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
4011 root->qgroup_meta_rsv_prealloc += num_bytes;
4012 else
4013 root->qgroup_meta_rsv_pertrans += num_bytes;
4014 spin_unlock(&root->qgroup_meta_rsv_lock);
4015 }
4016
sub_root_meta_rsv(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)4017 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
4018 enum btrfs_qgroup_rsv_type type)
4019 {
4020 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
4021 type != BTRFS_QGROUP_RSV_META_PERTRANS)
4022 return 0;
4023 if (num_bytes == 0)
4024 return 0;
4025
4026 spin_lock(&root->qgroup_meta_rsv_lock);
4027 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
4028 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
4029 num_bytes);
4030 root->qgroup_meta_rsv_prealloc -= num_bytes;
4031 } else {
4032 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
4033 num_bytes);
4034 root->qgroup_meta_rsv_pertrans -= num_bytes;
4035 }
4036 spin_unlock(&root->qgroup_meta_rsv_lock);
4037 return num_bytes;
4038 }
4039
btrfs_qgroup_reserve_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type,bool enforce)4040 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4041 enum btrfs_qgroup_rsv_type type, bool enforce)
4042 {
4043 struct btrfs_fs_info *fs_info = root->fs_info;
4044 int ret;
4045
4046 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4047 !is_fstree(root->root_key.objectid) || num_bytes == 0)
4048 return 0;
4049
4050 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4051 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
4052 ret = qgroup_reserve(root, num_bytes, enforce, type);
4053 if (ret < 0)
4054 return ret;
4055 /*
4056 * Record what we have reserved into root.
4057 *
4058 * To avoid quota disabled->enabled underflow.
4059 * In that case, we may try to free space we haven't reserved
4060 * (since quota was disabled), so record what we reserved into root.
4061 * And ensure later release won't underflow this number.
4062 */
4063 add_root_meta_rsv(root, num_bytes, type);
4064 return ret;
4065 }
4066
__btrfs_qgroup_reserve_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type,bool enforce,bool noflush)4067 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4068 enum btrfs_qgroup_rsv_type type, bool enforce,
4069 bool noflush)
4070 {
4071 int ret;
4072
4073 ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4074 if ((ret <= 0 && ret != -EDQUOT) || noflush)
4075 return ret;
4076
4077 ret = try_flush_qgroup(root);
4078 if (ret < 0)
4079 return ret;
4080 return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4081 }
4082
btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root * root)4083 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
4084 {
4085 struct btrfs_fs_info *fs_info = root->fs_info;
4086
4087 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4088 !is_fstree(root->root_key.objectid))
4089 return;
4090
4091 /* TODO: Update trace point to handle such free */
4092 trace_qgroup_meta_free_all_pertrans(root);
4093 /* Special value -1 means to free all reserved space */
4094 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
4095 BTRFS_QGROUP_RSV_META_PERTRANS);
4096 }
4097
__btrfs_qgroup_free_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)4098 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
4099 enum btrfs_qgroup_rsv_type type)
4100 {
4101 struct btrfs_fs_info *fs_info = root->fs_info;
4102
4103 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4104 !is_fstree(root->root_key.objectid))
4105 return;
4106
4107 /*
4108 * reservation for META_PREALLOC can happen before quota is enabled,
4109 * which can lead to underflow.
4110 * Here ensure we will only free what we really have reserved.
4111 */
4112 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
4113 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4114 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
4115 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
4116 num_bytes, type);
4117 }
4118
qgroup_convert_meta(struct btrfs_fs_info * fs_info,u64 ref_root,int num_bytes)4119 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
4120 int num_bytes)
4121 {
4122 struct btrfs_qgroup *qgroup;
4123 LIST_HEAD(qgroup_list);
4124
4125 if (num_bytes == 0)
4126 return;
4127 if (!fs_info->quota_root)
4128 return;
4129
4130 spin_lock(&fs_info->qgroup_lock);
4131 qgroup = find_qgroup_rb(fs_info, ref_root);
4132 if (!qgroup)
4133 goto out;
4134
4135 qgroup_iterator_add(&qgroup_list, qgroup);
4136 list_for_each_entry(qgroup, &qgroup_list, iterator) {
4137 struct btrfs_qgroup_list *glist;
4138
4139 qgroup_rsv_release(fs_info, qgroup, num_bytes,
4140 BTRFS_QGROUP_RSV_META_PREALLOC);
4141 if (!sb_rdonly(fs_info->sb))
4142 qgroup_rsv_add(fs_info, qgroup, num_bytes,
4143 BTRFS_QGROUP_RSV_META_PERTRANS);
4144
4145 list_for_each_entry(glist, &qgroup->groups, next_group)
4146 qgroup_iterator_add(&qgroup_list, glist->group);
4147 }
4148 out:
4149 qgroup_iterator_clean(&qgroup_list);
4150 spin_unlock(&fs_info->qgroup_lock);
4151 }
4152
btrfs_qgroup_convert_reserved_meta(struct btrfs_root * root,int num_bytes)4153 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4154 {
4155 struct btrfs_fs_info *fs_info = root->fs_info;
4156
