xref: /openbmc/linux/fs/btrfs/qgroup.c (revision f1288bdb)
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 
28 /*
29  * Helpers to access qgroup reservation
30  *
31  * Callers should ensure the lock context and type are valid
32  */
33 
34 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
35 {
36 	u64 ret = 0;
37 	int i;
38 
39 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
40 		ret += qgroup->rsv.values[i];
41 
42 	return ret;
43 }
44 
45 #ifdef CONFIG_BTRFS_DEBUG
46 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
47 {
48 	if (type == BTRFS_QGROUP_RSV_DATA)
49 		return "data";
50 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
51 		return "meta_pertrans";
52 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
53 		return "meta_prealloc";
54 	return NULL;
55 }
56 #endif
57 
58 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
59 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
60 			   enum btrfs_qgroup_rsv_type type)
61 {
62 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
63 	qgroup->rsv.values[type] += num_bytes;
64 }
65 
66 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
67 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
68 			       enum btrfs_qgroup_rsv_type type)
69 {
70 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
71 	if (qgroup->rsv.values[type] >= num_bytes) {
72 		qgroup->rsv.values[type] -= num_bytes;
73 		return;
74 	}
75 #ifdef CONFIG_BTRFS_DEBUG
76 	WARN_RATELIMIT(1,
77 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
78 		qgroup->qgroupid, qgroup_rsv_type_str(type),
79 		qgroup->rsv.values[type], num_bytes);
80 #endif
81 	qgroup->rsv.values[type] = 0;
82 }
83 
84 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
85 				     struct btrfs_qgroup *dest,
86 				     struct btrfs_qgroup *src)
87 {
88 	int i;
89 
90 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
91 		qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
92 }
93 
94 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
95 					 struct btrfs_qgroup *dest,
96 					  struct btrfs_qgroup *src)
97 {
98 	int i;
99 
100 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
102 }
103 
104 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
105 					   int mod)
106 {
107 	if (qg->old_refcnt < seq)
108 		qg->old_refcnt = seq;
109 	qg->old_refcnt += mod;
110 }
111 
112 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
113 					   int mod)
114 {
115 	if (qg->new_refcnt < seq)
116 		qg->new_refcnt = seq;
117 	qg->new_refcnt += mod;
118 }
119 
120 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
121 {
122 	if (qg->old_refcnt < seq)
123 		return 0;
124 	return qg->old_refcnt - seq;
125 }
126 
127 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
128 {
129 	if (qg->new_refcnt < seq)
130 		return 0;
131 	return qg->new_refcnt - seq;
132 }
133 
134 /*
135  * glue structure to represent the relations between qgroups.
136  */
137 struct btrfs_qgroup_list {
138 	struct list_head next_group;
139 	struct list_head next_member;
140 	struct btrfs_qgroup *group;
141 	struct btrfs_qgroup *member;
142 };
143 
144 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
145 {
146 	return (u64)(uintptr_t)qg;
147 }
148 
149 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
150 {
151 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
152 }
153 
154 static int
155 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
156 		   int init_flags);
157 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
158 
159 /* must be called with qgroup_ioctl_lock held */
160 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
161 					   u64 qgroupid)
162 {
163 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
164 	struct btrfs_qgroup *qgroup;
165 
166 	while (n) {
167 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
168 		if (qgroup->qgroupid < qgroupid)
169 			n = n->rb_left;
170 		else if (qgroup->qgroupid > qgroupid)
171 			n = n->rb_right;
172 		else
173 			return qgroup;
174 	}
175 	return NULL;
176 }
177 
178 /* must be called with qgroup_lock held */
179 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
180 					  u64 qgroupid)
181 {
182 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
183 	struct rb_node *parent = NULL;
184 	struct btrfs_qgroup *qgroup;
185 
186 	while (*p) {
187 		parent = *p;
188 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
189 
190 		if (qgroup->qgroupid < qgroupid)
191 			p = &(*p)->rb_left;
192 		else if (qgroup->qgroupid > qgroupid)
193 			p = &(*p)->rb_right;
194 		else
195 			return qgroup;
196 	}
197 
198 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
199 	if (!qgroup)
200 		return ERR_PTR(-ENOMEM);
201 
202 	qgroup->qgroupid = qgroupid;
203 	INIT_LIST_HEAD(&qgroup->groups);
204 	INIT_LIST_HEAD(&qgroup->members);
205 	INIT_LIST_HEAD(&qgroup->dirty);
206 
207 	rb_link_node(&qgroup->node, parent, p);
208 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
209 
210 	return qgroup;
211 }
212 
213 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
214 			    struct btrfs_qgroup *qgroup)
215 {
216 	struct btrfs_qgroup_list *list;
217 
218 	list_del(&qgroup->dirty);
219 	while (!list_empty(&qgroup->groups)) {
220 		list = list_first_entry(&qgroup->groups,
221 					struct btrfs_qgroup_list, next_group);
222 		list_del(&list->next_group);
223 		list_del(&list->next_member);
224 		kfree(list);
225 	}
226 
227 	while (!list_empty(&qgroup->members)) {
228 		list = list_first_entry(&qgroup->members,
229 					struct btrfs_qgroup_list, next_member);
230 		list_del(&list->next_group);
231 		list_del(&list->next_member);
232 		kfree(list);
233 	}
234 }
235 
236 /* must be called with qgroup_lock held */
237 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
238 {
239 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
240 
241 	if (!qgroup)
242 		return -ENOENT;
243 
244 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
245 	__del_qgroup_rb(fs_info, qgroup);
246 	return 0;
247 }
248 
249 /*
250  * Add relation specified by two qgroups.
251  *
252  * Must be called with qgroup_lock held.
253  *
254  * Return: 0        on success
255  *         -ENOENT  if one of the qgroups is NULL
256  *         <0       other errors
257  */
258 static int __add_relation_rb(struct btrfs_qgroup *member, struct btrfs_qgroup *parent)
259 {
260 	struct btrfs_qgroup_list *list;
261 
262 	if (!member || !parent)
263 		return -ENOENT;
264 
265 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
266 	if (!list)
267 		return -ENOMEM;
268 
269 	list->group = parent;
270 	list->member = member;
271 	list_add_tail(&list->next_group, &member->groups);
272 	list_add_tail(&list->next_member, &parent->members);
273 
274 	return 0;
275 }
276 
277 /*
278  * Add relation specified by two qgoup ids.
279  *
280  * Must be called with qgroup_lock held.
281  *
282  * Return: 0        on success
283  *         -ENOENT  if one of the ids does not exist
284  *         <0       other errors
285  */
286 static int add_relation_rb(struct btrfs_fs_info *fs_info, u64 memberid, u64 parentid)
287 {
288 	struct btrfs_qgroup *member;
289 	struct btrfs_qgroup *parent;
290 
291 	member = find_qgroup_rb(fs_info, memberid);
292 	parent = find_qgroup_rb(fs_info, parentid);
293 
294 	return __add_relation_rb(member, parent);
295 }
296 
297 /* Must be called with qgroup_lock held */
298 static int del_relation_rb(struct btrfs_fs_info *fs_info,
299 			   u64 memberid, u64 parentid)
300 {
301 	struct btrfs_qgroup *member;
302 	struct btrfs_qgroup *parent;
303 	struct btrfs_qgroup_list *list;
304 
305 	member = find_qgroup_rb(fs_info, memberid);
306 	parent = find_qgroup_rb(fs_info, parentid);
307 	if (!member || !parent)
308 		return -ENOENT;
309 
310 	list_for_each_entry(list, &member->groups, next_group) {
311 		if (list->group == parent) {
312 			list_del(&list->next_group);
313 			list_del(&list->next_member);
314 			kfree(list);
315 			return 0;
316 		}
317 	}
318 	return -ENOENT;
319 }
320 
321 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
322 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
323 			       u64 rfer, u64 excl)
324 {
325 	struct btrfs_qgroup *qgroup;
326 
327 	qgroup = find_qgroup_rb(fs_info, qgroupid);
328 	if (!qgroup)
329 		return -EINVAL;
330 	if (qgroup->rfer != rfer || qgroup->excl != excl)
331 		return -EINVAL;
332 	return 0;
333 }
334 #endif
335 
336 /*
337  * The full config is read in one go, only called from open_ctree()
338  * It doesn't use any locking, as at this point we're still single-threaded
339  */
340 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
341 {
342 	struct btrfs_key key;
343 	struct btrfs_key found_key;
344 	struct btrfs_root *quota_root = fs_info->quota_root;
345 	struct btrfs_path *path = NULL;
346 	struct extent_buffer *l;
347 	int slot;
348 	int ret = 0;
349 	u64 flags = 0;
350 	u64 rescan_progress = 0;
351 
352 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
353 		return 0;
354 
355 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
356 	if (!fs_info->qgroup_ulist) {
357 		ret = -ENOMEM;
358 		goto out;
359 	}
360 
361 	path = btrfs_alloc_path();
362 	if (!path) {
363 		ret = -ENOMEM;
364 		goto out;
365 	}
366 
367 	ret = btrfs_sysfs_add_qgroups(fs_info);
368 	if (ret < 0)
369 		goto out;
370 	/* default this to quota off, in case no status key is found */
371 	fs_info->qgroup_flags = 0;
372 
373 	/*
374 	 * pass 1: read status, all qgroup infos and limits
375 	 */
376 	key.objectid = 0;
377 	key.type = 0;
378 	key.offset = 0;
379 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
380 	if (ret)
381 		goto out;
382 
383 	while (1) {
384 		struct btrfs_qgroup *qgroup;
385 
386 		slot = path->slots[0];
387 		l = path->nodes[0];
388 		btrfs_item_key_to_cpu(l, &found_key, slot);
389 
390 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
391 			struct btrfs_qgroup_status_item *ptr;
392 
393 			ptr = btrfs_item_ptr(l, slot,
394 					     struct btrfs_qgroup_status_item);
395 
396 			if (btrfs_qgroup_status_version(l, ptr) !=
397 			    BTRFS_QGROUP_STATUS_VERSION) {
398 				btrfs_err(fs_info,
399 				 "old qgroup version, quota disabled");
400 				goto out;
401 			}
402 			if (btrfs_qgroup_status_generation(l, ptr) !=
403 			    fs_info->generation) {
404 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
405 				btrfs_err(fs_info,
406 					"qgroup generation mismatch, marked as inconsistent");
407 			}
408 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
409 									  ptr);
410 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
411 			goto next1;
412 		}
413 
414 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
415 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
416 			goto next1;
417 
418 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
419 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
420 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
421 			btrfs_err(fs_info, "inconsistent qgroup config");
422 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
423 		}
424 		if (!qgroup) {
425 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
426 			if (IS_ERR(qgroup)) {
427 				ret = PTR_ERR(qgroup);
428 				goto out;
429 			}
430 		}
431 		ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
432 		if (ret < 0)
433 			goto out;
434 
435 		switch (found_key.type) {
436 		case BTRFS_QGROUP_INFO_KEY: {
437 			struct btrfs_qgroup_info_item *ptr;
438 
439 			ptr = btrfs_item_ptr(l, slot,
440 					     struct btrfs_qgroup_info_item);
441 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
442 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
443 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
444 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
445 			/* generation currently unused */
446 			break;
447 		}
448 		case BTRFS_QGROUP_LIMIT_KEY: {
449 			struct btrfs_qgroup_limit_item *ptr;
450 
451 			ptr = btrfs_item_ptr(l, slot,
452 					     struct btrfs_qgroup_limit_item);
453 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
454 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
455 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
456 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
457 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
458 			break;
459 		}
460 		}
461 next1:
462 		ret = btrfs_next_item(quota_root, path);
463 		if (ret < 0)
464 			goto out;
465 		if (ret)
466 			break;
467 	}
468 	btrfs_release_path(path);
469 
470 	/*
471 	 * pass 2: read all qgroup relations
472 	 */
473 	key.objectid = 0;
474 	key.type = BTRFS_QGROUP_RELATION_KEY;
475 	key.offset = 0;
476 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
477 	if (ret)
478 		goto out;
479 	while (1) {
480 		slot = path->slots[0];
481 		l = path->nodes[0];
482 		btrfs_item_key_to_cpu(l, &found_key, slot);
483 
484 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
485 			goto next2;
486 
487 		if (found_key.objectid > found_key.offset) {
488 			/* parent <- member, not needed to build config */
489 			/* FIXME should we omit the key completely? */
490 			goto next2;
491 		}
492 
493 		ret = add_relation_rb(fs_info, found_key.objectid,
494 				      found_key.offset);
495 		if (ret == -ENOENT) {
496 			btrfs_warn(fs_info,
497 				"orphan qgroup relation 0x%llx->0x%llx",
498 				found_key.objectid, found_key.offset);
499 			ret = 0;	/* ignore the error */
500 		}
501 		if (ret)
502 			goto out;
503 next2:
504 		ret = btrfs_next_item(quota_root, path);
505 		if (ret < 0)
506 			goto out;
507 		if (ret)
508 			break;
509 	}
510 out:
511 	btrfs_free_path(path);
512 	fs_info->qgroup_flags |= flags;
513 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
514 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
515 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
516 		 ret >= 0)
517 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
518 
519 	if (ret < 0) {
520 		ulist_free(fs_info->qgroup_ulist);
521 		fs_info->qgroup_ulist = NULL;
522 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
523 		btrfs_sysfs_del_qgroups(fs_info);
524 	}
525 
526 	return ret < 0 ? ret : 0;
527 }
528 
529 /*
530  * Called in close_ctree() when quota is still enabled.  This verifies we don't
531  * leak some reserved space.
532  *
533  * Return false if no reserved space is left.
534  * Return true if some reserved space is leaked.
535  */
536 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
537 {
538 	struct rb_node *node;
539 	bool ret = false;
540 
541 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
542 		return ret;
543 	/*
544 	 * Since we're unmounting, there is no race and no need to grab qgroup
545 	 * lock.  And here we don't go post-order to provide a more user
546 	 * friendly sorted result.
547 	 */
548 	for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
549 		struct btrfs_qgroup *qgroup;
550 		int i;
551 
552 		qgroup = rb_entry(node, struct btrfs_qgroup, node);
553 		for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
554 			if (qgroup->rsv.values[i]) {
555 				ret = true;
556 				btrfs_warn(fs_info,
557 		"qgroup %hu/%llu has unreleased space, type %d rsv %llu",
558 				   btrfs_qgroup_level(qgroup->qgroupid),
559 				   btrfs_qgroup_subvolid(qgroup->qgroupid),
560 				   i, qgroup->rsv.values[i]);
561 			}
562 		}
563 	}
564 	return ret;
565 }
566 
567 /*
568  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
569  * first two are in single-threaded paths.And for the third one, we have set
570  * quota_root to be null with qgroup_lock held before, so it is safe to clean
571  * up the in-memory structures without qgroup_lock held.
572  */
573 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
574 {
575 	struct rb_node *n;
576 	struct btrfs_qgroup *qgroup;
577 
578 	while ((n = rb_first(&fs_info->qgroup_tree))) {
579 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
580 		rb_erase(n, &fs_info->qgroup_tree);
581 		__del_qgroup_rb(fs_info, qgroup);
582 		btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
583 		kfree(qgroup);
584 	}
585 	/*
586 	 * We call btrfs_free_qgroup_config() when unmounting
587 	 * filesystem and disabling quota, so we set qgroup_ulist
588 	 * to be null here to avoid double free.
