xref: /openbmc/linux/fs/btrfs/qgroup.c (revision a5b2c10c)
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 
15 #include "ctree.h"
16 #include "transaction.h"
17 #include "disk-io.h"
18 #include "locking.h"
19 #include "ulist.h"
20 #include "backref.h"
21 #include "extent_io.h"
22 #include "qgroup.h"
23 #include "block-group.h"
24 #include "sysfs.h"
25 
26 /* TODO XXX FIXME
27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
28  *  - reorganize keys
29  *  - compressed
30  *  - sync
31  *  - copy also limits on subvol creation
32  *  - limit
33  *  - caches for ulists
34  *  - performance benchmarks
35  *  - check all ioctl parameters
36  */
37 
38 /*
39  * Helpers to access qgroup reservation
40  *
41  * Callers should ensure the lock context and type are valid
42  */
43 
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45 {
46 	u64 ret = 0;
47 	int i;
48 
49 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 		ret += qgroup->rsv.values[i];
51 
52 	return ret;
53 }
54 
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57 {
58 	if (type == BTRFS_QGROUP_RSV_DATA)
59 		return "data";
60 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 		return "meta_pertrans";
62 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 		return "meta_prealloc";
64 	return NULL;
65 }
66 #endif
67 
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
70 			   enum btrfs_qgroup_rsv_type type)
71 {
72 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 	qgroup->rsv.values[type] += num_bytes;
74 }
75 
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
78 			       enum btrfs_qgroup_rsv_type type)
79 {
80 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 	if (qgroup->rsv.values[type] >= num_bytes) {
82 		qgroup->rsv.values[type] -= num_bytes;
83 		return;
84 	}
85 #ifdef CONFIG_BTRFS_DEBUG
86 	WARN_RATELIMIT(1,
87 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 		qgroup->qgroupid, qgroup_rsv_type_str(type),
89 		qgroup->rsv.values[type], num_bytes);
90 #endif
91 	qgroup->rsv.values[type] = 0;
92 }
93 
94 static void qgroup_rsv_add_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_add(fs_info, dest, src->rsv.values[i], i);
102 }
103 
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 					 struct btrfs_qgroup *dest,
106 					  struct btrfs_qgroup *src)
107 {
108 	int i;
109 
110 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
112 }
113 
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115 					   int mod)
116 {
117 	if (qg->old_refcnt < seq)
118 		qg->old_refcnt = seq;
119 	qg->old_refcnt += mod;
120 }
121 
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123 					   int mod)
124 {
125 	if (qg->new_refcnt < seq)
126 		qg->new_refcnt = seq;
127 	qg->new_refcnt += mod;
128 }
129 
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131 {
132 	if (qg->old_refcnt < seq)
133 		return 0;
134 	return qg->old_refcnt - seq;
135 }
136 
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138 {
139 	if (qg->new_refcnt < seq)
140 		return 0;
141 	return qg->new_refcnt - seq;
142 }
143 
144 /*
145  * glue structure to represent the relations between qgroups.
146  */
147 struct btrfs_qgroup_list {
148 	struct list_head next_group;
149 	struct list_head next_member;
150 	struct btrfs_qgroup *group;
151 	struct btrfs_qgroup *member;
152 };
153 
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 {
156 	return (u64)(uintptr_t)qg;
157 }
158 
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 {
161 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162 }
163 
164 static int
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 		   int init_flags);
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168 
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171 					   u64 qgroupid)
172 {
173 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 	struct btrfs_qgroup *qgroup;
175 
176 	while (n) {
177 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 		if (qgroup->qgroupid < qgroupid)
179 			n = n->rb_left;
180 		else if (qgroup->qgroupid > qgroupid)
181 			n = n->rb_right;
182 		else
183 			return qgroup;
184 	}
185 	return NULL;
186 }
187 
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190 					  u64 qgroupid)
191 {
192 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 	struct rb_node *parent = NULL;
194 	struct btrfs_qgroup *qgroup;
195 
196 	while (*p) {
197 		parent = *p;
198 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199 
200 		if (qgroup->qgroupid < qgroupid)
201 			p = &(*p)->rb_left;
202 		else if (qgroup->qgroupid > qgroupid)
203 			p = &(*p)->rb_right;
204 		else
205 			return qgroup;
206 	}
207 
208 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 	if (!qgroup)
210 		return ERR_PTR(-ENOMEM);
211 
212 	qgroup->qgroupid = qgroupid;
213 	INIT_LIST_HEAD(&qgroup->groups);
214 	INIT_LIST_HEAD(&qgroup->members);
215 	INIT_LIST_HEAD(&qgroup->dirty);
216 
217 	rb_link_node(&qgroup->node, parent, p);
218 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219 
220 	return qgroup;
221 }
222 
223 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
224 			    struct btrfs_qgroup *qgroup)
225 {
226 	struct btrfs_qgroup_list *list;
227 
228 	btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
229 	list_del(&qgroup->dirty);
230 	while (!list_empty(&qgroup->groups)) {
231 		list = list_first_entry(&qgroup->groups,
232 					struct btrfs_qgroup_list, next_group);
233 		list_del(&list->next_group);
234 		list_del(&list->next_member);
235 		kfree(list);
236 	}
237 
238 	while (!list_empty(&qgroup->members)) {
239 		list = list_first_entry(&qgroup->members,
240 					struct btrfs_qgroup_list, next_member);
241 		list_del(&list->next_group);
242 		list_del(&list->next_member);
243 		kfree(list);
244 	}
245 	kfree(qgroup);
246 }
247 
248 /* must be called with qgroup_lock held */
249 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
250 {
251 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
252 
253 	if (!qgroup)
254 		return -ENOENT;
255 
256 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
257 	__del_qgroup_rb(fs_info, qgroup);
258 	return 0;
259 }
260 
261 /* must be called with qgroup_lock held */
262 static int add_relation_rb(struct btrfs_fs_info *fs_info,
263 			   u64 memberid, u64 parentid)
264 {
265 	struct btrfs_qgroup *member;
266 	struct btrfs_qgroup *parent;
267 	struct btrfs_qgroup_list *list;
268 
269 	member = find_qgroup_rb(fs_info, memberid);
270 	parent = find_qgroup_rb(fs_info, parentid);
271 	if (!member || !parent)
272 		return -ENOENT;
273 
274 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
275 	if (!list)
276 		return -ENOMEM;
277 
278 	list->group = parent;
279 	list->member = member;
280 	list_add_tail(&list->next_group, &member->groups);
281 	list_add_tail(&list->next_member, &parent->members);
282 
283 	return 0;
284 }
285 
286 /* must be called with qgroup_lock held */
287 static int del_relation_rb(struct btrfs_fs_info *fs_info,
288 			   u64 memberid, u64 parentid)
289 {
290 	struct btrfs_qgroup *member;
291 	struct btrfs_qgroup *parent;
292 	struct btrfs_qgroup_list *list;
293 
294 	member = find_qgroup_rb(fs_info, memberid);
295 	parent = find_qgroup_rb(fs_info, parentid);
296 	if (!member || !parent)
297 		return -ENOENT;
298 
299 	list_for_each_entry(list, &member->groups, next_group) {
300 		if (list->group == parent) {
301 			list_del(&list->next_group);
302 			list_del(&list->next_member);
303 			kfree(list);
304 			return 0;
305 		}
306 	}
307 	return -ENOENT;
308 }
309 
310 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
311 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
312 			       u64 rfer, u64 excl)
313 {
314 	struct btrfs_qgroup *qgroup;
315 
316 	qgroup = find_qgroup_rb(fs_info, qgroupid);
317 	if (!qgroup)
318 		return -EINVAL;
319 	if (qgroup->rfer != rfer || qgroup->excl != excl)
320 		return -EINVAL;
321 	return 0;
322 }
323 #endif
324 
325 /*
326  * The full config is read in one go, only called from open_ctree()
327  * It doesn't use any locking, as at this point we're still single-threaded
328  */
329 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
330 {
331 	struct btrfs_key key;
332 	struct btrfs_key found_key;
333 	struct btrfs_root *quota_root = fs_info->quota_root;
334 	struct btrfs_path *path = NULL;
335 	struct extent_buffer *l;
336 	int slot;
337 	int ret = 0;
338 	u64 flags = 0;
339 	u64 rescan_progress = 0;
340 
341 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
342 		return 0;
343 
344 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
345 	if (!fs_info->qgroup_ulist) {
346 		ret = -ENOMEM;
347 		goto out;
348 	}
349 
350 	path = btrfs_alloc_path();
351 	if (!path) {
352 		ret = -ENOMEM;
353 		goto out;
354 	}
355 
356 	ret = btrfs_sysfs_add_qgroups(fs_info);
357 	if (ret < 0)
358 		goto out;
359 	/* default this to quota off, in case no status key is found */
360 	fs_info->qgroup_flags = 0;
361 
362 	/*
363 	 * pass 1: read status, all qgroup infos and limits
364 	 */
365 	key.objectid = 0;
366 	key.type = 0;
367 	key.offset = 0;
368 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
369 	if (ret)
370 		goto out;
371 
372 	while (1) {
373 		struct btrfs_qgroup *qgroup;
374 
375 		slot = path->slots[0];
376 		l = path->nodes[0];
377 		btrfs_item_key_to_cpu(l, &found_key, slot);
378 
379 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
380 			struct btrfs_qgroup_status_item *ptr;
381 
382 			ptr = btrfs_item_ptr(l, slot,
383 					     struct btrfs_qgroup_status_item);
384 
385 			if (btrfs_qgroup_status_version(l, ptr) !=
386 			    BTRFS_QGROUP_STATUS_VERSION) {
387 				btrfs_err(fs_info,
388 				 "old qgroup version, quota disabled");
389 				goto out;
390 			}
391 			if (btrfs_qgroup_status_generation(l, ptr) !=
392 			    fs_info->generation) {
393 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
394 				btrfs_err(fs_info,
395 					"qgroup generation mismatch, marked as inconsistent");
396 			}
397 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
398 									  ptr);
399 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
400 			goto next1;
401 		}
402 
403 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
404 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
405 			goto next1;
406 
407 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
408 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
409 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
410 			btrfs_err(fs_info, "inconsistent qgroup config");
411 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
412 		}
413 		if (!qgroup) {
414 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
415 			if (IS_ERR(qgroup)) {
416 				ret = PTR_ERR(qgroup);
417 				goto out;
418 			}
419 		}
420 		ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
421 		if (ret < 0)
422 			goto out;
423 
424 		switch (found_key.type) {
425 		case BTRFS_QGROUP_INFO_KEY: {
426 			struct btrfs_qgroup_info_item *ptr;
427 
428 			ptr = btrfs_item_ptr(l, slot,
429 					     struct btrfs_qgroup_info_item);
430 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
431 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
432 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
433 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
434 			/* generation currently unused */
435 			break;
436 		}
437 		case BTRFS_QGROUP_LIMIT_KEY: {
438 			struct btrfs_qgroup_limit_item *ptr;
439 
440 			ptr = btrfs_item_ptr(l, slot,
441 					     struct btrfs_qgroup_limit_item);
442 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
443 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
444 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
445 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
446 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
447 			break;
448 		}
449 		}
450 next1:
451 		ret = btrfs_next_item(quota_root, path);
452 		if (ret < 0)
453 			goto out;
454 		if (ret)
455 			break;
456 	}
457 	btrfs_release_path(path);
458 
459 	/*
460 	 * pass 2: read all qgroup relations
461 	 */
462 	key.objectid = 0;
463 	key.type = BTRFS_QGROUP_RELATION_KEY;
464 	key.offset = 0;
465 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
466 	if (ret)
467 		goto out;
468 	while (1) {
469 		slot = path->slots[0];
470 		l = path->nodes[0];
471 		btrfs_item_key_to_cpu(l, &found_key, slot);
472 
473 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
474 			goto next2;
475 
476 		if (found_key.objectid > found_key.offset) {
477 			/* parent <- member, not needed to build config */
478 			/* FIXME should we omit the key completely? */
479 			goto next2;
480 		}
481 
482 		ret = add_relation_rb(fs_info, found_key.objectid,
483 				      found_key.offset);
484 		if (ret == -ENOENT) {
485 			btrfs_warn(fs_info,
486 				"orphan qgroup relation 0x%llx->0x%llx",
487 				found_key.objectid, found_key.offset);
488 			ret = 0;	/* ignore the error */
489 		}
490 		if (ret)
491 			goto out;
492 next2:
493 		ret = btrfs_next_item(quota_root, path);
494 		if (ret < 0)
495 			goto out;
496 		if (ret)
497 			break;
498 	}
499 out:
500 	fs_info->qgroup_flags |= flags;
501 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
502 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
503 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
504 		 ret >= 0)
505 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
506 	btrfs_free_path(path);
507 
508 	if (ret < 0) {
509 		ulist_free(fs_info->qgroup_ulist);
510 		fs_info->qgroup_ulist = NULL;
511 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
512 		btrfs_sysfs_del_qgroups(fs_info);
513 	}
514 
515 	return ret < 0 ? ret : 0;
516 }
517 
518 /*
519  * Called in close_ctree() when quota is still enabled.  This verifies we don't
520  * leak some reserved space.
521  *
522  * Return false if no reserved space is left.
523  * Return true if some reserved space is leaked.
524  */
525 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
526 {
527 	struct rb_node *node;
528 	bool ret = false;
529 
530 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
531 		return ret;
532 	/*
533 	 * Since we're unmounting, there is no race and no need to grab qgroup
534 	 * lock.  And here we don't go post-order to provide a more user
535 	 * friendly sorted result.
536 	 */
537 	for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
538 		struct btrfs_qgroup *qgroup;
539 		int i;
540 
541 		qgroup = rb_entry(node, struct btrfs_qgroup, node);
542 		for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
543 			if (qgroup->rsv.values[i]) {
544 				ret = true;
545 				btrfs_warn(fs_info,
546 		"qgroup %hu/%llu has unreleased space, type %d rsv %llu",
547 				   btrfs_qgroup_level(qgroup->qgroupid),
548 				   btrfs_qgroup_subvolid(qgroup->qgroupid),
549 				   i, qgroup->rsv.values[i]);
550 			}
551 		}
552 	}
553 	return ret;
554 }
555 
556 /*
557  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
558  * first two are in single-threaded paths.And for the third one, we have set
559  * quota_root to be null with qgroup_lock held before, so it is safe to clean
560  * up the in-memory structures without qgroup_lock held.
561  */
562 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
563 {
564 	struct rb_node *n;
565 	struct btrfs_qgroup *qgroup;
566 
567 	while ((n = rb_first(&fs_info->qgroup_tree))) {
568 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
569 		rb_erase(n, &fs_info->qgroup_tree);
570 		__del_qgroup_rb(fs_info, qgroup);
571 	}
572 	/*
573 	 * We call btrfs_free_qgroup_config() when unmounting
574 	 * filesystem and disabling quota, so we set qgroup_ulist
575 	 * to be null here to avoid double free.
