xref: /linux/fs/btrfs/qgroup.c (revision 3839a7460721b87501134697b7b90c45dcc7825d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sizes.h>
15 
16 #include "ctree.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "locking.h"
20 #include "ulist.h"
21 #include "backref.h"
22 #include "extent_io.h"
23 #include "qgroup.h"
24 #include "block-group.h"
25 
26 /* TODO XXX FIXME
27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
28  *  - reorganize keys
29  *  - compressed
30  *  - sync
31  *  - copy also limits on subvol creation
32  *  - limit
33  *  - caches for ulists
34  *  - performance benchmarks
35  *  - check all ioctl parameters
36  */
37 
38 /*
39  * Helpers to access qgroup reservation
40  *
41  * Callers should ensure the lock context and type are valid
42  */
43 
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45 {
46 	u64 ret = 0;
47 	int i;
48 
49 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 		ret += qgroup->rsv.values[i];
51 
52 	return ret;
53 }
54 
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57 {
58 	if (type == BTRFS_QGROUP_RSV_DATA)
59 		return "data";
60 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 		return "meta_pertrans";
62 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 		return "meta_prealloc";
64 	return NULL;
65 }
66 #endif
67 
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
70 			   enum btrfs_qgroup_rsv_type type)
71 {
72 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 	qgroup->rsv.values[type] += num_bytes;
74 }
75 
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
78 			       enum btrfs_qgroup_rsv_type type)
79 {
80 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 	if (qgroup->rsv.values[type] >= num_bytes) {
82 		qgroup->rsv.values[type] -= num_bytes;
83 		return;
84 	}
85 #ifdef CONFIG_BTRFS_DEBUG
86 	WARN_RATELIMIT(1,
87 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 		qgroup->qgroupid, qgroup_rsv_type_str(type),
89 		qgroup->rsv.values[type], num_bytes);
90 #endif
91 	qgroup->rsv.values[type] = 0;
92 }
93 
94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 				     struct btrfs_qgroup *dest,
96 				     struct btrfs_qgroup *src)
97 {
98 	int i;
99 
100 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 		qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
102 }
103 
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 					 struct btrfs_qgroup *dest,
106 					  struct btrfs_qgroup *src)
107 {
108 	int i;
109 
110 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
112 }
113 
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115 					   int mod)
116 {
117 	if (qg->old_refcnt < seq)
118 		qg->old_refcnt = seq;
119 	qg->old_refcnt += mod;
120 }
121 
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123 					   int mod)
124 {
125 	if (qg->new_refcnt < seq)
126 		qg->new_refcnt = seq;
127 	qg->new_refcnt += mod;
128 }
129 
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131 {
132 	if (qg->old_refcnt < seq)
133 		return 0;
134 	return qg->old_refcnt - seq;
135 }
136 
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138 {
139 	if (qg->new_refcnt < seq)
140 		return 0;
141 	return qg->new_refcnt - seq;
142 }
143 
144 /*
145  * glue structure to represent the relations between qgroups.
146  */
147 struct btrfs_qgroup_list {
148 	struct list_head next_group;
149 	struct list_head next_member;
150 	struct btrfs_qgroup *group;
151 	struct btrfs_qgroup *member;
152 };
153 
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 {
156 	return (u64)(uintptr_t)qg;
157 }
158 
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 {
161 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162 }
163 
164 static int
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 		   int init_flags);
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168 
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171 					   u64 qgroupid)
172 {
173 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 	struct btrfs_qgroup *qgroup;
175 
176 	while (n) {
177 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 		if (qgroup->qgroupid < qgroupid)
179 			n = n->rb_left;
180 		else if (qgroup->qgroupid > qgroupid)
181 			n = n->rb_right;
182 		else
183 			return qgroup;
184 	}
185 	return NULL;
186 }
187 
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190 					  u64 qgroupid)
191 {
192 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 	struct rb_node *parent = NULL;
194 	struct btrfs_qgroup *qgroup;
195 
196 	while (*p) {
197 		parent = *p;
198 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199 
200 		if (qgroup->qgroupid < qgroupid)
201 			p = &(*p)->rb_left;
202 		else if (qgroup->qgroupid > qgroupid)
203 			p = &(*p)->rb_right;
204 		else
205 			return qgroup;
206 	}
207 
208 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 	if (!qgroup)
210 		return ERR_PTR(-ENOMEM);
211 
212 	qgroup->qgroupid = qgroupid;
213 	INIT_LIST_HEAD(&qgroup->groups);
214 	INIT_LIST_HEAD(&qgroup->members);
215 	INIT_LIST_HEAD(&qgroup->dirty);
216 
217 	rb_link_node(&qgroup->node, parent, p);
218 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219 
220 	return qgroup;
221 }
222 
223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
224 {
225 	struct btrfs_qgroup_list *list;
226 
227 	list_del(&qgroup->dirty);
228 	while (!list_empty(&qgroup->groups)) {
229 		list = list_first_entry(&qgroup->groups,
230 					struct btrfs_qgroup_list, next_group);
231 		list_del(&list->next_group);
232 		list_del(&list->next_member);
233 		kfree(list);
234 	}
235 
236 	while (!list_empty(&qgroup->members)) {
237 		list = list_first_entry(&qgroup->members,
238 					struct btrfs_qgroup_list, next_member);
239 		list_del(&list->next_group);
240 		list_del(&list->next_member);
241 		kfree(list);
242 	}
243 	kfree(qgroup);
244 }
245 
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248 {
249 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250 
251 	if (!qgroup)
252 		return -ENOENT;
253 
254 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 	__del_qgroup_rb(qgroup);
256 	return 0;
257 }
258 
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 			   u64 memberid, u64 parentid)
262 {
263 	struct btrfs_qgroup *member;
264 	struct btrfs_qgroup *parent;
265 	struct btrfs_qgroup_list *list;
266 
267 	member = find_qgroup_rb(fs_info, memberid);
268 	parent = find_qgroup_rb(fs_info, parentid);
269 	if (!member || !parent)
270 		return -ENOENT;
271 
272 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
273 	if (!list)
274 		return -ENOMEM;
275 
276 	list->group = parent;
277 	list->member = member;
278 	list_add_tail(&list->next_group, &member->groups);
279 	list_add_tail(&list->next_member, &parent->members);
280 
281 	return 0;
282 }
283 
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 			   u64 memberid, u64 parentid)
287 {
288 	struct btrfs_qgroup *member;
289 	struct btrfs_qgroup *parent;
290 	struct btrfs_qgroup_list *list;
291 
292 	member = find_qgroup_rb(fs_info, memberid);
293 	parent = find_qgroup_rb(fs_info, parentid);
294 	if (!member || !parent)
295 		return -ENOENT;
296 
297 	list_for_each_entry(list, &member->groups, next_group) {
298 		if (list->group == parent) {
299 			list_del(&list->next_group);
300 			list_del(&list->next_member);
301 			kfree(list);
302 			return 0;
303 		}
304 	}
305 	return -ENOENT;
306 }
307 
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310 			       u64 rfer, u64 excl)
311 {
312 	struct btrfs_qgroup *qgroup;
313 
314 	qgroup = find_qgroup_rb(fs_info, qgroupid);
315 	if (!qgroup)
316 		return -EINVAL;
317 	if (qgroup->rfer != rfer || qgroup->excl != excl)
318 		return -EINVAL;
319 	return 0;
320 }
321 #endif
322 
323 /*
324  * The full config is read in one go, only called from open_ctree()
325  * It doesn't use any locking, as at this point we're still single-threaded
326  */
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328 {
329 	struct btrfs_key key;
330 	struct btrfs_key found_key;
331 	struct btrfs_root *quota_root = fs_info->quota_root;
332 	struct btrfs_path *path = NULL;
333 	struct extent_buffer *l;
334 	int slot;
335 	int ret = 0;
336 	u64 flags = 0;
337 	u64 rescan_progress = 0;
338 
339 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
340 		return 0;
341 
342 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 	if (!fs_info->qgroup_ulist) {
344 		ret = -ENOMEM;
345 		goto out;
346 	}
347 
348 	path = btrfs_alloc_path();
349 	if (!path) {
350 		ret = -ENOMEM;
351 		goto out;
352 	}
353 
354 	/* default this to quota off, in case no status key is found */
355 	fs_info->qgroup_flags = 0;
356 
357 	/*
358 	 * pass 1: read status, all qgroup infos and limits
359 	 */
360 	key.objectid = 0;
361 	key.type = 0;
362 	key.offset = 0;
363 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
364 	if (ret)
365 		goto out;
366 
367 	while (1) {
368 		struct btrfs_qgroup *qgroup;
369 
370 		slot = path->slots[0];
371 		l = path->nodes[0];
372 		btrfs_item_key_to_cpu(l, &found_key, slot);
373 
374 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 			struct btrfs_qgroup_status_item *ptr;
376 
377 			ptr = btrfs_item_ptr(l, slot,
378 					     struct btrfs_qgroup_status_item);
379 
380 			if (btrfs_qgroup_status_version(l, ptr) !=
381 			    BTRFS_QGROUP_STATUS_VERSION) {
382 				btrfs_err(fs_info,
383 				 "old qgroup version, quota disabled");
384 				goto out;
385 			}
386 			if (btrfs_qgroup_status_generation(l, ptr) !=
387 			    fs_info->generation) {
388 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 				btrfs_err(fs_info,
390 					"qgroup generation mismatch, marked as inconsistent");
391 			}
392 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 									  ptr);
394 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
395 			goto next1;
396 		}
397 
398 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
400 			goto next1;
401 
402 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 			btrfs_err(fs_info, "inconsistent qgroup config");
406 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407 		}
408 		if (!qgroup) {
409 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 			if (IS_ERR(qgroup)) {
411 				ret = PTR_ERR(qgroup);
412 				goto out;
413 			}
414 		}
415 		switch (found_key.type) {
416 		case BTRFS_QGROUP_INFO_KEY: {
417 			struct btrfs_qgroup_info_item *ptr;
418 
419 			ptr = btrfs_item_ptr(l, slot,
420 					     struct btrfs_qgroup_info_item);
421 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 			/* generation currently unused */
426 			break;
427 		}
428 		case BTRFS_QGROUP_LIMIT_KEY: {
429 			struct btrfs_qgroup_limit_item *ptr;
430 
431 			ptr = btrfs_item_ptr(l, slot,
432 					     struct btrfs_qgroup_limit_item);
433 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438 			break;
439 		}
440 		}
441 next1:
442 		ret = btrfs_next_item(quota_root, path);
443 		if (ret < 0)
444 			goto out;
445 		if (ret)
446 			break;
447 	}
448 	btrfs_release_path(path);
449 
450 	/*
451 	 * pass 2: read all qgroup relations
452 	 */
453 	key.objectid = 0;
454 	key.type = BTRFS_QGROUP_RELATION_KEY;
455 	key.offset = 0;
456 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457 	if (ret)
458 		goto out;
459 	while (1) {
460 		slot = path->slots[0];
461 		l = path->nodes[0];
462 		btrfs_item_key_to_cpu(l, &found_key, slot);
463 
464 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465 			goto next2;
466 
467 		if (found_key.objectid > found_key.offset) {
468 			/* parent <- member, not needed to build config */
469 			/* FIXME should we omit the key completely? */
470 			goto next2;
471 		}
472 
473 		ret = add_relation_rb(fs_info, found_key.objectid,
474 				      found_key.offset);
475 		if (ret == -ENOENT) {
476 			btrfs_warn(fs_info,
477 				"orphan qgroup relation 0x%llx->0x%llx",
478 				found_key.objectid, found_key.offset);
479 			ret = 0;	/* ignore the error */
480 		}
481 		if (ret)
482 			goto out;
483 next2:
484 		ret = btrfs_next_item(quota_root, path);
485 		if (ret < 0)
486 			goto out;
487 		if (ret)
488 			break;
489 	}
490 out:
491 	fs_info->qgroup_flags |= flags;
492 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495 		 ret >= 0)
496 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 	btrfs_free_path(path);
498 
499 	if (ret < 0) {
500 		ulist_free(fs_info->qgroup_ulist);
501 		fs_info->qgroup_ulist = NULL;
502 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
503 	}
504 
505 	return ret < 0 ? ret : 0;
506 }
507 
508 /*
509  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510  * first two are in single-threaded paths.And for the third one, we have set
511  * quota_root to be null with qgroup_lock held before, so it is safe to clean
512  * up the in-memory structures without qgroup_lock held.
513  */
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515 {
516 	struct rb_node *n;
517 	struct btrfs_qgroup *qgroup;
518 
519 	while ((n = rb_first(&fs_info->qgroup_tree))) {
520 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 		rb_erase(n, &fs_info->qgroup_tree);
522 		__del_qgroup_rb(qgroup);
523 	}
524 	/*
525 	 * We call btrfs_free_qgroup_config() when unmounting
526 	 * filesystem and disabling quota, so we set qgroup_ulist
527 	 * to be null here to avoid double free.
528 	 */
529 	ulist_free(fs_info->qgroup_ulist);
530 	fs_info->qgroup_ulist = NULL;
531 }
532 
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 				    u64 dst)
535 {
536 	int ret;
537 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 	struct btrfs_path *path;
539 	struct btrfs_key key;
540 
541 	path = btrfs_alloc_path();
542 	if (!path)
543 		return -ENOMEM;
544 
545 	key.objectid = src;
546 	key.type = BTRFS_QGROUP_RELATION_KEY;
547 	key.offset = dst;
548 
549 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550 
551 	btrfs_mark_buffer_dirty(path->nodes[0]);
552 
553 	btrfs_free_path(path);
554 	return ret;
555 }
556 
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 				    u64 dst)
559 {
560 	int ret;
561 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 	struct btrfs_path *path;
563 	struct btrfs_key key;
564 
565 	path = btrfs_alloc_path();
566 	if (!path)
567 		return -ENOMEM;
568 
569 	key.objectid = src;
570 	key.type = BTRFS_QGROUP_RELATION_KEY;
571 	key.offset = dst;
572 
573 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574 	if (ret < 0)
575 		goto out;
576 
577 	if (ret > 0) {
578 		ret = -ENOENT;
579 		goto out;
580 	}
581 
582 	ret = btrfs_del_item(trans, quota_root, path);
583 out:
584 	btrfs_free_path(path);
585 	return ret;
586 }
587 
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 			   struct btrfs_root *quota_root, u64 qgroupid)
590 {
591 	int ret;
592 	struct btrfs_path *path;
593 	struct btrfs_qgroup_info_item *qgroup_info;
594 	struct btrfs_qgroup_limit_item *qgroup_limit;
595 	struct extent_buffer *leaf;
596 	struct btrfs_key key;
597 
598 	if (btrfs_is_testing(quota_root->fs_info))
599 		return 0;
600 
601 	path = btrfs_alloc_path();
602 	if (!path)
603 		return -ENOMEM;
604 
605 	key.objectid = 0;
606 	key.type = BTRFS_QGROUP_INFO_KEY;
607 	key.offset = qgroupid;
608 
609 	/*
610 	 * Avoid a transaction abort by catching -EEXIST here. In that
611 	 * case, we proceed by re-initializing the existing structure
612 	 * on disk.
