xref: /openbmc/linux/fs/btrfs/qgroup.c (revision dd5b2498)
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 
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, fs_info,
922 				       BTRFS_QUOTA_TREE_OBJECTID);
923 	if (IS_ERR(quota_root)) {
924 		ret =  PTR_ERR(quota_root);
925 		btrfs_abort_transaction(trans, ret);
926 		goto out;
927 	}
928 
929 	path = btrfs_alloc_path();
930 	if (!path) {
931 		ret = -ENOMEM;
932 		btrfs_abort_transaction(trans, ret);
933 		goto out_free_root;
934 	}
935 
936 	key.objectid = 0;
937 	key.type = BTRFS_QGROUP_STATUS_KEY;
938 	key.offset = 0;
939 
940 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
941 				      sizeof(*ptr));
942 	if (ret) {
943 		btrfs_abort_transaction(trans, ret);
944 		goto out_free_path;
945 	}
946 
947 	leaf = path->nodes[0];
948 	ptr = btrfs_item_ptr(leaf, path->slots[0],
949 				 struct btrfs_qgroup_status_item);
950 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
951 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
952 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
953 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
954 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
955 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
956 
957 	btrfs_mark_buffer_dirty(leaf);
958 
959 	key.objectid = 0;
960 	key.type = BTRFS_ROOT_REF_KEY;
961 	key.offset = 0;
962 
963 	btrfs_release_path(path);
964 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
965 	if (ret > 0)
966 		goto out_add_root;
967 	if (ret < 0) {
968 		btrfs_abort_transaction(trans, ret);
969 		goto out_free_path;
970 	}
971 
972 	while (1) {
973 		slot = path->slots[0];
974 		leaf = path->nodes[0];
975 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
976 
977 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
978 			ret = add_qgroup_item(trans, quota_root,
979 					      found_key.offset);
980 			if (ret) {
981 				btrfs_abort_transaction(trans, ret);
982 				goto out_free_path;
983 			}
984 
985 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
986 			if (IS_ERR(qgroup)) {
987 				ret = PTR_ERR(qgroup);
988 				btrfs_abort_transaction(trans, ret);
989 				goto out_free_path;
990 			}
991 		}
992 		ret = btrfs_next_item(tree_root, path);
993 		if (ret < 0) {
994 			btrfs_abort_transaction(trans, ret);
995 			goto out_free_path;
996 		}
997 		if (ret)
998 			break;
999 	}
1000 
1001 out_add_root:
1002 	btrfs_release_path(path);
1003 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1004 	if (ret) {
1005 		btrfs_abort_transaction(trans, ret);
1006 		goto out_free_path;
1007 	}
1008 
1009 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1010 	if (IS_ERR(qgroup)) {
1011 		ret = PTR_ERR(qgroup);
1012 		btrfs_abort_transaction(trans, ret);
1013 		goto out_free_path;
1014 	}
1015 
1016 	ret = btrfs_commit_transaction(trans);
1017 	trans = NULL;
1018 	if (ret)
1019 		goto out_free_path;
1020 
1021 	/*
1022 	 * Set quota enabled flag after committing the transaction, to avoid
1023 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1024 	 * creation.
1025 	 */
1026 	spin_lock(&fs_info->qgroup_lock);
1027 	fs_info->quota_root = quota_root;
1028 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1029 	spin_unlock(&fs_info->qgroup_lock);
1030 
1031 	ret = qgroup_rescan_init(fs_info, 0, 1);
1032 	if (!ret) {
1033 	        qgroup_rescan_zero_tracking(fs_info);
1034 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1035 	                         &fs_info->qgroup_rescan_work);
1036 	}
1037 
1038 out_free_path:
1039 	btrfs_free_path(path);
1040 out_free_root:
1041 	if (ret) {
1042 		free_extent_buffer(quota_root->node);
1043 		free_extent_buffer(quota_root->commit_root);
1044 		kfree(quota_root);
1045 	}
1046 out:
1047 	if (ret) {
1048 		ulist_free(fs_info->qgroup_ulist);
1049 		fs_info->qgroup_ulist = NULL;
1050 		if (trans)
1051 			btrfs_end_transaction(trans);
1052 	}
1053 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1054 	return ret;
1055 }
1056 
1057 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1058 {
1059 	struct btrfs_root *quota_root;
1060 	struct btrfs_trans_handle *trans = NULL;
1061 	int ret = 0;
1062 
1063 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1064 	if (!fs_info->quota_root)
1065 		goto out;
1066 
1067 	/*
1068 	 * 1 For the root item
1069 	 *
1070 	 * We should also reserve enough items for the quota tree deletion in
1071 	 * btrfs_clean_quota_tree but this is not done.
1072 	 */
1073 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1074 	if (IS_ERR(trans)) {
1075 		ret = PTR_ERR(trans);
1076 		goto out;
1077 	}
1078 
1079 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1080 	btrfs_qgroup_wait_for_completion(fs_info, false);
1081 	spin_lock(&fs_info->qgroup_lock);
1082 	quota_root = fs_info->quota_root;
1083 	fs_info->quota_root = NULL;
1084 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1085 	spin_unlock(&fs_info->qgroup_lock);
1086 
1087 	btrfs_free_qgroup_config(fs_info);
1088 
1089 	ret = btrfs_clean_quota_tree(trans, quota_root);
1090 	if (ret) {
1091 		btrfs_abort_transaction(trans, ret);
1092 		goto end_trans;
1093 	}
1094 
1095 	ret = btrfs_del_root(trans, &quota_root->root_key);
1096 	if (ret) {
1097 		btrfs_abort_transaction(trans, ret);
1098 		goto end_trans;
1099 	}
1100 
1101 	list_del(&quota_root->dirty_list);
1102 
1103 	btrfs_tree_lock(quota_root->node);
1104 	clean_tree_block(fs_info, quota_root->node);
1105 	btrfs_tree_unlock(quota_root->node);
1106 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1107 
1108 	free_extent_buffer(quota_root->node);
1109 	free_extent_buffer(quota_root->commit_root);
1110 	kfree(quota_root);
1111 
1112 end_trans:
1113 	ret = btrfs_end_transaction(trans);
1114 out:
1115 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1116 	return ret;
1117 }
1118 
1119 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1120 			 struct btrfs_qgroup *qgroup)
1121 {
1122 	if (list_empty(&qgroup->dirty))
1123 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1124 }
1125 
1126 /*
1127  * The easy accounting, we're updating qgroup relationship whose child qgroup
1128  * only has exclusive extents.
1129  *
1130  * In this case, all exclusive extents will also be exclusive for parent, so
1131  * excl/rfer just get added/removed.
1132  *
1133  * So is qgroup reservation space, which should also be added/removed to
1134  * parent.
1135  * Or when child tries to release reservation space, parent will underflow its
1136  * reservation (for relationship adding case).
1137  *
1138  * Caller should hold fs_info->qgroup_lock.
1139  */
1140 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1141 				    struct ulist *tmp, u64 ref_root,
1142 				    struct btrfs_qgroup *src, int sign)
1143 {
1144 	struct btrfs_qgroup *qgroup;
1145 	struct btrfs_qgroup_list *glist;
1146 	struct ulist_node *unode;
1147 	struct ulist_iterator uiter;
1148 	u64 num_bytes = src->excl;
1149 	int ret = 0;
1150 
1151 	qgroup = find_qgroup_rb(fs_info, ref_root);
1152 	if (!qgroup)
1153 		goto out;
1154 
1155 	qgroup->rfer += sign * num_bytes;
1156 	qgroup->rfer_cmpr += sign * num_bytes;
1157 
1158 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1159 	qgroup->excl += sign * num_bytes;
1160 	qgroup->excl_cmpr += sign * num_bytes;
1161 
1162 	if (sign > 0)
1163 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1164 	else
1165 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1166 
1167 	qgroup_dirty(fs_info, qgroup);
1168 
1169 	/* Get all of the parent groups that contain this qgroup */
1170 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1171 		ret = ulist_add(tmp, glist->group->qgroupid,
1172 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1173 		if (ret < 0)
1174 			goto out;
1175 	}
1176 
1177 	/* Iterate all of the parents and adjust their reference counts */
1178 	ULIST_ITER_INIT(&uiter);
1179 	while ((unode = ulist_next(tmp, &uiter))) {
1180 		qgroup = unode_aux_to_qgroup(unode);
1181 		qgroup->rfer += sign * num_bytes;
1182 		qgroup->rfer_cmpr += sign * num_bytes;
1183 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1184 		qgroup->excl += sign * num_bytes;
1185 		if (sign > 0)
1186 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1187 		else
1188 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1189 		qgroup->excl_cmpr += sign * num_bytes;
1190 		qgroup_dirty(fs_info, qgroup);
1191 
1192 		/* Add any parents of the parents */
1193 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1194 			ret = ulist_add(tmp, glist->group->qgroupid,
1195 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1196 			if (ret < 0)
1197 				goto out;
1198 		}
1199 	}
1200 	ret = 0;
1201 out:
1202 	return ret;
1203 }
1204 
1205 
1206 /*
1207  * Quick path for updating qgroup with only excl refs.
1208  *
1209  * In that case, just update all parent will be enough.
1210  * Or we needs to do a full rescan.
1211  * Caller should also hold fs_info->qgroup_lock.
1212  *
1213  * Return 0 for quick update, return >0 for need to full rescan
1214  * and mark INCONSISTENT flag.
1215  * Return < 0 for other error.