576 	 */
577 	ulist_free(fs_info->qgroup_ulist);
578 	fs_info->qgroup_ulist = NULL;
579 	btrfs_sysfs_del_qgroups(fs_info);
580 }
581 
582 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
583 				    u64 dst)
584 {
585 	int ret;
586 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
587 	struct btrfs_path *path;
588 	struct btrfs_key key;
589 
590 	path = btrfs_alloc_path();
591 	if (!path)
592 		return -ENOMEM;
593 
594 	key.objectid = src;
595 	key.type = BTRFS_QGROUP_RELATION_KEY;
596 	key.offset = dst;
597 
598 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
599 
600 	btrfs_mark_buffer_dirty(path->nodes[0]);
601 
602 	btrfs_free_path(path);
603 	return ret;
604 }
605 
606 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
607 				    u64 dst)
608 {
609 	int ret;
610 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
611 	struct btrfs_path *path;
612 	struct btrfs_key key;
613 
614 	path = btrfs_alloc_path();
615 	if (!path)
616 		return -ENOMEM;
617 
618 	key.objectid = src;
619 	key.type = BTRFS_QGROUP_RELATION_KEY;
620 	key.offset = dst;
621 
622 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
623 	if (ret < 0)
624 		goto out;
625 
626 	if (ret > 0) {
627 		ret = -ENOENT;
628 		goto out;
629 	}
630 
631 	ret = btrfs_del_item(trans, quota_root, path);
632 out:
633 	btrfs_free_path(path);
634 	return ret;
635 }
636 
637 static int add_qgroup_item(struct btrfs_trans_handle *trans,
638 			   struct btrfs_root *quota_root, u64 qgroupid)
639 {
640 	int ret;
641 	struct btrfs_path *path;
642 	struct btrfs_qgroup_info_item *qgroup_info;
643 	struct btrfs_qgroup_limit_item *qgroup_limit;
644 	struct extent_buffer *leaf;
645 	struct btrfs_key key;
646 
647 	if (btrfs_is_testing(quota_root->fs_info))
648 		return 0;
649 
650 	path = btrfs_alloc_path();
651 	if (!path)
652 		return -ENOMEM;
653 
654 	key.objectid = 0;
655 	key.type = BTRFS_QGROUP_INFO_KEY;
656 	key.offset = qgroupid;
657 
658 	/*
659 	 * Avoid a transaction abort by catching -EEXIST here. In that
660 	 * case, we proceed by re-initializing the existing structure
661 	 * on disk.
662 	 */
663 
664 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
665 				      sizeof(*qgroup_info));
666 	if (ret && ret != -EEXIST)
667 		goto out;
668 
669 	leaf = path->nodes[0];
670 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
671 				 struct btrfs_qgroup_info_item);
672 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
673 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
674 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
675 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
676 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
677 
678 	btrfs_mark_buffer_dirty(leaf);
679 
680 	btrfs_release_path(path);
681 
682 	key.type = BTRFS_QGROUP_LIMIT_KEY;
683 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
684 				      sizeof(*qgroup_limit));
685 	if (ret && ret != -EEXIST)
686 		goto out;
687 
688 	leaf = path->nodes[0];
689 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
690 				  struct btrfs_qgroup_limit_item);
691 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
692 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
693 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
694 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
695 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
696 
697 	btrfs_mark_buffer_dirty(leaf);
698 
699 	ret = 0;
700 out:
701 	btrfs_free_path(path);
702 	return ret;
703 }
704 
705 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
706 {
707 	int ret;
708 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
709 	struct btrfs_path *path;
710 	struct btrfs_key key;
711 
712 	path = btrfs_alloc_path();
713 	if (!path)
714 		return -ENOMEM;
715 
716 	key.objectid = 0;
717 	key.type = BTRFS_QGROUP_INFO_KEY;
718 	key.offset = qgroupid;
719 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
720 	if (ret < 0)
721 		goto out;
722 
723 	if (ret > 0) {
724 		ret = -ENOENT;
725 		goto out;
726 	}
727 
728 	ret = btrfs_del_item(trans, quota_root, path);
729 	if (ret)
730 		goto out;
731 
732 	btrfs_release_path(path);
733 
734 	key.type = BTRFS_QGROUP_LIMIT_KEY;
735 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
736 	if (ret < 0)
737 		goto out;
738 
739 	if (ret > 0) {
740 		ret = -ENOENT;
741 		goto out;
742 	}
743 
744 	ret = btrfs_del_item(trans, quota_root, path);
745 
746 out:
747 	btrfs_free_path(path);
748 	return ret;
749 }
750 
751 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
752 				    struct btrfs_qgroup *qgroup)
753 {
754 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
755 	struct btrfs_path *path;
756 	struct btrfs_key key;
757 	struct extent_buffer *l;
758 	struct btrfs_qgroup_limit_item *qgroup_limit;
759 	int ret;
760 	int slot;
761 
762 	key.objectid = 0;
763 	key.type = BTRFS_QGROUP_LIMIT_KEY;
764 	key.offset = qgroup->qgroupid;
765 
766 	path = btrfs_alloc_path();
767 	if (!path)
768 		return -ENOMEM;
769 
770 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
771 	if (ret > 0)
772 		ret = -ENOENT;
773 
774 	if (ret)
775 		goto out;
776 
777 	l = path->nodes[0];
778 	slot = path->slots[0];
779 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
780 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
781 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
782 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
783 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
784 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
785 
786 	btrfs_mark_buffer_dirty(l);
787 
788 out:
789 	btrfs_free_path(path);
790 	return ret;
791 }
792 
793 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
794 				   struct btrfs_qgroup *qgroup)
795 {
796 	struct btrfs_fs_info *fs_info = trans->fs_info;
797 	struct btrfs_root *quota_root = fs_info->quota_root;
798 	struct btrfs_path *path;
799 	struct btrfs_key key;
800 	struct extent_buffer *l;
801 	struct btrfs_qgroup_info_item *qgroup_info;
802 	int ret;
803 	int slot;
804 
805 	if (btrfs_is_testing(fs_info))
806 		return 0;
807 
808 	key.objectid = 0;
809 	key.type = BTRFS_QGROUP_INFO_KEY;
810 	key.offset = qgroup->qgroupid;
811 
812 	path = btrfs_alloc_path();
813 	if (!path)
814 		return -ENOMEM;
815 
816 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
817 	if (ret > 0)
818 		ret = -ENOENT;
819 
820 	if (ret)
821 		goto out;
822 
823 	l = path->nodes[0];
824 	slot = path->slots[0];
825 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
826 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
827 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
828 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
829 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
830 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
831 
832 	btrfs_mark_buffer_dirty(l);
833 
834 out:
835 	btrfs_free_path(path);
836 	return ret;
837 }
838 
839 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
840 {
841 	struct btrfs_fs_info *fs_info = trans->fs_info;
842 	struct btrfs_root *quota_root = fs_info->quota_root;
843 	struct btrfs_path *path;
844 	struct btrfs_key key;
845 	struct extent_buffer *l;
846 	struct btrfs_qgroup_status_item *ptr;
847 	int ret;
848 	int slot;
849 
850 	key.objectid = 0;
851 	key.type = BTRFS_QGROUP_STATUS_KEY;
852 	key.offset = 0;
853 
854 	path = btrfs_alloc_path();
855 	if (!path)
856 		return -ENOMEM;
857 
858 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
859 	if (ret > 0)
860 		ret = -ENOENT;
861 
862 	if (ret)
863 		goto out;
864 
865 	l = path->nodes[0];
866 	slot = path->slots[0];
867 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
868 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
869 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
870 	btrfs_set_qgroup_status_rescan(l, ptr,
871 				fs_info->qgroup_rescan_progress.objectid);
872 
873 	btrfs_mark_buffer_dirty(l);
874 
875 out:
876 	btrfs_free_path(path);
877 	return ret;
878 }
879 
880 /*
881  * called with qgroup_lock held
882  */
883 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
884 				  struct btrfs_root *root)
885 {
886 	struct btrfs_path *path;
887 	struct btrfs_key key;
888 	struct extent_buffer *leaf = NULL;
889 	int ret;
890 	int nr = 0;
891 
892 	path = btrfs_alloc_path();
893 	if (!path)
894 		return -ENOMEM;
895 
896 	path->leave_spinning = 1;
897 
898 	key.objectid = 0;
899 	key.offset = 0;
900 	key.type = 0;
901 
902 	while (1) {
903 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
904 		if (ret < 0)
905 			goto out;
906 		leaf = path->nodes[0];
907 		nr = btrfs_header_nritems(leaf);
908 		if (!nr)
909 			break;
910 		/*
911 		 * delete the leaf one by one
912 		 * since the whole tree is going
913 		 * to be deleted.
914 		 */
915 		path->slots[0] = 0;
916 		ret = btrfs_del_items(trans, root, path, 0, nr);
917 		if (ret)
918 			goto out;
919 
920 		btrfs_release_path(path);
921 	}
922 	ret = 0;
923 out:
924 	btrfs_free_path(path);
925 	return ret;
926 }
927 
928 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
929 {
930 	struct btrfs_root *quota_root;
931 	struct btrfs_root *tree_root = fs_info->tree_root;
932 	struct btrfs_path *path = NULL;
933 	struct btrfs_qgroup_status_item *ptr;
934 	struct extent_buffer *leaf;
935 	struct btrfs_key key;
936 	struct btrfs_key found_key;
937 	struct btrfs_qgroup *qgroup = NULL;
938 	struct btrfs_trans_handle *trans = NULL;
939 	int ret = 0;
940 	int slot;
941 
942 	mutex_lock(&fs_info->qgroup_ioctl_lock);
943 	if (fs_info->quota_root)
944 		goto out;
945 
946 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
947 	if (!fs_info->qgroup_ulist) {
948 		ret = -ENOMEM;
949 		goto out;
950 	}
951 
952 	ret = btrfs_sysfs_add_qgroups(fs_info);
953 	if (ret < 0)
954 		goto out;
955 	/*
956 	 * 1 for quota root item
957 	 * 1 for BTRFS_QGROUP_STATUS item
958 	 *
959 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
960 	 * per subvolume. However those are not currently reserved since it
961 	 * would be a lot of overkill.
962 	 */
963 	trans = btrfs_start_transaction(tree_root, 2);
964 	if (IS_ERR(trans)) {
965 		ret = PTR_ERR(trans);
966 		trans = NULL;
967 		goto out;
968 	}
969 
970 	/*
971 	 * initially create the quota tree
972 	 */
973 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
974 	if (IS_ERR(quota_root)) {
975 		ret =  PTR_ERR(quota_root);
976 		btrfs_abort_transaction(trans, ret);
977 		goto out;
978 	}
979 
980 	path = btrfs_alloc_path();
981 	if (!path) {
982 		ret = -ENOMEM;
983 		btrfs_abort_transaction(trans, ret);
984 		goto out_free_root;
985 	}
986 
987 	key.objectid = 0;
988 	key.type = BTRFS_QGROUP_STATUS_KEY;
989 	key.offset = 0;
990 
991 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
992 				      sizeof(*ptr));
993 	if (ret) {
994 		btrfs_abort_transaction(trans, ret);
995 		goto out_free_path;
996 	}
997 
998 	leaf = path->nodes[0];
999 	ptr = btrfs_item_ptr(leaf, path->slots[0],
1000 				 struct btrfs_qgroup_status_item);
1001 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1002 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1003 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1004 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1005 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1006 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1007 
1008 	btrfs_mark_buffer_dirty(leaf);
1009 
1010 	key.objectid = 0;
1011 	key.type = BTRFS_ROOT_REF_KEY;
1012 	key.offset = 0;
1013 
1014 	btrfs_release_path(path);
1015 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1016 	if (ret > 0)
1017 		goto out_add_root;
1018 	if (ret < 0) {
1019 		btrfs_abort_transaction(trans, ret);
1020 		goto out_free_path;
1021 	}
1022 
1023 	while (1) {
1024 		slot = path->slots[0];
1025 		leaf = path->nodes[0];
1026 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1027 
1028 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
1029 			ret = add_qgroup_item(trans, quota_root,
1030 					      found_key.offset);
1031 			if (ret) {
1032 				btrfs_abort_transaction(trans, ret);
1033 				goto out_free_path;
1034 			}
1035 
1036 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
1037 			if (IS_ERR(qgroup)) {
1038 				ret = PTR_ERR(qgroup);
1039 				btrfs_abort_transaction(trans, ret);
1040 				goto out_free_path;
1041 			}
1042 			ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1043 			if (ret < 0) {
1044 				btrfs_abort_transaction(trans, ret);
1045 				goto out_free_path;
1046 			}
1047 		}
1048 		ret = btrfs_next_item(tree_root, path);
1049 		if (ret < 0) {
1050 			btrfs_abort_transaction(trans, ret);
1051 			goto out_free_path;
1052 		}
1053 		if (ret)
1054 			break;
1055 	}
1056 
1057 out_add_root:
1058 	btrfs_release_path(path);
1059 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1060 	if (ret) {
1061 		btrfs_abort_transaction(trans, ret);
1062 		goto out_free_path;
1063 	}
1064 
1065 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1066 	if (IS_ERR(qgroup)) {
1067 		ret = PTR_ERR(qgroup);
1068 		btrfs_abort_transaction(trans, ret);
1069 		goto out_free_path;
1070 	}
1071 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1072 	if (ret < 0) {
1073 		btrfs_abort_transaction(trans, ret);
1074 		goto out_free_path;
1075 	}
1076 
1077 	ret = btrfs_commit_transaction(trans);
1078 	trans = NULL;
1079 	if (ret)
1080 		goto out_free_path;
1081 
1082 	/*
1083 	 * Set quota enabled flag after committing the transaction, to avoid
1084 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1085 	 * creation.
1086 	 */
1087 	spin_lock(&fs_info->qgroup_lock);
1088 	fs_info->quota_root = quota_root;
1089 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1090 	spin_unlock(&fs_info->qgroup_lock);
1091 
1092 	ret = qgroup_rescan_init(fs_info, 0, 1);
1093 	if (!ret) {
1094 	        qgroup_rescan_zero_tracking(fs_info);
1095 		fs_info->qgroup_rescan_running = true;
1096 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1097 	                         &fs_info->qgroup_rescan_work);
1098 	}
1099 
1100 out_free_path:
1101 	btrfs_free_path(path);
1102 out_free_root:
1103 	if (ret)
1104 		btrfs_put_root(quota_root);
1105 out:
1106 	if (ret) {
1107 		ulist_free(fs_info->qgroup_ulist);
1108 		fs_info->qgroup_ulist = NULL;
1109 		if (trans)
1110 			btrfs_end_transaction(trans);
1111 		btrfs_sysfs_del_qgroups(fs_info);
1112 	}
1113 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1114 	return ret;
1115 }
1116 
1117 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1118 {
1119 	struct btrfs_root *quota_root;
1120 	struct btrfs_trans_handle *trans = NULL;
1121 	int ret = 0;
1122 
1123 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1124 	if (!fs_info->quota_root)
1125 		goto out;
1126 
1127 	/*
1128 	 * 1 For the root item
1129 	 *
1130 	 * We should also reserve enough items for the quota tree deletion in
1131 	 * btrfs_clean_quota_tree but this is not done.
1132 	 */
1133 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1134 	if (IS_ERR(trans)) {
1135 		ret = PTR_ERR(trans);
1136 		goto out;
1137 	}
1138 
1139 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1140 	btrfs_qgroup_wait_for_completion(fs_info, false);
1141 	spin_lock(&fs_info->qgroup_lock);
1142 	quota_root = fs_info->quota_root;
1143 	fs_info->quota_root = NULL;
1144 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1145 	spin_unlock(&fs_info->qgroup_lock);
1146 
1147 	btrfs_free_qgroup_config(fs_info);
1148 
1149 	ret = btrfs_clean_quota_tree(trans, quota_root);
1150 	if (ret) {
1151 		btrfs_abort_transaction(trans, ret);
1152 		goto end_trans;
1153 	}
1154 
1155 	ret = btrfs_del_root(trans, &quota_root->root_key);
1156 	if (ret) {
1157 		btrfs_abort_transaction(trans, ret);
1158 		goto end_trans;
1159 	}
1160 
1161 	list_del(&quota_root->dirty_list);
1162 
1163 	btrfs_tree_lock(quota_root->node);
1164 	btrfs_clean_tree_block(quota_root->node);
1165 	btrfs_tree_unlock(quota_root->node);
1166 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1167 
1168 	btrfs_put_root(quota_root);
1169 
1170 end_trans:
1171 	ret = btrfs_end_transaction(trans);
1172 out:
1173 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1174 	return ret;
1175 }
1176 
1177 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1178 			 struct btrfs_qgroup *qgroup)
1179 {
1180 	if (list_empty(&qgroup->dirty))
1181 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1182 }
1183 
1184 /*
1185  * The easy accounting, we're updating qgroup relationship whose child qgroup
1186  * only has exclusive extents.
