xref: /openbmc/linux/fs/btrfs/qgroup.c (revision b2765275)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sizes.h>
15 
16 #include "ctree.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "locking.h"
20 #include "ulist.h"
21 #include "backref.h"
22 #include "extent_io.h"
23 #include "qgroup.h"
24 #include "block-group.h"
25 
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_qgroup *qgroup)
224 {
225 	struct btrfs_qgroup_list *list;
226 
227 	list_del(&qgroup->dirty);
228 	while (!list_empty(&qgroup->groups)) {
229 		list = list_first_entry(&qgroup->groups,
230 					struct btrfs_qgroup_list, next_group);
231 		list_del(&list->next_group);
232 		list_del(&list->next_member);
233 		kfree(list);
234 	}
235 
236 	while (!list_empty(&qgroup->members)) {
237 		list = list_first_entry(&qgroup->members,
238 					struct btrfs_qgroup_list, next_member);
239 		list_del(&list->next_group);
240 		list_del(&list->next_member);
241 		kfree(list);
242 	}
243 	kfree(qgroup);
244 }
245 
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248 {
249 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250 
251 	if (!qgroup)
252 		return -ENOENT;
253 
254 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 	__del_qgroup_rb(qgroup);
256 	return 0;
257 }
258 
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 			   u64 memberid, u64 parentid)
262 {
263 	struct btrfs_qgroup *member;
264 	struct btrfs_qgroup *parent;
265 	struct btrfs_qgroup_list *list;
266 
267 	member = find_qgroup_rb(fs_info, memberid);
268 	parent = find_qgroup_rb(fs_info, parentid);
269 	if (!member || !parent)
270 		return -ENOENT;
271 
272 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
273 	if (!list)
274 		return -ENOMEM;
275 
276 	list->group = parent;
277 	list->member = member;
278 	list_add_tail(&list->next_group, &member->groups);
279 	list_add_tail(&list->next_member, &parent->members);
280 
281 	return 0;
282 }
283 
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 			   u64 memberid, u64 parentid)
287 {
288 	struct btrfs_qgroup *member;
289 	struct btrfs_qgroup *parent;
290 	struct btrfs_qgroup_list *list;
291 
292 	member = find_qgroup_rb(fs_info, memberid);
293 	parent = find_qgroup_rb(fs_info, parentid);
294 	if (!member || !parent)
295 		return -ENOENT;
296 
297 	list_for_each_entry(list, &member->groups, next_group) {
298 		if (list->group == parent) {
299 			list_del(&list->next_group);
300 			list_del(&list->next_member);
301 			kfree(list);
302 			return 0;
303 		}
304 	}
305 	return -ENOENT;
306 }
307 
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310 			       u64 rfer, u64 excl)
311 {
312 	struct btrfs_qgroup *qgroup;
313 
314 	qgroup = find_qgroup_rb(fs_info, qgroupid);
315 	if (!qgroup)
316 		return -EINVAL;
317 	if (qgroup->rfer != rfer || qgroup->excl != excl)
318 		return -EINVAL;
319 	return 0;
320 }
321 #endif
322 
323 /*
324  * The full config is read in one go, only called from open_ctree()
325  * It doesn't use any locking, as at this point we're still single-threaded
326  */
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328 {
329 	struct btrfs_key key;
330 	struct btrfs_key found_key;
331 	struct btrfs_root *quota_root = fs_info->quota_root;
332 	struct btrfs_path *path = NULL;
333 	struct extent_buffer *l;
334 	int slot;
335 	int ret = 0;
336 	u64 flags = 0;
337 	u64 rescan_progress = 0;
338 
339 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
340 		return 0;
341 
342 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 	if (!fs_info->qgroup_ulist) {
344 		ret = -ENOMEM;
345 		goto out;
346 	}
347 
348 	path = btrfs_alloc_path();
349 	if (!path) {
350 		ret = -ENOMEM;
351 		goto out;
352 	}
353 
354 	/* default this to quota off, in case no status key is found */
355 	fs_info->qgroup_flags = 0;
356 
357 	/*
358 	 * pass 1: read status, all qgroup infos and limits
359 	 */
360 	key.objectid = 0;
361 	key.type = 0;
362 	key.offset = 0;
363 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
364 	if (ret)
365 		goto out;
366 
367 	while (1) {
368 		struct btrfs_qgroup *qgroup;
369 
370 		slot = path->slots[0];
371 		l = path->nodes[0];
372 		btrfs_item_key_to_cpu(l, &found_key, slot);
373 
374 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 			struct btrfs_qgroup_status_item *ptr;
376 
377 			ptr = btrfs_item_ptr(l, slot,
378 					     struct btrfs_qgroup_status_item);
379 
380 			if (btrfs_qgroup_status_version(l, ptr) !=
381 			    BTRFS_QGROUP_STATUS_VERSION) {
382 				btrfs_err(fs_info,
383 				 "old qgroup version, quota disabled");
384 				goto out;
385 			}
386 			if (btrfs_qgroup_status_generation(l, ptr) !=
387 			    fs_info->generation) {
388 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 				btrfs_err(fs_info,
390 					"qgroup generation mismatch, marked as inconsistent");
391 			}
392 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 									  ptr);
394 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
395 			goto next1;
396 		}
397 
398 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
400 			goto next1;
401 
402 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 			btrfs_err(fs_info, "inconsistent qgroup config");
406 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407 		}
408 		if (!qgroup) {
409 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 			if (IS_ERR(qgroup)) {
411 				ret = PTR_ERR(qgroup);
412 				goto out;
413 			}
414 		}
415 		switch (found_key.type) {
416 		case BTRFS_QGROUP_INFO_KEY: {
417 			struct btrfs_qgroup_info_item *ptr;
418 
419 			ptr = btrfs_item_ptr(l, slot,
420 					     struct btrfs_qgroup_info_item);
421 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 			/* generation currently unused */
426 			break;
427 		}
428 		case BTRFS_QGROUP_LIMIT_KEY: {
429 			struct btrfs_qgroup_limit_item *ptr;
430 
431 			ptr = btrfs_item_ptr(l, slot,
432 					     struct btrfs_qgroup_limit_item);
433 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438 			break;
439 		}
440 		}
441 next1:
442 		ret = btrfs_next_item(quota_root, path);
443 		if (ret < 0)
444 			goto out;
445 		if (ret)
446 			break;
447 	}
448 	btrfs_release_path(path);
449 
450 	/*
451 	 * pass 2: read all qgroup relations
452 	 */
453 	key.objectid = 0;
454 	key.type = BTRFS_QGROUP_RELATION_KEY;
455 	key.offset = 0;
456 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457 	if (ret)
458 		goto out;
459 	while (1) {
460 		slot = path->slots[0];
461 		l = path->nodes[0];
462 		btrfs_item_key_to_cpu(l, &found_key, slot);
463 
464 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465 			goto next2;
466 
467 		if (found_key.objectid > found_key.offset) {
468 			/* parent <- member, not needed to build config */
469 			/* FIXME should we omit the key completely? */
470 			goto next2;
471 		}
472 
473 		ret = add_relation_rb(fs_info, found_key.objectid,
474 				      found_key.offset);
475 		if (ret == -ENOENT) {
476 			btrfs_warn(fs_info,
477 				"orphan qgroup relation 0x%llx->0x%llx",
478 				found_key.objectid, found_key.offset);
479 			ret = 0;	/* ignore the error */
480 		}
481 		if (ret)
482 			goto out;
483 next2:
484 		ret = btrfs_next_item(quota_root, path);
485 		if (ret < 0)
486 			goto out;
487 		if (ret)
488 			break;
489 	}
490 out:
491 	fs_info->qgroup_flags |= flags;
492 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495 		 ret >= 0)
496 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 	btrfs_free_path(path);
498 
499 	if (ret < 0) {
500 		ulist_free(fs_info->qgroup_ulist);
501 		fs_info->qgroup_ulist = NULL;
502 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
503 	}
504 
505 	return ret < 0 ? ret : 0;
506 }
507 
508 /*
509  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510  * first two are in single-threaded paths.And for the third one, we have set
511  * quota_root to be null with qgroup_lock held before, so it is safe to clean
512  * up the in-memory structures without qgroup_lock held.
513  */
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515 {
516 	struct rb_node *n;
517 	struct btrfs_qgroup *qgroup;
518 
519 	while ((n = rb_first(&fs_info->qgroup_tree))) {
520 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 		rb_erase(n, &fs_info->qgroup_tree);
522 		__del_qgroup_rb(qgroup);
523 	}
524 	/*
525 	 * We call btrfs_free_qgroup_config() when unmounting
526 	 * filesystem and disabling quota, so we set qgroup_ulist
527 	 * to be null here to avoid double free.
528 	 */
529 	ulist_free(fs_info->qgroup_ulist);
530 	fs_info->qgroup_ulist = NULL;
531 }
532 
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 				    u64 dst)
535 {
536 	int ret;
537 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 	struct btrfs_path *path;
539 	struct btrfs_key key;
540 
541 	path = btrfs_alloc_path();
542 	if (!path)
543 		return -ENOMEM;
544 
545 	key.objectid = src;
546 	key.type = BTRFS_QGROUP_RELATION_KEY;
547 	key.offset = dst;
548 
549 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550 
551 	btrfs_mark_buffer_dirty(path->nodes[0]);
552 
553 	btrfs_free_path(path);
554 	return ret;
555 }
556 
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 				    u64 dst)
559 {
560 	int ret;
561 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 	struct btrfs_path *path;
563 	struct btrfs_key key;
564 
565 	path = btrfs_alloc_path();
566 	if (!path)
567 		return -ENOMEM;
568 
569 	key.objectid = src;
570 	key.type = BTRFS_QGROUP_RELATION_KEY;
571 	key.offset = dst;
572 
573 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574 	if (ret < 0)
575 		goto out;
576 
577 	if (ret > 0) {
578 		ret = -ENOENT;
579 		goto out;
580 	}
581 
582 	ret = btrfs_del_item(trans, quota_root, path);
583 out:
584 	btrfs_free_path(path);
585 	return ret;
586 }
587 
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 			   struct btrfs_root *quota_root, u64 qgroupid)
590 {
591 	int ret;
592 	struct btrfs_path *path;
593 	struct btrfs_qgroup_info_item *qgroup_info;
594 	struct btrfs_qgroup_limit_item *qgroup_limit;
595 	struct extent_buffer *leaf;
596 	struct btrfs_key key;
597 
598 	if (btrfs_is_testing(quota_root->fs_info))
599 		return 0;
600 
601 	path = btrfs_alloc_path();
602 	if (!path)
603 		return -ENOMEM;
604 
605 	key.objectid = 0;
606 	key.type = BTRFS_QGROUP_INFO_KEY;
607 	key.offset = qgroupid;
608 
609 	/*
610 	 * Avoid a transaction abort by catching -EEXIST here. In that
611 	 * case, we proceed by re-initializing the existing structure
612 	 * on disk.
613 	 */
614 
615 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 				      sizeof(*qgroup_info));
617 	if (ret && ret != -EEXIST)
618 		goto out;
619 
620 	leaf = path->nodes[0];
621 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 				 struct btrfs_qgroup_info_item);
623 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628 
629 	btrfs_mark_buffer_dirty(leaf);
630 
631 	btrfs_release_path(path);
632 
633 	key.type = BTRFS_QGROUP_LIMIT_KEY;
634 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 				      sizeof(*qgroup_limit));
636 	if (ret && ret != -EEXIST)
637 		goto out;
638 
639 	leaf = path->nodes[0];
640 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 				  struct btrfs_qgroup_limit_item);
642 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647 
648 	btrfs_mark_buffer_dirty(leaf);
649 
650 	ret = 0;
651 out:
652 	btrfs_free_path(path);
653 	return ret;
654 }
655 
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
657 {
658 	int ret;
659 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 	struct btrfs_path *path;
661 	struct btrfs_key key;
662 
663 	path = btrfs_alloc_path();
664 	if (!path)
665 		return -ENOMEM;
666 
667 	key.objectid = 0;
668 	key.type = BTRFS_QGROUP_INFO_KEY;
669 	key.offset = qgroupid;
670 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 	if (ret < 0)
672 		goto out;
673 
674 	if (ret > 0) {
675 		ret = -ENOENT;
676 		goto out;
677 	}
678 
679 	ret = btrfs_del_item(trans, quota_root, path);
680 	if (ret)
681 		goto out;
682 
683 	btrfs_release_path(path);
684 
685 	key.type = BTRFS_QGROUP_LIMIT_KEY;
686 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687 	if (ret < 0)
688 		goto out;
689 
690 	if (ret > 0) {
691 		ret = -ENOENT;
692 		goto out;
693 	}
694 
695 	ret = btrfs_del_item(trans, quota_root, path);
696 
697 out:
698 	btrfs_free_path(path);
699 	return ret;
700 }
701 
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 				    struct btrfs_qgroup *qgroup)
704 {
705 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 	struct btrfs_path *path;
707 	struct btrfs_key key;
708 	struct extent_buffer *l;
709 	struct btrfs_qgroup_limit_item *qgroup_limit;
710 	int ret;
711 	int slot;
712 
713 	key.objectid = 0;
714 	key.type = BTRFS_QGROUP_LIMIT_KEY;
715 	key.offset = qgroup->qgroupid;
716 
717 	path = btrfs_alloc_path();
718 	if (!path)
719 		return -ENOMEM;
720 
721 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
722 	if (ret > 0)
723 		ret = -ENOENT;
724 
725 	if (ret)
726 		goto out;
727 
728 	l = path->nodes[0];
729 	slot = path->slots[0];
730 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736 
737 	btrfs_mark_buffer_dirty(l);
738 
739 out:
740 	btrfs_free_path(path);
741 	return ret;
742 }
743 
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 				   struct btrfs_qgroup *qgroup)
746 {
747 	struct btrfs_fs_info *fs_info = trans->fs_info;
748 	struct btrfs_root *quota_root = fs_info->quota_root;
749 	struct btrfs_path *path;
750 	struct btrfs_key key;
751 	struct extent_buffer *l;
752 	struct btrfs_qgroup_info_item *qgroup_info;
753 	int ret;
754 	int slot;
755 
756 	if (btrfs_is_testing(fs_info))
757 		return 0;
758 
759 	key.objectid = 0;
760 	key.type = BTRFS_QGROUP_INFO_KEY;
761 	key.offset = qgroup->qgroupid;
762 
763 	path = btrfs_alloc_path();
764 	if (!path)
765 		return -ENOMEM;
766 
767 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
768 	if (ret > 0)
769 		ret = -ENOENT;
770 
771 	if (ret)
772 		goto out;
773 
774 	l = path->nodes[0];
775 	slot = path->slots[0];
776 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782 
783 	btrfs_mark_buffer_dirty(l);
784 
785 out:
786 	btrfs_free_path(path);
787 	return ret;
788 }
789 
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
791 {
792 	struct btrfs_fs_info *fs_info = trans->fs_info;
793 	struct btrfs_root *quota_root = fs_info->quota_root;
794 	struct btrfs_path *path;
795 	struct btrfs_key key;
796 	struct extent_buffer *l;
797 	struct btrfs_qgroup_status_item *ptr;
798 	int ret;
799 	int slot;
800 
801 	key.objectid = 0;
802 	key.type = BTRFS_QGROUP_STATUS_KEY;
803 	key.offset = 0;
804 
805 	path = btrfs_alloc_path();
806 	if (!path)
807 		return -ENOMEM;
808 
809 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
810 	if (ret > 0)
811 		ret = -ENOENT;
812 
813 	if (ret)
814 		goto out;
815 
816 	l = path->nodes[0];
817 	slot = path->slots[0];
818 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 	btrfs_set_qgroup_status_rescan(l, ptr,
822 				fs_info->qgroup_rescan_progress.objectid);
823 
824 	btrfs_mark_buffer_dirty(l);
825 
826 out:
827 	btrfs_free_path(path);
828 	return ret;
829 }
830 
831 /*
832  * called with qgroup_lock held
833  */
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 				  struct btrfs_root *root)
836 {
837 	struct btrfs_path *path;
838 	struct btrfs_key key;
839 	struct extent_buffer *leaf = NULL;
840 	int ret;
841 	int nr = 0;
842 
843 	path = btrfs_alloc_path();
844 	if (!path)
845 		return -ENOMEM;
846 
847 	path->leave_spinning = 1;
848 
849 	key.objectid = 0;
850 	key.offset = 0;
851 	key.type = 0;
852 
853 	while (1) {
854 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855 		if (ret < 0)
856 			goto out;
857 		leaf = path->nodes[0];
858 		nr = btrfs_header_nritems(leaf);
859 		if (!nr)
860 			break;
861 		/*
862 		 * delete the leaf one by one
863 		 * since the whole tree is going
864 		 * to be deleted.