1216  */
1217 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1218 				   struct ulist *tmp, u64 src, u64 dst,
1219 				   int sign)
1220 {
1221 	struct btrfs_qgroup *qgroup;
1222 	int ret = 1;
1223 	int err = 0;
1224 
1225 	qgroup = find_qgroup_rb(fs_info, src);
1226 	if (!qgroup)
1227 		goto out;
1228 	if (qgroup->excl == qgroup->rfer) {
1229 		ret = 0;
1230 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1231 					       qgroup, sign);
1232 		if (err < 0) {
1233 			ret = err;
1234 			goto out;
1235 		}
1236 	}
1237 out:
1238 	if (ret)
1239 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1240 	return ret;
1241 }
1242 
1243 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1244 			      u64 dst)
1245 {
1246 	struct btrfs_fs_info *fs_info = trans->fs_info;
1247 	struct btrfs_root *quota_root;
1248 	struct btrfs_qgroup *parent;
1249 	struct btrfs_qgroup *member;
1250 	struct btrfs_qgroup_list *list;
1251 	struct ulist *tmp;
1252 	int ret = 0;
1253 
1254 	/* Check the level of src and dst first */
1255 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1256 		return -EINVAL;
1257 
1258 	tmp = ulist_alloc(GFP_KERNEL);
1259 	if (!tmp)
1260 		return -ENOMEM;
1261 
1262 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1263 	quota_root = fs_info->quota_root;
1264 	if (!quota_root) {
1265 		ret = -EINVAL;
1266 		goto out;
1267 	}
1268 	member = find_qgroup_rb(fs_info, src);
1269 	parent = find_qgroup_rb(fs_info, dst);
1270 	if (!member || !parent) {
1271 		ret = -EINVAL;
1272 		goto out;
1273 	}
1274 
1275 	/* check if such qgroup relation exist firstly */
1276 	list_for_each_entry(list, &member->groups, next_group) {
1277 		if (list->group == parent) {
1278 			ret = -EEXIST;
1279 			goto out;
1280 		}
1281 	}
1282 
1283 	ret = add_qgroup_relation_item(trans, src, dst);
1284 	if (ret)
1285 		goto out;
1286 
1287 	ret = add_qgroup_relation_item(trans, dst, src);
1288 	if (ret) {
1289 		del_qgroup_relation_item(trans, src, dst);
1290 		goto out;
1291 	}
1292 
1293 	spin_lock(&fs_info->qgroup_lock);
1294 	ret = add_relation_rb(fs_info, src, dst);
1295 	if (ret < 0) {
1296 		spin_unlock(&fs_info->qgroup_lock);
1297 		goto out;
1298 	}
1299 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1300 	spin_unlock(&fs_info->qgroup_lock);
1301 out:
1302 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1303 	ulist_free(tmp);
1304 	return ret;
1305 }
1306 
1307 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1308 				 u64 dst)
1309 {
1310 	struct btrfs_fs_info *fs_info = trans->fs_info;
1311 	struct btrfs_root *quota_root;
1312 	struct btrfs_qgroup *parent;
1313 	struct btrfs_qgroup *member;
1314 	struct btrfs_qgroup_list *list;
1315 	struct ulist *tmp;
1316 	int ret = 0;
1317 	int err;
1318 
1319 	tmp = ulist_alloc(GFP_KERNEL);
1320 	if (!tmp)
1321 		return -ENOMEM;
1322 
1323 	quota_root = fs_info->quota_root;
1324 	if (!quota_root) {
1325 		ret = -EINVAL;
1326 		goto out;
1327 	}
1328 
1329 	member = find_qgroup_rb(fs_info, src);
1330 	parent = find_qgroup_rb(fs_info, dst);
1331 	if (!member || !parent) {
1332 		ret = -EINVAL;
1333 		goto out;
1334 	}
1335 
1336 	/* check if such qgroup relation exist firstly */
1337 	list_for_each_entry(list, &member->groups, next_group) {
1338 		if (list->group == parent)
1339 			goto exist;
1340 	}
1341 	ret = -ENOENT;
1342 	goto out;
1343 exist:
1344 	ret = del_qgroup_relation_item(trans, src, dst);
1345 	err = del_qgroup_relation_item(trans, dst, src);
1346 	if (err && !ret)
1347 		ret = err;
1348 
1349 	spin_lock(&fs_info->qgroup_lock);
1350 	del_relation_rb(fs_info, src, dst);
1351 	ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1352 	spin_unlock(&fs_info->qgroup_lock);
1353 out:
1354 	ulist_free(tmp);
1355 	return ret;
1356 }
1357 
1358 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1359 			      u64 dst)
1360 {
1361 	struct btrfs_fs_info *fs_info = trans->fs_info;
1362 	int ret = 0;
1363 
1364 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1365 	ret = __del_qgroup_relation(trans, src, dst);
1366 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1367 
1368 	return ret;
1369 }
1370 
1371 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1372 {
1373 	struct btrfs_fs_info *fs_info = trans->fs_info;
1374 	struct btrfs_root *quota_root;
1375 	struct btrfs_qgroup *qgroup;
1376 	int ret = 0;
1377 
1378 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1379 	quota_root = fs_info->quota_root;
1380 	if (!quota_root) {
1381 		ret = -EINVAL;
1382 		goto out;
1383 	}
1384 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1385 	if (qgroup) {
1386 		ret = -EEXIST;
1387 		goto out;
1388 	}
1389 
1390 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1391 	if (ret)
1392 		goto out;
1393 
1394 	spin_lock(&fs_info->qgroup_lock);
1395 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1396 	spin_unlock(&fs_info->qgroup_lock);
1397 
1398 	if (IS_ERR(qgroup))
1399 		ret = PTR_ERR(qgroup);
1400 out:
1401 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1402 	return ret;
1403 }
1404 
1405 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1406 {
1407 	struct btrfs_fs_info *fs_info = trans->fs_info;
1408 	struct btrfs_root *quota_root;
1409 	struct btrfs_qgroup *qgroup;
1410 	struct btrfs_qgroup_list *list;
1411 	int ret = 0;
1412 
1413 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1414 	quota_root = fs_info->quota_root;
1415 	if (!quota_root) {
1416 		ret = -EINVAL;
1417 		goto out;
1418 	}
1419 
1420 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1421 	if (!qgroup) {
1422 		ret = -ENOENT;
1423 		goto out;
1424 	}
1425 
1426 	/* Check if there are no children of this qgroup */
1427 	if (!list_empty(&qgroup->members)) {
1428 		ret = -EBUSY;
1429 		goto out;
1430 	}
1431 
1432 	ret = del_qgroup_item(trans, qgroupid);
1433 	if (ret && ret != -ENOENT)
1434 		goto out;
1435 
1436 	while (!list_empty(&qgroup->groups)) {
1437 		list = list_first_entry(&qgroup->groups,
1438 					struct btrfs_qgroup_list, next_group);
1439 		ret = __del_qgroup_relation(trans, qgroupid,
1440 					    list->group->qgroupid);
1441 		if (ret)
1442 			goto out;
1443 	}
1444 
1445 	spin_lock(&fs_info->qgroup_lock);
1446 	del_qgroup_rb(fs_info, qgroupid);
1447 	spin_unlock(&fs_info->qgroup_lock);
1448 out:
1449 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1450 	return ret;
1451 }
1452 
1453 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1454 		       struct btrfs_qgroup_limit *limit)
1455 {
1456 	struct btrfs_fs_info *fs_info = trans->fs_info;
1457 	struct btrfs_root *quota_root;
1458 	struct btrfs_qgroup *qgroup;
1459 	int ret = 0;
1460 	/* Sometimes we would want to clear the limit on this qgroup.
1461 	 * To meet this requirement, we treat the -1 as a special value
1462 	 * which tell kernel to clear the limit on this qgroup.
1463 	 */
1464 	const u64 CLEAR_VALUE = -1;
1465 
1466 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1467 	quota_root = fs_info->quota_root;
1468 	if (!quota_root) {
1469 		ret = -EINVAL;
1470 		goto out;
1471 	}
1472 
1473 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1474 	if (!qgroup) {
1475 		ret = -ENOENT;
1476 		goto out;
1477 	}
1478 
1479 	spin_lock(&fs_info->qgroup_lock);
1480 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1481 		if (limit->max_rfer == CLEAR_VALUE) {
1482 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1483 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1484 			qgroup->max_rfer = 0;
1485 		} else {
1486 			qgroup->max_rfer = limit->max_rfer;
1487 		}
1488 	}
1489 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1490 		if (limit->max_excl == CLEAR_VALUE) {
1491 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1492 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1493 			qgroup->max_excl = 0;
1494 		} else {
1495 			qgroup->max_excl = limit->max_excl;
1496 		}
1497 	}
1498 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1499 		if (limit->rsv_rfer == CLEAR_VALUE) {
1500 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1501 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1502 			qgroup->rsv_rfer = 0;
1503 		} else {
1504 			qgroup->rsv_rfer = limit->rsv_rfer;
1505 		}
1506 	}
1507 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1508 		if (limit->rsv_excl == CLEAR_VALUE) {
1509 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1510 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1511 			qgroup->rsv_excl = 0;
1512 		} else {
1513 			qgroup->rsv_excl = limit->rsv_excl;
1514 		}
1515 	}
1516 	qgroup->lim_flags |= limit->flags;
1517 
1518 	spin_unlock(&fs_info->qgroup_lock);
1519 
1520 	ret = update_qgroup_limit_item(trans, qgroup);
1521 	if (ret) {
1522 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1523 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1524 		       qgroupid);
1525 	}
1526 
1527 out:
1528 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1529 	return ret;
1530 }
1531 
1532 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1533 				struct btrfs_delayed_ref_root *delayed_refs,
1534 				struct btrfs_qgroup_extent_record *record)
1535 {
1536 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1537 	struct rb_node *parent_node = NULL;
1538 	struct btrfs_qgroup_extent_record *entry;
1539 	u64 bytenr = record->bytenr;
1540 
1541 	lockdep_assert_held(&delayed_refs->lock);
1542 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1543 
1544 	while (*p) {
1545 		parent_node = *p;
1546 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1547 				 node);
1548 		if (bytenr < entry->bytenr) {
1549 			p = &(*p)->rb_left;
1550 		} else if (bytenr > entry->bytenr) {
1551 			p = &(*p)->rb_right;
1552 		} else {
1553 			if (record->data_rsv && !entry->data_rsv) {
1554 				entry->data_rsv = record->data_rsv;
1555 				entry->data_rsv_refroot =
1556 					record->data_rsv_refroot;
1557 			}
1558 			return 1;
1559 		}
1560 	}
1561 
1562 	rb_link_node(&record->node, parent_node, p);
1563 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1564 	return 0;
1565 }
1566 
1567 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1568 				   struct btrfs_qgroup_extent_record *qrecord)
1569 {
1570 	struct ulist *old_root;
1571 	u64 bytenr = qrecord->bytenr;
1572 	int ret;
1573 
1574 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1575 	if (ret < 0) {
1576 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1577 		btrfs_warn(fs_info,
1578 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1579 			ret);
1580 		return 0;
1581 	}
1582 
1583 	/*
1584 	 * Here we don't need to get the lock of
1585 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1586 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1587 	 *
1588 	 * So modifying qrecord->old_roots is safe here
1589 	 */
1590 	qrecord->old_roots = old_root;
1591 	return 0;
1592 }
1593 
1594 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1595 			      u64 num_bytes, gfp_t gfp_flag)
1596 {
1597 	struct btrfs_fs_info *fs_info = trans->fs_info;
1598 	struct btrfs_qgroup_extent_record *record;
1599 	struct btrfs_delayed_ref_root *delayed_refs;
1600 	int ret;
1601 
1602 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1603 	    || bytenr == 0 || num_bytes == 0)
1604 		return 0;
1605 	record = kzalloc(sizeof(*record), gfp_flag);
1606 	if (!record)
1607 		return -ENOMEM;
1608 
1609 	delayed_refs = &trans->transaction->delayed_refs;
1610 	record->bytenr = bytenr;
1611 	record->num_bytes = num_bytes;
1612 	record->old_roots = NULL;
1613 
1614 	spin_lock(&delayed_refs->lock);
1615 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1616 	spin_unlock(&delayed_refs->lock);
1617 	if (ret > 0) {
1618 		kfree(record);
1619 		return 0;
1620 	}
1621 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1622 }
1623 
1624 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1625 				  struct extent_buffer *eb)
1626 {
1627 	struct btrfs_fs_info *fs_info = trans->fs_info;
1628 	int nr = btrfs_header_nritems(eb);
1629 	int i, extent_type, ret;
1630 	struct btrfs_key key;
1631 	struct btrfs_file_extent_item *fi;
1632 	u64 bytenr, num_bytes;
1633 
1634 	/* We can be called directly from walk_up_proc() */
1635 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1636 		return 0;
1637 
1638 	for (i = 0; i < nr; i++) {
1639 		btrfs_item_key_to_cpu(eb, &key, i);
1640 
1641 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1642 			continue;
1643 
1644 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1645 		/* filter out non qgroup-accountable extents  */
1646 		extent_type = btrfs_file_extent_type(eb, fi);
1647 
1648 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1649 			continue;
1650 
1651 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1652 		if (!bytenr)
1653 			continue;
1654 
1655 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1656 
1657 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1658 						GFP_NOFS);
1659 		if (ret)
1660 			return ret;
1661 	}
1662 	cond_resched();
1663 	return 0;
1664 }
1665 
1666 /*
1667  * Walk up the tree from the bottom, freeing leaves and any interior
1668  * nodes which have had all slots visited. If a node (leaf or
1669  * interior) is freed, the node above it will have it's slot
1670  * incremented. The root node will never be freed.
1671  *
1672  * At the end of this function, we should have a path which has all
1673  * slots incremented to the next position for a search. If we need to
1674  * read a new node it will be NULL and the node above it will have the
1675  * correct slot selected for a later read.
1676  *
1677  * If we increment the root nodes slot counter past the number of
1678  * elements, 1 is returned to signal completion of the search.