589 	 */
590 	ulist_free(fs_info->qgroup_ulist);
591 	fs_info->qgroup_ulist = NULL;
592 	btrfs_sysfs_del_qgroups(fs_info);
593 }
594 
595 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
596 				    u64 dst)
597 {
598 	int ret;
599 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
600 	struct btrfs_path *path;
601 	struct btrfs_key key;
602 
603 	path = btrfs_alloc_path();
604 	if (!path)
605 		return -ENOMEM;
606 
607 	key.objectid = src;
608 	key.type = BTRFS_QGROUP_RELATION_KEY;
609 	key.offset = dst;
610 
611 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
612 
613 	btrfs_mark_buffer_dirty(path->nodes[0]);
614 
615 	btrfs_free_path(path);
616 	return ret;
617 }
618 
619 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
620 				    u64 dst)
621 {
622 	int ret;
623 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
624 	struct btrfs_path *path;
625 	struct btrfs_key key;
626 
627 	path = btrfs_alloc_path();
628 	if (!path)
629 		return -ENOMEM;
630 
631 	key.objectid = src;
632 	key.type = BTRFS_QGROUP_RELATION_KEY;
633 	key.offset = dst;
634 
635 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
636 	if (ret < 0)
637 		goto out;
638 
639 	if (ret > 0) {
640 		ret = -ENOENT;
641 		goto out;
642 	}
643 
644 	ret = btrfs_del_item(trans, quota_root, path);
645 out:
646 	btrfs_free_path(path);
647 	return ret;
648 }
649 
650 static int add_qgroup_item(struct btrfs_trans_handle *trans,
651 			   struct btrfs_root *quota_root, u64 qgroupid)
652 {
653 	int ret;
654 	struct btrfs_path *path;
655 	struct btrfs_qgroup_info_item *qgroup_info;
656 	struct btrfs_qgroup_limit_item *qgroup_limit;
657 	struct extent_buffer *leaf;
658 	struct btrfs_key key;
659 
660 	if (btrfs_is_testing(quota_root->fs_info))
661 		return 0;
662 
663 	path = btrfs_alloc_path();
664 	if (!path)
665 		return -ENOMEM;
666 
667 	key.objectid = 0;
668 	key.type = BTRFS_QGROUP_INFO_KEY;
669 	key.offset = qgroupid;
670 
671 	/*
672 	 * Avoid a transaction abort by catching -EEXIST here. In that
673 	 * case, we proceed by re-initializing the existing structure
674 	 * on disk.
675 	 */
676 
677 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
678 				      sizeof(*qgroup_info));
679 	if (ret && ret != -EEXIST)
680 		goto out;
681 
682 	leaf = path->nodes[0];
683 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
684 				 struct btrfs_qgroup_info_item);
685 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
686 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
687 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
688 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
689 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
690 
691 	btrfs_mark_buffer_dirty(leaf);
692 
693 	btrfs_release_path(path);
694 
695 	key.type = BTRFS_QGROUP_LIMIT_KEY;
696 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
697 				      sizeof(*qgroup_limit));
698 	if (ret && ret != -EEXIST)
699 		goto out;
700 
701 	leaf = path->nodes[0];
702 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
703 				  struct btrfs_qgroup_limit_item);
704 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
705 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
706 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
707 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
708 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
709 
710 	btrfs_mark_buffer_dirty(leaf);
711 
712 	ret = 0;
713 out:
714 	btrfs_free_path(path);
715 	return ret;
716 }
717 
718 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
719 {
720 	int ret;
721 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
722 	struct btrfs_path *path;
723 	struct btrfs_key key;
724 
725 	path = btrfs_alloc_path();
726 	if (!path)
727 		return -ENOMEM;
728 
729 	key.objectid = 0;
730 	key.type = BTRFS_QGROUP_INFO_KEY;
731 	key.offset = qgroupid;
732 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
733 	if (ret < 0)
734 		goto out;
735 
736 	if (ret > 0) {
737 		ret = -ENOENT;
738 		goto out;
739 	}
740 
741 	ret = btrfs_del_item(trans, quota_root, path);
742 	if (ret)
743 		goto out;
744 
745 	btrfs_release_path(path);
746 
747 	key.type = BTRFS_QGROUP_LIMIT_KEY;
748 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
749 	if (ret < 0)
750 		goto out;
751 
752 	if (ret > 0) {
753 		ret = -ENOENT;
754 		goto out;
755 	}
756 
757 	ret = btrfs_del_item(trans, quota_root, path);
758 
759 out:
760 	btrfs_free_path(path);
761 	return ret;
762 }
763 
764 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
765 				    struct btrfs_qgroup *qgroup)
766 {
767 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
768 	struct btrfs_path *path;
769 	struct btrfs_key key;
770 	struct extent_buffer *l;
771 	struct btrfs_qgroup_limit_item *qgroup_limit;
772 	int ret;
773 	int slot;
774 
775 	key.objectid = 0;
776 	key.type = BTRFS_QGROUP_LIMIT_KEY;
777 	key.offset = qgroup->qgroupid;
778 
779 	path = btrfs_alloc_path();
780 	if (!path)
781 		return -ENOMEM;
782 
783 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
784 	if (ret > 0)
785 		ret = -ENOENT;
786 
787 	if (ret)
788 		goto out;
789 
790 	l = path->nodes[0];
791 	slot = path->slots[0];
792 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
793 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
794 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
795 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
796 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
797 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
798 
799 	btrfs_mark_buffer_dirty(l);
800 
801 out:
802 	btrfs_free_path(path);
803 	return ret;
804 }
805 
806 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
807 				   struct btrfs_qgroup *qgroup)
808 {
809 	struct btrfs_fs_info *fs_info = trans->fs_info;
810 	struct btrfs_root *quota_root = fs_info->quota_root;
811 	struct btrfs_path *path;
812 	struct btrfs_key key;
813 	struct extent_buffer *l;
814 	struct btrfs_qgroup_info_item *qgroup_info;
815 	int ret;
816 	int slot;
817 
818 	if (btrfs_is_testing(fs_info))
819 		return 0;
820 
821 	key.objectid = 0;
822 	key.type = BTRFS_QGROUP_INFO_KEY;
823 	key.offset = qgroup->qgroupid;
824 
825 	path = btrfs_alloc_path();
826 	if (!path)
827 		return -ENOMEM;
828 
829 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
830 	if (ret > 0)
831 		ret = -ENOENT;
832 
833 	if (ret)
834 		goto out;
835 
836 	l = path->nodes[0];
837 	slot = path->slots[0];
838 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
839 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
840 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
841 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
842 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
843 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
844 
845 	btrfs_mark_buffer_dirty(l);
846 
847 out:
848 	btrfs_free_path(path);
849 	return ret;
850 }
851 
852 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
853 {
854 	struct btrfs_fs_info *fs_info = trans->fs_info;
855 	struct btrfs_root *quota_root = fs_info->quota_root;
856 	struct btrfs_path *path;
857 	struct btrfs_key key;
858 	struct extent_buffer *l;
859 	struct btrfs_qgroup_status_item *ptr;
860 	int ret;
861 	int slot;
862 
863 	key.objectid = 0;
864 	key.type = BTRFS_QGROUP_STATUS_KEY;
865 	key.offset = 0;
866 
867 	path = btrfs_alloc_path();
868 	if (!path)
869 		return -ENOMEM;
870 
871 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
872 	if (ret > 0)
873 		ret = -ENOENT;
874 
875 	if (ret)
876 		goto out;
877 
878 	l = path->nodes[0];
879 	slot = path->slots[0];
880 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
881 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
882 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
883 	btrfs_set_qgroup_status_rescan(l, ptr,
884 				fs_info->qgroup_rescan_progress.objectid);
885 
886 	btrfs_mark_buffer_dirty(l);
887 
888 out:
889 	btrfs_free_path(path);
890 	return ret;
891 }
892 
893 /*
894  * called with qgroup_lock held
895  */
896 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
897 				  struct btrfs_root *root)
898 {
899 	struct btrfs_path *path;
900 	struct btrfs_key key;
901 	struct extent_buffer *leaf = NULL;
902 	int ret;
903 	int nr = 0;
904 
905 	path = btrfs_alloc_path();
906 	if (!path)
907 		return -ENOMEM;
908 
909 	key.objectid = 0;
910 	key.offset = 0;
911 	key.type = 0;
912 
913 	while (1) {
914 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
915 		if (ret < 0)
916 			goto out;
917 		leaf = path->nodes[0];
918 		nr = btrfs_header_nritems(leaf);
919 		if (!nr)
920 			break;
921 		/*
922 		 * delete the leaf one by one
923 		 * since the whole tree is going
924 		 * to be deleted.
925 		 */
926 		path->slots[0] = 0;
927 		ret = btrfs_del_items(trans, root, path, 0, nr);
928 		if (ret)
929 			goto out;
930 
931 		btrfs_release_path(path);
932 	}
933 	ret = 0;
934 out:
935 	btrfs_free_path(path);
936 	return ret;
937 }
938 
939 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
940 {
941 	struct btrfs_root *quota_root;
942 	struct btrfs_root *tree_root = fs_info->tree_root;
943 	struct btrfs_path *path = NULL;
944 	struct btrfs_qgroup_status_item *ptr;
945 	struct extent_buffer *leaf;
946 	struct btrfs_key key;
947 	struct btrfs_key found_key;
948 	struct btrfs_qgroup *qgroup = NULL;
949 	struct btrfs_trans_handle *trans = NULL;
950 	struct ulist *ulist = NULL;
951 	int ret = 0;
952 	int slot;
953 
954 	/*
955 	 * We need to have subvol_sem write locked, to prevent races between
956 	 * concurrent tasks trying to enable quotas, because we will unlock
957 	 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
958 	 * and before setting BTRFS_FS_QUOTA_ENABLED.
959 	 */
960 	lockdep_assert_held_write(&fs_info->subvol_sem);
961 
962 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
963 		btrfs_err(fs_info,
964 			  "qgroups are currently unsupported in extent tree v2");
965 		return -EINVAL;
966 	}
967 
968 	mutex_lock(&fs_info->qgroup_ioctl_lock);
969 	if (fs_info->quota_root)
970 		goto out;
971 
972 	ulist = ulist_alloc(GFP_KERNEL);
973 	if (!ulist) {
974 		ret = -ENOMEM;
975 		goto out;
976 	}
977 
978 	ret = btrfs_sysfs_add_qgroups(fs_info);
979 	if (ret < 0)
980 		goto out;
981 
982 	/*
983 	 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
984 	 * avoid lock acquisition inversion problems (reported by lockdep) between
985 	 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
986 	 * start a transaction.
987 	 * After we started the transaction lock qgroup_ioctl_lock again and
988 	 * check if someone else created the quota root in the meanwhile. If so,
989 	 * just return success and release the transaction handle.
990 	 *
991 	 * Also we don't need to worry about someone else calling
992 	 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
993 	 * that function returns 0 (success) when the sysfs entries already exist.
994 	 */
995 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
996 
997 	/*
998 	 * 1 for quota root item
999 	 * 1 for BTRFS_QGROUP_STATUS item
1000 	 *
1001 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1002 	 * per subvolume. However those are not currently reserved since it
1003 	 * would be a lot of overkill.
1004 	 */
1005 	trans = btrfs_start_transaction(tree_root, 2);
1006 
1007 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1008 	if (IS_ERR(trans)) {
1009 		ret = PTR_ERR(trans);
1010 		trans = NULL;
1011 		goto out;
1012 	}
1013 
1014 	if (fs_info->quota_root)
1015 		goto out;
1016 
1017 	fs_info->qgroup_ulist = ulist;
1018 	ulist = NULL;
1019 
1020 	/*
1021 	 * initially create the quota tree
1022 	 */
1023 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1024 	if (IS_ERR(quota_root)) {
1025 		ret =  PTR_ERR(quota_root);
1026 		btrfs_abort_transaction(trans, ret);
1027 		goto out;
1028 	}
1029 
1030 	path = btrfs_alloc_path();
1031 	if (!path) {
1032 		ret = -ENOMEM;
1033 		btrfs_abort_transaction(trans, ret);
1034 		goto out_free_root;
1035 	}
1036 
1037 	key.objectid = 0;
1038 	key.type = BTRFS_QGROUP_STATUS_KEY;
1039 	key.offset = 0;
1040 
1041 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1042 				      sizeof(*ptr));
1043 	if (ret) {
1044 		btrfs_abort_transaction(trans, ret);
1045 		goto out_free_path;
1046 	}
1047 
1048 	leaf = path->nodes[0];
1049 	ptr = btrfs_item_ptr(leaf, path->slots[0],
1050 				 struct btrfs_qgroup_status_item);
1051 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1052 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1053 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1054 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1055 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1056 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1057 
1058 	btrfs_mark_buffer_dirty(leaf);
1059 
1060 	key.objectid = 0;
1061 	key.type = BTRFS_ROOT_REF_KEY;
1062 	key.offset = 0;
1063 
1064 	btrfs_release_path(path);
1065 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1066 	if (ret > 0)
1067 		goto out_add_root;
1068 	if (ret < 0) {
1069 		btrfs_abort_transaction(trans, ret);
1070 		goto out_free_path;
1071 	}
1072 
1073 	while (1) {
1074 		slot = path->slots[0];
1075 		leaf = path->nodes[0];
1076 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1077 
1078 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
1079 
1080 			/* Release locks on tree_root before we access quota_root */
1081 			btrfs_release_path(path);
1082 
1083 			ret = add_qgroup_item(trans, quota_root,
1084 					      found_key.offset);
1085 			if (ret) {
1086 				btrfs_abort_transaction(trans, ret);
1087 				goto out_free_path;
1088 			}
1089 
1090 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
1091 			if (IS_ERR(qgroup)) {
1092 				ret = PTR_ERR(qgroup);
1093 				btrfs_abort_transaction(trans, ret);
1094 				goto out_free_path;
1095 			}
1096 			ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1097 			if (ret < 0) {
1098 				btrfs_abort_transaction(trans, ret);
1099 				goto out_free_path;
1100 			}
1101 			ret = btrfs_search_slot_for_read(tree_root, &found_key,
1102 							 path, 1, 0);
1103 			if (ret < 0) {
1104 				btrfs_abort_transaction(trans, ret);
1105 				goto out_free_path;
1106 			}
1107 			if (ret > 0) {
1108 				/*
1109 				 * Shouldn't happen, but in case it does we
1110 				 * don't need to do the btrfs_next_item, just
1111 				 * continue.
1112 				 */
1113 				continue;
1114 			}
1115 		}
1116 		ret = btrfs_next_item(tree_root, path);
1117 		if (ret < 0) {
1118 			btrfs_abort_transaction(trans, ret);
1119 			goto out_free_path;
1120 		}
1121 		if (ret)
1122 			break;
1123 	}
1124 
1125 out_add_root:
1126 	btrfs_release_path(path);
1127 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1128 	if (ret) {
1129 		btrfs_abort_transaction(trans, ret);
1130 		goto out_free_path;
1131 	}
1132 
1133 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1134 	if (IS_ERR(qgroup)) {
1135 		ret = PTR_ERR(qgroup);
1136 		btrfs_abort_transaction(trans, ret);
1137 		goto out_free_path;
1138 	}
1139 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1140 	if (ret < 0) {
1141 		btrfs_abort_transaction(trans, ret);
1142 		goto out_free_path;
1143 	}
1144 
1145 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1146 	/*
1147 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1148 	 * a deadlock with tasks concurrently doing other qgroup operations, such
1149 	 * adding/removing qgroups or adding/deleting qgroup relations for example,
1150 	 * because all qgroup operations first start or join a transaction and then
1151 	 * lock the qgroup_ioctl_lock mutex.
1152 	 * We are safe from a concurrent task trying to enable quotas, by calling
1153 	 * this function, since we are serialized by fs_info->subvol_sem.
1154 	 */
1155 	ret = btrfs_commit_transaction(trans);
1156 	trans = NULL;
1157 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1158 	if (ret)
1159 		goto out_free_path;
1160 
1161 	/*
1162 	 * Set quota enabled flag after committing the transaction, to avoid
1163 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1164 	 * creation.
1165 	 */
1166 	spin_lock(&fs_info->qgroup_lock);
1167 	fs_info->quota_root = quota_root;
1168 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1169 	spin_unlock(&fs_info->qgroup_lock);
1170 
1171 	ret = qgroup_rescan_init(fs_info, 0, 1);
1172 	if (!ret) {
1173 	        qgroup_rescan_zero_tracking(fs_info);
1174 		fs_info->qgroup_rescan_running = true;
1175 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1176 	                         &fs_info->qgroup_rescan_work);
1177 	}
1178 
1179 out_free_path:
1180 	btrfs_free_path(path);
1181 out_free_root:
1182 	if (ret)
1183 		btrfs_put_root(quota_root);
1184 out:
1185 	if (ret) {
1186 		ulist_free(fs_info->qgroup_ulist);
1187 		fs_info->qgroup_ulist = NULL;
1188 		btrfs_sysfs_del_qgroups(fs_info);
1189 	}
1190 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1191 	if (ret && trans)
1192 		btrfs_end_transaction(trans);
1193 	else if (trans)
1194 		ret = btrfs_end_transaction(trans);
1195 	ulist_free(ulist);
1196 	return ret;
1197 }
1198 
1199 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1200 {
1201 	struct btrfs_root *quota_root;
1202 	struct btrfs_trans_handle *trans = NULL;
1203 	int ret = 0;
1204 
1205 	/*
1206 	 * We need to have subvol_sem write locked, to prevent races between
1207 	 * concurrent tasks trying to disable quotas, because we will unlock
1208 	 * and relock qgroup_ioctl_lock across BTRFS_FS_QUOTA_ENABLED changes.