865 		 */
866 		path->slots[0] = 0;
867 		ret = btrfs_del_items(trans, root, path, 0, nr);
868 		if (ret)
869 			goto out;
870 
871 		btrfs_release_path(path);
872 	}
873 	ret = 0;
874 out:
875 	btrfs_free_path(path);
876 	return ret;
877 }
878 
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
880 {
881 	struct btrfs_root *quota_root;
882 	struct btrfs_root *tree_root = fs_info->tree_root;
883 	struct btrfs_path *path = NULL;
884 	struct btrfs_qgroup_status_item *ptr;
885 	struct extent_buffer *leaf;
886 	struct btrfs_key key;
887 	struct btrfs_key found_key;
888 	struct btrfs_qgroup *qgroup = NULL;
889 	struct btrfs_trans_handle *trans = NULL;
890 	int ret = 0;
891 	int slot;
892 
893 	mutex_lock(&fs_info->qgroup_ioctl_lock);
894 	if (fs_info->quota_root)
895 		goto out;
896 
897 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 	if (!fs_info->qgroup_ulist) {
899 		ret = -ENOMEM;
900 		goto out;
901 	}
902 
903 	/*
904 	 * 1 for quota root item
905 	 * 1 for BTRFS_QGROUP_STATUS item
906 	 *
907 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
908 	 * per subvolume. However those are not currently reserved since it
909 	 * would be a lot of overkill.
910 	 */
911 	trans = btrfs_start_transaction(tree_root, 2);
912 	if (IS_ERR(trans)) {
913 		ret = PTR_ERR(trans);
914 		trans = NULL;
915 		goto out;
916 	}
917 
918 	/*
919 	 * initially create the quota tree
920 	 */
921 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
922 	if (IS_ERR(quota_root)) {
923 		ret =  PTR_ERR(quota_root);
924 		btrfs_abort_transaction(trans, ret);
925 		goto out;
926 	}
927 
928 	path = btrfs_alloc_path();
929 	if (!path) {
930 		ret = -ENOMEM;
931 		btrfs_abort_transaction(trans, ret);
932 		goto out_free_root;
933 	}
934 
935 	key.objectid = 0;
936 	key.type = BTRFS_QGROUP_STATUS_KEY;
937 	key.offset = 0;
938 
939 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
940 				      sizeof(*ptr));
941 	if (ret) {
942 		btrfs_abort_transaction(trans, ret);
943 		goto out_free_path;
944 	}
945 
946 	leaf = path->nodes[0];
947 	ptr = btrfs_item_ptr(leaf, path->slots[0],
948 				 struct btrfs_qgroup_status_item);
949 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
950 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
951 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
952 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
953 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
954 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
955 
956 	btrfs_mark_buffer_dirty(leaf);
957 
958 	key.objectid = 0;
959 	key.type = BTRFS_ROOT_REF_KEY;
960 	key.offset = 0;
961 
962 	btrfs_release_path(path);
963 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
964 	if (ret > 0)
965 		goto out_add_root;
966 	if (ret < 0) {
967 		btrfs_abort_transaction(trans, ret);
968 		goto out_free_path;
969 	}
970 
971 	while (1) {
972 		slot = path->slots[0];
973 		leaf = path->nodes[0];
974 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
975 
976 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
977 			ret = add_qgroup_item(trans, quota_root,
978 					      found_key.offset);
979 			if (ret) {
980 				btrfs_abort_transaction(trans, ret);
981 				goto out_free_path;
982 			}
983 
984 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 			if (IS_ERR(qgroup)) {
986 				ret = PTR_ERR(qgroup);
987 				btrfs_abort_transaction(trans, ret);
988 				goto out_free_path;
989 			}
990 		}
991 		ret = btrfs_next_item(tree_root, path);
992 		if (ret < 0) {
993 			btrfs_abort_transaction(trans, ret);
994 			goto out_free_path;
995 		}
996 		if (ret)
997 			break;
998 	}
999 
1000 out_add_root:
1001 	btrfs_release_path(path);
1002 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1003 	if (ret) {
1004 		btrfs_abort_transaction(trans, ret);
1005 		goto out_free_path;
1006 	}
1007 
1008 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 	if (IS_ERR(qgroup)) {
1010 		ret = PTR_ERR(qgroup);
1011 		btrfs_abort_transaction(trans, ret);
1012 		goto out_free_path;
1013 	}
1014 
1015 	ret = btrfs_commit_transaction(trans);
1016 	trans = NULL;
1017 	if (ret)
1018 		goto out_free_path;
1019 
1020 	/*
1021 	 * Set quota enabled flag after committing the transaction, to avoid
1022 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1023 	 * creation.
1024 	 */
1025 	spin_lock(&fs_info->qgroup_lock);
1026 	fs_info->quota_root = quota_root;
1027 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1028 	spin_unlock(&fs_info->qgroup_lock);
1029 
1030 	ret = qgroup_rescan_init(fs_info, 0, 1);
1031 	if (!ret) {
1032 	        qgroup_rescan_zero_tracking(fs_info);
1033 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1034 	                         &fs_info->qgroup_rescan_work);
1035 	}
1036 
1037 out_free_path:
1038 	btrfs_free_path(path);
1039 out_free_root:
1040 	if (ret) {
1041 		free_extent_buffer(quota_root->node);
1042 		free_extent_buffer(quota_root->commit_root);
1043 		kfree(quota_root);
1044 	}
1045 out:
1046 	if (ret) {
1047 		ulist_free(fs_info->qgroup_ulist);
1048 		fs_info->qgroup_ulist = NULL;
1049 		if (trans)
1050 			btrfs_end_transaction(trans);
1051 	}
1052 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1053 	return ret;
1054 }
1055 
1056 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1057 {
1058 	struct btrfs_root *quota_root;
1059 	struct btrfs_trans_handle *trans = NULL;
1060 	int ret = 0;
1061 
1062 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1063 	if (!fs_info->quota_root)
1064 		goto out;
1065 
1066 	/*
1067 	 * 1 For the root item
1068 	 *
1069 	 * We should also reserve enough items for the quota tree deletion in
1070 	 * btrfs_clean_quota_tree but this is not done.
1071 	 */
1072 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1073 	if (IS_ERR(trans)) {
1074 		ret = PTR_ERR(trans);
1075 		goto out;
1076 	}
1077 
1078 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1079 	btrfs_qgroup_wait_for_completion(fs_info, false);
1080 	spin_lock(&fs_info->qgroup_lock);
1081 	quota_root = fs_info->quota_root;
1082 	fs_info->quota_root = NULL;
1083 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1084 	spin_unlock(&fs_info->qgroup_lock);
1085 
1086 	btrfs_free_qgroup_config(fs_info);
1087 
1088 	ret = btrfs_clean_quota_tree(trans, quota_root);
1089 	if (ret) {
1090 		btrfs_abort_transaction(trans, ret);
1091 		goto end_trans;
1092 	}
1093 
1094 	ret = btrfs_del_root(trans, &quota_root->root_key);
1095 	if (ret) {
1096 		btrfs_abort_transaction(trans, ret);
1097 		goto end_trans;
1098 	}
1099 
1100 	list_del(&quota_root->dirty_list);
1101 
1102 	btrfs_tree_lock(quota_root->node);
1103 	btrfs_clean_tree_block(quota_root->node);
1104 	btrfs_tree_unlock(quota_root->node);
1105 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1106 
1107 	free_extent_buffer(quota_root->node);
1108 	free_extent_buffer(quota_root->commit_root);
1109 	kfree(quota_root);
1110 
1111 end_trans:
1112 	ret = btrfs_end_transaction(trans);
1113 out:
1114 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1115 	return ret;
1116 }
1117 
1118 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1119 			 struct btrfs_qgroup *qgroup)
1120 {
1121 	if (list_empty(&qgroup->dirty))
1122 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1123 }
1124 
1125 /*
1126  * The easy accounting, we're updating qgroup relationship whose child qgroup
1127  * only has exclusive extents.
1128  *
1129  * In this case, all exclusive extents will also be exclusive for parent, so
1130  * excl/rfer just get added/removed.
1131  *
1132  * So is qgroup reservation space, which should also be added/removed to
1133  * parent.
1134  * Or when child tries to release reservation space, parent will underflow its
1135  * reservation (for relationship adding case).
1136  *
1137  * Caller should hold fs_info->qgroup_lock.
1138  */
1139 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1140 				    struct ulist *tmp, u64 ref_root,
1141 				    struct btrfs_qgroup *src, int sign)
1142 {
1143 	struct btrfs_qgroup *qgroup;
1144 	struct btrfs_qgroup_list *glist;
1145 	struct ulist_node *unode;
1146 	struct ulist_iterator uiter;
1147 	u64 num_bytes = src->excl;
1148 	int ret = 0;
1149 
1150 	qgroup = find_qgroup_rb(fs_info, ref_root);
1151 	if (!qgroup)
1152 		goto out;
1153 
1154 	qgroup->rfer += sign * num_bytes;
1155 	qgroup->rfer_cmpr += sign * num_bytes;
1156 
1157 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1158 	qgroup->excl += sign * num_bytes;
1159 	qgroup->excl_cmpr += sign * num_bytes;
1160 
1161 	if (sign > 0)
1162 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1163 	else
1164 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1165 
1166 	qgroup_dirty(fs_info, qgroup);
1167 
1168 	/* Get all of the parent groups that contain this qgroup */
1169 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1170 		ret = ulist_add(tmp, glist->group->qgroupid,
1171 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1172 		if (ret < 0)
1173 			goto out;
1174 	}
1175 
1176 	/* Iterate all of the parents and adjust their reference counts */
1177 	ULIST_ITER_INIT(&uiter);
1178 	while ((unode = ulist_next(tmp, &uiter))) {
1179 		qgroup = unode_aux_to_qgroup(unode);
1180 		qgroup->rfer += sign * num_bytes;
1181 		qgroup->rfer_cmpr += sign * num_bytes;
1182 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1183 		qgroup->excl += sign * num_bytes;
1184 		if (sign > 0)
1185 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1186 		else
1187 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1188 		qgroup->excl_cmpr += sign * num_bytes;
1189 		qgroup_dirty(fs_info, qgroup);
1190 
1191 		/* Add any parents of the parents */
1192 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1193 			ret = ulist_add(tmp, glist->group->qgroupid,
1194 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1195 			if (ret < 0)
1196 				goto out;
1197 		}
1198 	}
1199 	ret = 0;
1200 out:
1201 	return ret;
1202 }
1203 
1204 
1205 /*
1206  * Quick path for updating qgroup with only excl refs.
1207  *
1208  * In that case, just update all parent will be enough.
1209  * Or we needs to do a full rescan.
1210  * Caller should also hold fs_info->qgroup_lock.
1211  *
1212  * Return 0 for quick update, return >0 for need to full rescan
1213  * and mark INCONSISTENT flag.
1214  * Return < 0 for other error.