613 	 */
614 
615 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 				      sizeof(*qgroup_info));
617 	if (ret && ret != -EEXIST)
618 		goto out;
619 
620 	leaf = path->nodes[0];
621 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 				 struct btrfs_qgroup_info_item);
623 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628 
629 	btrfs_mark_buffer_dirty(leaf);
630 
631 	btrfs_release_path(path);
632 
633 	key.type = BTRFS_QGROUP_LIMIT_KEY;
634 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 				      sizeof(*qgroup_limit));
636 	if (ret && ret != -EEXIST)
637 		goto out;
638 
639 	leaf = path->nodes[0];
640 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 				  struct btrfs_qgroup_limit_item);
642 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647 
648 	btrfs_mark_buffer_dirty(leaf);
649 
650 	ret = 0;
651 out:
652 	btrfs_free_path(path);
653 	return ret;
654 }
655 
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
657 {
658 	int ret;
659 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 	struct btrfs_path *path;
661 	struct btrfs_key key;
662 
663 	path = btrfs_alloc_path();
664 	if (!path)
665 		return -ENOMEM;
666 
667 	key.objectid = 0;
668 	key.type = BTRFS_QGROUP_INFO_KEY;
669 	key.offset = qgroupid;
670 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 	if (ret < 0)
672 		goto out;
673 
674 	if (ret > 0) {
675 		ret = -ENOENT;
676 		goto out;
677 	}
678 
679 	ret = btrfs_del_item(trans, quota_root, path);
680 	if (ret)
681 		goto out;
682 
683 	btrfs_release_path(path);
684 
685 	key.type = BTRFS_QGROUP_LIMIT_KEY;
686 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687 	if (ret < 0)
688 		goto out;
689 
690 	if (ret > 0) {
691 		ret = -ENOENT;
692 		goto out;
693 	}
694 
695 	ret = btrfs_del_item(trans, quota_root, path);
696 
697 out:
698 	btrfs_free_path(path);
699 	return ret;
700 }
701 
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 				    struct btrfs_qgroup *qgroup)
704 {
705 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 	struct btrfs_path *path;
707 	struct btrfs_key key;
708 	struct extent_buffer *l;
709 	struct btrfs_qgroup_limit_item *qgroup_limit;
710 	int ret;
711 	int slot;
712 
713 	key.objectid = 0;
714 	key.type = BTRFS_QGROUP_LIMIT_KEY;
715 	key.offset = qgroup->qgroupid;
716 
717 	path = btrfs_alloc_path();
718 	if (!path)
719 		return -ENOMEM;
720 
721 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
722 	if (ret > 0)
723 		ret = -ENOENT;
724 
725 	if (ret)
726 		goto out;
727 
728 	l = path->nodes[0];
729 	slot = path->slots[0];
730 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736 
737 	btrfs_mark_buffer_dirty(l);
738 
739 out:
740 	btrfs_free_path(path);
741 	return ret;
742 }
743 
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 				   struct btrfs_qgroup *qgroup)
746 {
747 	struct btrfs_fs_info *fs_info = trans->fs_info;
748 	struct btrfs_root *quota_root = fs_info->quota_root;
749 	struct btrfs_path *path;
750 	struct btrfs_key key;
751 	struct extent_buffer *l;
752 	struct btrfs_qgroup_info_item *qgroup_info;
753 	int ret;
754 	int slot;
755 
756 	if (btrfs_is_testing(fs_info))
757 		return 0;
758 
759 	key.objectid = 0;
760 	key.type = BTRFS_QGROUP_INFO_KEY;
761 	key.offset = qgroup->qgroupid;
762 
763 	path = btrfs_alloc_path();
764 	if (!path)
765 		return -ENOMEM;
766 
767 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
768 	if (ret > 0)
769 		ret = -ENOENT;
770 
771 	if (ret)
772 		goto out;
773 
774 	l = path->nodes[0];
775 	slot = path->slots[0];
776 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782 
783 	btrfs_mark_buffer_dirty(l);
784 
785 out:
786 	btrfs_free_path(path);
787 	return ret;
788 }
789 
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
791 {
792 	struct btrfs_fs_info *fs_info = trans->fs_info;
793 	struct btrfs_root *quota_root = fs_info->quota_root;
794 	struct btrfs_path *path;
795 	struct btrfs_key key;
796 	struct extent_buffer *l;
797 	struct btrfs_qgroup_status_item *ptr;
798 	int ret;
799 	int slot;
800 
801 	key.objectid = 0;
802 	key.type = BTRFS_QGROUP_STATUS_KEY;
803 	key.offset = 0;
804 
805 	path = btrfs_alloc_path();
806 	if (!path)
807 		return -ENOMEM;
808 
809 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
810 	if (ret > 0)
811 		ret = -ENOENT;
812 
813 	if (ret)
814 		goto out;
815 
816 	l = path->nodes[0];
817 	slot = path->slots[0];
818 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 	btrfs_set_qgroup_status_rescan(l, ptr,
822 				fs_info->qgroup_rescan_progress.objectid);
823 
824 	btrfs_mark_buffer_dirty(l);
825 
826 out:
827 	btrfs_free_path(path);
828 	return ret;
829 }
830 
831 /*
832  * called with qgroup_lock held
833  */
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 				  struct btrfs_root *root)
836 {
837 	struct btrfs_path *path;
838 	struct btrfs_key key;
839 	struct extent_buffer *leaf = NULL;
840 	int ret;
841 	int nr = 0;
842 
843 	path = btrfs_alloc_path();
844 	if (!path)
845 		return -ENOMEM;
846 
847 	path->leave_spinning = 1;
848 
849 	key.objectid = 0;
850 	key.offset = 0;
851 	key.type = 0;
852 
853 	while (1) {
854 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855 		if (ret < 0)
856 			goto out;
857 		leaf = path->nodes[0];
858 		nr = btrfs_header_nritems(leaf);
859 		if (!nr)
860 			break;
861 		/*
862 		 * delete the leaf one by one
863 		 * since the whole tree is going
864 		 * to be deleted.
865 		 */
866 		path->slots[0] = 0;
867 		ret = btrfs_del_items(trans, root, path, 0, nr);
868 		if (ret)
869 			goto out;
870 
871 		btrfs_release_path(path);
872 	}
873 	ret = 0;
874 out:
875 	btrfs_free_path(path);
876 	return ret;
877 }
878 
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
880 {
881 	struct btrfs_root *quota_root;
882 	struct btrfs_root *tree_root = fs_info->tree_root;
883 	struct btrfs_path *path = NULL;
884 	struct btrfs_qgroup_status_item *ptr;
885 	struct extent_buffer *leaf;
886 	struct btrfs_key key;
887 	struct btrfs_key found_key;
888 	struct btrfs_qgroup *qgroup = NULL;
889 	struct btrfs_trans_handle *trans = NULL;
890 	int ret = 0;
891 	int slot;
892 
893 	mutex_lock(&fs_info->qgroup_ioctl_lock);
894 	if (fs_info->quota_root)
895 		goto out;
896 
897 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 	if (!fs_info->qgroup_ulist) {
899 		ret = -ENOMEM;
900 		goto out;
901 	}
902 
903 	/*
904 	 * 1 for quota root item
905 	 * 1 for BTRFS_QGROUP_STATUS item
906 	 *
907 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
908 	 * per subvolume. However those are not currently reserved since it
909 	 * would be a lot of overkill.
910 	 */
911 	trans = btrfs_start_transaction(tree_root, 2);
912 	if (IS_ERR(trans)) {
913 		ret = PTR_ERR(trans);
914 		trans = NULL;
915 		goto out;
916 	}
917 
918 	/*
919 	 * initially create the quota tree
920 	 */
921 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
922 	if (IS_ERR(quota_root)) {
923 		ret =  PTR_ERR(quota_root);
924 		btrfs_abort_transaction(trans, ret);
925 		goto out;
926 	}
927 
928 	path = btrfs_alloc_path();
929 	if (!path) {
930 		ret = -ENOMEM;
931 		btrfs_abort_transaction(trans, ret);
932 		goto out_free_root;
933 	}
934 
935 	key.objectid = 0;
936 	key.type = BTRFS_QGROUP_STATUS_KEY;
937 	key.offset = 0;
938 
939 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
940 				      sizeof(*ptr));
941 	if (ret) {
942 		btrfs_abort_transaction(trans, ret);
943 		goto out_free_path;
944 	}
945 
946 	leaf = path->nodes[0];
947 	ptr = btrfs_item_ptr(leaf, path->slots[0],
948 				 struct btrfs_qgroup_status_item);
949 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
950 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
951 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
952 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
953 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
954 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
955 
956 	btrfs_mark_buffer_dirty(leaf);
957 
958 	key.objectid = 0;
959 	key.type = BTRFS_ROOT_REF_KEY;
960 	key.offset = 0;
961 
962 	btrfs_release_path(path);
963 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
964 	if (ret > 0)
965 		goto out_add_root;
966 	if (ret < 0) {
967 		btrfs_abort_transaction(trans, ret);
968 		goto out_free_path;
969 	}
970 
971 	while (1) {
972 		slot = path->slots[0];
973 		leaf = path->nodes[0];
974 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
975 
976 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
977 			ret = add_qgroup_item(trans, quota_root,
978 					      found_key.offset);
979 			if (ret) {
980 				btrfs_abort_transaction(trans, ret);
981 				goto out_free_path;
982 			}
983 
984 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 			if (IS_ERR(qgroup)) {
986 				ret = PTR_ERR(qgroup);
987 				btrfs_abort_transaction(trans, ret);
988 				goto out_free_path;
989 			}
990 		}
991 		ret = btrfs_next_item(tree_root, path);
992 		if (ret < 0) {
993 			btrfs_abort_transaction(trans, ret);
994 			goto out_free_path;
995 		}
996 		if (ret)
997 			break;
998 	}
999 
1000 out_add_root:
1001 	btrfs_release_path(path);
1002 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1003 	if (ret) {
1004 		btrfs_abort_transaction(trans, ret);
1005 		goto out_free_path;
1006 	}
1007 
1008 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 	if (IS_ERR(qgroup)) {
1010 		ret = PTR_ERR(qgroup);
1011 		btrfs_abort_transaction(trans, ret);
1012 		goto out_free_path;
1013 	}
1014 
1015 	ret = btrfs_commit_transaction(trans);
1016 	trans = NULL;
1017 	if (ret)
1018 		goto out_free_path;
1019 
1020 	/*
1021 	 * Set quota enabled flag after committing the transaction, to avoid
1022 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1023 	 * creation.
1024 	 */
1025 	spin_lock(&fs_info->qgroup_lock);
1026 	fs_info->quota_root = quota_root;
1027 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1028 	spin_unlock(&fs_info->qgroup_lock);
1029 
1030 	ret = qgroup_rescan_init(fs_info, 0, 1);
1031 	if (!ret) {
1032 	        qgroup_rescan_zero_tracking(fs_info);
1033 		fs_info->qgroup_rescan_running = true;
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 		btrfs_put_root(quota_root);
1043 out:
1044 	if (ret) {
1045 		ulist_free(fs_info->qgroup_ulist);
1046 		fs_info->qgroup_ulist = NULL;
1047 		if (trans)
1048 			btrfs_end_transaction(trans);
1049 	}
1050 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1051 	return ret;
1052 }
1053 
1054 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1055 {
1056 	struct btrfs_root *quota_root;
1057 	struct btrfs_trans_handle *trans = NULL;
1058 	int ret = 0;
1059 
1060 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1061 	if (!fs_info->quota_root)
1062 		goto out;
1063 
1064 	/*
1065 	 * 1 For the root item
1066 	 *
1067 	 * We should also reserve enough items for the quota tree deletion in
1068 	 * btrfs_clean_quota_tree but this is not done.
1069 	 */
1070 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1071 	if (IS_ERR(trans)) {
1072 		ret = PTR_ERR(trans);
1073 		goto out;
1074 	}
1075 
1076 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1077 	btrfs_qgroup_wait_for_completion(fs_info, false);
1078 	spin_lock(&fs_info->qgroup_lock);
1079 	quota_root = fs_info->quota_root;
1080 	fs_info->quota_root = NULL;
1081 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1082 	spin_unlock(&fs_info->qgroup_lock);
1083 
1084 	btrfs_free_qgroup_config(fs_info);
1085 
1086 	ret = btrfs_clean_quota_tree(trans, quota_root);
1087 	if (ret) {
1088 		btrfs_abort_transaction(trans, ret);
1089 		goto end_trans;
1090 	}
1091 
1092 	ret = btrfs_del_root(trans, &quota_root->root_key);
1093 	if (ret) {
1094 		btrfs_abort_transaction(trans, ret);
1095 		goto end_trans;
1096 	}
1097 
1098 	list_del(&quota_root->dirty_list);
1099 
1100 	btrfs_tree_lock(quota_root->node);
1101 	btrfs_clean_tree_block(quota_root->node);
1102 	btrfs_tree_unlock(quota_root->node);
1103 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1104 
1105 	btrfs_put_root(quota_root);
1106 
1107 end_trans:
1108 	ret = btrfs_end_transaction(trans);
1109 out:
1110 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1111 	return ret;
1112 }
1113 
1114 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1115 			 struct btrfs_qgroup *qgroup)
1116 {
1117 	if (list_empty(&qgroup->dirty))
1118 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1119 }
1120 
1121 /*
1122  * The easy accounting, we're updating qgroup relationship whose child qgroup
1123  * only has exclusive extents.
1124  *
1125  * In this case, all exclusive extents will also be exclusive for parent, so
1126  * excl/rfer just get added/removed.
1127  *
1128  * So is qgroup reservation space, which should also be added/removed to
1129  * parent.
1130  * Or when child tries to release reservation space, parent will underflow its
1131  * reservation (for relationship adding case).
1132  *
1133  * Caller should hold fs_info->qgroup_lock.