1209 	 */
1210 	lockdep_assert_held_write(&fs_info->subvol_sem);
1211 
1212 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1213 	if (!fs_info->quota_root)
1214 		goto out;
1215 
1216 	/*
1217 	 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1218 	 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1219 	 * to lock that mutex while holding a transaction handle and the rescan
1220 	 * worker needs to commit a transaction.
1221 	 */
1222 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1223 
1224 	/*
1225 	 * Request qgroup rescan worker to complete and wait for it. This wait
1226 	 * must be done before transaction start for quota disable since it may
1227 	 * deadlock with transaction by the qgroup rescan worker.
1228 	 */
1229 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1230 	btrfs_qgroup_wait_for_completion(fs_info, false);
1231 
1232 	/*
1233 	 * 1 For the root item
1234 	 *
1235 	 * We should also reserve enough items for the quota tree deletion in
1236 	 * btrfs_clean_quota_tree but this is not done.
1237 	 *
1238 	 * Also, we must always start a transaction without holding the mutex
1239 	 * qgroup_ioctl_lock, see btrfs_quota_enable().
1240 	 */
1241 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1242 
1243 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1244 	if (IS_ERR(trans)) {
1245 		ret = PTR_ERR(trans);
1246 		trans = NULL;
1247 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1248 		goto out;
1249 	}
1250 
1251 	if (!fs_info->quota_root)
1252 		goto out;
1253 
1254 	spin_lock(&fs_info->qgroup_lock);
1255 	quota_root = fs_info->quota_root;
1256 	fs_info->quota_root = NULL;
1257 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1258 	spin_unlock(&fs_info->qgroup_lock);
1259 
1260 	btrfs_free_qgroup_config(fs_info);
1261 
1262 	ret = btrfs_clean_quota_tree(trans, quota_root);
1263 	if (ret) {
1264 		btrfs_abort_transaction(trans, ret);
1265 		goto out;
1266 	}
1267 
1268 	ret = btrfs_del_root(trans, &quota_root->root_key);
1269 	if (ret) {
1270 		btrfs_abort_transaction(trans, ret);
1271 		goto out;
1272 	}
1273 
1274 	list_del(&quota_root->dirty_list);
1275 
1276 	btrfs_tree_lock(quota_root->node);
1277 	btrfs_clean_tree_block(quota_root->node);
1278 	btrfs_tree_unlock(quota_root->node);
1279 	btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1280 			      quota_root->node, 0, 1);
1281 
1282 	btrfs_put_root(quota_root);
1283 
1284 out:
1285 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1286 	if (ret && trans)
1287 		btrfs_end_transaction(trans);
1288 	else if (trans)
1289 		ret = btrfs_end_transaction(trans);
1290 
1291 	return ret;
1292 }
1293 
1294 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1295 			 struct btrfs_qgroup *qgroup)
1296 {
1297 	if (list_empty(&qgroup->dirty))
1298 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1299 }
1300 
1301 /*
1302  * The easy accounting, we're updating qgroup relationship whose child qgroup
1303  * only has exclusive extents.
1304  *
1305  * In this case, all exclusive extents will also be exclusive for parent, so
1306  * excl/rfer just get added/removed.
1307  *
1308  * So is qgroup reservation space, which should also be added/removed to
1309  * parent.
1310  * Or when child tries to release reservation space, parent will underflow its
1311  * reservation (for relationship adding case).
1312  *
1313  * Caller should hold fs_info->qgroup_lock.
1314  */
1315 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1316 				    struct ulist *tmp, u64 ref_root,
1317 				    struct btrfs_qgroup *src, int sign)
1318 {
1319 	struct btrfs_qgroup *qgroup;
1320 	struct btrfs_qgroup_list *glist;
1321 	struct ulist_node *unode;
1322 	struct ulist_iterator uiter;
1323 	u64 num_bytes = src->excl;
1324 	int ret = 0;
1325 
1326 	qgroup = find_qgroup_rb(fs_info, ref_root);
1327 	if (!qgroup)
1328 		goto out;
1329 
1330 	qgroup->rfer += sign * num_bytes;
1331 	qgroup->rfer_cmpr += sign * num_bytes;
1332 
1333 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1334 	qgroup->excl += sign * num_bytes;
1335 	qgroup->excl_cmpr += sign * num_bytes;
1336 
1337 	if (sign > 0)
1338 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1339 	else
1340 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1341 
1342 	qgroup_dirty(fs_info, qgroup);
1343 
1344 	/* Get all of the parent groups that contain this qgroup */
1345 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1346 		ret = ulist_add(tmp, glist->group->qgroupid,
1347 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1348 		if (ret < 0)
1349 			goto out;
1350 	}
1351 
1352 	/* Iterate all of the parents and adjust their reference counts */
1353 	ULIST_ITER_INIT(&uiter);
1354 	while ((unode = ulist_next(tmp, &uiter))) {
1355 		qgroup = unode_aux_to_qgroup(unode);
1356 		qgroup->rfer += sign * num_bytes;
1357 		qgroup->rfer_cmpr += sign * num_bytes;
1358 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1359 		qgroup->excl += sign * num_bytes;
1360 		if (sign > 0)
1361 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1362 		else
1363 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1364 		qgroup->excl_cmpr += sign * num_bytes;
1365 		qgroup_dirty(fs_info, qgroup);
1366 
1367 		/* Add any parents of the parents */
1368 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1369 			ret = ulist_add(tmp, glist->group->qgroupid,
1370 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1371 			if (ret < 0)
1372 				goto out;
1373 		}
1374 	}
1375 	ret = 0;
1376 out:
1377 	return ret;
1378 }
1379 
1380 
1381 /*
1382  * Quick path for updating qgroup with only excl refs.
1383  *
1384  * In that case, just update all parent will be enough.
1385  * Or we needs to do a full rescan.
1386  * Caller should also hold fs_info->qgroup_lock.
1387  *
1388  * Return 0 for quick update, return >0 for need to full rescan
1389  * and mark INCONSISTENT flag.
1390  * Return < 0 for other error.
1391  */
1392 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1393 				   struct ulist *tmp, u64 src, u64 dst,
1394 				   int sign)
1395 {
1396 	struct btrfs_qgroup *qgroup;
1397 	int ret = 1;
1398 	int err = 0;
1399 
1400 	qgroup = find_qgroup_rb(fs_info, src);
1401 	if (!qgroup)
1402 		goto out;
1403 	if (qgroup->excl == qgroup->rfer) {
1404 		ret = 0;
1405 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1406 					       qgroup, sign);
1407 		if (err < 0) {
1408 			ret = err;
1409 			goto out;
1410 		}
1411 	}
1412 out:
1413 	if (ret)
1414 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1415 	return ret;
1416 }
1417 
1418 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1419 			      u64 dst)
1420 {
1421 	struct btrfs_fs_info *fs_info = trans->fs_info;
1422 	struct btrfs_qgroup *parent;
1423 	struct btrfs_qgroup *member;
1424 	struct btrfs_qgroup_list *list;
1425 	struct ulist *tmp;
1426 	unsigned int nofs_flag;
1427 	int ret = 0;
1428 
1429 	/* Check the level of src and dst first */
1430 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1431 		return -EINVAL;
1432 
1433 	/* We hold a transaction handle open, must do a NOFS allocation. */
1434 	nofs_flag = memalloc_nofs_save();
1435 	tmp = ulist_alloc(GFP_KERNEL);
1436 	memalloc_nofs_restore(nofs_flag);
1437 	if (!tmp)
1438 		return -ENOMEM;
1439 
1440 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1441 	if (!fs_info->quota_root) {
1442 		ret = -ENOTCONN;
1443 		goto out;
1444 	}
1445 	member = find_qgroup_rb(fs_info, src);
1446 	parent = find_qgroup_rb(fs_info, dst);
1447 	if (!member || !parent) {
1448 		ret = -EINVAL;
1449 		goto out;
1450 	}
1451 
1452 	/* check if such qgroup relation exist firstly */
1453 	list_for_each_entry(list, &member->groups, next_group) {
1454 		if (list->group == parent) {
1455 			ret = -EEXIST;
1456 			goto out;
1457 		}
1458 	}
1459 
1460 	ret = add_qgroup_relation_item(trans, src, dst);
1461 	if (ret)
1462 		goto out;
1463 
1464 	ret = add_qgroup_relation_item(trans, dst, src);
1465 	if (ret) {
1466 		del_qgroup_relation_item(trans, src, dst);
1467 		goto out;
1468 	}
1469 
1470 	spin_lock(&fs_info->qgroup_lock);
1471 	ret = __add_relation_rb(member, parent);
1472 	if (ret < 0) {
1473 		spin_unlock(&fs_info->qgroup_lock);
1474 		goto out;
1475 	}
1476 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1477 	spin_unlock(&fs_info->qgroup_lock);
1478 out:
1479 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1480 	ulist_free(tmp);
1481 	return ret;
1482 }
1483 
1484 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1485 				 u64 dst)
1486 {
1487 	struct btrfs_fs_info *fs_info = trans->fs_info;
1488 	struct btrfs_qgroup *parent;
1489 	struct btrfs_qgroup *member;
1490 	struct btrfs_qgroup_list *list;
1491 	struct ulist *tmp;
1492 	bool found = false;
1493 	unsigned int nofs_flag;
1494 	int ret = 0;
1495 	int ret2;
1496 
1497 	/* We hold a transaction handle open, must do a NOFS allocation. */
1498 	nofs_flag = memalloc_nofs_save();
1499 	tmp = ulist_alloc(GFP_KERNEL);
1500 	memalloc_nofs_restore(nofs_flag);
1501 	if (!tmp)
1502 		return -ENOMEM;
1503 
1504 	if (!fs_info->quota_root) {
1505 		ret = -ENOTCONN;
1506 		goto out;
1507 	}
1508 
1509 	member = find_qgroup_rb(fs_info, src);
1510 	parent = find_qgroup_rb(fs_info, dst);
1511 	/*
1512 	 * The parent/member pair doesn't exist, then try to delete the dead
1513 	 * relation items only.
1514 	 */
1515 	if (!member || !parent)
1516 		goto delete_item;
1517 
1518 	/* check if such qgroup relation exist firstly */
1519 	list_for_each_entry(list, &member->groups, next_group) {
1520 		if (list->group == parent) {
1521 			found = true;
1522 			break;
1523 		}
1524 	}
1525 
1526 delete_item:
1527 	ret = del_qgroup_relation_item(trans, src, dst);
1528 	if (ret < 0 && ret != -ENOENT)
1529 		goto out;
1530 	ret2 = del_qgroup_relation_item(trans, dst, src);
1531 	if (ret2 < 0 && ret2 != -ENOENT)
1532 		goto out;
1533 
1534 	/* At least one deletion succeeded, return 0 */
1535 	if (!ret || !ret2)
1536 		ret = 0;
1537 
1538 	if (found) {
1539 		spin_lock(&fs_info->qgroup_lock);
1540 		del_relation_rb(fs_info, src, dst);
1541 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1542 		spin_unlock(&fs_info->qgroup_lock);
1543 	}
1544 out:
1545 	ulist_free(tmp);
1546 	return ret;
1547 }
1548 
1549 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1550 			      u64 dst)
1551 {
1552 	struct btrfs_fs_info *fs_info = trans->fs_info;
1553 	int ret = 0;
1554 
1555 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1556 	ret = __del_qgroup_relation(trans, src, dst);
1557 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1558 
1559 	return ret;
1560 }
1561 
1562 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1563 {
1564 	struct btrfs_fs_info *fs_info = trans->fs_info;
1565 	struct btrfs_root *quota_root;
1566 	struct btrfs_qgroup *qgroup;
1567 	int ret = 0;
1568 
1569 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1570 	if (!fs_info->quota_root) {
1571 		ret = -ENOTCONN;
1572 		goto out;
1573 	}
1574 	quota_root = fs_info->quota_root;
1575 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1576 	if (qgroup) {
1577 		ret = -EEXIST;
1578 		goto out;
1579 	}
1580 
1581 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1582 	if (ret)
1583 		goto out;
1584 
1585 	spin_lock(&fs_info->qgroup_lock);
1586 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1587 	spin_unlock(&fs_info->qgroup_lock);
1588 
1589 	if (IS_ERR(qgroup)) {
1590 		ret = PTR_ERR(qgroup);
1591 		goto out;
1592 	}
1593 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1594 out:
1595 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1596 	return ret;
1597 }
1598 
1599 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1600 {
1601 	struct btrfs_fs_info *fs_info = trans->fs_info;
1602 	struct btrfs_qgroup *qgroup;
1603 	struct btrfs_qgroup_list *list;
1604 	int ret = 0;
1605 
1606 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1607 	if (!fs_info->quota_root) {
1608 		ret = -ENOTCONN;
1609 		goto out;
1610 	}
1611 
1612 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1613 	if (!qgroup) {
1614 		ret = -ENOENT;
1615 		goto out;
1616 	}
1617 
1618 	/* Check if there are no children of this qgroup */
1619 	if (!list_empty(&qgroup->members)) {
1620 		ret = -EBUSY;
1621 		goto out;
1622 	}
1623 
1624 	ret = del_qgroup_item(trans, qgroupid);
1625 	if (ret && ret != -ENOENT)
1626 		goto out;
1627 
1628 	while (!list_empty(&qgroup->groups)) {
1629 		list = list_first_entry(&qgroup->groups,
1630 					struct btrfs_qgroup_list, next_group);
1631 		ret = __del_qgroup_relation(trans, qgroupid,
1632 					    list->group->qgroupid);
1633 		if (ret)
1634 			goto out;
1635 	}
1636 
1637 	spin_lock(&fs_info->qgroup_lock);
1638 	del_qgroup_rb(fs_info, qgroupid);
1639 	spin_unlock(&fs_info->qgroup_lock);
1640 
1641 	/*
1642 	 * Remove the qgroup from sysfs now without holding the qgroup_lock
1643 	 * spinlock, since the sysfs_remove_group() function needs to take
1644 	 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1645 	 */
1646 	btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1647 	kfree(qgroup);
1648 out:
1649 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1650 	return ret;
1651 }
1652 
1653 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1654 		       struct btrfs_qgroup_limit *limit)
1655 {
1656 	struct btrfs_fs_info *fs_info = trans->fs_info;
1657 	struct btrfs_qgroup *qgroup;
1658 	int ret = 0;
1659 	/* Sometimes we would want to clear the limit on this qgroup.
1660 	 * To meet this requirement, we treat the -1 as a special value
1661 	 * which tell kernel to clear the limit on this qgroup.