1215  */
1216 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1217 				   struct ulist *tmp, u64 src, u64 dst,
1218 				   int sign)
1219 {
1220 	struct btrfs_qgroup *qgroup;
1221 	int ret = 1;
1222 	int err = 0;
1223 
1224 	qgroup = find_qgroup_rb(fs_info, src);
1225 	if (!qgroup)
1226 		goto out;
1227 	if (qgroup->excl == qgroup->rfer) {
1228 		ret = 0;
1229 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1230 					       qgroup, sign);
1231 		if (err < 0) {
1232 			ret = err;
1233 			goto out;
1234 		}
1235 	}
1236 out:
1237 	if (ret)
1238 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1239 	return ret;
1240 }
1241 
1242 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1243 			      u64 dst)
1244 {
1245 	struct btrfs_fs_info *fs_info = trans->fs_info;
1246 	struct btrfs_qgroup *parent;
1247 	struct btrfs_qgroup *member;
1248 	struct btrfs_qgroup_list *list;
1249 	struct ulist *tmp;
1250 	int ret = 0;
1251 
1252 	/* Check the level of src and dst first */
1253 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1254 		return -EINVAL;
1255 
1256 	tmp = ulist_alloc(GFP_KERNEL);
1257 	if (!tmp)
1258 		return -ENOMEM;
1259 
1260 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1261 	if (!fs_info->quota_root) {
1262 		ret = -ENOTCONN;
1263 		goto out;
1264 	}
1265 	member = find_qgroup_rb(fs_info, src);
1266 	parent = find_qgroup_rb(fs_info, dst);
1267 	if (!member || !parent) {
1268 		ret = -EINVAL;
1269 		goto out;
1270 	}
1271 
1272 	/* check if such qgroup relation exist firstly */
1273 	list_for_each_entry(list, &member->groups, next_group) {
1274 		if (list->group == parent) {
1275 			ret = -EEXIST;
1276 			goto out;
1277 		}
1278 	}
1279 
1280 	ret = add_qgroup_relation_item(trans, src, dst);
1281 	if (ret)
1282 		goto out;
1283 
1284 	ret = add_qgroup_relation_item(trans, dst, src);
1285 	if (ret) {
1286 		del_qgroup_relation_item(trans, src, dst);
1287 		goto out;
1288 	}
1289 
1290 	spin_lock(&fs_info->qgroup_lock);
1291 	ret = add_relation_rb(fs_info, src, dst);
1292 	if (ret < 0) {
1293 		spin_unlock(&fs_info->qgroup_lock);
1294 		goto out;
1295 	}
1296 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1297 	spin_unlock(&fs_info->qgroup_lock);
1298 out:
1299 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1300 	ulist_free(tmp);
1301 	return ret;
1302 }
1303 
1304 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1305 				 u64 dst)
1306 {
1307 	struct btrfs_fs_info *fs_info = trans->fs_info;
1308 	struct btrfs_qgroup *parent;
1309 	struct btrfs_qgroup *member;
1310 	struct btrfs_qgroup_list *list;
1311 	struct ulist *tmp;
1312 	bool found = false;
1313 	int ret = 0;
1314 	int ret2;
1315 
1316 	tmp = ulist_alloc(GFP_KERNEL);
1317 	if (!tmp)
1318 		return -ENOMEM;
1319 
1320 	if (!fs_info->quota_root) {
1321 		ret = -ENOTCONN;
1322 		goto out;
1323 	}
1324 
1325 	member = find_qgroup_rb(fs_info, src);
1326 	parent = find_qgroup_rb(fs_info, dst);
1327 	/*
1328 	 * The parent/member pair doesn't exist, then try to delete the dead
1329 	 * relation items only.
1330 	 */
1331 	if (!member || !parent)
1332 		goto delete_item;
1333 
1334 	/* check if such qgroup relation exist firstly */
1335 	list_for_each_entry(list, &member->groups, next_group) {
1336 		if (list->group == parent) {
1337 			found = true;
1338 			break;
1339 		}
1340 	}
1341 
1342 delete_item:
1343 	ret = del_qgroup_relation_item(trans, src, dst);
1344 	if (ret < 0 && ret != -ENOENT)
1345 		goto out;
1346 	ret2 = del_qgroup_relation_item(trans, dst, src);
1347 	if (ret2 < 0 && ret2 != -ENOENT)
1348 		goto out;
1349 
1350 	/* At least one deletion succeeded, return 0 */
1351 	if (!ret || !ret2)
1352 		ret = 0;
1353 
1354 	if (found) {
1355 		spin_lock(&fs_info->qgroup_lock);
1356 		del_relation_rb(fs_info, src, dst);
1357 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1358 		spin_unlock(&fs_info->qgroup_lock);
1359 	}
1360 out:
1361 	ulist_free(tmp);
1362 	return ret;
1363 }
1364 
1365 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1366 			      u64 dst)
1367 {
1368 	struct btrfs_fs_info *fs_info = trans->fs_info;
1369 	int ret = 0;
1370 
1371 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1372 	ret = __del_qgroup_relation(trans, src, dst);
1373 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1374 
1375 	return ret;
1376 }
1377 
1378 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1379 {
1380 	struct btrfs_fs_info *fs_info = trans->fs_info;
1381 	struct btrfs_root *quota_root;
1382 	struct btrfs_qgroup *qgroup;
1383 	int ret = 0;
1384 
1385 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1386 	if (!fs_info->quota_root) {
1387 		ret = -ENOTCONN;
1388 		goto out;
1389 	}
1390 	quota_root = fs_info->quota_root;
1391 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1392 	if (qgroup) {
1393 		ret = -EEXIST;
1394 		goto out;
1395 	}
1396 
1397 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1398 	if (ret)
1399 		goto out;
1400 
1401 	spin_lock(&fs_info->qgroup_lock);
1402 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1403 	spin_unlock(&fs_info->qgroup_lock);
1404 
1405 	if (IS_ERR(qgroup))
1406 		ret = PTR_ERR(qgroup);
1407 out:
1408 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1409 	return ret;
1410 }
1411 
1412 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1413 {
1414 	struct btrfs_fs_info *fs_info = trans->fs_info;
1415 	struct btrfs_qgroup *qgroup;
1416 	struct btrfs_qgroup_list *list;
1417 	int ret = 0;
1418 
1419 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1420 	if (!fs_info->quota_root) {
1421 		ret = -ENOTCONN;
1422 		goto out;
1423 	}
1424 
1425 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1426 	if (!qgroup) {
1427 		ret = -ENOENT;
1428 		goto out;
1429 	}
1430 
1431 	/* Check if there are no children of this qgroup */
1432 	if (!list_empty(&qgroup->members)) {
1433 		ret = -EBUSY;
1434 		goto out;
1435 	}
1436 
1437 	ret = del_qgroup_item(trans, qgroupid);
1438 	if (ret && ret != -ENOENT)
1439 		goto out;
1440 
1441 	while (!list_empty(&qgroup->groups)) {
1442 		list = list_first_entry(&qgroup->groups,
1443 					struct btrfs_qgroup_list, next_group);
1444 		ret = __del_qgroup_relation(trans, qgroupid,
1445 					    list->group->qgroupid);
1446 		if (ret)
1447 			goto out;
1448 	}
1449 
1450 	spin_lock(&fs_info->qgroup_lock);
1451 	del_qgroup_rb(fs_info, qgroupid);
1452 	spin_unlock(&fs_info->qgroup_lock);
1453 out:
1454 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1455 	return ret;
1456 }
1457 
1458 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1459 		       struct btrfs_qgroup_limit *limit)
1460 {
1461 	struct btrfs_fs_info *fs_info = trans->fs_info;
1462 	struct btrfs_qgroup *qgroup;
1463 	int ret = 0;
1464 	/* Sometimes we would want to clear the limit on this qgroup.
1465 	 * To meet this requirement, we treat the -1 as a special value
1466 	 * which tell kernel to clear the limit on this qgroup.
1467 	 */
1468 	const u64 CLEAR_VALUE = -1;
1469 
1470 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1471 	if (!fs_info->quota_root) {
1472 		ret = -ENOTCONN;
1473 		goto out;
1474 	}
1475 
1476 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1477 	if (!qgroup) {
1478 		ret = -ENOENT;
1479 		goto out;
1480 	}
1481 
1482 	spin_lock(&fs_info->qgroup_lock);
1483 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1484 		if (limit->max_rfer == CLEAR_VALUE) {
1485 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1486 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1487 			qgroup->max_rfer = 0;
1488 		} else {
1489 			qgroup->max_rfer = limit->max_rfer;
1490 		}
1491 	}
1492 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1493 		if (limit->max_excl == CLEAR_VALUE) {
1494 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1495 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1496 			qgroup->max_excl = 0;
1497 		} else {
1498 			qgroup->max_excl = limit->max_excl;
1499 		}
1500 	}
1501 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1502 		if (limit->rsv_rfer == CLEAR_VALUE) {
1503 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1504 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1505 			qgroup->rsv_rfer = 0;
1506 		} else {
1507 			qgroup->rsv_rfer = limit->rsv_rfer;
1508 		}
1509 	}
1510 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1511 		if (limit->rsv_excl == CLEAR_VALUE) {
1512 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1513 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1514 			qgroup->rsv_excl = 0;
1515 		} else {
1516 			qgroup->rsv_excl = limit->rsv_excl;
1517 		}
1518 	}
1519 	qgroup->lim_flags |= limit->flags;
1520 
1521 	spin_unlock(&fs_info->qgroup_lock);
1522 
1523 	ret = update_qgroup_limit_item(trans, qgroup);
1524 	if (ret) {
1525 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1526 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1527 		       qgroupid);
1528 	}
1529 
1530 out:
1531 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1532 	return ret;
1533 }
1534 
1535 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1536 				struct btrfs_delayed_ref_root *delayed_refs,
1537 				struct btrfs_qgroup_extent_record *record)
1538 {
1539 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1540 	struct rb_node *parent_node = NULL;
1541 	struct btrfs_qgroup_extent_record *entry;
1542 	u64 bytenr = record->bytenr;
1543 
1544 	lockdep_assert_held(&delayed_refs->lock);
1545 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1546 
1547 	while (*p) {
1548 		parent_node = *p;
1549 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1550 				 node);
1551 		if (bytenr < entry->bytenr) {
1552 			p = &(*p)->rb_left;
1553 		} else if (bytenr > entry->bytenr) {
1554 			p = &(*p)->rb_right;
1555 		} else {
1556 			if (record->data_rsv && !entry->data_rsv) {
1557 				entry->data_rsv = record->data_rsv;
1558 				entry->data_rsv_refroot =
1559 					record->data_rsv_refroot;
1560 			}
1561 			return 1;
1562 		}
1563 	}
1564 
1565 	rb_link_node(&record->node, parent_node, p);
1566 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1567 	return 0;
1568 }
1569 
1570 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1571 				   struct btrfs_qgroup_extent_record *qrecord)
1572 {
1573 	struct ulist *old_root;
1574 	u64 bytenr = qrecord->bytenr;
1575 	int ret;
1576 
1577 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1578 	if (ret < 0) {
1579 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1580 		btrfs_warn(fs_info,
1581 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1582 			ret);
1583 		return 0;
1584 	}
1585 
1586 	/*
1587 	 * Here we don't need to get the lock of
1588 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1589 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1590 	 *
1591 	 * So modifying qrecord->old_roots is safe here
1592 	 */
1593 	qrecord->old_roots = old_root;
1594 	return 0;
1595 }
1596 
1597 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1598 			      u64 num_bytes, gfp_t gfp_flag)
1599 {
1600 	struct btrfs_fs_info *fs_info = trans->fs_info;
1601 	struct btrfs_qgroup_extent_record *record;
1602 	struct btrfs_delayed_ref_root *delayed_refs;
1603 	int ret;
1604 
1605 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1606 	    || bytenr == 0 || num_bytes == 0)
1607 		return 0;
1608 	record = kzalloc(sizeof(*record), gfp_flag);
1609 	if (!record)
1610 		return -ENOMEM;
1611 
1612 	delayed_refs = &trans->transaction->delayed_refs;
1613 	record->bytenr = bytenr;
1614 	record->num_bytes = num_bytes;
1615 	record->old_roots = NULL;
1616 
1617 	spin_lock(&delayed_refs->lock);
1618 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1619 	spin_unlock(&delayed_refs->lock);
1620 	if (ret > 0) {
1621 		kfree(record);
1622 		return 0;
1623 	}
1624 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1625 }
1626 
1627 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1628 				  struct extent_buffer *eb)
1629 {
1630 	struct btrfs_fs_info *fs_info = trans->fs_info;
1631 	int nr = btrfs_header_nritems(eb);
1632 	int i, extent_type, ret;
1633 	struct btrfs_key key;
1634 	struct btrfs_file_extent_item *fi;
1635 	u64 bytenr, num_bytes;
1636 
1637 	/* We can be called directly from walk_up_proc() */
1638 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1639 		return 0;
1640 
1641 	for (i = 0; i < nr; i++) {
1642 		btrfs_item_key_to_cpu(eb, &key, i);
1643 
1644 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1645 			continue;
1646 
1647 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1648 		/* filter out non qgroup-accountable extents  */
1649 		extent_type = btrfs_file_extent_type(eb, fi);
1650 
1651 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1652 			continue;
1653 
1654 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1655 		if (!bytenr)
1656 			continue;
1657 
1658 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1659 
1660 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1661 						GFP_NOFS);
1662 		if (ret)
1663 			return ret;
1664 	}
1665 	cond_resched();
1666 	return 0;
1667 }
1668 
1669 /*
1670  * Walk up the tree from the bottom, freeing leaves and any interior
1671  * nodes which have had all slots visited. If a node (leaf or
1672  * interior) is freed, the node above it will have it's slot
1673  * incremented. The root node will never be freed.
1674  *
1675  * At the end of this function, we should have a path which has all
1676  * slots incremented to the next position for a search. If we need to
1677  * read a new node it will be NULL and the node above it will have the
1678  * correct slot selected for a later read.
1679  *
1680  * If we increment the root nodes slot counter past the number of
1681  * elements, 1 is returned to signal completion of the search.
1682  */
1683 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1684 {
1685 	int level = 0;
1686 	int nr, slot;
1687 	struct extent_buffer *eb;
1688 
1689 	if (root_level == 0)
1690 		return 1;
1691 
1692 	while (level <= root_level) {
1693 		eb = path->nodes[level];
1694 		nr = btrfs_header_nritems(eb);
1695 		path->slots[level]++;
1696 		slot = path->slots[level];
1697 		if (slot >= nr || level == 0) {
1698 			/*
1699 			 * Don't free the root -  we will detect this
1700 			 * condition after our loop and return a
1701 			 * positive value for caller to stop walking the tree.
1702 			 */
1703 			if (level != root_level) {
1704 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1705 				path->locks[level] = 0;
1706 
1707 				free_extent_buffer(eb);
1708 				path->nodes[level] = NULL;
1709 				path->slots[level] = 0;
1710 			}
1711 		} else {
1712 			/*
1713 			 * We have a valid slot to walk back down
1714 			 * from. Stop here so caller can process these
1715 			 * new nodes.
