xref: /openbmc/linux/fs/btrfs/qgroup.c (revision 532c2919)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sizes.h>
15 
16 #include "ctree.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "locking.h"
20 #include "ulist.h"
21 #include "backref.h"
22 #include "extent_io.h"
23 #include "qgroup.h"
24 #include "block-group.h"
25 
26 /* TODO XXX FIXME
27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
28  *  - reorganize keys
29  *  - compressed
30  *  - sync
31  *  - copy also limits on subvol creation
32  *  - limit
33  *  - caches for ulists
34  *  - performance benchmarks
35  *  - check all ioctl parameters
36  */
37 
38 /*
39  * Helpers to access qgroup reservation
40  *
41  * Callers should ensure the lock context and type are valid
42  */
43 
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45 {
46 	u64 ret = 0;
47 	int i;
48 
49 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 		ret += qgroup->rsv.values[i];
51 
52 	return ret;
53 }
54 
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57 {
58 	if (type == BTRFS_QGROUP_RSV_DATA)
59 		return "data";
60 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 		return "meta_pertrans";
62 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 		return "meta_prealloc";
64 	return NULL;
65 }
66 #endif
67 
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
70 			   enum btrfs_qgroup_rsv_type type)
71 {
72 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 	qgroup->rsv.values[type] += num_bytes;
74 }
75 
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
78 			       enum btrfs_qgroup_rsv_type type)
79 {
80 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 	if (qgroup->rsv.values[type] >= num_bytes) {
82 		qgroup->rsv.values[type] -= num_bytes;
83 		return;
84 	}
85 #ifdef CONFIG_BTRFS_DEBUG
86 	WARN_RATELIMIT(1,
87 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 		qgroup->qgroupid, qgroup_rsv_type_str(type),
89 		qgroup->rsv.values[type], num_bytes);
90 #endif
91 	qgroup->rsv.values[type] = 0;
92 }
93 
94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 				     struct btrfs_qgroup *dest,
96 				     struct btrfs_qgroup *src)
97 {
98 	int i;
99 
100 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 		qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
102 }
103 
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 					 struct btrfs_qgroup *dest,
106 					  struct btrfs_qgroup *src)
107 {
108 	int i;
109 
110 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
112 }
113 
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115 					   int mod)
116 {
117 	if (qg->old_refcnt < seq)
118 		qg->old_refcnt = seq;
119 	qg->old_refcnt += mod;
120 }
121 
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123 					   int mod)
124 {
125 	if (qg->new_refcnt < seq)
126 		qg->new_refcnt = seq;
127 	qg->new_refcnt += mod;
128 }
129 
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131 {
132 	if (qg->old_refcnt < seq)
133 		return 0;
134 	return qg->old_refcnt - seq;
135 }
136 
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138 {
139 	if (qg->new_refcnt < seq)
140 		return 0;
141 	return qg->new_refcnt - seq;
142 }
143 
144 /*
145  * glue structure to represent the relations between qgroups.
146  */
147 struct btrfs_qgroup_list {
148 	struct list_head next_group;
149 	struct list_head next_member;
150 	struct btrfs_qgroup *group;
151 	struct btrfs_qgroup *member;
152 };
153 
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 {
156 	return (u64)(uintptr_t)qg;
157 }
158 
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 {
161 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162 }
163 
164 static int
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 		   int init_flags);
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168 
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171 					   u64 qgroupid)
172 {
173 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 	struct btrfs_qgroup *qgroup;
175 
176 	while (n) {
177 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 		if (qgroup->qgroupid < qgroupid)
179 			n = n->rb_left;
180 		else if (qgroup->qgroupid > qgroupid)
181 			n = n->rb_right;
182 		else
183 			return qgroup;
184 	}
185 	return NULL;
186 }
187 
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190 					  u64 qgroupid)
191 {
192 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 	struct rb_node *parent = NULL;
194 	struct btrfs_qgroup *qgroup;
195 
196 	while (*p) {
197 		parent = *p;
198 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199 
200 		if (qgroup->qgroupid < qgroupid)
201 			p = &(*p)->rb_left;
202 		else if (qgroup->qgroupid > qgroupid)
203 			p = &(*p)->rb_right;
204 		else
205 			return qgroup;
206 	}
207 
208 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 	if (!qgroup)
210 		return ERR_PTR(-ENOMEM);
211 
212 	qgroup->qgroupid = qgroupid;
213 	INIT_LIST_HEAD(&qgroup->groups);
214 	INIT_LIST_HEAD(&qgroup->members);
215 	INIT_LIST_HEAD(&qgroup->dirty);
216 
217 	rb_link_node(&qgroup->node, parent, p);
218 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219 
220 	return qgroup;
221 }
222 
223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
224 {
225 	struct btrfs_qgroup_list *list;
226 
227 	list_del(&qgroup->dirty);
228 	while (!list_empty(&qgroup->groups)) {
229 		list = list_first_entry(&qgroup->groups,
230 					struct btrfs_qgroup_list, next_group);
231 		list_del(&list->next_group);
232 		list_del(&list->next_member);
233 		kfree(list);
234 	}
235 
236 	while (!list_empty(&qgroup->members)) {
237 		list = list_first_entry(&qgroup->members,
238 					struct btrfs_qgroup_list, next_member);
239 		list_del(&list->next_group);
240 		list_del(&list->next_member);
241 		kfree(list);
242 	}
243 	kfree(qgroup);
244 }
245 
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248 {
249 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250 
251 	if (!qgroup)
252 		return -ENOENT;
253 
254 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 	__del_qgroup_rb(qgroup);
256 	return 0;
257 }
258 
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 			   u64 memberid, u64 parentid)
262 {
263 	struct btrfs_qgroup *member;
264 	struct btrfs_qgroup *parent;
265 	struct btrfs_qgroup_list *list;
266 
267 	member = find_qgroup_rb(fs_info, memberid);
268 	parent = find_qgroup_rb(fs_info, parentid);
269 	if (!member || !parent)
270 		return -ENOENT;
271 
272 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
273 	if (!list)
274 		return -ENOMEM;
275 
276 	list->group = parent;
277 	list->member = member;
278 	list_add_tail(&list->next_group, &member->groups);
279 	list_add_tail(&list->next_member, &parent->members);
280 
281 	return 0;
282 }
283 
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 			   u64 memberid, u64 parentid)
287 {
288 	struct btrfs_qgroup *member;
289 	struct btrfs_qgroup *parent;
290 	struct btrfs_qgroup_list *list;
291 
292 	member = find_qgroup_rb(fs_info, memberid);
293 	parent = find_qgroup_rb(fs_info, parentid);
294 	if (!member || !parent)
295 		return -ENOENT;
296 
297 	list_for_each_entry(list, &member->groups, next_group) {
298 		if (list->group == parent) {
299 			list_del(&list->next_group);
300 			list_del(&list->next_member);
301 			kfree(list);
302 			return 0;
303 		}
304 	}
305 	return -ENOENT;
306 }
307 
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310 			       u64 rfer, u64 excl)
311 {
312 	struct btrfs_qgroup *qgroup;
313 
314 	qgroup = find_qgroup_rb(fs_info, qgroupid);
315 	if (!qgroup)
316 		return -EINVAL;
317 	if (qgroup->rfer != rfer || qgroup->excl != excl)
318 		return -EINVAL;
319 	return 0;
320 }
321 #endif
322 
323 /*
324  * The full config is read in one go, only called from open_ctree()
325  * It doesn't use any locking, as at this point we're still single-threaded
326  */
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328 {
329 	struct btrfs_key key;
330 	struct btrfs_key found_key;
331 	struct btrfs_root *quota_root = fs_info->quota_root;
332 	struct btrfs_path *path = NULL;
333 	struct extent_buffer *l;
334 	int slot;
335 	int ret = 0;
336 	u64 flags = 0;
337 	u64 rescan_progress = 0;
338 
339 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
340 		return 0;
341 
342 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 	if (!fs_info->qgroup_ulist) {
344 		ret = -ENOMEM;
345 		goto out;
346 	}
347 
348 	path = btrfs_alloc_path();
349 	if (!path) {
350 		ret = -ENOMEM;
351 		goto out;
352 	}
353 
354 	/* default this to quota off, in case no status key is found */
355 	fs_info->qgroup_flags = 0;
356 
357 	/*
358 	 * pass 1: read status, all qgroup infos and limits
359 	 */
360 	key.objectid = 0;
361 	key.type = 0;
362 	key.offset = 0;
363 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
364 	if (ret)
365 		goto out;
366 
367 	while (1) {
368 		struct btrfs_qgroup *qgroup;
369 
370 		slot = path->slots[0];
371 		l = path->nodes[0];
372 		btrfs_item_key_to_cpu(l, &found_key, slot);
373 
374 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 			struct btrfs_qgroup_status_item *ptr;
376 
377 			ptr = btrfs_item_ptr(l, slot,
378 					     struct btrfs_qgroup_status_item);
379 
380 			if (btrfs_qgroup_status_version(l, ptr) !=
381 			    BTRFS_QGROUP_STATUS_VERSION) {
382 				btrfs_err(fs_info,
383 				 "old qgroup version, quota disabled");
384 				goto out;
385 			}
386 			if (btrfs_qgroup_status_generation(l, ptr) !=
387 			    fs_info->generation) {
388 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 				btrfs_err(fs_info,
390 					"qgroup generation mismatch, marked as inconsistent");
391 			}
392 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 									  ptr);
394 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
395 			goto next1;
396 		}
397 
398 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
400 			goto next1;
401 
402 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 			btrfs_err(fs_info, "inconsistent qgroup config");
406 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407 		}
408 		if (!qgroup) {
409 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 			if (IS_ERR(qgroup)) {
411 				ret = PTR_ERR(qgroup);
412 				goto out;
413 			}
414 		}
415 		switch (found_key.type) {
416 		case BTRFS_QGROUP_INFO_KEY: {
417 			struct btrfs_qgroup_info_item *ptr;
418 
419 			ptr = btrfs_item_ptr(l, slot,
420 					     struct btrfs_qgroup_info_item);
421 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 			/* generation currently unused */
426 			break;
427 		}
428 		case BTRFS_QGROUP_LIMIT_KEY: {
429 			struct btrfs_qgroup_limit_item *ptr;
430 
431 			ptr = btrfs_item_ptr(l, slot,
432 					     struct btrfs_qgroup_limit_item);
433 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438 			break;
439 		}
440 		}
441 next1:
442 		ret = btrfs_next_item(quota_root, path);
443 		if (ret < 0)
444 			goto out;
445 		if (ret)
446 			break;
447 	}
448 	btrfs_release_path(path);
449 
450 	/*
451 	 * pass 2: read all qgroup relations
452 	 */
453 	key.objectid = 0;
454 	key.type = BTRFS_QGROUP_RELATION_KEY;
455 	key.offset = 0;
456 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457 	if (ret)
458 		goto out;
459 	while (1) {
460 		slot = path->slots[0];
461 		l = path->nodes[0];
462 		btrfs_item_key_to_cpu(l, &found_key, slot);
463 
464 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465 			goto next2;
466 
467 		if (found_key.objectid > found_key.offset) {
468 			/* parent <- member, not needed to build config */
469 			/* FIXME should we omit the key completely? */
470 			goto next2;
471 		}
472 
473 		ret = add_relation_rb(fs_info, found_key.objectid,
474 				      found_key.offset);
475 		if (ret == -ENOENT) {
476 			btrfs_warn(fs_info,
477 				"orphan qgroup relation 0x%llx->0x%llx",
478 				found_key.objectid, found_key.offset);
479 			ret = 0;	/* ignore the error */
480 		}
481 		if (ret)
482 			goto out;
483 next2:
484 		ret = btrfs_next_item(quota_root, path);
485 		if (ret < 0)
486 			goto out;
487 		if (ret)
488 			break;
489 	}
490 out:
491 	fs_info->qgroup_flags |= flags;
492 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495 		 ret >= 0)
496 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 	btrfs_free_path(path);
498 
499 	if (ret < 0) {
500 		ulist_free(fs_info->qgroup_ulist);
501 		fs_info->qgroup_ulist = NULL;
502 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
503 	}
504 
505 	return ret < 0 ? ret : 0;
506 }
507 
508 /*
509  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510  * first two are in single-threaded paths.And for the third one, we have set
511  * quota_root to be null with qgroup_lock held before, so it is safe to clean
512  * up the in-memory structures without qgroup_lock held.