1134  */
1135 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1136 				    struct ulist *tmp, u64 ref_root,
1137 				    struct btrfs_qgroup *src, int sign)
1138 {
1139 	struct btrfs_qgroup *qgroup;
1140 	struct btrfs_qgroup_list *glist;
1141 	struct ulist_node *unode;
1142 	struct ulist_iterator uiter;
1143 	u64 num_bytes = src->excl;
1144 	int ret = 0;
1145 
1146 	qgroup = find_qgroup_rb(fs_info, ref_root);
1147 	if (!qgroup)
1148 		goto out;
1149 
1150 	qgroup->rfer += sign * num_bytes;
1151 	qgroup->rfer_cmpr += sign * num_bytes;
1152 
1153 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1154 	qgroup->excl += sign * num_bytes;
1155 	qgroup->excl_cmpr += sign * num_bytes;
1156 
1157 	if (sign > 0)
1158 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1159 	else
1160 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1161 
1162 	qgroup_dirty(fs_info, qgroup);
1163 
1164 	/* Get all of the parent groups that contain this qgroup */
1165 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1166 		ret = ulist_add(tmp, glist->group->qgroupid,
1167 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1168 		if (ret < 0)
1169 			goto out;
1170 	}
1171 
1172 	/* Iterate all of the parents and adjust their reference counts */
1173 	ULIST_ITER_INIT(&uiter);
1174 	while ((unode = ulist_next(tmp, &uiter))) {
1175 		qgroup = unode_aux_to_qgroup(unode);
1176 		qgroup->rfer += sign * num_bytes;
1177 		qgroup->rfer_cmpr += sign * num_bytes;
1178 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1179 		qgroup->excl += sign * num_bytes;
1180 		if (sign > 0)
1181 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1182 		else
1183 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1184 		qgroup->excl_cmpr += sign * num_bytes;
1185 		qgroup_dirty(fs_info, qgroup);
1186 
1187 		/* Add any parents of the parents */
1188 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1189 			ret = ulist_add(tmp, glist->group->qgroupid,
1190 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1191 			if (ret < 0)
1192 				goto out;
1193 		}
1194 	}
1195 	ret = 0;
1196 out:
1197 	return ret;
1198 }
1199 
1200 
1201 /*
1202  * Quick path for updating qgroup with only excl refs.
1203  *
1204  * In that case, just update all parent will be enough.
1205  * Or we needs to do a full rescan.
1206  * Caller should also hold fs_info->qgroup_lock.
1207  *
1208  * Return 0 for quick update, return >0 for need to full rescan
1209  * and mark INCONSISTENT flag.
1210  * Return < 0 for other error.
1211  */
1212 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1213 				   struct ulist *tmp, u64 src, u64 dst,
1214 				   int sign)
1215 {
1216 	struct btrfs_qgroup *qgroup;
1217 	int ret = 1;
1218 	int err = 0;
1219 
1220 	qgroup = find_qgroup_rb(fs_info, src);
1221 	if (!qgroup)
1222 		goto out;
1223 	if (qgroup->excl == qgroup->rfer) {
1224 		ret = 0;
1225 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1226 					       qgroup, sign);
1227 		if (err < 0) {
1228 			ret = err;
1229 			goto out;
1230 		}
1231 	}
1232 out:
1233 	if (ret)
1234 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1235 	return ret;
1236 }
1237 
1238 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1239 			      u64 dst)
1240 {
1241 	struct btrfs_fs_info *fs_info = trans->fs_info;
1242 	struct btrfs_qgroup *parent;
1243 	struct btrfs_qgroup *member;
1244 	struct btrfs_qgroup_list *list;
1245 	struct ulist *tmp;
1246 	int ret = 0;
1247 
1248 	/* Check the level of src and dst first */
1249 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1250 		return -EINVAL;
1251 
1252 	tmp = ulist_alloc(GFP_KERNEL);
1253 	if (!tmp)
1254 		return -ENOMEM;
1255 
1256 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1257 	if (!fs_info->quota_root) {
1258 		ret = -ENOTCONN;
1259 		goto out;
1260 	}
1261 	member = find_qgroup_rb(fs_info, src);
1262 	parent = find_qgroup_rb(fs_info, dst);
1263 	if (!member || !parent) {
1264 		ret = -EINVAL;
1265 		goto out;
1266 	}
1267 
1268 	/* check if such qgroup relation exist firstly */
1269 	list_for_each_entry(list, &member->groups, next_group) {
1270 		if (list->group == parent) {
1271 			ret = -EEXIST;
1272 			goto out;
1273 		}
1274 	}
1275 
1276 	ret = add_qgroup_relation_item(trans, src, dst);
1277 	if (ret)
1278 		goto out;
1279 
1280 	ret = add_qgroup_relation_item(trans, dst, src);
1281 	if (ret) {
1282 		del_qgroup_relation_item(trans, src, dst);
1283 		goto out;
1284 	}
1285 
1286 	spin_lock(&fs_info->qgroup_lock);
1287 	ret = add_relation_rb(fs_info, src, dst);
1288 	if (ret < 0) {
1289 		spin_unlock(&fs_info->qgroup_lock);
1290 		goto out;
1291 	}
1292 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1293 	spin_unlock(&fs_info->qgroup_lock);
1294 out:
1295 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1296 	ulist_free(tmp);
1297 	return ret;
1298 }
1299 
1300 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1301 				 u64 dst)
1302 {
1303 	struct btrfs_fs_info *fs_info = trans->fs_info;
1304 	struct btrfs_qgroup *parent;
1305 	struct btrfs_qgroup *member;
1306 	struct btrfs_qgroup_list *list;
1307 	struct ulist *tmp;
1308 	bool found = false;
1309 	int ret = 0;
1310 	int ret2;
1311 
1312 	tmp = ulist_alloc(GFP_KERNEL);
1313 	if (!tmp)
1314 		return -ENOMEM;
1315 
1316 	if (!fs_info->quota_root) {
1317 		ret = -ENOTCONN;
1318 		goto out;
1319 	}
1320 
1321 	member = find_qgroup_rb(fs_info, src);
1322 	parent = find_qgroup_rb(fs_info, dst);
1323 	/*
1324 	 * The parent/member pair doesn't exist, then try to delete the dead
1325 	 * relation items only.
1326 	 */
1327 	if (!member || !parent)
1328 		goto delete_item;
1329 
1330 	/* check if such qgroup relation exist firstly */
1331 	list_for_each_entry(list, &member->groups, next_group) {
1332 		if (list->group == parent) {
1333 			found = true;
1334 			break;
1335 		}
1336 	}
1337 
1338 delete_item:
1339 	ret = del_qgroup_relation_item(trans, src, dst);
1340 	if (ret < 0 && ret != -ENOENT)
1341 		goto out;
1342 	ret2 = del_qgroup_relation_item(trans, dst, src);
1343 	if (ret2 < 0 && ret2 != -ENOENT)
1344 		goto out;
1345 
1346 	/* At least one deletion succeeded, return 0 */
1347 	if (!ret || !ret2)
1348 		ret = 0;
1349 
1350 	if (found) {
1351 		spin_lock(&fs_info->qgroup_lock);
1352 		del_relation_rb(fs_info, src, dst);
1353 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1354 		spin_unlock(&fs_info->qgroup_lock);
1355 	}
1356 out:
1357 	ulist_free(tmp);
1358 	return ret;
1359 }
1360 
1361 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1362 			      u64 dst)
1363 {
1364 	struct btrfs_fs_info *fs_info = trans->fs_info;
1365 	int ret = 0;
1366 
1367 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1368 	ret = __del_qgroup_relation(trans, src, dst);
1369 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1370 
1371 	return ret;
1372 }
1373 
1374 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1375 {
1376 	struct btrfs_fs_info *fs_info = trans->fs_info;
1377 	struct btrfs_root *quota_root;
1378 	struct btrfs_qgroup *qgroup;
1379 	int ret = 0;
1380 
1381 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1382 	if (!fs_info->quota_root) {
1383 		ret = -ENOTCONN;
1384 		goto out;
1385 	}
1386 	quota_root = fs_info->quota_root;
1387 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1388 	if (qgroup) {
1389 		ret = -EEXIST;
1390 		goto out;
1391 	}
1392 
1393 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1394 	if (ret)
1395 		goto out;
1396 
1397 	spin_lock(&fs_info->qgroup_lock);
1398 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1399 	spin_unlock(&fs_info->qgroup_lock);
1400 
1401 	if (IS_ERR(qgroup))
1402 		ret = PTR_ERR(qgroup);
1403 out:
1404 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1405 	return ret;
1406 }
1407 
1408 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1409 {
1410 	struct btrfs_fs_info *fs_info = trans->fs_info;
1411 	struct btrfs_qgroup *qgroup;
1412 	struct btrfs_qgroup_list *list;
1413 	int ret = 0;
1414 
1415 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1416 	if (!fs_info->quota_root) {
1417 		ret = -ENOTCONN;
1418 		goto out;
1419 	}
1420 
1421 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1422 	if (!qgroup) {
1423 		ret = -ENOENT;
1424 		goto out;
1425 	}
1426 
1427 	/* Check if there are no children of this qgroup */
1428 	if (!list_empty(&qgroup->members)) {
1429 		ret = -EBUSY;
1430 		goto out;
1431 	}
1432 
1433 	ret = del_qgroup_item(trans, qgroupid);
1434 	if (ret && ret != -ENOENT)
1435 		goto out;
1436 
1437 	while (!list_empty(&qgroup->groups)) {
1438 		list = list_first_entry(&qgroup->groups,
1439 					struct btrfs_qgroup_list, next_group);
1440 		ret = __del_qgroup_relation(trans, qgroupid,
1441 					    list->group->qgroupid);
1442 		if (ret)
1443 			goto out;
1444 	}
1445 
1446 	spin_lock(&fs_info->qgroup_lock);
1447 	del_qgroup_rb(fs_info, qgroupid);
1448 	spin_unlock(&fs_info->qgroup_lock);
1449 out:
1450 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1451 	return ret;
1452 }
1453 
1454 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1455 		       struct btrfs_qgroup_limit *limit)
1456 {
1457 	struct btrfs_fs_info *fs_info = trans->fs_info;
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 	if (!fs_info->quota_root) {
1468 		ret = -ENOTCONN;
1469 		goto out;
1470 	}
1471 
1472 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1473 	if (!qgroup) {
1474 		ret = -ENOENT;
1475 		goto out;
1476 	}
1477 
1478 	spin_lock(&fs_info->qgroup_lock);
1479 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1480 		if (limit->max_rfer == CLEAR_VALUE) {
1481 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1482 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1483 			qgroup->max_rfer = 0;
1484 		} else {
1485 			qgroup->max_rfer = limit->max_rfer;
1486 		}
1487 	}
1488 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1489 		if (limit->max_excl == CLEAR_VALUE) {
1490 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1491 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1492 			qgroup->max_excl = 0;
1493 		} else {
1494 			qgroup->max_excl = limit->max_excl;
1495 		}
1496 	}
1497 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1498 		if (limit->rsv_rfer == CLEAR_VALUE) {
1499 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1500 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1501 			qgroup->rsv_rfer = 0;
1502 		} else {
1503 			qgroup->rsv_rfer = limit->rsv_rfer;
1504 		}
1505 	}
1506 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1507 		if (limit->rsv_excl == CLEAR_VALUE) {
1508 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1509 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1510 			qgroup->rsv_excl = 0;
1511 		} else {
1512 			qgroup->rsv_excl = limit->rsv_excl;
1513 		}
1514 	}
1515 	qgroup->lim_flags |= limit->flags;
1516 
1517 	spin_unlock(&fs_info->qgroup_lock);
1518 
1519 	ret = update_qgroup_limit_item(trans, qgroup);
1520 	if (ret) {
1521 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1522 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1523 		       qgroupid);
1524 	}
1525 
1526 out:
1527 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1528 	return ret;
1529 }
1530 
1531 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1532 				struct btrfs_delayed_ref_root *delayed_refs,
1533 				struct btrfs_qgroup_extent_record *record)
1534 {
1535 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1536 	struct rb_node *parent_node = NULL;
1537 	struct btrfs_qgroup_extent_record *entry;
1538 	u64 bytenr = record->bytenr;
1539 
1540 	lockdep_assert_held(&delayed_refs->lock);
1541 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1542 
1543 	while (*p) {
1544 		parent_node = *p;
1545 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1546 				 node);
1547 		if (bytenr < entry->bytenr) {
1548 			p = &(*p)->rb_left;
1549 		} else if (bytenr > entry->bytenr) {
1550 			p = &(*p)->rb_right;
1551 		} else {
1552 			if (record->data_rsv && !entry->data_rsv) {
1553 				entry->data_rsv = record->data_rsv;
1554 				entry->data_rsv_refroot =
1555 					record->data_rsv_refroot;
1556 			}
1557 			return 1;
1558 		}
1559 	}
1560 
1561 	rb_link_node(&record->node, parent_node, p);
1562 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1563 	return 0;
1564 }
1565 
1566 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1567 				   struct btrfs_qgroup_extent_record *qrecord)
1568 {
1569 	struct ulist *old_root;
1570 	u64 bytenr = qrecord->bytenr;
1571 	int ret;
1572 
1573 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1574 	if (ret < 0) {
1575 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1576 		btrfs_warn(fs_info,
1577 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1578 			ret);
1579 		return 0;
1580 	}
1581 
1582 	/*
1583 	 * Here we don't need to get the lock of
1584 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1585 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1586 	 *
1587 	 * So modifying qrecord->old_roots is safe here
1588 	 */
1589 	qrecord->old_roots = old_root;
1590 	return 0;
1591 }
1592 
1593 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1594 			      u64 num_bytes, gfp_t gfp_flag)
1595 {
1596 	struct btrfs_fs_info *fs_info = trans->fs_info;
1597 	struct btrfs_qgroup_extent_record *record;
1598 	struct btrfs_delayed_ref_root *delayed_refs;
1599 	int ret;
1600 
1601 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1602 	    || bytenr == 0 || num_bytes == 0)
1603 		return 0;
1604 	record = kzalloc(sizeof(*record), gfp_flag);
1605 	if (!record)
1606 		return -ENOMEM;
1607 
1608 	delayed_refs = &trans->transaction->delayed_refs;
1609 	record->bytenr = bytenr;
1610 	record->num_bytes = num_bytes;
1611 	record->old_roots = NULL;
1612 
1613 	spin_lock(&delayed_refs->lock);
1614 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1615 	spin_unlock(&delayed_refs->lock);
1616 	if (ret > 0) {
1617 		kfree(record);
1618 		return 0;
1619 	}
1620 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1621 }
1622 
1623 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1624 				  struct extent_buffer *eb)
1625 {
1626 	struct btrfs_fs_info *fs_info = trans->fs_info;
1627 	int nr = btrfs_header_nritems(eb);
1628 	int i, extent_type, ret;
1629 	struct btrfs_key key;
1630 	struct btrfs_file_extent_item *fi;
1631 	u64 bytenr, num_bytes;
1632 
1633 	/* We can be called directly from walk_up_proc() */
1634 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1635 		return 0;
1636 
1637 	for (i = 0; i < nr; i++) {
1638 		btrfs_item_key_to_cpu(eb, &key, i);
1639 
1640 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1641 			continue;
1642 
1643 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1644 		/* filter out non qgroup-accountable extents  */
1645 		extent_type = btrfs_file_extent_type(eb, fi);
1646 
1647 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1648 			continue;
1649 
1650 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1651 		if (!bytenr)
1652 			continue;
1653 
1654 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1655 
1656 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1657 						GFP_NOFS);
1658 		if (ret)
1659 			return ret;
1660 	}
1661 	cond_resched();
1662 	return 0;
1663 }
1664 
1665 /*
1666  * Walk up the tree from the bottom, freeing leaves and any interior
1667  * nodes which have had all slots visited. If a node (leaf or
1668  * interior) is freed, the node above it will have it's slot
1669  * incremented. The root node will never be freed.