1716 			 */
1717 			break;
1718 		}
1719 
1720 		level++;
1721 	}
1722 
1723 	eb = path->nodes[root_level];
1724 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1725 		return 1;
1726 
1727 	return 0;
1728 }
1729 
1730 /*
1731  * Helper function to trace a subtree tree block swap.
1732  *
1733  * The swap will happen in highest tree block, but there may be a lot of
1734  * tree blocks involved.
1735  *
1736  * For example:
1737  *  OO = Old tree blocks
1738  *  NN = New tree blocks allocated during balance
1739  *
1740  *           File tree (257)                  Reloc tree for 257
1741  * L2              OO                                NN
1742  *               /    \                            /    \
1743  * L1          OO      OO (a)                    OO      NN (a)
1744  *            / \     / \                       / \     / \
1745  * L0       OO   OO OO   OO                   OO   OO NN   NN
1746  *                  (b)  (c)                          (b)  (c)
1747  *
1748  * When calling qgroup_trace_extent_swap(), we will pass:
1749  * @src_eb = OO(a)
1750  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1751  * @dst_level = 0
1752  * @root_level = 1
1753  *
1754  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1755  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1756  *
1757  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1758  *
1759  * 1) Tree search from @src_eb
1760  *    It should acts as a simplified btrfs_search_slot().
1761  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1762  *    (first key).
1763  *
1764  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1765  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1766  *    They should be marked during previous (@dst_level = 1) iteration.
1767  *
1768  * 3) Mark file extents in leaves dirty
1769  *    We don't have good way to pick out new file extents only.
1770  *    So we still follow the old method by scanning all file extents in
1771  *    the leave.
1772  *
1773  * This function can free us from keeping two paths, thus later we only need
1774  * to care about how to iterate all new tree blocks in reloc tree.
1775  */
1776 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1777 				    struct extent_buffer *src_eb,
1778 				    struct btrfs_path *dst_path,
1779 				    int dst_level, int root_level,
1780 				    bool trace_leaf)
1781 {
1782 	struct btrfs_key key;
1783 	struct btrfs_path *src_path;
1784 	struct btrfs_fs_info *fs_info = trans->fs_info;
1785 	u32 nodesize = fs_info->nodesize;
1786 	int cur_level = root_level;
1787 	int ret;
1788 
1789 	BUG_ON(dst_level > root_level);
1790 	/* Level mismatch */
1791 	if (btrfs_header_level(src_eb) != root_level)
1792 		return -EINVAL;
1793 
1794 	src_path = btrfs_alloc_path();
1795 	if (!src_path) {
1796 		ret = -ENOMEM;
1797 		goto out;
1798 	}
1799 
1800 	if (dst_level)
1801 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1802 	else
1803 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1804 
1805 	/* For src_path */
1806 	atomic_inc(&src_eb->refs);
1807 	src_path->nodes[root_level] = src_eb;
1808 	src_path->slots[root_level] = dst_path->slots[root_level];
1809 	src_path->locks[root_level] = 0;
1810 
1811 	/* A simplified version of btrfs_search_slot() */
1812 	while (cur_level >= dst_level) {
1813 		struct btrfs_key src_key;
1814 		struct btrfs_key dst_key;
1815 
1816 		if (src_path->nodes[cur_level] == NULL) {
1817 			struct btrfs_key first_key;
1818 			struct extent_buffer *eb;
1819 			int parent_slot;
1820 			u64 child_gen;
1821 			u64 child_bytenr;
1822 
1823 			eb = src_path->nodes[cur_level + 1];
1824 			parent_slot = src_path->slots[cur_level + 1];
1825 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1826 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1827 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1828 
1829 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1830 					     cur_level, &first_key);
1831 			if (IS_ERR(eb)) {
1832 				ret = PTR_ERR(eb);
1833 				goto out;
1834 			} else if (!extent_buffer_uptodate(eb)) {
1835 				free_extent_buffer(eb);
1836 				ret = -EIO;
1837 				goto out;
1838 			}
1839 
1840 			src_path->nodes[cur_level] = eb;
1841 
1842 			btrfs_tree_read_lock(eb);
1843 			btrfs_set_lock_blocking_read(eb);
1844 			src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1845 		}
1846 
1847 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1848 		if (cur_level) {
1849 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1850 					&dst_key, dst_path->slots[cur_level]);
1851 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1852 					&src_key, src_path->slots[cur_level]);
1853 		} else {
1854 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1855 					&dst_key, dst_path->slots[cur_level]);
1856 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1857 					&src_key, src_path->slots[cur_level]);
1858 		}
1859 		/* Content mismatch, something went wrong */
1860 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1861 			ret = -ENOENT;
1862 			goto out;
1863 		}
1864 		cur_level--;
1865 	}
1866 
1867 	/*
1868 	 * Now both @dst_path and @src_path have been populated, record the tree
1869 	 * blocks for qgroup accounting.
1870 	 */
1871 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1872 			nodesize, GFP_NOFS);
1873 	if (ret < 0)
1874 		goto out;
1875 	ret = btrfs_qgroup_trace_extent(trans,
1876 			dst_path->nodes[dst_level]->start,
1877 			nodesize, GFP_NOFS);
1878 	if (ret < 0)
1879 		goto out;
1880 
1881 	/* Record leaf file extents */
1882 	if (dst_level == 0 && trace_leaf) {
1883 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1884 		if (ret < 0)
1885 			goto out;
1886 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1887 	}
1888 out:
1889 	btrfs_free_path(src_path);
1890 	return ret;
1891 }
1892 
1893 /*
1894  * Helper function to do recursive generation-aware depth-first search, to
1895  * locate all new tree blocks in a subtree of reloc tree.
1896  *
1897  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1898  *         reloc tree
1899  * L2         NN (a)
1900  *          /    \
1901  * L1    OO        NN (b)
1902  *      /  \      /  \
1903  * L0  OO  OO    OO  NN
1904  *               (c) (d)
1905  * If we pass:
1906  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1907  * @cur_level = 1
1908  * @root_level = 1
1909  *
1910  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1911  * above tree blocks along with their counter parts in file tree.
1912  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1913  * won't affect OO(c).
1914  */
1915 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1916 					   struct extent_buffer *src_eb,
1917 					   struct btrfs_path *dst_path,
1918 					   int cur_level, int root_level,
1919 					   u64 last_snapshot, bool trace_leaf)
1920 {
1921 	struct btrfs_fs_info *fs_info = trans->fs_info;
1922 	struct extent_buffer *eb;
1923 	bool need_cleanup = false;
1924 	int ret = 0;
1925 	int i;
1926 
1927 	/* Level sanity check */
1928 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1929 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1930 	    root_level < cur_level) {
1931 		btrfs_err_rl(fs_info,
1932 			"%s: bad levels, cur_level=%d root_level=%d",
1933 			__func__, cur_level, root_level);
1934 		return -EUCLEAN;
1935 	}
1936 
1937 	/* Read the tree block if needed */
1938 	if (dst_path->nodes[cur_level] == NULL) {
1939 		struct btrfs_key first_key;
1940 		int parent_slot;
1941 		u64 child_gen;
1942 		u64 child_bytenr;
1943 
1944 		/*
1945 		 * dst_path->nodes[root_level] must be initialized before
1946 		 * calling this function.
1947 		 */
1948 		if (cur_level == root_level) {
1949 			btrfs_err_rl(fs_info,
1950 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1951 				__func__, root_level, root_level, cur_level);
1952 			return -EUCLEAN;
1953 		}
1954 
1955 		/*
1956 		 * We need to get child blockptr/gen from parent before we can
1957 		 * read it.
1958 		  */
1959 		eb = dst_path->nodes[cur_level + 1];
1960 		parent_slot = dst_path->slots[cur_level + 1];
1961 		child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1962 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1963 		btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1964 
1965 		/* This node is old, no need to trace */
1966 		if (child_gen < last_snapshot)
1967 			goto out;
1968 
1969 		eb = read_tree_block(fs_info, child_bytenr, child_gen,
1970 				     cur_level, &first_key);
1971 		if (IS_ERR(eb)) {
1972 			ret = PTR_ERR(eb);
1973 			goto out;
1974 		} else if (!extent_buffer_uptodate(eb)) {
1975 			free_extent_buffer(eb);
1976 			ret = -EIO;
1977 			goto out;
1978 		}
1979 
1980 		dst_path->nodes[cur_level] = eb;
1981 		dst_path->slots[cur_level] = 0;
1982 
1983 		btrfs_tree_read_lock(eb);
1984 		btrfs_set_lock_blocking_read(eb);
1985 		dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1986 		need_cleanup = true;
1987 	}
1988 
1989 	/* Now record this tree block and its counter part for qgroups */
1990 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1991 				       root_level, trace_leaf);
1992 	if (ret < 0)
1993 		goto cleanup;
1994 
1995 	eb = dst_path->nodes[cur_level];
1996 
1997 	if (cur_level > 0) {
1998 		/* Iterate all child tree blocks */
1999 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2000 			/* Skip old tree blocks as they won't be swapped */
2001 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2002 				continue;
2003 			dst_path->slots[cur_level] = i;
2004 
2005 			/* Recursive call (at most 7 times) */
2006 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2007 					dst_path, cur_level - 1, root_level,
2008 					last_snapshot, trace_leaf);
2009 			if (ret < 0)
2010 				goto cleanup;
2011 		}
2012 	}
2013 
2014 cleanup:
2015 	if (need_cleanup) {
2016 		/* Clean up */
2017 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2018 				     dst_path->locks[cur_level]);
2019 		free_extent_buffer(dst_path->nodes[cur_level]);
2020 		dst_path->nodes[cur_level] = NULL;
2021 		dst_path->slots[cur_level] = 0;
2022 		dst_path->locks[cur_level] = 0;
2023 	}
2024 out:
2025 	return ret;
2026 }
2027 
2028 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2029 				struct extent_buffer *src_eb,
2030 				struct extent_buffer *dst_eb,
2031 				u64 last_snapshot, bool trace_leaf)
2032 {
2033 	struct btrfs_fs_info *fs_info = trans->fs_info;
2034 	struct btrfs_path *dst_path = NULL;
2035 	int level;
2036 	int ret;
2037 
2038 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2039 		return 0;
2040 
2041 	/* Wrong parameter order */
2042 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2043 		btrfs_err_rl(fs_info,
2044 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2045 			     btrfs_header_generation(src_eb),
2046 			     btrfs_header_generation(dst_eb));
2047 		return -EUCLEAN;
2048 	}
2049 
2050 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2051 		ret = -EIO;
2052 		goto out;
2053 	}
2054 
2055 	level = btrfs_header_level(dst_eb);
2056 	dst_path = btrfs_alloc_path();
2057 	if (!dst_path) {
2058 		ret = -ENOMEM;
2059 		goto out;
2060 	}
2061 	/* For dst_path */
2062 	atomic_inc(&dst_eb->refs);
2063 	dst_path->nodes[level] = dst_eb;
2064 	dst_path->slots[level] = 0;
2065 	dst_path->locks[level] = 0;
2066 
2067 	/* Do the generation aware breadth-first search */
2068 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2069 					      level, last_snapshot, trace_leaf);
2070 	if (ret < 0)
2071 		goto out;
2072 	ret = 0;
2073 
2074 out:
2075 	btrfs_free_path(dst_path);
2076 	if (ret < 0)
2077 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2078 	return ret;
2079 }
2080 
2081 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2082 			       struct extent_buffer *root_eb,
2083 			       u64 root_gen, int root_level)
2084 {
2085 	struct btrfs_fs_info *fs_info = trans->fs_info;
2086 	int ret = 0;
2087 	int level;
2088 	struct extent_buffer *eb = root_eb;
2089 	struct btrfs_path *path = NULL;
2090 
2091 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2092 	BUG_ON(root_eb == NULL);
2093 
2094 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2095 		return 0;
2096 
2097 	if (!extent_buffer_uptodate(root_eb)) {
2098 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2099 		if (ret)
2100 			goto out;
2101 	}
2102 
2103 	if (root_level == 0) {
2104 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2105 		goto out;
2106 	}
2107 
2108 	path = btrfs_alloc_path();
2109 	if (!path)
2110 		return -ENOMEM;
2111 
2112 	/*
2113 	 * Walk down the tree.  Missing extent blocks are filled in as
2114 	 * we go. Metadata is accounted every time we read a new
2115 	 * extent block.
2116 	 *
2117 	 * When we reach a leaf, we account for file extent items in it,
2118 	 * walk back up the tree (adjusting slot pointers as we go)
2119 	 * and restart the search process.
2120 	 */
2121 	atomic_inc(&root_eb->refs);	/* For path */
2122 	path->nodes[root_level] = root_eb;
2123 	path->slots[root_level] = 0;
2124 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2125 walk_down:
2126 	level = root_level;
2127 	while (level >= 0) {
2128 		if (path->nodes[level] == NULL) {
2129 			struct btrfs_key first_key;
2130 			int parent_slot;
2131 			u64 child_gen;
2132 			u64 child_bytenr;
2133 
2134 			/*
2135 			 * We need to get child blockptr/gen from parent before
2136 			 * we can read it.