513  */
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515 {
516 	struct rb_node *n;
517 	struct btrfs_qgroup *qgroup;
518 
519 	while ((n = rb_first(&fs_info->qgroup_tree))) {
520 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 		rb_erase(n, &fs_info->qgroup_tree);
522 		__del_qgroup_rb(qgroup);
523 	}
524 	/*
525 	 * We call btrfs_free_qgroup_config() when unmounting
526 	 * filesystem and disabling quota, so we set qgroup_ulist
527 	 * to be null here to avoid double free.
528 	 */
529 	ulist_free(fs_info->qgroup_ulist);
530 	fs_info->qgroup_ulist = NULL;
531 }
532 
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534 				    u64 dst)
535 {
536 	int ret;
537 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 	struct btrfs_path *path;
539 	struct btrfs_key key;
540 
541 	path = btrfs_alloc_path();
542 	if (!path)
543 		return -ENOMEM;
544 
545 	key.objectid = src;
546 	key.type = BTRFS_QGROUP_RELATION_KEY;
547 	key.offset = dst;
548 
549 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550 
551 	btrfs_mark_buffer_dirty(path->nodes[0]);
552 
553 	btrfs_free_path(path);
554 	return ret;
555 }
556 
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558 				    u64 dst)
559 {
560 	int ret;
561 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 	struct btrfs_path *path;
563 	struct btrfs_key key;
564 
565 	path = btrfs_alloc_path();
566 	if (!path)
567 		return -ENOMEM;
568 
569 	key.objectid = src;
570 	key.type = BTRFS_QGROUP_RELATION_KEY;
571 	key.offset = dst;
572 
573 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574 	if (ret < 0)
575 		goto out;
576 
577 	if (ret > 0) {
578 		ret = -ENOENT;
579 		goto out;
580 	}
581 
582 	ret = btrfs_del_item(trans, quota_root, path);
583 out:
584 	btrfs_free_path(path);
585 	return ret;
586 }
587 
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 			   struct btrfs_root *quota_root, u64 qgroupid)
590 {
591 	int ret;
592 	struct btrfs_path *path;
593 	struct btrfs_qgroup_info_item *qgroup_info;
594 	struct btrfs_qgroup_limit_item *qgroup_limit;
595 	struct extent_buffer *leaf;
596 	struct btrfs_key key;
597 
598 	if (btrfs_is_testing(quota_root->fs_info))
599 		return 0;
600 
601 	path = btrfs_alloc_path();
602 	if (!path)
603 		return -ENOMEM;
604 
605 	key.objectid = 0;
606 	key.type = BTRFS_QGROUP_INFO_KEY;
607 	key.offset = qgroupid;
608 
609 	/*
610 	 * Avoid a transaction abort by catching -EEXIST here. In that
611 	 * case, we proceed by re-initializing the existing structure
612 	 * on disk.
613 	 */
614 
615 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 				      sizeof(*qgroup_info));
617 	if (ret && ret != -EEXIST)
618 		goto out;
619 
620 	leaf = path->nodes[0];
621 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 				 struct btrfs_qgroup_info_item);
623 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628 
629 	btrfs_mark_buffer_dirty(leaf);
630 
631 	btrfs_release_path(path);
632 
633 	key.type = BTRFS_QGROUP_LIMIT_KEY;
634 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 				      sizeof(*qgroup_limit));
636 	if (ret && ret != -EEXIST)
637 		goto out;
638 
639 	leaf = path->nodes[0];
640 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 				  struct btrfs_qgroup_limit_item);
642 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647 
648 	btrfs_mark_buffer_dirty(leaf);
649 
650 	ret = 0;
651 out:
652 	btrfs_free_path(path);
653 	return ret;
654 }
655 
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
657 {
658 	int ret;
659 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 	struct btrfs_path *path;
661 	struct btrfs_key key;
662 
663 	path = btrfs_alloc_path();
664 	if (!path)
665 		return -ENOMEM;
666 
667 	key.objectid = 0;
668 	key.type = BTRFS_QGROUP_INFO_KEY;
669 	key.offset = qgroupid;
670 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 	if (ret < 0)
672 		goto out;
673 
674 	if (ret > 0) {
675 		ret = -ENOENT;
676 		goto out;
677 	}
678 
679 	ret = btrfs_del_item(trans, quota_root, path);
680 	if (ret)
681 		goto out;
682 
683 	btrfs_release_path(path);
684 
685 	key.type = BTRFS_QGROUP_LIMIT_KEY;
686 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687 	if (ret < 0)
688 		goto out;
689 
690 	if (ret > 0) {
691 		ret = -ENOENT;
692 		goto out;
693 	}
694 
695 	ret = btrfs_del_item(trans, quota_root, path);
696 
697 out:
698 	btrfs_free_path(path);
699 	return ret;
700 }
701 
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 				    struct btrfs_qgroup *qgroup)
704 {
705 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 	struct btrfs_path *path;
707 	struct btrfs_key key;
708 	struct extent_buffer *l;
709 	struct btrfs_qgroup_limit_item *qgroup_limit;
710 	int ret;
711 	int slot;
712 
713 	key.objectid = 0;
714 	key.type = BTRFS_QGROUP_LIMIT_KEY;
715 	key.offset = qgroup->qgroupid;
716 
717 	path = btrfs_alloc_path();
718 	if (!path)
719 		return -ENOMEM;
720 
721 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
722 	if (ret > 0)
723 		ret = -ENOENT;
724 
725 	if (ret)
726 		goto out;
727 
728 	l = path->nodes[0];
729 	slot = path->slots[0];
730 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736 
737 	btrfs_mark_buffer_dirty(l);
738 
739 out:
740 	btrfs_free_path(path);
741 	return ret;
742 }
743 
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 				   struct btrfs_qgroup *qgroup)
746 {
747 	struct btrfs_fs_info *fs_info = trans->fs_info;
748 	struct btrfs_root *quota_root = fs_info->quota_root;
749 	struct btrfs_path *path;
750 	struct btrfs_key key;
751 	struct extent_buffer *l;
752 	struct btrfs_qgroup_info_item *qgroup_info;
753 	int ret;
754 	int slot;
755 
756 	if (btrfs_is_testing(fs_info))
757 		return 0;
758 
759 	key.objectid = 0;
760 	key.type = BTRFS_QGROUP_INFO_KEY;
761 	key.offset = qgroup->qgroupid;
762 
763 	path = btrfs_alloc_path();
764 	if (!path)
765 		return -ENOMEM;
766 
767 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
768 	if (ret > 0)
769 		ret = -ENOENT;
770 
771 	if (ret)
772 		goto out;
773 
774 	l = path->nodes[0];
775 	slot = path->slots[0];
776 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782 
783 	btrfs_mark_buffer_dirty(l);
784 
785 out:
786 	btrfs_free_path(path);
787 	return ret;
788 }
789 
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
791 {
792 	struct btrfs_fs_info *fs_info = trans->fs_info;
793 	struct btrfs_root *quota_root = fs_info->quota_root;
794 	struct btrfs_path *path;
795 	struct btrfs_key key;
796 	struct extent_buffer *l;
797 	struct btrfs_qgroup_status_item *ptr;
798 	int ret;
799 	int slot;
800 
801 	key.objectid = 0;
802 	key.type = BTRFS_QGROUP_STATUS_KEY;
803 	key.offset = 0;
804 
805 	path = btrfs_alloc_path();
806 	if (!path)
807 		return -ENOMEM;
808 
809 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
810 	if (ret > 0)
811 		ret = -ENOENT;
812 
813 	if (ret)
814 		goto out;
815 
816 	l = path->nodes[0];
817 	slot = path->slots[0];
818 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 	btrfs_set_qgroup_status_rescan(l, ptr,
822 				fs_info->qgroup_rescan_progress.objectid);
823 
824 	btrfs_mark_buffer_dirty(l);
825 
826 out:
827 	btrfs_free_path(path);
828 	return ret;
829 }
830 
831 /*
832  * called with qgroup_lock held
833  */
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 				  struct btrfs_root *root)
836 {
837 	struct btrfs_path *path;
838 	struct btrfs_key key;
839 	struct extent_buffer *leaf = NULL;
840 	int ret;
841 	int nr = 0;
842 
843 	path = btrfs_alloc_path();
844 	if (!path)
845 		return -ENOMEM;
846 
847 	path->leave_spinning = 1;
848 
849 	key.objectid = 0;
850 	key.offset = 0;
851 	key.type = 0;
852 
853 	while (1) {
854 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855 		if (ret < 0)
856 			goto out;
857 		leaf = path->nodes[0];
858 		nr = btrfs_header_nritems(leaf);
859 		if (!nr)
860 			break;
861 		/*
862 		 * delete the leaf one by one
863 		 * since the whole tree is going
864 		 * to be deleted.