1187  *
1188  * In this case, all exclusive extents will also be exclusive for parent, so
1189  * excl/rfer just get added/removed.
1190  *
1191  * So is qgroup reservation space, which should also be added/removed to
1192  * parent.
1193  * Or when child tries to release reservation space, parent will underflow its
1194  * reservation (for relationship adding case).
1195  *
1196  * Caller should hold fs_info->qgroup_lock.
1197  */
1198 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1199 				    struct ulist *tmp, u64 ref_root,
1200 				    struct btrfs_qgroup *src, int sign)
1201 {
1202 	struct btrfs_qgroup *qgroup;
1203 	struct btrfs_qgroup_list *glist;
1204 	struct ulist_node *unode;
1205 	struct ulist_iterator uiter;
1206 	u64 num_bytes = src->excl;
1207 	int ret = 0;
1208 
1209 	qgroup = find_qgroup_rb(fs_info, ref_root);
1210 	if (!qgroup)
1211 		goto out;
1212 
1213 	qgroup->rfer += sign * num_bytes;
1214 	qgroup->rfer_cmpr += sign * num_bytes;
1215 
1216 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1217 	qgroup->excl += sign * num_bytes;
1218 	qgroup->excl_cmpr += sign * num_bytes;
1219 
1220 	if (sign > 0)
1221 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1222 	else
1223 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1224 
1225 	qgroup_dirty(fs_info, qgroup);
1226 
1227 	/* Get all of the parent groups that contain this qgroup */
1228 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1229 		ret = ulist_add(tmp, glist->group->qgroupid,
1230 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1231 		if (ret < 0)
1232 			goto out;
1233 	}
1234 
1235 	/* Iterate all of the parents and adjust their reference counts */
1236 	ULIST_ITER_INIT(&uiter);
1237 	while ((unode = ulist_next(tmp, &uiter))) {
1238 		qgroup = unode_aux_to_qgroup(unode);
1239 		qgroup->rfer += sign * num_bytes;
1240 		qgroup->rfer_cmpr += sign * num_bytes;
1241 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1242 		qgroup->excl += sign * num_bytes;
1243 		if (sign > 0)
1244 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1245 		else
1246 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1247 		qgroup->excl_cmpr += sign * num_bytes;
1248 		qgroup_dirty(fs_info, qgroup);
1249 
1250 		/* Add any parents of the parents */
1251 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1252 			ret = ulist_add(tmp, glist->group->qgroupid,
1253 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1254 			if (ret < 0)
1255 				goto out;
1256 		}
1257 	}
1258 	ret = 0;
1259 out:
1260 	return ret;
1261 }
1262 
1263 
1264 /*
1265  * Quick path for updating qgroup with only excl refs.
1266  *
1267  * In that case, just update all parent will be enough.
1268  * Or we needs to do a full rescan.
1269  * Caller should also hold fs_info->qgroup_lock.
1270  *
1271  * Return 0 for quick update, return >0 for need to full rescan
1272  * and mark INCONSISTENT flag.
1273  * Return < 0 for other error.
1274  */
1275 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1276 				   struct ulist *tmp, u64 src, u64 dst,
1277 				   int sign)
1278 {
1279 	struct btrfs_qgroup *qgroup;
1280 	int ret = 1;
1281 	int err = 0;
1282 
1283 	qgroup = find_qgroup_rb(fs_info, src);
1284 	if (!qgroup)
1285 		goto out;
1286 	if (qgroup->excl == qgroup->rfer) {
1287 		ret = 0;
1288 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1289 					       qgroup, sign);
1290 		if (err < 0) {
1291 			ret = err;
1292 			goto out;
1293 		}
1294 	}
1295 out:
1296 	if (ret)
1297 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1298 	return ret;
1299 }
1300 
1301 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1302 			      u64 dst)
1303 {
1304 	struct btrfs_fs_info *fs_info = trans->fs_info;
1305 	struct btrfs_qgroup *parent;
1306 	struct btrfs_qgroup *member;
1307 	struct btrfs_qgroup_list *list;
1308 	struct ulist *tmp;
1309 	int ret = 0;
1310 
1311 	/* Check the level of src and dst first */
1312 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1313 		return -EINVAL;
1314 
1315 	tmp = ulist_alloc(GFP_KERNEL);
1316 	if (!tmp)
1317 		return -ENOMEM;
1318 
1319 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1320 	if (!fs_info->quota_root) {
1321 		ret = -ENOTCONN;
1322 		goto out;
1323 	}
1324 	member = find_qgroup_rb(fs_info, src);
1325 	parent = find_qgroup_rb(fs_info, dst);
1326 	if (!member || !parent) {
1327 		ret = -EINVAL;
1328 		goto out;
1329 	}
1330 
1331 	/* check if such qgroup relation exist firstly */
1332 	list_for_each_entry(list, &member->groups, next_group) {
1333 		if (list->group == parent) {
1334 			ret = -EEXIST;
1335 			goto out;
1336 		}
1337 	}
1338 
1339 	ret = add_qgroup_relation_item(trans, src, dst);
1340 	if (ret)
1341 		goto out;
1342 
1343 	ret = add_qgroup_relation_item(trans, dst, src);
1344 	if (ret) {
1345 		del_qgroup_relation_item(trans, src, dst);
1346 		goto out;
1347 	}
1348 
1349 	spin_lock(&fs_info->qgroup_lock);
1350 	ret = add_relation_rb(fs_info, src, dst);
1351 	if (ret < 0) {
1352 		spin_unlock(&fs_info->qgroup_lock);
1353 		goto out;
1354 	}
1355 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1356 	spin_unlock(&fs_info->qgroup_lock);
1357 out:
1358 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1359 	ulist_free(tmp);
1360 	return ret;
1361 }
1362 
1363 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1364 				 u64 dst)
1365 {
1366 	struct btrfs_fs_info *fs_info = trans->fs_info;
1367 	struct btrfs_qgroup *parent;
1368 	struct btrfs_qgroup *member;
1369 	struct btrfs_qgroup_list *list;
1370 	struct ulist *tmp;
1371 	bool found = false;
1372 	int ret = 0;
1373 	int ret2;
1374 
1375 	tmp = ulist_alloc(GFP_KERNEL);
1376 	if (!tmp)
1377 		return -ENOMEM;
1378 
1379 	if (!fs_info->quota_root) {
1380 		ret = -ENOTCONN;
1381 		goto out;
1382 	}
1383 
1384 	member = find_qgroup_rb(fs_info, src);
1385 	parent = find_qgroup_rb(fs_info, dst);
1386 	/*
1387 	 * The parent/member pair doesn't exist, then try to delete the dead
1388 	 * relation items only.
1389 	 */
1390 	if (!member || !parent)
1391 		goto delete_item;
1392 
1393 	/* check if such qgroup relation exist firstly */
1394 	list_for_each_entry(list, &member->groups, next_group) {
1395 		if (list->group == parent) {
1396 			found = true;
1397 			break;
1398 		}
1399 	}
1400 
1401 delete_item:
1402 	ret = del_qgroup_relation_item(trans, src, dst);
1403 	if (ret < 0 && ret != -ENOENT)
1404 		goto out;
1405 	ret2 = del_qgroup_relation_item(trans, dst, src);
1406 	if (ret2 < 0 && ret2 != -ENOENT)
1407 		goto out;
1408 
1409 	/* At least one deletion succeeded, return 0 */
1410 	if (!ret || !ret2)
1411 		ret = 0;
1412 
1413 	if (found) {
1414 		spin_lock(&fs_info->qgroup_lock);
1415 		del_relation_rb(fs_info, src, dst);
1416 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1417 		spin_unlock(&fs_info->qgroup_lock);
1418 	}
1419 out:
1420 	ulist_free(tmp);
1421 	return ret;
1422 }
1423 
1424 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1425 			      u64 dst)
1426 {
1427 	struct btrfs_fs_info *fs_info = trans->fs_info;
1428 	int ret = 0;
1429 
1430 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1431 	ret = __del_qgroup_relation(trans, src, dst);
1432 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1433 
1434 	return ret;
1435 }
1436 
1437 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1438 {
1439 	struct btrfs_fs_info *fs_info = trans->fs_info;
1440 	struct btrfs_root *quota_root;
1441 	struct btrfs_qgroup *qgroup;
1442 	int ret = 0;
1443 
1444 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1445 	if (!fs_info->quota_root) {
1446 		ret = -ENOTCONN;
1447 		goto out;
1448 	}
1449 	quota_root = fs_info->quota_root;
1450 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1451 	if (qgroup) {
1452 		ret = -EEXIST;
1453 		goto out;
1454 	}
1455 
1456 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1457 	if (ret)
1458 		goto out;
1459 
1460 	spin_lock(&fs_info->qgroup_lock);
1461 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1462 	spin_unlock(&fs_info->qgroup_lock);
1463 
1464 	if (IS_ERR(qgroup)) {
1465 		ret = PTR_ERR(qgroup);
1466 		goto out;
1467 	}
1468 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1469 out:
1470 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1471 	return ret;
1472 }
1473 
1474 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1475 {
1476 	struct btrfs_fs_info *fs_info = trans->fs_info;
1477 	struct btrfs_qgroup *qgroup;
1478 	struct btrfs_qgroup_list *list;
1479 	int ret = 0;
1480 
1481 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1482 	if (!fs_info->quota_root) {
1483 		ret = -ENOTCONN;
1484 		goto out;
1485 	}
1486 
1487 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1488 	if (!qgroup) {
1489 		ret = -ENOENT;
1490 		goto out;
1491 	}
1492 
1493 	/* Check if there are no children of this qgroup */
1494 	if (!list_empty(&qgroup->members)) {
1495 		ret = -EBUSY;
1496 		goto out;
1497 	}
1498 
1499 	ret = del_qgroup_item(trans, qgroupid);
1500 	if (ret && ret != -ENOENT)
1501 		goto out;
1502 
1503 	while (!list_empty(&qgroup->groups)) {
1504 		list = list_first_entry(&qgroup->groups,
1505 					struct btrfs_qgroup_list, next_group);
1506 		ret = __del_qgroup_relation(trans, qgroupid,
1507 					    list->group->qgroupid);
1508 		if (ret)
1509 			goto out;
1510 	}
1511 
1512 	spin_lock(&fs_info->qgroup_lock);
1513 	del_qgroup_rb(fs_info, qgroupid);
1514 	spin_unlock(&fs_info->qgroup_lock);
1515 out:
1516 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1517 	return ret;
1518 }
1519 
1520 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1521 		       struct btrfs_qgroup_limit *limit)
1522 {
1523 	struct btrfs_fs_info *fs_info = trans->fs_info;
1524 	struct btrfs_qgroup *qgroup;
1525 	int ret = 0;
1526 	/* Sometimes we would want to clear the limit on this qgroup.
1527 	 * To meet this requirement, we treat the -1 as a special value
1528 	 * which tell kernel to clear the limit on this qgroup.
1529 	 */
1530 	const u64 CLEAR_VALUE = -1;
1531 
1532 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1533 	if (!fs_info->quota_root) {
1534 		ret = -ENOTCONN;
1535 		goto out;
1536 	}
1537 
1538 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1539 	if (!qgroup) {
1540 		ret = -ENOENT;
1541 		goto out;
1542 	}
1543 
1544 	spin_lock(&fs_info->qgroup_lock);
1545 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1546 		if (limit->max_rfer == CLEAR_VALUE) {
1547 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1548 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1549 			qgroup->max_rfer = 0;
1550 		} else {
1551 			qgroup->max_rfer = limit->max_rfer;
1552 		}
1553 	}
1554 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1555 		if (limit->max_excl == CLEAR_VALUE) {
1556 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1557 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1558 			qgroup->max_excl = 0;
1559 		} else {
1560 			qgroup->max_excl = limit->max_excl;
1561 		}
1562 	}
1563 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1564 		if (limit->rsv_rfer == CLEAR_VALUE) {
1565 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1566 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1567 			qgroup->rsv_rfer = 0;
1568 		} else {
1569 			qgroup->rsv_rfer = limit->rsv_rfer;
1570 		}
1571 	}
1572 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1573 		if (limit->rsv_excl == CLEAR_VALUE) {
1574 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1575 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1576 			qgroup->rsv_excl = 0;
1577 		} else {
1578 			qgroup->rsv_excl = limit->rsv_excl;
1579 		}
1580 	}
1581 	qgroup->lim_flags |= limit->flags;
1582 
1583 	spin_unlock(&fs_info->qgroup_lock);
1584 
1585 	ret = update_qgroup_limit_item(trans, qgroup);
1586 	if (ret) {
1587 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1588 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1589 		       qgroupid);
1590 	}
1591 
1592 out:
1593 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1594 	return ret;
1595 }
1596 
1597 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1598 				struct btrfs_delayed_ref_root *delayed_refs,
1599 				struct btrfs_qgroup_extent_record *record)
1600 {
1601 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1602 	struct rb_node *parent_node = NULL;
1603 	struct btrfs_qgroup_extent_record *entry;
1604 	u64 bytenr = record->bytenr;
1605 
1606 	lockdep_assert_held(&delayed_refs->lock);
1607 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1608 
1609 	while (*p) {
1610 		parent_node = *p;
1611 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1612 				 node);
1613 		if (bytenr < entry->bytenr) {
1614 			p = &(*p)->rb_left;
1615 		} else if (bytenr > entry->bytenr) {
1616 			p = &(*p)->rb_right;
1617 		} else {
1618 			if (record->data_rsv && !entry->data_rsv) {
1619 				entry->data_rsv = record->data_rsv;
1620 				entry->data_rsv_refroot =
1621 					record->data_rsv_refroot;
1622 			}
1623 			return 1;
1624 		}
1625 	}
1626 
1627 	rb_link_node(&record->node, parent_node, p);
1628 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1629 	return 0;
1630 }
1631 
1632 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1633 				   struct btrfs_qgroup_extent_record *qrecord)
1634 {
1635 	struct ulist *old_root;
1636 	u64 bytenr = qrecord->bytenr;
1637 	int ret;
1638 
1639 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1640 	if (ret < 0) {
1641 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1642 		btrfs_warn(fs_info,
1643 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1644 			ret);
1645 		return 0;
1646 	}
1647 
1648 	/*
1649 	 * Here we don't need to get the lock of
1650 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1651 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1652 	 *
1653 	 * So modifying qrecord->old_roots is safe here
1654 	 */
1655 	qrecord->old_roots = old_root;
1656 	return 0;
1657 }
1658 
1659 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1660 			      u64 num_bytes, gfp_t gfp_flag)
1661 {
1662 	struct btrfs_fs_info *fs_info = trans->fs_info;
1663 	struct btrfs_qgroup_extent_record *record;
1664 	struct btrfs_delayed_ref_root *delayed_refs;
1665 	int ret;
1666 
1667 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1668 	    || bytenr == 0 || num_bytes == 0)
1669 		return 0;
1670 	record = kzalloc(sizeof(*record), gfp_flag);
1671 	if (!record)
1672 		return -ENOMEM;
1673 
1674 	delayed_refs = &trans->transaction->delayed_refs;
1675 	record->bytenr = bytenr;
1676 	record->num_bytes = num_bytes;
1677 	record->old_roots = NULL;
1678 
1679 	spin_lock(&delayed_refs->lock);
1680 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1681 	spin_unlock(&delayed_refs->lock);
1682 	if (ret > 0) {
1683 		kfree(record);
1684 		return 0;
1685 	}
1686 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1687 }
1688 
1689 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1690 				  struct extent_buffer *eb)
1691 {
1692 	struct btrfs_fs_info *fs_info = trans->fs_info;
1693 	int nr = btrfs_header_nritems(eb);
1694 	int i, extent_type, ret;
1695 	struct btrfs_key key;
1696 	struct btrfs_file_extent_item *fi;
1697 	u64 bytenr, num_bytes;
1698 
1699 	/* We can be called directly from walk_up_proc() */
1700 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1701 		return 0;
1702 
1703 	for (i = 0; i < nr; i++) {
1704 		btrfs_item_key_to_cpu(eb, &key, i);
1705 
1706 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1707 			continue;
1708 
1709 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1710 		/* filter out non qgroup-accountable extents  */
1711 		extent_type = btrfs_file_extent_type(eb, fi);
1712 
1713 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1714 			continue;
1715 
1716 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1717 		if (!bytenr)
1718 			continue;
1719 
1720 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1721 
1722 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1723 						GFP_NOFS);
1724 		if (ret)
1725 			return ret;
1726 	}
1727 	cond_resched();
1728 	return 0;
1729 }
1730 
1731 /*
1732  * Walk up the tree from the bottom, freeing leaves and any interior
1733  * nodes which have had all slots visited. If a node (leaf or
1734  * interior) is freed, the node above it will have it's slot
1735  * incremented. The root node will never be freed.