1679  */
1680 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1681 {
1682 	int level = 0;
1683 	int nr, slot;
1684 	struct extent_buffer *eb;
1685 
1686 	if (root_level == 0)
1687 		return 1;
1688 
1689 	while (level <= root_level) {
1690 		eb = path->nodes[level];
1691 		nr = btrfs_header_nritems(eb);
1692 		path->slots[level]++;
1693 		slot = path->slots[level];
1694 		if (slot >= nr || level == 0) {
1695 			/*
1696 			 * Don't free the root -  we will detect this
1697 			 * condition after our loop and return a
1698 			 * positive value for caller to stop walking the tree.
1699 			 */
1700 			if (level != root_level) {
1701 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1702 				path->locks[level] = 0;
1703 
1704 				free_extent_buffer(eb);
1705 				path->nodes[level] = NULL;
1706 				path->slots[level] = 0;
1707 			}
1708 		} else {
1709 			/*
1710 			 * We have a valid slot to walk back down
1711 			 * from. Stop here so caller can process these
1712 			 * new nodes.
1713 			 */
1714 			break;
1715 		}
1716 
1717 		level++;
1718 	}
1719 
1720 	eb = path->nodes[root_level];
1721 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1722 		return 1;
1723 
1724 	return 0;
1725 }
1726 
1727 /*
1728  * Helper function to trace a subtree tree block swap.
1729  *
1730  * The swap will happen in highest tree block, but there may be a lot of
1731  * tree blocks involved.
1732  *
1733  * For example:
1734  *  OO = Old tree blocks
1735  *  NN = New tree blocks allocated during balance
1736  *
1737  *           File tree (257)                  Reloc tree for 257
1738  * L2              OO                                NN
1739  *               /    \                            /    \
1740  * L1          OO      OO (a)                    OO      NN (a)
1741  *            / \     / \                       / \     / \
1742  * L0       OO   OO OO   OO                   OO   OO NN   NN
1743  *                  (b)  (c)                          (b)  (c)
1744  *
1745  * When calling qgroup_trace_extent_swap(), we will pass:
1746  * @src_eb = OO(a)
1747  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1748  * @dst_level = 0
1749  * @root_level = 1
1750  *
1751  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1752  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1753  *
1754  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1755  *
1756  * 1) Tree search from @src_eb
1757  *    It should acts as a simplified btrfs_search_slot().
1758  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1759  *    (first key).
1760  *
1761  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1762  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1763  *    They should be marked during previous (@dst_level = 1) iteration.
1764  *
1765  * 3) Mark file extents in leaves dirty
1766  *    We don't have good way to pick out new file extents only.
1767  *    So we still follow the old method by scanning all file extents in
1768  *    the leave.
1769  *
1770  * This function can free us from keeping two paths, thus later we only need
1771  * to care about how to iterate all new tree blocks in reloc tree.
1772  */
1773 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1774 				    struct extent_buffer *src_eb,
1775 				    struct btrfs_path *dst_path,
1776 				    int dst_level, int root_level,
1777 				    bool trace_leaf)
1778 {
1779 	struct btrfs_key key;
1780 	struct btrfs_path *src_path;
1781 	struct btrfs_fs_info *fs_info = trans->fs_info;
1782 	u32 nodesize = fs_info->nodesize;
1783 	int cur_level = root_level;
1784 	int ret;
1785 
1786 	BUG_ON(dst_level > root_level);
1787 	/* Level mismatch */
1788 	if (btrfs_header_level(src_eb) != root_level)
1789 		return -EINVAL;
1790 
1791 	src_path = btrfs_alloc_path();
1792 	if (!src_path) {
1793 		ret = -ENOMEM;
1794 		goto out;
1795 	}
1796 
1797 	if (dst_level)
1798 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1799 	else
1800 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1801 
1802 	/* For src_path */
1803 	extent_buffer_get(src_eb);
1804 	src_path->nodes[root_level] = src_eb;
1805 	src_path->slots[root_level] = dst_path->slots[root_level];
1806 	src_path->locks[root_level] = 0;
1807 
1808 	/* A simplified version of btrfs_search_slot() */
1809 	while (cur_level >= dst_level) {
1810 		struct btrfs_key src_key;
1811 		struct btrfs_key dst_key;
1812 
1813 		if (src_path->nodes[cur_level] == NULL) {
1814 			struct btrfs_key first_key;
1815 			struct extent_buffer *eb;
1816 			int parent_slot;
1817 			u64 child_gen;
1818 			u64 child_bytenr;
1819 
1820 			eb = src_path->nodes[cur_level + 1];
1821 			parent_slot = src_path->slots[cur_level + 1];
1822 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1823 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1824 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1825 
1826 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1827 					     cur_level, &first_key);
1828 			if (IS_ERR(eb)) {
1829 				ret = PTR_ERR(eb);
1830 				goto out;
1831 			} else if (!extent_buffer_uptodate(eb)) {
1832 				free_extent_buffer(eb);
1833 				ret = -EIO;
1834 				goto out;
1835 			}
1836 
1837 			src_path->nodes[cur_level] = eb;
1838 
1839 			btrfs_tree_read_lock(eb);
1840 			btrfs_set_lock_blocking_read(eb);
1841 			src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1842 		}
1843 
1844 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1845 		if (cur_level) {
1846 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1847 					&dst_key, dst_path->slots[cur_level]);
1848 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1849 					&src_key, src_path->slots[cur_level]);
1850 		} else {
1851 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1852 					&dst_key, dst_path->slots[cur_level]);
1853 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1854 					&src_key, src_path->slots[cur_level]);
1855 		}
1856 		/* Content mismatch, something went wrong */
1857 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1858 			ret = -ENOENT;
1859 			goto out;
1860 		}
1861 		cur_level--;
1862 	}
1863 
1864 	/*
1865 	 * Now both @dst_path and @src_path have been populated, record the tree
1866 	 * blocks for qgroup accounting.
1867 	 */
1868 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1869 			nodesize, GFP_NOFS);
1870 	if (ret < 0)
1871 		goto out;
1872 	ret = btrfs_qgroup_trace_extent(trans,
1873 			dst_path->nodes[dst_level]->start,
1874 			nodesize, GFP_NOFS);
1875 	if (ret < 0)
1876 		goto out;
1877 
1878 	/* Record leaf file extents */
1879 	if (dst_level == 0 && trace_leaf) {
1880 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1881 		if (ret < 0)
1882 			goto out;
1883 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1884 	}
1885 out:
1886 	btrfs_free_path(src_path);
1887 	return ret;
1888 }
1889 
1890 /*
1891  * Helper function to do recursive generation-aware depth-first search, to
1892  * locate all new tree blocks in a subtree of reloc tree.
1893  *
1894  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1895  *         reloc tree
1896  * L2         NN (a)
1897  *          /    \
1898  * L1    OO        NN (b)
1899  *      /  \      /  \
1900  * L0  OO  OO    OO  NN
1901  *               (c) (d)
1902  * If we pass:
1903  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1904  * @cur_level = 1
1905  * @root_level = 1
1906  *
1907  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1908  * above tree blocks along with their counter parts in file tree.
1909  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1910  * won't affect OO(c).
1911  */
1912 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1913 					   struct extent_buffer *src_eb,
1914 					   struct btrfs_path *dst_path,
1915 					   int cur_level, int root_level,
1916 					   u64 last_snapshot, bool trace_leaf)
1917 {
1918 	struct btrfs_fs_info *fs_info = trans->fs_info;
1919 	struct extent_buffer *eb;
1920 	bool need_cleanup = false;
1921 	int ret = 0;
1922 	int i;
1923 
1924 	/* Level sanity check */
1925 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1926 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1927 	    root_level < cur_level) {
1928 		btrfs_err_rl(fs_info,
1929 			"%s: bad levels, cur_level=%d root_level=%d",
1930 			__func__, cur_level, root_level);
1931 		return -EUCLEAN;
1932 	}
1933 
1934 	/* Read the tree block if needed */
1935 	if (dst_path->nodes[cur_level] == NULL) {
1936 		struct btrfs_key first_key;
1937 		int parent_slot;
1938 		u64 child_gen;
1939 		u64 child_bytenr;
1940 
1941 		/*
1942 		 * dst_path->nodes[root_level] must be initialized before
1943 		 * calling this function.
1944 		 */
1945 		if (cur_level == root_level) {
1946 			btrfs_err_rl(fs_info,
1947 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1948 				__func__, root_level, root_level, cur_level);
1949 			return -EUCLEAN;
1950 		}
1951 
1952 		/*
1953 		 * We need to get child blockptr/gen from parent before we can
1954 		 * read it.
1955 		  */
1956 		eb = dst_path->nodes[cur_level + 1];
1957 		parent_slot = dst_path->slots[cur_level + 1];
1958 		child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1959 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1960 		btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1961 
1962 		/* This node is old, no need to trace */
1963 		if (child_gen < last_snapshot)
1964 			goto out;
1965 
1966 		eb = read_tree_block(fs_info, child_bytenr, child_gen,
1967 				     cur_level, &first_key);
1968 		if (IS_ERR(eb)) {
1969 			ret = PTR_ERR(eb);
1970 			goto out;
1971 		} else if (!extent_buffer_uptodate(eb)) {
1972 			free_extent_buffer(eb);
1973 			ret = -EIO;
1974 			goto out;
1975 		}
1976 
1977 		dst_path->nodes[cur_level] = eb;
1978 		dst_path->slots[cur_level] = 0;
1979 
1980 		btrfs_tree_read_lock(eb);
1981 		btrfs_set_lock_blocking_read(eb);
1982 		dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1983 		need_cleanup = true;
1984 	}
1985 
1986 	/* Now record this tree block and its counter part for qgroups */
1987 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1988 				       root_level, trace_leaf);
1989 	if (ret < 0)
1990 		goto cleanup;
1991 
1992 	eb = dst_path->nodes[cur_level];
1993 
1994 	if (cur_level > 0) {
1995 		/* Iterate all child tree blocks */
1996 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
1997 			/* Skip old tree blocks as they won't be swapped */
1998 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1999 				continue;
2000 			dst_path->slots[cur_level] = i;
2001 
2002 			/* Recursive call (at most 7 times) */
2003 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2004 					dst_path, cur_level - 1, root_level,
2005 					last_snapshot, trace_leaf);
2006 			if (ret < 0)
2007 				goto cleanup;
2008 		}
2009 	}
2010 
2011 cleanup:
2012 	if (need_cleanup) {
2013 		/* Clean up */
2014 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2015 				     dst_path->locks[cur_level]);
2016 		free_extent_buffer(dst_path->nodes[cur_level]);
2017 		dst_path->nodes[cur_level] = NULL;
2018 		dst_path->slots[cur_level] = 0;
2019 		dst_path->locks[cur_level] = 0;
2020 	}
2021 out:
2022 	return ret;
2023 }
2024 
2025 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2026 				struct extent_buffer *src_eb,
2027 				struct extent_buffer *dst_eb,
2028 				u64 last_snapshot, bool trace_leaf)
2029 {
2030 	struct btrfs_fs_info *fs_info = trans->fs_info;
2031 	struct btrfs_path *dst_path = NULL;
2032 	int level;
2033 	int ret;
2034 
2035 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2036 		return 0;
2037 
2038 	/* Wrong parameter order */
2039 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2040 		btrfs_err_rl(fs_info,
2041 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2042 			     btrfs_header_generation(src_eb),
2043 			     btrfs_header_generation(dst_eb));
2044 		return -EUCLEAN;
2045 	}
2046 
2047 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2048 		ret = -EIO;
2049 		goto out;
2050 	}
2051 
2052 	level = btrfs_header_level(dst_eb);
2053 	dst_path = btrfs_alloc_path();
2054 	if (!dst_path) {
2055 		ret = -ENOMEM;
2056 		goto out;
2057 	}
2058 	/* For dst_path */
2059 	extent_buffer_get(dst_eb);
2060 	dst_path->nodes[level] = dst_eb;
2061 	dst_path->slots[level] = 0;
2062 	dst_path->locks[level] = 0;
2063 
2064 	/* Do the generation aware breadth-first search */
2065 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2066 					      level, last_snapshot, trace_leaf);
2067 	if (ret < 0)
2068 		goto out;
2069 	ret = 0;
2070 
2071 out:
2072 	btrfs_free_path(dst_path);
2073 	if (ret < 0)
2074 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2075 	return ret;
2076 }
2077 
2078 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2079 			       struct extent_buffer *root_eb,
2080 			       u64 root_gen, int root_level)
2081 {
2082 	struct btrfs_fs_info *fs_info = trans->fs_info;
2083 	int ret = 0;
2084 	int level;
2085 	struct extent_buffer *eb = root_eb;
2086 	struct btrfs_path *path = NULL;
2087 
2088 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2089 	BUG_ON(root_eb == NULL);
2090 
2091 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2092 		return 0;
2093 
2094 	if (!extent_buffer_uptodate(root_eb)) {
2095 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2096 		if (ret)
2097 			goto out;
2098 	}
2099 
2100 	if (root_level == 0) {
2101 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2102 		goto out;
2103 	}
2104 
2105 	path = btrfs_alloc_path();
2106 	if (!path)
2107 		return -ENOMEM;
2108 
2109 	/*
2110 	 * Walk down the tree.  Missing extent blocks are filled in as
2111 	 * we go. Metadata is accounted every time we read a new
2112 	 * extent block.