2137 			  */
2138 			eb = path->nodes[level + 1];
2139 			parent_slot = path->slots[level + 1];
2140 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2141 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2142 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2143 
2144 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
2145 					     level, &first_key);
2146 			if (IS_ERR(eb)) {
2147 				ret = PTR_ERR(eb);
2148 				goto out;
2149 			} else if (!extent_buffer_uptodate(eb)) {
2150 				free_extent_buffer(eb);
2151 				ret = -EIO;
2152 				goto out;
2153 			}
2154 
2155 			path->nodes[level] = eb;
2156 			path->slots[level] = 0;
2157 
2158 			btrfs_tree_read_lock(eb);
2159 			btrfs_set_lock_blocking_read(eb);
2160 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2161 
2162 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2163 							fs_info->nodesize,
2164 							GFP_NOFS);
2165 			if (ret)
2166 				goto out;
2167 		}
2168 
2169 		if (level == 0) {
2170 			ret = btrfs_qgroup_trace_leaf_items(trans,
2171 							    path->nodes[level]);
2172 			if (ret)
2173 				goto out;
2174 
2175 			/* Nonzero return here means we completed our search */
2176 			ret = adjust_slots_upwards(path, root_level);
2177 			if (ret)
2178 				break;
2179 
2180 			/* Restart search with new slots */
2181 			goto walk_down;
2182 		}
2183 
2184 		level--;
2185 	}
2186 
2187 	ret = 0;
2188 out:
2189 	btrfs_free_path(path);
2190 
2191 	return ret;
2192 }
2193 
2194 #define UPDATE_NEW	0
2195 #define UPDATE_OLD	1
2196 /*
2197  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2198  */
2199 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2200 				struct ulist *roots, struct ulist *tmp,
2201 				struct ulist *qgroups, u64 seq, int update_old)
2202 {
2203 	struct ulist_node *unode;
2204 	struct ulist_iterator uiter;
2205 	struct ulist_node *tmp_unode;
2206 	struct ulist_iterator tmp_uiter;
2207 	struct btrfs_qgroup *qg;
2208 	int ret = 0;
2209 
2210 	if (!roots)
2211 		return 0;
2212 	ULIST_ITER_INIT(&uiter);
2213 	while ((unode = ulist_next(roots, &uiter))) {
2214 		qg = find_qgroup_rb(fs_info, unode->val);
2215 		if (!qg)
2216 			continue;
2217 
2218 		ulist_reinit(tmp);
2219 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2220 				GFP_ATOMIC);
2221 		if (ret < 0)
2222 			return ret;
2223 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2224 		if (ret < 0)
2225 			return ret;
2226 		ULIST_ITER_INIT(&tmp_uiter);
2227 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2228 			struct btrfs_qgroup_list *glist;
2229 
2230 			qg = unode_aux_to_qgroup(tmp_unode);
2231 			if (update_old)
2232 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2233 			else
2234 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2235 			list_for_each_entry(glist, &qg->groups, next_group) {
2236 				ret = ulist_add(qgroups, glist->group->qgroupid,
2237 						qgroup_to_aux(glist->group),
2238 						GFP_ATOMIC);
2239 				if (ret < 0)
2240 					return ret;
2241 				ret = ulist_add(tmp, glist->group->qgroupid,
2242 						qgroup_to_aux(glist->group),
2243 						GFP_ATOMIC);
2244 				if (ret < 0)
2245 					return ret;
2246 			}
2247 		}
2248 	}
2249 	return 0;
2250 }
2251 
2252 /*
2253  * Update qgroup rfer/excl counters.
2254  * Rfer update is easy, codes can explain themselves.
2255  *
2256  * Excl update is tricky, the update is split into 2 part.
2257  * Part 1: Possible exclusive <-> sharing detect:
2258  *	|	A	|	!A	|
2259  *  -------------------------------------
2260  *  B	|	*	|	-	|
2261  *  -------------------------------------
2262  *  !B	|	+	|	**	|
2263  *  -------------------------------------
2264  *
2265  * Conditions:
2266  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2267  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2268  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2269  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2270  *
2271  * Results:
2272  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2273  * *: Definitely not changed.		**: Possible unchanged.
2274  *
2275  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2276  *
2277  * To make the logic clear, we first use condition A and B to split
2278  * combination into 4 results.
2279  *
2280  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2281  * only on variant maybe 0.
2282  *
2283  * Lastly, check result **, since there are 2 variants maybe 0, split them
2284  * again(2x2).
2285  * But this time we don't need to consider other things, the codes and logic
2286  * is easy to understand now.
2287  */
2288 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2289 				  struct ulist *qgroups,
2290 				  u64 nr_old_roots,
2291 				  u64 nr_new_roots,
2292 				  u64 num_bytes, u64 seq)
2293 {
2294 	struct ulist_node *unode;
2295 	struct ulist_iterator uiter;
2296 	struct btrfs_qgroup *qg;
2297 	u64 cur_new_count, cur_old_count;
2298 
2299 	ULIST_ITER_INIT(&uiter);
2300 	while ((unode = ulist_next(qgroups, &uiter))) {
2301 		bool dirty = false;
2302 
2303 		qg = unode_aux_to_qgroup(unode);
2304 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2305 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2306 
2307 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2308 					     cur_new_count);
2309 
2310 		/* Rfer update part */
2311 		if (cur_old_count == 0 && cur_new_count > 0) {
2312 			qg->rfer += num_bytes;
2313 			qg->rfer_cmpr += num_bytes;
2314 			dirty = true;
2315 		}
2316 		if (cur_old_count > 0 && cur_new_count == 0) {
2317 			qg->rfer -= num_bytes;
2318 			qg->rfer_cmpr -= num_bytes;
2319 			dirty = true;
2320 		}
2321 
2322 		/* Excl update part */
2323 		/* Exclusive/none -> shared case */
2324 		if (cur_old_count == nr_old_roots &&
2325 		    cur_new_count < nr_new_roots) {
2326 			/* Exclusive -> shared */
2327 			if (cur_old_count != 0) {
2328 				qg->excl -= num_bytes;
2329 				qg->excl_cmpr -= num_bytes;
2330 				dirty = true;
2331 			}
2332 		}
2333 
2334 		/* Shared -> exclusive/none case */
2335 		if (cur_old_count < nr_old_roots &&
2336 		    cur_new_count == nr_new_roots) {
2337 			/* Shared->exclusive */
2338 			if (cur_new_count != 0) {
2339 				qg->excl += num_bytes;
2340 				qg->excl_cmpr += num_bytes;
2341 				dirty = true;
2342 			}
2343 		}
2344 
2345 		/* Exclusive/none -> exclusive/none case */
2346 		if (cur_old_count == nr_old_roots &&
2347 		    cur_new_count == nr_new_roots) {
2348 			if (cur_old_count == 0) {
2349 				/* None -> exclusive/none */
2350 
2351 				if (cur_new_count != 0) {
2352 					/* None -> exclusive */
2353 					qg->excl += num_bytes;
2354 					qg->excl_cmpr += num_bytes;
2355 					dirty = true;
2356 				}
2357 				/* None -> none, nothing changed */
2358 			} else {
2359 				/* Exclusive -> exclusive/none */
2360 
2361 				if (cur_new_count == 0) {
2362 					/* Exclusive -> none */
2363 					qg->excl -= num_bytes;
2364 					qg->excl_cmpr -= num_bytes;
2365 					dirty = true;
2366 				}
2367 				/* Exclusive -> exclusive, nothing changed */
2368 			}
2369 		}
2370 
2371 		if (dirty)
2372 			qgroup_dirty(fs_info, qg);
2373 	}
2374 	return 0;
2375 }
2376 
2377 /*
2378  * Check if the @roots potentially is a list of fs tree roots
2379  *
2380  * Return 0 for definitely not a fs/subvol tree roots ulist
2381  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2382  *          one as well)
2383  */
2384 static int maybe_fs_roots(struct ulist *roots)
2385 {
2386 	struct ulist_node *unode;
2387 	struct ulist_iterator uiter;
2388 
2389 	/* Empty one, still possible for fs roots */
2390 	if (!roots || roots->nnodes == 0)
2391 		return 1;
2392 
2393 	ULIST_ITER_INIT(&uiter);
2394 	unode = ulist_next(roots, &uiter);
2395 	if (!unode)
2396 		return 1;
2397 
2398 	/*
2399 	 * If it contains fs tree roots, then it must belong to fs/subvol
2400 	 * trees.
2401 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2402 	 */
2403 	return is_fstree(unode->val);
2404 }
2405 
2406 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2407 				u64 num_bytes, struct ulist *old_roots,
2408 				struct ulist *new_roots)
2409 {
2410 	struct btrfs_fs_info *fs_info = trans->fs_info;
2411 	struct ulist *qgroups = NULL;
2412 	struct ulist *tmp = NULL;
2413 	u64 seq;
2414 	u64 nr_new_roots = 0;
2415 	u64 nr_old_roots = 0;
2416 	int ret = 0;
2417 
2418 	/*
2419 	 * If quotas get disabled meanwhile, the resouces need to be freed and
2420 	 * we can't just exit here.
2421 	 */
2422 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2423 		goto out_free;
2424 
2425 	if (new_roots) {
2426 		if (!maybe_fs_roots(new_roots))
2427 			goto out_free;
2428 		nr_new_roots = new_roots->nnodes;
2429 	}
2430 	if (old_roots) {
2431 		if (!maybe_fs_roots(old_roots))
2432 			goto out_free;
2433 		nr_old_roots = old_roots->nnodes;
2434 	}
2435 
2436 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2437 	if (nr_old_roots == 0 && nr_new_roots == 0)
2438 		goto out_free;
2439 
2440 	BUG_ON(!fs_info->quota_root);
2441 
2442 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2443 					num_bytes, nr_old_roots, nr_new_roots);
2444 
2445 	qgroups = ulist_alloc(GFP_NOFS);
2446 	if (!qgroups) {
2447 		ret = -ENOMEM;
2448 		goto out_free;
2449 	}
2450 	tmp = ulist_alloc(GFP_NOFS);
2451 	if (!tmp) {
2452 		ret = -ENOMEM;
2453 		goto out_free;
2454 	}
2455 
2456 	mutex_lock(&fs_info->qgroup_rescan_lock);
2457 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2458 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2459 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2460 			ret = 0;
2461 			goto out_free;
2462 		}
2463 	}
2464 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2465 
2466 	spin_lock(&fs_info->qgroup_lock);
2467 	seq = fs_info->qgroup_seq;
2468 
2469 	/* Update old refcnts using old_roots */
2470 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2471 				   UPDATE_OLD);
2472 	if (ret < 0)
2473 		goto out;
2474 
2475 	/* Update new refcnts using new_roots */
2476 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2477 				   UPDATE_NEW);
2478 	if (ret < 0)
2479 		goto out;
2480 
2481 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2482 			       num_bytes, seq);
2483 
2484 	/*
2485 	 * Bump qgroup_seq to avoid seq overlap
2486 	 */
2487 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2488 out:
2489 	spin_unlock(&fs_info->qgroup_lock);
2490 out_free:
2491 	ulist_free(tmp);
2492 	ulist_free(qgroups);
2493 	ulist_free(old_roots);
2494 	ulist_free(new_roots);
2495 	return ret;
2496 }
2497 
2498 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2499 {
2500 	struct btrfs_fs_info *fs_info = trans->fs_info;
2501 	struct btrfs_qgroup_extent_record *record;
2502 	struct btrfs_delayed_ref_root *delayed_refs;
2503 	struct ulist *new_roots = NULL;
2504 	struct rb_node *node;
2505 	u64 num_dirty_extents = 0;
2506 	u64 qgroup_to_skip;
2507 	int ret = 0;
2508 
2509 	delayed_refs = &trans->transaction->delayed_refs;
2510 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2511 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2512 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2513 				  node);
2514 
2515 		num_dirty_extents++;
2516 		trace_btrfs_qgroup_account_extents(fs_info, record);
2517 
2518 		if (!ret) {
2519 			/*
2520 			 * Old roots should be searched when inserting qgroup
2521 			 * extent record
2522 			 */
2523 			if (WARN_ON(!record->old_roots)) {
2524 				/* Search commit root to find old_roots */
2525 				ret = btrfs_find_all_roots(NULL, fs_info,
2526 						record->bytenr, 0,
2527 						&record->old_roots, false);
2528 				if (ret < 0)
2529 					goto cleanup;
2530 			}
2531 
2532 			/* Free the reserved data space */
2533 			btrfs_qgroup_free_refroot(fs_info,
2534 					record->data_rsv_refroot,
2535 					record->data_rsv,
2536 					BTRFS_QGROUP_RSV_DATA);
2537 			/*
2538 			 * Use SEQ_LAST as time_seq to do special search, which
2539 			 * doesn't lock tree or delayed_refs and search current
2540 			 * root. It's safe inside commit_transaction().
2541 			 */
2542 			ret = btrfs_find_all_roots(trans, fs_info,
2543 				record->bytenr, SEQ_LAST, &new_roots, false);
2544 			if (ret < 0)
2545 				goto cleanup;
2546 			if (qgroup_to_skip) {
2547 				ulist_del(new_roots, qgroup_to_skip, 0);
2548 				ulist_del(record->old_roots, qgroup_to_skip,
2549 					  0);
2550 			}
2551 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2552 							  record->num_bytes,
2553 							  record->old_roots,
2554 							  new_roots);
2555 			record->old_roots = NULL;
2556 			new_roots = NULL;
2557 		}
2558 cleanup:
2559 		ulist_free(record->old_roots);
2560 		ulist_free(new_roots);
2561 		new_roots = NULL;
2562 		rb_erase(node, &delayed_refs->dirty_extent_root);
2563 		kfree(record);
2564 
2565 	}
2566 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2567 				       num_dirty_extents);
2568 	return ret;
2569 }
2570 
2571 /*
2572  * called from commit_transaction. Writes all changed qgroups to disk.
2573  */
2574 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2575 {
2576 	struct btrfs_fs_info *fs_info = trans->fs_info;
2577 	int ret = 0;
2578 
2579 	if (!fs_info->quota_root)
2580 		return ret;
2581 
2582 	spin_lock(&fs_info->qgroup_lock);
2583 	while (!list_empty(&fs_info->dirty_qgroups)) {
2584 		struct btrfs_qgroup *qgroup;
2585 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2586 					  struct btrfs_qgroup, dirty);
2587 		list_del_init(&qgroup->dirty);
2588 		spin_unlock(&fs_info->qgroup_lock);
2589 		ret = update_qgroup_info_item(trans, qgroup);
2590 		if (ret)
2591 			fs_info->qgroup_flags |=
2592 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2593 		ret = update_qgroup_limit_item(trans, qgroup);
2594 		if (ret)
2595 			fs_info->qgroup_flags |=
2596 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2597 		spin_lock(&fs_info->qgroup_lock);
2598 	}
2599 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2600 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2601 	else
2602 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2603 	spin_unlock(&fs_info->qgroup_lock);
2604 
2605 	ret = update_qgroup_status_item(trans);
2606 	if (ret)
2607 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2608 
2609 	return ret;
2610 }
2611 
2612 /*
2613  * Copy the accounting information between qgroups. This is necessary
2614  * when a snapshot or a subvolume is created. Throwing an error will
2615  * cause a transaction abort so we take extra care here to only error
2616  * when a readonly fs is a reasonable outcome.