1662 	 */
1663 	const u64 CLEAR_VALUE = -1;
1664 
1665 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1666 	if (!fs_info->quota_root) {
1667 		ret = -ENOTCONN;
1668 		goto out;
1669 	}
1670 
1671 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1672 	if (!qgroup) {
1673 		ret = -ENOENT;
1674 		goto out;
1675 	}
1676 
1677 	spin_lock(&fs_info->qgroup_lock);
1678 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1679 		if (limit->max_rfer == CLEAR_VALUE) {
1680 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1681 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1682 			qgroup->max_rfer = 0;
1683 		} else {
1684 			qgroup->max_rfer = limit->max_rfer;
1685 		}
1686 	}
1687 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1688 		if (limit->max_excl == CLEAR_VALUE) {
1689 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1690 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1691 			qgroup->max_excl = 0;
1692 		} else {
1693 			qgroup->max_excl = limit->max_excl;
1694 		}
1695 	}
1696 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1697 		if (limit->rsv_rfer == CLEAR_VALUE) {
1698 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1699 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1700 			qgroup->rsv_rfer = 0;
1701 		} else {
1702 			qgroup->rsv_rfer = limit->rsv_rfer;
1703 		}
1704 	}
1705 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1706 		if (limit->rsv_excl == CLEAR_VALUE) {
1707 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1708 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1709 			qgroup->rsv_excl = 0;
1710 		} else {
1711 			qgroup->rsv_excl = limit->rsv_excl;
1712 		}
1713 	}
1714 	qgroup->lim_flags |= limit->flags;
1715 
1716 	spin_unlock(&fs_info->qgroup_lock);
1717 
1718 	ret = update_qgroup_limit_item(trans, qgroup);
1719 	if (ret) {
1720 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1721 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1722 		       qgroupid);
1723 	}
1724 
1725 out:
1726 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1727 	return ret;
1728 }
1729 
1730 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1731 				struct btrfs_delayed_ref_root *delayed_refs,
1732 				struct btrfs_qgroup_extent_record *record)
1733 {
1734 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1735 	struct rb_node *parent_node = NULL;
1736 	struct btrfs_qgroup_extent_record *entry;
1737 	u64 bytenr = record->bytenr;
1738 
1739 	lockdep_assert_held(&delayed_refs->lock);
1740 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1741 
1742 	while (*p) {
1743 		parent_node = *p;
1744 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1745 				 node);
1746 		if (bytenr < entry->bytenr) {
1747 			p = &(*p)->rb_left;
1748 		} else if (bytenr > entry->bytenr) {
1749 			p = &(*p)->rb_right;
1750 		} else {
1751 			if (record->data_rsv && !entry->data_rsv) {
1752 				entry->data_rsv = record->data_rsv;
1753 				entry->data_rsv_refroot =
1754 					record->data_rsv_refroot;
1755 			}
1756 			return 1;
1757 		}
1758 	}
1759 
1760 	rb_link_node(&record->node, parent_node, p);
1761 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1762 	return 0;
1763 }
1764 
1765 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1766 				   struct btrfs_qgroup_extent_record *qrecord)
1767 {
1768 	struct ulist *old_root;
1769 	u64 bytenr = qrecord->bytenr;
1770 	int ret;
1771 
1772 	/*
1773 	 * We are always called in a context where we are already holding a
1774 	 * transaction handle. Often we are called when adding a data delayed
1775 	 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1776 	 * in which case we will be holding a write lock on extent buffer from a
1777 	 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1778 	 * acquire fs_info->commit_root_sem, because that is a higher level lock
1779 	 * that must be acquired before locking any extent buffers.
1780 	 *
1781 	 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1782 	 * but we can't pass it a non-NULL transaction handle, because otherwise
1783 	 * it would not use commit roots and would lock extent buffers, causing
1784 	 * a deadlock if it ends up trying to read lock the same extent buffer
1785 	 * that was previously write locked at btrfs_truncate_inode_items().
1786 	 *
1787 	 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1788 	 * explicitly tell it to not acquire the commit_root_sem - if we are
1789 	 * holding a transaction handle we don't need its protection.
1790 	 */
1791 	ASSERT(trans != NULL);
1792 
1793 	ret = btrfs_find_all_roots(NULL, trans->fs_info, bytenr, 0, &old_root,
1794 				   true);
1795 	if (ret < 0) {
1796 		trans->fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1797 		btrfs_warn(trans->fs_info,
1798 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1799 			ret);
1800 		return 0;
1801 	}
1802 
1803 	/*
1804 	 * Here we don't need to get the lock of
1805 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1806 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1807 	 *
1808 	 * So modifying qrecord->old_roots is safe here
1809 	 */
1810 	qrecord->old_roots = old_root;
1811 	return 0;
1812 }
1813 
1814 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1815 			      u64 num_bytes, gfp_t gfp_flag)
1816 {
1817 	struct btrfs_fs_info *fs_info = trans->fs_info;
1818 	struct btrfs_qgroup_extent_record *record;
1819 	struct btrfs_delayed_ref_root *delayed_refs;
1820 	int ret;
1821 
1822 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1823 	    || bytenr == 0 || num_bytes == 0)
1824 		return 0;
1825 	record = kzalloc(sizeof(*record), gfp_flag);
1826 	if (!record)
1827 		return -ENOMEM;
1828 
1829 	delayed_refs = &trans->transaction->delayed_refs;
1830 	record->bytenr = bytenr;
1831 	record->num_bytes = num_bytes;
1832 	record->old_roots = NULL;
1833 
1834 	spin_lock(&delayed_refs->lock);
1835 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1836 	spin_unlock(&delayed_refs->lock);
1837 	if (ret > 0) {
1838 		kfree(record);
1839 		return 0;
1840 	}
1841 	return btrfs_qgroup_trace_extent_post(trans, record);
1842 }
1843 
1844 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1845 				  struct extent_buffer *eb)
1846 {
1847 	struct btrfs_fs_info *fs_info = trans->fs_info;
1848 	int nr = btrfs_header_nritems(eb);
1849 	int i, extent_type, ret;
1850 	struct btrfs_key key;
1851 	struct btrfs_file_extent_item *fi;
1852 	u64 bytenr, num_bytes;
1853 
1854 	/* We can be called directly from walk_up_proc() */
1855 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1856 		return 0;
1857 
1858 	for (i = 0; i < nr; i++) {
1859 		btrfs_item_key_to_cpu(eb, &key, i);
1860 
1861 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1862 			continue;
1863 
1864 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1865 		/* filter out non qgroup-accountable extents  */
1866 		extent_type = btrfs_file_extent_type(eb, fi);
1867 
1868 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1869 			continue;
1870 
1871 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1872 		if (!bytenr)
1873 			continue;
1874 
1875 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1876 
1877 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1878 						GFP_NOFS);
1879 		if (ret)
1880 			return ret;
1881 	}
1882 	cond_resched();
1883 	return 0;
1884 }
1885 
1886 /*
1887  * Walk up the tree from the bottom, freeing leaves and any interior
1888  * nodes which have had all slots visited. If a node (leaf or
1889  * interior) is freed, the node above it will have it's slot
1890  * incremented. The root node will never be freed.
1891  *
1892  * At the end of this function, we should have a path which has all
1893  * slots incremented to the next position for a search. If we need to
1894  * read a new node it will be NULL and the node above it will have the
1895  * correct slot selected for a later read.
1896  *
1897  * If we increment the root nodes slot counter past the number of
1898  * elements, 1 is returned to signal completion of the search.
1899  */
1900 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1901 {
1902 	int level = 0;
1903 	int nr, slot;
1904 	struct extent_buffer *eb;
1905 
1906 	if (root_level == 0)
1907 		return 1;
1908 
1909 	while (level <= root_level) {
1910 		eb = path->nodes[level];
1911 		nr = btrfs_header_nritems(eb);
1912 		path->slots[level]++;
1913 		slot = path->slots[level];
1914 		if (slot >= nr || level == 0) {
1915 			/*
1916 			 * Don't free the root -  we will detect this
1917 			 * condition after our loop and return a
1918 			 * positive value for caller to stop walking the tree.
1919 			 */
1920 			if (level != root_level) {
1921 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1922 				path->locks[level] = 0;
1923 
1924 				free_extent_buffer(eb);
1925 				path->nodes[level] = NULL;
1926 				path->slots[level] = 0;
1927 			}
1928 		} else {
1929 			/*
1930 			 * We have a valid slot to walk back down
1931 			 * from. Stop here so caller can process these
1932 			 * new nodes.
1933 			 */
1934 			break;
1935 		}
1936 
1937 		level++;
1938 	}
1939 
1940 	eb = path->nodes[root_level];
1941 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1942 		return 1;
1943 
1944 	return 0;
1945 }
1946 
1947 /*
1948  * Helper function to trace a subtree tree block swap.
1949  *
1950  * The swap will happen in highest tree block, but there may be a lot of
1951  * tree blocks involved.
1952  *
1953  * For example:
1954  *  OO = Old tree blocks
1955  *  NN = New tree blocks allocated during balance
1956  *
1957  *           File tree (257)                  Reloc tree for 257
1958  * L2              OO                                NN
1959  *               /    \                            /    \
1960  * L1          OO      OO (a)                    OO      NN (a)
1961  *            / \     / \                       / \     / \
1962  * L0       OO   OO OO   OO                   OO   OO NN   NN
1963  *                  (b)  (c)                          (b)  (c)
1964  *
1965  * When calling qgroup_trace_extent_swap(), we will pass:
1966  * @src_eb = OO(a)
1967  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1968  * @dst_level = 0
1969  * @root_level = 1
1970  *
1971  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1972  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1973  *
1974  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1975  *
1976  * 1) Tree search from @src_eb
1977  *    It should acts as a simplified btrfs_search_slot().
1978  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1979  *    (first key).
1980  *
1981  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1982  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1983  *    They should be marked during previous (@dst_level = 1) iteration.
1984  *
1985  * 3) Mark file extents in leaves dirty
1986  *    We don't have good way to pick out new file extents only.
1987  *    So we still follow the old method by scanning all file extents in
1988  *    the leave.
1989  *
1990  * This function can free us from keeping two paths, thus later we only need
1991  * to care about how to iterate all new tree blocks in reloc tree.
1992  */
1993 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1994 				    struct extent_buffer *src_eb,
1995 				    struct btrfs_path *dst_path,
1996 				    int dst_level, int root_level,
1997 				    bool trace_leaf)
1998 {
1999 	struct btrfs_key key;
2000 	struct btrfs_path *src_path;
2001 	struct btrfs_fs_info *fs_info = trans->fs_info;
2002 	u32 nodesize = fs_info->nodesize;
2003 	int cur_level = root_level;
2004 	int ret;
2005 
2006 	BUG_ON(dst_level > root_level);
2007 	/* Level mismatch */
2008 	if (btrfs_header_level(src_eb) != root_level)
2009 		return -EINVAL;
2010 
2011 	src_path = btrfs_alloc_path();
2012 	if (!src_path) {
2013 		ret = -ENOMEM;
2014 		goto out;
2015 	}
2016 
2017 	if (dst_level)
2018 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2019 	else
2020 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2021 
2022 	/* For src_path */
2023 	atomic_inc(&src_eb->refs);
2024 	src_path->nodes[root_level] = src_eb;
2025 	src_path->slots[root_level] = dst_path->slots[root_level];
2026 	src_path->locks[root_level] = 0;
2027 
2028 	/* A simplified version of btrfs_search_slot() */
2029 	while (cur_level >= dst_level) {
2030 		struct btrfs_key src_key;
2031 		struct btrfs_key dst_key;
2032 
2033 		if (src_path->nodes[cur_level] == NULL) {
2034 			struct extent_buffer *eb;
2035 			int parent_slot;
2036 
2037 			eb = src_path->nodes[cur_level + 1];
2038 			parent_slot = src_path->slots[cur_level + 1];
2039 
2040 			eb = btrfs_read_node_slot(eb, parent_slot);
2041 			if (IS_ERR(eb)) {
2042 				ret = PTR_ERR(eb);
2043 				goto out;
2044 			}
2045 
2046 			src_path->nodes[cur_level] = eb;
2047 
2048 			btrfs_tree_read_lock(eb);
2049 			src_path->locks[cur_level] = BTRFS_READ_LOCK;
2050 		}
2051 
2052 		src_path->slots[cur_level] = dst_path->slots[cur_level];
2053 		if (cur_level) {
2054 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2055 					&dst_key, dst_path->slots[cur_level]);
2056 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2057 					&src_key, src_path->slots[cur_level]);
2058 		} else {
2059 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2060 					&dst_key, dst_path->slots[cur_level]);
2061 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2062 					&src_key, src_path->slots[cur_level]);
2063 		}
2064 		/* Content mismatch, something went wrong */
2065 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2066 			ret = -ENOENT;
2067 			goto out;
2068 		}
2069 		cur_level--;
2070 	}
2071 
2072 	/*
2073 	 * Now both @dst_path and @src_path have been populated, record the tree
2074 	 * blocks for qgroup accounting.
2075 	 */
2076 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2077 			nodesize, GFP_NOFS);
2078 	if (ret < 0)
2079 		goto out;
2080 	ret = btrfs_qgroup_trace_extent(trans,
2081 			dst_path->nodes[dst_level]->start,
2082 			nodesize, GFP_NOFS);
2083 	if (ret < 0)
2084 		goto out;
2085 
2086 	/* Record leaf file extents */
2087 	if (dst_level == 0 && trace_leaf) {
2088 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2089 		if (ret < 0)
2090 			goto out;
2091 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2092 	}
2093 out:
2094 	btrfs_free_path(src_path);
2095 	return ret;
2096 }
2097 
2098 /*
2099  * Helper function to do recursive generation-aware depth-first search, to
2100  * locate all new tree blocks in a subtree of reloc tree.
2101  *
2102  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2103  *         reloc tree
2104  * L2         NN (a)
2105  *          /    \
2106  * L1    OO        NN (b)
2107  *      /  \      /  \
2108  * L0  OO  OO    OO  NN
2109  *               (c) (d)
2110  * If we pass:
2111  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2112  * @cur_level = 1
2113  * @root_level = 1
2114  *
2115  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2116  * above tree blocks along with their counter parts in file tree.
2117  * While during search, old tree blocks OO(c) will be skipped as tree block swap
2118  * won't affect OO(c).
2119  */
2120 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2121 					   struct extent_buffer *src_eb,
2122 					   struct btrfs_path *dst_path,
2123 					   int cur_level, int root_level,
2124 					   u64 last_snapshot, bool trace_leaf)
2125 {
2126 	struct btrfs_fs_info *fs_info = trans->fs_info;
2127 	struct extent_buffer *eb;
2128 	bool need_cleanup = false;
2129 	int ret = 0;
2130 	int i;
2131 
2132 	/* Level sanity check */
2133 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2134 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2135 	    root_level < cur_level) {
2136 		btrfs_err_rl(fs_info,
2137 			"%s: bad levels, cur_level=%d root_level=%d",
2138 			__func__, cur_level, root_level);
2139 		return -EUCLEAN;
2140 	}
2141 
2142 	/* Read the tree block if needed */
2143 	if (dst_path->nodes[cur_level] == NULL) {
2144 		int parent_slot;
2145 		u64 child_gen;
2146 
2147 		/*
2148 		 * dst_path->nodes[root_level] must be initialized before
2149 		 * calling this function.
2150 		 */
2151 		if (cur_level == root_level) {
2152 			btrfs_err_rl(fs_info,
2153 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2154 				__func__, root_level, root_level, cur_level);
2155 			return -EUCLEAN;
2156 		}
2157 
2158 		/*
2159 		 * We need to get child blockptr/gen from parent before we can
2160 		 * read it.