2113 	 *
2114 	 * When we reach a leaf, we account for file extent items in it,
2115 	 * walk back up the tree (adjusting slot pointers as we go)
2116 	 * and restart the search process.
2117 	 */
2118 	extent_buffer_get(root_eb); /* For path */
2119 	path->nodes[root_level] = root_eb;
2120 	path->slots[root_level] = 0;
2121 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2122 walk_down:
2123 	level = root_level;
2124 	while (level >= 0) {
2125 		if (path->nodes[level] == NULL) {
2126 			struct btrfs_key first_key;
2127 			int parent_slot;
2128 			u64 child_gen;
2129 			u64 child_bytenr;
2130 
2131 			/*
2132 			 * We need to get child blockptr/gen from parent before
2133 			 * we can read it.
2134 			  */
2135 			eb = path->nodes[level + 1];
2136 			parent_slot = path->slots[level + 1];
2137 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2138 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2139 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2140 
2141 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
2142 					     level, &first_key);
2143 			if (IS_ERR(eb)) {
2144 				ret = PTR_ERR(eb);
2145 				goto out;
2146 			} else if (!extent_buffer_uptodate(eb)) {
2147 				free_extent_buffer(eb);
2148 				ret = -EIO;
2149 				goto out;
2150 			}
2151 
2152 			path->nodes[level] = eb;
2153 			path->slots[level] = 0;
2154 
2155 			btrfs_tree_read_lock(eb);
2156 			btrfs_set_lock_blocking_read(eb);
2157 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2158 
2159 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2160 							fs_info->nodesize,
2161 							GFP_NOFS);
2162 			if (ret)
2163 				goto out;
2164 		}
2165 
2166 		if (level == 0) {
2167 			ret = btrfs_qgroup_trace_leaf_items(trans,
2168 							    path->nodes[level]);
2169 			if (ret)
2170 				goto out;
2171 
2172 			/* Nonzero return here means we completed our search */
2173 			ret = adjust_slots_upwards(path, root_level);
2174 			if (ret)
2175 				break;
2176 
2177 			/* Restart search with new slots */
2178 			goto walk_down;
2179 		}
2180 
2181 		level--;
2182 	}
2183 
2184 	ret = 0;
2185 out:
2186 	btrfs_free_path(path);
2187 
2188 	return ret;
2189 }
2190 
2191 #define UPDATE_NEW	0
2192 #define UPDATE_OLD	1
2193 /*
2194  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2195  */
2196 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2197 				struct ulist *roots, struct ulist *tmp,
2198 				struct ulist *qgroups, u64 seq, int update_old)
2199 {
2200 	struct ulist_node *unode;
2201 	struct ulist_iterator uiter;
2202 	struct ulist_node *tmp_unode;
2203 	struct ulist_iterator tmp_uiter;
2204 	struct btrfs_qgroup *qg;
2205 	int ret = 0;
2206 
2207 	if (!roots)
2208 		return 0;
2209 	ULIST_ITER_INIT(&uiter);
2210 	while ((unode = ulist_next(roots, &uiter))) {
2211 		qg = find_qgroup_rb(fs_info, unode->val);
2212 		if (!qg)
2213 			continue;
2214 
2215 		ulist_reinit(tmp);
2216 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2217 				GFP_ATOMIC);
2218 		if (ret < 0)
2219 			return ret;
2220 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2221 		if (ret < 0)
2222 			return ret;
2223 		ULIST_ITER_INIT(&tmp_uiter);
2224 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2225 			struct btrfs_qgroup_list *glist;
2226 
2227 			qg = unode_aux_to_qgroup(tmp_unode);
2228 			if (update_old)
2229 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2230 			else
2231 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2232 			list_for_each_entry(glist, &qg->groups, next_group) {
2233 				ret = ulist_add(qgroups, glist->group->qgroupid,
2234 						qgroup_to_aux(glist->group),
2235 						GFP_ATOMIC);
2236 				if (ret < 0)
2237 					return ret;
2238 				ret = ulist_add(tmp, glist->group->qgroupid,
2239 						qgroup_to_aux(glist->group),
2240 						GFP_ATOMIC);
2241 				if (ret < 0)
2242 					return ret;
2243 			}
2244 		}
2245 	}
2246 	return 0;
2247 }
2248 
2249 /*
2250  * Update qgroup rfer/excl counters.
2251  * Rfer update is easy, codes can explain themselves.
2252  *
2253  * Excl update is tricky, the update is split into 2 part.
2254  * Part 1: Possible exclusive <-> sharing detect:
2255  *	|	A	|	!A	|
2256  *  -------------------------------------
2257  *  B	|	*	|	-	|
2258  *  -------------------------------------
2259  *  !B	|	+	|	**	|
2260  *  -------------------------------------
2261  *
2262  * Conditions:
2263  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2264  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2265  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2266  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2267  *
2268  * Results:
2269  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2270  * *: Definitely not changed.		**: Possible unchanged.
2271  *
2272  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2273  *
2274  * To make the logic clear, we first use condition A and B to split
2275  * combination into 4 results.
2276  *
2277  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2278  * only on variant maybe 0.
2279  *
2280  * Lastly, check result **, since there are 2 variants maybe 0, split them
2281  * again(2x2).
2282  * But this time we don't need to consider other things, the codes and logic
2283  * is easy to understand now.
2284  */
2285 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2286 				  struct ulist *qgroups,
2287 				  u64 nr_old_roots,
2288 				  u64 nr_new_roots,
2289 				  u64 num_bytes, u64 seq)
2290 {
2291 	struct ulist_node *unode;
2292 	struct ulist_iterator uiter;
2293 	struct btrfs_qgroup *qg;
2294 	u64 cur_new_count, cur_old_count;
2295 
2296 	ULIST_ITER_INIT(&uiter);
2297 	while ((unode = ulist_next(qgroups, &uiter))) {
2298 		bool dirty = false;
2299 
2300 		qg = unode_aux_to_qgroup(unode);
2301 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2302 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2303 
2304 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2305 					     cur_new_count);
2306 
2307 		/* Rfer update part */
2308 		if (cur_old_count == 0 && cur_new_count > 0) {
2309 			qg->rfer += num_bytes;
2310 			qg->rfer_cmpr += num_bytes;
2311 			dirty = true;
2312 		}
2313 		if (cur_old_count > 0 && cur_new_count == 0) {
2314 			qg->rfer -= num_bytes;
2315 			qg->rfer_cmpr -= num_bytes;
2316 			dirty = true;
2317 		}
2318 
2319 		/* Excl update part */
2320 		/* Exclusive/none -> shared case */
2321 		if (cur_old_count == nr_old_roots &&
2322 		    cur_new_count < nr_new_roots) {
2323 			/* Exclusive -> shared */
2324 			if (cur_old_count != 0) {
2325 				qg->excl -= num_bytes;
2326 				qg->excl_cmpr -= num_bytes;
2327 				dirty = true;
2328 			}
2329 		}
2330 
2331 		/* Shared -> exclusive/none case */
2332 		if (cur_old_count < nr_old_roots &&
2333 		    cur_new_count == nr_new_roots) {
2334 			/* Shared->exclusive */
2335 			if (cur_new_count != 0) {
2336 				qg->excl += num_bytes;
2337 				qg->excl_cmpr += num_bytes;
2338 				dirty = true;
2339 			}
2340 		}
2341 
2342 		/* Exclusive/none -> exclusive/none case */
2343 		if (cur_old_count == nr_old_roots &&
2344 		    cur_new_count == nr_new_roots) {
2345 			if (cur_old_count == 0) {
2346 				/* None -> exclusive/none */
2347 
2348 				if (cur_new_count != 0) {
2349 					/* None -> exclusive */
2350 					qg->excl += num_bytes;
2351 					qg->excl_cmpr += num_bytes;
2352 					dirty = true;
2353 				}
2354 				/* None -> none, nothing changed */
2355 			} else {
2356 				/* Exclusive -> exclusive/none */
2357 
2358 				if (cur_new_count == 0) {
2359 					/* Exclusive -> none */
2360 					qg->excl -= num_bytes;
2361 					qg->excl_cmpr -= num_bytes;
2362 					dirty = true;
2363 				}
2364 				/* Exclusive -> exclusive, nothing changed */
2365 			}
2366 		}
2367 
2368 		if (dirty)
2369 			qgroup_dirty(fs_info, qg);
2370 	}
2371 	return 0;
2372 }
2373 
2374 /*
2375  * Check if the @roots potentially is a list of fs tree roots
2376  *
2377  * Return 0 for definitely not a fs/subvol tree roots ulist
2378  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2379  *          one as well)
2380  */
2381 static int maybe_fs_roots(struct ulist *roots)
2382 {
2383 	struct ulist_node *unode;
2384 	struct ulist_iterator uiter;
2385 
2386 	/* Empty one, still possible for fs roots */
2387 	if (!roots || roots->nnodes == 0)
2388 		return 1;
2389 
2390 	ULIST_ITER_INIT(&uiter);
2391 	unode = ulist_next(roots, &uiter);
2392 	if (!unode)
2393 		return 1;
2394 
2395 	/*
2396 	 * If it contains fs tree roots, then it must belong to fs/subvol
2397 	 * trees.