865 		 */
866 		path->slots[0] = 0;
867 		ret = btrfs_del_items(trans, root, path, 0, nr);
868 		if (ret)
869 			goto out;
870 
871 		btrfs_release_path(path);
872 	}
873 	ret = 0;
874 out:
875 	btrfs_free_path(path);
876 	return ret;
877 }
878 
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
880 {
881 	struct btrfs_root *quota_root;
882 	struct btrfs_root *tree_root = fs_info->tree_root;
883 	struct btrfs_path *path = NULL;
884 	struct btrfs_qgroup_status_item *ptr;
885 	struct extent_buffer *leaf;
886 	struct btrfs_key key;
887 	struct btrfs_key found_key;
888 	struct btrfs_qgroup *qgroup = NULL;
889 	struct btrfs_trans_handle *trans = NULL;
890 	int ret = 0;
891 	int slot;
892 
893 	mutex_lock(&fs_info->qgroup_ioctl_lock);
894 	if (fs_info->quota_root)
895 		goto out;
896 
897 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 	if (!fs_info->qgroup_ulist) {
899 		ret = -ENOMEM;
900 		goto out;
901 	}
902 
903 	/*
904 	 * 1 for quota root item
905 	 * 1 for BTRFS_QGROUP_STATUS item
906 	 *
907 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
908 	 * per subvolume. However those are not currently reserved since it
909 	 * would be a lot of overkill.
910 	 */
911 	trans = btrfs_start_transaction(tree_root, 2);
912 	if (IS_ERR(trans)) {
913 		ret = PTR_ERR(trans);
914 		trans = NULL;
915 		goto out;
916 	}
917 
918 	/*
919 	 * initially create the quota tree
920 	 */
921 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
922 	if (IS_ERR(quota_root)) {
923 		ret =  PTR_ERR(quota_root);
924 		btrfs_abort_transaction(trans, ret);
925 		goto out;
926 	}
927 
928 	path = btrfs_alloc_path();
929 	if (!path) {
930 		ret = -ENOMEM;
931 		btrfs_abort_transaction(trans, ret);
932 		goto out_free_root;
933 	}
934 
935 	key.objectid = 0;
936 	key.type = BTRFS_QGROUP_STATUS_KEY;
937 	key.offset = 0;
938 
939 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
940 				      sizeof(*ptr));
941 	if (ret) {
942 		btrfs_abort_transaction(trans, ret);
943 		goto out_free_path;
944 	}
945 
946 	leaf = path->nodes[0];
947 	ptr = btrfs_item_ptr(leaf, path->slots[0],
948 				 struct btrfs_qgroup_status_item);
949 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
950 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
951 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
952 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
953 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
954 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
955 
956 	btrfs_mark_buffer_dirty(leaf);
957 
958 	key.objectid = 0;
959 	key.type = BTRFS_ROOT_REF_KEY;
960 	key.offset = 0;
961 
962 	btrfs_release_path(path);
963 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
964 	if (ret > 0)
965 		goto out_add_root;
966 	if (ret < 0) {
967 		btrfs_abort_transaction(trans, ret);
968 		goto out_free_path;
969 	}
970 
971 	while (1) {
972 		slot = path->slots[0];
973 		leaf = path->nodes[0];
974 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
975 
976 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
977 			ret = add_qgroup_item(trans, quota_root,
978 					      found_key.offset);
979 			if (ret) {
980 				btrfs_abort_transaction(trans, ret);
981 				goto out_free_path;
982 			}
983 
984 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 			if (IS_ERR(qgroup)) {
986 				ret = PTR_ERR(qgroup);
987 				btrfs_abort_transaction(trans, ret);
988 				goto out_free_path;
989 			}
990 		}
991 		ret = btrfs_next_item(tree_root, path);
992 		if (ret < 0) {
993 			btrfs_abort_transaction(trans, ret);
994 			goto out_free_path;
995 		}
996 		if (ret)
997 			break;
998 	}
999 
1000 out_add_root:
1001 	btrfs_release_path(path);
1002 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1003 	if (ret) {
1004 		btrfs_abort_transaction(trans, ret);
1005 		goto out_free_path;
1006 	}
1007 
1008 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 	if (IS_ERR(qgroup)) {
1010 		ret = PTR_ERR(qgroup);
1011 		btrfs_abort_transaction(trans, ret);
1012 		goto out_free_path;
1013 	}
1014 
1015 	ret = btrfs_commit_transaction(trans);
1016 	trans = NULL;
1017 	if (ret)
1018 		goto out_free_path;
1019 
1020 	/*
1021 	 * Set quota enabled flag after committing the transaction, to avoid
1022 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1023 	 * creation.
1024 	 */
1025 	spin_lock(&fs_info->qgroup_lock);
1026 	fs_info->quota_root = quota_root;
1027 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1028 	spin_unlock(&fs_info->qgroup_lock);
1029 
1030 	ret = qgroup_rescan_init(fs_info, 0, 1);
1031 	if (!ret) {
1032 	        qgroup_rescan_zero_tracking(fs_info);
1033 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1034 	                         &fs_info->qgroup_rescan_work);
1035 	}
1036 
1037 out_free_path:
1038 	btrfs_free_path(path);
1039 out_free_root:
1040 	if (ret) {
1041 		free_extent_buffer(quota_root->node);
1042 		free_extent_buffer(quota_root->commit_root);
1043 		kfree(quota_root);
1044 	}
1045 out:
1046 	if (ret) {
1047 		ulist_free(fs_info->qgroup_ulist);
1048 		fs_info->qgroup_ulist = NULL;
1049 		if (trans)
1050 			btrfs_end_transaction(trans);
1051 	}
1052 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1053 	return ret;
1054 }
1055 
1056 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1057 {
1058 	struct btrfs_root *quota_root;
1059 	struct btrfs_trans_handle *trans = NULL;
1060 	int ret = 0;
1061 
1062 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1063 	if (!fs_info->quota_root)
1064 		goto out;
1065 
1066 	/*
1067 	 * 1 For the root item
1068 	 *
1069 	 * We should also reserve enough items for the quota tree deletion in
1070 	 * btrfs_clean_quota_tree but this is not done.
1071 	 */
1072 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1073 	if (IS_ERR(trans)) {
1074 		ret = PTR_ERR(trans);
1075 		goto out;
1076 	}
1077 
1078 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1079 	btrfs_qgroup_wait_for_completion(fs_info, false);
1080 	spin_lock(&fs_info->qgroup_lock);
1081 	quota_root = fs_info->quota_root;
1082 	fs_info->quota_root = NULL;
1083 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1084 	spin_unlock(&fs_info->qgroup_lock);
1085 
1086 	btrfs_free_qgroup_config(fs_info);
1087 
1088 	ret = btrfs_clean_quota_tree(trans, quota_root);
1089 	if (ret) {
1090 		btrfs_abort_transaction(trans, ret);
1091 		goto end_trans;
1092 	}
1093 
1094 	ret = btrfs_del_root(trans, &quota_root->root_key);
1095 	if (ret) {
1096 		btrfs_abort_transaction(trans, ret);
1097 		goto end_trans;
1098 	}
1099 
1100 	list_del(&quota_root->dirty_list);
1101 
1102 	btrfs_tree_lock(quota_root->node);
1103 	btrfs_clean_tree_block(quota_root->node);
1104 	btrfs_tree_unlock(quota_root->node);
1105 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1106 
1107 	free_extent_buffer(quota_root->node);
1108 	free_extent_buffer(quota_root->commit_root);
1109 	kfree(quota_root);
1110 
1111 end_trans:
1112 	ret = btrfs_end_transaction(trans);
1113 out:
1114 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1115 	return ret;
1116 }
1117 
1118 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1119 			 struct btrfs_qgroup *qgroup)
1120 {
1121 	if (list_empty(&qgroup->dirty))
1122 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1123 }
1124 
1125 /*
1126  * The easy accounting, we're updating qgroup relationship whose child qgroup
1127  * only has exclusive extents.
1128  *
1129  * In this case, all exclusive extents will also be exclusive for parent, so
1130  * excl/rfer just get added/removed.
1131  *
1132  * So is qgroup reservation space, which should also be added/removed to
1133  * parent.
1134  * Or when child tries to release reservation space, parent will underflow its
1135  * reservation (for relationship adding case).
1136  *
1137  * Caller should hold fs_info->qgroup_lock.
1138  */
1139 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1140 				    struct ulist *tmp, u64 ref_root,
1141 				    struct btrfs_qgroup *src, int sign)
1142 {
1143 	struct btrfs_qgroup *qgroup;
1144 	struct btrfs_qgroup_list *glist;
1145 	struct ulist_node *unode;
1146 	struct ulist_iterator uiter;
1147 	u64 num_bytes = src->excl;
1148 	int ret = 0;
1149 
1150 	qgroup = find_qgroup_rb(fs_info, ref_root);
1151 	if (!qgroup)
1152 		goto out;
1153 
1154 	qgroup->rfer += sign * num_bytes;
1155 	qgroup->rfer_cmpr += sign * num_bytes;
1156 
1157 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1158 	qgroup->excl += sign * num_bytes;
1159 	qgroup->excl_cmpr += sign * num_bytes;
1160 
1161 	if (sign > 0)
1162 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1163 	else
1164 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1165 
1166 	qgroup_dirty(fs_info, qgroup);
1167 
1168 	/* Get all of the parent groups that contain this qgroup */
1169 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1170 		ret = ulist_add(tmp, glist->group->qgroupid,
1171 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1172 		if (ret < 0)
1173 			goto out;
1174 	}
1175 
1176 	/* Iterate all of the parents and adjust their reference counts */
1177 	ULIST_ITER_INIT(&uiter);
1178 	while ((unode = ulist_next(tmp, &uiter))) {
1179 		qgroup = unode_aux_to_qgroup(unode);
1180 		qgroup->rfer += sign * num_bytes;
1181 		qgroup->rfer_cmpr += sign * num_bytes;
1182 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1183 		qgroup->excl += sign * num_bytes;
1184 		if (sign > 0)
1185 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1186 		else
1187 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1188 		qgroup->excl_cmpr += sign * num_bytes;
1189 		qgroup_dirty(fs_info, qgroup);
1190 
1191 		/* Add any parents of the parents */
1192 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1193 			ret = ulist_add(tmp, glist->group->qgroupid,
1194 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1195 			if (ret < 0)
1196 				goto out;
1197 		}
1198 	}
1199 	ret = 0;
1200 out:
1201 	return ret;
1202 }
1203 
1204 
1205 /*
1206  * Quick path for updating qgroup with only excl refs.
1207  *
1208  * In that case, just update all parent will be enough.
1209  * Or we needs to do a full rescan.
1210  * Caller should also hold fs_info->qgroup_lock.
1211  *
1212  * Return 0 for quick update, return >0 for need to full rescan
1213  * and mark INCONSISTENT flag.
1214  * Return < 0 for other error.