2617  */
2618 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2619 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2620 {
2621 	int ret = 0;
2622 	int i;
2623 	u64 *i_qgroups;
2624 	bool committing = false;
2625 	struct btrfs_fs_info *fs_info = trans->fs_info;
2626 	struct btrfs_root *quota_root;
2627 	struct btrfs_qgroup *srcgroup;
2628 	struct btrfs_qgroup *dstgroup;
2629 	u32 level_size = 0;
2630 	u64 nums;
2631 
2632 	/*
2633 	 * There are only two callers of this function.
2634 	 *
2635 	 * One in create_subvol() in the ioctl context, which needs to hold
2636 	 * the qgroup_ioctl_lock.
2637 	 *
2638 	 * The other one in create_pending_snapshot() where no other qgroup
2639 	 * code can modify the fs as they all need to either start a new trans
2640 	 * or hold a trans handler, thus we don't need to hold
2641 	 * qgroup_ioctl_lock.
2642 	 * This would avoid long and complex lock chain and make lockdep happy.
2643 	 */
2644 	spin_lock(&fs_info->trans_lock);
2645 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2646 		committing = true;
2647 	spin_unlock(&fs_info->trans_lock);
2648 
2649 	if (!committing)
2650 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2651 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2652 		goto out;
2653 
2654 	quota_root = fs_info->quota_root;
2655 	if (!quota_root) {
2656 		ret = -EINVAL;
2657 		goto out;
2658 	}
2659 
2660 	if (inherit) {
2661 		i_qgroups = (u64 *)(inherit + 1);
2662 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2663 		       2 * inherit->num_excl_copies;
2664 		for (i = 0; i < nums; ++i) {
2665 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2666 
2667 			/*
2668 			 * Zero out invalid groups so we can ignore
2669 			 * them later.
2670 			 */
2671 			if (!srcgroup ||
2672 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2673 				*i_qgroups = 0ULL;
2674 
2675 			++i_qgroups;
2676 		}
2677 	}
2678 
2679 	/*
2680 	 * create a tracking group for the subvol itself
2681 	 */
2682 	ret = add_qgroup_item(trans, quota_root, objectid);
2683 	if (ret)
2684 		goto out;
2685 
2686 	/*
2687 	 * add qgroup to all inherited groups
2688 	 */
2689 	if (inherit) {
2690 		i_qgroups = (u64 *)(inherit + 1);
2691 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2692 			if (*i_qgroups == 0)
2693 				continue;
2694 			ret = add_qgroup_relation_item(trans, objectid,
2695 						       *i_qgroups);
2696 			if (ret && ret != -EEXIST)
2697 				goto out;
2698 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2699 						       objectid);
2700 			if (ret && ret != -EEXIST)
2701 				goto out;
2702 		}
2703 		ret = 0;
2704 	}
2705 
2706 
2707 	spin_lock(&fs_info->qgroup_lock);
2708 
2709 	dstgroup = add_qgroup_rb(fs_info, objectid);
2710 	if (IS_ERR(dstgroup)) {
2711 		ret = PTR_ERR(dstgroup);
2712 		goto unlock;
2713 	}
2714 
2715 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2716 		dstgroup->lim_flags = inherit->lim.flags;
2717 		dstgroup->max_rfer = inherit->lim.max_rfer;
2718 		dstgroup->max_excl = inherit->lim.max_excl;
2719 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2720 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2721 
2722 		ret = update_qgroup_limit_item(trans, dstgroup);
2723 		if (ret) {
2724 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2725 			btrfs_info(fs_info,
2726 				   "unable to update quota limit for %llu",
2727 				   dstgroup->qgroupid);
2728 			goto unlock;
2729 		}
2730 	}
2731 
2732 	if (srcid) {
2733 		srcgroup = find_qgroup_rb(fs_info, srcid);
2734 		if (!srcgroup)
2735 			goto unlock;
2736 
2737 		/*
2738 		 * We call inherit after we clone the root in order to make sure
2739 		 * our counts don't go crazy, so at this point the only
2740 		 * difference between the two roots should be the root node.
2741 		 */
2742 		level_size = fs_info->nodesize;
2743 		dstgroup->rfer = srcgroup->rfer;
2744 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2745 		dstgroup->excl = level_size;
2746 		dstgroup->excl_cmpr = level_size;
2747 		srcgroup->excl = level_size;
2748 		srcgroup->excl_cmpr = level_size;
2749 
2750 		/* inherit the limit info */
2751 		dstgroup->lim_flags = srcgroup->lim_flags;
2752 		dstgroup->max_rfer = srcgroup->max_rfer;
2753 		dstgroup->max_excl = srcgroup->max_excl;
2754 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2755 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2756 
2757 		qgroup_dirty(fs_info, dstgroup);
2758 		qgroup_dirty(fs_info, srcgroup);
2759 	}
2760 
2761 	if (!inherit)
2762 		goto unlock;
2763 
2764 	i_qgroups = (u64 *)(inherit + 1);
2765 	for (i = 0; i < inherit->num_qgroups; ++i) {
2766 		if (*i_qgroups) {
2767 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2768 			if (ret)
2769 				goto unlock;
2770 		}
2771 		++i_qgroups;
2772 	}
2773 
2774 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2775 		struct btrfs_qgroup *src;
2776 		struct btrfs_qgroup *dst;
2777 
2778 		if (!i_qgroups[0] || !i_qgroups[1])
2779 			continue;
2780 
2781 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2782 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2783 
2784 		if (!src || !dst) {
2785 			ret = -EINVAL;
2786 			goto unlock;
2787 		}
2788 
2789 		dst->rfer = src->rfer - level_size;
2790 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2791 	}
2792 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2793 		struct btrfs_qgroup *src;
2794 		struct btrfs_qgroup *dst;
2795 
2796 		if (!i_qgroups[0] || !i_qgroups[1])
2797 			continue;
2798 
2799 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2800 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2801 
2802 		if (!src || !dst) {
2803 			ret = -EINVAL;
2804 			goto unlock;
2805 		}
2806 
2807 		dst->excl = src->excl + level_size;
2808 		dst->excl_cmpr = src->excl_cmpr + level_size;
2809 	}
2810 
2811 unlock:
2812 	spin_unlock(&fs_info->qgroup_lock);
2813 out:
2814 	if (!committing)
2815 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
2816 	return ret;
2817 }
2818 
2819 /*
2820  * Two limits to commit transaction in advance.
2821  *
2822  * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2823  * For SIZE, it will be in byte unit as threshold.
2824  */
2825 #define QGROUP_FREE_RATIO		32
2826 #define QGROUP_FREE_SIZE		SZ_32M
2827 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2828 				const struct btrfs_qgroup *qg, u64 num_bytes)
2829 {
2830 	u64 free;
2831 	u64 threshold;
2832 
2833 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2834 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2835 		return false;
2836 
2837 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2838 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2839 		return false;
2840 
2841 	/*
2842 	 * Even if we passed the check, it's better to check if reservation
2843 	 * for meta_pertrans is pushing us near limit.
2844 	 * If there is too much pertrans reservation or it's near the limit,
2845 	 * let's try commit transaction to free some, using transaction_kthread
2846 	 */
2847 	if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2848 			      BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2849 		if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2850 			free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2851 			threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2852 					  QGROUP_FREE_SIZE);
2853 		} else {
2854 			free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2855 			threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2856 					  QGROUP_FREE_SIZE);
2857 		}
2858 
2859 		/*
2860 		 * Use transaction_kthread to commit transaction, so we no
2861 		 * longer need to bother nested transaction nor lock context.
2862 		 */
2863 		if (free < threshold)
2864 			btrfs_commit_transaction_locksafe(fs_info);
2865 	}
2866 
2867 	return true;
2868 }
2869 
2870 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2871 			  enum btrfs_qgroup_rsv_type type)
2872 {
2873 	struct btrfs_qgroup *qgroup;
2874 	struct btrfs_fs_info *fs_info = root->fs_info;
2875 	u64 ref_root = root->root_key.objectid;
2876 	int ret = 0;
2877 	struct ulist_node *unode;
2878 	struct ulist_iterator uiter;
2879 
2880 	if (!is_fstree(ref_root))
2881 		return 0;
2882 
2883 	if (num_bytes == 0)
2884 		return 0;
2885 
2886 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2887 	    capable(CAP_SYS_RESOURCE))
2888 		enforce = false;
2889 
2890 	spin_lock(&fs_info->qgroup_lock);
2891 	if (!fs_info->quota_root)
2892 		goto out;
2893 
2894 	qgroup = find_qgroup_rb(fs_info, ref_root);
2895 	if (!qgroup)
2896 		goto out;
2897 
2898 	/*
2899 	 * in a first step, we check all affected qgroups if any limits would
2900 	 * be exceeded
2901 	 */
2902 	ulist_reinit(fs_info->qgroup_ulist);
2903 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2904 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2905 	if (ret < 0)
2906 		goto out;
2907 	ULIST_ITER_INIT(&uiter);
2908 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2909 		struct btrfs_qgroup *qg;
2910 		struct btrfs_qgroup_list *glist;
2911 
2912 		qg = unode_aux_to_qgroup(unode);
2913 
2914 		if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2915 			ret = -EDQUOT;
2916 			goto out;
2917 		}
2918 
2919 		list_for_each_entry(glist, &qg->groups, next_group) {
2920 			ret = ulist_add(fs_info->qgroup_ulist,
2921 					glist->group->qgroupid,
2922 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2923 			if (ret < 0)
2924 				goto out;
2925 		}
2926 	}
2927 	ret = 0;
2928 	/*
2929 	 * no limits exceeded, now record the reservation into all qgroups
2930 	 */
2931 	ULIST_ITER_INIT(&uiter);
2932 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2933 		struct btrfs_qgroup *qg;
2934 
2935 		qg = unode_aux_to_qgroup(unode);
2936 
2937 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
2938 	}
2939 
2940 out:
2941 	spin_unlock(&fs_info->qgroup_lock);
2942 	return ret;
2943 }
2944 
2945 /*
2946  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2947  * qgroup).
2948  *
2949  * Will handle all higher level qgroup too.
2950  *
2951  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2952  * This special case is only used for META_PERTRANS type.
2953  */
2954 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2955 			       u64 ref_root, u64 num_bytes,
2956 			       enum btrfs_qgroup_rsv_type type)
2957 {
2958 	struct btrfs_qgroup *qgroup;
2959 	struct ulist_node *unode;
2960 	struct ulist_iterator uiter;
2961 	int ret = 0;
2962 
2963 	if (!is_fstree(ref_root))
2964 		return;
2965 
2966 	if (num_bytes == 0)
2967 		return;
2968 
2969 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2970 		WARN(1, "%s: Invalid type to free", __func__);
2971 		return;
2972 	}
2973 	spin_lock(&fs_info->qgroup_lock);
2974 
2975 	if (!fs_info->quota_root)
2976 		goto out;
2977 
2978 	qgroup = find_qgroup_rb(fs_info, ref_root);
2979 	if (!qgroup)
2980 		goto out;
2981 
2982 	if (num_bytes == (u64)-1)
2983 		/*
2984 		 * We're freeing all pertrans rsv, get reserved value from
2985 		 * level 0 qgroup as real num_bytes to free.
2986 		 */
2987 		num_bytes = qgroup->rsv.values[type];
2988 
2989 	ulist_reinit(fs_info->qgroup_ulist);
2990 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2991 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2992 	if (ret < 0)
2993 		goto out;
2994 	ULIST_ITER_INIT(&uiter);
2995 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2996 		struct btrfs_qgroup *qg;
2997 		struct btrfs_qgroup_list *glist;
2998 
2999 		qg = unode_aux_to_qgroup(unode);
3000 
3001 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3002 
3003 		list_for_each_entry(glist, &qg->groups, next_group) {
3004 			ret = ulist_add(fs_info->qgroup_ulist,
3005 					glist->group->qgroupid,
3006 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3007 			if (ret < 0)
3008 				goto out;
3009 		}
3010 	}
3011 
3012 out:
3013 	spin_unlock(&fs_info->qgroup_lock);
3014 }
3015 
3016 /*
3017  * Check if the leaf is the last leaf. Which means all node pointers
3018  * are at their last position.
3019  */
3020 static bool is_last_leaf(struct btrfs_path *path)
3021 {
3022 	int i;
3023 
3024 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3025 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3026 			return false;
3027 	}
3028 	return true;
3029 }
3030 
3031 /*
3032  * returns < 0 on error, 0 when more leafs are to be scanned.
3033  * returns 1 when done.
3034  */
3035 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3036 			      struct btrfs_path *path)
3037 {
3038 	struct btrfs_fs_info *fs_info = trans->fs_info;
3039 	struct btrfs_key found;
3040 	struct extent_buffer *scratch_leaf = NULL;
3041 	struct ulist *roots = NULL;
3042 	u64 num_bytes;
3043 	bool done;
3044 	int slot;
3045 	int ret;
3046 
3047 	mutex_lock(&fs_info->qgroup_rescan_lock);
3048 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3049 					 &fs_info->qgroup_rescan_progress,
3050 					 path, 1, 0);
3051 
3052 	btrfs_debug(fs_info,
3053 		"current progress key (%llu %u %llu), search_slot ret %d",
3054 		fs_info->qgroup_rescan_progress.objectid,
3055 		fs_info->qgroup_rescan_progress.type,
3056 		fs_info->qgroup_rescan_progress.offset, ret);
3057 
3058 	if (ret) {
3059 		/*
3060 		 * The rescan is about to end, we will not be scanning any
3061 		 * further blocks. We cannot unset the RESCAN flag here, because
3062 		 * we want to commit the transaction if everything went well.
3063 		 * To make the live accounting work in this phase, we set our
3064 		 * scan progress pointer such that every real extent objectid
3065 		 * will be smaller.