2398 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2399 	 */
2400 	return is_fstree(unode->val);
2401 }
2402 
2403 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2404 				u64 num_bytes, struct ulist *old_roots,
2405 				struct ulist *new_roots)
2406 {
2407 	struct btrfs_fs_info *fs_info = trans->fs_info;
2408 	struct ulist *qgroups = NULL;
2409 	struct ulist *tmp = NULL;
2410 	u64 seq;
2411 	u64 nr_new_roots = 0;
2412 	u64 nr_old_roots = 0;
2413 	int ret = 0;
2414 
2415 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2416 		return 0;
2417 
2418 	if (new_roots) {
2419 		if (!maybe_fs_roots(new_roots))
2420 			goto out_free;
2421 		nr_new_roots = new_roots->nnodes;
2422 	}
2423 	if (old_roots) {
2424 		if (!maybe_fs_roots(old_roots))
2425 			goto out_free;
2426 		nr_old_roots = old_roots->nnodes;
2427 	}
2428 
2429 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2430 	if (nr_old_roots == 0 && nr_new_roots == 0)
2431 		goto out_free;
2432 
2433 	BUG_ON(!fs_info->quota_root);
2434 
2435 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2436 					num_bytes, nr_old_roots, nr_new_roots);
2437 
2438 	qgroups = ulist_alloc(GFP_NOFS);
2439 	if (!qgroups) {
2440 		ret = -ENOMEM;
2441 		goto out_free;
2442 	}
2443 	tmp = ulist_alloc(GFP_NOFS);
2444 	if (!tmp) {
2445 		ret = -ENOMEM;
2446 		goto out_free;
2447 	}
2448 
2449 	mutex_lock(&fs_info->qgroup_rescan_lock);
2450 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2451 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2452 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2453 			ret = 0;
2454 			goto out_free;
2455 		}
2456 	}
2457 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2458 
2459 	spin_lock(&fs_info->qgroup_lock);
2460 	seq = fs_info->qgroup_seq;
2461 
2462 	/* Update old refcnts using old_roots */
2463 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2464 				   UPDATE_OLD);
2465 	if (ret < 0)
2466 		goto out;
2467 
2468 	/* Update new refcnts using new_roots */
2469 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2470 				   UPDATE_NEW);
2471 	if (ret < 0)
2472 		goto out;
2473 
2474 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2475 			       num_bytes, seq);
2476 
2477 	/*
2478 	 * Bump qgroup_seq to avoid seq overlap
2479 	 */
2480 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2481 out:
2482 	spin_unlock(&fs_info->qgroup_lock);
2483 out_free:
2484 	ulist_free(tmp);
2485 	ulist_free(qgroups);
2486 	ulist_free(old_roots);
2487 	ulist_free(new_roots);
2488 	return ret;
2489 }
2490 
2491 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2492 {
2493 	struct btrfs_fs_info *fs_info = trans->fs_info;
2494 	struct btrfs_qgroup_extent_record *record;
2495 	struct btrfs_delayed_ref_root *delayed_refs;
2496 	struct ulist *new_roots = NULL;
2497 	struct rb_node *node;
2498 	u64 num_dirty_extents = 0;
2499 	u64 qgroup_to_skip;
2500 	int ret = 0;
2501 
2502 	delayed_refs = &trans->transaction->delayed_refs;
2503 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2504 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2505 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2506 				  node);
2507 
2508 		num_dirty_extents++;
2509 		trace_btrfs_qgroup_account_extents(fs_info, record);
2510 
2511 		if (!ret) {
2512 			/*
2513 			 * Old roots should be searched when inserting qgroup
2514 			 * extent record
2515 			 */
2516 			if (WARN_ON(!record->old_roots)) {
2517 				/* Search commit root to find old_roots */
2518 				ret = btrfs_find_all_roots(NULL, fs_info,
2519 						record->bytenr, 0,
2520 						&record->old_roots, false);
2521 				if (ret < 0)
2522 					goto cleanup;
2523 			}
2524 
2525 			/* Free the reserved data space */
2526 			btrfs_qgroup_free_refroot(fs_info,
2527 					record->data_rsv_refroot,
2528 					record->data_rsv,
2529 					BTRFS_QGROUP_RSV_DATA);
2530 			/*
2531 			 * Use SEQ_LAST as time_seq to do special search, which
2532 			 * doesn't lock tree or delayed_refs and search current
2533 			 * root. It's safe inside commit_transaction().
2534 			 */
2535 			ret = btrfs_find_all_roots(trans, fs_info,
2536 				record->bytenr, SEQ_LAST, &new_roots, false);
2537 			if (ret < 0)
2538 				goto cleanup;
2539 			if (qgroup_to_skip) {
2540 				ulist_del(new_roots, qgroup_to_skip, 0);
2541 				ulist_del(record->old_roots, qgroup_to_skip,
2542 					  0);
2543 			}
2544 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2545 							  record->num_bytes,
2546 							  record->old_roots,
2547 							  new_roots);
2548 			record->old_roots = NULL;
2549 			new_roots = NULL;
2550 		}
2551 cleanup:
2552 		ulist_free(record->old_roots);
2553 		ulist_free(new_roots);
2554 		new_roots = NULL;
2555 		rb_erase(node, &delayed_refs->dirty_extent_root);
2556 		kfree(record);
2557 
2558 	}
2559 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2560 				       num_dirty_extents);
2561 	return ret;
2562 }
2563 
2564 /*
2565  * called from commit_transaction. Writes all changed qgroups to disk.
2566  */
2567 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2568 {
2569 	struct btrfs_fs_info *fs_info = trans->fs_info;
2570 	struct btrfs_root *quota_root = fs_info->quota_root;
2571 	int ret = 0;
2572 
2573 	if (!quota_root)
2574 		return ret;
2575 
2576 	spin_lock(&fs_info->qgroup_lock);
2577 	while (!list_empty(&fs_info->dirty_qgroups)) {
2578 		struct btrfs_qgroup *qgroup;
2579 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2580 					  struct btrfs_qgroup, dirty);
2581 		list_del_init(&qgroup->dirty);
2582 		spin_unlock(&fs_info->qgroup_lock);
2583 		ret = update_qgroup_info_item(trans, qgroup);
2584 		if (ret)
2585 			fs_info->qgroup_flags |=
2586 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2587 		ret = update_qgroup_limit_item(trans, qgroup);
2588 		if (ret)
2589 			fs_info->qgroup_flags |=
2590 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2591 		spin_lock(&fs_info->qgroup_lock);
2592 	}
2593 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2594 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2595 	else
2596 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2597 	spin_unlock(&fs_info->qgroup_lock);
2598 
2599 	ret = update_qgroup_status_item(trans);
2600 	if (ret)
2601 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2602 
2603 	return ret;
2604 }
2605 
2606 /*
2607  * Copy the accounting information between qgroups. This is necessary
2608  * when a snapshot or a subvolume is created. Throwing an error will
2609  * cause a transaction abort so we take extra care here to only error
2610  * when a readonly fs is a reasonable outcome.
2611  */
2612 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2613 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2614 {
2615 	int ret = 0;
2616 	int i;
2617 	u64 *i_qgroups;
2618 	struct btrfs_fs_info *fs_info = trans->fs_info;
2619 	struct btrfs_root *quota_root;
2620 	struct btrfs_qgroup *srcgroup;
2621 	struct btrfs_qgroup *dstgroup;
2622 	u32 level_size = 0;
2623 	u64 nums;
2624 
2625 	mutex_lock(&fs_info->qgroup_ioctl_lock);
2626 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2627 		goto out;
2628 
2629 	quota_root = fs_info->quota_root;
2630 	if (!quota_root) {
2631 		ret = -EINVAL;
2632 		goto out;
2633 	}
2634 
2635 	if (inherit) {
2636 		i_qgroups = (u64 *)(inherit + 1);
2637 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2638 		       2 * inherit->num_excl_copies;
2639 		for (i = 0; i < nums; ++i) {
2640 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2641 
2642 			/*
2643 			 * Zero out invalid groups so we can ignore
2644 			 * them later.
2645 			 */
2646 			if (!srcgroup ||
2647 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2648 				*i_qgroups = 0ULL;
2649 
2650 			++i_qgroups;
2651 		}
2652 	}
2653 
2654 	/*
2655 	 * create a tracking group for the subvol itself
2656 	 */
2657 	ret = add_qgroup_item(trans, quota_root, objectid);
2658 	if (ret)
2659 		goto out;
2660 
2661 	/*
2662 	 * add qgroup to all inherited groups
2663 	 */
2664 	if (inherit) {
2665 		i_qgroups = (u64 *)(inherit + 1);
2666 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2667 			if (*i_qgroups == 0)
2668 				continue;
2669 			ret = add_qgroup_relation_item(trans, objectid,
2670 						       *i_qgroups);
2671 			if (ret && ret != -EEXIST)
2672 				goto out;
2673 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2674 						       objectid);
2675 			if (ret && ret != -EEXIST)
2676 				goto out;
2677 		}
2678 		ret = 0;
2679 	}
2680 
2681 
2682 	spin_lock(&fs_info->qgroup_lock);
2683 
2684 	dstgroup = add_qgroup_rb(fs_info, objectid);
2685 	if (IS_ERR(dstgroup)) {
2686 		ret = PTR_ERR(dstgroup);
2687 		goto unlock;
2688 	}
2689 
2690 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2691 		dstgroup->lim_flags = inherit->lim.flags;
2692 		dstgroup->max_rfer = inherit->lim.max_rfer;
2693 		dstgroup->max_excl = inherit->lim.max_excl;
2694 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2695 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2696 
2697 		ret = update_qgroup_limit_item(trans, dstgroup);
2698 		if (ret) {
2699 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2700 			btrfs_info(fs_info,
2701 				   "unable to update quota limit for %llu",
2702 				   dstgroup->qgroupid);
2703 			goto unlock;
2704 		}
2705 	}
2706 
2707 	if (srcid) {
2708 		srcgroup = find_qgroup_rb(fs_info, srcid);
2709 		if (!srcgroup)
2710 			goto unlock;
2711 
2712 		/*
2713 		 * We call inherit after we clone the root in order to make sure
2714 		 * our counts don't go crazy, so at this point the only
2715 		 * difference between the two roots should be the root node.
2716 		 */
2717 		level_size = fs_info->nodesize;
2718 		dstgroup->rfer = srcgroup->rfer;
2719 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2720 		dstgroup->excl = level_size;
2721 		dstgroup->excl_cmpr = level_size;
2722 		srcgroup->excl = level_size;
2723 		srcgroup->excl_cmpr = level_size;
2724 
2725 		/* inherit the limit info */
2726 		dstgroup->lim_flags = srcgroup->lim_flags;
2727 		dstgroup->max_rfer = srcgroup->max_rfer;
2728 		dstgroup->max_excl = srcgroup->max_excl;
2729 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2730 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2731 
2732 		qgroup_dirty(fs_info, dstgroup);
2733 		qgroup_dirty(fs_info, srcgroup);
2734 	}
2735 
2736 	if (!inherit)
2737 		goto unlock;
2738 
2739 	i_qgroups = (u64 *)(inherit + 1);
2740 	for (i = 0; i < inherit->num_qgroups; ++i) {
2741 		if (*i_qgroups) {
2742 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2743 			if (ret)
2744 				goto unlock;
2745 		}
2746 		++i_qgroups;
2747 	}
2748 
2749 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2750 		struct btrfs_qgroup *src;
2751 		struct btrfs_qgroup *dst;
2752 
2753 		if (!i_qgroups[0] || !i_qgroups[1])
2754 			continue;
2755 
2756 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2757 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2758 
2759 		if (!src || !dst) {
2760 			ret = -EINVAL;
2761 			goto unlock;
2762 		}
2763 
2764 		dst->rfer = src->rfer - level_size;
2765 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2766 	}
2767 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2768 		struct btrfs_qgroup *src;
2769 		struct btrfs_qgroup *dst;
2770 
2771 		if (!i_qgroups[0] || !i_qgroups[1])
2772 			continue;
2773 
2774 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2775 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2776 
2777 		if (!src || !dst) {
2778 			ret = -EINVAL;
2779 			goto unlock;
2780 		}
2781 
2782 		dst->excl = src->excl + level_size;
2783 		dst->excl_cmpr = src->excl_cmpr + level_size;
2784 	}
2785 
2786 unlock:
2787 	spin_unlock(&fs_info->qgroup_lock);
2788 out:
2789 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
2790 	return ret;
2791 }
2792 
2793 /*
2794  * Two limits to commit transaction in advance.
2795  *
2796  * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2797  * For SIZE, it will be in byte unit as threshold.
2798  */
2799 #define QGROUP_FREE_RATIO		32
2800 #define QGROUP_FREE_SIZE		SZ_32M
2801 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2802 				const struct btrfs_qgroup *qg, u64 num_bytes)
2803 {
2804 	u64 free;
2805 	u64 threshold;
2806 
2807 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2808 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2809 		return false;
2810 
2811 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2812 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2813 		return false;
2814 
2815 	/*
2816 	 * Even if we passed the check, it's better to check if reservation
2817 	 * for meta_pertrans is pushing us near limit.