1670  *
1671  * At the end of this function, we should have a path which has all
1672  * slots incremented to the next position for a search. If we need to
1673  * read a new node it will be NULL and the node above it will have the
1674  * correct slot selected for a later read.
1675  *
1676  * If we increment the root nodes slot counter past the number of
1677  * elements, 1 is returned to signal completion of the search.
1678  */
1679 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1680 {
1681 	int level = 0;
1682 	int nr, slot;
1683 	struct extent_buffer *eb;
1684 
1685 	if (root_level == 0)
1686 		return 1;
1687 
1688 	while (level <= root_level) {
1689 		eb = path->nodes[level];
1690 		nr = btrfs_header_nritems(eb);
1691 		path->slots[level]++;
1692 		slot = path->slots[level];
1693 		if (slot >= nr || level == 0) {
1694 			/*
1695 			 * Don't free the root -  we will detect this
1696 			 * condition after our loop and return a
1697 			 * positive value for caller to stop walking the tree.
1698 			 */
1699 			if (level != root_level) {
1700 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1701 				path->locks[level] = 0;
1702 
1703 				free_extent_buffer(eb);
1704 				path->nodes[level] = NULL;
1705 				path->slots[level] = 0;
1706 			}
1707 		} else {
1708 			/*
1709 			 * We have a valid slot to walk back down
1710 			 * from. Stop here so caller can process these
1711 			 * new nodes.
1712 			 */
1713 			break;
1714 		}
1715 
1716 		level++;
1717 	}
1718 
1719 	eb = path->nodes[root_level];
1720 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1721 		return 1;
1722 
1723 	return 0;
1724 }
1725 
1726 /*
1727  * Helper function to trace a subtree tree block swap.
1728  *
1729  * The swap will happen in highest tree block, but there may be a lot of
1730  * tree blocks involved.
1731  *
1732  * For example:
1733  *  OO = Old tree blocks
1734  *  NN = New tree blocks allocated during balance
1735  *
1736  *           File tree (257)                  Reloc tree for 257
1737  * L2              OO                                NN
1738  *               /    \                            /    \
1739  * L1          OO      OO (a)                    OO      NN (a)
1740  *            / \     / \                       / \     / \
1741  * L0       OO   OO OO   OO                   OO   OO NN   NN
1742  *                  (b)  (c)                          (b)  (c)
1743  *
1744  * When calling qgroup_trace_extent_swap(), we will pass:
1745  * @src_eb = OO(a)
1746  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1747  * @dst_level = 0
1748  * @root_level = 1
1749  *
1750  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1751  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1752  *
1753  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1754  *
1755  * 1) Tree search from @src_eb
1756  *    It should acts as a simplified btrfs_search_slot().
1757  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1758  *    (first key).
1759  *
1760  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1761  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1762  *    They should be marked during previous (@dst_level = 1) iteration.
1763  *
1764  * 3) Mark file extents in leaves dirty
1765  *    We don't have good way to pick out new file extents only.
1766  *    So we still follow the old method by scanning all file extents in
1767  *    the leave.
1768  *
1769  * This function can free us from keeping two paths, thus later we only need
1770  * to care about how to iterate all new tree blocks in reloc tree.
1771  */
1772 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1773 				    struct extent_buffer *src_eb,
1774 				    struct btrfs_path *dst_path,
1775 				    int dst_level, int root_level,
1776 				    bool trace_leaf)
1777 {
1778 	struct btrfs_key key;
1779 	struct btrfs_path *src_path;
1780 	struct btrfs_fs_info *fs_info = trans->fs_info;
1781 	u32 nodesize = fs_info->nodesize;
1782 	int cur_level = root_level;
1783 	int ret;
1784 
1785 	BUG_ON(dst_level > root_level);
1786 	/* Level mismatch */
1787 	if (btrfs_header_level(src_eb) != root_level)
1788 		return -EINVAL;
1789 
1790 	src_path = btrfs_alloc_path();
1791 	if (!src_path) {
1792 		ret = -ENOMEM;
1793 		goto out;
1794 	}
1795 
1796 	if (dst_level)
1797 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1798 	else
1799 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1800 
1801 	/* For src_path */
1802 	atomic_inc(&src_eb->refs);
1803 	src_path->nodes[root_level] = src_eb;
1804 	src_path->slots[root_level] = dst_path->slots[root_level];
1805 	src_path->locks[root_level] = 0;
1806 
1807 	/* A simplified version of btrfs_search_slot() */
1808 	while (cur_level >= dst_level) {
1809 		struct btrfs_key src_key;
1810 		struct btrfs_key dst_key;
1811 
1812 		if (src_path->nodes[cur_level] == NULL) {
1813 			struct btrfs_key first_key;
1814 			struct extent_buffer *eb;
1815 			int parent_slot;
1816 			u64 child_gen;
1817 			u64 child_bytenr;
1818 
1819 			eb = src_path->nodes[cur_level + 1];
1820 			parent_slot = src_path->slots[cur_level + 1];
1821 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1822 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1823 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1824 
1825 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1826 					     cur_level, &first_key);
1827 			if (IS_ERR(eb)) {
1828 				ret = PTR_ERR(eb);
1829 				goto out;
1830 			} else if (!extent_buffer_uptodate(eb)) {
1831 				free_extent_buffer(eb);
1832 				ret = -EIO;
1833 				goto out;
1834 			}
1835 
1836 			src_path->nodes[cur_level] = eb;
1837 
1838 			btrfs_tree_read_lock(eb);
1839 			btrfs_set_lock_blocking_read(eb);
1840 			src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1841 		}
1842 
1843 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1844 		if (cur_level) {
1845 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1846 					&dst_key, dst_path->slots[cur_level]);
1847 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1848 					&src_key, src_path->slots[cur_level]);
1849 		} else {
1850 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1851 					&dst_key, dst_path->slots[cur_level]);
1852 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1853 					&src_key, src_path->slots[cur_level]);
1854 		}
1855 		/* Content mismatch, something went wrong */
1856 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1857 			ret = -ENOENT;
1858 			goto out;
1859 		}
1860 		cur_level--;
1861 	}
1862 
1863 	/*
1864 	 * Now both @dst_path and @src_path have been populated, record the tree
1865 	 * blocks for qgroup accounting.
1866 	 */
1867 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1868 			nodesize, GFP_NOFS);
1869 	if (ret < 0)
1870 		goto out;
1871 	ret = btrfs_qgroup_trace_extent(trans,
1872 			dst_path->nodes[dst_level]->start,
1873 			nodesize, GFP_NOFS);
1874 	if (ret < 0)
1875 		goto out;
1876 
1877 	/* Record leaf file extents */
1878 	if (dst_level == 0 && trace_leaf) {
1879 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1880 		if (ret < 0)
1881 			goto out;
1882 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1883 	}
1884 out:
1885 	btrfs_free_path(src_path);
1886 	return ret;
1887 }
1888 
1889 /*
1890  * Helper function to do recursive generation-aware depth-first search, to
1891  * locate all new tree blocks in a subtree of reloc tree.
1892  *
1893  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1894  *         reloc tree
1895  * L2         NN (a)
1896  *          /    \
1897  * L1    OO        NN (b)
1898  *      /  \      /  \
1899  * L0  OO  OO    OO  NN
1900  *               (c) (d)
1901  * If we pass:
1902  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1903  * @cur_level = 1
1904  * @root_level = 1
1905  *
1906  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1907  * above tree blocks along with their counter parts in file tree.
1908  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1909  * won't affect OO(c).
1910  */
1911 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1912 					   struct extent_buffer *src_eb,
1913 					   struct btrfs_path *dst_path,
1914 					   int cur_level, int root_level,
1915 					   u64 last_snapshot, bool trace_leaf)
1916 {
1917 	struct btrfs_fs_info *fs_info = trans->fs_info;
1918 	struct extent_buffer *eb;
1919 	bool need_cleanup = false;
1920 	int ret = 0;
1921 	int i;
1922 
1923 	/* Level sanity check */
1924 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1925 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1926 	    root_level < cur_level) {
1927 		btrfs_err_rl(fs_info,
1928 			"%s: bad levels, cur_level=%d root_level=%d",
1929 			__func__, cur_level, root_level);
1930 		return -EUCLEAN;
1931 	}
1932 
1933 	/* Read the tree block if needed */
1934 	if (dst_path->nodes[cur_level] == NULL) {
1935 		struct btrfs_key first_key;
1936 		int parent_slot;
1937 		u64 child_gen;
1938 		u64 child_bytenr;
1939 
1940 		/*
1941 		 * dst_path->nodes[root_level] must be initialized before
1942 		 * calling this function.
1943 		 */
1944 		if (cur_level == root_level) {
1945 			btrfs_err_rl(fs_info,
1946 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1947 				__func__, root_level, root_level, cur_level);
1948 			return -EUCLEAN;
1949 		}
1950 
1951 		/*
1952 		 * We need to get child blockptr/gen from parent before we can
1953 		 * read it.
1954 		  */
1955 		eb = dst_path->nodes[cur_level + 1];
1956 		parent_slot = dst_path->slots[cur_level + 1];
1957 		child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1958 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1959 		btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1960 
1961 		/* This node is old, no need to trace */
1962 		if (child_gen < last_snapshot)
1963 			goto out;
1964 
1965 		eb = read_tree_block(fs_info, child_bytenr, child_gen,
1966 				     cur_level, &first_key);
1967 		if (IS_ERR(eb)) {
1968 			ret = PTR_ERR(eb);
1969 			goto out;
1970 		} else if (!extent_buffer_uptodate(eb)) {
1971 			free_extent_buffer(eb);
1972 			ret = -EIO;
1973 			goto out;
1974 		}
1975 
1976 		dst_path->nodes[cur_level] = eb;
1977 		dst_path->slots[cur_level] = 0;
1978 
1979 		btrfs_tree_read_lock(eb);
1980 		btrfs_set_lock_blocking_read(eb);
1981 		dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1982 		need_cleanup = true;
1983 	}
1984 
1985 	/* Now record this tree block and its counter part for qgroups */
1986 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1987 				       root_level, trace_leaf);
1988 	if (ret < 0)
1989 		goto cleanup;
1990 
1991 	eb = dst_path->nodes[cur_level];
1992 
1993 	if (cur_level > 0) {
1994 		/* Iterate all child tree blocks */
1995 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
1996 			/* Skip old tree blocks as they won't be swapped */
1997 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1998 				continue;
1999 			dst_path->slots[cur_level] = i;
2000 
2001 			/* Recursive call (at most 7 times) */
2002 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2003 					dst_path, cur_level - 1, root_level,
2004 					last_snapshot, trace_leaf);
2005 			if (ret < 0)
2006 				goto cleanup;
2007 		}
2008 	}
2009 
2010 cleanup:
2011 	if (need_cleanup) {
2012 		/* Clean up */
2013 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2014 				     dst_path->locks[cur_level]);
2015 		free_extent_buffer(dst_path->nodes[cur_level]);
2016 		dst_path->nodes[cur_level] = NULL;
2017 		dst_path->slots[cur_level] = 0;
2018 		dst_path->locks[cur_level] = 0;
2019 	}
2020 out:
2021 	return ret;
2022 }
2023 
2024 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2025 				struct extent_buffer *src_eb,
2026 				struct extent_buffer *dst_eb,
2027 				u64 last_snapshot, bool trace_leaf)
2028 {
2029 	struct btrfs_fs_info *fs_info = trans->fs_info;
2030 	struct btrfs_path *dst_path = NULL;
2031 	int level;
2032 	int ret;
2033 
2034 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2035 		return 0;
2036 
2037 	/* Wrong parameter order */
2038 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2039 		btrfs_err_rl(fs_info,
2040 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2041 			     btrfs_header_generation(src_eb),
2042 			     btrfs_header_generation(dst_eb));
2043 		return -EUCLEAN;
2044 	}
2045 
2046 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2047 		ret = -EIO;
2048 		goto out;
2049 	}
2050 
2051 	level = btrfs_header_level(dst_eb);
2052 	dst_path = btrfs_alloc_path();
2053 	if (!dst_path) {
2054 		ret = -ENOMEM;
2055 		goto out;
2056 	}
2057 	/* For dst_path */
2058 	atomic_inc(&dst_eb->refs);
2059 	dst_path->nodes[level] = dst_eb;
2060 	dst_path->slots[level] = 0;
2061 	dst_path->locks[level] = 0;
2062 
2063 	/* Do the generation aware breadth-first search */
2064 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2065 					      level, last_snapshot, trace_leaf);
2066 	if (ret < 0)
2067 		goto out;
2068 	ret = 0;
2069 
2070 out:
2071 	btrfs_free_path(dst_path);
2072 	if (ret < 0)
2073 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2074 	return ret;
2075 }
2076 
2077 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2078 			       struct extent_buffer *root_eb,
2079 			       u64 root_gen, int root_level)
2080 {
2081 	struct btrfs_fs_info *fs_info = trans->fs_info;
2082 	int ret = 0;
2083 	int level;
2084 	struct extent_buffer *eb = root_eb;
2085 	struct btrfs_path *path = NULL;
2086 
2087 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2088 	BUG_ON(root_eb == NULL);
2089 
2090 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2091 		return 0;
2092 
2093 	if (!extent_buffer_uptodate(root_eb)) {
2094 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2095 		if (ret)
2096 			goto out;
2097 	}
2098 
2099 	if (root_level == 0) {
2100 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2101 		goto out;
2102 	}
2103 
2104 	path = btrfs_alloc_path();
2105 	if (!path)
2106 		return -ENOMEM;
2107 
2108 	/*
2109 	 * Walk down the tree.  Missing extent blocks are filled in as
2110 	 * we go. Metadata is accounted every time we read a new
2111 	 * extent block.
2112 	 *
2113 	 * When we reach a leaf, we account for file extent items in it,
2114 	 * walk back up the tree (adjusting slot pointers as we go)
2115 	 * and restart the search process.