1215  */
1216 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1217 				   struct ulist *tmp, u64 src, u64 dst,
1218 				   int sign)
1219 {
1220 	struct btrfs_qgroup *qgroup;
1221 	int ret = 1;
1222 	int err = 0;
1223 
1224 	qgroup = find_qgroup_rb(fs_info, src);
1225 	if (!qgroup)
1226 		goto out;
1227 	if (qgroup->excl == qgroup->rfer) {
1228 		ret = 0;
1229 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1230 					       qgroup, sign);
1231 		if (err < 0) {
1232 			ret = err;
1233 			goto out;
1234 		}
1235 	}
1236 out:
1237 	if (ret)
1238 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1239 	return ret;
1240 }
1241 
1242 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1243 			      u64 dst)
1244 {
1245 	struct btrfs_fs_info *fs_info = trans->fs_info;
1246 	struct btrfs_root *quota_root;
1247 	struct btrfs_qgroup *parent;
1248 	struct btrfs_qgroup *member;
1249 	struct btrfs_qgroup_list *list;
1250 	struct ulist *tmp;
1251 	int ret = 0;
1252 
1253 	/* Check the level of src and dst first */
1254 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1255 		return -EINVAL;
1256 
1257 	tmp = ulist_alloc(GFP_KERNEL);
1258 	if (!tmp)
1259 		return -ENOMEM;
1260 
1261 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1262 	quota_root = fs_info->quota_root;
1263 	if (!quota_root) {
1264 		ret = -EINVAL;
1265 		goto out;
1266 	}
1267 	member = find_qgroup_rb(fs_info, src);
1268 	parent = find_qgroup_rb(fs_info, dst);
1269 	if (!member || !parent) {
1270 		ret = -EINVAL;
1271 		goto out;
1272 	}
1273 
1274 	/* check if such qgroup relation exist firstly */
1275 	list_for_each_entry(list, &member->groups, next_group) {
1276 		if (list->group == parent) {
1277 			ret = -EEXIST;
1278 			goto out;
1279 		}
1280 	}
1281 
1282 	ret = add_qgroup_relation_item(trans, src, dst);
1283 	if (ret)
1284 		goto out;
1285 
1286 	ret = add_qgroup_relation_item(trans, dst, src);
1287 	if (ret) {
1288 		del_qgroup_relation_item(trans, src, dst);
1289 		goto out;
1290 	}
1291 
1292 	spin_lock(&fs_info->qgroup_lock);
1293 	ret = add_relation_rb(fs_info, src, dst);
1294 	if (ret < 0) {
1295 		spin_unlock(&fs_info->qgroup_lock);
1296 		goto out;
1297 	}
1298 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1299 	spin_unlock(&fs_info->qgroup_lock);
1300 out:
1301 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1302 	ulist_free(tmp);
1303 	return ret;
1304 }
1305 
1306 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1307 				 u64 dst)
1308 {
1309 	struct btrfs_fs_info *fs_info = trans->fs_info;
1310 	struct btrfs_root *quota_root;
1311 	struct btrfs_qgroup *parent;
1312 	struct btrfs_qgroup *member;
1313 	struct btrfs_qgroup_list *list;
1314 	struct ulist *tmp;
1315 	bool found = false;
1316 	int ret = 0;
1317 	int ret2;
1318 
1319 	tmp = ulist_alloc(GFP_KERNEL);
1320 	if (!tmp)
1321 		return -ENOMEM;
1322 
1323 	quota_root = fs_info->quota_root;
1324 	if (!quota_root) {
1325 		ret = -EINVAL;
1326 		goto out;
1327 	}
1328 
1329 	member = find_qgroup_rb(fs_info, src);
1330 	parent = find_qgroup_rb(fs_info, dst);
1331 	/*
1332 	 * The parent/member pair doesn't exist, then try to delete the dead
1333 	 * relation items only.
1334 	 */
1335 	if (!member || !parent)
1336 		goto delete_item;
1337 
1338 	/* check if such qgroup relation exist firstly */
1339 	list_for_each_entry(list, &member->groups, next_group) {
1340 		if (list->group == parent) {
1341 			found = true;
1342 			break;
1343 		}
1344 	}
1345 
1346 delete_item:
1347 	ret = del_qgroup_relation_item(trans, src, dst);
1348 	if (ret < 0 && ret != -ENOENT)
1349 		goto out;
1350 	ret2 = del_qgroup_relation_item(trans, dst, src);
1351 	if (ret2 < 0 && ret2 != -ENOENT)
1352 		goto out;
1353 
1354 	/* At least one deletion succeeded, return 0 */
1355 	if (!ret || !ret2)
1356 		ret = 0;
1357 
1358 	if (found) {
1359 		spin_lock(&fs_info->qgroup_lock);
1360 		del_relation_rb(fs_info, src, dst);
1361 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1362 		spin_unlock(&fs_info->qgroup_lock);
1363 	}
1364 out:
1365 	ulist_free(tmp);
1366 	return ret;
1367 }
1368 
1369 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1370 			      u64 dst)
1371 {
1372 	struct btrfs_fs_info *fs_info = trans->fs_info;
1373 	int ret = 0;
1374 
1375 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1376 	ret = __del_qgroup_relation(trans, src, dst);
1377 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1378 
1379 	return ret;
1380 }
1381 
1382 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1383 {
1384 	struct btrfs_fs_info *fs_info = trans->fs_info;
1385 	struct btrfs_root *quota_root;
1386 	struct btrfs_qgroup *qgroup;
1387 	int ret = 0;
1388 
1389 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1390 	quota_root = fs_info->quota_root;
1391 	if (!quota_root) {
1392 		ret = -EINVAL;
1393 		goto out;
1394 	}
1395 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1396 	if (qgroup) {
1397 		ret = -EEXIST;
1398 		goto out;
1399 	}
1400 
1401 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1402 	if (ret)
1403 		goto out;
1404 
1405 	spin_lock(&fs_info->qgroup_lock);
1406 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1407 	spin_unlock(&fs_info->qgroup_lock);
1408 
1409 	if (IS_ERR(qgroup))
1410 		ret = PTR_ERR(qgroup);
1411 out:
1412 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1413 	return ret;
1414 }
1415 
1416 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1417 {
1418 	struct btrfs_fs_info *fs_info = trans->fs_info;
1419 	struct btrfs_root *quota_root;
1420 	struct btrfs_qgroup *qgroup;
1421 	struct btrfs_qgroup_list *list;
1422 	int ret = 0;
1423 
1424 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1425 	quota_root = fs_info->quota_root;
1426 	if (!quota_root) {
1427 		ret = -EINVAL;
1428 		goto out;
1429 	}
1430 
1431 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1432 	if (!qgroup) {
1433 		ret = -ENOENT;
1434 		goto out;
1435 	}
1436 
1437 	/* Check if there are no children of this qgroup */
1438 	if (!list_empty(&qgroup->members)) {
1439 		ret = -EBUSY;
1440 		goto out;
1441 	}
1442 
1443 	ret = del_qgroup_item(trans, qgroupid);
1444 	if (ret && ret != -ENOENT)
1445 		goto out;
1446 
1447 	while (!list_empty(&qgroup->groups)) {
1448 		list = list_first_entry(&qgroup->groups,
1449 					struct btrfs_qgroup_list, next_group);
1450 		ret = __del_qgroup_relation(trans, qgroupid,
1451 					    list->group->qgroupid);
1452 		if (ret)
1453 			goto out;
1454 	}
1455 
1456 	spin_lock(&fs_info->qgroup_lock);
1457 	del_qgroup_rb(fs_info, qgroupid);
1458 	spin_unlock(&fs_info->qgroup_lock);
1459 out:
1460 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1461 	return ret;
1462 }
1463 
1464 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1465 		       struct btrfs_qgroup_limit *limit)
1466 {
1467 	struct btrfs_fs_info *fs_info = trans->fs_info;
1468 	struct btrfs_root *quota_root;
1469 	struct btrfs_qgroup *qgroup;
1470 	int ret = 0;
1471 	/* Sometimes we would want to clear the limit on this qgroup.
1472 	 * To meet this requirement, we treat the -1 as a special value
1473 	 * which tell kernel to clear the limit on this qgroup.
1474 	 */
1475 	const u64 CLEAR_VALUE = -1;
1476 
1477 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1478 	quota_root = fs_info->quota_root;
1479 	if (!quota_root) {
1480 		ret = -EINVAL;
1481 		goto out;
1482 	}
1483 
1484 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1485 	if (!qgroup) {
1486 		ret = -ENOENT;
1487 		goto out;
1488 	}
1489 
1490 	spin_lock(&fs_info->qgroup_lock);
1491 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1492 		if (limit->max_rfer == CLEAR_VALUE) {
1493 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1494 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1495 			qgroup->max_rfer = 0;
1496 		} else {
1497 			qgroup->max_rfer = limit->max_rfer;
1498 		}
1499 	}
1500 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1501 		if (limit->max_excl == CLEAR_VALUE) {
1502 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1503 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1504 			qgroup->max_excl = 0;
1505 		} else {
1506 			qgroup->max_excl = limit->max_excl;
1507 		}
1508 	}
1509 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1510 		if (limit->rsv_rfer == CLEAR_VALUE) {
1511 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1512 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1513 			qgroup->rsv_rfer = 0;
1514 		} else {
1515 			qgroup->rsv_rfer = limit->rsv_rfer;
1516 		}
1517 	}
1518 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1519 		if (limit->rsv_excl == CLEAR_VALUE) {
1520 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1521 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1522 			qgroup->rsv_excl = 0;
1523 		} else {
1524 			qgroup->rsv_excl = limit->rsv_excl;
1525 		}
1526 	}
1527 	qgroup->lim_flags |= limit->flags;
1528 
1529 	spin_unlock(&fs_info->qgroup_lock);
1530 
1531 	ret = update_qgroup_limit_item(trans, qgroup);
1532 	if (ret) {
1533 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1534 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1535 		       qgroupid);
1536 	}
1537 
1538 out:
1539 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1540 	return ret;
1541 }
1542 
1543 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1544 				struct btrfs_delayed_ref_root *delayed_refs,
1545 				struct btrfs_qgroup_extent_record *record)
1546 {
1547 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1548 	struct rb_node *parent_node = NULL;
1549 	struct btrfs_qgroup_extent_record *entry;
1550 	u64 bytenr = record->bytenr;
1551 
1552 	lockdep_assert_held(&delayed_refs->lock);
1553 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1554 
1555 	while (*p) {
1556 		parent_node = *p;
1557 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1558 				 node);
1559 		if (bytenr < entry->bytenr) {
1560 			p = &(*p)->rb_left;
1561 		} else if (bytenr > entry->bytenr) {
1562 			p = &(*p)->rb_right;
1563 		} else {
1564 			if (record->data_rsv && !entry->data_rsv) {
1565 				entry->data_rsv = record->data_rsv;
1566 				entry->data_rsv_refroot =
1567 					record->data_rsv_refroot;
1568 			}
1569 			return 1;
1570 		}
1571 	}
1572 
1573 	rb_link_node(&record->node, parent_node, p);
1574 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1575 	return 0;
1576 }
1577 
1578 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1579 				   struct btrfs_qgroup_extent_record *qrecord)
1580 {
1581 	struct ulist *old_root;
1582 	u64 bytenr = qrecord->bytenr;
1583 	int ret;
1584 
1585 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1586 	if (ret < 0) {
1587 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1588 		btrfs_warn(fs_info,
1589 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1590 			ret);
1591 		return 0;
1592 	}
1593 
1594 	/*
1595 	 * Here we don't need to get the lock of
1596 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1597 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1598 	 *
1599 	 * So modifying qrecord->old_roots is safe here
1600 	 */
1601 	qrecord->old_roots = old_root;
1602 	return 0;
1603 }
1604 
1605 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1606 			      u64 num_bytes, gfp_t gfp_flag)
1607 {
1608 	struct btrfs_fs_info *fs_info = trans->fs_info;
1609 	struct btrfs_qgroup_extent_record *record;
1610 	struct btrfs_delayed_ref_root *delayed_refs;
1611 	int ret;
1612 
1613 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1614 	    || bytenr == 0 || num_bytes == 0)
1615 		return 0;
1616 	record = kzalloc(sizeof(*record), gfp_flag);
1617 	if (!record)
1618 		return -ENOMEM;
1619 
1620 	delayed_refs = &trans->transaction->delayed_refs;
1621 	record->bytenr = bytenr;
1622 	record->num_bytes = num_bytes;
1623 	record->old_roots = NULL;
1624 
1625 	spin_lock(&delayed_refs->lock);
1626 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1627 	spin_unlock(&delayed_refs->lock);
1628 	if (ret > 0) {
1629 		kfree(record);
1630 		return 0;
1631 	}
1632 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1633 }
1634 
1635 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1636 				  struct extent_buffer *eb)
1637 {
1638 	struct btrfs_fs_info *fs_info = trans->fs_info;
1639 	int nr = btrfs_header_nritems(eb);
1640 	int i, extent_type, ret;
1641 	struct btrfs_key key;
1642 	struct btrfs_file_extent_item *fi;
1643 	u64 bytenr, num_bytes;
1644 
1645 	/* We can be called directly from walk_up_proc() */
1646 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1647 		return 0;
1648 
1649 	for (i = 0; i < nr; i++) {
1650 		btrfs_item_key_to_cpu(eb, &key, i);
1651 
1652 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1653 			continue;
1654 
1655 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1656 		/* filter out non qgroup-accountable extents  */
1657 		extent_type = btrfs_file_extent_type(eb, fi);
1658 
1659 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1660 			continue;
1661 
1662 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1663 		if (!bytenr)
1664 			continue;
1665 
1666 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1667 
1668 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1669 						GFP_NOFS);
1670 		if (ret)
1671 			return ret;
1672 	}
1673 	cond_resched();
1674 	return 0;
1675 }
1676 
1677 /*
1678  * Walk up the tree from the bottom, freeing leaves and any interior
1679  * nodes which have had all slots visited. If a node (leaf or
1680  * interior) is freed, the node above it will have it's slot
1681  * incremented. The root node will never be freed.