2818 	 * If there is too much pertrans reservation or it's near the limit,
2819 	 * let's try commit transaction to free some, using transaction_kthread
2820 	 */
2821 	if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2822 			      BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2823 		if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2824 			free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2825 			threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2826 					  QGROUP_FREE_SIZE);
2827 		} else {
2828 			free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2829 			threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2830 					  QGROUP_FREE_SIZE);
2831 		}
2832 
2833 		/*
2834 		 * Use transaction_kthread to commit transaction, so we no
2835 		 * longer need to bother nested transaction nor lock context.
2836 		 */
2837 		if (free < threshold)
2838 			btrfs_commit_transaction_locksafe(fs_info);
2839 	}
2840 
2841 	return true;
2842 }
2843 
2844 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2845 			  enum btrfs_qgroup_rsv_type type)
2846 {
2847 	struct btrfs_root *quota_root;
2848 	struct btrfs_qgroup *qgroup;
2849 	struct btrfs_fs_info *fs_info = root->fs_info;
2850 	u64 ref_root = root->root_key.objectid;
2851 	int ret = 0;
2852 	struct ulist_node *unode;
2853 	struct ulist_iterator uiter;
2854 
2855 	if (!is_fstree(ref_root))
2856 		return 0;
2857 
2858 	if (num_bytes == 0)
2859 		return 0;
2860 
2861 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2862 	    capable(CAP_SYS_RESOURCE))
2863 		enforce = false;
2864 
2865 	spin_lock(&fs_info->qgroup_lock);
2866 	quota_root = fs_info->quota_root;
2867 	if (!quota_root)
2868 		goto out;
2869 
2870 	qgroup = find_qgroup_rb(fs_info, ref_root);
2871 	if (!qgroup)
2872 		goto out;
2873 
2874 	/*
2875 	 * in a first step, we check all affected qgroups if any limits would
2876 	 * be exceeded
2877 	 */
2878 	ulist_reinit(fs_info->qgroup_ulist);
2879 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2880 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2881 	if (ret < 0)
2882 		goto out;
2883 	ULIST_ITER_INIT(&uiter);
2884 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2885 		struct btrfs_qgroup *qg;
2886 		struct btrfs_qgroup_list *glist;
2887 
2888 		qg = unode_aux_to_qgroup(unode);
2889 
2890 		if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2891 			ret = -EDQUOT;
2892 			goto out;
2893 		}
2894 
2895 		list_for_each_entry(glist, &qg->groups, next_group) {
2896 			ret = ulist_add(fs_info->qgroup_ulist,
2897 					glist->group->qgroupid,
2898 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2899 			if (ret < 0)
2900 				goto out;
2901 		}
2902 	}
2903 	ret = 0;
2904 	/*
2905 	 * no limits exceeded, now record the reservation into all qgroups
2906 	 */
2907 	ULIST_ITER_INIT(&uiter);
2908 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2909 		struct btrfs_qgroup *qg;
2910 
2911 		qg = unode_aux_to_qgroup(unode);
2912 
2913 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
2914 	}
2915 
2916 out:
2917 	spin_unlock(&fs_info->qgroup_lock);
2918 	return ret;
2919 }
2920 
2921 /*
2922  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2923  * qgroup).
2924  *
2925  * Will handle all higher level qgroup too.
2926  *
2927  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2928  * This special case is only used for META_PERTRANS type.
2929  */
2930 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2931 			       u64 ref_root, u64 num_bytes,
2932 			       enum btrfs_qgroup_rsv_type type)
2933 {
2934 	struct btrfs_root *quota_root;
2935 	struct btrfs_qgroup *qgroup;
2936 	struct ulist_node *unode;
2937 	struct ulist_iterator uiter;
2938 	int ret = 0;
2939 
2940 	if (!is_fstree(ref_root))
2941 		return;
2942 
2943 	if (num_bytes == 0)
2944 		return;
2945 
2946 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2947 		WARN(1, "%s: Invalid type to free", __func__);
2948 		return;
2949 	}
2950 	spin_lock(&fs_info->qgroup_lock);
2951 
2952 	quota_root = fs_info->quota_root;
2953 	if (!quota_root)
2954 		goto out;
2955 
2956 	qgroup = find_qgroup_rb(fs_info, ref_root);
2957 	if (!qgroup)
2958 		goto out;
2959 
2960 	if (num_bytes == (u64)-1)
2961 		/*
2962 		 * We're freeing all pertrans rsv, get reserved value from
2963 		 * level 0 qgroup as real num_bytes to free.
2964 		 */
2965 		num_bytes = qgroup->rsv.values[type];
2966 
2967 	ulist_reinit(fs_info->qgroup_ulist);
2968 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2969 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2970 	if (ret < 0)
2971 		goto out;
2972 	ULIST_ITER_INIT(&uiter);
2973 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2974 		struct btrfs_qgroup *qg;
2975 		struct btrfs_qgroup_list *glist;
2976 
2977 		qg = unode_aux_to_qgroup(unode);
2978 
2979 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
2980 
2981 		list_for_each_entry(glist, &qg->groups, next_group) {
2982 			ret = ulist_add(fs_info->qgroup_ulist,
2983 					glist->group->qgroupid,
2984 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2985 			if (ret < 0)
2986 				goto out;
2987 		}
2988 	}
2989 
2990 out:
2991 	spin_unlock(&fs_info->qgroup_lock);
2992 }
2993 
2994 /*
2995  * Check if the leaf is the last leaf. Which means all node pointers
2996  * are at their last position.
2997  */
2998 static bool is_last_leaf(struct btrfs_path *path)
2999 {
3000 	int i;
3001 
3002 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3003 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3004 			return false;
3005 	}
3006 	return true;
3007 }
3008 
3009 /*
3010  * returns < 0 on error, 0 when more leafs are to be scanned.
3011  * returns 1 when done.
3012  */
3013 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3014 			      struct btrfs_path *path)
3015 {
3016 	struct btrfs_fs_info *fs_info = trans->fs_info;
3017 	struct btrfs_key found;
3018 	struct extent_buffer *scratch_leaf = NULL;
3019 	struct ulist *roots = NULL;
3020 	u64 num_bytes;
3021 	bool done;
3022 	int slot;
3023 	int ret;
3024 
3025 	mutex_lock(&fs_info->qgroup_rescan_lock);
3026 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3027 					 &fs_info->qgroup_rescan_progress,
3028 					 path, 1, 0);
3029 
3030 	btrfs_debug(fs_info,
3031 		"current progress key (%llu %u %llu), search_slot ret %d",
3032 		fs_info->qgroup_rescan_progress.objectid,
3033 		fs_info->qgroup_rescan_progress.type,
3034 		fs_info->qgroup_rescan_progress.offset, ret);
3035 
3036 	if (ret) {
3037 		/*
3038 		 * The rescan is about to end, we will not be scanning any
3039 		 * further blocks. We cannot unset the RESCAN flag here, because
3040 		 * we want to commit the transaction if everything went well.
3041 		 * To make the live accounting work in this phase, we set our
3042 		 * scan progress pointer such that every real extent objectid
3043 		 * will be smaller.
3044 		 */
3045 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3046 		btrfs_release_path(path);
3047 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3048 		return ret;
3049 	}
3050 	done = is_last_leaf(path);
3051 
3052 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3053 			      btrfs_header_nritems(path->nodes[0]) - 1);
3054 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3055 
3056 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3057 	if (!scratch_leaf) {
3058 		ret = -ENOMEM;
3059 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3060 		goto out;
3061 	}
3062 	slot = path->slots[0];
3063 	btrfs_release_path(path);
3064 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3065 
3066 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3067 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3068 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3069 		    found.type != BTRFS_METADATA_ITEM_KEY)
3070 			continue;
3071 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3072 			num_bytes = fs_info->nodesize;
3073 		else
3074 			num_bytes = found.offset;
3075 
3076 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3077 					   &roots, false);
3078 		if (ret < 0)
3079 			goto out;
3080 		/* For rescan, just pass old_roots as NULL */
3081 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3082 						  num_bytes, NULL, roots);
3083 		if (ret < 0)
3084 			goto out;
3085 	}
3086 out:
3087 	if (scratch_leaf)
3088 		free_extent_buffer(scratch_leaf);
3089 
3090 	if (done && !ret) {
3091 		ret = 1;
3092 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3093 	}
3094 	return ret;
3095 }
3096 
3097 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3098 {
3099 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3100 						     qgroup_rescan_work);
3101 	struct btrfs_path *path;
3102 	struct btrfs_trans_handle *trans = NULL;
3103 	int err = -ENOMEM;
3104 	int ret = 0;
3105 
3106 	path = btrfs_alloc_path();
3107 	if (!path)
3108 		goto out;
3109 	/*
3110 	 * Rescan should only search for commit root, and any later difference
3111 	 * should be recorded by qgroup
3112 	 */
3113 	path->search_commit_root = 1;
3114 	path->skip_locking = 1;
3115 
3116 	err = 0;
3117 	while (!err && !btrfs_fs_closing(fs_info)) {
3118 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3119 		if (IS_ERR(trans)) {
3120 			err = PTR_ERR(trans);
3121 			break;
3122 		}
3123 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3124 			err = -EINTR;
3125 		} else {
3126 			err = qgroup_rescan_leaf(trans, path);
3127 		}
3128 		if (err > 0)
3129 			btrfs_commit_transaction(trans);
3130 		else
3131 			btrfs_end_transaction(trans);
3132 	}
3133 
3134 out:
3135 	btrfs_free_path(path);
3136 
3137 	mutex_lock(&fs_info->qgroup_rescan_lock);
3138 	if (!btrfs_fs_closing(fs_info))
3139 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3140 
3141 	if (err > 0 &&
3142 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3143 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3144 	} else if (err < 0) {
3145 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3146 	}
3147 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3148 
3149 	/*
3150 	 * only update status, since the previous part has already updated the
3151 	 * qgroup info.
3152 	 */
3153 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3154 	if (IS_ERR(trans)) {
3155 		err = PTR_ERR(trans);
3156 		btrfs_err(fs_info,
3157 			  "fail to start transaction for status update: %d",
3158 			  err);
3159 		goto done;
3160 	}
3161 	ret = update_qgroup_status_item(trans);
3162 	if (ret < 0) {
3163 		err = ret;
3164 		btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3165 	}
3166 	btrfs_end_transaction(trans);
3167 
3168 	if (btrfs_fs_closing(fs_info)) {
3169 		btrfs_info(fs_info, "qgroup scan paused");
3170 	} else if (err >= 0) {
3171 		btrfs_info(fs_info, "qgroup scan completed%s",
3172 			err > 0 ? " (inconsistency flag cleared)" : "");
3173 	} else {
3174 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3175 	}
3176 
3177 done:
3178 	mutex_lock(&fs_info->qgroup_rescan_lock);
3179 	fs_info->qgroup_rescan_running = false;
3180 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3181 	complete_all(&fs_info->qgroup_rescan_completion);
3182 }
3183 
3184 /*
3185  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3186  * memory required for the rescan context.