2116 	 */
2117 	atomic_inc(&root_eb->refs);	/* For path */
2118 	path->nodes[root_level] = root_eb;
2119 	path->slots[root_level] = 0;
2120 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2121 walk_down:
2122 	level = root_level;
2123 	while (level >= 0) {
2124 		if (path->nodes[level] == NULL) {
2125 			struct btrfs_key first_key;
2126 			int parent_slot;
2127 			u64 child_gen;
2128 			u64 child_bytenr;
2129 
2130 			/*
2131 			 * We need to get child blockptr/gen from parent before
2132 			 * we can read it.
2133 			  */
2134 			eb = path->nodes[level + 1];
2135 			parent_slot = path->slots[level + 1];
2136 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2137 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2138 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2139 
2140 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
2141 					     level, &first_key);
2142 			if (IS_ERR(eb)) {
2143 				ret = PTR_ERR(eb);
2144 				goto out;
2145 			} else if (!extent_buffer_uptodate(eb)) {
2146 				free_extent_buffer(eb);
2147 				ret = -EIO;
2148 				goto out;
2149 			}
2150 
2151 			path->nodes[level] = eb;
2152 			path->slots[level] = 0;
2153 
2154 			btrfs_tree_read_lock(eb);
2155 			btrfs_set_lock_blocking_read(eb);
2156 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2157 
2158 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2159 							fs_info->nodesize,
2160 							GFP_NOFS);
2161 			if (ret)
2162 				goto out;
2163 		}
2164 
2165 		if (level == 0) {
2166 			ret = btrfs_qgroup_trace_leaf_items(trans,
2167 							    path->nodes[level]);
2168 			if (ret)
2169 				goto out;
2170 
2171 			/* Nonzero return here means we completed our search */
2172 			ret = adjust_slots_upwards(path, root_level);
2173 			if (ret)
2174 				break;
2175 
2176 			/* Restart search with new slots */
2177 			goto walk_down;
2178 		}
2179 
2180 		level--;
2181 	}
2182 
2183 	ret = 0;
2184 out:
2185 	btrfs_free_path(path);
2186 
2187 	return ret;
2188 }
2189 
2190 #define UPDATE_NEW	0
2191 #define UPDATE_OLD	1
2192 /*
2193  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2194  */
2195 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2196 				struct ulist *roots, struct ulist *tmp,
2197 				struct ulist *qgroups, u64 seq, int update_old)
2198 {
2199 	struct ulist_node *unode;
2200 	struct ulist_iterator uiter;
2201 	struct ulist_node *tmp_unode;
2202 	struct ulist_iterator tmp_uiter;
2203 	struct btrfs_qgroup *qg;
2204 	int ret = 0;
2205 
2206 	if (!roots)
2207 		return 0;
2208 	ULIST_ITER_INIT(&uiter);
2209 	while ((unode = ulist_next(roots, &uiter))) {
2210 		qg = find_qgroup_rb(fs_info, unode->val);
2211 		if (!qg)
2212 			continue;
2213 
2214 		ulist_reinit(tmp);
2215 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2216 				GFP_ATOMIC);
2217 		if (ret < 0)
2218 			return ret;
2219 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2220 		if (ret < 0)
2221 			return ret;
2222 		ULIST_ITER_INIT(&tmp_uiter);
2223 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2224 			struct btrfs_qgroup_list *glist;
2225 
2226 			qg = unode_aux_to_qgroup(tmp_unode);
2227 			if (update_old)
2228 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2229 			else
2230 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2231 			list_for_each_entry(glist, &qg->groups, next_group) {
2232 				ret = ulist_add(qgroups, glist->group->qgroupid,
2233 						qgroup_to_aux(glist->group),
2234 						GFP_ATOMIC);
2235 				if (ret < 0)
2236 					return ret;
2237 				ret = ulist_add(tmp, glist->group->qgroupid,
2238 						qgroup_to_aux(glist->group),
2239 						GFP_ATOMIC);
2240 				if (ret < 0)
2241 					return ret;
2242 			}
2243 		}
2244 	}
2245 	return 0;
2246 }
2247 
2248 /*
2249  * Update qgroup rfer/excl counters.
2250  * Rfer update is easy, codes can explain themselves.
2251  *
2252  * Excl update is tricky, the update is split into 2 part.
2253  * Part 1: Possible exclusive <-> sharing detect:
2254  *	|	A	|	!A	|
2255  *  -------------------------------------
2256  *  B	|	*	|	-	|
2257  *  -------------------------------------
2258  *  !B	|	+	|	**	|
2259  *  -------------------------------------
2260  *
2261  * Conditions:
2262  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2263  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2264  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2265  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2266  *
2267  * Results:
2268  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2269  * *: Definitely not changed.		**: Possible unchanged.
2270  *
2271  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2272  *
2273  * To make the logic clear, we first use condition A and B to split
2274  * combination into 4 results.
2275  *
2276  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2277  * only on variant maybe 0.
2278  *
2279  * Lastly, check result **, since there are 2 variants maybe 0, split them
2280  * again(2x2).
2281  * But this time we don't need to consider other things, the codes and logic
2282  * is easy to understand now.
2283  */
2284 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2285 				  struct ulist *qgroups,
2286 				  u64 nr_old_roots,
2287 				  u64 nr_new_roots,
2288 				  u64 num_bytes, u64 seq)
2289 {
2290 	struct ulist_node *unode;
2291 	struct ulist_iterator uiter;
2292 	struct btrfs_qgroup *qg;
2293 	u64 cur_new_count, cur_old_count;
2294 
2295 	ULIST_ITER_INIT(&uiter);
2296 	while ((unode = ulist_next(qgroups, &uiter))) {
2297 		bool dirty = false;
2298 
2299 		qg = unode_aux_to_qgroup(unode);
2300 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2301 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2302 
2303 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2304 					     cur_new_count);
2305 
2306 		/* Rfer update part */
2307 		if (cur_old_count == 0 && cur_new_count > 0) {
2308 			qg->rfer += num_bytes;
2309 			qg->rfer_cmpr += num_bytes;
2310 			dirty = true;
2311 		}
2312 		if (cur_old_count > 0 && cur_new_count == 0) {
2313 			qg->rfer -= num_bytes;
2314 			qg->rfer_cmpr -= num_bytes;
2315 			dirty = true;
2316 		}
2317 
2318 		/* Excl update part */
2319 		/* Exclusive/none -> shared case */
2320 		if (cur_old_count == nr_old_roots &&
2321 		    cur_new_count < nr_new_roots) {
2322 			/* Exclusive -> shared */
2323 			if (cur_old_count != 0) {
2324 				qg->excl -= num_bytes;
2325 				qg->excl_cmpr -= num_bytes;
2326 				dirty = true;
2327 			}
2328 		}
2329 
2330 		/* Shared -> exclusive/none case */
2331 		if (cur_old_count < nr_old_roots &&
2332 		    cur_new_count == nr_new_roots) {
2333 			/* Shared->exclusive */
2334 			if (cur_new_count != 0) {
2335 				qg->excl += num_bytes;
2336 				qg->excl_cmpr += num_bytes;
2337 				dirty = true;
2338 			}
2339 		}
2340 
2341 		/* Exclusive/none -> exclusive/none case */
2342 		if (cur_old_count == nr_old_roots &&
2343 		    cur_new_count == nr_new_roots) {
2344 			if (cur_old_count == 0) {
2345 				/* None -> exclusive/none */
2346 
2347 				if (cur_new_count != 0) {
2348 					/* None -> exclusive */
2349 					qg->excl += num_bytes;
2350 					qg->excl_cmpr += num_bytes;
2351 					dirty = true;
2352 				}
2353 				/* None -> none, nothing changed */
2354 			} else {
2355 				/* Exclusive -> exclusive/none */
2356 
2357 				if (cur_new_count == 0) {
2358 					/* Exclusive -> none */
2359 					qg->excl -= num_bytes;
2360 					qg->excl_cmpr -= num_bytes;
2361 					dirty = true;
2362 				}
2363 				/* Exclusive -> exclusive, nothing changed */
2364 			}
2365 		}
2366 
2367 		if (dirty)
2368 			qgroup_dirty(fs_info, qg);
2369 	}
2370 	return 0;
2371 }
2372 
2373 /*
2374  * Check if the @roots potentially is a list of fs tree roots
2375  *
2376  * Return 0 for definitely not a fs/subvol tree roots ulist
2377  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2378  *          one as well)
2379  */
2380 static int maybe_fs_roots(struct ulist *roots)
2381 {
2382 	struct ulist_node *unode;
2383 	struct ulist_iterator uiter;
2384 
2385 	/* Empty one, still possible for fs roots */
2386 	if (!roots || roots->nnodes == 0)
2387 		return 1;
2388 
2389 	ULIST_ITER_INIT(&uiter);
2390 	unode = ulist_next(roots, &uiter);
2391 	if (!unode)
2392 		return 1;
2393 
2394 	/*
2395 	 * If it contains fs tree roots, then it must belong to fs/subvol
2396 	 * trees.
2397 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2398 	 */
2399 	return is_fstree(unode->val);
2400 }
2401 
2402 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2403 				u64 num_bytes, struct ulist *old_roots,
2404 				struct ulist *new_roots)
2405 {
2406 	struct btrfs_fs_info *fs_info = trans->fs_info;
2407 	struct ulist *qgroups = NULL;
2408 	struct ulist *tmp = NULL;
2409 	u64 seq;
2410 	u64 nr_new_roots = 0;
2411 	u64 nr_old_roots = 0;
2412 	int ret = 0;
2413 
2414 	/*
2415 	 * If quotas get disabled meanwhile, the resouces need to be freed and
2416 	 * we can't just exit here.
2417 	 */
2418 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2419 		goto out_free;
2420 
2421 	if (new_roots) {
2422 		if (!maybe_fs_roots(new_roots))
2423 			goto out_free;
2424 		nr_new_roots = new_roots->nnodes;
2425 	}
2426 	if (old_roots) {
2427 		if (!maybe_fs_roots(old_roots))
2428 			goto out_free;
2429 		nr_old_roots = old_roots->nnodes;
2430 	}
2431 
2432 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2433 	if (nr_old_roots == 0 && nr_new_roots == 0)
2434 		goto out_free;
2435 
2436 	BUG_ON(!fs_info->quota_root);
2437 
2438 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2439 					num_bytes, nr_old_roots, nr_new_roots);
2440 
2441 	qgroups = ulist_alloc(GFP_NOFS);
2442 	if (!qgroups) {
2443 		ret = -ENOMEM;
2444 		goto out_free;
2445 	}
2446 	tmp = ulist_alloc(GFP_NOFS);
2447 	if (!tmp) {
2448 		ret = -ENOMEM;
2449 		goto out_free;
2450 	}
2451 
2452 	mutex_lock(&fs_info->qgroup_rescan_lock);
2453 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2454 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2455 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2456 			ret = 0;
2457 			goto out_free;
2458 		}
2459 	}
2460 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2461 
2462 	spin_lock(&fs_info->qgroup_lock);
2463 	seq = fs_info->qgroup_seq;
2464 
2465 	/* Update old refcnts using old_roots */
2466 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2467 				   UPDATE_OLD);
2468 	if (ret < 0)
2469 		goto out;
2470 
2471 	/* Update new refcnts using new_roots */
2472 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2473 				   UPDATE_NEW);
2474 	if (ret < 0)
2475 		goto out;
2476 
2477 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2478 			       num_bytes, seq);
2479 
2480 	/*
2481 	 * Bump qgroup_seq to avoid seq overlap
2482 	 */
2483 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2484 out:
2485 	spin_unlock(&fs_info->qgroup_lock);
2486 out_free:
2487 	ulist_free(tmp);
2488 	ulist_free(qgroups);
2489 	ulist_free(old_roots);
2490 	ulist_free(new_roots);
2491 	return ret;
2492 }
2493 
2494 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2495 {
2496 	struct btrfs_fs_info *fs_info = trans->fs_info;
2497 	struct btrfs_qgroup_extent_record *record;
2498 	struct btrfs_delayed_ref_root *delayed_refs;
2499 	struct ulist *new_roots = NULL;
2500 	struct rb_node *node;
2501 	u64 num_dirty_extents = 0;
2502 	u64 qgroup_to_skip;
2503 	int ret = 0;
2504 
2505 	delayed_refs = &trans->transaction->delayed_refs;
2506 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2507 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2508 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2509 				  node);
2510 
2511 		num_dirty_extents++;
2512 		trace_btrfs_qgroup_account_extents(fs_info, record);
2513 
2514 		if (!ret) {
2515 			/*
2516 			 * Old roots should be searched when inserting qgroup
2517 			 * extent record
2518 			 */
2519 			if (WARN_ON(!record->old_roots)) {
2520 				/* Search commit root to find old_roots */
2521 				ret = btrfs_find_all_roots(NULL, fs_info,
2522 						record->bytenr, 0,
2523 						&record->old_roots, false);
2524 				if (ret < 0)
2525 					goto cleanup;
2526 			}
2527 
2528 			/* Free the reserved data space */
2529 			btrfs_qgroup_free_refroot(fs_info,
2530 					record->data_rsv_refroot,
2531 					record->data_rsv,
2532 					BTRFS_QGROUP_RSV_DATA);
2533 			/*
2534 			 * Use SEQ_LAST as time_seq to do special search, which
2535 			 * doesn't lock tree or delayed_refs and search current
2536 			 * root. It's safe inside commit_transaction().
2537 			 */
2538 			ret = btrfs_find_all_roots(trans, fs_info,
2539 				record->bytenr, SEQ_LAST, &new_roots, false);
2540 			if (ret < 0)
2541 				goto cleanup;
2542 			if (qgroup_to_skip) {
2543 				ulist_del(new_roots, qgroup_to_skip, 0);
2544 				ulist_del(record->old_roots, qgroup_to_skip,
2545 					  0);
2546 			}
2547 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2548 							  record->num_bytes,
2549 							  record->old_roots,
2550 							  new_roots);
2551 			record->old_roots = NULL;
2552 			new_roots = NULL;
2553 		}
2554 cleanup:
2555 		ulist_free(record->old_roots);
2556 		ulist_free(new_roots);
2557 		new_roots = NULL;
2558 		rb_erase(node, &delayed_refs->dirty_extent_root);
2559 		kfree(record);
2560 
2561 	}
2562 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2563 				       num_dirty_extents);
2564 	return ret;
2565 }
2566 
2567 /*
2568  * called from commit_transaction. Writes all changed qgroups to disk.