2161 		  */
2162 		eb = dst_path->nodes[cur_level + 1];
2163 		parent_slot = dst_path->slots[cur_level + 1];
2164 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2165 
2166 		/* This node is old, no need to trace */
2167 		if (child_gen < last_snapshot)
2168 			goto out;
2169 
2170 		eb = btrfs_read_node_slot(eb, parent_slot);
2171 		if (IS_ERR(eb)) {
2172 			ret = PTR_ERR(eb);
2173 			goto out;
2174 		}
2175 
2176 		dst_path->nodes[cur_level] = eb;
2177 		dst_path->slots[cur_level] = 0;
2178 
2179 		btrfs_tree_read_lock(eb);
2180 		dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2181 		need_cleanup = true;
2182 	}
2183 
2184 	/* Now record this tree block and its counter part for qgroups */
2185 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2186 				       root_level, trace_leaf);
2187 	if (ret < 0)
2188 		goto cleanup;
2189 
2190 	eb = dst_path->nodes[cur_level];
2191 
2192 	if (cur_level > 0) {
2193 		/* Iterate all child tree blocks */
2194 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2195 			/* Skip old tree blocks as they won't be swapped */
2196 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2197 				continue;
2198 			dst_path->slots[cur_level] = i;
2199 
2200 			/* Recursive call (at most 7 times) */
2201 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2202 					dst_path, cur_level - 1, root_level,
2203 					last_snapshot, trace_leaf);
2204 			if (ret < 0)
2205 				goto cleanup;
2206 		}
2207 	}
2208 
2209 cleanup:
2210 	if (need_cleanup) {
2211 		/* Clean up */
2212 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2213 				     dst_path->locks[cur_level]);
2214 		free_extent_buffer(dst_path->nodes[cur_level]);
2215 		dst_path->nodes[cur_level] = NULL;
2216 		dst_path->slots[cur_level] = 0;
2217 		dst_path->locks[cur_level] = 0;
2218 	}
2219 out:
2220 	return ret;
2221 }
2222 
2223 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2224 				struct extent_buffer *src_eb,
2225 				struct extent_buffer *dst_eb,
2226 				u64 last_snapshot, bool trace_leaf)
2227 {
2228 	struct btrfs_fs_info *fs_info = trans->fs_info;
2229 	struct btrfs_path *dst_path = NULL;
2230 	int level;
2231 	int ret;
2232 
2233 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2234 		return 0;
2235 
2236 	/* Wrong parameter order */
2237 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2238 		btrfs_err_rl(fs_info,
2239 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2240 			     btrfs_header_generation(src_eb),
2241 			     btrfs_header_generation(dst_eb));
2242 		return -EUCLEAN;
2243 	}
2244 
2245 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2246 		ret = -EIO;
2247 		goto out;
2248 	}
2249 
2250 	level = btrfs_header_level(dst_eb);
2251 	dst_path = btrfs_alloc_path();
2252 	if (!dst_path) {
2253 		ret = -ENOMEM;
2254 		goto out;
2255 	}
2256 	/* For dst_path */
2257 	atomic_inc(&dst_eb->refs);
2258 	dst_path->nodes[level] = dst_eb;
2259 	dst_path->slots[level] = 0;
2260 	dst_path->locks[level] = 0;
2261 
2262 	/* Do the generation aware breadth-first search */
2263 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2264 					      level, last_snapshot, trace_leaf);
2265 	if (ret < 0)
2266 		goto out;
2267 	ret = 0;
2268 
2269 out:
2270 	btrfs_free_path(dst_path);
2271 	if (ret < 0)
2272 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2273 	return ret;
2274 }
2275 
2276 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2277 			       struct extent_buffer *root_eb,
2278 			       u64 root_gen, int root_level)
2279 {
2280 	struct btrfs_fs_info *fs_info = trans->fs_info;
2281 	int ret = 0;
2282 	int level;
2283 	struct extent_buffer *eb = root_eb;
2284 	struct btrfs_path *path = NULL;
2285 
2286 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2287 	BUG_ON(root_eb == NULL);
2288 
2289 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2290 		return 0;
2291 
2292 	if (!extent_buffer_uptodate(root_eb)) {
2293 		ret = btrfs_read_extent_buffer(root_eb, root_gen, root_level, NULL);
2294 		if (ret)
2295 			goto out;
2296 	}
2297 
2298 	if (root_level == 0) {
2299 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2300 		goto out;
2301 	}
2302 
2303 	path = btrfs_alloc_path();
2304 	if (!path)
2305 		return -ENOMEM;
2306 
2307 	/*
2308 	 * Walk down the tree.  Missing extent blocks are filled in as
2309 	 * we go. Metadata is accounted every time we read a new
2310 	 * extent block.
2311 	 *
2312 	 * When we reach a leaf, we account for file extent items in it,
2313 	 * walk back up the tree (adjusting slot pointers as we go)
2314 	 * and restart the search process.
2315 	 */
2316 	atomic_inc(&root_eb->refs);	/* For path */
2317 	path->nodes[root_level] = root_eb;
2318 	path->slots[root_level] = 0;
2319 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2320 walk_down:
2321 	level = root_level;
2322 	while (level >= 0) {
2323 		if (path->nodes[level] == NULL) {
2324 			int parent_slot;
2325 			u64 child_bytenr;
2326 
2327 			/*
2328 			 * We need to get child blockptr from parent before we
2329 			 * can read it.
2330 			  */
2331 			eb = path->nodes[level + 1];
2332 			parent_slot = path->slots[level + 1];
2333 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2334 
2335 			eb = btrfs_read_node_slot(eb, parent_slot);
2336 			if (IS_ERR(eb)) {
2337 				ret = PTR_ERR(eb);
2338 				goto out;
2339 			}
2340 
2341 			path->nodes[level] = eb;
2342 			path->slots[level] = 0;
2343 
2344 			btrfs_tree_read_lock(eb);
2345 			path->locks[level] = BTRFS_READ_LOCK;
2346 
2347 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2348 							fs_info->nodesize,
2349 							GFP_NOFS);
2350 			if (ret)
2351 				goto out;
2352 		}
2353 
2354 		if (level == 0) {
2355 			ret = btrfs_qgroup_trace_leaf_items(trans,
2356 							    path->nodes[level]);
2357 			if (ret)
2358 				goto out;
2359 
2360 			/* Nonzero return here means we completed our search */
2361 			ret = adjust_slots_upwards(path, root_level);
2362 			if (ret)
2363 				break;
2364 
2365 			/* Restart search with new slots */
2366 			goto walk_down;
2367 		}
2368 
2369 		level--;
2370 	}
2371 
2372 	ret = 0;
2373 out:
2374 	btrfs_free_path(path);
2375 
2376 	return ret;
2377 }
2378 
2379 #define UPDATE_NEW	0
2380 #define UPDATE_OLD	1
2381 /*
2382  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2383  */
2384 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2385 				struct ulist *roots, struct ulist *tmp,
2386 				struct ulist *qgroups, u64 seq, int update_old)
2387 {
2388 	struct ulist_node *unode;
2389 	struct ulist_iterator uiter;
2390 	struct ulist_node *tmp_unode;
2391 	struct ulist_iterator tmp_uiter;
2392 	struct btrfs_qgroup *qg;
2393 	int ret = 0;
2394 
2395 	if (!roots)
2396 		return 0;
2397 	ULIST_ITER_INIT(&uiter);
2398 	while ((unode = ulist_next(roots, &uiter))) {
2399 		qg = find_qgroup_rb(fs_info, unode->val);
2400 		if (!qg)
2401 			continue;
2402 
2403 		ulist_reinit(tmp);
2404 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2405 				GFP_ATOMIC);
2406 		if (ret < 0)
2407 			return ret;
2408 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2409 		if (ret < 0)
2410 			return ret;
2411 		ULIST_ITER_INIT(&tmp_uiter);
2412 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2413 			struct btrfs_qgroup_list *glist;
2414 
2415 			qg = unode_aux_to_qgroup(tmp_unode);
2416 			if (update_old)
2417 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2418 			else
2419 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2420 			list_for_each_entry(glist, &qg->groups, next_group) {
2421 				ret = ulist_add(qgroups, glist->group->qgroupid,
2422 						qgroup_to_aux(glist->group),
2423 						GFP_ATOMIC);
2424 				if (ret < 0)
2425 					return ret;
2426 				ret = ulist_add(tmp, glist->group->qgroupid,
2427 						qgroup_to_aux(glist->group),
2428 						GFP_ATOMIC);
2429 				if (ret < 0)
2430 					return ret;
2431 			}
2432 		}
2433 	}
2434 	return 0;
2435 }
2436 
2437 /*
2438  * Update qgroup rfer/excl counters.
2439  * Rfer update is easy, codes can explain themselves.
2440  *
2441  * Excl update is tricky, the update is split into 2 parts.
2442  * Part 1: Possible exclusive <-> sharing detect:
2443  *	|	A	|	!A	|
2444  *  -------------------------------------
2445  *  B	|	*	|	-	|
2446  *  -------------------------------------
2447  *  !B	|	+	|	**	|
2448  *  -------------------------------------
2449  *
2450  * Conditions:
2451  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2452  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2453  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2454  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2455  *
2456  * Results:
2457  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2458  * *: Definitely not changed.		**: Possible unchanged.
2459  *
2460  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2461  *
2462  * To make the logic clear, we first use condition A and B to split
2463  * combination into 4 results.
2464  *
2465  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2466  * only on variant maybe 0.
2467  *
2468  * Lastly, check result **, since there are 2 variants maybe 0, split them
2469  * again(2x2).
2470  * But this time we don't need to consider other things, the codes and logic
2471  * is easy to understand now.
2472  */
2473 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2474 				  struct ulist *qgroups,
2475 				  u64 nr_old_roots,
2476 				  u64 nr_new_roots,
2477 				  u64 num_bytes, u64 seq)
2478 {
2479 	struct ulist_node *unode;
2480 	struct ulist_iterator uiter;
2481 	struct btrfs_qgroup *qg;
2482 	u64 cur_new_count, cur_old_count;
2483 
2484 	ULIST_ITER_INIT(&uiter);
2485 	while ((unode = ulist_next(qgroups, &uiter))) {
2486 		bool dirty = false;
2487 
2488 		qg = unode_aux_to_qgroup(unode);
2489 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2490 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2491 
2492 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2493 					     cur_new_count);
2494 
2495 		/* Rfer update part */
2496 		if (cur_old_count == 0 && cur_new_count > 0) {
2497 			qg->rfer += num_bytes;
2498 			qg->rfer_cmpr += num_bytes;
2499 			dirty = true;
2500 		}
2501 		if (cur_old_count > 0 && cur_new_count == 0) {
2502 			qg->rfer -= num_bytes;
2503 			qg->rfer_cmpr -= num_bytes;
2504 			dirty = true;
2505 		}
2506 
2507 		/* Excl update part */
2508 		/* Exclusive/none -> shared case */
2509 		if (cur_old_count == nr_old_roots &&
2510 		    cur_new_count < nr_new_roots) {
2511 			/* Exclusive -> shared */
2512 			if (cur_old_count != 0) {
2513 				qg->excl -= num_bytes;
2514 				qg->excl_cmpr -= num_bytes;
2515 				dirty = true;
2516 			}
2517 		}
2518 
2519 		/* Shared -> exclusive/none case */
2520 		if (cur_old_count < nr_old_roots &&
2521 		    cur_new_count == nr_new_roots) {
2522 			/* Shared->exclusive */
2523 			if (cur_new_count != 0) {
2524 				qg->excl += num_bytes;
2525 				qg->excl_cmpr += num_bytes;
2526 				dirty = true;
2527 			}
2528 		}
2529 
2530 		/* Exclusive/none -> exclusive/none case */
2531 		if (cur_old_count == nr_old_roots &&
2532 		    cur_new_count == nr_new_roots) {
2533 			if (cur_old_count == 0) {
2534 				/* None -> exclusive/none */
2535 
2536 				if (cur_new_count != 0) {
2537 					/* None -> exclusive */
2538 					qg->excl += num_bytes;
2539 					qg->excl_cmpr += num_bytes;
2540 					dirty = true;
2541 				}
2542 				/* None -> none, nothing changed */
2543 			} else {
2544 				/* Exclusive -> exclusive/none */
2545 
2546 				if (cur_new_count == 0) {
2547 					/* Exclusive -> none */
2548 					qg->excl -= num_bytes;
2549 					qg->excl_cmpr -= num_bytes;
2550 					dirty = true;
2551 				}
2552 				/* Exclusive -> exclusive, nothing changed */
2553 			}
2554 		}
2555 
2556 		if (dirty)
2557 			qgroup_dirty(fs_info, qg);
2558 	}
2559 	return 0;
2560 }
2561 
2562 /*
2563  * Check if the @roots potentially is a list of fs tree roots
2564  *
2565  * Return 0 for definitely not a fs/subvol tree roots ulist
2566  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2567  *          one as well)
2568  */
2569 static int maybe_fs_roots(struct ulist *roots)
2570 {
2571 	struct ulist_node *unode;
2572 	struct ulist_iterator uiter;
2573 
2574 	/* Empty one, still possible for fs roots */
2575 	if (!roots || roots->nnodes == 0)
2576 		return 1;
2577 
2578 	ULIST_ITER_INIT(&uiter);
2579 	unode = ulist_next(roots, &uiter);
2580 	if (!unode)
2581 		return 1;
2582 
2583 	/*
2584 	 * If it contains fs tree roots, then it must belong to fs/subvol
2585 	 * trees.
2586 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2587 	 */
2588 	return is_fstree(unode->val);
2589 }
2590 
2591 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2592 				u64 num_bytes, struct ulist *old_roots,
2593 				struct ulist *new_roots)
2594 {
2595 	struct btrfs_fs_info *fs_info = trans->fs_info;
2596 	struct ulist *qgroups = NULL;
2597 	struct ulist *tmp = NULL;
2598 	u64 seq;
2599 	u64 nr_new_roots = 0;
2600 	u64 nr_old_roots = 0;
2601 	int ret = 0;
2602 
2603 	/*
2604 	 * If quotas get disabled meanwhile, the resources need to be freed and
2605 	 * we can't just exit here.
2606 	 */
2607 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2608 		goto out_free;
2609 
2610 	if (new_roots) {
2611 		if (!maybe_fs_roots(new_roots))
2612 			goto out_free;
2613 		nr_new_roots = new_roots->nnodes;
2614 	}
2615 	if (old_roots) {
2616 		if (!maybe_fs_roots(old_roots))
2617 			goto out_free;
2618 		nr_old_roots = old_roots->nnodes;
2619 	}
2620 
2621 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2622 	if (nr_old_roots == 0 && nr_new_roots == 0)
2623 		goto out_free;
2624 
2625 	BUG_ON(!fs_info->quota_root);
2626 
2627 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2628 					num_bytes, nr_old_roots, nr_new_roots);
2629 
2630 	qgroups = ulist_alloc(GFP_NOFS);
2631 	if (!qgroups) {
2632 		ret = -ENOMEM;
2633 		goto out_free;
2634 	}
2635 	tmp = ulist_alloc(GFP_NOFS);
2636 	if (!tmp) {
2637 		ret = -ENOMEM;
2638 		goto out_free;
2639 	}
2640 
2641 	mutex_lock(&fs_info->qgroup_rescan_lock);
2642 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2643 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2644 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2645 			ret = 0;
2646 			goto out_free;
2647 		}
2648 	}
2649 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2650 
2651 	spin_lock(&fs_info->qgroup_lock);
2652 	seq = fs_info->qgroup_seq;
2653 
2654 	/* Update old refcnts using old_roots */
2655 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2656 				   UPDATE_OLD);
2657 	if (ret < 0)
2658 		goto out;
2659 
2660 	/* Update new refcnts using new_roots */
2661 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2662 				   UPDATE_NEW);
2663 	if (ret < 0)
2664 		goto out;
2665 
2666 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2667 			       num_bytes, seq);
2668 
2669 	/*
2670 	 * Bump qgroup_seq to avoid seq overlap
2671 	 */
2672 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2673 out:
2674 	spin_unlock(&fs_info->qgroup_lock);
2675 out_free:
2676 	ulist_free(tmp);
2677 	ulist_free(qgroups);
2678 	ulist_free(old_roots);
2679 	ulist_free(new_roots);
2680 	return ret;
2681 }
2682 
2683 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2684 {
2685 	struct btrfs_fs_info *fs_info = trans->fs_info;
2686 	struct btrfs_qgroup_extent_record *record;
2687 	struct btrfs_delayed_ref_root *delayed_refs;
2688 	struct ulist *new_roots = NULL;
2689 	struct rb_node *node;
2690 	u64 num_dirty_extents = 0;
2691 	u64 qgroup_to_skip;
2692 	int ret = 0;
2693 
2694 	delayed_refs = &trans->transaction->delayed_refs;
2695 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2696 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2697 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2698 				  node);
2699 
2700 		num_dirty_extents++;
2701 		trace_btrfs_qgroup_account_extents(fs_info, record);
2702 
2703 		if (!ret) {
2704 			/*
2705 			 * Old roots should be searched when inserting qgroup
2706 			 * extent record
2707 			 */
2708 			if (WARN_ON(!record->old_roots)) {
2709 				/* Search commit root to find old_roots */
2710 				ret = btrfs_find_all_roots(NULL, fs_info,
2711 						record->bytenr, 0,
2712 						&record->old_roots, false);
2713 				if (ret < 0)
2714 					goto cleanup;
2715 			}
2716 
2717 			/* Free the reserved data space */
2718 			btrfs_qgroup_free_refroot(fs_info,
2719 					record->data_rsv_refroot,
2720 					record->data_rsv,
2721 					BTRFS_QGROUP_RSV_DATA);
2722 			/*
2723 			 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2724 			 * which doesn't lock tree or delayed_refs and search
2725 			 * current root. It's safe inside commit_transaction().