3187  */
3188 static int
3189 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3190 		   int init_flags)
3191 {
3192 	int ret = 0;
3193 
3194 	if (!init_flags) {
3195 		/* we're resuming qgroup rescan at mount time */
3196 		if (!(fs_info->qgroup_flags &
3197 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3198 			btrfs_warn(fs_info,
3199 			"qgroup rescan init failed, qgroup is not enabled");
3200 			ret = -EINVAL;
3201 		} else if (!(fs_info->qgroup_flags &
3202 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3203 			btrfs_warn(fs_info,
3204 			"qgroup rescan init failed, qgroup rescan is not queued");
3205 			ret = -EINVAL;
3206 		}
3207 
3208 		if (ret)
3209 			return ret;
3210 	}
3211 
3212 	mutex_lock(&fs_info->qgroup_rescan_lock);
3213 	spin_lock(&fs_info->qgroup_lock);
3214 
3215 	if (init_flags) {
3216 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3217 			btrfs_warn(fs_info,
3218 				   "qgroup rescan is already in progress");
3219 			ret = -EINPROGRESS;
3220 		} else if (!(fs_info->qgroup_flags &
3221 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3222 			btrfs_warn(fs_info,
3223 			"qgroup rescan init failed, qgroup is not enabled");
3224 			ret = -EINVAL;
3225 		}
3226 
3227 		if (ret) {
3228 			spin_unlock(&fs_info->qgroup_lock);
3229 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3230 			return ret;
3231 		}
3232 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3233 	}
3234 
3235 	memset(&fs_info->qgroup_rescan_progress, 0,
3236 		sizeof(fs_info->qgroup_rescan_progress));
3237 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3238 	init_completion(&fs_info->qgroup_rescan_completion);
3239 	fs_info->qgroup_rescan_running = true;
3240 
3241 	spin_unlock(&fs_info->qgroup_lock);
3242 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3243 
3244 	memset(&fs_info->qgroup_rescan_work, 0,
3245 	       sizeof(fs_info->qgroup_rescan_work));
3246 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3247 			btrfs_qgroup_rescan_helper,
3248 			btrfs_qgroup_rescan_worker, NULL, NULL);
3249 	return 0;
3250 }
3251 
3252 static void
3253 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3254 {
3255 	struct rb_node *n;
3256 	struct btrfs_qgroup *qgroup;
3257 
3258 	spin_lock(&fs_info->qgroup_lock);
3259 	/* clear all current qgroup tracking information */
3260 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3261 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3262 		qgroup->rfer = 0;
3263 		qgroup->rfer_cmpr = 0;
3264 		qgroup->excl = 0;
3265 		qgroup->excl_cmpr = 0;
3266 		qgroup_dirty(fs_info, qgroup);
3267 	}
3268 	spin_unlock(&fs_info->qgroup_lock);
3269 }
3270 
3271 int
3272 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3273 {
3274 	int ret = 0;
3275 	struct btrfs_trans_handle *trans;
3276 
3277 	ret = qgroup_rescan_init(fs_info, 0, 1);
3278 	if (ret)
3279 		return ret;
3280 
3281 	/*
3282 	 * We have set the rescan_progress to 0, which means no more
3283 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3284 	 * However, btrfs_qgroup_account_ref may be right after its call
3285 	 * to btrfs_find_all_roots, in which case it would still do the
3286 	 * accounting.
3287 	 * To solve this, we're committing the transaction, which will
3288 	 * ensure we run all delayed refs and only after that, we are
3289 	 * going to clear all tracking information for a clean start.
3290 	 */
3291 
3292 	trans = btrfs_join_transaction(fs_info->fs_root);
3293 	if (IS_ERR(trans)) {
3294 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3295 		return PTR_ERR(trans);
3296 	}
3297 	ret = btrfs_commit_transaction(trans);
3298 	if (ret) {
3299 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3300 		return ret;
3301 	}
3302 
3303 	qgroup_rescan_zero_tracking(fs_info);
3304 
3305 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3306 			 &fs_info->qgroup_rescan_work);
3307 
3308 	return 0;
3309 }
3310 
3311 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3312 				     bool interruptible)
3313 {
3314 	int running;
3315 	int ret = 0;
3316 
3317 	mutex_lock(&fs_info->qgroup_rescan_lock);
3318 	spin_lock(&fs_info->qgroup_lock);
3319 	running = fs_info->qgroup_rescan_running;
3320 	spin_unlock(&fs_info->qgroup_lock);
3321 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3322 
3323 	if (!running)
3324 		return 0;
3325 
3326 	if (interruptible)
3327 		ret = wait_for_completion_interruptible(
3328 					&fs_info->qgroup_rescan_completion);
3329 	else
3330 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3331 
3332 	return ret;
3333 }
3334 
3335 /*
3336  * this is only called from open_ctree where we're still single threaded, thus
3337  * locking is omitted here.
3338  */
3339 void
3340 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3341 {
3342 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3343 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3344 				 &fs_info->qgroup_rescan_work);
3345 }
3346 
3347 /*
3348  * Reserve qgroup space for range [start, start + len).
3349  *
3350  * This function will either reserve space from related qgroups or doing
3351  * nothing if the range is already reserved.
3352  *
3353  * Return 0 for successful reserve
3354  * Return <0 for error (including -EQUOT)
3355  *
3356  * NOTE: this function may sleep for memory allocation.
3357  *       if btrfs_qgroup_reserve_data() is called multiple times with
3358  *       same @reserved, caller must ensure when error happens it's OK
3359  *       to free *ALL* reserved space.
3360  */
3361 int btrfs_qgroup_reserve_data(struct inode *inode,
3362 			struct extent_changeset **reserved_ret, u64 start,
3363 			u64 len)
3364 {
3365 	struct btrfs_root *root = BTRFS_I(inode)->root;
3366 	struct ulist_node *unode;
3367 	struct ulist_iterator uiter;
3368 	struct extent_changeset *reserved;
3369 	u64 orig_reserved;
3370 	u64 to_reserve;
3371 	int ret;
3372 
3373 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3374 	    !is_fstree(root->root_key.objectid) || len == 0)
3375 		return 0;
3376 
3377 	/* @reserved parameter is mandatory for qgroup */
3378 	if (WARN_ON(!reserved_ret))
3379 		return -EINVAL;
3380 	if (!*reserved_ret) {
3381 		*reserved_ret = extent_changeset_alloc();
3382 		if (!*reserved_ret)
3383 			return -ENOMEM;
3384 	}
3385 	reserved = *reserved_ret;
3386 	/* Record already reserved space */
3387 	orig_reserved = reserved->bytes_changed;
3388 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3389 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3390 
3391 	/* Newly reserved space */
3392 	to_reserve = reserved->bytes_changed - orig_reserved;
3393 	trace_btrfs_qgroup_reserve_data(inode, start, len,
3394 					to_reserve, QGROUP_RESERVE);
3395 	if (ret < 0)
3396 		goto cleanup;
3397 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3398 	if (ret < 0)
3399 		goto cleanup;
3400 
3401 	return ret;
3402 
3403 cleanup:
3404 	/* cleanup *ALL* already reserved ranges */
3405 	ULIST_ITER_INIT(&uiter);
3406 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3407 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3408 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3409 	extent_changeset_release(reserved);
3410 	return ret;
3411 }
3412 
3413 /* Free ranges specified by @reserved, normally in error path */
3414 static int qgroup_free_reserved_data(struct inode *inode,
3415 			struct extent_changeset *reserved, u64 start, u64 len)
3416 {
3417 	struct btrfs_root *root = BTRFS_I(inode)->root;
3418 	struct ulist_node *unode;
3419 	struct ulist_iterator uiter;
3420 	struct extent_changeset changeset;
3421 	int freed = 0;
3422 	int ret;
3423 
3424 	extent_changeset_init(&changeset);
3425 	len = round_up(start + len, root->fs_info->sectorsize);
3426 	start = round_down(start, root->fs_info->sectorsize);
3427 
3428 	ULIST_ITER_INIT(&uiter);
3429 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3430 		u64 range_start = unode->val;
3431 		/* unode->aux is the inclusive end */
3432 		u64 range_len = unode->aux - range_start + 1;
3433 		u64 free_start;
3434 		u64 free_len;
3435 
3436 		extent_changeset_release(&changeset);
3437 
3438 		/* Only free range in range [start, start + len) */
3439 		if (range_start >= start + len ||
3440 		    range_start + range_len <= start)
3441 			continue;
3442 		free_start = max(range_start, start);
3443 		free_len = min(start + len, range_start + range_len) -
3444 			   free_start;
3445 		/*
3446 		 * TODO: To also modify reserved->ranges_reserved to reflect
3447 		 * the modification.
3448 		 *
3449 		 * However as long as we free qgroup reserved according to
3450 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3451 		 * So not need to rush.
3452 		 */
3453 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3454 				free_start, free_start + free_len - 1,
3455 				EXTENT_QGROUP_RESERVED, &changeset);
3456 		if (ret < 0)
3457 			goto out;
3458 		freed += changeset.bytes_changed;
3459 	}
3460 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3461 				  BTRFS_QGROUP_RSV_DATA);
3462 	ret = freed;
3463 out:
3464 	extent_changeset_release(&changeset);
3465 	return ret;
3466 }
3467 
3468 static int __btrfs_qgroup_release_data(struct inode *inode,
3469 			struct extent_changeset *reserved, u64 start, u64 len,
3470 			int free)
3471 {
3472 	struct extent_changeset changeset;
3473 	int trace_op = QGROUP_RELEASE;
3474 	int ret;
3475 
3476 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3477 		      &BTRFS_I(inode)->root->fs_info->flags))
3478 		return 0;
3479 
3480 	/* In release case, we shouldn't have @reserved */
3481 	WARN_ON(!free && reserved);
3482 	if (free && reserved)
3483 		return qgroup_free_reserved_data(inode, reserved, start, len);
3484 	extent_changeset_init(&changeset);
3485 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3486 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3487 	if (ret < 0)
3488 		goto out;
3489 
3490 	if (free)
3491 		trace_op = QGROUP_FREE;
3492 	trace_btrfs_qgroup_release_data(inode, start, len,
3493 					changeset.bytes_changed, trace_op);
3494 	if (free)
3495 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3496 				BTRFS_I(inode)->root->root_key.objectid,
3497 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3498 	ret = changeset.bytes_changed;
3499 out:
3500 	extent_changeset_release(&changeset);
3501 	return ret;
3502 }
3503 
3504 /*
3505  * Free a reserved space range from io_tree and related qgroups
3506  *
3507  * Should be called when a range of pages get invalidated before reaching disk.
3508  * Or for error cleanup case.
3509  * if @reserved is given, only reserved range in [@start, @start + @len) will
3510  * be freed.
3511  *
3512  * For data written to disk, use btrfs_qgroup_release_data().
3513  *
3514  * NOTE: This function may sleep for memory allocation.
3515  */
3516 int btrfs_qgroup_free_data(struct inode *inode,
3517 			struct extent_changeset *reserved, u64 start, u64 len)
3518 {
3519 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3520 }
3521 
3522 /*
3523  * Release a reserved space range from io_tree only.
3524  *
3525  * Should be called when a range of pages get written to disk and corresponding
3526  * FILE_EXTENT is inserted into corresponding root.
3527  *
3528  * Since new qgroup accounting framework will only update qgroup numbers at
3529  * commit_transaction() time, its reserved space shouldn't be freed from
3530  * related qgroups.
3531  *
3532  * But we should release the range from io_tree, to allow further write to be
3533  * COWed.
3534  *
3535  * NOTE: This function may sleep for memory allocation.
3536  */
3537 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3538 {
3539 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3540 }
3541 
3542 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3543 			      enum btrfs_qgroup_rsv_type type)
3544 {
3545 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3546 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3547 		return;
3548 	if (num_bytes == 0)
3549 		return;
3550 
3551 	spin_lock(&root->qgroup_meta_rsv_lock);
3552 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3553 		root->qgroup_meta_rsv_prealloc += num_bytes;
3554 	else
3555 		root->qgroup_meta_rsv_pertrans += num_bytes;
3556 	spin_unlock(&root->qgroup_meta_rsv_lock);
3557 }
3558 
3559 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3560 			     enum btrfs_qgroup_rsv_type type)
3561 {
3562 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3563 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3564 		return 0;
3565 	if (num_bytes == 0)
3566 		return 0;
3567 
3568 	spin_lock(&root->qgroup_meta_rsv_lock);
3569 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3570 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3571 				  num_bytes);
3572 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3573 	} else {
3574 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3575 				  num_bytes);
3576 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3577 	}
3578 	spin_unlock(&root->qgroup_meta_rsv_lock);
3579 	return num_bytes;
3580 }
3581 
3582 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3583 				enum btrfs_qgroup_rsv_type type, bool enforce)
3584 {
3585 	struct btrfs_fs_info *fs_info = root->fs_info;
3586 	int ret;
3587 
3588 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3589 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3590 		return 0;
3591 
3592 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3593 	trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3594 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3595 	if (ret < 0)
3596 		return ret;
3597 	/*
3598 	 * Record what we have reserved into root.