1682  *
1683  * At the end of this function, we should have a path which has all
1684  * slots incremented to the next position for a search. If we need to
1685  * read a new node it will be NULL and the node above it will have the
1686  * correct slot selected for a later read.
1687  *
1688  * If we increment the root nodes slot counter past the number of
1689  * elements, 1 is returned to signal completion of the search.
1690  */
1691 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1692 {
1693 	int level = 0;
1694 	int nr, slot;
1695 	struct extent_buffer *eb;
1696 
1697 	if (root_level == 0)
1698 		return 1;
1699 
1700 	while (level <= root_level) {
1701 		eb = path->nodes[level];
1702 		nr = btrfs_header_nritems(eb);
1703 		path->slots[level]++;
1704 		slot = path->slots[level];
1705 		if (slot >= nr || level == 0) {
1706 			/*
1707 			 * Don't free the root -  we will detect this
1708 			 * condition after our loop and return a
1709 			 * positive value for caller to stop walking the tree.
1710 			 */
1711 			if (level != root_level) {
1712 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1713 				path->locks[level] = 0;
1714 
1715 				free_extent_buffer(eb);
1716 				path->nodes[level] = NULL;
1717 				path->slots[level] = 0;
1718 			}
1719 		} else {
1720 			/*
1721 			 * We have a valid slot to walk back down
1722 			 * from. Stop here so caller can process these
1723 			 * new nodes.
1724 			 */
1725 			break;
1726 		}
1727 
1728 		level++;
1729 	}
1730 
1731 	eb = path->nodes[root_level];
1732 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1733 		return 1;
1734 
1735 	return 0;
1736 }
1737 
1738 /*
1739  * Helper function to trace a subtree tree block swap.
1740  *
1741  * The swap will happen in highest tree block, but there may be a lot of
1742  * tree blocks involved.
1743  *
1744  * For example:
1745  *  OO = Old tree blocks
1746  *  NN = New tree blocks allocated during balance
1747  *
1748  *           File tree (257)                  Reloc tree for 257
1749  * L2              OO                                NN
1750  *               /    \                            /    \
1751  * L1          OO      OO (a)                    OO      NN (a)
1752  *            / \     / \                       / \     / \
1753  * L0       OO   OO OO   OO                   OO   OO NN   NN
1754  *                  (b)  (c)                          (b)  (c)
1755  *
1756  * When calling qgroup_trace_extent_swap(), we will pass:
1757  * @src_eb = OO(a)
1758  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1759  * @dst_level = 0
1760  * @root_level = 1
1761  *
1762  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1763  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1764  *
1765  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1766  *
1767  * 1) Tree search from @src_eb
1768  *    It should acts as a simplified btrfs_search_slot().
1769  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1770  *    (first key).
1771  *
1772  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1773  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1774  *    They should be marked during previous (@dst_level = 1) iteration.
1775  *
1776  * 3) Mark file extents in leaves dirty
1777  *    We don't have good way to pick out new file extents only.
1778  *    So we still follow the old method by scanning all file extents in
1779  *    the leave.
1780  *
1781  * This function can free us from keeping two paths, thus later we only need
1782  * to care about how to iterate all new tree blocks in reloc tree.
1783  */
1784 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1785 				    struct extent_buffer *src_eb,
1786 				    struct btrfs_path *dst_path,
1787 				    int dst_level, int root_level,
1788 				    bool trace_leaf)
1789 {
1790 	struct btrfs_key key;
1791 	struct btrfs_path *src_path;
1792 	struct btrfs_fs_info *fs_info = trans->fs_info;
1793 	u32 nodesize = fs_info->nodesize;
1794 	int cur_level = root_level;
1795 	int ret;
1796 
1797 	BUG_ON(dst_level > root_level);
1798 	/* Level mismatch */
1799 	if (btrfs_header_level(src_eb) != root_level)
1800 		return -EINVAL;
1801 
1802 	src_path = btrfs_alloc_path();
1803 	if (!src_path) {
1804 		ret = -ENOMEM;
1805 		goto out;
1806 	}
1807 
1808 	if (dst_level)
1809 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1810 	else
1811 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1812 
1813 	/* For src_path */
1814 	atomic_inc(&src_eb->refs);
1815 	src_path->nodes[root_level] = src_eb;
1816 	src_path->slots[root_level] = dst_path->slots[root_level];
1817 	src_path->locks[root_level] = 0;
1818 
1819 	/* A simplified version of btrfs_search_slot() */
1820 	while (cur_level >= dst_level) {
1821 		struct btrfs_key src_key;
1822 		struct btrfs_key dst_key;
1823 
1824 		if (src_path->nodes[cur_level] == NULL) {
1825 			struct btrfs_key first_key;
1826 			struct extent_buffer *eb;
1827 			int parent_slot;
1828 			u64 child_gen;
1829 			u64 child_bytenr;
1830 
1831 			eb = src_path->nodes[cur_level + 1];
1832 			parent_slot = src_path->slots[cur_level + 1];
1833 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1834 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1835 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1836 
1837 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1838 					     cur_level, &first_key);
1839 			if (IS_ERR(eb)) {
1840 				ret = PTR_ERR(eb);
1841 				goto out;
1842 			} else if (!extent_buffer_uptodate(eb)) {
1843 				free_extent_buffer(eb);
1844 				ret = -EIO;
1845 				goto out;
1846 			}
1847 
1848 			src_path->nodes[cur_level] = eb;
1849 
1850 			btrfs_tree_read_lock(eb);
1851 			btrfs_set_lock_blocking_read(eb);
1852 			src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1853 		}
1854 
1855 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1856 		if (cur_level) {
1857 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1858 					&dst_key, dst_path->slots[cur_level]);
1859 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1860 					&src_key, src_path->slots[cur_level]);
1861 		} else {
1862 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1863 					&dst_key, dst_path->slots[cur_level]);
1864 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1865 					&src_key, src_path->slots[cur_level]);
1866 		}
1867 		/* Content mismatch, something went wrong */
1868 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1869 			ret = -ENOENT;
1870 			goto out;
1871 		}
1872 		cur_level--;
1873 	}
1874 
1875 	/*
1876 	 * Now both @dst_path and @src_path have been populated, record the tree
1877 	 * blocks for qgroup accounting.
1878 	 */
1879 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1880 			nodesize, GFP_NOFS);
1881 	if (ret < 0)
1882 		goto out;
1883 	ret = btrfs_qgroup_trace_extent(trans,
1884 			dst_path->nodes[dst_level]->start,
1885 			nodesize, GFP_NOFS);
1886 	if (ret < 0)
1887 		goto out;
1888 
1889 	/* Record leaf file extents */
1890 	if (dst_level == 0 && trace_leaf) {
1891 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1892 		if (ret < 0)
1893 			goto out;
1894 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1895 	}
1896 out:
1897 	btrfs_free_path(src_path);
1898 	return ret;
1899 }
1900 
1901 /*
1902  * Helper function to do recursive generation-aware depth-first search, to
1903  * locate all new tree blocks in a subtree of reloc tree.
1904  *
1905  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1906  *         reloc tree
1907  * L2         NN (a)
1908  *          /    \
1909  * L1    OO        NN (b)
1910  *      /  \      /  \
1911  * L0  OO  OO    OO  NN
1912  *               (c) (d)
1913  * If we pass:
1914  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1915  * @cur_level = 1
1916  * @root_level = 1
1917  *
1918  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1919  * above tree blocks along with their counter parts in file tree.
1920  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1921  * won't affect OO(c).
1922  */
1923 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1924 					   struct extent_buffer *src_eb,
1925 					   struct btrfs_path *dst_path,
1926 					   int cur_level, int root_level,
1927 					   u64 last_snapshot, bool trace_leaf)
1928 {
1929 	struct btrfs_fs_info *fs_info = trans->fs_info;
1930 	struct extent_buffer *eb;
1931 	bool need_cleanup = false;
1932 	int ret = 0;
1933 	int i;
1934 
1935 	/* Level sanity check */
1936 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1937 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1938 	    root_level < cur_level) {
1939 		btrfs_err_rl(fs_info,
1940 			"%s: bad levels, cur_level=%d root_level=%d",
1941 			__func__, cur_level, root_level);
1942 		return -EUCLEAN;
1943 	}
1944 
1945 	/* Read the tree block if needed */
1946 	if (dst_path->nodes[cur_level] == NULL) {
1947 		struct btrfs_key first_key;
1948 		int parent_slot;
1949 		u64 child_gen;
1950 		u64 child_bytenr;
1951 
1952 		/*
1953 		 * dst_path->nodes[root_level] must be initialized before
1954 		 * calling this function.