2569  */
2570 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2571 {
2572 	struct btrfs_fs_info *fs_info = trans->fs_info;
2573 	int ret = 0;
2574 
2575 	if (!fs_info->quota_root)
2576 		return ret;
2577 
2578 	spin_lock(&fs_info->qgroup_lock);
2579 	while (!list_empty(&fs_info->dirty_qgroups)) {
2580 		struct btrfs_qgroup *qgroup;
2581 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2582 					  struct btrfs_qgroup, dirty);
2583 		list_del_init(&qgroup->dirty);
2584 		spin_unlock(&fs_info->qgroup_lock);
2585 		ret = update_qgroup_info_item(trans, qgroup);
2586 		if (ret)
2587 			fs_info->qgroup_flags |=
2588 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2589 		ret = update_qgroup_limit_item(trans, qgroup);
2590 		if (ret)
2591 			fs_info->qgroup_flags |=
2592 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2593 		spin_lock(&fs_info->qgroup_lock);
2594 	}
2595 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2596 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2597 	else
2598 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2599 	spin_unlock(&fs_info->qgroup_lock);
2600 
2601 	ret = update_qgroup_status_item(trans);
2602 	if (ret)
2603 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2604 
2605 	return ret;
2606 }
2607 
2608 /*
2609  * Copy the accounting information between qgroups. This is necessary
2610  * when a snapshot or a subvolume is created. Throwing an error will
2611  * cause a transaction abort so we take extra care here to only error
2612  * when a readonly fs is a reasonable outcome.
2613  */
2614 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2615 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2616 {
2617 	int ret = 0;
2618 	int i;
2619 	u64 *i_qgroups;
2620 	bool committing = false;
2621 	struct btrfs_fs_info *fs_info = trans->fs_info;
2622 	struct btrfs_root *quota_root;
2623 	struct btrfs_qgroup *srcgroup;
2624 	struct btrfs_qgroup *dstgroup;
2625 	bool need_rescan = false;
2626 	u32 level_size = 0;
2627 	u64 nums;
2628 
2629 	/*
2630 	 * There are only two callers of this function.
2631 	 *
2632 	 * One in create_subvol() in the ioctl context, which needs to hold
2633 	 * the qgroup_ioctl_lock.
2634 	 *
2635 	 * The other one in create_pending_snapshot() where no other qgroup
2636 	 * code can modify the fs as they all need to either start a new trans
2637 	 * or hold a trans handler, thus we don't need to hold
2638 	 * qgroup_ioctl_lock.
2639 	 * This would avoid long and complex lock chain and make lockdep happy.
2640 	 */
2641 	spin_lock(&fs_info->trans_lock);
2642 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2643 		committing = true;
2644 	spin_unlock(&fs_info->trans_lock);
2645 
2646 	if (!committing)
2647 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2648 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2649 		goto out;
2650 
2651 	quota_root = fs_info->quota_root;
2652 	if (!quota_root) {
2653 		ret = -EINVAL;
2654 		goto out;
2655 	}
2656 
2657 	if (inherit) {
2658 		i_qgroups = (u64 *)(inherit + 1);
2659 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2660 		       2 * inherit->num_excl_copies;
2661 		for (i = 0; i < nums; ++i) {
2662 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2663 
2664 			/*
2665 			 * Zero out invalid groups so we can ignore
2666 			 * them later.
2667 			 */
2668 			if (!srcgroup ||
2669 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2670 				*i_qgroups = 0ULL;
2671 
2672 			++i_qgroups;
2673 		}
2674 	}
2675 
2676 	/*
2677 	 * create a tracking group for the subvol itself
2678 	 */
2679 	ret = add_qgroup_item(trans, quota_root, objectid);
2680 	if (ret)
2681 		goto out;
2682 
2683 	/*
2684 	 * add qgroup to all inherited groups
2685 	 */
2686 	if (inherit) {
2687 		i_qgroups = (u64 *)(inherit + 1);
2688 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2689 			if (*i_qgroups == 0)
2690 				continue;
2691 			ret = add_qgroup_relation_item(trans, objectid,
2692 						       *i_qgroups);
2693 			if (ret && ret != -EEXIST)
2694 				goto out;
2695 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2696 						       objectid);
2697 			if (ret && ret != -EEXIST)
2698 				goto out;
2699 		}
2700 		ret = 0;
2701 	}
2702 
2703 
2704 	spin_lock(&fs_info->qgroup_lock);
2705 
2706 	dstgroup = add_qgroup_rb(fs_info, objectid);
2707 	if (IS_ERR(dstgroup)) {
2708 		ret = PTR_ERR(dstgroup);
2709 		goto unlock;
2710 	}
2711 
2712 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2713 		dstgroup->lim_flags = inherit->lim.flags;
2714 		dstgroup->max_rfer = inherit->lim.max_rfer;
2715 		dstgroup->max_excl = inherit->lim.max_excl;
2716 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2717 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2718 
2719 		ret = update_qgroup_limit_item(trans, dstgroup);
2720 		if (ret) {
2721 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2722 			btrfs_info(fs_info,
2723 				   "unable to update quota limit for %llu",
2724 				   dstgroup->qgroupid);
2725 			goto unlock;
2726 		}
2727 	}
2728 
2729 	if (srcid) {
2730 		srcgroup = find_qgroup_rb(fs_info, srcid);
2731 		if (!srcgroup)
2732 			goto unlock;
2733 
2734 		/*
2735 		 * We call inherit after we clone the root in order to make sure
2736 		 * our counts don't go crazy, so at this point the only
2737 		 * difference between the two roots should be the root node.
2738 		 */
2739 		level_size = fs_info->nodesize;
2740 		dstgroup->rfer = srcgroup->rfer;
2741 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2742 		dstgroup->excl = level_size;
2743 		dstgroup->excl_cmpr = level_size;
2744 		srcgroup->excl = level_size;
2745 		srcgroup->excl_cmpr = level_size;
2746 
2747 		/* inherit the limit info */
2748 		dstgroup->lim_flags = srcgroup->lim_flags;
2749 		dstgroup->max_rfer = srcgroup->max_rfer;
2750 		dstgroup->max_excl = srcgroup->max_excl;
2751 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2752 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2753 
2754 		qgroup_dirty(fs_info, dstgroup);
2755 		qgroup_dirty(fs_info, srcgroup);
2756 	}
2757 
2758 	if (!inherit)
2759 		goto unlock;
2760 
2761 	i_qgroups = (u64 *)(inherit + 1);
2762 	for (i = 0; i < inherit->num_qgroups; ++i) {
2763 		if (*i_qgroups) {
2764 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2765 			if (ret)
2766 				goto unlock;
2767 		}
2768 		++i_qgroups;
2769 
2770 		/*
2771 		 * If we're doing a snapshot, and adding the snapshot to a new
2772 		 * qgroup, the numbers are guaranteed to be incorrect.
2773 		 */
2774 		if (srcid)
2775 			need_rescan = true;
2776 	}
2777 
2778 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2779 		struct btrfs_qgroup *src;
2780 		struct btrfs_qgroup *dst;
2781 
2782 		if (!i_qgroups[0] || !i_qgroups[1])
2783 			continue;
2784 
2785 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2786 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2787 
2788 		if (!src || !dst) {
2789 			ret = -EINVAL;
2790 			goto unlock;
2791 		}
2792 
2793 		dst->rfer = src->rfer - level_size;
2794 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2795 
2796 		/* Manually tweaking numbers certainly needs a rescan */
2797 		need_rescan = true;
2798 	}
2799 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2800 		struct btrfs_qgroup *src;
2801 		struct btrfs_qgroup *dst;
2802 
2803 		if (!i_qgroups[0] || !i_qgroups[1])
2804 			continue;
2805 
2806 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2807 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2808 
2809 		if (!src || !dst) {
2810 			ret = -EINVAL;
2811 			goto unlock;
2812 		}
2813 
2814 		dst->excl = src->excl + level_size;
2815 		dst->excl_cmpr = src->excl_cmpr + level_size;
2816 		need_rescan = true;
2817 	}
2818 
2819 unlock:
2820 	spin_unlock(&fs_info->qgroup_lock);
2821 out:
2822 	if (!committing)
2823 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
2824 	if (need_rescan)
2825 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2826 	return ret;
2827 }
2828 
2829 /*
2830  * Two limits to commit transaction in advance.
2831  *
2832  * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2833  * For SIZE, it will be in byte unit as threshold.
2834  */
2835 #define QGROUP_FREE_RATIO		32
2836 #define QGROUP_FREE_SIZE		SZ_32M
2837 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2838 				const struct btrfs_qgroup *qg, u64 num_bytes)
2839 {
2840 	u64 free;
2841 	u64 threshold;
2842 
2843 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2844 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2845 		return false;
2846 
2847 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2848 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2849 		return false;
2850 
2851 	/*
2852 	 * Even if we passed the check, it's better to check if reservation
2853 	 * for meta_pertrans is pushing us near limit.
2854 	 * If there is too much pertrans reservation or it's near the limit,
2855 	 * let's try commit transaction to free some, using transaction_kthread
2856 	 */
2857 	if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2858 			      BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2859 		if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2860 			free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2861 			threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2862 					  QGROUP_FREE_SIZE);
2863 		} else {
2864 			free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2865 			threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2866 					  QGROUP_FREE_SIZE);
2867 		}
2868 
2869 		/*
2870 		 * Use transaction_kthread to commit transaction, so we no
2871 		 * longer need to bother nested transaction nor lock context.
2872 		 */
2873 		if (free < threshold)
2874 			btrfs_commit_transaction_locksafe(fs_info);
2875 	}
2876 
2877 	return true;
2878 }
2879 
2880 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2881 			  enum btrfs_qgroup_rsv_type type)
2882 {
2883 	struct btrfs_qgroup *qgroup;
2884 	struct btrfs_fs_info *fs_info = root->fs_info;
2885 	u64 ref_root = root->root_key.objectid;
2886 	int ret = 0;
2887 	struct ulist_node *unode;
2888 	struct ulist_iterator uiter;
2889 
2890 	if (!is_fstree(ref_root))
2891 		return 0;
2892 
2893 	if (num_bytes == 0)
2894 		return 0;
2895 
2896 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2897 	    capable(CAP_SYS_RESOURCE))
2898 		enforce = false;
2899 
2900 	spin_lock(&fs_info->qgroup_lock);
2901 	if (!fs_info->quota_root)
2902 		goto out;
2903 
2904 	qgroup = find_qgroup_rb(fs_info, ref_root);
2905 	if (!qgroup)
2906 		goto out;
2907 
2908 	/*
2909 	 * in a first step, we check all affected qgroups if any limits would
2910 	 * be exceeded
2911 	 */
2912 	ulist_reinit(fs_info->qgroup_ulist);
2913 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2914 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2915 	if (ret < 0)
2916 		goto out;
2917 	ULIST_ITER_INIT(&uiter);
2918 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2919 		struct btrfs_qgroup *qg;
2920 		struct btrfs_qgroup_list *glist;
2921 
2922 		qg = unode_aux_to_qgroup(unode);
2923 
2924 		if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2925 			ret = -EDQUOT;
2926 			goto out;
2927 		}
2928 
2929 		list_for_each_entry(glist, &qg->groups, next_group) {
2930 			ret = ulist_add(fs_info->qgroup_ulist,
2931 					glist->group->qgroupid,
2932 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2933 			if (ret < 0)
2934 				goto out;
2935 		}
2936 	}
2937 	ret = 0;
2938 	/*
2939 	 * no limits exceeded, now record the reservation into all qgroups
2940 	 */
2941 	ULIST_ITER_INIT(&uiter);
2942 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2943 		struct btrfs_qgroup *qg;
2944 
2945 		qg = unode_aux_to_qgroup(unode);
2946 
2947 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
2948 	}
2949 
2950 out:
2951 	spin_unlock(&fs_info->qgroup_lock);
2952 	return ret;
2953 }
2954 
2955 /*
2956  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2957  * qgroup).
2958  *
2959  * Will handle all higher level qgroup too.
2960  *
2961  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2962  * This special case is only used for META_PERTRANS type.
2963  */
2964 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2965 			       u64 ref_root, u64 num_bytes,
2966 			       enum btrfs_qgroup_rsv_type type)
2967 {
2968 	struct btrfs_qgroup *qgroup;
2969 	struct ulist_node *unode;
2970 	struct ulist_iterator uiter;
2971 	int ret = 0;
2972 
2973 	if (!is_fstree(ref_root))
2974 		return;
2975 
2976 	if (num_bytes == 0)
2977 		return;
2978 
2979 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2980 		WARN(1, "%s: Invalid type to free", __func__);
2981 		return;
2982 	}
2983 	spin_lock(&fs_info->qgroup_lock);
2984 
2985 	if (!fs_info->quota_root)
2986 		goto out;
2987 
2988 	qgroup = find_qgroup_rb(fs_info, ref_root);
2989 	if (!qgroup)
2990 		goto out;
2991 
2992 	if (num_bytes == (u64)-1)
2993 		/*
2994 		 * We're freeing all pertrans rsv, get reserved value from
2995 		 * level 0 qgroup as real num_bytes to free.
2996 		 */
2997 		num_bytes = qgroup->rsv.values[type];
2998 
2999 	ulist_reinit(fs_info->qgroup_ulist);
3000 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3001 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3002 	if (ret < 0)
3003 		goto out;
3004 	ULIST_ITER_INIT(&uiter);
3005 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3006 		struct btrfs_qgroup *qg;
3007 		struct btrfs_qgroup_list *glist;
3008 
3009 		qg = unode_aux_to_qgroup(unode);
3010 
3011 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3012 
3013 		list_for_each_entry(glist, &qg->groups, next_group) {
3014 			ret = ulist_add(fs_info->qgroup_ulist,
3015 					glist->group->qgroupid,
3016 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3017 			if (ret < 0)
3018 				goto out;
3019 		}
3020 	}
3021 
3022 out:
3023 	spin_unlock(&fs_info->qgroup_lock);
3024 }
3025 
3026 /*
3027  * Check if the leaf is the last leaf. Which means all node pointers
3028  * are at their last position.
3029  */
3030 static bool is_last_leaf(struct btrfs_path *path)
3031 {
3032 	int i;
3033 
3034 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3035 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3036 			return false;
3037 	}
3038 	return true;
3039 }
3040 
3041 /*
3042  * returns < 0 on error, 0 when more leafs are to be scanned.
3043  * returns 1 when done.
3044  */
3045 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3046 			      struct btrfs_path *path)
3047 {
3048 	struct btrfs_fs_info *fs_info = trans->fs_info;
3049 	struct btrfs_key found;
3050 	struct extent_buffer *scratch_leaf = NULL;
3051 	struct ulist *roots = NULL;
3052 	u64 num_bytes;
3053 	bool done;
3054 	int slot;
3055 	int ret;
3056 
3057 	mutex_lock(&fs_info->qgroup_rescan_lock);
3058 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3059 					 &fs_info->qgroup_rescan_progress,
3060 					 path, 1, 0);
3061 
3062 	btrfs_debug(fs_info,
3063 		"current progress key (%llu %u %llu), search_slot ret %d",
3064 		fs_info->qgroup_rescan_progress.objectid,
3065 		fs_info->qgroup_rescan_progress.type,
3066 		fs_info->qgroup_rescan_progress.offset, ret);
3067 
3068 	if (ret) {
3069 		/*
3070 		 * The rescan is about to end, we will not be scanning any
3071 		 * further blocks. We cannot unset the RESCAN flag here, because
3072 		 * we want to commit the transaction if everything went well.