3599 	 *
3600 	 * To avoid quota disabled->enabled underflow.
3601 	 * In that case, we may try to free space we haven't reserved
3602 	 * (since quota was disabled), so record what we reserved into root.
3603 	 * And ensure later release won't underflow this number.
3604 	 */
3605 	add_root_meta_rsv(root, num_bytes, type);
3606 	return ret;
3607 }
3608 
3609 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3610 {
3611 	struct btrfs_fs_info *fs_info = root->fs_info;
3612 
3613 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3614 	    !is_fstree(root->root_key.objectid))
3615 		return;
3616 
3617 	/* TODO: Update trace point to handle such free */
3618 	trace_qgroup_meta_free_all_pertrans(root);
3619 	/* Special value -1 means to free all reserved space */
3620 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3621 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3622 }
3623 
3624 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3625 			      enum btrfs_qgroup_rsv_type type)
3626 {
3627 	struct btrfs_fs_info *fs_info = root->fs_info;
3628 
3629 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3630 	    !is_fstree(root->root_key.objectid))
3631 		return;
3632 
3633 	/*
3634 	 * reservation for META_PREALLOC can happen before quota is enabled,
3635 	 * which can lead to underflow.
3636 	 * Here ensure we will only free what we really have reserved.
3637 	 */
3638 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3639 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3640 	trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3641 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3642 				  num_bytes, type);
3643 }
3644 
3645 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3646 				int num_bytes)
3647 {
3648 	struct btrfs_root *quota_root = fs_info->quota_root;
3649 	struct btrfs_qgroup *qgroup;
3650 	struct ulist_node *unode;
3651 	struct ulist_iterator uiter;
3652 	int ret = 0;
3653 
3654 	if (num_bytes == 0)
3655 		return;
3656 	if (!quota_root)
3657 		return;
3658 
3659 	spin_lock(&fs_info->qgroup_lock);
3660 	qgroup = find_qgroup_rb(fs_info, ref_root);
3661 	if (!qgroup)
3662 		goto out;
3663 	ulist_reinit(fs_info->qgroup_ulist);
3664 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3665 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3666 	if (ret < 0)
3667 		goto out;
3668 	ULIST_ITER_INIT(&uiter);
3669 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3670 		struct btrfs_qgroup *qg;
3671 		struct btrfs_qgroup_list *glist;
3672 
3673 		qg = unode_aux_to_qgroup(unode);
3674 
3675 		qgroup_rsv_release(fs_info, qg, num_bytes,
3676 				BTRFS_QGROUP_RSV_META_PREALLOC);
3677 		qgroup_rsv_add(fs_info, qg, num_bytes,
3678 				BTRFS_QGROUP_RSV_META_PERTRANS);
3679 		list_for_each_entry(glist, &qg->groups, next_group) {
3680 			ret = ulist_add(fs_info->qgroup_ulist,
3681 					glist->group->qgroupid,
3682 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3683 			if (ret < 0)
3684 				goto out;
3685 		}
3686 	}
3687 out:
3688 	spin_unlock(&fs_info->qgroup_lock);
3689 }
3690 
3691 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3692 {
3693 	struct btrfs_fs_info *fs_info = root->fs_info;
3694 
3695 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3696 	    !is_fstree(root->root_key.objectid))
3697 		return;
3698 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3699 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3700 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3701 	trace_qgroup_meta_convert(root, num_bytes);
3702 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3703 }
3704 
3705 /*
3706  * Check qgroup reserved space leaking, normally at destroy inode
3707  * time
3708  */
3709 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3710 {
3711 	struct extent_changeset changeset;
3712 	struct ulist_node *unode;
3713 	struct ulist_iterator iter;
3714 	int ret;
3715 
3716 	extent_changeset_init(&changeset);
3717 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3718 			EXTENT_QGROUP_RESERVED, &changeset);
3719 
3720 	WARN_ON(ret < 0);
3721 	if (WARN_ON(changeset.bytes_changed)) {
3722 		ULIST_ITER_INIT(&iter);
3723 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3724 			btrfs_warn(BTRFS_I(inode)->root->fs_info,
3725 				"leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3726 				inode->i_ino, unode->val, unode->aux);
3727 		}
3728 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3729 				BTRFS_I(inode)->root->root_key.objectid,
3730 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3731 
3732 	}
3733 	extent_changeset_release(&changeset);
3734 }
3735 
3736 void btrfs_qgroup_init_swapped_blocks(
3737 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3738 {
3739 	int i;
3740 
3741 	spin_lock_init(&swapped_blocks->lock);
3742 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3743 		swapped_blocks->blocks[i] = RB_ROOT;
3744 	swapped_blocks->swapped = false;
3745 }
3746 
3747 /*
3748  * Delete all swapped blocks record of @root.
3749  * Every record here means we skipped a full subtree scan for qgroup.
3750  *
3751  * Gets called when committing one transaction.
3752  */
3753 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3754 {
3755 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3756 	int i;
3757 
3758 	swapped_blocks = &root->swapped_blocks;
3759 
3760 	spin_lock(&swapped_blocks->lock);
3761 	if (!swapped_blocks->swapped)
3762 		goto out;
3763 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3764 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
3765 		struct btrfs_qgroup_swapped_block *entry;
3766 		struct btrfs_qgroup_swapped_block *next;
3767 
3768 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3769 						     node)
3770 			kfree(entry);
3771 		swapped_blocks->blocks[i] = RB_ROOT;
3772 	}
3773 	swapped_blocks->swapped = false;
3774 out:
3775 	spin_unlock(&swapped_blocks->lock);
3776 }
3777 
3778 /*
3779  * Add subtree roots record into @subvol_root.
3780  *
3781  * @subvol_root:	tree root of the subvolume tree get swapped
3782  * @bg:			block group under balance
3783  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
3784  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
3785  *			BOTH POINTERS ARE BEFORE TREE SWAP
3786  * @last_snapshot:	last snapshot generation of the subvolume tree
3787  */
3788 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3789 		struct btrfs_root *subvol_root,
3790 		struct btrfs_block_group_cache *bg,
3791 		struct extent_buffer *subvol_parent, int subvol_slot,
3792 		struct extent_buffer *reloc_parent, int reloc_slot,
3793 		u64 last_snapshot)
3794 {
3795 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3796 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3797 	struct btrfs_qgroup_swapped_block *block;
3798 	struct rb_node **cur;
3799 	struct rb_node *parent = NULL;
3800 	int level = btrfs_header_level(subvol_parent) - 1;
3801 	int ret = 0;
3802 
3803 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3804 		return 0;
3805 
3806 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3807 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3808 		btrfs_err_rl(fs_info,
3809 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3810 			__func__,
3811 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3812 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3813 		return -EUCLEAN;
3814 	}
3815 
3816 	block = kmalloc(sizeof(*block), GFP_NOFS);
3817 	if (!block) {
3818 		ret = -ENOMEM;
3819 		goto out;
3820 	}
3821 
3822 	/*
3823 	 * @reloc_parent/slot is still before swap, while @block is going to
3824 	 * record the bytenr after swap, so we do the swap here.
3825 	 */
3826 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3827 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3828 							     reloc_slot);
3829 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3830 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3831 							    subvol_slot);
3832 	block->last_snapshot = last_snapshot;
3833 	block->level = level;
3834 	if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
3835 		block->trace_leaf = true;
3836 	else
3837 		block->trace_leaf = false;
3838 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3839 
3840 	/* Insert @block into @blocks */
3841 	spin_lock(&blocks->lock);
3842 	cur = &blocks->blocks[level].rb_node;
3843 	while (*cur) {
3844 		struct btrfs_qgroup_swapped_block *entry;
3845 
3846 		parent = *cur;
3847 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3848 				 node);
3849 
3850 		if (entry->subvol_bytenr < block->subvol_bytenr) {
3851 			cur = &(*cur)->rb_left;
3852 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
3853 			cur = &(*cur)->rb_right;
3854 		} else {
3855 			if (entry->subvol_generation !=
3856 					block->subvol_generation ||
3857 			    entry->reloc_bytenr != block->reloc_bytenr ||
3858 			    entry->reloc_generation !=
3859 					block->reloc_generation) {
3860 				/*
3861 				 * Duplicated but mismatch entry found.
3862 				 * Shouldn't happen.
3863 				 *
3864 				 * Marking qgroup inconsistent should be enough
3865 				 * for end users.
3866 				 */
3867 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3868 				ret = -EEXIST;
3869 			}
3870 			kfree(block);
3871 			goto out_unlock;
3872 		}
3873 	}
3874 	rb_link_node(&block->node, parent, cur);
3875 	rb_insert_color(&block->node, &blocks->blocks[level]);
3876 	blocks->swapped = true;
3877 out_unlock:
3878 	spin_unlock(&blocks->lock);
3879 out:
3880 	if (ret < 0)
3881 		fs_info->qgroup_flags |=
3882 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3883 	return ret;
3884 }
3885 
3886 /*
3887  * Check if the tree block is a subtree root, and if so do the needed
3888  * delayed subtree trace for qgroup.
3889  *
3890  * This is called during btrfs_cow_block().
3891  */
3892 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3893 					 struct btrfs_root *root,
3894 					 struct extent_buffer *subvol_eb)
3895 {
3896 	struct btrfs_fs_info *fs_info = root->fs_info;
3897 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3898 	struct btrfs_qgroup_swapped_block *block;
3899 	struct extent_buffer *reloc_eb = NULL;
3900 	struct rb_node *node;
3901 	bool found = false;
3902 	bool swapped = false;
3903 	int level = btrfs_header_level(subvol_eb);
3904 	int ret = 0;
3905 	int i;
3906 
3907 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3908 		return 0;
3909 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3910 		return 0;
3911 
3912 	spin_lock(&blocks->lock);
3913 	if (!blocks->swapped) {
3914 		spin_unlock(&blocks->lock);
3915 		return 0;
3916 	}
3917 	node = blocks->blocks[level].rb_node;
3918 
3919 	while (node) {
3920 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3921 		if (block->subvol_bytenr < subvol_eb->start) {
3922 			node = node->rb_left;
3923 		} else if (block->subvol_bytenr > subvol_eb->start) {
3924 			node = node->rb_right;
3925 		} else {
3926 			found = true;
3927 			break;
3928 		}
3929 	}
3930 	if (!found) {
3931 		spin_unlock(&blocks->lock);
3932 		goto out;
3933 	}
3934 	/* Found one, remove it from @blocks first and update blocks->swapped */
3935 	rb_erase(&block->node, &blocks->blocks[level]);
3936 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3937 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3938 			swapped = true;
3939 			break;
3940 		}
3941 	}
3942 	blocks->swapped = swapped;
3943 	spin_unlock(&blocks->lock);
3944 
3945 	/* Read out reloc subtree root */
3946 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3947 				   block->reloc_generation, block->level,
3948 				   &block->first_key);
3949 	if (IS_ERR(reloc_eb)) {
3950 		ret = PTR_ERR(reloc_eb);
3951 		reloc_eb = NULL;
3952 		goto free_out;
3953 	}
3954 	if (!extent_buffer_uptodate(reloc_eb)) {
3955 		ret = -EIO;
3956 		goto free_out;
3957 	}
3958 
3959 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3960 			block->last_snapshot, block->trace_leaf);
3961 free_out:
3962 	kfree(block);
3963 	free_extent_buffer(reloc_eb);
3964 out:
3965 	if (ret < 0) {
3966 		btrfs_err_rl(fs_info,
3967 			     "failed to account subtree at bytenr %llu: %d",
3968 			     subvol_eb->start, ret);
3969 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3970 	}
3971 	return ret;
3972 }
3973