3066 		 */
3067 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3068 		btrfs_release_path(path);
3069 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3070 		return ret;
3071 	}
3072 	done = is_last_leaf(path);
3073 
3074 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3075 			      btrfs_header_nritems(path->nodes[0]) - 1);
3076 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3077 
3078 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3079 	if (!scratch_leaf) {
3080 		ret = -ENOMEM;
3081 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3082 		goto out;
3083 	}
3084 	slot = path->slots[0];
3085 	btrfs_release_path(path);
3086 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3087 
3088 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3089 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3090 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3091 		    found.type != BTRFS_METADATA_ITEM_KEY)
3092 			continue;
3093 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3094 			num_bytes = fs_info->nodesize;
3095 		else
3096 			num_bytes = found.offset;
3097 
3098 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3099 					   &roots, false);
3100 		if (ret < 0)
3101 			goto out;
3102 		/* For rescan, just pass old_roots as NULL */
3103 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3104 						  num_bytes, NULL, roots);
3105 		if (ret < 0)
3106 			goto out;
3107 	}
3108 out:
3109 	if (scratch_leaf)
3110 		free_extent_buffer(scratch_leaf);
3111 
3112 	if (done && !ret) {
3113 		ret = 1;
3114 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3115 	}
3116 	return ret;
3117 }
3118 
3119 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3120 {
3121 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3122 						     qgroup_rescan_work);
3123 	struct btrfs_path *path;
3124 	struct btrfs_trans_handle *trans = NULL;
3125 	int err = -ENOMEM;
3126 	int ret = 0;
3127 
3128 	path = btrfs_alloc_path();
3129 	if (!path)
3130 		goto out;
3131 	/*
3132 	 * Rescan should only search for commit root, and any later difference
3133 	 * should be recorded by qgroup
3134 	 */
3135 	path->search_commit_root = 1;
3136 	path->skip_locking = 1;
3137 
3138 	err = 0;
3139 	while (!err && !btrfs_fs_closing(fs_info)) {
3140 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3141 		if (IS_ERR(trans)) {
3142 			err = PTR_ERR(trans);
3143 			break;
3144 		}
3145 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3146 			err = -EINTR;
3147 		} else {
3148 			err = qgroup_rescan_leaf(trans, path);
3149 		}
3150 		if (err > 0)
3151 			btrfs_commit_transaction(trans);
3152 		else
3153 			btrfs_end_transaction(trans);
3154 	}
3155 
3156 out:
3157 	btrfs_free_path(path);
3158 
3159 	mutex_lock(&fs_info->qgroup_rescan_lock);
3160 	if (err > 0 &&
3161 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3162 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3163 	} else if (err < 0) {
3164 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3165 	}
3166 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3167 
3168 	/*
3169 	 * only update status, since the previous part has already updated the
3170 	 * qgroup info.
3171 	 */
3172 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3173 	if (IS_ERR(trans)) {
3174 		err = PTR_ERR(trans);
3175 		trans = NULL;
3176 		btrfs_err(fs_info,
3177 			  "fail to start transaction for status update: %d",
3178 			  err);
3179 	}
3180 
3181 	mutex_lock(&fs_info->qgroup_rescan_lock);
3182 	if (!btrfs_fs_closing(fs_info))
3183 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3184 	if (trans) {
3185 		ret = update_qgroup_status_item(trans);
3186 		if (ret < 0) {
3187 			err = ret;
3188 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3189 				  err);
3190 		}
3191 	}
3192 	fs_info->qgroup_rescan_running = false;
3193 	complete_all(&fs_info->qgroup_rescan_completion);
3194 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3195 
3196 	if (!trans)
3197 		return;
3198 
3199 	btrfs_end_transaction(trans);
3200 
3201 	if (btrfs_fs_closing(fs_info)) {
3202 		btrfs_info(fs_info, "qgroup scan paused");
3203 	} else if (err >= 0) {
3204 		btrfs_info(fs_info, "qgroup scan completed%s",
3205 			err > 0 ? " (inconsistency flag cleared)" : "");
3206 	} else {
3207 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3208 	}
3209 }
3210 
3211 /*
3212  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3213  * memory required for the rescan context.
3214  */
3215 static int
3216 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3217 		   int init_flags)
3218 {
3219 	int ret = 0;
3220 
3221 	if (!init_flags) {
3222 		/* we're resuming qgroup rescan at mount time */
3223 		if (!(fs_info->qgroup_flags &
3224 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3225 			btrfs_warn(fs_info,
3226 			"qgroup rescan init failed, qgroup rescan is not queued");
3227 			ret = -EINVAL;
3228 		} else if (!(fs_info->qgroup_flags &
3229 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3230 			btrfs_warn(fs_info,
3231 			"qgroup rescan init failed, qgroup is not enabled");
3232 			ret = -EINVAL;
3233 		}
3234 
3235 		if (ret)
3236 			return ret;
3237 	}
3238 
3239 	mutex_lock(&fs_info->qgroup_rescan_lock);
3240 	spin_lock(&fs_info->qgroup_lock);
3241 
3242 	if (init_flags) {
3243 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3244 			btrfs_warn(fs_info,
3245 				   "qgroup rescan is already in progress");
3246 			ret = -EINPROGRESS;
3247 		} else if (!(fs_info->qgroup_flags &
3248 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3249 			btrfs_warn(fs_info,
3250 			"qgroup rescan init failed, qgroup is not enabled");
3251 			ret = -EINVAL;
3252 		}
3253 
3254 		if (ret) {
3255 			spin_unlock(&fs_info->qgroup_lock);
3256 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3257 			return ret;
3258 		}
3259 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3260 	}
3261 
3262 	memset(&fs_info->qgroup_rescan_progress, 0,
3263 		sizeof(fs_info->qgroup_rescan_progress));
3264 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3265 	init_completion(&fs_info->qgroup_rescan_completion);
3266 	fs_info->qgroup_rescan_running = true;
3267 
3268 	spin_unlock(&fs_info->qgroup_lock);
3269 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3270 
3271 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3272 			btrfs_qgroup_rescan_worker, NULL, NULL);
3273 	return 0;
3274 }
3275 
3276 static void
3277 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3278 {
3279 	struct rb_node *n;
3280 	struct btrfs_qgroup *qgroup;
3281 
3282 	spin_lock(&fs_info->qgroup_lock);
3283 	/* clear all current qgroup tracking information */
3284 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3285 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3286 		qgroup->rfer = 0;
3287 		qgroup->rfer_cmpr = 0;
3288 		qgroup->excl = 0;
3289 		qgroup->excl_cmpr = 0;
3290 		qgroup_dirty(fs_info, qgroup);
3291 	}
3292 	spin_unlock(&fs_info->qgroup_lock);
3293 }
3294 
3295 int
3296 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3297 {
3298 	int ret = 0;
3299 	struct btrfs_trans_handle *trans;
3300 
3301 	ret = qgroup_rescan_init(fs_info, 0, 1);
3302 	if (ret)
3303 		return ret;
3304 
3305 	/*
3306 	 * We have set the rescan_progress to 0, which means no more
3307 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3308 	 * However, btrfs_qgroup_account_ref may be right after its call
3309 	 * to btrfs_find_all_roots, in which case it would still do the
3310 	 * accounting.
3311 	 * To solve this, we're committing the transaction, which will
3312 	 * ensure we run all delayed refs and only after that, we are
3313 	 * going to clear all tracking information for a clean start.
3314 	 */
3315 
3316 	trans = btrfs_join_transaction(fs_info->fs_root);
3317 	if (IS_ERR(trans)) {
3318 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3319 		return PTR_ERR(trans);
3320 	}
3321 	ret = btrfs_commit_transaction(trans);
3322 	if (ret) {
3323 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3324 		return ret;
3325 	}
3326 
3327 	qgroup_rescan_zero_tracking(fs_info);
3328 
3329 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3330 			 &fs_info->qgroup_rescan_work);
3331 
3332 	return 0;
3333 }
3334 
3335 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3336 				     bool interruptible)
3337 {
3338 	int running;
3339 	int ret = 0;
3340 
3341 	mutex_lock(&fs_info->qgroup_rescan_lock);
3342 	spin_lock(&fs_info->qgroup_lock);
3343 	running = fs_info->qgroup_rescan_running;
3344 	spin_unlock(&fs_info->qgroup_lock);
3345 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3346 
3347 	if (!running)
3348 		return 0;
3349 
3350 	if (interruptible)
3351 		ret = wait_for_completion_interruptible(
3352 					&fs_info->qgroup_rescan_completion);
3353 	else
3354 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3355 
3356 	return ret;
3357 }
3358 
3359 /*
3360  * this is only called from open_ctree where we're still single threaded, thus
3361  * locking is omitted here.
3362  */
3363 void
3364 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3365 {
3366 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3367 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3368 				 &fs_info->qgroup_rescan_work);
3369 }
3370 
3371 /*
3372  * Reserve qgroup space for range [start, start + len).
3373  *
3374  * This function will either reserve space from related qgroups or doing
3375  * nothing if the range is already reserved.
3376  *
3377  * Return 0 for successful reserve
3378  * Return <0 for error (including -EQUOT)
3379  *
3380  * NOTE: this function may sleep for memory allocation.
3381  *       if btrfs_qgroup_reserve_data() is called multiple times with
3382  *       same @reserved, caller must ensure when error happens it's OK
3383  *       to free *ALL* reserved space.
3384  */
3385 int btrfs_qgroup_reserve_data(struct inode *inode,
3386 			struct extent_changeset **reserved_ret, u64 start,
3387 			u64 len)
3388 {
3389 	struct btrfs_root *root = BTRFS_I(inode)->root;
3390 	struct ulist_node *unode;
3391 	struct ulist_iterator uiter;
3392 	struct extent_changeset *reserved;
3393 	u64 orig_reserved;
3394 	u64 to_reserve;
3395 	int ret;
3396 
3397 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3398 	    !is_fstree(root->root_key.objectid) || len == 0)
3399 		return 0;
3400 
3401 	/* @reserved parameter is mandatory for qgroup */
3402 	if (WARN_ON(!reserved_ret))
3403 		return -EINVAL;
3404 	if (!*reserved_ret) {
3405 		*reserved_ret = extent_changeset_alloc();
3406 		if (!*reserved_ret)
3407 			return -ENOMEM;
3408 	}
3409 	reserved = *reserved_ret;
3410 	/* Record already reserved space */
3411 	orig_reserved = reserved->bytes_changed;
3412 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3413 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3414 
3415 	/* Newly reserved space */
3416 	to_reserve = reserved->bytes_changed - orig_reserved;
3417 	trace_btrfs_qgroup_reserve_data(inode, start, len,
3418 					to_reserve, QGROUP_RESERVE);
3419 	if (ret < 0)
3420 		goto cleanup;
3421 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3422 	if (ret < 0)
3423 		goto cleanup;
3424 
3425 	return ret;
3426 
3427 cleanup:
3428 	/* cleanup *ALL* already reserved ranges */
3429 	ULIST_ITER_INIT(&uiter);
3430 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3431 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3432 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3433 	/* Also free data bytes of already reserved one */
3434 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3435 				  orig_reserved, BTRFS_QGROUP_RSV_DATA);
3436 	extent_changeset_release(reserved);
3437 	return ret;
3438 }
3439 
3440 /* Free ranges specified by @reserved, normally in error path */
3441 static int qgroup_free_reserved_data(struct inode *inode,
3442 			struct extent_changeset *reserved, u64 start, u64 len)
3443 {
3444 	struct btrfs_root *root = BTRFS_I(inode)->root;
3445 	struct ulist_node *unode;
3446 	struct ulist_iterator uiter;
3447 	struct extent_changeset changeset;
3448 	int freed = 0;
3449 	int ret;
3450 
3451 	extent_changeset_init(&changeset);
3452 	len = round_up(start + len, root->fs_info->sectorsize);
3453 	start = round_down(start, root->fs_info->sectorsize);
3454 
3455 	ULIST_ITER_INIT(&uiter);
3456 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3457 		u64 range_start = unode->val;
3458 		/* unode->aux is the inclusive end */
3459 		u64 range_len = unode->aux - range_start + 1;
3460 		u64 free_start;
3461 		u64 free_len;
3462 
3463 		extent_changeset_release(&changeset);
3464 
3465 		/* Only free range in range [start, start + len) */
3466 		if (range_start >= start + len ||
3467 		    range_start + range_len <= start)
3468 			continue;
3469 		free_start = max(range_start, start);
3470 		free_len = min(start + len, range_start + range_len) -
3471 			   free_start;
3472 		/*
3473 		 * TODO: To also modify reserved->ranges_reserved to reflect
3474 		 * the modification.
3475 		 *
3476 		 * However as long as we free qgroup reserved according to
3477 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3478 		 * So not need to rush.
3479 		 */
3480 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree,
3481 				free_start, free_start + free_len - 1,
3482 				EXTENT_QGROUP_RESERVED, &changeset);
3483 		if (ret < 0)
3484 			goto out;
3485 		freed += changeset.bytes_changed;
3486 	}
3487 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3488 				  BTRFS_QGROUP_RSV_DATA);
3489 	ret = freed;
3490 out:
3491 	extent_changeset_release(&changeset);
3492 	return ret;
3493 }
3494 
3495 static int __btrfs_qgroup_release_data(struct inode *inode,
3496 			struct extent_changeset *reserved, u64 start, u64 len,
3497 			int free)
3498 {
3499 	struct extent_changeset changeset;
3500 	int trace_op = QGROUP_RELEASE;
3501 	int ret;
3502 
3503 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3504 		      &BTRFS_I(inode)->root->fs_info->flags))
3505 		return 0;
3506 
3507 	/* In release case, we shouldn't have @reserved */
3508 	WARN_ON(!free && reserved);
3509 	if (free && reserved)
3510 		return qgroup_free_reserved_data(inode, reserved, start, len);
3511 	extent_changeset_init(&changeset);
3512 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3513 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3514 	if (ret < 0)
3515 		goto out;
3516 
3517 	if (free)
3518 		trace_op = QGROUP_FREE;
3519 	trace_btrfs_qgroup_release_data(inode, start, len,
3520 					changeset.bytes_changed, trace_op);
3521 	if (free)
3522 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3523 				BTRFS_I(inode)->root->root_key.objectid,
3524 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3525 	ret = changeset.bytes_changed;
3526 out:
3527 	extent_changeset_release(&changeset);
3528 	return ret;
3529 }
3530 
3531 /*
3532  * Free a reserved space range from io_tree and related qgroups
3533  *
3534  * Should be called when a range of pages get invalidated before reaching disk.
3535  * Or for error cleanup case.
3536  * if @reserved is given, only reserved range in [@start, @start + @len) will
3537  * be freed.
3538  *
3539  * For data written to disk, use btrfs_qgroup_release_data().
3540  *
3541  * NOTE: This function may sleep for memory allocation.