2726 			 */
2727 			ret = btrfs_find_all_roots(trans, fs_info,
2728 			   record->bytenr, BTRFS_SEQ_LAST, &new_roots, false);
2729 			if (ret < 0)
2730 				goto cleanup;
2731 			if (qgroup_to_skip) {
2732 				ulist_del(new_roots, qgroup_to_skip, 0);
2733 				ulist_del(record->old_roots, qgroup_to_skip,
2734 					  0);
2735 			}
2736 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2737 							  record->num_bytes,
2738 							  record->old_roots,
2739 							  new_roots);
2740 			record->old_roots = NULL;
2741 			new_roots = NULL;
2742 		}
2743 cleanup:
2744 		ulist_free(record->old_roots);
2745 		ulist_free(new_roots);
2746 		new_roots = NULL;
2747 		rb_erase(node, &delayed_refs->dirty_extent_root);
2748 		kfree(record);
2749 
2750 	}
2751 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2752 				       num_dirty_extents);
2753 	return ret;
2754 }
2755 
2756 /*
2757  * called from commit_transaction. Writes all changed qgroups to disk.
2758  */
2759 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2760 {
2761 	struct btrfs_fs_info *fs_info = trans->fs_info;
2762 	int ret = 0;
2763 
2764 	if (!fs_info->quota_root)
2765 		return ret;
2766 
2767 	spin_lock(&fs_info->qgroup_lock);
2768 	while (!list_empty(&fs_info->dirty_qgroups)) {
2769 		struct btrfs_qgroup *qgroup;
2770 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2771 					  struct btrfs_qgroup, dirty);
2772 		list_del_init(&qgroup->dirty);
2773 		spin_unlock(&fs_info->qgroup_lock);
2774 		ret = update_qgroup_info_item(trans, qgroup);
2775 		if (ret)
2776 			fs_info->qgroup_flags |=
2777 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2778 		ret = update_qgroup_limit_item(trans, qgroup);
2779 		if (ret)
2780 			fs_info->qgroup_flags |=
2781 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2782 		spin_lock(&fs_info->qgroup_lock);
2783 	}
2784 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2785 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2786 	else
2787 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2788 	spin_unlock(&fs_info->qgroup_lock);
2789 
2790 	ret = update_qgroup_status_item(trans);
2791 	if (ret)
2792 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2793 
2794 	return ret;
2795 }
2796 
2797 /*
2798  * Copy the accounting information between qgroups. This is necessary
2799  * when a snapshot or a subvolume is created. Throwing an error will
2800  * cause a transaction abort so we take extra care here to only error
2801  * when a readonly fs is a reasonable outcome.
2802  */
2803 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2804 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2805 {
2806 	int ret = 0;
2807 	int i;
2808 	u64 *i_qgroups;
2809 	bool committing = false;
2810 	struct btrfs_fs_info *fs_info = trans->fs_info;
2811 	struct btrfs_root *quota_root;
2812 	struct btrfs_qgroup *srcgroup;
2813 	struct btrfs_qgroup *dstgroup;
2814 	bool need_rescan = false;
2815 	u32 level_size = 0;
2816 	u64 nums;
2817 
2818 	/*
2819 	 * There are only two callers of this function.
2820 	 *
2821 	 * One in create_subvol() in the ioctl context, which needs to hold
2822 	 * the qgroup_ioctl_lock.
2823 	 *
2824 	 * The other one in create_pending_snapshot() where no other qgroup
2825 	 * code can modify the fs as they all need to either start a new trans
2826 	 * or hold a trans handler, thus we don't need to hold
2827 	 * qgroup_ioctl_lock.
2828 	 * This would avoid long and complex lock chain and make lockdep happy.
2829 	 */
2830 	spin_lock(&fs_info->trans_lock);
2831 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2832 		committing = true;
2833 	spin_unlock(&fs_info->trans_lock);
2834 
2835 	if (!committing)
2836 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2837 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2838 		goto out;
2839 
2840 	quota_root = fs_info->quota_root;
2841 	if (!quota_root) {
2842 		ret = -EINVAL;
2843 		goto out;
2844 	}
2845 
2846 	if (inherit) {
2847 		i_qgroups = (u64 *)(inherit + 1);
2848 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2849 		       2 * inherit->num_excl_copies;
2850 		for (i = 0; i < nums; ++i) {
2851 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2852 
2853 			/*
2854 			 * Zero out invalid groups so we can ignore
2855 			 * them later.
2856 			 */
2857 			if (!srcgroup ||
2858 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2859 				*i_qgroups = 0ULL;
2860 
2861 			++i_qgroups;
2862 		}
2863 	}
2864 
2865 	/*
2866 	 * create a tracking group for the subvol itself
2867 	 */
2868 	ret = add_qgroup_item(trans, quota_root, objectid);
2869 	if (ret)
2870 		goto out;
2871 
2872 	/*
2873 	 * add qgroup to all inherited groups
2874 	 */
2875 	if (inherit) {
2876 		i_qgroups = (u64 *)(inherit + 1);
2877 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2878 			if (*i_qgroups == 0)
2879 				continue;
2880 			ret = add_qgroup_relation_item(trans, objectid,
2881 						       *i_qgroups);
2882 			if (ret && ret != -EEXIST)
2883 				goto out;
2884 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2885 						       objectid);
2886 			if (ret && ret != -EEXIST)
2887 				goto out;
2888 		}
2889 		ret = 0;
2890 	}
2891 
2892 
2893 	spin_lock(&fs_info->qgroup_lock);
2894 
2895 	dstgroup = add_qgroup_rb(fs_info, objectid);
2896 	if (IS_ERR(dstgroup)) {
2897 		ret = PTR_ERR(dstgroup);
2898 		goto unlock;
2899 	}
2900 
2901 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2902 		dstgroup->lim_flags = inherit->lim.flags;
2903 		dstgroup->max_rfer = inherit->lim.max_rfer;
2904 		dstgroup->max_excl = inherit->lim.max_excl;
2905 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2906 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2907 
2908 		ret = update_qgroup_limit_item(trans, dstgroup);
2909 		if (ret) {
2910 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2911 			btrfs_info(fs_info,
2912 				   "unable to update quota limit for %llu",
2913 				   dstgroup->qgroupid);
2914 			goto unlock;
2915 		}
2916 	}
2917 
2918 	if (srcid) {
2919 		srcgroup = find_qgroup_rb(fs_info, srcid);
2920 		if (!srcgroup)
2921 			goto unlock;
2922 
2923 		/*
2924 		 * We call inherit after we clone the root in order to make sure
2925 		 * our counts don't go crazy, so at this point the only
2926 		 * difference between the two roots should be the root node.
2927 		 */
2928 		level_size = fs_info->nodesize;
2929 		dstgroup->rfer = srcgroup->rfer;
2930 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2931 		dstgroup->excl = level_size;
2932 		dstgroup->excl_cmpr = level_size;
2933 		srcgroup->excl = level_size;
2934 		srcgroup->excl_cmpr = level_size;
2935 
2936 		/* inherit the limit info */
2937 		dstgroup->lim_flags = srcgroup->lim_flags;
2938 		dstgroup->max_rfer = srcgroup->max_rfer;
2939 		dstgroup->max_excl = srcgroup->max_excl;
2940 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2941 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2942 
2943 		qgroup_dirty(fs_info, dstgroup);
2944 		qgroup_dirty(fs_info, srcgroup);
2945 	}
2946 
2947 	if (!inherit)
2948 		goto unlock;
2949 
2950 	i_qgroups = (u64 *)(inherit + 1);
2951 	for (i = 0; i < inherit->num_qgroups; ++i) {
2952 		if (*i_qgroups) {
2953 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2954 			if (ret)
2955 				goto unlock;
2956 		}
2957 		++i_qgroups;
2958 
2959 		/*
2960 		 * If we're doing a snapshot, and adding the snapshot to a new
2961 		 * qgroup, the numbers are guaranteed to be incorrect.
2962 		 */
2963 		if (srcid)
2964 			need_rescan = true;
2965 	}
2966 
2967 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2968 		struct btrfs_qgroup *src;
2969 		struct btrfs_qgroup *dst;
2970 
2971 		if (!i_qgroups[0] || !i_qgroups[1])
2972 			continue;
2973 
2974 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2975 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2976 
2977 		if (!src || !dst) {
2978 			ret = -EINVAL;
2979 			goto unlock;
2980 		}
2981 
2982 		dst->rfer = src->rfer - level_size;
2983 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2984 
2985 		/* Manually tweaking numbers certainly needs a rescan */
2986 		need_rescan = true;
2987 	}
2988 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2989 		struct btrfs_qgroup *src;
2990 		struct btrfs_qgroup *dst;
2991 
2992 		if (!i_qgroups[0] || !i_qgroups[1])
2993 			continue;
2994 
2995 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2996 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2997 
2998 		if (!src || !dst) {
2999 			ret = -EINVAL;
3000 			goto unlock;
3001 		}
3002 
3003 		dst->excl = src->excl + level_size;
3004 		dst->excl_cmpr = src->excl_cmpr + level_size;
3005 		need_rescan = true;
3006 	}
3007 
3008 unlock:
3009 	spin_unlock(&fs_info->qgroup_lock);
3010 	if (!ret)
3011 		ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3012 out:
3013 	if (!committing)
3014 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
3015 	if (need_rescan)
3016 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3017 	return ret;
3018 }
3019 
3020 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3021 {
3022 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3023 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3024 		return false;
3025 
3026 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3027 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3028 		return false;
3029 
3030 	return true;
3031 }
3032 
3033 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3034 			  enum btrfs_qgroup_rsv_type type)
3035 {
3036 	struct btrfs_qgroup *qgroup;
3037 	struct btrfs_fs_info *fs_info = root->fs_info;
3038 	u64 ref_root = root->root_key.objectid;
3039 	int ret = 0;
3040 	struct ulist_node *unode;
3041 	struct ulist_iterator uiter;
3042 
3043 	if (!is_fstree(ref_root))
3044 		return 0;
3045 
3046 	if (num_bytes == 0)
3047 		return 0;
3048 
3049 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3050 	    capable(CAP_SYS_RESOURCE))
3051 		enforce = false;
3052 
3053 	spin_lock(&fs_info->qgroup_lock);
3054 	if (!fs_info->quota_root)
3055 		goto out;
3056 
3057 	qgroup = find_qgroup_rb(fs_info, ref_root);
3058 	if (!qgroup)
3059 		goto out;
3060 
3061 	/*
3062 	 * in a first step, we check all affected qgroups if any limits would
3063 	 * be exceeded
3064 	 */
3065 	ulist_reinit(fs_info->qgroup_ulist);
3066 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3067 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3068 	if (ret < 0)
3069 		goto out;
3070 	ULIST_ITER_INIT(&uiter);
3071 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3072 		struct btrfs_qgroup *qg;
3073 		struct btrfs_qgroup_list *glist;
3074 
3075 		qg = unode_aux_to_qgroup(unode);
3076 
3077 		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3078 			ret = -EDQUOT;
3079 			goto out;
3080 		}
3081 
3082 		list_for_each_entry(glist, &qg->groups, next_group) {
3083 			ret = ulist_add(fs_info->qgroup_ulist,
3084 					glist->group->qgroupid,
3085 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3086 			if (ret < 0)
3087 				goto out;
3088 		}
3089 	}
3090 	ret = 0;
3091 	/*
3092 	 * no limits exceeded, now record the reservation into all qgroups
3093 	 */
3094 	ULIST_ITER_INIT(&uiter);
3095 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3096 		struct btrfs_qgroup *qg;
3097 
3098 		qg = unode_aux_to_qgroup(unode);
3099 
3100 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
3101 	}
3102 
3103 out:
3104 	spin_unlock(&fs_info->qgroup_lock);
3105 	return ret;
3106 }
3107 
3108 /*
3109  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3110  * qgroup).
3111  *
3112  * Will handle all higher level qgroup too.
3113  *
3114  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3115  * This special case is only used for META_PERTRANS type.
3116  */
3117 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3118 			       u64 ref_root, u64 num_bytes,
3119 			       enum btrfs_qgroup_rsv_type type)
3120 {
3121 	struct btrfs_qgroup *qgroup;
3122 	struct ulist_node *unode;
3123 	struct ulist_iterator uiter;
3124 	int ret = 0;
3125 
3126 	if (!is_fstree(ref_root))
3127 		return;
3128 
3129 	if (num_bytes == 0)
3130 		return;
3131 
3132 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3133 		WARN(1, "%s: Invalid type to free", __func__);
3134 		return;
3135 	}
3136 	spin_lock(&fs_info->qgroup_lock);
3137 
3138 	if (!fs_info->quota_root)
3139 		goto out;
3140 
3141 	qgroup = find_qgroup_rb(fs_info, ref_root);
3142 	if (!qgroup)
3143 		goto out;
3144 
3145 	if (num_bytes == (u64)-1)
3146 		/*
3147 		 * We're freeing all pertrans rsv, get reserved value from
3148 		 * level 0 qgroup as real num_bytes to free.
3149 		 */
3150 		num_bytes = qgroup->rsv.values[type];
3151 
3152 	ulist_reinit(fs_info->qgroup_ulist);
3153 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3154 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3155 	if (ret < 0)
3156 		goto out;
3157 	ULIST_ITER_INIT(&uiter);
3158 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3159 		struct btrfs_qgroup *qg;
3160 		struct btrfs_qgroup_list *glist;
3161 
3162 		qg = unode_aux_to_qgroup(unode);
3163 
3164 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3165 
3166 		list_for_each_entry(glist, &qg->groups, next_group) {
3167 			ret = ulist_add(fs_info->qgroup_ulist,
3168 					glist->group->qgroupid,
3169 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3170 			if (ret < 0)
3171 				goto out;
3172 		}
3173 	}
3174 
3175 out:
3176 	spin_unlock(&fs_info->qgroup_lock);
3177 }
3178 
3179 /*
3180  * Check if the leaf is the last leaf. Which means all node pointers
3181  * are at their last position.
3182  */
3183 static bool is_last_leaf(struct btrfs_path *path)
3184 {
3185 	int i;
3186 
3187 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3188 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3189 			return false;
3190 	}
3191 	return true;
3192 }
3193 
3194 /*
3195  * returns < 0 on error, 0 when more leafs are to be scanned.
3196  * returns 1 when done.
3197  */
3198 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3199 			      struct btrfs_path *path)
3200 {
3201 	struct btrfs_fs_info *fs_info = trans->fs_info;
3202 	struct btrfs_root *extent_root;
3203 	struct btrfs_key found;
3204 	struct extent_buffer *scratch_leaf = NULL;
3205 	struct ulist *roots = NULL;
3206 	u64 num_bytes;
3207 	bool done;
3208 	int slot;
3209 	int ret;
3210 
3211 	mutex_lock(&fs_info->qgroup_rescan_lock);
3212 	extent_root = btrfs_extent_root(fs_info,
3213 				fs_info->qgroup_rescan_progress.objectid);
3214 	ret = btrfs_search_slot_for_read(extent_root,
3215 					 &fs_info->qgroup_rescan_progress,
3216 					 path, 1, 0);
3217 
3218 	btrfs_debug(fs_info,
3219 		"current progress key (%llu %u %llu), search_slot ret %d",
3220 		fs_info->qgroup_rescan_progress.objectid,
3221 		fs_info->qgroup_rescan_progress.type,
3222 		fs_info->qgroup_rescan_progress.offset, ret);
3223 
3224 	if (ret) {
3225 		/*
3226 		 * The rescan is about to end, we will not be scanning any
3227 		 * further blocks. We cannot unset the RESCAN flag here, because
3228 		 * we want to commit the transaction if everything went well.
3229 		 * To make the live accounting work in this phase, we set our
3230 		 * scan progress pointer such that every real extent objectid
3231 		 * will be smaller.