1955 		 */
1956 		if (cur_level == root_level) {
1957 			btrfs_err_rl(fs_info,
1958 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1959 				__func__, root_level, root_level, cur_level);
1960 			return -EUCLEAN;
1961 		}
1962 
1963 		/*
1964 		 * We need to get child blockptr/gen from parent before we can
1965 		 * read it.
1966 		  */
1967 		eb = dst_path->nodes[cur_level + 1];
1968 		parent_slot = dst_path->slots[cur_level + 1];
1969 		child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1970 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1971 		btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1972 
1973 		/* This node is old, no need to trace */
1974 		if (child_gen < last_snapshot)
1975 			goto out;
1976 
1977 		eb = read_tree_block(fs_info, child_bytenr, child_gen,
1978 				     cur_level, &first_key);
1979 		if (IS_ERR(eb)) {
1980 			ret = PTR_ERR(eb);
1981 			goto out;
1982 		} else if (!extent_buffer_uptodate(eb)) {
1983 			free_extent_buffer(eb);
1984 			ret = -EIO;
1985 			goto out;
1986 		}
1987 
1988 		dst_path->nodes[cur_level] = eb;
1989 		dst_path->slots[cur_level] = 0;
1990 
1991 		btrfs_tree_read_lock(eb);
1992 		btrfs_set_lock_blocking_read(eb);
1993 		dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1994 		need_cleanup = true;
1995 	}
1996 
1997 	/* Now record this tree block and its counter part for qgroups */
1998 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1999 				       root_level, trace_leaf);
2000 	if (ret < 0)
2001 		goto cleanup;
2002 
2003 	eb = dst_path->nodes[cur_level];
2004 
2005 	if (cur_level > 0) {
2006 		/* Iterate all child tree blocks */
2007 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2008 			/* Skip old tree blocks as they won't be swapped */
2009 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2010 				continue;
2011 			dst_path->slots[cur_level] = i;
2012 
2013 			/* Recursive call (at most 7 times) */
2014 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2015 					dst_path, cur_level - 1, root_level,
2016 					last_snapshot, trace_leaf);
2017 			if (ret < 0)
2018 				goto cleanup;
2019 		}
2020 	}
2021 
2022 cleanup:
2023 	if (need_cleanup) {
2024 		/* Clean up */
2025 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2026 				     dst_path->locks[cur_level]);
2027 		free_extent_buffer(dst_path->nodes[cur_level]);
2028 		dst_path->nodes[cur_level] = NULL;
2029 		dst_path->slots[cur_level] = 0;
2030 		dst_path->locks[cur_level] = 0;
2031 	}
2032 out:
2033 	return ret;
2034 }
2035 
2036 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2037 				struct extent_buffer *src_eb,
2038 				struct extent_buffer *dst_eb,
2039 				u64 last_snapshot, bool trace_leaf)
2040 {
2041 	struct btrfs_fs_info *fs_info = trans->fs_info;
2042 	struct btrfs_path *dst_path = NULL;
2043 	int level;
2044 	int ret;
2045 
2046 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2047 		return 0;
2048 
2049 	/* Wrong parameter order */
2050 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2051 		btrfs_err_rl(fs_info,
2052 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2053 			     btrfs_header_generation(src_eb),
2054 			     btrfs_header_generation(dst_eb));
2055 		return -EUCLEAN;
2056 	}
2057 
2058 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2059 		ret = -EIO;
2060 		goto out;
2061 	}
2062 
2063 	level = btrfs_header_level(dst_eb);
2064 	dst_path = btrfs_alloc_path();
2065 	if (!dst_path) {
2066 		ret = -ENOMEM;
2067 		goto out;
2068 	}
2069 	/* For dst_path */
2070 	atomic_inc(&dst_eb->refs);
2071 	dst_path->nodes[level] = dst_eb;
2072 	dst_path->slots[level] = 0;
2073 	dst_path->locks[level] = 0;
2074 
2075 	/* Do the generation aware breadth-first search */
2076 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2077 					      level, last_snapshot, trace_leaf);
2078 	if (ret < 0)
2079 		goto out;
2080 	ret = 0;
2081 
2082 out:
2083 	btrfs_free_path(dst_path);
2084 	if (ret < 0)
2085 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2086 	return ret;
2087 }
2088 
2089 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2090 			       struct extent_buffer *root_eb,
2091 			       u64 root_gen, int root_level)
2092 {
2093 	struct btrfs_fs_info *fs_info = trans->fs_info;
2094 	int ret = 0;
2095 	int level;
2096 	struct extent_buffer *eb = root_eb;
2097 	struct btrfs_path *path = NULL;
2098 
2099 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2100 	BUG_ON(root_eb == NULL);
2101 
2102 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2103 		return 0;
2104 
2105 	if (!extent_buffer_uptodate(root_eb)) {
2106 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2107 		if (ret)
2108 			goto out;
2109 	}
2110 
2111 	if (root_level == 0) {
2112 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2113 		goto out;
2114 	}
2115 
2116 	path = btrfs_alloc_path();
2117 	if (!path)
2118 		return -ENOMEM;
2119 
2120 	/*
2121 	 * Walk down the tree.  Missing extent blocks are filled in as
2122 	 * we go. Metadata is accounted every time we read a new
2123 	 * extent block.
2124 	 *
2125 	 * When we reach a leaf, we account for file extent items in it,
2126 	 * walk back up the tree (adjusting slot pointers as we go)
2127 	 * and restart the search process.
2128 	 */
2129 	atomic_inc(&root_eb->refs);	/* For path */
2130 	path->nodes[root_level] = root_eb;
2131 	path->slots[root_level] = 0;
2132 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2133 walk_down:
2134 	level = root_level;
2135 	while (level >= 0) {
2136 		if (path->nodes[level] == NULL) {
2137 			struct btrfs_key first_key;
2138 			int parent_slot;
2139 			u64 child_gen;
2140 			u64 child_bytenr;
2141 
2142 			/*
2143 			 * We need to get child blockptr/gen from parent before
2144 			 * we can read it.
2145 			  */
2146 			eb = path->nodes[level + 1];
2147 			parent_slot = path->slots[level + 1];
2148 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2149 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2150 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2151 
2152 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
2153 					     level, &first_key);
2154 			if (IS_ERR(eb)) {
2155 				ret = PTR_ERR(eb);
2156 				goto out;
2157 			} else if (!extent_buffer_uptodate(eb)) {
2158 				free_extent_buffer(eb);
2159 				ret = -EIO;
2160 				goto out;
2161 			}
2162 
2163 			path->nodes[level] = eb;
2164 			path->slots[level] = 0;
2165 
2166 			btrfs_tree_read_lock(eb);
2167 			btrfs_set_lock_blocking_read(eb);
2168 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2169 
2170 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2171 							fs_info->nodesize,
2172 							GFP_NOFS);
2173 			if (ret)
2174 				goto out;
2175 		}
2176 
2177 		if (level == 0) {
2178 			ret = btrfs_qgroup_trace_leaf_items(trans,
2179 							    path->nodes[level]);
2180 			if (ret)
2181 				goto out;
2182 
2183 			/* Nonzero return here means we completed our search */
2184 			ret = adjust_slots_upwards(path, root_level);
2185 			if (ret)
2186 				break;
2187 
2188 			/* Restart search with new slots */
2189 			goto walk_down;
2190 		}
2191 
2192 		level--;
2193 	}
2194 
2195 	ret = 0;
2196 out:
2197 	btrfs_free_path(path);
2198 
2199 	return ret;
2200 }
2201 
2202 #define UPDATE_NEW	0
2203 #define UPDATE_OLD	1
2204 /*
2205  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2206  */
2207 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2208 				struct ulist *roots, struct ulist *tmp,
2209 				struct ulist *qgroups, u64 seq, int update_old)
2210 {
2211 	struct ulist_node *unode;
2212 	struct ulist_iterator uiter;
2213 	struct ulist_node *tmp_unode;
2214 	struct ulist_iterator tmp_uiter;
2215 	struct btrfs_qgroup *qg;
2216 	int ret = 0;
2217 
2218 	if (!roots)
2219 		return 0;
2220 	ULIST_ITER_INIT(&uiter);
2221 	while ((unode = ulist_next(roots, &uiter))) {
2222 		qg = find_qgroup_rb(fs_info, unode->val);
2223 		if (!qg)
2224 			continue;
2225 
2226 		ulist_reinit(tmp);
2227 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2228 				GFP_ATOMIC);
2229 		if (ret < 0)
2230 			return ret;
2231 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2232 		if (ret < 0)
2233 			return ret;
2234 		ULIST_ITER_INIT(&tmp_uiter);
2235 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2236 			struct btrfs_qgroup_list *glist;
2237 
2238 			qg = unode_aux_to_qgroup(tmp_unode);
2239 			if (update_old)
2240 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2241 			else
2242 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2243 			list_for_each_entry(glist, &qg->groups, next_group) {
2244 				ret = ulist_add(qgroups, glist->group->qgroupid,
2245 						qgroup_to_aux(glist->group),
2246 						GFP_ATOMIC);
2247 				if (ret < 0)
2248 					return ret;
2249 				ret = ulist_add(tmp, glist->group->qgroupid,
2250 						qgroup_to_aux(glist->group),
2251 						GFP_ATOMIC);
2252 				if (ret < 0)
2253 					return ret;
2254 			}
2255 		}
2256 	}
2257 	return 0;
2258 }
2259 
2260 /*
2261  * Update qgroup rfer/excl counters.
2262  * Rfer update is easy, codes can explain themselves.
2263  *
2264  * Excl update is tricky, the update is split into 2 part.
2265  * Part 1: Possible exclusive <-> sharing detect:
2266  *	|	A	|	!A	|
2267  *  -------------------------------------
2268  *  B	|	*	|	-	|
2269  *  -------------------------------------
2270  *  !B	|	+	|	**	|
2271  *  -------------------------------------
2272  *
2273  * Conditions:
2274  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2275  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2276  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2277  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2278  *
2279  * Results:
2280  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2281  * *: Definitely not changed.		**: Possible unchanged.
2282  *
2283  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2284  *
2285  * To make the logic clear, we first use condition A and B to split
2286  * combination into 4 results.
2287  *
2288  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2289  * only on variant maybe 0.
2290  *
2291  * Lastly, check result **, since there are 2 variants maybe 0, split them
2292  * again(2x2).
2293  * But this time we don't need to consider other things, the codes and logic
2294  * is easy to understand now.