4157 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4158 !is_fstree(root->root_key.objectid))
4159 return;
4160 /* Same as btrfs_qgroup_free_meta_prealloc() */
4161 num_bytes = sub_root_meta_rsv(root, num_bytes,
4162 BTRFS_QGROUP_RSV_META_PREALLOC);
4163 trace_qgroup_meta_convert(root, num_bytes);
4164 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4165 }
4166
4167 /*
4168 * Check qgroup reserved space leaking, normally at destroy inode
4169 * time
4170 */
btrfs_qgroup_check_reserved_leak(struct btrfs_inode * inode)4171 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4172 {
4173 struct extent_changeset changeset;
4174 struct ulist_node *unode;
4175 struct ulist_iterator iter;
4176 int ret;
4177
4178 extent_changeset_init(&changeset);
4179 ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4180 EXTENT_QGROUP_RESERVED, &changeset);
4181
4182 WARN_ON(ret < 0);
4183 if (WARN_ON(changeset.bytes_changed)) {
4184 ULIST_ITER_INIT(&iter);
4185 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4186 btrfs_warn(inode->root->fs_info,
4187 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4188 btrfs_ino(inode), unode->val, unode->aux);
4189 }
4190 btrfs_qgroup_free_refroot(inode->root->fs_info,
4191 inode->root->root_key.objectid,
4192 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4193
4194 }
4195 extent_changeset_release(&changeset);
4196 }
4197
btrfs_qgroup_init_swapped_blocks(struct btrfs_qgroup_swapped_blocks * swapped_blocks)4198 void btrfs_qgroup_init_swapped_blocks(
4199 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4200 {
4201 int i;
4202
4203 spin_lock_init(&swapped_blocks->lock);
4204 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4205 swapped_blocks->blocks[i] = RB_ROOT;
4206 swapped_blocks->swapped = false;
4207 }
4208
4209 /*
4210 * Delete all swapped blocks record of @root.
4211 * Every record here means we skipped a full subtree scan for qgroup.
4212 *
4213 * Gets called when committing one transaction.
4214 */
btrfs_qgroup_clean_swapped_blocks(struct btrfs_root * root)4215 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4216 {
4217 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4218 int i;
4219
4220 swapped_blocks = &root->swapped_blocks;
4221
4222 spin_lock(&swapped_blocks->lock);
4223 if (!swapped_blocks->swapped)
4224 goto out;
4225 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4226 struct rb_root *cur_root = &swapped_blocks->blocks[i];
4227 struct btrfs_qgroup_swapped_block *entry;
4228 struct btrfs_qgroup_swapped_block *next;
4229
4230 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4231 node)
4232 kfree(entry);
4233 swapped_blocks->blocks[i] = RB_ROOT;
4234 }
4235 swapped_blocks->swapped = false;
4236 out:
4237 spin_unlock(&swapped_blocks->lock);
4238 }
4239
4240 /*
4241 * Add subtree roots record into @subvol_root.
4242 *
4243 * @subvol_root: tree root of the subvolume tree get swapped
4244 * @bg: block group under balance
4245 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4246 * @reloc_parent/slot: pointer to the subtree root in reloc tree
4247 * BOTH POINTERS ARE BEFORE TREE SWAP
4248 * @last_snapshot: last snapshot generation of the subvolume tree
4249 */
btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle * trans,struct btrfs_root * subvol_root,struct btrfs_block_group * bg,struct extent_buffer * subvol_parent,int subvol_slot,struct extent_buffer * reloc_parent,int reloc_slot,u64 last_snapshot)4250 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4251 struct btrfs_root *subvol_root,
4252 struct btrfs_block_group *bg,
4253 struct extent_buffer *subvol_parent, int subvol_slot,
4254 struct extent_buffer *reloc_parent, int reloc_slot,
4255 u64 last_snapshot)
4256 {
4257 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4258 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4259 struct btrfs_qgroup_swapped_block *block;
4260 struct rb_node **cur;
4261 struct rb_node *parent = NULL;
4262 int level = btrfs_header_level(subvol_parent) - 1;
4263 int ret = 0;
4264
4265 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4266 return 0;
4267
4268 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4269 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4270 btrfs_err_rl(fs_info,
4271 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4272 __func__,
4273 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4274 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4275 return -EUCLEAN;
4276 }
4277
4278 block = kmalloc(sizeof(*block), GFP_NOFS);
4279 if (!block) {
4280 ret = -ENOMEM;
4281 goto out;
4282 }
4283
4284 /*
4285 * @reloc_parent/slot is still before swap, while @block is going to
4286 * record the bytenr after swap, so we do the swap here.
4287 */
4288 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4289 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4290 reloc_slot);
4291 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4292 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4293 subvol_slot);
4294 block->last_snapshot = last_snapshot;
4295 block->level = level;
4296
4297 /*
4298 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4299 * no one else can modify tree blocks thus we qgroup will not change
4300 * no matter the value of trace_leaf.