3232 		 */
3233 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3234 		btrfs_release_path(path);
3235 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3236 		return ret;
3237 	}
3238 	done = is_last_leaf(path);
3239 
3240 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3241 			      btrfs_header_nritems(path->nodes[0]) - 1);
3242 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3243 
3244 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3245 	if (!scratch_leaf) {
3246 		ret = -ENOMEM;
3247 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3248 		goto out;
3249 	}
3250 	slot = path->slots[0];
3251 	btrfs_release_path(path);
3252 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3253 
3254 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3255 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3256 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3257 		    found.type != BTRFS_METADATA_ITEM_KEY)
3258 			continue;
3259 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3260 			num_bytes = fs_info->nodesize;
3261 		else
3262 			num_bytes = found.offset;
3263 
3264 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3265 					   &roots, false);
3266 		if (ret < 0)
3267 			goto out;
3268 		/* For rescan, just pass old_roots as NULL */
3269 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3270 						  num_bytes, NULL, roots);
3271 		if (ret < 0)
3272 			goto out;
3273 	}
3274 out:
3275 	if (scratch_leaf)
3276 		free_extent_buffer(scratch_leaf);
3277 
3278 	if (done && !ret) {
3279 		ret = 1;
3280 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3281 	}
3282 	return ret;
3283 }
3284 
3285 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3286 {
3287 	return btrfs_fs_closing(fs_info) ||
3288 		test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3289 		!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
3290 }
3291 
3292 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3293 {
3294 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3295 						     qgroup_rescan_work);
3296 	struct btrfs_path *path;
3297 	struct btrfs_trans_handle *trans = NULL;
3298 	int err = -ENOMEM;
3299 	int ret = 0;
3300 	bool stopped = false;
3301 
3302 	path = btrfs_alloc_path();
3303 	if (!path)
3304 		goto out;
3305 	/*
3306 	 * Rescan should only search for commit root, and any later difference
3307 	 * should be recorded by qgroup
3308 	 */
3309 	path->search_commit_root = 1;
3310 	path->skip_locking = 1;
3311 
3312 	err = 0;
3313 	while (!err && !(stopped = rescan_should_stop(fs_info))) {
3314 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3315 		if (IS_ERR(trans)) {
3316 			err = PTR_ERR(trans);
3317 			break;
3318 		}
3319 
3320 		err = qgroup_rescan_leaf(trans, path);
3321 
3322 		if (err > 0)
3323 			btrfs_commit_transaction(trans);
3324 		else
3325 			btrfs_end_transaction(trans);
3326 	}
3327 
3328 out:
3329 	btrfs_free_path(path);
3330 
3331 	mutex_lock(&fs_info->qgroup_rescan_lock);
3332 	if (err > 0 &&
3333 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3334 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3335 	} else if (err < 0 || stopped) {
3336 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3337 	}
3338 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3339 
3340 	/*
3341 	 * only update status, since the previous part has already updated the
3342 	 * qgroup info.
3343 	 */
3344 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3345 	if (IS_ERR(trans)) {
3346 		err = PTR_ERR(trans);
3347 		trans = NULL;
3348 		btrfs_err(fs_info,
3349 			  "fail to start transaction for status update: %d",
3350 			  err);
3351 	}
3352 
3353 	mutex_lock(&fs_info->qgroup_rescan_lock);
3354 	if (!stopped)
3355 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3356 	if (trans) {
3357 		ret = update_qgroup_status_item(trans);
3358 		if (ret < 0) {
3359 			err = ret;
3360 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3361 				  err);
3362 		}
3363 	}
3364 	fs_info->qgroup_rescan_running = false;
3365 	complete_all(&fs_info->qgroup_rescan_completion);
3366 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3367 
3368 	if (!trans)
3369 		return;
3370 
3371 	btrfs_end_transaction(trans);
3372 
3373 	if (stopped) {
3374 		btrfs_info(fs_info, "qgroup scan paused");
3375 	} else if (err >= 0) {
3376 		btrfs_info(fs_info, "qgroup scan completed%s",
3377 			err > 0 ? " (inconsistency flag cleared)" : "");
3378 	} else {
3379 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3380 	}
3381 }
3382 
3383 /*
3384  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3385  * memory required for the rescan context.
3386  */
3387 static int
3388 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3389 		   int init_flags)
3390 {
3391 	int ret = 0;
3392 
3393 	if (!init_flags) {
3394 		/* we're resuming qgroup rescan at mount time */
3395 		if (!(fs_info->qgroup_flags &
3396 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3397 			btrfs_warn(fs_info,
3398 			"qgroup rescan init failed, qgroup rescan is not queued");
3399 			ret = -EINVAL;
3400 		} else if (!(fs_info->qgroup_flags &
3401 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3402 			btrfs_warn(fs_info,
3403 			"qgroup rescan init failed, qgroup is not enabled");
3404 			ret = -EINVAL;
3405 		}
3406 
3407 		if (ret)
3408 			return ret;
3409 	}
3410 
3411 	mutex_lock(&fs_info->qgroup_rescan_lock);
3412 
3413 	if (init_flags) {
3414 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3415 			btrfs_warn(fs_info,
3416 				   "qgroup rescan is already in progress");
3417 			ret = -EINPROGRESS;
3418 		} else if (!(fs_info->qgroup_flags &
3419 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3420 			btrfs_warn(fs_info,
3421 			"qgroup rescan init failed, qgroup is not enabled");
3422 			ret = -EINVAL;
3423 		} else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3424 			/* Quota disable is in progress */
3425 			ret = -EBUSY;
3426 		}
3427 
3428 		if (ret) {
3429 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3430 			return ret;
3431 		}
3432 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3433 	}
3434 
3435 	memset(&fs_info->qgroup_rescan_progress, 0,
3436 		sizeof(fs_info->qgroup_rescan_progress));
3437 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3438 	init_completion(&fs_info->qgroup_rescan_completion);
3439 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3440 
3441 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3442 			btrfs_qgroup_rescan_worker, NULL, NULL);
3443 	return 0;
3444 }
3445 
3446 static void
3447 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3448 {
3449 	struct rb_node *n;
3450 	struct btrfs_qgroup *qgroup;
3451 
3452 	spin_lock(&fs_info->qgroup_lock);
3453 	/* clear all current qgroup tracking information */
3454 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3455 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3456 		qgroup->rfer = 0;
3457 		qgroup->rfer_cmpr = 0;
3458 		qgroup->excl = 0;
3459 		qgroup->excl_cmpr = 0;
3460 		qgroup_dirty(fs_info, qgroup);
3461 	}
3462 	spin_unlock(&fs_info->qgroup_lock);
3463 }
3464 
3465 int
3466 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3467 {
3468 	int ret = 0;
3469 	struct btrfs_trans_handle *trans;
3470 
3471 	ret = qgroup_rescan_init(fs_info, 0, 1);
3472 	if (ret)
3473 		return ret;
3474 
3475 	/*
3476 	 * We have set the rescan_progress to 0, which means no more
3477 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3478 	 * However, btrfs_qgroup_account_ref may be right after its call
3479 	 * to btrfs_find_all_roots, in which case it would still do the
3480 	 * accounting.
3481 	 * To solve this, we're committing the transaction, which will
3482 	 * ensure we run all delayed refs and only after that, we are
3483 	 * going to clear all tracking information for a clean start.
3484 	 */
3485 
3486 	trans = btrfs_join_transaction(fs_info->fs_root);
3487 	if (IS_ERR(trans)) {
3488 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3489 		return PTR_ERR(trans);
3490 	}
3491 	ret = btrfs_commit_transaction(trans);
3492 	if (ret) {
3493 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3494 		return ret;
3495 	}
3496 
3497 	qgroup_rescan_zero_tracking(fs_info);
3498 
3499 	mutex_lock(&fs_info->qgroup_rescan_lock);
3500 	fs_info->qgroup_rescan_running = true;
3501 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3502 			 &fs_info->qgroup_rescan_work);
3503 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3504 
3505 	return 0;
3506 }
3507 
3508 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3509 				     bool interruptible)
3510 {
3511 	int running;
3512 	int ret = 0;
3513 
3514 	mutex_lock(&fs_info->qgroup_rescan_lock);
3515 	running = fs_info->qgroup_rescan_running;
3516 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3517 
3518 	if (!running)
3519 		return 0;
3520 
3521 	if (interruptible)
3522 		ret = wait_for_completion_interruptible(
3523 					&fs_info->qgroup_rescan_completion);
3524 	else
3525 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3526 
3527 	return ret;
3528 }
3529 
3530 /*
3531  * this is only called from open_ctree where we're still single threaded, thus
3532  * locking is omitted here.
3533  */
3534 void
3535 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3536 {
3537 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3538 		mutex_lock(&fs_info->qgroup_rescan_lock);
3539 		fs_info->qgroup_rescan_running = true;
3540 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3541 				 &fs_info->qgroup_rescan_work);
3542 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3543 	}
3544 }
3545 
3546 #define rbtree_iterate_from_safe(node, next, start)				\
3547        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3548 
3549 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3550 				  struct extent_changeset *reserved, u64 start,
3551 				  u64 len)
3552 {
3553 	struct rb_node *node;
3554 	struct rb_node *next;
3555 	struct ulist_node *entry;
3556 	int ret = 0;
3557 
3558 	node = reserved->range_changed.root.rb_node;
3559 	if (!node)
3560 		return 0;
3561 	while (node) {
3562 		entry = rb_entry(node, struct ulist_node, rb_node);
3563 		if (entry->val < start)
3564 			node = node->rb_right;
3565 		else
3566 			node = node->rb_left;
3567 	}
3568 
3569 	if (entry->val > start && rb_prev(&entry->rb_node))
3570 		entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3571 				 rb_node);
3572 
3573 	rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3574 		u64 entry_start;
3575 		u64 entry_end;
3576 		u64 entry_len;
3577 		int clear_ret;
3578 
3579 		entry = rb_entry(node, struct ulist_node, rb_node);
3580 		entry_start = entry->val;
3581 		entry_end = entry->aux;
3582 		entry_len = entry_end - entry_start + 1;
3583 
3584 		if (entry_start >= start + len)
3585 			break;
3586 		if (entry_start + entry_len <= start)
3587 			continue;
3588 		/*
3589 		 * Now the entry is in [start, start + len), revert the
3590 		 * EXTENT_QGROUP_RESERVED bit.
3591 		 */
3592 		clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3593 					      entry_end, EXTENT_QGROUP_RESERVED);
3594 		if (!ret && clear_ret < 0)
3595 			ret = clear_ret;
3596 
3597 		ulist_del(&reserved->range_changed, entry->val, entry->aux);
3598 		if (likely(reserved->bytes_changed >= entry_len)) {
3599 			reserved->bytes_changed -= entry_len;
3600 		} else {
3601 			WARN_ON(1);
3602 			reserved->bytes_changed = 0;
3603 		}
3604 	}
3605 
3606 	return ret;
3607 }
3608 
3609 /*
3610  * Try to free some space for qgroup.
3611  *
3612  * For qgroup, there are only 3 ways to free qgroup space:
3613  * - Flush nodatacow write
3614  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3615  *   In theory, we should only flush nodatacow inodes, but it's not yet
3616  *   possible, so we need to flush the whole root.
3617  *
3618  * - Wait for ordered extents
3619  *   When ordered extents are finished, their reserved metadata is finally
3620  *   converted to per_trans status, which can be freed by later commit
3621  *   transaction.
3622  *
3623  * - Commit transaction
3624  *   This would free the meta_per_trans space.
3625  *   In theory this shouldn't provide much space, but any more qgroup space
3626  *   is needed.
3627  */
3628 static int try_flush_qgroup(struct btrfs_root *root)
3629 {
3630 	struct btrfs_trans_handle *trans;
3631 	int ret;
3632 
3633 	/* Can't hold an open transaction or we run the risk of deadlocking. */
3634 	ASSERT(current->journal_info == NULL);
3635 	if (WARN_ON(current->journal_info))
3636 		return 0;
3637 
3638 	/*
3639 	 * We don't want to run flush again and again, so if there is a running
3640 	 * one, we won't try to start a new flush, but exit directly.
3641 	 */
3642 	if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3643 		wait_event(root->qgroup_flush_wait,
3644 			!test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3645 		return 0;
3646 	}
3647 
3648 	ret = btrfs_start_delalloc_snapshot(root, true);
3649 	if (ret < 0)
3650 		goto out;
3651 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3652 
3653 	trans = btrfs_join_transaction(root);
3654 	if (IS_ERR(trans)) {
3655 		ret = PTR_ERR(trans);
3656 		goto out;
3657 	}
3658 
3659 	ret = btrfs_commit_transaction(trans);
3660 out:
3661 	clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3662 	wake_up(&root->qgroup_flush_wait);
3663 	return ret;
3664 }
3665 
3666 static int qgroup_reserve_data(struct btrfs_inode *inode,
3667 			struct extent_changeset **reserved_ret, u64 start,
3668 			u64 len)
3669 {
3670 	struct btrfs_root *root = inode->root;
3671 	struct extent_changeset *reserved;
3672 	bool new_reserved = false;
3673 	u64 orig_reserved;
3674 	u64 to_reserve;
3675 	int ret;
3676 
3677 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3678 	    !is_fstree(root->root_key.objectid) || len == 0)
3679 		return 0;
3680 
3681 	/* @reserved parameter is mandatory for qgroup */
3682 	if (WARN_ON(!reserved_ret))
3683 		return -EINVAL;
3684 	if (!*reserved_ret) {
3685 		new_reserved = true;
3686 		*reserved_ret = extent_changeset_alloc();
3687 		if (!*reserved_ret)
3688 			return -ENOMEM;
3689 	}
3690 	reserved = *reserved_ret;
3691 	/* Record already reserved space */
3692 	orig_reserved = reserved->bytes_changed;
3693 	ret = set_record_extent_bits(&inode->io_tree, start,
3694 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3695 
3696 	/* Newly reserved space */
3697 	to_reserve = reserved->bytes_changed - orig_reserved;
3698 	trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3699 					to_reserve, QGROUP_RESERVE);
3700 	if (ret < 0)
3701 		goto out;
3702 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3703 	if (ret < 0)
3704 		goto cleanup;
3705 
3706 	return ret;
3707 
3708 cleanup:
3709 	qgroup_unreserve_range(inode, reserved, start, len);
3710 out:
3711 	if (new_reserved) {
3712 		extent_changeset_free(reserved);
3713 		*reserved_ret = NULL;
3714 	}
3715 	return ret;
3716 }
3717 
3718 /*
3719  * Reserve qgroup space for range [start, start + len).
3720  *
3721  * This function will either reserve space from related qgroups or do nothing
3722  * if the range is already reserved.
3723  *
3724  * Return 0 for successful reservation
3725  * Return <0 for error (including -EQUOT)
3726  *
3727  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3728  *	 commit transaction. So caller should not hold any dirty page locked.
3729  */
3730 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3731 			struct extent_changeset **reserved_ret, u64 start,
3732 			u64 len)
3733 {
3734 	int ret;
3735 
3736 	ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3737 	if (ret <= 0 && ret != -EDQUOT)
3738 		return ret;
3739 
3740 	ret = try_flush_qgroup(inode->root);
3741 	if (ret < 0)
3742 		return ret;
3743 	return qgroup_reserve_data(inode, reserved_ret, start, len);
3744 }
3745 
3746 /* Free ranges specified by @reserved, normally in error path */
3747 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3748 			struct extent_changeset *reserved, u64 start, u64 len)
3749 {
3750 	struct btrfs_root *root = inode->root;
3751 	struct ulist_node *unode;
3752 	struct ulist_iterator uiter;
3753 	struct extent_changeset changeset;
3754 	int freed = 0;
3755 	int ret;
3756 
3757 	extent_changeset_init(&changeset);
3758 	len = round_up(start + len, root->fs_info->sectorsize);
3759 	start = round_down(start, root->fs_info->sectorsize);
3760 
3761 	ULIST_ITER_INIT(&uiter);
3762 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3763 		u64 range_start = unode->val;
3764 		/* unode->aux is the inclusive end */
3765 		u64 range_len = unode->aux - range_start + 1;
3766 		u64 free_start;
3767 		u64 free_len;
3768 
3769 		extent_changeset_release(&changeset);
3770 
3771 		/* Only free range in range [start, start + len) */
3772 		if (range_start >= start + len ||
3773 		    range_start + range_len <= start)
3774 			continue;
3775 		free_start = max(range_start, start);
3776 		free_len = min(start + len, range_start + range_len) -
3777 			   free_start;
3778 		/*
3779 		 * TODO: To also modify reserved->ranges_reserved to reflect
3780 		 * the modification.
3781 		 *
3782 		 * However as long as we free qgroup reserved according to
3783 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3784 		 * So not need to rush.