2295  */
2296 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2297 				  struct ulist *qgroups,
2298 				  u64 nr_old_roots,
2299 				  u64 nr_new_roots,
2300 				  u64 num_bytes, u64 seq)
2301 {
2302 	struct ulist_node *unode;
2303 	struct ulist_iterator uiter;
2304 	struct btrfs_qgroup *qg;
2305 	u64 cur_new_count, cur_old_count;
2306 
2307 	ULIST_ITER_INIT(&uiter);
2308 	while ((unode = ulist_next(qgroups, &uiter))) {
2309 		bool dirty = false;
2310 
2311 		qg = unode_aux_to_qgroup(unode);
2312 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2313 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2314 
2315 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2316 					     cur_new_count);
2317 
2318 		/* Rfer update part */
2319 		if (cur_old_count == 0 && cur_new_count > 0) {
2320 			qg->rfer += num_bytes;
2321 			qg->rfer_cmpr += num_bytes;
2322 			dirty = true;
2323 		}
2324 		if (cur_old_count > 0 && cur_new_count == 0) {
2325 			qg->rfer -= num_bytes;
2326 			qg->rfer_cmpr -= num_bytes;
2327 			dirty = true;
2328 		}
2329 
2330 		/* Excl update part */
2331 		/* Exclusive/none -> shared case */
2332 		if (cur_old_count == nr_old_roots &&
2333 		    cur_new_count < nr_new_roots) {
2334 			/* Exclusive -> shared */
2335 			if (cur_old_count != 0) {
2336 				qg->excl -= num_bytes;
2337 				qg->excl_cmpr -= num_bytes;
2338 				dirty = true;
2339 			}
2340 		}
2341 
2342 		/* Shared -> exclusive/none case */
2343 		if (cur_old_count < nr_old_roots &&
2344 		    cur_new_count == nr_new_roots) {
2345 			/* Shared->exclusive */
2346 			if (cur_new_count != 0) {
2347 				qg->excl += num_bytes;
2348 				qg->excl_cmpr += num_bytes;
2349 				dirty = true;
2350 			}
2351 		}
2352 
2353 		/* Exclusive/none -> exclusive/none case */
2354 		if (cur_old_count == nr_old_roots &&
2355 		    cur_new_count == nr_new_roots) {
2356 			if (cur_old_count == 0) {
2357 				/* None -> exclusive/none */
2358 
2359 				if (cur_new_count != 0) {
2360 					/* None -> exclusive */
2361 					qg->excl += num_bytes;
2362 					qg->excl_cmpr += num_bytes;
2363 					dirty = true;
2364 				}
2365 				/* None -> none, nothing changed */
2366 			} else {
2367 				/* Exclusive -> exclusive/none */
2368 
2369 				if (cur_new_count == 0) {
2370 					/* Exclusive -> none */
2371 					qg->excl -= num_bytes;
2372 					qg->excl_cmpr -= num_bytes;
2373 					dirty = true;
2374 				}
2375 				/* Exclusive -> exclusive, nothing changed */
2376 			}
2377 		}
2378 
2379 		if (dirty)
2380 			qgroup_dirty(fs_info, qg);
2381 	}
2382 	return 0;
2383 }
2384 
2385 /*
2386  * Check if the @roots potentially is a list of fs tree roots
2387  *
2388  * Return 0 for definitely not a fs/subvol tree roots ulist
2389  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2390  *          one as well)
2391  */
2392 static int maybe_fs_roots(struct ulist *roots)
2393 {
2394 	struct ulist_node *unode;
2395 	struct ulist_iterator uiter;
2396 
2397 	/* Empty one, still possible for fs roots */
2398 	if (!roots || roots->nnodes == 0)
2399 		return 1;
2400 
2401 	ULIST_ITER_INIT(&uiter);
2402 	unode = ulist_next(roots, &uiter);
2403 	if (!unode)
2404 		return 1;
2405 
2406 	/*
2407 	 * If it contains fs tree roots, then it must belong to fs/subvol
2408 	 * trees.
2409 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2410 	 */
2411 	return is_fstree(unode->val);
2412 }
2413 
2414 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2415 				u64 num_bytes, struct ulist *old_roots,
2416 				struct ulist *new_roots)
2417 {
2418 	struct btrfs_fs_info *fs_info = trans->fs_info;
2419 	struct ulist *qgroups = NULL;
2420 	struct ulist *tmp = NULL;
2421 	u64 seq;
2422 	u64 nr_new_roots = 0;
2423 	u64 nr_old_roots = 0;
2424 	int ret = 0;
2425 
2426 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2427 		return 0;
2428 
2429 	if (new_roots) {
2430 		if (!maybe_fs_roots(new_roots))
2431 			goto out_free;
2432 		nr_new_roots = new_roots->nnodes;
2433 	}
2434 	if (old_roots) {
2435 		if (!maybe_fs_roots(old_roots))
2436 			goto out_free;
2437 		nr_old_roots = old_roots->nnodes;
2438 	}
2439 
2440 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2441 	if (nr_old_roots == 0 && nr_new_roots == 0)
2442 		goto out_free;
2443 
2444 	BUG_ON(!fs_info->quota_root);
2445 
2446 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2447 					num_bytes, nr_old_roots, nr_new_roots);
2448 
2449 	qgroups = ulist_alloc(GFP_NOFS);
2450 	if (!qgroups) {
2451 		ret = -ENOMEM;
2452 		goto out_free;
2453 	}
2454 	tmp = ulist_alloc(GFP_NOFS);
2455 	if (!tmp) {
2456 		ret = -ENOMEM;
2457 		goto out_free;
2458 	}
2459 
2460 	mutex_lock(&fs_info->qgroup_rescan_lock);
2461 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2462 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2463 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2464 			ret = 0;
2465 			goto out_free;
2466 		}
2467 	}
2468 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2469 
2470 	spin_lock(&fs_info->qgroup_lock);
2471 	seq = fs_info->qgroup_seq;
2472 
2473 	/* Update old refcnts using old_roots */
2474 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2475 				   UPDATE_OLD);
2476 	if (ret < 0)
2477 		goto out;
2478 
2479 	/* Update new refcnts using new_roots */
2480 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2481 				   UPDATE_NEW);
2482 	if (ret < 0)
2483 		goto out;
2484 
2485 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2486 			       num_bytes, seq);
2487 
2488 	/*
2489 	 * Bump qgroup_seq to avoid seq overlap
2490 	 */
2491 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2492 out:
2493 	spin_unlock(&fs_info->qgroup_lock);
2494 out_free:
2495 	ulist_free(tmp);
2496 	ulist_free(qgroups);
2497 	ulist_free(old_roots);
2498 	ulist_free(new_roots);
2499 	return ret;
2500 }
2501 
2502 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2503 {
2504 	struct btrfs_fs_info *fs_info = trans->fs_info;
2505 	struct btrfs_qgroup_extent_record *record;
2506 	struct btrfs_delayed_ref_root *delayed_refs;
2507 	struct ulist *new_roots = NULL;
2508 	struct rb_node *node;
2509 	u64 num_dirty_extents = 0;
2510 	u64 qgroup_to_skip;
2511 	int ret = 0;
2512 
2513 	delayed_refs = &trans->transaction->delayed_refs;
2514 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2515 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2516 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2517 				  node);
2518 
2519 		num_dirty_extents++;
2520 		trace_btrfs_qgroup_account_extents(fs_info, record);
2521 
2522 		if (!ret) {
2523 			/*
2524 			 * Old roots should be searched when inserting qgroup
2525 			 * extent record
2526 			 */
2527 			if (WARN_ON(!record->old_roots)) {
2528 				/* Search commit root to find old_roots */
2529 				ret = btrfs_find_all_roots(NULL, fs_info,
2530 						record->bytenr, 0,
2531 						&record->old_roots, false);
2532 				if (ret < 0)
2533 					goto cleanup;
2534 			}
2535 
2536 			/* Free the reserved data space */
2537 			btrfs_qgroup_free_refroot(fs_info,
2538 					record->data_rsv_refroot,
2539 					record->data_rsv,
2540 					BTRFS_QGROUP_RSV_DATA);
2541 			/*
2542 			 * Use SEQ_LAST as time_seq to do special search, which
2543 			 * doesn't lock tree or delayed_refs and search current
2544 			 * root. It's safe inside commit_transaction().
2545 			 */
2546 			ret = btrfs_find_all_roots(trans, fs_info,
2547 				record->bytenr, SEQ_LAST, &new_roots, false);
2548 			if (ret < 0)
2549 				goto cleanup;
2550 			if (qgroup_to_skip) {
2551 				ulist_del(new_roots, qgroup_to_skip, 0);
2552 				ulist_del(record->old_roots, qgroup_to_skip,
2553 					  0);
2554 			}
2555 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2556 							  record->num_bytes,
2557 							  record->old_roots,
2558 							  new_roots);
2559 			record->old_roots = NULL;
2560 			new_roots = NULL;
2561 		}
2562 cleanup:
2563 		ulist_free(record->old_roots);
2564 		ulist_free(new_roots);
2565 		new_roots = NULL;
2566 		rb_erase(node, &delayed_refs->dirty_extent_root);
2567 		kfree(record);
2568 
2569 	}
2570 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2571 				       num_dirty_extents);
2572 	return ret;
2573 }
2574 
2575 /*
2576  * called from commit_transaction. Writes all changed qgroups to disk.
2577  */
2578 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2579 {
2580 	struct btrfs_fs_info *fs_info = trans->fs_info;
2581 	struct btrfs_root *quota_root = fs_info->quota_root;
2582 	int ret = 0;
2583 
2584 	if (!quota_root)
2585 		return ret;
2586 
2587 	spin_lock(&fs_info->qgroup_lock);
2588 	while (!list_empty(&fs_info->dirty_qgroups)) {
2589 		struct btrfs_qgroup *qgroup;
2590 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2591 					  struct btrfs_qgroup, dirty);
2592 		list_del_init(&qgroup->dirty);
2593 		spin_unlock(&fs_info->qgroup_lock);
2594 		ret = update_qgroup_info_item(trans, qgroup);
2595 		if (ret)
2596 			fs_info->qgroup_flags |=
2597 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2598 		ret = update_qgroup_limit_item(trans, qgroup);
2599 		if (ret)
2600 			fs_info->qgroup_flags |=
2601 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2602 		spin_lock(&fs_info->qgroup_lock);
2603 	}
2604 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2605 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2606 	else
2607 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2608 	spin_unlock(&fs_info->qgroup_lock);
2609 
2610 	ret = update_qgroup_status_item(trans);
2611 	if (ret)
2612 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2613 
2614 	return ret;
2615 }
2616 
2617 /*
2618  * Copy the accounting information between qgroups. This is necessary
2619  * when a snapshot or a subvolume is created. Throwing an error will
2620  * cause a transaction abort so we take extra care here to only error
2621  * when a readonly fs is a reasonable outcome.
2622  */
2623 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2624 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2625 {
2626 	int ret = 0;
2627 	int i;
2628 	u64 *i_qgroups;
2629 	bool committing = false;
2630 	struct btrfs_fs_info *fs_info = trans->fs_info;
2631 	struct btrfs_root *quota_root;
2632 	struct btrfs_qgroup *srcgroup;
2633 	struct btrfs_qgroup *dstgroup;
2634 	u32 level_size = 0;
2635 	u64 nums;
2636 
2637 	/*
2638 	 * There are only two callers of this function.