4301 */
4302 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4303 block->trace_leaf = true;
4304 else
4305 block->trace_leaf = false;
4306 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4307
4308 /* Insert @block into @blocks */
4309 spin_lock(&blocks->lock);
4310 cur = &blocks->blocks[level].rb_node;
4311 while (*cur) {
4312 struct btrfs_qgroup_swapped_block *entry;
4313
4314 parent = *cur;
4315 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4316 node);
4317
4318 if (entry->subvol_bytenr < block->subvol_bytenr) {
4319 cur = &(*cur)->rb_left;
4320 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4321 cur = &(*cur)->rb_right;
4322 } else {
4323 if (entry->subvol_generation !=
4324 block->subvol_generation ||
4325 entry->reloc_bytenr != block->reloc_bytenr ||
4326 entry->reloc_generation !=
4327 block->reloc_generation) {
4328 /*
4329 * Duplicated but mismatch entry found.
4330 * Shouldn't happen.
4331 *
4332 * Marking qgroup inconsistent should be enough
4333 * for end users.
4334 */
4335 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4336 ret = -EEXIST;
4337 }
4338 kfree(block);
4339 goto out_unlock;
4340 }
4341 }
4342 rb_link_node(&block->node, parent, cur);
4343 rb_insert_color(&block->node, &blocks->blocks[level]);
4344 blocks->swapped = true;
4345 out_unlock:
4346 spin_unlock(&blocks->lock);
4347 out:
4348 if (ret < 0)
4349 qgroup_mark_inconsistent(fs_info);
4350 return ret;
4351 }
4352
4353 /*
4354 * Check if the tree block is a subtree root, and if so do the needed
4355 * delayed subtree trace for qgroup.
4356 *
4357 * This is called during btrfs_cow_block().
4358 */
btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * subvol_eb)4359 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4360 struct btrfs_root *root,
4361 struct extent_buffer *subvol_eb)
4362 {
4363 struct btrfs_fs_info *fs_info = root->fs_info;
4364 struct btrfs_tree_parent_check check = { 0 };
4365 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4366 struct btrfs_qgroup_swapped_block *block;
4367 struct extent_buffer *reloc_eb = NULL;
4368 struct rb_node *node;
4369 bool found = false;
4370 bool swapped = false;
4371 int level = btrfs_header_level(subvol_eb);
4372 int ret = 0;
4373 int i;
4374
4375 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4376 return 0;
4377 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4378 return 0;
4379
4380 spin_lock(&blocks->lock);
4381 if (!blocks->swapped) {
4382 spin_unlock(&blocks->lock);
4383 return 0;
4384 }
4385 node = blocks->blocks[level].rb_node;
4386
4387 while (node) {
4388 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4389 if (block->subvol_bytenr < subvol_eb->start) {
4390 node = node->rb_left;
4391 } else if (block->subvol_bytenr > subvol_eb->start) {
4392 node = node->rb_right;
4393 } else {
4394 found = true;
4395 break;
4396 }
4397 }
4398 if (!found) {
4399 spin_unlock(&blocks->lock);
4400 goto out;
4401 }
4402 /* Found one, remove it from @blocks first and update blocks->swapped */
4403 rb_erase(&block->node, &blocks->blocks[level]);
4404 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4405 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4406 swapped = true;
4407 break;
4408 }
4409 }
4410 blocks->swapped = swapped;
4411 spin_unlock(&blocks->lock);
4412
4413 check.level = block->level;
4414 check.transid = block->reloc_generation;
4415 check.has_first_key = true;
4416 memcpy(&check.first_key, &block->first_key, sizeof(check.first_key));
4417
4418 /* Read out reloc subtree root */
4419 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, &check);
4420 if (IS_ERR(reloc_eb)) {
4421 ret = PTR_ERR(reloc_eb);
4422 reloc_eb = NULL;
4423 goto free_out;
4424 }
4425 if (!extent_buffer_uptodate(reloc_eb)) {
4426 ret = -EIO;
4427 goto free_out;
4428 }
4429
4430 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4431 block->last_snapshot, block->trace_leaf);
4432 free_out:
4433 kfree(block);
4434 free_extent_buffer(reloc_eb);
4435 out:
4436 if (ret < 0) {
4437 btrfs_err_rl(fs_info,
4438 "failed to account subtree at bytenr %llu: %d",
4439 subvol_eb->start, ret);
4440 qgroup_mark_inconsistent(fs_info);
4441 }
4442 return ret;
4443 }
4444
btrfs_qgroup_destroy_extent_records(struct btrfs_transaction * trans)4445 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4446 {
4447 struct btrfs_qgroup_extent_record *entry;
4448 struct btrfs_qgroup_extent_record *next;
4449 struct rb_root *root;
4450
4451 root = &trans->delayed_refs.dirty_extent_root;
4452 rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4453 ulist_free(entry->old_roots);
4454 kfree(entry);
4455 }
4456 *root = RB_ROOT;
4457 }
4458