1736  *
1737  * At the end of this function, we should have a path which has all
1738  * slots incremented to the next position for a search. If we need to
1739  * read a new node it will be NULL and the node above it will have the
1740  * correct slot selected for a later read.
1741  *
1742  * If we increment the root nodes slot counter past the number of
1743  * elements, 1 is returned to signal completion of the search.
1744  */
1745 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1746 {
1747 	int level = 0;
1748 	int nr, slot;
1749 	struct extent_buffer *eb;
1750 
1751 	if (root_level == 0)
1752 		return 1;
1753 
1754 	while (level <= root_level) {
1755 		eb = path->nodes[level];
1756 		nr = btrfs_header_nritems(eb);
1757 		path->slots[level]++;
1758 		slot = path->slots[level];
1759 		if (slot >= nr || level == 0) {
1760 			/*
1761 			 * Don't free the root -  we will detect this
1762 			 * condition after our loop and return a
1763 			 * positive value for caller to stop walking the tree.
1764 			 */
1765 			if (level != root_level) {
1766 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1767 				path->locks[level] = 0;
1768 
1769 				free_extent_buffer(eb);
1770 				path->nodes[level] = NULL;
1771 				path->slots[level] = 0;
1772 			}
1773 		} else {
1774 			/*
1775 			 * We have a valid slot to walk back down
1776 			 * from. Stop here so caller can process these
1777 			 * new nodes.
1778 			 */
1779 			break;
1780 		}
1781 
1782 		level++;
1783 	}
1784 
1785 	eb = path->nodes[root_level];
1786 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1787 		return 1;
1788 
1789 	return 0;
1790 }
1791 
1792 /*
1793  * Helper function to trace a subtree tree block swap.
1794  *
1795  * The swap will happen in highest tree block, but there may be a lot of
1796  * tree blocks involved.
1797  *
1798  * For example:
1799  *  OO = Old tree blocks
1800  *  NN = New tree blocks allocated during balance
1801  *
1802  *           File tree (257)                  Reloc tree for 257
1803  * L2              OO                                NN
1804  *               /    \                            /    \
1805  * L1          OO      OO (a)                    OO      NN (a)
1806  *            / \     / \                       / \     / \
1807  * L0       OO   OO OO   OO                   OO   OO NN   NN
1808  *                  (b)  (c)                          (b)  (c)
1809  *
1810  * When calling qgroup_trace_extent_swap(), we will pass:
1811  * @src_eb = OO(a)
1812  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1813  * @dst_level = 0
1814  * @root_level = 1
1815  *
1816  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1817  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1818  *
1819  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1820  *
1821  * 1) Tree search from @src_eb
1822  *    It should acts as a simplified btrfs_search_slot().
1823  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1824  *    (first key).
1825  *
1826  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1827  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1828  *    They should be marked during previous (@dst_level = 1) iteration.
1829  *
1830  * 3) Mark file extents in leaves dirty
1831  *    We don't have good way to pick out new file extents only.
1832  *    So we still follow the old method by scanning all file extents in
1833  *    the leave.
1834  *
1835  * This function can free us from keeping two paths, thus later we only need
1836  * to care about how to iterate all new tree blocks in reloc tree.
1837  */
1838 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1839 				    struct extent_buffer *src_eb,
1840 				    struct btrfs_path *dst_path,
1841 				    int dst_level, int root_level,
1842 				    bool trace_leaf)
1843 {
1844 	struct btrfs_key key;
1845 	struct btrfs_path *src_path;
1846 	struct btrfs_fs_info *fs_info = trans->fs_info;
1847 	u32 nodesize = fs_info->nodesize;
1848 	int cur_level = root_level;
1849 	int ret;
1850 
1851 	BUG_ON(dst_level > root_level);
1852 	/* Level mismatch */
1853 	if (btrfs_header_level(src_eb) != root_level)
1854 		return -EINVAL;
1855 
1856 	src_path = btrfs_alloc_path();
1857 	if (!src_path) {
1858 		ret = -ENOMEM;
1859 		goto out;
1860 	}
1861 
1862 	if (dst_level)
1863 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1864 	else
1865 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1866 
1867 	/* For src_path */
1868 	atomic_inc(&src_eb->refs);
1869 	src_path->nodes[root_level] = src_eb;
1870 	src_path->slots[root_level] = dst_path->slots[root_level];
1871 	src_path->locks[root_level] = 0;
1872 
1873 	/* A simplified version of btrfs_search_slot() */
1874 	while (cur_level >= dst_level) {
1875 		struct btrfs_key src_key;
1876 		struct btrfs_key dst_key;
1877 
1878 		if (src_path->nodes[cur_level] == NULL) {
1879 			struct btrfs_key first_key;
1880 			struct extent_buffer *eb;
1881 			int parent_slot;
1882 			u64 child_gen;
1883 			u64 child_bytenr;
1884 
1885 			eb = src_path->nodes[cur_level + 1];
1886 			parent_slot = src_path->slots[cur_level + 1];
1887 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1888 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1889 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1890 
1891 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1892 					     cur_level, &first_key);
1893 			if (IS_ERR(eb)) {
1894 				ret = PTR_ERR(eb);
1895 				goto out;
1896 			} else if (!extent_buffer_uptodate(eb)) {
1897 				free_extent_buffer(eb);
1898 				ret = -EIO;
1899 				goto out;
1900 			}
1901 
1902 			src_path->nodes[cur_level] = eb;
1903 
1904 			btrfs_tree_read_lock(eb);
1905 			btrfs_set_lock_blocking_read(eb);
1906 			src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1907 		}
1908 
1909 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1910 		if (cur_level) {
1911 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1912 					&dst_key, dst_path->slots[cur_level]);
1913 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1914 					&src_key, src_path->slots[cur_level]);
1915 		} else {
1916 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1917 					&dst_key, dst_path->slots[cur_level]);
1918 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1919 					&src_key, src_path->slots[cur_level]);
1920 		}
1921 		/* Content mismatch, something went wrong */
1922 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1923 			ret = -ENOENT;
1924 			goto out;
1925 		}
1926 		cur_level--;
1927 	}
1928 
1929 	/*
1930 	 * Now both @dst_path and @src_path have been populated, record the tree
1931 	 * blocks for qgroup accounting.
1932 	 */
1933 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1934 			nodesize, GFP_NOFS);
1935 	if (ret < 0)
1936 		goto out;
1937 	ret = btrfs_qgroup_trace_extent(trans,
1938 			dst_path->nodes[dst_level]->start,
1939 			nodesize, GFP_NOFS);
1940 	if (ret < 0)
1941 		goto out;
1942 
1943 	/* Record leaf file extents */
1944 	if (dst_level == 0 && trace_leaf) {
1945 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1946 		if (ret < 0)
1947 			goto out;
1948 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1949 	}
1950 out:
1951 	btrfs_free_path(src_path);
1952 	return ret;
1953 }
1954 
1955 /*
1956  * Helper function to do recursive generation-aware depth-first search, to
1957  * locate all new tree blocks in a subtree of reloc tree.
1958  *
1959  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1960  *         reloc tree
1961  * L2         NN (a)
1962  *          /    \
1963  * L1    OO        NN (b)
1964  *      /  \      /  \
1965  * L0  OO  OO    OO  NN
1966  *               (c) (d)
1967  * If we pass:
1968  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1969  * @cur_level = 1
1970  * @root_level = 1
1971  *
1972  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1973  * above tree blocks along with their counter parts in file tree.
1974  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1975  * won't affect OO(c).
1976  */
1977 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1978 					   struct extent_buffer *src_eb,
1979 					   struct btrfs_path *dst_path,
1980 					   int cur_level, int root_level,
1981 					   u64 last_snapshot, bool trace_leaf)
1982 {
1983 	struct btrfs_fs_info *fs_info = trans->fs_info;
1984 	struct extent_buffer *eb;
1985 	bool need_cleanup = false;
1986 	int ret = 0;
1987 	int i;
1988 
1989 	/* Level sanity check */
1990 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1991 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1992 	    root_level < cur_level) {
1993 		btrfs_err_rl(fs_info,
1994 			"%s: bad levels, cur_level=%d root_level=%d",
1995 			__func__, cur_level, root_level);
1996 		return -EUCLEAN;
1997 	}
1998 
1999 	/* Read the tree block if needed */
2000 	if (dst_path->nodes[cur_level] == NULL) {
2001 		struct btrfs_key first_key;
2002 		int parent_slot;
2003 		u64 child_gen;
2004 		u64 child_bytenr;
2005 
2006 		/*
2007 		 * dst_path->nodes[root_level] must be initialized before
2008 		 * calling this function.
2009 		 */
2010 		if (cur_level == root_level) {
2011 			btrfs_err_rl(fs_info,
2012 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2013 				__func__, root_level, root_level, cur_level);
2014 			return -EUCLEAN;
2015 		}
2016 
2017 		/*
2018 		 * We need to get child blockptr/gen from parent before we can
2019 		 * read it.
2020 		  */
2021 		eb = dst_path->nodes[cur_level + 1];
2022 		parent_slot = dst_path->slots[cur_level + 1];
2023 		child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2024 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2025 		btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2026 
2027 		/* This node is old, no need to trace */
2028 		if (child_gen < last_snapshot)
2029 			goto out;
2030 
2031 		eb = read_tree_block(fs_info, child_bytenr, child_gen,
2032 				     cur_level, &first_key);
2033 		if (IS_ERR(eb)) {
2034 			ret = PTR_ERR(eb);
2035 			goto out;
2036 		} else if (!extent_buffer_uptodate(eb)) {
2037 			free_extent_buffer(eb);
2038 			ret = -EIO;
2039 			goto out;
2040 		}
2041 
2042 		dst_path->nodes[cur_level] = eb;
2043 		dst_path->slots[cur_level] = 0;
2044 
2045 		btrfs_tree_read_lock(eb);
2046 		btrfs_set_lock_blocking_read(eb);
2047 		dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
2048 		need_cleanup = true;
2049 	}
2050 
2051 	/* Now record this tree block and its counter part for qgroups */
2052 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2053 				       root_level, trace_leaf);
2054 	if (ret < 0)
2055 		goto cleanup;
2056 
2057 	eb = dst_path->nodes[cur_level];
2058 
2059 	if (cur_level > 0) {
2060 		/* Iterate all child tree blocks */
2061 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2062 			/* Skip old tree blocks as they won't be swapped */
2063 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2064 				continue;
2065 			dst_path->slots[cur_level] = i;
2066 
2067 			/* Recursive call (at most 7 times) */
2068 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2069 					dst_path, cur_level - 1, root_level,
2070 					last_snapshot, trace_leaf);
2071 			if (ret < 0)
2072 				goto cleanup;
2073 		}
2074 	}
2075 
2076 cleanup:
2077 	if (need_cleanup) {
2078 		/* Clean up */
2079 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2080 				     dst_path->locks[cur_level]);
2081 		free_extent_buffer(dst_path->nodes[cur_level]);
2082 		dst_path->nodes[cur_level] = NULL;
2083 		dst_path->slots[cur_level] = 0;
2084 		dst_path->locks[cur_level] = 0;
2085 	}
2086 out:
2087 	return ret;
2088 }
2089 
2090 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2091 				struct extent_buffer *src_eb,
2092 				struct extent_buffer *dst_eb,
2093 				u64 last_snapshot, bool trace_leaf)
2094 {
2095 	struct btrfs_fs_info *fs_info = trans->fs_info;
2096 	struct btrfs_path *dst_path = NULL;
2097 	int level;
2098 	int ret;
2099 
2100 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2101 		return 0;
2102 
2103 	/* Wrong parameter order */
2104 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2105 		btrfs_err_rl(fs_info,
2106 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2107 			     btrfs_header_generation(src_eb),
2108 			     btrfs_header_generation(dst_eb));
2109 		return -EUCLEAN;
2110 	}
2111 
2112 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2113 		ret = -EIO;
2114 		goto out;
2115 	}
2116 
2117 	level = btrfs_header_level(dst_eb);
2118 	dst_path = btrfs_alloc_path();
2119 	if (!dst_path) {
2120 		ret = -ENOMEM;
2121 		goto out;
2122 	}
2123 	/* For dst_path */
2124 	atomic_inc(&dst_eb->refs);
2125 	dst_path->nodes[level] = dst_eb;
2126 	dst_path->slots[level] = 0;
2127 	dst_path->locks[level] = 0;
2128 
2129 	/* Do the generation aware breadth-first search */
2130 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2131 					      level, last_snapshot, trace_leaf);
2132 	if (ret < 0)
2133 		goto out;
2134 	ret = 0;
2135 
2136 out:
2137 	btrfs_free_path(dst_path);
2138 	if (ret < 0)
2139 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2140 	return ret;
2141 }
2142 
2143 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2144 			       struct extent_buffer *root_eb,
2145 			       u64 root_gen, int root_level)
2146 {
2147 	struct btrfs_fs_info *fs_info = trans->fs_info;
2148 	int ret = 0;
2149 	int level;
2150 	struct extent_buffer *eb = root_eb;
2151 	struct btrfs_path *path = NULL;
2152 
2153 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2154 	BUG_ON(root_eb == NULL);
2155 
2156 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2157 		return 0;
2158 
2159 	if (!extent_buffer_uptodate(root_eb)) {
2160 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2161 		if (ret)
2162 			goto out;
2163 	}
2164 
2165 	if (root_level == 0) {
2166 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2167 		goto out;
2168 	}
2169 
2170 	path = btrfs_alloc_path();
2171 	if (!path)
2172 		return -ENOMEM;
2173 
2174 	/*
2175 	 * Walk down the tree.  Missing extent blocks are filled in as
2176 	 * we go. Metadata is accounted every time we read a new
2177 	 * extent block.