3073 		 * To make the live accounting work in this phase, we set our
3074 		 * scan progress pointer such that every real extent objectid
3075 		 * will be smaller.
3076 		 */
3077 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3078 		btrfs_release_path(path);
3079 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3080 		return ret;
3081 	}
3082 	done = is_last_leaf(path);
3083 
3084 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3085 			      btrfs_header_nritems(path->nodes[0]) - 1);
3086 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3087 
3088 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3089 	if (!scratch_leaf) {
3090 		ret = -ENOMEM;
3091 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3092 		goto out;
3093 	}
3094 	slot = path->slots[0];
3095 	btrfs_release_path(path);
3096 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3097 
3098 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3099 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3100 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3101 		    found.type != BTRFS_METADATA_ITEM_KEY)
3102 			continue;
3103 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3104 			num_bytes = fs_info->nodesize;
3105 		else
3106 			num_bytes = found.offset;
3107 
3108 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3109 					   &roots, false);
3110 		if (ret < 0)
3111 			goto out;
3112 		/* For rescan, just pass old_roots as NULL */
3113 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3114 						  num_bytes, NULL, roots);
3115 		if (ret < 0)
3116 			goto out;
3117 	}
3118 out:
3119 	if (scratch_leaf)
3120 		free_extent_buffer(scratch_leaf);
3121 
3122 	if (done && !ret) {
3123 		ret = 1;
3124 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3125 	}
3126 	return ret;
3127 }
3128 
3129 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3130 {
3131 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3132 						     qgroup_rescan_work);
3133 	struct btrfs_path *path;
3134 	struct btrfs_trans_handle *trans = NULL;
3135 	int err = -ENOMEM;
3136 	int ret = 0;
3137 
3138 	path = btrfs_alloc_path();
3139 	if (!path)
3140 		goto out;
3141 	/*
3142 	 * Rescan should only search for commit root, and any later difference
3143 	 * should be recorded by qgroup
3144 	 */
3145 	path->search_commit_root = 1;
3146 	path->skip_locking = 1;
3147 
3148 	err = 0;
3149 	while (!err && !btrfs_fs_closing(fs_info)) {
3150 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3151 		if (IS_ERR(trans)) {
3152 			err = PTR_ERR(trans);
3153 			break;
3154 		}
3155 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3156 			err = -EINTR;
3157 		} else {
3158 			err = qgroup_rescan_leaf(trans, path);
3159 		}
3160 		if (err > 0)
3161 			btrfs_commit_transaction(trans);
3162 		else
3163 			btrfs_end_transaction(trans);
3164 	}
3165 
3166 out:
3167 	btrfs_free_path(path);
3168 
3169 	mutex_lock(&fs_info->qgroup_rescan_lock);
3170 	if (err > 0 &&
3171 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3172 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3173 	} else if (err < 0) {
3174 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3175 	}
3176 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3177 
3178 	/*
3179 	 * only update status, since the previous part has already updated the
3180 	 * qgroup info.
3181 	 */
3182 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3183 	if (IS_ERR(trans)) {
3184 		err = PTR_ERR(trans);
3185 		trans = NULL;
3186 		btrfs_err(fs_info,
3187 			  "fail to start transaction for status update: %d",
3188 			  err);
3189 	}
3190 
3191 	mutex_lock(&fs_info->qgroup_rescan_lock);
3192 	if (!btrfs_fs_closing(fs_info))
3193 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3194 	if (trans) {
3195 		ret = update_qgroup_status_item(trans);
3196 		if (ret < 0) {
3197 			err = ret;
3198 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3199 				  err);
3200 		}
3201 	}
3202 	fs_info->qgroup_rescan_running = false;
3203 	complete_all(&fs_info->qgroup_rescan_completion);
3204 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3205 
3206 	if (!trans)
3207 		return;
3208 
3209 	btrfs_end_transaction(trans);
3210 
3211 	if (btrfs_fs_closing(fs_info)) {
3212 		btrfs_info(fs_info, "qgroup scan paused");
3213 	} else if (err >= 0) {
3214 		btrfs_info(fs_info, "qgroup scan completed%s",
3215 			err > 0 ? " (inconsistency flag cleared)" : "");
3216 	} else {
3217 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3218 	}
3219 }
3220 
3221 /*
3222  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3223  * memory required for the rescan context.
3224  */
3225 static int
3226 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3227 		   int init_flags)
3228 {
3229 	int ret = 0;
3230 
3231 	if (!init_flags) {
3232 		/* we're resuming qgroup rescan at mount time */
3233 		if (!(fs_info->qgroup_flags &
3234 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3235 			btrfs_warn(fs_info,
3236 			"qgroup rescan init failed, qgroup rescan is not queued");
3237 			ret = -EINVAL;
3238 		} else if (!(fs_info->qgroup_flags &
3239 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3240 			btrfs_warn(fs_info,
3241 			"qgroup rescan init failed, qgroup is not enabled");
3242 			ret = -EINVAL;
3243 		}
3244 
3245 		if (ret)
3246 			return ret;
3247 	}
3248 
3249 	mutex_lock(&fs_info->qgroup_rescan_lock);
3250 
3251 	if (init_flags) {
3252 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3253 			btrfs_warn(fs_info,
3254 				   "qgroup rescan is already in progress");
3255 			ret = -EINPROGRESS;
3256 		} else if (!(fs_info->qgroup_flags &
3257 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3258 			btrfs_warn(fs_info,
3259 			"qgroup rescan init failed, qgroup is not enabled");
3260 			ret = -EINVAL;
3261 		}
3262 
3263 		if (ret) {
3264 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3265 			return ret;
3266 		}
3267 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3268 	}
3269 
3270 	memset(&fs_info->qgroup_rescan_progress, 0,
3271 		sizeof(fs_info->qgroup_rescan_progress));
3272 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3273 	init_completion(&fs_info->qgroup_rescan_completion);
3274 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3275 
3276 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3277 			btrfs_qgroup_rescan_worker, NULL, NULL);
3278 	return 0;
3279 }
3280 
3281 static void
3282 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3283 {
3284 	struct rb_node *n;
3285 	struct btrfs_qgroup *qgroup;
3286 
3287 	spin_lock(&fs_info->qgroup_lock);
3288 	/* clear all current qgroup tracking information */
3289 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3290 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3291 		qgroup->rfer = 0;
3292 		qgroup->rfer_cmpr = 0;
3293 		qgroup->excl = 0;
3294 		qgroup->excl_cmpr = 0;
3295 		qgroup_dirty(fs_info, qgroup);
3296 	}
3297 	spin_unlock(&fs_info->qgroup_lock);
3298 }
3299 
3300 int
3301 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3302 {
3303 	int ret = 0;
3304 	struct btrfs_trans_handle *trans;
3305 
3306 	ret = qgroup_rescan_init(fs_info, 0, 1);
3307 	if (ret)
3308 		return ret;
3309 
3310 	/*
3311 	 * We have set the rescan_progress to 0, which means no more
3312 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3313 	 * However, btrfs_qgroup_account_ref may be right after its call
3314 	 * to btrfs_find_all_roots, in which case it would still do the
3315 	 * accounting.
3316 	 * To solve this, we're committing the transaction, which will
3317 	 * ensure we run all delayed refs and only after that, we are
3318 	 * going to clear all tracking information for a clean start.
3319 	 */
3320 
3321 	trans = btrfs_join_transaction(fs_info->fs_root);
3322 	if (IS_ERR(trans)) {
3323 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3324 		return PTR_ERR(trans);
3325 	}
3326 	ret = btrfs_commit_transaction(trans);
3327 	if (ret) {
3328 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3329 		return ret;
3330 	}
3331 
3332 	qgroup_rescan_zero_tracking(fs_info);
3333 
3334 	mutex_lock(&fs_info->qgroup_rescan_lock);
3335 	fs_info->qgroup_rescan_running = true;
3336 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3337 			 &fs_info->qgroup_rescan_work);
3338 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3339 
3340 	return 0;
3341 }
3342 
3343 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3344 				     bool interruptible)
3345 {
3346 	int running;
3347 	int ret = 0;
3348 
3349 	mutex_lock(&fs_info->qgroup_rescan_lock);
3350 	running = fs_info->qgroup_rescan_running;
3351 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3352 
3353 	if (!running)
3354 		return 0;
3355 
3356 	if (interruptible)
3357 		ret = wait_for_completion_interruptible(
3358 					&fs_info->qgroup_rescan_completion);
3359 	else
3360 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3361 
3362 	return ret;
3363 }
3364 
3365 /*
3366  * this is only called from open_ctree where we're still single threaded, thus
3367  * locking is omitted here.
3368  */
3369 void
3370 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3371 {
3372 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3373 		mutex_lock(&fs_info->qgroup_rescan_lock);
3374 		fs_info->qgroup_rescan_running = true;
3375 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3376 				 &fs_info->qgroup_rescan_work);
3377 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3378 	}
3379 }
3380 
3381 /*
3382  * Reserve qgroup space for range [start, start + len).
3383  *
3384  * This function will either reserve space from related qgroups or doing
3385  * nothing if the range is already reserved.
3386  *
3387  * Return 0 for successful reserve
3388  * Return <0 for error (including -EQUOT)
3389  *
3390  * NOTE: this function may sleep for memory allocation.
3391  *       if btrfs_qgroup_reserve_data() is called multiple times with
3392  *       same @reserved, caller must ensure when error happens it's OK
3393  *       to free *ALL* reserved space.
3394  */
3395 int btrfs_qgroup_reserve_data(struct inode *inode,
3396 			struct extent_changeset **reserved_ret, u64 start,
3397 			u64 len)
3398 {
3399 	struct btrfs_root *root = BTRFS_I(inode)->root;
3400 	struct ulist_node *unode;
3401 	struct ulist_iterator uiter;
3402 	struct extent_changeset *reserved;
3403 	u64 orig_reserved;
3404 	u64 to_reserve;
3405 	int ret;
3406 
3407 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3408 	    !is_fstree(root->root_key.objectid) || len == 0)
3409 		return 0;
3410 
3411 	/* @reserved parameter is mandatory for qgroup */
3412 	if (WARN_ON(!reserved_ret))
3413 		return -EINVAL;
3414 	if (!*reserved_ret) {
3415 		*reserved_ret = extent_changeset_alloc();
3416 		if (!*reserved_ret)
3417 			return -ENOMEM;
3418 	}
3419 	reserved = *reserved_ret;
3420 	/* Record already reserved space */
3421 	orig_reserved = reserved->bytes_changed;
3422 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3423 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3424 
3425 	/* Newly reserved space */
3426 	to_reserve = reserved->bytes_changed - orig_reserved;
3427 	trace_btrfs_qgroup_reserve_data(inode, start, len,
3428 					to_reserve, QGROUP_RESERVE);
3429 	if (ret < 0)
3430 		goto cleanup;
3431 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3432 	if (ret < 0)
3433 		goto cleanup;
3434 
3435 	return ret;
3436 
3437 cleanup:
3438 	/* cleanup *ALL* already reserved ranges */
3439 	ULIST_ITER_INIT(&uiter);
3440 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3441 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3442 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3443 	/* Also free data bytes of already reserved one */
3444 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3445 				  orig_reserved, BTRFS_QGROUP_RSV_DATA);
3446 	extent_changeset_release(reserved);
3447 	return ret;
3448 }
3449 
3450 /* Free ranges specified by @reserved, normally in error path */
3451 static int qgroup_free_reserved_data(struct inode *inode,
3452 			struct extent_changeset *reserved, u64 start, u64 len)
3453 {
3454 	struct btrfs_root *root = BTRFS_I(inode)->root;
3455 	struct ulist_node *unode;
3456 	struct ulist_iterator uiter;
3457 	struct extent_changeset changeset;
3458 	int freed = 0;
3459 	int ret;
3460 
3461 	extent_changeset_init(&changeset);
3462 	len = round_up(start + len, root->fs_info->sectorsize);
3463 	start = round_down(start, root->fs_info->sectorsize);
3464 
3465 	ULIST_ITER_INIT(&uiter);
3466 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3467 		u64 range_start = unode->val;
3468 		/* unode->aux is the inclusive end */
3469 		u64 range_len = unode->aux - range_start + 1;
3470 		u64 free_start;
3471 		u64 free_len;
3472 
3473 		extent_changeset_release(&changeset);
3474 
3475 		/* Only free range in range [start, start + len) */
3476 		if (range_start >= start + len ||
3477 		    range_start + range_len <= start)
3478 			continue;
3479 		free_start = max(range_start, start);
3480 		free_len = min(start + len, range_start + range_len) -
3481 			   free_start;
3482 		/*
3483 		 * TODO: To also modify reserved->ranges_reserved to reflect
3484 		 * the modification.
3485 		 *
3486 		 * However as long as we free qgroup reserved according to
3487 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3488 		 * So not need to rush.
3489 		 */
3490 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree,
3491 				free_start, free_start + free_len - 1,
3492 				EXTENT_QGROUP_RESERVED, &changeset);
3493 		if (ret < 0)
3494 			goto out;
3495 		freed += changeset.bytes_changed;
3496 	}
3497 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3498 				  BTRFS_QGROUP_RSV_DATA);
3499 	ret = freed;
3500 out:
3501 	extent_changeset_release(&changeset);
3502 	return ret;
3503 }
3504 
3505 static int __btrfs_qgroup_release_data(struct inode *inode,
3506 			struct extent_changeset *reserved, u64 start, u64 len,
3507 			int free)
3508 {
3509 	struct extent_changeset changeset;
3510 	int trace_op = QGROUP_RELEASE;
3511 	int ret;
3512 
3513 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3514 		      &BTRFS_I(inode)->root->fs_info->flags))
3515 		return 0;
3516 
3517 	/* In release case, we shouldn't have @reserved */
3518 	WARN_ON(!free && reserved);
3519 	if (free && reserved)
3520 		return qgroup_free_reserved_data(inode, reserved, start, len);
3521 	extent_changeset_init(&changeset);
3522 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3523 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3524 	if (ret < 0)
3525 		goto out;
3526 
3527 	if (free)
3528 		trace_op = QGROUP_FREE;
3529 	trace_btrfs_qgroup_release_data(inode, start, len,
3530 					changeset.bytes_changed, trace_op);
3531 	if (free)
3532 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3533 				BTRFS_I(inode)->root->root_key.objectid,
3534 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3535 	ret = changeset.bytes_changed;
3536 out:
3537 	extent_changeset_release(&changeset);
3538 	return ret;
3539 }
3540 
3541 /*
3542  * Free a reserved space range from io_tree and related qgroups
3543  *
3544  * Should be called when a range of pages get invalidated before reaching disk.