3542  */
3543 int btrfs_qgroup_free_data(struct inode *inode,
3544 			struct extent_changeset *reserved, u64 start, u64 len)
3545 {
3546 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3547 }
3548 
3549 /*
3550  * Release a reserved space range from io_tree only.
3551  *
3552  * Should be called when a range of pages get written to disk and corresponding
3553  * FILE_EXTENT is inserted into corresponding root.
3554  *
3555  * Since new qgroup accounting framework will only update qgroup numbers at
3556  * commit_transaction() time, its reserved space shouldn't be freed from
3557  * related qgroups.
3558  *
3559  * But we should release the range from io_tree, to allow further write to be
3560  * COWed.
3561  *
3562  * NOTE: This function may sleep for memory allocation.
3563  */
3564 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3565 {
3566 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3567 }
3568 
3569 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3570 			      enum btrfs_qgroup_rsv_type type)
3571 {
3572 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3573 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3574 		return;
3575 	if (num_bytes == 0)
3576 		return;
3577 
3578 	spin_lock(&root->qgroup_meta_rsv_lock);
3579 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3580 		root->qgroup_meta_rsv_prealloc += num_bytes;
3581 	else
3582 		root->qgroup_meta_rsv_pertrans += num_bytes;
3583 	spin_unlock(&root->qgroup_meta_rsv_lock);
3584 }
3585 
3586 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3587 			     enum btrfs_qgroup_rsv_type type)
3588 {
3589 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3590 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3591 		return 0;
3592 	if (num_bytes == 0)
3593 		return 0;
3594 
3595 	spin_lock(&root->qgroup_meta_rsv_lock);
3596 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3597 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3598 				  num_bytes);
3599 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3600 	} else {
3601 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3602 				  num_bytes);
3603 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3604 	}
3605 	spin_unlock(&root->qgroup_meta_rsv_lock);
3606 	return num_bytes;
3607 }
3608 
3609 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3610 				enum btrfs_qgroup_rsv_type type, bool enforce)
3611 {
3612 	struct btrfs_fs_info *fs_info = root->fs_info;
3613 	int ret;
3614 
3615 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3616 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3617 		return 0;
3618 
3619 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3620 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3621 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3622 	if (ret < 0)
3623 		return ret;
3624 	/*
3625 	 * Record what we have reserved into root.
3626 	 *
3627 	 * To avoid quota disabled->enabled underflow.
3628 	 * In that case, we may try to free space we haven't reserved
3629 	 * (since quota was disabled), so record what we reserved into root.
3630 	 * And ensure later release won't underflow this number.
3631 	 */
3632 	add_root_meta_rsv(root, num_bytes, type);
3633 	return ret;
3634 }
3635 
3636 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3637 {
3638 	struct btrfs_fs_info *fs_info = root->fs_info;
3639 
3640 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3641 	    !is_fstree(root->root_key.objectid))
3642 		return;
3643 
3644 	/* TODO: Update trace point to handle such free */
3645 	trace_qgroup_meta_free_all_pertrans(root);
3646 	/* Special value -1 means to free all reserved space */
3647 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3648 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3649 }
3650 
3651 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3652 			      enum btrfs_qgroup_rsv_type type)
3653 {
3654 	struct btrfs_fs_info *fs_info = root->fs_info;
3655 
3656 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3657 	    !is_fstree(root->root_key.objectid))
3658 		return;
3659 
3660 	/*
3661 	 * reservation for META_PREALLOC can happen before quota is enabled,
3662 	 * which can lead to underflow.
3663 	 * Here ensure we will only free what we really have reserved.
3664 	 */
3665 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3666 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3667 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3668 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3669 				  num_bytes, type);
3670 }
3671 
3672 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3673 				int num_bytes)
3674 {
3675 	struct btrfs_qgroup *qgroup;
3676 	struct ulist_node *unode;
3677 	struct ulist_iterator uiter;
3678 	int ret = 0;
3679 
3680 	if (num_bytes == 0)
3681 		return;
3682 	if (!fs_info->quota_root)
3683 		return;
3684 
3685 	spin_lock(&fs_info->qgroup_lock);
3686 	qgroup = find_qgroup_rb(fs_info, ref_root);
3687 	if (!qgroup)
3688 		goto out;
3689 	ulist_reinit(fs_info->qgroup_ulist);
3690 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3691 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3692 	if (ret < 0)
3693 		goto out;
3694 	ULIST_ITER_INIT(&uiter);
3695 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3696 		struct btrfs_qgroup *qg;
3697 		struct btrfs_qgroup_list *glist;
3698 
3699 		qg = unode_aux_to_qgroup(unode);
3700 
3701 		qgroup_rsv_release(fs_info, qg, num_bytes,
3702 				BTRFS_QGROUP_RSV_META_PREALLOC);
3703 		qgroup_rsv_add(fs_info, qg, num_bytes,
3704 				BTRFS_QGROUP_RSV_META_PERTRANS);
3705 		list_for_each_entry(glist, &qg->groups, next_group) {
3706 			ret = ulist_add(fs_info->qgroup_ulist,
3707 					glist->group->qgroupid,
3708 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3709 			if (ret < 0)
3710 				goto out;
3711 		}
3712 	}
3713 out:
3714 	spin_unlock(&fs_info->qgroup_lock);
3715 }
3716 
3717 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3718 {
3719 	struct btrfs_fs_info *fs_info = root->fs_info;
3720 
3721 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3722 	    !is_fstree(root->root_key.objectid))
3723 		return;
3724 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3725 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3726 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3727 	trace_qgroup_meta_convert(root, num_bytes);
3728 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3729 }
3730 
3731 /*
3732  * Check qgroup reserved space leaking, normally at destroy inode
3733  * time
3734  */
3735 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3736 {
3737 	struct extent_changeset changeset;
3738 	struct ulist_node *unode;
3739 	struct ulist_iterator iter;
3740 	int ret;
3741 
3742 	extent_changeset_init(&changeset);
3743 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3744 			EXTENT_QGROUP_RESERVED, &changeset);
3745 
3746 	WARN_ON(ret < 0);
3747 	if (WARN_ON(changeset.bytes_changed)) {
3748 		ULIST_ITER_INIT(&iter);
3749 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3750 			btrfs_warn(BTRFS_I(inode)->root->fs_info,
3751 				"leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3752 				inode->i_ino, unode->val, unode->aux);
3753 		}
3754 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3755 				BTRFS_I(inode)->root->root_key.objectid,
3756 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3757 
3758 	}
3759 	extent_changeset_release(&changeset);
3760 }
3761 
3762 void btrfs_qgroup_init_swapped_blocks(
3763 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3764 {
3765 	int i;
3766 
3767 	spin_lock_init(&swapped_blocks->lock);
3768 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3769 		swapped_blocks->blocks[i] = RB_ROOT;
3770 	swapped_blocks->swapped = false;
3771 }
3772 
3773 /*
3774  * Delete all swapped blocks record of @root.
3775  * Every record here means we skipped a full subtree scan for qgroup.
3776  *
3777  * Gets called when committing one transaction.
3778  */
3779 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3780 {
3781 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3782 	int i;
3783 
3784 	swapped_blocks = &root->swapped_blocks;
3785 
3786 	spin_lock(&swapped_blocks->lock);
3787 	if (!swapped_blocks->swapped)
3788 		goto out;
3789 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3790 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
3791 		struct btrfs_qgroup_swapped_block *entry;
3792 		struct btrfs_qgroup_swapped_block *next;
3793 
3794 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3795 						     node)
3796 			kfree(entry);
3797 		swapped_blocks->blocks[i] = RB_ROOT;
3798 	}
3799 	swapped_blocks->swapped = false;
3800 out:
3801 	spin_unlock(&swapped_blocks->lock);
3802 }
3803 
3804 /*
3805  * Add subtree roots record into @subvol_root.
3806  *
3807  * @subvol_root:	tree root of the subvolume tree get swapped
3808  * @bg:			block group under balance
3809  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
3810  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
3811  *			BOTH POINTERS ARE BEFORE TREE SWAP
3812  * @last_snapshot:	last snapshot generation of the subvolume tree
3813  */
3814 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3815 		struct btrfs_root *subvol_root,
3816 		struct btrfs_block_group *bg,
3817 		struct extent_buffer *subvol_parent, int subvol_slot,
3818 		struct extent_buffer *reloc_parent, int reloc_slot,
3819 		u64 last_snapshot)
3820 {
3821 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3822 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3823 	struct btrfs_qgroup_swapped_block *block;
3824 	struct rb_node **cur;
3825 	struct rb_node *parent = NULL;
3826 	int level = btrfs_header_level(subvol_parent) - 1;
3827 	int ret = 0;
3828 
3829 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3830 		return 0;
3831 
3832 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3833 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3834 		btrfs_err_rl(fs_info,
3835 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3836 			__func__,
3837 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3838 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3839 		return -EUCLEAN;
3840 	}
3841 
3842 	block = kmalloc(sizeof(*block), GFP_NOFS);
3843 	if (!block) {
3844 		ret = -ENOMEM;
3845 		goto out;
3846 	}
3847 
3848 	/*
3849 	 * @reloc_parent/slot is still before swap, while @block is going to
3850 	 * record the bytenr after swap, so we do the swap here.
3851 	 */
3852 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3853 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3854 							     reloc_slot);
3855 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3856 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3857 							    subvol_slot);
3858 	block->last_snapshot = last_snapshot;
3859 	block->level = level;
3860 
3861 	/*
3862 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3863 	 * no one else can modify tree blocks thus we qgroup will not change
3864 	 * no matter the value of trace_leaf.
3865 	 */
3866 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
3867 		block->trace_leaf = true;
3868 	else
3869 		block->trace_leaf = false;
3870 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3871 
3872 	/* Insert @block into @blocks */
3873 	spin_lock(&blocks->lock);
3874 	cur = &blocks->blocks[level].rb_node;
3875 	while (*cur) {
3876 		struct btrfs_qgroup_swapped_block *entry;
3877 
3878 		parent = *cur;
3879 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3880 				 node);
3881 
3882 		if (entry->subvol_bytenr < block->subvol_bytenr) {
3883 			cur = &(*cur)->rb_left;
3884 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
3885 			cur = &(*cur)->rb_right;
3886 		} else {
3887 			if (entry->subvol_generation !=
3888 					block->subvol_generation ||
3889 			    entry->reloc_bytenr != block->reloc_bytenr ||
3890 			    entry->reloc_generation !=
3891 					block->reloc_generation) {
3892 				/*
3893 				 * Duplicated but mismatch entry found.
3894 				 * Shouldn't happen.
3895 				 *
3896 				 * Marking qgroup inconsistent should be enough
3897 				 * for end users.
3898 				 */
3899 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3900 				ret = -EEXIST;
3901 			}
3902 			kfree(block);
3903 			goto out_unlock;
3904 		}
3905 	}
3906 	rb_link_node(&block->node, parent, cur);
3907 	rb_insert_color(&block->node, &blocks->blocks[level]);
3908 	blocks->swapped = true;
3909 out_unlock:
3910 	spin_unlock(&blocks->lock);
3911 out:
3912 	if (ret < 0)
3913 		fs_info->qgroup_flags |=
3914 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3915 	return ret;
3916 }
3917 
3918 /*
3919  * Check if the tree block is a subtree root, and if so do the needed
3920  * delayed subtree trace for qgroup.
3921  *
3922  * This is called during btrfs_cow_block().
3923  */
3924 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3925 					 struct btrfs_root *root,
3926 					 struct extent_buffer *subvol_eb)
3927 {
3928 	struct btrfs_fs_info *fs_info = root->fs_info;
3929 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3930 	struct btrfs_qgroup_swapped_block *block;
3931 	struct extent_buffer *reloc_eb = NULL;
3932 	struct rb_node *node;
3933 	bool found = false;
3934 	bool swapped = false;
3935 	int level = btrfs_header_level(subvol_eb);
3936 	int ret = 0;
3937 	int i;
3938 
3939 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3940 		return 0;
3941 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3942 		return 0;
3943 
3944 	spin_lock(&blocks->lock);
3945 	if (!blocks->swapped) {
3946 		spin_unlock(&blocks->lock);
3947 		return 0;
3948 	}
3949 	node = blocks->blocks[level].rb_node;
3950 
3951 	while (node) {
3952 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3953 		if (block->subvol_bytenr < subvol_eb->start) {
3954 			node = node->rb_left;
3955 		} else if (block->subvol_bytenr > subvol_eb->start) {
3956 			node = node->rb_right;
3957 		} else {
3958 			found = true;
3959 			break;
3960 		}
3961 	}
3962 	if (!found) {
3963 		spin_unlock(&blocks->lock);
3964 		goto out;
3965 	}
3966 	/* Found one, remove it from @blocks first and update blocks->swapped */
3967 	rb_erase(&block->node, &blocks->blocks[level]);
3968 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3969 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3970 			swapped = true;
3971 			break;
3972 		}
3973 	}
3974 	blocks->swapped = swapped;
3975 	spin_unlock(&blocks->lock);
3976 
3977 	/* Read out reloc subtree root */
3978 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3979 				   block->reloc_generation, block->level,
3980 				   &block->first_key);
3981 	if (IS_ERR(reloc_eb)) {
3982 		ret = PTR_ERR(reloc_eb);
3983 		reloc_eb = NULL;
3984 		goto free_out;
3985 	}
3986 	if (!extent_buffer_uptodate(reloc_eb)) {
3987 		ret = -EIO;
3988 		goto free_out;
3989 	}
3990 
3991 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3992 			block->last_snapshot, block->trace_leaf);
3993 free_out:
3994 	kfree(block);
3995 	free_extent_buffer(reloc_eb);
3996 out:
3997 	if (ret < 0) {
3998 		btrfs_err_rl(fs_info,
3999 			     "failed to account subtree at bytenr %llu: %d",
4000 			     subvol_eb->start, ret);
4001 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4002 	}
4003 	return ret;
4004 }
4005 
4006 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4007 {
4008 	struct btrfs_qgroup_extent_record *entry;
4009 	struct btrfs_qgroup_extent_record *next;
4010 	struct rb_root *root;
4011 
4012 	root = &trans->delayed_refs.dirty_extent_root;
4013 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4014 		ulist_free(entry->old_roots);
4015 		kfree(entry);
4016 	}
4017 }
4018