2178 	 *
2179 	 * When we reach a leaf, we account for file extent items in it,
2180 	 * walk back up the tree (adjusting slot pointers as we go)
2181 	 * and restart the search process.
2182 	 */
2183 	atomic_inc(&root_eb->refs);	/* For path */
2184 	path->nodes[root_level] = root_eb;
2185 	path->slots[root_level] = 0;
2186 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2187 walk_down:
2188 	level = root_level;
2189 	while (level >= 0) {
2190 		if (path->nodes[level] == NULL) {
2191 			struct btrfs_key first_key;
2192 			int parent_slot;
2193 			u64 child_gen;
2194 			u64 child_bytenr;
2195 
2196 			/*
2197 			 * We need to get child blockptr/gen from parent before
2198 			 * we can read it.
2199 			  */
2200 			eb = path->nodes[level + 1];
2201 			parent_slot = path->slots[level + 1];
2202 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2203 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2204 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2205 
2206 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
2207 					     level, &first_key);
2208 			if (IS_ERR(eb)) {
2209 				ret = PTR_ERR(eb);
2210 				goto out;
2211 			} else if (!extent_buffer_uptodate(eb)) {
2212 				free_extent_buffer(eb);
2213 				ret = -EIO;
2214 				goto out;
2215 			}
2216 
2217 			path->nodes[level] = eb;
2218 			path->slots[level] = 0;
2219 
2220 			btrfs_tree_read_lock(eb);
2221 			btrfs_set_lock_blocking_read(eb);
2222 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2223 
2224 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2225 							fs_info->nodesize,
2226 							GFP_NOFS);
2227 			if (ret)
2228 				goto out;
2229 		}
2230 
2231 		if (level == 0) {
2232 			ret = btrfs_qgroup_trace_leaf_items(trans,
2233 							    path->nodes[level]);
2234 			if (ret)
2235 				goto out;
2236 
2237 			/* Nonzero return here means we completed our search */
2238 			ret = adjust_slots_upwards(path, root_level);
2239 			if (ret)
2240 				break;
2241 
2242 			/* Restart search with new slots */
2243 			goto walk_down;
2244 		}
2245 
2246 		level--;
2247 	}
2248 
2249 	ret = 0;
2250 out:
2251 	btrfs_free_path(path);
2252 
2253 	return ret;
2254 }
2255 
2256 #define UPDATE_NEW	0
2257 #define UPDATE_OLD	1
2258 /*
2259  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2260  */
2261 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2262 				struct ulist *roots, struct ulist *tmp,
2263 				struct ulist *qgroups, u64 seq, int update_old)
2264 {
2265 	struct ulist_node *unode;
2266 	struct ulist_iterator uiter;
2267 	struct ulist_node *tmp_unode;
2268 	struct ulist_iterator tmp_uiter;
2269 	struct btrfs_qgroup *qg;
2270 	int ret = 0;
2271 
2272 	if (!roots)
2273 		return 0;
2274 	ULIST_ITER_INIT(&uiter);
2275 	while ((unode = ulist_next(roots, &uiter))) {
2276 		qg = find_qgroup_rb(fs_info, unode->val);
2277 		if (!qg)
2278 			continue;
2279 
2280 		ulist_reinit(tmp);
2281 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2282 				GFP_ATOMIC);
2283 		if (ret < 0)
2284 			return ret;
2285 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2286 		if (ret < 0)
2287 			return ret;
2288 		ULIST_ITER_INIT(&tmp_uiter);
2289 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2290 			struct btrfs_qgroup_list *glist;
2291 
2292 			qg = unode_aux_to_qgroup(tmp_unode);
2293 			if (update_old)
2294 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2295 			else
2296 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2297 			list_for_each_entry(glist, &qg->groups, next_group) {
2298 				ret = ulist_add(qgroups, glist->group->qgroupid,
2299 						qgroup_to_aux(glist->group),
2300 						GFP_ATOMIC);
2301 				if (ret < 0)
2302 					return ret;
2303 				ret = ulist_add(tmp, glist->group->qgroupid,
2304 						qgroup_to_aux(glist->group),
2305 						GFP_ATOMIC);
2306 				if (ret < 0)
2307 					return ret;
2308 			}
2309 		}
2310 	}
2311 	return 0;
2312 }
2313 
2314 /*
2315  * Update qgroup rfer/excl counters.
2316  * Rfer update is easy, codes can explain themselves.
2317  *
2318  * Excl update is tricky, the update is split into 2 part.
2319  * Part 1: Possible exclusive <-> sharing detect:
2320  *	|	A	|	!A	|
2321  *  -------------------------------------
2322  *  B	|	*	|	-	|
2323  *  -------------------------------------
2324  *  !B	|	+	|	**	|
2325  *  -------------------------------------
2326  *
2327  * Conditions:
2328  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2329  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2330  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2331  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2332  *
2333  * Results:
2334  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2335  * *: Definitely not changed.		**: Possible unchanged.
2336  *
2337  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2338  *
2339  * To make the logic clear, we first use condition A and B to split
2340  * combination into 4 results.
2341  *
2342  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2343  * only on variant maybe 0.
2344  *
2345  * Lastly, check result **, since there are 2 variants maybe 0, split them
2346  * again(2x2).
2347  * But this time we don't need to consider other things, the codes and logic
2348  * is easy to understand now.
2349  */
2350 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2351 				  struct ulist *qgroups,
2352 				  u64 nr_old_roots,
2353 				  u64 nr_new_roots,
2354 				  u64 num_bytes, u64 seq)
2355 {
2356 	struct ulist_node *unode;
2357 	struct ulist_iterator uiter;
2358 	struct btrfs_qgroup *qg;
2359 	u64 cur_new_count, cur_old_count;
2360 
2361 	ULIST_ITER_INIT(&uiter);
2362 	while ((unode = ulist_next(qgroups, &uiter))) {
2363 		bool dirty = false;
2364 
2365 		qg = unode_aux_to_qgroup(unode);
2366 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2367 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2368 
2369 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2370 					     cur_new_count);
2371 
2372 		/* Rfer update part */
2373 		if (cur_old_count == 0 && cur_new_count > 0) {
2374 			qg->rfer += num_bytes;
2375 			qg->rfer_cmpr += num_bytes;
2376 			dirty = true;
2377 		}
2378 		if (cur_old_count > 0 && cur_new_count == 0) {
2379 			qg->rfer -= num_bytes;
2380 			qg->rfer_cmpr -= num_bytes;
2381 			dirty = true;
2382 		}
2383 
2384 		/* Excl update part */
2385 		/* Exclusive/none -> shared case */
2386 		if (cur_old_count == nr_old_roots &&
2387 		    cur_new_count < nr_new_roots) {
2388 			/* Exclusive -> shared */
2389 			if (cur_old_count != 0) {
2390 				qg->excl -= num_bytes;
2391 				qg->excl_cmpr -= num_bytes;
2392 				dirty = true;
2393 			}
2394 		}
2395 
2396 		/* Shared -> exclusive/none case */
2397 		if (cur_old_count < nr_old_roots &&
2398 		    cur_new_count == nr_new_roots) {
2399 			/* Shared->exclusive */
2400 			if (cur_new_count != 0) {
2401 				qg->excl += num_bytes;
2402 				qg->excl_cmpr += num_bytes;
2403 				dirty = true;
2404 			}
2405 		}
2406 
2407 		/* Exclusive/none -> exclusive/none case */
2408 		if (cur_old_count == nr_old_roots &&
2409 		    cur_new_count == nr_new_roots) {
2410 			if (cur_old_count == 0) {
2411 				/* None -> exclusive/none */
2412 
2413 				if (cur_new_count != 0) {
2414 					/* None -> exclusive */
2415 					qg->excl += num_bytes;
2416 					qg->excl_cmpr += num_bytes;
2417 					dirty = true;
2418 				}
2419 				/* None -> none, nothing changed */
2420 			} else {
2421 				/* Exclusive -> exclusive/none */
2422 
2423 				if (cur_new_count == 0) {
2424 					/* Exclusive -> none */
2425 					qg->excl -= num_bytes;
2426 					qg->excl_cmpr -= num_bytes;
2427 					dirty = true;
2428 				}
2429 				/* Exclusive -> exclusive, nothing changed */
2430 			}
2431 		}
2432 
2433 		if (dirty)
2434 			qgroup_dirty(fs_info, qg);
2435 	}
2436 	return 0;
2437 }
2438 
2439 /*
2440  * Check if the @roots potentially is a list of fs tree roots
2441  *
2442  * Return 0 for definitely not a fs/subvol tree roots ulist
2443  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2444  *          one as well)
2445  */
2446 static int maybe_fs_roots(struct ulist *roots)
2447 {
2448 	struct ulist_node *unode;
2449 	struct ulist_iterator uiter;
2450 
2451 	/* Empty one, still possible for fs roots */
2452 	if (!roots || roots->nnodes == 0)
2453 		return 1;
2454 
2455 	ULIST_ITER_INIT(&uiter);
2456 	unode = ulist_next(roots, &uiter);
2457 	if (!unode)
2458 		return 1;
2459 
2460 	/*
2461 	 * If it contains fs tree roots, then it must belong to fs/subvol
2462 	 * trees.
2463 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2464 	 */
2465 	return is_fstree(unode->val);
2466 }
2467 
2468 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2469 				u64 num_bytes, struct ulist *old_roots,
2470 				struct ulist *new_roots)
2471 {
2472 	struct btrfs_fs_info *fs_info = trans->fs_info;
2473 	struct ulist *qgroups = NULL;
2474 	struct ulist *tmp = NULL;
2475 	u64 seq;
2476 	u64 nr_new_roots = 0;
2477 	u64 nr_old_roots = 0;
2478 	int ret = 0;
2479 
2480 	/*
2481 	 * If quotas get disabled meanwhile, the resouces need to be freed and
2482 	 * we can't just exit here.
2483 	 */
2484 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2485 		goto out_free;
2486 
2487 	if (new_roots) {
2488 		if (!maybe_fs_roots(new_roots))
2489 			goto out_free;
2490 		nr_new_roots = new_roots->nnodes;
2491 	}
2492 	if (old_roots) {
2493 		if (!maybe_fs_roots(old_roots))
2494 			goto out_free;
2495 		nr_old_roots = old_roots->nnodes;
2496 	}
2497 
2498 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2499 	if (nr_old_roots == 0 && nr_new_roots == 0)
2500 		goto out_free;
2501 
2502 	BUG_ON(!fs_info->quota_root);
2503 
2504 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2505 					num_bytes, nr_old_roots, nr_new_roots);
2506 
2507 	qgroups = ulist_alloc(GFP_NOFS);
2508 	if (!qgroups) {
2509 		ret = -ENOMEM;
2510 		goto out_free;
2511 	}
2512 	tmp = ulist_alloc(GFP_NOFS);
2513 	if (!tmp) {
2514 		ret = -ENOMEM;
2515 		goto out_free;
2516 	}
2517 
2518 	mutex_lock(&fs_info->qgroup_rescan_lock);
2519 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2520 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2521 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2522 			ret = 0;
2523 			goto out_free;
2524 		}
2525 	}
2526 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2527 
2528 	spin_lock(&fs_info->qgroup_lock);
2529 	seq = fs_info->qgroup_seq;
2530 
2531 	/* Update old refcnts using old_roots */
2532 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2533 				   UPDATE_OLD);
2534 	if (ret < 0)
2535 		goto out;
2536 
2537 	/* Update new refcnts using new_roots */
2538 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2539 				   UPDATE_NEW);
2540 	if (ret < 0)
2541 		goto out;
2542 
2543 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2544 			       num_bytes, seq);
2545 
2546 	/*
2547 	 * Bump qgroup_seq to avoid seq overlap
2548 	 */
2549 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2550 out:
2551 	spin_unlock(&fs_info->qgroup_lock);
2552 out_free:
2553 	ulist_free(tmp);
2554 	ulist_free(qgroups);
2555 	ulist_free(old_roots);
2556 	ulist_free(new_roots);
2557 	return ret;
2558 }
2559 
2560 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2561 {
2562 	struct btrfs_fs_info *fs_info = trans->fs_info;
2563 	struct btrfs_qgroup_extent_record *record;
2564 	struct btrfs_delayed_ref_root *delayed_refs;
2565 	struct ulist *new_roots = NULL;
2566 	struct rb_node *node;
2567 	u64 num_dirty_extents = 0;
2568 	u64 qgroup_to_skip;
2569 	int ret = 0;
2570 
2571 	delayed_refs = &trans->transaction->delayed_refs;
2572 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2573 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2574 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2575 				  node);
2576 
2577 		num_dirty_extents++;
2578 		trace_btrfs_qgroup_account_extents(fs_info, record);
2579 
2580 		if (!ret) {
2581 			/*
2582 			 * Old roots should be searched when inserting qgroup
2583 			 * extent record
2584 			 */
2585 			if (WARN_ON(!record->old_roots)) {
2586 				/* Search commit root to find old_roots */
2587 				ret = btrfs_find_all_roots(NULL, fs_info,
2588 						record->bytenr, 0,
2589 						&record->old_roots, false);
2590 				if (ret < 0)
2591 					goto cleanup;
2592 			}
2593 
2594 			/* Free the reserved data space */
2595 			btrfs_qgroup_free_refroot(fs_info,
2596 					record->data_rsv_refroot,
2597 					record->data_rsv,
2598 					BTRFS_QGROUP_RSV_DATA);
2599 			/*
2600 			 * Use SEQ_LAST as time_seq to do special search, which
2601 			 * doesn't lock tree or delayed_refs and search current
2602 			 * root. It's safe inside commit_transaction().
2603 			 */
2604 			ret = btrfs_find_all_roots(trans, fs_info,
2605 				record->bytenr, SEQ_LAST, &new_roots, false);
2606 			if (ret < 0)
2607 				goto cleanup;
2608 			if (qgroup_to_skip) {
2609 				ulist_del(new_roots, qgroup_to_skip, 0);
2610 				ulist_del(record->old_roots, qgroup_to_skip,
2611 					  0);
2612 			}
2613 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2614 							  record->num_bytes,
2615 							  record->old_roots,
2616 							  new_roots);
2617 			record->old_roots = NULL;
2618 			new_roots = NULL;
2619 		}
2620 cleanup:
2621 		ulist_free(record->old_roots);
2622 		ulist_free(new_roots);
2623 		new_roots = NULL;
2624 		rb_erase(node, &delayed_refs->dirty_extent_root);
2625 		kfree(record);
2626 
2627 	}
2628 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2629 				       num_dirty_extents);
2630 	return ret;
2631 }
2632 
2633 /*
2634  * called from commit_transaction. Writes all changed qgroups to disk.
2635  */
2636 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2637 {
2638 	struct btrfs_fs_info *fs_info = trans->fs_info;
2639 	int ret = 0;
2640 
2641 	if (!fs_info->quota_root)
2642 		return ret;
2643 
2644 	spin_lock(&fs_info->qgroup_lock);
2645 	while (!list_empty(&fs_info->dirty_qgroups)) {
2646 		struct btrfs_qgroup *qgroup;
2647 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2648 					  struct btrfs_qgroup, dirty);
2649 		list_del_init(&qgroup->dirty);
2650 		spin_unlock(&fs_info->qgroup_lock);
2651 		ret = update_qgroup_info_item(trans, qgroup);
2652 		if (ret)
2653 			fs_info->qgroup_flags |=
2654 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2655 		ret = update_qgroup_limit_item(trans, qgroup);
2656 		if (ret)
2657 			fs_info->qgroup_flags |=
2658 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2659 		spin_lock(&fs_info->qgroup_lock);
2660 	}
2661 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2662 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2663 	else
2664 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2665 	spin_unlock(&fs_info->qgroup_lock);
2666 
2667 	ret = update_qgroup_status_item(trans);
2668 	if (ret)
2669 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2670 
2671 	return ret;
2672 }
2673 
2674 /*
2675  * Copy the accounting information between qgroups. This is necessary
2676  * when a snapshot or a subvolume is created. Throwing an error will
2677  * cause a transaction abort so we take extra care here to only error
2678  * when a readonly fs is a reasonable outcome.