3545  * Or for error cleanup case.
3546  * if @reserved is given, only reserved range in [@start, @start + @len) will
3547  * be freed.
3548  *
3549  * For data written to disk, use btrfs_qgroup_release_data().
3550  *
3551  * NOTE: This function may sleep for memory allocation.
3552  */
3553 int btrfs_qgroup_free_data(struct inode *inode,
3554 			struct extent_changeset *reserved, u64 start, u64 len)
3555 {
3556 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3557 }
3558 
3559 /*
3560  * Release a reserved space range from io_tree only.
3561  *
3562  * Should be called when a range of pages get written to disk and corresponding
3563  * FILE_EXTENT is inserted into corresponding root.
3564  *
3565  * Since new qgroup accounting framework will only update qgroup numbers at
3566  * commit_transaction() time, its reserved space shouldn't be freed from
3567  * related qgroups.
3568  *
3569  * But we should release the range from io_tree, to allow further write to be
3570  * COWed.
3571  *
3572  * NOTE: This function may sleep for memory allocation.
3573  */
3574 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3575 {
3576 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3577 }
3578 
3579 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3580 			      enum btrfs_qgroup_rsv_type type)
3581 {
3582 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3583 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3584 		return;
3585 	if (num_bytes == 0)
3586 		return;
3587 
3588 	spin_lock(&root->qgroup_meta_rsv_lock);
3589 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3590 		root->qgroup_meta_rsv_prealloc += num_bytes;
3591 	else
3592 		root->qgroup_meta_rsv_pertrans += num_bytes;
3593 	spin_unlock(&root->qgroup_meta_rsv_lock);
3594 }
3595 
3596 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3597 			     enum btrfs_qgroup_rsv_type type)
3598 {
3599 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3600 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3601 		return 0;
3602 	if (num_bytes == 0)
3603 		return 0;
3604 
3605 	spin_lock(&root->qgroup_meta_rsv_lock);
3606 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3607 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3608 				  num_bytes);
3609 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3610 	} else {
3611 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3612 				  num_bytes);
3613 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3614 	}
3615 	spin_unlock(&root->qgroup_meta_rsv_lock);
3616 	return num_bytes;
3617 }
3618 
3619 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3620 				enum btrfs_qgroup_rsv_type type, bool enforce)
3621 {
3622 	struct btrfs_fs_info *fs_info = root->fs_info;
3623 	int ret;
3624 
3625 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3626 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3627 		return 0;
3628 
3629 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3630 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3631 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3632 	if (ret < 0)
3633 		return ret;
3634 	/*
3635 	 * Record what we have reserved into root.
3636 	 *
3637 	 * To avoid quota disabled->enabled underflow.
3638 	 * In that case, we may try to free space we haven't reserved
3639 	 * (since quota was disabled), so record what we reserved into root.
3640 	 * And ensure later release won't underflow this number.
3641 	 */
3642 	add_root_meta_rsv(root, num_bytes, type);
3643 	return ret;
3644 }
3645 
3646 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3647 {
3648 	struct btrfs_fs_info *fs_info = root->fs_info;
3649 
3650 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3651 	    !is_fstree(root->root_key.objectid))
3652 		return;
3653 
3654 	/* TODO: Update trace point to handle such free */
3655 	trace_qgroup_meta_free_all_pertrans(root);
3656 	/* Special value -1 means to free all reserved space */
3657 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3658 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3659 }
3660 
3661 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3662 			      enum btrfs_qgroup_rsv_type type)
3663 {
3664 	struct btrfs_fs_info *fs_info = root->fs_info;
3665 
3666 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3667 	    !is_fstree(root->root_key.objectid))
3668 		return;
3669 
3670 	/*
3671 	 * reservation for META_PREALLOC can happen before quota is enabled,
3672 	 * which can lead to underflow.
3673 	 * Here ensure we will only free what we really have reserved.
3674 	 */
3675 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3676 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3677 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3678 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3679 				  num_bytes, type);
3680 }
3681 
3682 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3683 				int num_bytes)
3684 {
3685 	struct btrfs_qgroup *qgroup;
3686 	struct ulist_node *unode;
3687 	struct ulist_iterator uiter;
3688 	int ret = 0;
3689 
3690 	if (num_bytes == 0)
3691 		return;
3692 	if (!fs_info->quota_root)
3693 		return;
3694 
3695 	spin_lock(&fs_info->qgroup_lock);
3696 	qgroup = find_qgroup_rb(fs_info, ref_root);
3697 	if (!qgroup)
3698 		goto out;
3699 	ulist_reinit(fs_info->qgroup_ulist);
3700 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3701 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3702 	if (ret < 0)
3703 		goto out;
3704 	ULIST_ITER_INIT(&uiter);
3705 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3706 		struct btrfs_qgroup *qg;
3707 		struct btrfs_qgroup_list *glist;
3708 
3709 		qg = unode_aux_to_qgroup(unode);
3710 
3711 		qgroup_rsv_release(fs_info, qg, num_bytes,
3712 				BTRFS_QGROUP_RSV_META_PREALLOC);
3713 		qgroup_rsv_add(fs_info, qg, num_bytes,
3714 				BTRFS_QGROUP_RSV_META_PERTRANS);
3715 		list_for_each_entry(glist, &qg->groups, next_group) {
3716 			ret = ulist_add(fs_info->qgroup_ulist,
3717 					glist->group->qgroupid,
3718 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3719 			if (ret < 0)
3720 				goto out;
3721 		}
3722 	}
3723 out:
3724 	spin_unlock(&fs_info->qgroup_lock);
3725 }
3726 
3727 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3728 {
3729 	struct btrfs_fs_info *fs_info = root->fs_info;
3730 
3731 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3732 	    !is_fstree(root->root_key.objectid))
3733 		return;
3734 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3735 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3736 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3737 	trace_qgroup_meta_convert(root, num_bytes);
3738 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3739 }
3740 
3741 /*
3742  * Check qgroup reserved space leaking, normally at destroy inode
3743  * time
3744  */
3745 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3746 {
3747 	struct extent_changeset changeset;
3748 	struct ulist_node *unode;
3749 	struct ulist_iterator iter;
3750 	int ret;
3751 
3752 	extent_changeset_init(&changeset);
3753 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3754 			EXTENT_QGROUP_RESERVED, &changeset);
3755 
3756 	WARN_ON(ret < 0);
3757 	if (WARN_ON(changeset.bytes_changed)) {
3758 		ULIST_ITER_INIT(&iter);
3759 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3760 			btrfs_warn(BTRFS_I(inode)->root->fs_info,
3761 				"leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3762 				inode->i_ino, unode->val, unode->aux);
3763 		}
3764 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3765 				BTRFS_I(inode)->root->root_key.objectid,
3766 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3767 
3768 	}
3769 	extent_changeset_release(&changeset);
3770 }
3771 
3772 void btrfs_qgroup_init_swapped_blocks(
3773 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3774 {
3775 	int i;
3776 
3777 	spin_lock_init(&swapped_blocks->lock);
3778 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3779 		swapped_blocks->blocks[i] = RB_ROOT;
3780 	swapped_blocks->swapped = false;
3781 }
3782 
3783 /*
3784  * Delete all swapped blocks record of @root.
3785  * Every record here means we skipped a full subtree scan for qgroup.
3786  *
3787  * Gets called when committing one transaction.
3788  */
3789 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3790 {
3791 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3792 	int i;
3793 
3794 	swapped_blocks = &root->swapped_blocks;
3795 
3796 	spin_lock(&swapped_blocks->lock);
3797 	if (!swapped_blocks->swapped)
3798 		goto out;
3799 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3800 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
3801 		struct btrfs_qgroup_swapped_block *entry;
3802 		struct btrfs_qgroup_swapped_block *next;
3803 
3804 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3805 						     node)
3806 			kfree(entry);
3807 		swapped_blocks->blocks[i] = RB_ROOT;
3808 	}
3809 	swapped_blocks->swapped = false;
3810 out:
3811 	spin_unlock(&swapped_blocks->lock);
3812 }
3813 
3814 /*
3815  * Add subtree roots record into @subvol_root.
3816  *
3817  * @subvol_root:	tree root of the subvolume tree get swapped
3818  * @bg:			block group under balance
3819  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
3820  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
3821  *			BOTH POINTERS ARE BEFORE TREE SWAP
3822  * @last_snapshot:	last snapshot generation of the subvolume tree
3823  */
3824 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3825 		struct btrfs_root *subvol_root,
3826 		struct btrfs_block_group *bg,
3827 		struct extent_buffer *subvol_parent, int subvol_slot,
3828 		struct extent_buffer *reloc_parent, int reloc_slot,
3829 		u64 last_snapshot)
3830 {
3831 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3832 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3833 	struct btrfs_qgroup_swapped_block *block;
3834 	struct rb_node **cur;
3835 	struct rb_node *parent = NULL;
3836 	int level = btrfs_header_level(subvol_parent) - 1;
3837 	int ret = 0;
3838 
3839 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3840 		return 0;
3841 
3842 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3843 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3844 		btrfs_err_rl(fs_info,
3845 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3846 			__func__,
3847 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3848 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3849 		return -EUCLEAN;
3850 	}
3851 
3852 	block = kmalloc(sizeof(*block), GFP_NOFS);
3853 	if (!block) {
3854 		ret = -ENOMEM;
3855 		goto out;
3856 	}
3857 
3858 	/*
3859 	 * @reloc_parent/slot is still before swap, while @block is going to
3860 	 * record the bytenr after swap, so we do the swap here.
3861 	 */
3862 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3863 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3864 							     reloc_slot);
3865 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3866 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3867 							    subvol_slot);
3868 	block->last_snapshot = last_snapshot;
3869 	block->level = level;
3870 
3871 	/*
3872 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3873 	 * no one else can modify tree blocks thus we qgroup will not change
3874 	 * no matter the value of trace_leaf.
3875 	 */
3876 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
3877 		block->trace_leaf = true;
3878 	else
3879 		block->trace_leaf = false;
3880 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3881 
3882 	/* Insert @block into @blocks */
3883 	spin_lock(&blocks->lock);
3884 	cur = &blocks->blocks[level].rb_node;
3885 	while (*cur) {
3886 		struct btrfs_qgroup_swapped_block *entry;
3887 
3888 		parent = *cur;
3889 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3890 				 node);
3891 
3892 		if (entry->subvol_bytenr < block->subvol_bytenr) {
3893 			cur = &(*cur)->rb_left;
3894 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
3895 			cur = &(*cur)->rb_right;
3896 		} else {
3897 			if (entry->subvol_generation !=
3898 					block->subvol_generation ||
3899 			    entry->reloc_bytenr != block->reloc_bytenr ||
3900 			    entry->reloc_generation !=
3901 					block->reloc_generation) {
3902 				/*
3903 				 * Duplicated but mismatch entry found.
3904 				 * Shouldn't happen.
3905 				 *
3906 				 * Marking qgroup inconsistent should be enough
3907 				 * for end users.
3908 				 */
3909 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3910 				ret = -EEXIST;
3911 			}
3912 			kfree(block);
3913 			goto out_unlock;
3914 		}
3915 	}
3916 	rb_link_node(&block->node, parent, cur);
3917 	rb_insert_color(&block->node, &blocks->blocks[level]);
3918 	blocks->swapped = true;
3919 out_unlock:
3920 	spin_unlock(&blocks->lock);
3921 out:
3922 	if (ret < 0)
3923 		fs_info->qgroup_flags |=
3924 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3925 	return ret;
3926 }
3927 
3928 /*
3929  * Check if the tree block is a subtree root, and if so do the needed
3930  * delayed subtree trace for qgroup.
3931  *
3932  * This is called during btrfs_cow_block().
3933  */
3934 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3935 					 struct btrfs_root *root,
3936 					 struct extent_buffer *subvol_eb)
3937 {
3938 	struct btrfs_fs_info *fs_info = root->fs_info;
3939 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3940 	struct btrfs_qgroup_swapped_block *block;
3941 	struct extent_buffer *reloc_eb = NULL;
3942 	struct rb_node *node;
3943 	bool found = false;
3944 	bool swapped = false;
3945 	int level = btrfs_header_level(subvol_eb);
3946 	int ret = 0;
3947 	int i;
3948 
3949 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3950 		return 0;
3951 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3952 		return 0;
3953 
3954 	spin_lock(&blocks->lock);
3955 	if (!blocks->swapped) {
3956 		spin_unlock(&blocks->lock);
3957 		return 0;
3958 	}
3959 	node = blocks->blocks[level].rb_node;
3960 
3961 	while (node) {
3962 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3963 		if (block->subvol_bytenr < subvol_eb->start) {
3964 			node = node->rb_left;
3965 		} else if (block->subvol_bytenr > subvol_eb->start) {
3966 			node = node->rb_right;
3967 		} else {
3968 			found = true;
3969 			break;
3970 		}
3971 	}
3972 	if (!found) {
3973 		spin_unlock(&blocks->lock);
3974 		goto out;
3975 	}
3976 	/* Found one, remove it from @blocks first and update blocks->swapped */
3977 	rb_erase(&block->node, &blocks->blocks[level]);
3978 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3979 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3980 			swapped = true;
3981 			break;
3982 		}
3983 	}
3984 	blocks->swapped = swapped;
3985 	spin_unlock(&blocks->lock);
3986 
3987 	/* Read out reloc subtree root */
3988 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3989 				   block->reloc_generation, block->level,
3990 				   &block->first_key);
3991 	if (IS_ERR(reloc_eb)) {
3992 		ret = PTR_ERR(reloc_eb);
3993 		reloc_eb = NULL;
3994 		goto free_out;
3995 	}
3996 	if (!extent_buffer_uptodate(reloc_eb)) {
3997 		ret = -EIO;
3998 		goto free_out;
3999 	}
4000 
4001 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4002 			block->last_snapshot, block->trace_leaf);
4003 free_out:
4004 	kfree(block);
4005 	free_extent_buffer(reloc_eb);
4006 out:
4007 	if (ret < 0) {
4008 		btrfs_err_rl(fs_info,
4009 			     "failed to account subtree at bytenr %llu: %d",
4010 			     subvol_eb->start, ret);
4011 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4012 	}
4013 	return ret;
4014 }
4015 
4016 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4017 {
4018 	struct btrfs_qgroup_extent_record *entry;
4019 	struct btrfs_qgroup_extent_record *next;
4020 	struct rb_root *root;
4021 
4022 	root = &trans->delayed_refs.dirty_extent_root;
4023 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4024 		ulist_free(entry->old_roots);
4025 		kfree(entry);
4026 	}
4027 }
4028