3785 		 */
3786 		ret = clear_record_extent_bits(&inode->io_tree, free_start,
3787 				free_start + free_len - 1,
3788 				EXTENT_QGROUP_RESERVED, &changeset);
3789 		if (ret < 0)
3790 			goto out;
3791 		freed += changeset.bytes_changed;
3792 	}
3793 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3794 				  BTRFS_QGROUP_RSV_DATA);
3795 	ret = freed;
3796 out:
3797 	extent_changeset_release(&changeset);
3798 	return ret;
3799 }
3800 
3801 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3802 			struct extent_changeset *reserved, u64 start, u64 len,
3803 			int free)
3804 {
3805 	struct extent_changeset changeset;
3806 	int trace_op = QGROUP_RELEASE;
3807 	int ret;
3808 
3809 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3810 		return 0;
3811 
3812 	/* In release case, we shouldn't have @reserved */
3813 	WARN_ON(!free && reserved);
3814 	if (free && reserved)
3815 		return qgroup_free_reserved_data(inode, reserved, start, len);
3816 	extent_changeset_init(&changeset);
3817 	ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3818 				       EXTENT_QGROUP_RESERVED, &changeset);
3819 	if (ret < 0)
3820 		goto out;
3821 
3822 	if (free)
3823 		trace_op = QGROUP_FREE;
3824 	trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3825 					changeset.bytes_changed, trace_op);
3826 	if (free)
3827 		btrfs_qgroup_free_refroot(inode->root->fs_info,
3828 				inode->root->root_key.objectid,
3829 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3830 	ret = changeset.bytes_changed;
3831 out:
3832 	extent_changeset_release(&changeset);
3833 	return ret;
3834 }
3835 
3836 /*
3837  * Free a reserved space range from io_tree and related qgroups
3838  *
3839  * Should be called when a range of pages get invalidated before reaching disk.
3840  * Or for error cleanup case.
3841  * if @reserved is given, only reserved range in [@start, @start + @len) will
3842  * be freed.
3843  *
3844  * For data written to disk, use btrfs_qgroup_release_data().
3845  *
3846  * NOTE: This function may sleep for memory allocation.
3847  */
3848 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3849 			struct extent_changeset *reserved, u64 start, u64 len)
3850 {
3851 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3852 }
3853 
3854 /*
3855  * Release a reserved space range from io_tree only.
3856  *
3857  * Should be called when a range of pages get written to disk and corresponding
3858  * FILE_EXTENT is inserted into corresponding root.
3859  *
3860  * Since new qgroup accounting framework will only update qgroup numbers at
3861  * commit_transaction() time, its reserved space shouldn't be freed from
3862  * related qgroups.
3863  *
3864  * But we should release the range from io_tree, to allow further write to be
3865  * COWed.
3866  *
3867  * NOTE: This function may sleep for memory allocation.
3868  */
3869 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3870 {
3871 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3872 }
3873 
3874 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3875 			      enum btrfs_qgroup_rsv_type type)
3876 {
3877 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3878 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3879 		return;
3880 	if (num_bytes == 0)
3881 		return;
3882 
3883 	spin_lock(&root->qgroup_meta_rsv_lock);
3884 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3885 		root->qgroup_meta_rsv_prealloc += num_bytes;
3886 	else
3887 		root->qgroup_meta_rsv_pertrans += num_bytes;
3888 	spin_unlock(&root->qgroup_meta_rsv_lock);
3889 }
3890 
3891 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3892 			     enum btrfs_qgroup_rsv_type type)
3893 {
3894 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3895 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3896 		return 0;
3897 	if (num_bytes == 0)
3898 		return 0;
3899 
3900 	spin_lock(&root->qgroup_meta_rsv_lock);
3901 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3902 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3903 				  num_bytes);
3904 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3905 	} else {
3906 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3907 				  num_bytes);
3908 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3909 	}
3910 	spin_unlock(&root->qgroup_meta_rsv_lock);
3911 	return num_bytes;
3912 }
3913 
3914 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3915 			      enum btrfs_qgroup_rsv_type type, bool enforce)
3916 {
3917 	struct btrfs_fs_info *fs_info = root->fs_info;
3918 	int ret;
3919 
3920 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3921 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3922 		return 0;
3923 
3924 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3925 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3926 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3927 	if (ret < 0)
3928 		return ret;
3929 	/*
3930 	 * Record what we have reserved into root.
3931 	 *
3932 	 * To avoid quota disabled->enabled underflow.
3933 	 * In that case, we may try to free space we haven't reserved
3934 	 * (since quota was disabled), so record what we reserved into root.
3935 	 * And ensure later release won't underflow this number.
3936 	 */
3937 	add_root_meta_rsv(root, num_bytes, type);
3938 	return ret;
3939 }
3940 
3941 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3942 				enum btrfs_qgroup_rsv_type type, bool enforce,
3943 				bool noflush)
3944 {
3945 	int ret;
3946 
3947 	ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3948 	if ((ret <= 0 && ret != -EDQUOT) || noflush)
3949 		return ret;
3950 
3951 	ret = try_flush_qgroup(root);
3952 	if (ret < 0)
3953 		return ret;
3954 	return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3955 }
3956 
3957 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3958 {
3959 	struct btrfs_fs_info *fs_info = root->fs_info;
3960 
3961 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3962 	    !is_fstree(root->root_key.objectid))
3963 		return;
3964 
3965 	/* TODO: Update trace point to handle such free */
3966 	trace_qgroup_meta_free_all_pertrans(root);
3967 	/* Special value -1 means to free all reserved space */
3968 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3969 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3970 }
3971 
3972 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3973 			      enum btrfs_qgroup_rsv_type type)
3974 {
3975 	struct btrfs_fs_info *fs_info = root->fs_info;
3976 
3977 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3978 	    !is_fstree(root->root_key.objectid))
3979 		return;
3980 
3981 	/*
3982 	 * reservation for META_PREALLOC can happen before quota is enabled,
3983 	 * which can lead to underflow.
3984 	 * Here ensure we will only free what we really have reserved.
3985 	 */
3986 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3987 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3988 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3989 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3990 				  num_bytes, type);
3991 }
3992 
3993 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3994 				int num_bytes)
3995 {
3996 	struct btrfs_qgroup *qgroup;
3997 	struct ulist_node *unode;
3998 	struct ulist_iterator uiter;
3999 	int ret = 0;
4000 
4001 	if (num_bytes == 0)
4002 		return;
4003 	if (!fs_info->quota_root)
4004 		return;
4005 
4006 	spin_lock(&fs_info->qgroup_lock);
4007 	qgroup = find_qgroup_rb(fs_info, ref_root);
4008 	if (!qgroup)
4009 		goto out;
4010 	ulist_reinit(fs_info->qgroup_ulist);
4011 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
4012 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
4013 	if (ret < 0)
4014 		goto out;
4015 	ULIST_ITER_INIT(&uiter);
4016 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
4017 		struct btrfs_qgroup *qg;
4018 		struct btrfs_qgroup_list *glist;
4019 
4020 		qg = unode_aux_to_qgroup(unode);
4021 
4022 		qgroup_rsv_release(fs_info, qg, num_bytes,
4023 				BTRFS_QGROUP_RSV_META_PREALLOC);
4024 		qgroup_rsv_add(fs_info, qg, num_bytes,
4025 				BTRFS_QGROUP_RSV_META_PERTRANS);
4026 		list_for_each_entry(glist, &qg->groups, next_group) {
4027 			ret = ulist_add(fs_info->qgroup_ulist,
4028 					glist->group->qgroupid,
4029 					qgroup_to_aux(glist->group), GFP_ATOMIC);
4030 			if (ret < 0)
4031 				goto out;
4032 		}
4033 	}
4034 out:
4035 	spin_unlock(&fs_info->qgroup_lock);
4036 }
4037 
4038 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4039 {
4040 	struct btrfs_fs_info *fs_info = root->fs_info;
4041 
4042 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4043 	    !is_fstree(root->root_key.objectid))
4044 		return;
4045 	/* Same as btrfs_qgroup_free_meta_prealloc() */
4046 	num_bytes = sub_root_meta_rsv(root, num_bytes,
4047 				      BTRFS_QGROUP_RSV_META_PREALLOC);
4048 	trace_qgroup_meta_convert(root, num_bytes);
4049 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4050 }
4051 
4052 /*
4053  * Check qgroup reserved space leaking, normally at destroy inode
4054  * time
4055  */
4056 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4057 {
4058 	struct extent_changeset changeset;
4059 	struct ulist_node *unode;
4060 	struct ulist_iterator iter;
4061 	int ret;
4062 
4063 	extent_changeset_init(&changeset);
4064 	ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4065 			EXTENT_QGROUP_RESERVED, &changeset);
4066 
4067 	WARN_ON(ret < 0);
4068 	if (WARN_ON(changeset.bytes_changed)) {
4069 		ULIST_ITER_INIT(&iter);
4070 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4071 			btrfs_warn(inode->root->fs_info,
4072 		"leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4073 				btrfs_ino(inode), unode->val, unode->aux);
4074 		}
4075 		btrfs_qgroup_free_refroot(inode->root->fs_info,
4076 				inode->root->root_key.objectid,
4077 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4078 
4079 	}
4080 	extent_changeset_release(&changeset);
4081 }
4082 
4083 void btrfs_qgroup_init_swapped_blocks(
4084 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4085 {
4086 	int i;
4087 
4088 	spin_lock_init(&swapped_blocks->lock);
4089 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4090 		swapped_blocks->blocks[i] = RB_ROOT;
4091 	swapped_blocks->swapped = false;
4092 }
4093 
4094 /*
4095  * Delete all swapped blocks record of @root.
4096  * Every record here means we skipped a full subtree scan for qgroup.
4097  *
4098  * Gets called when committing one transaction.
4099  */
4100 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4101 {
4102 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4103 	int i;
4104 
4105 	swapped_blocks = &root->swapped_blocks;
4106 
4107 	spin_lock(&swapped_blocks->lock);
4108 	if (!swapped_blocks->swapped)
4109 		goto out;
4110 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4111 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
4112 		struct btrfs_qgroup_swapped_block *entry;
4113 		struct btrfs_qgroup_swapped_block *next;
4114 
4115 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4116 						     node)
4117 			kfree(entry);
4118 		swapped_blocks->blocks[i] = RB_ROOT;
4119 	}
4120 	swapped_blocks->swapped = false;
4121 out:
4122 	spin_unlock(&swapped_blocks->lock);
4123 }
4124 
4125 /*
4126  * Add subtree roots record into @subvol_root.
4127  *
4128  * @subvol_root:	tree root of the subvolume tree get swapped
4129  * @bg:			block group under balance
4130  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
4131  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
4132  *			BOTH POINTERS ARE BEFORE TREE SWAP
4133  * @last_snapshot:	last snapshot generation of the subvolume tree
4134  */
4135 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4136 		struct btrfs_root *subvol_root,
4137 		struct btrfs_block_group *bg,
4138 		struct extent_buffer *subvol_parent, int subvol_slot,
4139 		struct extent_buffer *reloc_parent, int reloc_slot,
4140 		u64 last_snapshot)
4141 {
4142 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4143 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4144 	struct btrfs_qgroup_swapped_block *block;
4145 	struct rb_node **cur;
4146 	struct rb_node *parent = NULL;
4147 	int level = btrfs_header_level(subvol_parent) - 1;
4148 	int ret = 0;
4149 
4150 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4151 		return 0;
4152 
4153 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4154 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4155 		btrfs_err_rl(fs_info,
4156 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4157 			__func__,
4158 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4159 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4160 		return -EUCLEAN;
4161 	}
4162 
4163 	block = kmalloc(sizeof(*block), GFP_NOFS);
4164 	if (!block) {
4165 		ret = -ENOMEM;
4166 		goto out;
4167 	}
4168 
4169 	/*
4170 	 * @reloc_parent/slot is still before swap, while @block is going to
4171 	 * record the bytenr after swap, so we do the swap here.
4172 	 */
4173 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4174 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4175 							     reloc_slot);
4176 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4177 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4178 							    subvol_slot);
4179 	block->last_snapshot = last_snapshot;
4180 	block->level = level;
4181 
4182 	/*
4183 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4184 	 * no one else can modify tree blocks thus we qgroup will not change
4185 	 * no matter the value of trace_leaf.
4186 	 */
4187 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4188 		block->trace_leaf = true;
4189 	else
4190 		block->trace_leaf = false;
4191 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4192 
4193 	/* Insert @block into @blocks */
4194 	spin_lock(&blocks->lock);
4195 	cur = &blocks->blocks[level].rb_node;
4196 	while (*cur) {
4197 		struct btrfs_qgroup_swapped_block *entry;
4198 
4199 		parent = *cur;
4200 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4201 				 node);
4202 
4203 		if (entry->subvol_bytenr < block->subvol_bytenr) {
4204 			cur = &(*cur)->rb_left;
4205 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
4206 			cur = &(*cur)->rb_right;
4207 		} else {
4208 			if (entry->subvol_generation !=
4209 					block->subvol_generation ||
4210 			    entry->reloc_bytenr != block->reloc_bytenr ||
4211 			    entry->reloc_generation !=
4212 					block->reloc_generation) {
4213 				/*
4214 				 * Duplicated but mismatch entry found.
4215 				 * Shouldn't happen.
4216 				 *
4217 				 * Marking qgroup inconsistent should be enough
4218 				 * for end users.
4219 				 */
4220 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4221 				ret = -EEXIST;
4222 			}
4223 			kfree(block);
4224 			goto out_unlock;
4225 		}
4226 	}
4227 	rb_link_node(&block->node, parent, cur);
4228 	rb_insert_color(&block->node, &blocks->blocks[level]);
4229 	blocks->swapped = true;
4230 out_unlock:
4231 	spin_unlock(&blocks->lock);
4232 out:
4233 	if (ret < 0)
4234 		fs_info->qgroup_flags |=
4235 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4236 	return ret;
4237 }
4238 
4239 /*
4240  * Check if the tree block is a subtree root, and if so do the needed
4241  * delayed subtree trace for qgroup.
4242  *
4243  * This is called during btrfs_cow_block().
4244  */
4245 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4246 					 struct btrfs_root *root,
4247 					 struct extent_buffer *subvol_eb)
4248 {
4249 	struct btrfs_fs_info *fs_info = root->fs_info;
4250 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4251 	struct btrfs_qgroup_swapped_block *block;
4252 	struct extent_buffer *reloc_eb = NULL;
4253 	struct rb_node *node;
4254 	bool found = false;
4255 	bool swapped = false;
4256 	int level = btrfs_header_level(subvol_eb);
4257 	int ret = 0;
4258 	int i;
4259 
4260 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4261 		return 0;
4262 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4263 		return 0;
4264 
4265 	spin_lock(&blocks->lock);
4266 	if (!blocks->swapped) {
4267 		spin_unlock(&blocks->lock);
4268 		return 0;
4269 	}
4270 	node = blocks->blocks[level].rb_node;
4271 
4272 	while (node) {
4273 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4274 		if (block->subvol_bytenr < subvol_eb->start) {
4275 			node = node->rb_left;
4276 		} else if (block->subvol_bytenr > subvol_eb->start) {
4277 			node = node->rb_right;
4278 		} else {
4279 			found = true;
4280 			break;
4281 		}
4282 	}
4283 	if (!found) {
4284 		spin_unlock(&blocks->lock);
4285 		goto out;
4286 	}
4287 	/* Found one, remove it from @blocks first and update blocks->swapped */
4288 	rb_erase(&block->node, &blocks->blocks[level]);
4289 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4290 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4291 			swapped = true;
4292 			break;
4293 		}
4294 	}
4295 	blocks->swapped = swapped;
4296 	spin_unlock(&blocks->lock);
4297 
4298 	/* Read out reloc subtree root */
4299 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, 0,
4300 				   block->reloc_generation, block->level,
4301 				   &block->first_key);
4302 	if (IS_ERR(reloc_eb)) {
4303 		ret = PTR_ERR(reloc_eb);
4304 		reloc_eb = NULL;
4305 		goto free_out;
4306 	}
4307 	if (!extent_buffer_uptodate(reloc_eb)) {
4308 		ret = -EIO;
4309 		goto free_out;
4310 	}
4311 
4312 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4313 			block->last_snapshot, block->trace_leaf);
4314 free_out:
4315 	kfree(block);
4316 	free_extent_buffer(reloc_eb);
4317 out:
4318 	if (ret < 0) {
4319 		btrfs_err_rl(fs_info,
4320 			     "failed to account subtree at bytenr %llu: %d",
4321 			     subvol_eb->start, ret);
4322 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4323 	}
4324 	return ret;
4325 }
4326 
4327 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4328 {
4329 	struct btrfs_qgroup_extent_record *entry;
4330 	struct btrfs_qgroup_extent_record *next;
4331 	struct rb_root *root;
4332 
4333 	root = &trans->delayed_refs.dirty_extent_root;
4334 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4335 		ulist_free(entry->old_roots);
4336 		kfree(entry);
4337 	}
4338 }
4339