2679  */
2680 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2681 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2682 {
2683 	int ret = 0;
2684 	int i;
2685 	u64 *i_qgroups;
2686 	bool committing = false;
2687 	struct btrfs_fs_info *fs_info = trans->fs_info;
2688 	struct btrfs_root *quota_root;
2689 	struct btrfs_qgroup *srcgroup;
2690 	struct btrfs_qgroup *dstgroup;
2691 	bool need_rescan = false;
2692 	u32 level_size = 0;
2693 	u64 nums;
2694 
2695 	/*
2696 	 * There are only two callers of this function.
2697 	 *
2698 	 * One in create_subvol() in the ioctl context, which needs to hold
2699 	 * the qgroup_ioctl_lock.
2700 	 *
2701 	 * The other one in create_pending_snapshot() where no other qgroup
2702 	 * code can modify the fs as they all need to either start a new trans
2703 	 * or hold a trans handler, thus we don't need to hold
2704 	 * qgroup_ioctl_lock.
2705 	 * This would avoid long and complex lock chain and make lockdep happy.
2706 	 */
2707 	spin_lock(&fs_info->trans_lock);
2708 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2709 		committing = true;
2710 	spin_unlock(&fs_info->trans_lock);
2711 
2712 	if (!committing)
2713 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2714 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2715 		goto out;
2716 
2717 	quota_root = fs_info->quota_root;
2718 	if (!quota_root) {
2719 		ret = -EINVAL;
2720 		goto out;
2721 	}
2722 
2723 	if (inherit) {
2724 		i_qgroups = (u64 *)(inherit + 1);
2725 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2726 		       2 * inherit->num_excl_copies;
2727 		for (i = 0; i < nums; ++i) {
2728 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2729 
2730 			/*
2731 			 * Zero out invalid groups so we can ignore
2732 			 * them later.
2733 			 */
2734 			if (!srcgroup ||
2735 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2736 				*i_qgroups = 0ULL;
2737 
2738 			++i_qgroups;
2739 		}
2740 	}
2741 
2742 	/*
2743 	 * create a tracking group for the subvol itself
2744 	 */
2745 	ret = add_qgroup_item(trans, quota_root, objectid);
2746 	if (ret)
2747 		goto out;
2748 
2749 	/*
2750 	 * add qgroup to all inherited groups
2751 	 */
2752 	if (inherit) {
2753 		i_qgroups = (u64 *)(inherit + 1);
2754 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2755 			if (*i_qgroups == 0)
2756 				continue;
2757 			ret = add_qgroup_relation_item(trans, objectid,
2758 						       *i_qgroups);
2759 			if (ret && ret != -EEXIST)
2760 				goto out;
2761 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2762 						       objectid);
2763 			if (ret && ret != -EEXIST)
2764 				goto out;
2765 		}
2766 		ret = 0;
2767 	}
2768 
2769 
2770 	spin_lock(&fs_info->qgroup_lock);
2771 
2772 	dstgroup = add_qgroup_rb(fs_info, objectid);
2773 	if (IS_ERR(dstgroup)) {
2774 		ret = PTR_ERR(dstgroup);
2775 		goto unlock;
2776 	}
2777 
2778 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2779 		dstgroup->lim_flags = inherit->lim.flags;
2780 		dstgroup->max_rfer = inherit->lim.max_rfer;
2781 		dstgroup->max_excl = inherit->lim.max_excl;
2782 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2783 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2784 
2785 		ret = update_qgroup_limit_item(trans, dstgroup);
2786 		if (ret) {
2787 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2788 			btrfs_info(fs_info,
2789 				   "unable to update quota limit for %llu",
2790 				   dstgroup->qgroupid);
2791 			goto unlock;
2792 		}
2793 	}
2794 
2795 	if (srcid) {
2796 		srcgroup = find_qgroup_rb(fs_info, srcid);
2797 		if (!srcgroup)
2798 			goto unlock;
2799 
2800 		/*
2801 		 * We call inherit after we clone the root in order to make sure
2802 		 * our counts don't go crazy, so at this point the only
2803 		 * difference between the two roots should be the root node.
2804 		 */
2805 		level_size = fs_info->nodesize;
2806 		dstgroup->rfer = srcgroup->rfer;
2807 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2808 		dstgroup->excl = level_size;
2809 		dstgroup->excl_cmpr = level_size;
2810 		srcgroup->excl = level_size;
2811 		srcgroup->excl_cmpr = level_size;
2812 
2813 		/* inherit the limit info */
2814 		dstgroup->lim_flags = srcgroup->lim_flags;
2815 		dstgroup->max_rfer = srcgroup->max_rfer;
2816 		dstgroup->max_excl = srcgroup->max_excl;
2817 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2818 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2819 
2820 		qgroup_dirty(fs_info, dstgroup);
2821 		qgroup_dirty(fs_info, srcgroup);
2822 	}
2823 
2824 	if (!inherit)
2825 		goto unlock;
2826 
2827 	i_qgroups = (u64 *)(inherit + 1);
2828 	for (i = 0; i < inherit->num_qgroups; ++i) {
2829 		if (*i_qgroups) {
2830 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2831 			if (ret)
2832 				goto unlock;
2833 		}
2834 		++i_qgroups;
2835 
2836 		/*
2837 		 * If we're doing a snapshot, and adding the snapshot to a new
2838 		 * qgroup, the numbers are guaranteed to be incorrect.
2839 		 */
2840 		if (srcid)
2841 			need_rescan = true;
2842 	}
2843 
2844 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2845 		struct btrfs_qgroup *src;
2846 		struct btrfs_qgroup *dst;
2847 
2848 		if (!i_qgroups[0] || !i_qgroups[1])
2849 			continue;
2850 
2851 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2852 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2853 
2854 		if (!src || !dst) {
2855 			ret = -EINVAL;
2856 			goto unlock;
2857 		}
2858 
2859 		dst->rfer = src->rfer - level_size;
2860 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2861 
2862 		/* Manually tweaking numbers certainly needs a rescan */
2863 		need_rescan = true;
2864 	}
2865 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2866 		struct btrfs_qgroup *src;
2867 		struct btrfs_qgroup *dst;
2868 
2869 		if (!i_qgroups[0] || !i_qgroups[1])
2870 			continue;
2871 
2872 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2873 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2874 
2875 		if (!src || !dst) {
2876 			ret = -EINVAL;
2877 			goto unlock;
2878 		}
2879 
2880 		dst->excl = src->excl + level_size;
2881 		dst->excl_cmpr = src->excl_cmpr + level_size;
2882 		need_rescan = true;
2883 	}
2884 
2885 unlock:
2886 	spin_unlock(&fs_info->qgroup_lock);
2887 	if (!ret)
2888 		ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
2889 out:
2890 	if (!committing)
2891 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
2892 	if (need_rescan)
2893 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2894 	return ret;
2895 }
2896 
2897 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2898 {
2899 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2900 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2901 		return false;
2902 
2903 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2904 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2905 		return false;
2906 
2907 	return true;
2908 }
2909 
2910 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2911 			  enum btrfs_qgroup_rsv_type type)
2912 {
2913 	struct btrfs_qgroup *qgroup;
2914 	struct btrfs_fs_info *fs_info = root->fs_info;
2915 	u64 ref_root = root->root_key.objectid;
2916 	int ret = 0;
2917 	struct ulist_node *unode;
2918 	struct ulist_iterator uiter;
2919 
2920 	if (!is_fstree(ref_root))
2921 		return 0;
2922 
2923 	if (num_bytes == 0)
2924 		return 0;
2925 
2926 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2927 	    capable(CAP_SYS_RESOURCE))
2928 		enforce = false;
2929 
2930 	spin_lock(&fs_info->qgroup_lock);
2931 	if (!fs_info->quota_root)
2932 		goto out;
2933 
2934 	qgroup = find_qgroup_rb(fs_info, ref_root);
2935 	if (!qgroup)
2936 		goto out;
2937 
2938 	/*
2939 	 * in a first step, we check all affected qgroups if any limits would
2940 	 * be exceeded
2941 	 */
2942 	ulist_reinit(fs_info->qgroup_ulist);
2943 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2944 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2945 	if (ret < 0)
2946 		goto out;
2947 	ULIST_ITER_INIT(&uiter);
2948 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2949 		struct btrfs_qgroup *qg;
2950 		struct btrfs_qgroup_list *glist;
2951 
2952 		qg = unode_aux_to_qgroup(unode);
2953 
2954 		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2955 			ret = -EDQUOT;
2956 			goto out;
2957 		}
2958 
2959 		list_for_each_entry(glist, &qg->groups, next_group) {
2960 			ret = ulist_add(fs_info->qgroup_ulist,
2961 					glist->group->qgroupid,
2962 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2963 			if (ret < 0)
2964 				goto out;
2965 		}
2966 	}
2967 	ret = 0;
2968 	/*
2969 	 * no limits exceeded, now record the reservation into all qgroups
2970 	 */
2971 	ULIST_ITER_INIT(&uiter);
2972 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2973 		struct btrfs_qgroup *qg;
2974 
2975 		qg = unode_aux_to_qgroup(unode);
2976 
2977 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
2978 	}
2979 
2980 out:
2981 	spin_unlock(&fs_info->qgroup_lock);
2982 	return ret;
2983 }
2984 
2985 /*
2986  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2987  * qgroup).
2988  *
2989  * Will handle all higher level qgroup too.
2990  *
2991  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2992  * This special case is only used for META_PERTRANS type.
2993  */
2994 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2995 			       u64 ref_root, u64 num_bytes,
2996 			       enum btrfs_qgroup_rsv_type type)
2997 {
2998 	struct btrfs_qgroup *qgroup;
2999 	struct ulist_node *unode;
3000 	struct ulist_iterator uiter;
3001 	int ret = 0;
3002 
3003 	if (!is_fstree(ref_root))
3004 		return;
3005 
3006 	if (num_bytes == 0)
3007 		return;
3008 
3009 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3010 		WARN(1, "%s: Invalid type to free", __func__);
3011 		return;
3012 	}
3013 	spin_lock(&fs_info->qgroup_lock);
3014 
3015 	if (!fs_info->quota_root)
3016 		goto out;
3017 
3018 	qgroup = find_qgroup_rb(fs_info, ref_root);
3019 	if (!qgroup)
3020 		goto out;
3021 
3022 	if (num_bytes == (u64)-1)
3023 		/*
3024 		 * We're freeing all pertrans rsv, get reserved value from
3025 		 * level 0 qgroup as real num_bytes to free.
3026 		 */
3027 		num_bytes = qgroup->rsv.values[type];
3028 
3029 	ulist_reinit(fs_info->qgroup_ulist);
3030 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3031 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3032 	if (ret < 0)
3033 		goto out;
3034 	ULIST_ITER_INIT(&uiter);
3035 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3036 		struct btrfs_qgroup *qg;
3037 		struct btrfs_qgroup_list *glist;
3038 
3039 		qg = unode_aux_to_qgroup(unode);
3040 
3041 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3042 
3043 		list_for_each_entry(glist, &qg->groups, next_group) {
3044 			ret = ulist_add(fs_info->qgroup_ulist,
3045 					glist->group->qgroupid,
3046 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3047 			if (ret < 0)
3048 				goto out;
3049 		}
3050 	}
3051 
3052 out:
3053 	spin_unlock(&fs_info->qgroup_lock);
3054 }
3055 
3056 /*
3057  * Check if the leaf is the last leaf. Which means all node pointers
3058  * are at their last position.
3059  */
3060 static bool is_last_leaf(struct btrfs_path *path)
3061 {
3062 	int i;
3063 
3064 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3065 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3066 			return false;
3067 	}
3068 	return true;
3069 }
3070 
3071 /*
3072  * returns < 0 on error, 0 when more leafs are to be scanned.
3073  * returns 1 when done.
3074  */
3075 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3076 			      struct btrfs_path *path)
3077 {
3078 	struct btrfs_fs_info *fs_info = trans->fs_info;
3079 	struct btrfs_key found;
3080 	struct extent_buffer *scratch_leaf = NULL;
3081 	struct ulist *roots = NULL;
3082 	u64 num_bytes;
3083 	bool done;
3084 	int slot;
3085 	int ret;
3086 
3087 	mutex_lock(&fs_info->qgroup_rescan_lock);
3088 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3089 					 &fs_info->qgroup_rescan_progress,
3090 					 path, 1, 0);
3091 
3092 	btrfs_debug(fs_info,
3093 		"current progress key (%llu %u %llu), search_slot ret %d",
3094 		fs_info->qgroup_rescan_progress.objectid,
3095 		fs_info->qgroup_rescan_progress.type,
3096 		fs_info->qgroup_rescan_progress.offset, ret);
3097 
3098 	if (ret) {
3099 		/*
3100 		 * The rescan is about to end, we will not be scanning any
3101 		 * further blocks. We cannot unset the RESCAN flag here, because
3102 		 * we want to commit the transaction if everything went well.
3103 		 * To make the live accounting work in this phase, we set our
3104 		 * scan progress pointer such that every real extent objectid
3105 		 * will be smaller.
3106 		 */
3107 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3108 		btrfs_release_path(path);
3109 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3110 		return ret;
3111 	}
3112 	done = is_last_leaf(path);
3113 
3114 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3115 			      btrfs_header_nritems(path->nodes[0]) - 1);
3116 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3117 
3118 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3119 	if (!scratch_leaf) {
3120 		ret = -ENOMEM;
3121 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3122 		goto out;
3123 	}
3124 	slot = path->slots[0];
3125 	btrfs_release_path(path);
3126 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3127 
3128 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3129 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3130 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3131 		    found.type != BTRFS_METADATA_ITEM_KEY)
3132 			continue;
3133 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3134 			num_bytes = fs_info->nodesize;
3135 		else
3136 			num_bytes = found.offset;
3137 
3138 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3139 					   &roots, false);
3140 		if (ret < 0)
3141 			goto out;
3142 		/* For rescan, just pass old_roots as NULL */
3143 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3144 						  num_bytes, NULL, roots);
3145 		if (ret < 0)
3146 			goto out;
3147 	}
3148 out:
3149 	if (scratch_leaf)
3150 		free_extent_buffer(scratch_leaf);
3151 
3152 	if (done && !ret) {
3153 		ret = 1;
3154 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3155 	}
3156 	return ret;
3157 }
3158 
3159 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3160 {
3161 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3162 						     qgroup_rescan_work);
3163 	struct btrfs_path *path;
3164 	struct btrfs_trans_handle *trans = NULL;
3165 	int err = -ENOMEM;
3166 	int ret = 0;
3167 
3168 	path = btrfs_alloc_path();
3169 	if (!path)
3170 		goto out;
3171 	/*
3172 	 * Rescan should only search for commit root, and any later difference
3173 	 * should be recorded by qgroup
3174 	 */
3175 	path->search_commit_root = 1;
3176 	path->skip_locking = 1;
3177 
3178 	err = 0;
3179 	while (!err && !btrfs_fs_closing(fs_info)) {
3180 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3181 		if (IS_ERR(trans)) {
3182 			err = PTR_ERR(trans);
3183 			break;
3184 		}
3185 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3186 			err = -EINTR;
3187 		} else {
3188 			err = qgroup_rescan_leaf(trans, path);
3189 		}
3190 		if (err > 0)
3191 			btrfs_commit_transaction(trans);
3192 		else
3193 			btrfs_end_transaction(trans);
3194 	}
3195 
3196 out:
3197 	btrfs_free_path(path);
3198 
3199 	mutex_lock(&fs_info->qgroup_rescan_lock);
3200 	if (err > 0 &&
3201 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3202 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3203 	} else if (err < 0) {
3204 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3205 	}
3206 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3207 
3208 	/*
3209 	 * only update status, since the previous part has already updated the
3210 	 * qgroup info.
3211 	 */
3212 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3213 	if (IS_ERR(trans)) {
3214 		err = PTR_ERR(trans);
3215 		trans = NULL;
3216 		btrfs_err(fs_info,
3217 			  "fail to start transaction for status update: %d",
3218 			  err);
3219 	}
3220 
3221 	mutex_lock(&fs_info->qgroup_rescan_lock);
3222 	if (!btrfs_fs_closing(fs_info))
3223 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3224 	if (trans) {
3225 		ret = update_qgroup_status_item(trans);
3226 		if (ret < 0) {
3227 			err = ret;
3228 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3229 				  err);
3230 		}
3231 	}
3232 	fs_info->qgroup_rescan_running = false;
3233 	complete_all(&fs_info->qgroup_rescan_completion);
3234 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3235 
3236 	if (!trans)
3237 		return;
3238 
3239 	btrfs_end_transaction(trans);
3240 
3241 	if (btrfs_fs_closing(fs_info)) {
3242 		btrfs_info(fs_info, "qgroup scan paused");
3243 	} else if (err >= 0) {
3244 		btrfs_info(fs_info, "qgroup scan completed%s",
3245 			err > 0 ? " (inconsistency flag cleared)" : "");
3246 	} else {
3247 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3248 	}
3249 }
3250 
3251 /*
3252  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3253  * memory required for the rescan context.
3254  */
3255 static int
3256 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3257 		   int init_flags)
3258 {
3259 	int ret = 0;
3260 
3261 	if (!init_flags) {
3262 		/* we're resuming qgroup rescan at mount time */
3263 		if (!(fs_info->qgroup_flags &
3264 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3265 			btrfs_warn(fs_info,
3266 			"qgroup rescan init failed, qgroup rescan is not queued");
3267 			ret = -EINVAL;
3268 		} else if (!(fs_info->qgroup_flags &
3269 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3270 			btrfs_warn(fs_info,
3271 			"qgroup rescan init failed, qgroup is not enabled");
3272 			ret = -EINVAL;
3273 		}
3274 
3275 		if (ret)
3276 			return ret;
3277 	}
3278 
3279 	mutex_lock(&fs_info->qgroup_rescan_lock);
3280 
3281 	if (init_flags) {
3282 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3283 			btrfs_warn(fs_info,
3284 				   "qgroup rescan is already in progress");
3285 			ret = -EINPROGRESS;
3286 		} else if (!(fs_info->qgroup_flags &
3287 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3288 			btrfs_warn(fs_info,
3289 			"qgroup rescan init failed, qgroup is not enabled");
3290 			ret = -EINVAL;
3291 		}
3292 
3293 		if (ret) {
3294 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3295 			return ret;
3296 		}
3297 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3298 	}
3299 
3300 	memset(&fs_info->qgroup_rescan_progress, 0,
3301 		sizeof(fs_info->qgroup_rescan_progress));
3302 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3303 	init_completion(&fs_info->qgroup_rescan_completion);
3304 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3305 
3306 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3307 			btrfs_qgroup_rescan_worker, NULL, NULL);
3308 	return 0;
3309 }
3310 
3311 static void
3312 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3313 {
3314 	struct rb_node *n;
3315 	struct btrfs_qgroup *qgroup;
3316 
3317 	spin_lock(&fs_info->qgroup_lock);
3318 	/* clear all current qgroup tracking information */
3319 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3320 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3321 		qgroup->rfer = 0;
3322 		qgroup->rfer_cmpr = 0;
3323 		qgroup->excl = 0;
3324 		qgroup->excl_cmpr = 0;
3325 		qgroup_dirty(fs_info, qgroup);
3326 	}
3327 	spin_unlock(&fs_info->qgroup_lock);
3328 }
3329 
3330 int
3331 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3332 {
3333 	int ret = 0;
3334 	struct btrfs_trans_handle *trans;
3335 
3336 	ret = qgroup_rescan_init(fs_info, 0, 1);
3337 	if (ret)
3338 		return ret;
3339 
3340 	/*
3341 	 * We have set the rescan_progress to 0, which means no more
3342 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3343 	 * However, btrfs_qgroup_account_ref may be right after its call
3344 	 * to btrfs_find_all_roots, in which case it would still do the
3345 	 * accounting.
3346 	 * To solve this, we're committing the transaction, which will
3347 	 * ensure we run all delayed refs and only after that, we are
3348 	 * going to clear all tracking information for a clean start.
3349 	 */
3350 
3351 	trans = btrfs_join_transaction(fs_info->fs_root);
3352 	if (IS_ERR(trans)) {
3353 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3354 		return PTR_ERR(trans);
3355 	}
3356 	ret = btrfs_commit_transaction(trans);
3357 	if (ret) {
3358 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3359 		return ret;
3360 	}
3361 
3362 	qgroup_rescan_zero_tracking(fs_info);
3363 
3364 	mutex_lock(&fs_info->qgroup_rescan_lock);
3365 	fs_info->qgroup_rescan_running = true;
3366 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3367 			 &fs_info->qgroup_rescan_work);
3368 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3369 
3370 	return 0;
3371 }
3372 
3373 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3374 				     bool interruptible)
3375 {
3376 	int running;
3377 	int ret = 0;
3378 
3379 	mutex_lock(&fs_info->qgroup_rescan_lock);
3380 	running = fs_info->qgroup_rescan_running;
3381 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3382 
3383 	if (!running)
3384 		return 0;
3385 
3386 	if (interruptible)
3387 		ret = wait_for_completion_interruptible(
3388 					&fs_info->qgroup_rescan_completion);
3389 	else
3390 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3391 
3392 	return ret;
3393 }
3394 
3395 /*
3396  * this is only called from open_ctree where we're still single threaded, thus
3397  * locking is omitted here.
3398  */
3399 void
3400 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3401 {
3402 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3403 		mutex_lock(&fs_info->qgroup_rescan_lock);
3404 		fs_info->qgroup_rescan_running = true;
3405 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3406 				 &fs_info->qgroup_rescan_work);
3407 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3408 	}
3409 }
3410 
3411 #define rbtree_iterate_from_safe(node, next, start)				\
3412        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3413 
3414 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3415 				  struct extent_changeset *reserved, u64 start,
3416 				  u64 len)
3417 {
3418 	struct rb_node *node;
3419 	struct rb_node *next;
3420 	struct ulist_node *entry = NULL;
3421 	int ret = 0;
3422 
3423 	node = reserved->range_changed.root.rb_node;
3424 	while (node) {
3425 		entry = rb_entry(node, struct ulist_node, rb_node);
3426 		if (entry->val < start)
3427 			node = node->rb_right;
3428 		else if (entry)
3429 			node = node->rb_left;
3430 		else
3431 			break;
3432 	}
3433 
3434 	/* Empty changeset */
3435 	if (!entry)
3436 		return 0;
3437 
3438 	if (entry->val > start && rb_prev(&entry->rb_node))
3439 		entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3440 				 rb_node);
3441 
3442 	rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3443 		u64 entry_start;
3444 		u64 entry_end;
3445 		u64 entry_len;
3446 		int clear_ret;
3447 
3448 		entry = rb_entry(node, struct ulist_node, rb_node);
3449 		entry_start = entry->val;
3450 		entry_end = entry->aux;
3451 		entry_len = entry_end - entry_start + 1;
3452 
3453 		if (entry_start >= start + len)
3454 			break;
3455 		if (entry_start + entry_len <= start)
3456 			continue;
3457 		/*
3458 		 * Now the entry is in [start, start + len), revert the
3459 		 * EXTENT_QGROUP_RESERVED bit.
3460 		 */
3461 		clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3462 					      entry_end, EXTENT_QGROUP_RESERVED);
3463 		if (!ret && clear_ret < 0)
3464 			ret = clear_ret;
3465 
3466 		ulist_del(&reserved->range_changed, entry->val, entry->aux);
3467 		if (likely(reserved->bytes_changed >= entry_len)) {
3468 			reserved->bytes_changed -= entry_len;
3469 		} else {
3470 			WARN_ON(1);
3471 			reserved->bytes_changed = 0;
3472 		}
3473 	}
3474 
3475 	return ret;
3476 }
3477 
3478 /*
3479  * Try to free some space for qgroup.
3480  *
3481  * For qgroup, there are only 3 ways to free qgroup space:
3482  * - Flush nodatacow write
3483  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3484  *   In theory, we should only flush nodatacow inodes, but it's not yet
3485  *   possible, so we need to flush the whole root.
3486  *
3487  * - Wait for ordered extents
3488  *   When ordered extents are finished, their reserved metadata is finally
3489  *   converted to per_trans status, which can be freed by later commit
3490  *   transaction.
3491  *
3492  * - Commit transaction
3493  *   This would free the meta_per_trans space.
3494  *   In theory this shouldn't provide much space, but any more qgroup space
3495  *   is needed.
3496  */
3497 static int try_flush_qgroup(struct btrfs_root *root)
3498 {
3499 	struct btrfs_trans_handle *trans;
3500 	int ret;
3501 
3502 	/*
3503 	 * We don't want to run flush again and again, so if there is a running
3504 	 * one, we won't try to start a new flush, but exit directly.
3505 	 */
3506 	if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3507 		wait_event(root->qgroup_flush_wait,
3508 			!test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3509 		return 0;
3510 	}
3511 
3512 	ret = btrfs_start_delalloc_snapshot(root);
3513 	if (ret < 0)
3514 		goto out;
3515 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3516 
3517 	trans = btrfs_join_transaction(root);
3518 	if (IS_ERR(trans)) {
3519 		ret = PTR_ERR(trans);
3520 		goto out;
3521 	}
3522 
3523 	ret = btrfs_commit_transaction(trans);
3524 out:
3525 	clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3526 	wake_up(&root->qgroup_flush_wait);
3527 	return ret;
3528 }
3529 
3530 static int qgroup_reserve_data(struct btrfs_inode *inode,
3531 			struct extent_changeset **reserved_ret, u64 start,
3532 			u64 len)
3533 {
3534 	struct btrfs_root *root = inode->root;
3535 	struct extent_changeset *reserved;
3536 	bool new_reserved = false;
3537 	u64 orig_reserved;
3538 	u64 to_reserve;
3539 	int ret;
3540 
3541 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3542 	    !is_fstree(root->root_key.objectid) || len == 0)
3543 		return 0;
3544 
3545 	/* @reserved parameter is mandatory for qgroup */
3546 	if (WARN_ON(!reserved_ret))
3547 		return -EINVAL;
3548 	if (!*reserved_ret) {
3549 		new_reserved = true;
3550 		*reserved_ret = extent_changeset_alloc();
3551 		if (!*reserved_ret)
3552 			return -ENOMEM;
3553 	}
3554 	reserved = *reserved_ret;
3555 	/* Record already reserved space */
3556 	orig_reserved = reserved->bytes_changed;
3557 	ret = set_record_extent_bits(&inode->io_tree, start,
3558 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3559 
3560 	/* Newly reserved space */
3561 	to_reserve = reserved->bytes_changed - orig_reserved;
3562 	trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3563 					to_reserve, QGROUP_RESERVE);
3564 	if (ret < 0)
3565 		goto out;
3566 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3567 	if (ret < 0)
3568 		goto cleanup;
3569 
3570 	return ret;
3571 
3572 cleanup:
3573 	qgroup_unreserve_range(inode, reserved, start, len);
3574 out:
3575 	if (new_reserved) {
3576 		extent_changeset_release(reserved);
3577 		kfree(reserved);
3578 		*reserved_ret = NULL;
3579 	}
3580 	return ret;
3581 }
3582 
3583 /*
3584  * Reserve qgroup space for range [start, start + len).
3585  *
3586  * This function will either reserve space from related qgroups or do nothing
3587  * if the range is already reserved.
3588  *
3589  * Return 0 for successful reservation
3590  * Return <0 for error (including -EQUOT)
3591  *
3592  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3593  *	 commit transaction. So caller should not hold any dirty page locked.
3594  */
3595 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3596 			struct extent_changeset **reserved_ret, u64 start,
3597 			u64 len)
3598 {
3599 	int ret;
3600 
3601 	ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3602 	if (ret <= 0 && ret != -EDQUOT)
3603 		return ret;
3604 
3605 	ret = try_flush_qgroup(inode->root);
3606 	if (ret < 0)
3607 		return ret;
3608 	return qgroup_reserve_data(inode, reserved_ret, start, len);
3609 }
3610 
3611 /* Free ranges specified by @reserved, normally in error path */
3612 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3613 			struct extent_changeset *reserved, u64 start, u64 len)
3614 {
3615 	struct btrfs_root *root = inode->root;
3616 	struct ulist_node *unode;
3617 	struct ulist_iterator uiter;
3618 	struct extent_changeset changeset;
3619 	int freed = 0;
3620 	int ret;
3621 
3622 	extent_changeset_init(&changeset);
3623 	len = round_up(start + len, root->fs_info->sectorsize);
3624 	start = round_down(start, root->fs_info->sectorsize);
3625 
3626 	ULIST_ITER_INIT(&uiter);
3627 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3628 		u64 range_start = unode->val;
3629 		/* unode->aux is the inclusive end */
3630 		u64 range_len = unode->aux - range_start + 1;
3631 		u64 free_start;
3632 		u64 free_len;
3633 
3634 		extent_changeset_release(&changeset);
3635 
3636 		/* Only free range in range [start, start + len) */
3637 		if (range_start >= start + len ||
3638 		    range_start + range_len <= start)
3639 			continue;
3640 		free_start = max(range_start, start);
3641 		free_len = min(start + len, range_start + range_len) -
3642 			   free_start;
3643 		/*
3644 		 * TODO: To also modify reserved->ranges_reserved to reflect
3645 		 * the modification.
3646 		 *
3647 		 * However as long as we free qgroup reserved according to
3648 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3649 		 * So not need to rush.
3650 		 */
3651 		ret = clear_record_extent_bits(&inode->io_tree, free_start,
3652 				free_start + free_len - 1,
3653 				EXTENT_QGROUP_RESERVED, &changeset);
3654 		if (ret < 0)
3655 			goto out;
3656 		freed += changeset.bytes_changed;
3657 	}
3658 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3659 				  BTRFS_QGROUP_RSV_DATA);
3660 	ret = freed;
3661 out:
3662 	extent_changeset_release(&changeset);
3663 	return ret;
3664 }
3665 
3666 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3667 			struct extent_changeset *reserved, u64 start, u64 len,
3668 			int free)
3669 {
3670 	struct extent_changeset changeset;
3671 	int trace_op = QGROUP_RELEASE;
3672 	int ret;
3673 
3674 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3675 		return 0;
3676 
3677 	/* In release case, we shouldn't have @reserved */
3678 	WARN_ON(!free && reserved);
3679 	if (free && reserved)
3680 		return qgroup_free_reserved_data(inode, reserved, start, len);
3681 	extent_changeset_init(&changeset);
3682 	ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3683 				       EXTENT_QGROUP_RESERVED, &changeset);
3684 	if (ret < 0)
3685 		goto out;
3686 
3687 	if (free)
3688 		trace_op = QGROUP_FREE;
3689 	trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3690 					changeset.bytes_changed, trace_op);
3691 	if (free)
3692 		btrfs_qgroup_free_refroot(inode->root->fs_info,
3693 				inode->root->root_key.objectid,
3694 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3695 	ret = changeset.bytes_changed;
3696 out:
3697 	extent_changeset_release(&changeset);
3698 	return ret;
3699 }
3700 
3701 /*
3702  * Free a reserved space range from io_tree and related qgroups
3703  *
3704  * Should be called when a range of pages get invalidated before reaching disk.
3705  * Or for error cleanup case.
3706  * if @reserved is given, only reserved range in [@start, @start + @len) will
3707  * be freed.
3708  *
3709  * For data written to disk, use btrfs_qgroup_release_data().
3710  *
3711  * NOTE: This function may sleep for memory allocation.
3712  */
3713 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3714 			struct extent_changeset *reserved, u64 start, u64 len)
3715 {
3716 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3717 }
3718 
3719 /*
3720  * Release a reserved space range from io_tree only.
3721  *
3722  * Should be called when a range of pages get written to disk and corresponding
3723  * FILE_EXTENT is inserted into corresponding root.
3724  *
3725  * Since new qgroup accounting framework will only update qgroup numbers at
3726  * commit_transaction() time, its reserved space shouldn't be freed from
3727  * related qgroups.
3728  *
3729  * But we should release the range from io_tree, to allow further write to be
3730  * COWed.
3731  *
3732  * NOTE: This function may sleep for memory allocation.
3733  */
3734 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3735 {
3736 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3737 }
3738 
3739 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3740 			      enum btrfs_qgroup_rsv_type type)
3741 {
3742 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3743 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3744 		return;
3745 	if (num_bytes == 0)
3746 		return;
3747 
3748 	spin_lock(&root->qgroup_meta_rsv_lock);
3749 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3750 		root->qgroup_meta_rsv_prealloc += num_bytes;
3751 	else
3752 		root->qgroup_meta_rsv_pertrans += num_bytes;
3753 	spin_unlock(&root->qgroup_meta_rsv_lock);
3754 }
3755 
3756 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3757 			     enum btrfs_qgroup_rsv_type type)
3758 {
3759 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3760 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3761 		return 0;
3762 	if (num_bytes == 0)
3763 		return 0;
3764 
3765 	spin_lock(&root->qgroup_meta_rsv_lock);
3766 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3767 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3768 				  num_bytes);
3769 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3770 	} else {
3771 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3772 				  num_bytes);
3773 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3774 	}
3775 	spin_unlock(&root->qgroup_meta_rsv_lock);
3776 	return num_bytes;
3777 }
3778 
3779 static int qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3780 				enum btrfs_qgroup_rsv_type type, bool enforce)
3781 {
3782 	struct btrfs_fs_info *fs_info = root->fs_info;
3783 	int ret;
3784 
3785 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3786 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3787 		return 0;
3788 
3789 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3790 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3791 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3792 	if (ret < 0)
3793 		return ret;
3794 	/*
3795 	 * Record what we have reserved into root.
3796 	 *
3797 	 * To avoid quota disabled->enabled underflow.
3798 	 * In that case, we may try to free space we haven't reserved
3799 	 * (since quota was disabled), so record what we reserved into root.
3800 	 * And ensure later release won't underflow this number.
3801 	 */
3802 	add_root_meta_rsv(root, num_bytes, type);
3803 	return ret;
3804 }
3805 
3806 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3807 				enum btrfs_qgroup_rsv_type type, bool enforce)
3808 {
3809 	int ret;
3810 
3811 	ret = qgroup_reserve_meta(root, num_bytes, type, enforce);
3812 	if (ret <= 0 && ret != -EDQUOT)
3813 		return ret;
3814 
3815 	ret = try_flush_qgroup(root);
3816 	if (ret < 0)
3817 		return ret;
3818 	return qgroup_reserve_meta(root, num_bytes, type, enforce);
3819 }
3820 
3821 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3822 {
3823 	struct btrfs_fs_info *fs_info = root->fs_info;
3824 
3825 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3826 	    !is_fstree(root->root_key.objectid))
3827 		return;
3828 
3829 	/* TODO: Update trace point to handle such free */
3830 	trace_qgroup_meta_free_all_pertrans(root);
3831 	/* Special value -1 means to free all reserved space */
3832 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3833 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3834 }
3835 
3836 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3837 			      enum btrfs_qgroup_rsv_type type)
3838 {
3839 	struct btrfs_fs_info *fs_info = root->fs_info;
3840 
3841 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3842 	    !is_fstree(root->root_key.objectid))
3843 		return;
3844 
3845 	/*
3846 	 * reservation for META_PREALLOC can happen before quota is enabled,
3847 	 * which can lead to underflow.
3848 	 * Here ensure we will only free what we really have reserved.
3849 	 */
3850 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3851 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3852 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3853 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3854 				  num_bytes, type);
3855 }
3856 
3857 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3858 				int num_bytes)
3859 {
3860 	struct btrfs_qgroup *qgroup;
3861 	struct ulist_node *unode;
3862 	struct ulist_iterator uiter;
3863 	int ret = 0;
3864 
3865 	if (num_bytes == 0)
3866 		return;
3867 	if (!fs_info->quota_root)
3868 		return;
3869 
3870 	spin_lock(&fs_info->qgroup_lock);
3871 	qgroup = find_qgroup_rb(fs_info, ref_root);
3872 	if (!qgroup)
3873 		goto out;
3874 	ulist_reinit(fs_info->qgroup_ulist);
3875 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3876 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3877 	if (ret < 0)
3878 		goto out;
3879 	ULIST_ITER_INIT(&uiter);
3880 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3881 		struct btrfs_qgroup *qg;
3882 		struct btrfs_qgroup_list *glist;
3883 
3884 		qg = unode_aux_to_qgroup(unode);
3885 
3886 		qgroup_rsv_release(fs_info, qg, num_bytes,
3887 				BTRFS_QGROUP_RSV_META_PREALLOC);
3888 		qgroup_rsv_add(fs_info, qg, num_bytes,
3889 				BTRFS_QGROUP_RSV_META_PERTRANS);
3890 		list_for_each_entry(glist, &qg->groups, next_group) {
3891 			ret = ulist_add(fs_info->qgroup_ulist,
3892 					glist->group->qgroupid,
3893 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3894 			if (ret < 0)
3895 				goto out;
3896 		}
3897 	}
3898 out:
3899 	spin_unlock(&fs_info->qgroup_lock);
3900 }
3901 
3902 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3903 {
3904 	struct btrfs_fs_info *fs_info = root->fs_info;
3905 
3906 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3907 	    !is_fstree(root->root_key.objectid))
3908 		return;
3909 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3910 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3911 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3912 	trace_qgroup_meta_convert(root, num_bytes);
3913 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3914 }
3915 
3916 /*
3917  * Check qgroup reserved space leaking, normally at destroy inode
3918  * time
3919  */
3920 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
3921 {
3922 	struct extent_changeset changeset;
3923 	struct ulist_node *unode;
3924 	struct ulist_iterator iter;
3925 	int ret;
3926 
3927 	extent_changeset_init(&changeset);
3928 	ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
3929 			EXTENT_QGROUP_RESERVED, &changeset);
3930 
3931 	WARN_ON(ret < 0);
3932 	if (WARN_ON(changeset.bytes_changed)) {
3933 		ULIST_ITER_INIT(&iter);
3934 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3935 			btrfs_warn(inode->root->fs_info,
3936 		"leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
3937 				btrfs_ino(inode), unode->val, unode->aux);
3938 		}
3939 		btrfs_qgroup_free_refroot(inode->root->fs_info,
3940 				inode->root->root_key.objectid,
3941 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3942 
3943 	}
3944 	extent_changeset_release(&changeset);
3945 }
3946 
3947 void btrfs_qgroup_init_swapped_blocks(
3948 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3949 {
3950 	int i;
3951 
3952 	spin_lock_init(&swapped_blocks->lock);
3953 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3954 		swapped_blocks->blocks[i] = RB_ROOT;
3955 	swapped_blocks->swapped = false;
3956 }
3957 
3958 /*
3959  * Delete all swapped blocks record of @root.
3960  * Every record here means we skipped a full subtree scan for qgroup.
3961  *
3962  * Gets called when committing one transaction.
3963  */
3964 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3965 {
3966 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3967 	int i;
3968 
3969 	swapped_blocks = &root->swapped_blocks;
3970 
3971 	spin_lock(&swapped_blocks->lock);
3972 	if (!swapped_blocks->swapped)
3973 		goto out;
3974 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3975 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
3976 		struct btrfs_qgroup_swapped_block *entry;
3977 		struct btrfs_qgroup_swapped_block *next;
3978 
3979 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3980 						     node)
3981 			kfree(entry);
3982 		swapped_blocks->blocks[i] = RB_ROOT;
3983 	}
3984 	swapped_blocks->swapped = false;
3985 out:
3986 	spin_unlock(&swapped_blocks->lock);
3987 }
3988 
3989 /*
3990  * Add subtree roots record into @subvol_root.
3991  *
3992  * @subvol_root:	tree root of the subvolume tree get swapped
3993  * @bg:			block group under balance
3994  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
3995  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
3996  *			BOTH POINTERS ARE BEFORE TREE SWAP
3997  * @last_snapshot:	last snapshot generation of the subvolume tree
3998  */
3999 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4000 		struct btrfs_root *subvol_root,
4001 		struct btrfs_block_group *bg,
4002 		struct extent_buffer *subvol_parent, int subvol_slot,
4003 		struct extent_buffer *reloc_parent, int reloc_slot,
4004 		u64 last_snapshot)
4005 {
4006 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4007 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4008 	struct btrfs_qgroup_swapped_block *block;
4009 	struct rb_node **cur;
4010 	struct rb_node *parent = NULL;
4011 	int level = btrfs_header_level(subvol_parent) - 1;
4012 	int ret = 0;
4013 
4014 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4015 		return 0;
4016 
4017 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4018 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4019 		btrfs_err_rl(fs_info,
4020 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4021 			__func__,
4022 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4023 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4024 		return -EUCLEAN;
4025 	}
4026 
4027 	block = kmalloc(sizeof(*block), GFP_NOFS);
4028 	if (!block) {
4029 		ret = -ENOMEM;
4030 		goto out;
4031 	}
4032 
4033 	/*
4034 	 * @reloc_parent/slot is still before swap, while @block is going to
4035 	 * record the bytenr after swap, so we do the swap here.
4036 	 */
4037 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4038 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4039 							     reloc_slot);
4040 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4041 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4042 							    subvol_slot);
4043 	block->last_snapshot = last_snapshot;
4044 	block->level = level;
4045 
4046 	/*
4047 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4048 	 * no one else can modify tree blocks thus we qgroup will not change
4049 	 * no matter the value of trace_leaf.
4050 	 */
4051 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4052 		block->trace_leaf = true;
4053 	else
4054 		block->trace_leaf = false;
4055 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4056 
4057 	/* Insert @block into @blocks */
4058 	spin_lock(&blocks->lock);
4059 	cur = &blocks->blocks[level].rb_node;
4060 	while (*cur) {
4061 		struct btrfs_qgroup_swapped_block *entry;
4062 
4063 		parent = *cur;
4064 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4065 				 node);
4066 
4067 		if (entry->subvol_bytenr < block->subvol_bytenr) {
4068 			cur = &(*cur)->rb_left;
4069 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
4070 			cur = &(*cur)->rb_right;
4071 		} else {
4072 			if (entry->subvol_generation !=
4073 					block->subvol_generation ||
4074 			    entry->reloc_bytenr != block->reloc_bytenr ||
4075 			    entry->reloc_generation !=
4076 					block->reloc_generation) {
4077 				/*
4078 				 * Duplicated but mismatch entry found.
4079 				 * Shouldn't happen.
4080 				 *
4081 				 * Marking qgroup inconsistent should be enough
4082 				 * for end users.
4083 				 */
4084 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4085 				ret = -EEXIST;
4086 			}
4087 			kfree(block);
4088 			goto out_unlock;
4089 		}
4090 	}
4091 	rb_link_node(&block->node, parent, cur);
4092 	rb_insert_color(&block->node, &blocks->blocks[level]);
4093 	blocks->swapped = true;
4094 out_unlock:
4095 	spin_unlock(&blocks->lock);
4096 out:
4097 	if (ret < 0)
4098 		fs_info->qgroup_flags |=
4099 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4100 	return ret;
4101 }
4102 
4103 /*
4104  * Check if the tree block is a subtree root, and if so do the needed
4105  * delayed subtree trace for qgroup.
4106  *
4107  * This is called during btrfs_cow_block().
4108  */
4109 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4110 					 struct btrfs_root *root,
4111 					 struct extent_buffer *subvol_eb)
4112 {
4113 	struct btrfs_fs_info *fs_info = root->fs_info;
4114 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4115 	struct btrfs_qgroup_swapped_block *block;
4116 	struct extent_buffer *reloc_eb = NULL;
4117 	struct rb_node *node;
4118 	bool found = false;
4119 	bool swapped = false;
4120 	int level = btrfs_header_level(subvol_eb);
4121 	int ret = 0;
4122 	int i;
4123 
4124 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4125 		return 0;
4126 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4127 		return 0;
4128 
4129 	spin_lock(&blocks->lock);
4130 	if (!blocks->swapped) {
4131 		spin_unlock(&blocks->lock);
4132 		return 0;
4133 	}
4134 	node = blocks->blocks[level].rb_node;
4135 
4136 	while (node) {
4137 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4138 		if (block->subvol_bytenr < subvol_eb->start) {
4139 			node = node->rb_left;
4140 		} else if (block->subvol_bytenr > subvol_eb->start) {
4141 			node = node->rb_right;
4142 		} else {
4143 			found = true;
4144 			break;
4145 		}
4146 	}
4147 	if (!found) {
4148 		spin_unlock(&blocks->lock);
4149 		goto out;
4150 	}
4151 	/* Found one, remove it from @blocks first and update blocks->swapped */
4152 	rb_erase(&block->node, &blocks->blocks[level]);
4153 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4154 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4155 			swapped = true;
4156 			break;
4157 		}
4158 	}
4159 	blocks->swapped = swapped;
4160 	spin_unlock(&blocks->lock);
4161 
4162 	/* Read out reloc subtree root */
4163 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
4164 				   block->reloc_generation, block->level,
4165 				   &block->first_key);
4166 	if (IS_ERR(reloc_eb)) {
4167 		ret = PTR_ERR(reloc_eb);
4168 		reloc_eb = NULL;
4169 		goto free_out;
4170 	}
4171 	if (!extent_buffer_uptodate(reloc_eb)) {
4172 		ret = -EIO;
4173 		goto free_out;
4174 	}
4175 
4176 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4177 			block->last_snapshot, block->trace_leaf);
4178 free_out:
4179 	kfree(block);
4180 	free_extent_buffer(reloc_eb);
4181 out:
4182 	if (ret < 0) {
4183 		btrfs_err_rl(fs_info,
4184 			     "failed to account subtree at bytenr %llu: %d",
4185 			     subvol_eb->start, ret);
4186 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4187 	}
4188 	return ret;
4189 }
4190 
4191 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4192 {
4193 	struct btrfs_qgroup_extent_record *entry;
4194 	struct btrfs_qgroup_extent_record *next;
4195 	struct rb_root *root;
4196 
4197 	root = &trans->delayed_refs.dirty_extent_root;
4198 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4199 		ulist_free(entry->old_roots);
4200 		kfree(entry);
4201 	}
4202 }
4203