2639 	 *
2640 	 * One in create_subvol() in the ioctl context, which needs to hold
2641 	 * the qgroup_ioctl_lock.
2642 	 *
2643 	 * The other one in create_pending_snapshot() where no other qgroup
2644 	 * code can modify the fs as they all need to either start a new trans
2645 	 * or hold a trans handler, thus we don't need to hold
2646 	 * qgroup_ioctl_lock.
2647 	 * This would avoid long and complex lock chain and make lockdep happy.
2648 	 */
2649 	spin_lock(&fs_info->trans_lock);
2650 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2651 		committing = true;
2652 	spin_unlock(&fs_info->trans_lock);
2653 
2654 	if (!committing)
2655 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2656 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2657 		goto out;
2658 
2659 	quota_root = fs_info->quota_root;
2660 	if (!quota_root) {
2661 		ret = -EINVAL;
2662 		goto out;
2663 	}
2664 
2665 	if (inherit) {
2666 		i_qgroups = (u64 *)(inherit + 1);
2667 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2668 		       2 * inherit->num_excl_copies;
2669 		for (i = 0; i < nums; ++i) {
2670 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2671 
2672 			/*
2673 			 * Zero out invalid groups so we can ignore
2674 			 * them later.
2675 			 */
2676 			if (!srcgroup ||
2677 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2678 				*i_qgroups = 0ULL;
2679 
2680 			++i_qgroups;
2681 		}
2682 	}
2683 
2684 	/*
2685 	 * create a tracking group for the subvol itself
2686 	 */
2687 	ret = add_qgroup_item(trans, quota_root, objectid);
2688 	if (ret)
2689 		goto out;
2690 
2691 	/*
2692 	 * add qgroup to all inherited groups
2693 	 */
2694 	if (inherit) {
2695 		i_qgroups = (u64 *)(inherit + 1);
2696 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2697 			if (*i_qgroups == 0)
2698 				continue;
2699 			ret = add_qgroup_relation_item(trans, objectid,
2700 						       *i_qgroups);
2701 			if (ret && ret != -EEXIST)
2702 				goto out;
2703 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2704 						       objectid);
2705 			if (ret && ret != -EEXIST)
2706 				goto out;
2707 		}
2708 		ret = 0;
2709 	}
2710 
2711 
2712 	spin_lock(&fs_info->qgroup_lock);
2713 
2714 	dstgroup = add_qgroup_rb(fs_info, objectid);
2715 	if (IS_ERR(dstgroup)) {
2716 		ret = PTR_ERR(dstgroup);
2717 		goto unlock;
2718 	}
2719 
2720 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2721 		dstgroup->lim_flags = inherit->lim.flags;
2722 		dstgroup->max_rfer = inherit->lim.max_rfer;
2723 		dstgroup->max_excl = inherit->lim.max_excl;
2724 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2725 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2726 
2727 		ret = update_qgroup_limit_item(trans, dstgroup);
2728 		if (ret) {
2729 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2730 			btrfs_info(fs_info,
2731 				   "unable to update quota limit for %llu",
2732 				   dstgroup->qgroupid);
2733 			goto unlock;
2734 		}
2735 	}
2736 
2737 	if (srcid) {
2738 		srcgroup = find_qgroup_rb(fs_info, srcid);
2739 		if (!srcgroup)
2740 			goto unlock;
2741 
2742 		/*
2743 		 * We call inherit after we clone the root in order to make sure
2744 		 * our counts don't go crazy, so at this point the only
2745 		 * difference between the two roots should be the root node.
2746 		 */
2747 		level_size = fs_info->nodesize;
2748 		dstgroup->rfer = srcgroup->rfer;
2749 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2750 		dstgroup->excl = level_size;
2751 		dstgroup->excl_cmpr = level_size;
2752 		srcgroup->excl = level_size;
2753 		srcgroup->excl_cmpr = level_size;
2754 
2755 		/* inherit the limit info */
2756 		dstgroup->lim_flags = srcgroup->lim_flags;
2757 		dstgroup->max_rfer = srcgroup->max_rfer;
2758 		dstgroup->max_excl = srcgroup->max_excl;
2759 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2760 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2761 
2762 		qgroup_dirty(fs_info, dstgroup);
2763 		qgroup_dirty(fs_info, srcgroup);
2764 	}
2765 
2766 	if (!inherit)
2767 		goto unlock;
2768 
2769 	i_qgroups = (u64 *)(inherit + 1);
2770 	for (i = 0; i < inherit->num_qgroups; ++i) {
2771 		if (*i_qgroups) {
2772 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2773 			if (ret)
2774 				goto unlock;
2775 		}
2776 		++i_qgroups;
2777 	}
2778 
2779 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2780 		struct btrfs_qgroup *src;
2781 		struct btrfs_qgroup *dst;
2782 
2783 		if (!i_qgroups[0] || !i_qgroups[1])
2784 			continue;
2785 
2786 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2787 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2788 
2789 		if (!src || !dst) {
2790 			ret = -EINVAL;
2791 			goto unlock;
2792 		}
2793 
2794 		dst->rfer = src->rfer - level_size;
2795 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2796 	}
2797 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2798 		struct btrfs_qgroup *src;
2799 		struct btrfs_qgroup *dst;
2800 
2801 		if (!i_qgroups[0] || !i_qgroups[1])
2802 			continue;
2803 
2804 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2805 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2806 
2807 		if (!src || !dst) {
2808 			ret = -EINVAL;
2809 			goto unlock;
2810 		}
2811 
2812 		dst->excl = src->excl + level_size;
2813 		dst->excl_cmpr = src->excl_cmpr + level_size;
2814 	}
2815 
2816 unlock:
2817 	spin_unlock(&fs_info->qgroup_lock);
2818 out:
2819 	if (!committing)
2820 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
2821 	return ret;
2822 }
2823 
2824 /*
2825  * Two limits to commit transaction in advance.
2826  *
2827  * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2828  * For SIZE, it will be in byte unit as threshold.
2829  */
2830 #define QGROUP_FREE_RATIO		32
2831 #define QGROUP_FREE_SIZE		SZ_32M
2832 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2833 				const struct btrfs_qgroup *qg, u64 num_bytes)
2834 {
2835 	u64 free;
2836 	u64 threshold;
2837 
2838 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2839 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2840 		return false;
2841 
2842 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2843 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2844 		return false;
2845 
2846 	/*
2847 	 * Even if we passed the check, it's better to check if reservation
2848 	 * for meta_pertrans is pushing us near limit.
2849 	 * If there is too much pertrans reservation or it's near the limit,
2850 	 * let's try commit transaction to free some, using transaction_kthread
2851 	 */
2852 	if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2853 			      BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2854 		if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2855 			free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2856 			threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2857 					  QGROUP_FREE_SIZE);
2858 		} else {
2859 			free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2860 			threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2861 					  QGROUP_FREE_SIZE);
2862 		}
2863 
2864 		/*
2865 		 * Use transaction_kthread to commit transaction, so we no
2866 		 * longer need to bother nested transaction nor lock context.
2867 		 */
2868 		if (free < threshold)
2869 			btrfs_commit_transaction_locksafe(fs_info);
2870 	}
2871 
2872 	return true;
2873 }
2874 
2875 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2876 			  enum btrfs_qgroup_rsv_type type)
2877 {
2878 	struct btrfs_root *quota_root;
2879 	struct btrfs_qgroup *qgroup;
2880 	struct btrfs_fs_info *fs_info = root->fs_info;
2881 	u64 ref_root = root->root_key.objectid;
2882 	int ret = 0;
2883 	struct ulist_node *unode;
2884 	struct ulist_iterator uiter;
2885 
2886 	if (!is_fstree(ref_root))
2887 		return 0;
2888 
2889 	if (num_bytes == 0)
2890 		return 0;
2891 
2892 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2893 	    capable(CAP_SYS_RESOURCE))
2894 		enforce = false;
2895 
2896 	spin_lock(&fs_info->qgroup_lock);
2897 	quota_root = fs_info->quota_root;
2898 	if (!quota_root)
2899 		goto out;
2900 
2901 	qgroup = find_qgroup_rb(fs_info, ref_root);
2902 	if (!qgroup)
2903 		goto out;
2904 
2905 	/*
2906 	 * in a first step, we check all affected qgroups if any limits would
2907 	 * be exceeded
2908 	 */
2909 	ulist_reinit(fs_info->qgroup_ulist);
2910 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2911 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2912 	if (ret < 0)
2913 		goto out;
2914 	ULIST_ITER_INIT(&uiter);
2915 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2916 		struct btrfs_qgroup *qg;
2917 		struct btrfs_qgroup_list *glist;
2918 
2919 		qg = unode_aux_to_qgroup(unode);
2920 
2921 		if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2922 			ret = -EDQUOT;
2923 			goto out;
2924 		}
2925 
2926 		list_for_each_entry(glist, &qg->groups, next_group) {
2927 			ret = ulist_add(fs_info->qgroup_ulist,
2928 					glist->group->qgroupid,
2929 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2930 			if (ret < 0)
2931 				goto out;
2932 		}
2933 	}
2934 	ret = 0;
2935 	/*
2936 	 * no limits exceeded, now record the reservation into all qgroups
2937 	 */
2938 	ULIST_ITER_INIT(&uiter);
2939 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2940 		struct btrfs_qgroup *qg;
2941 
2942 		qg = unode_aux_to_qgroup(unode);
2943 
2944 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
2945 	}
2946 
2947 out:
2948 	spin_unlock(&fs_info->qgroup_lock);
2949 	return ret;
2950 }
2951 
2952 /*
2953  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2954  * qgroup).
2955  *
2956  * Will handle all higher level qgroup too.
2957  *
2958  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2959  * This special case is only used for META_PERTRANS type.
2960  */
2961 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2962 			       u64 ref_root, u64 num_bytes,
2963 			       enum btrfs_qgroup_rsv_type type)
2964 {
2965 	struct btrfs_root *quota_root;
2966 	struct btrfs_qgroup *qgroup;
2967 	struct ulist_node *unode;
2968 	struct ulist_iterator uiter;
2969 	int ret = 0;
2970 
2971 	if (!is_fstree(ref_root))
2972 		return;
2973 
2974 	if (num_bytes == 0)
2975 		return;
2976 
2977 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2978 		WARN(1, "%s: Invalid type to free", __func__);
2979 		return;
2980 	}
2981 	spin_lock(&fs_info->qgroup_lock);
2982 
2983 	quota_root = fs_info->quota_root;
2984 	if (!quota_root)
2985 		goto out;
2986 
2987 	qgroup = find_qgroup_rb(fs_info, ref_root);
2988 	if (!qgroup)
2989 		goto out;
2990 
2991 	if (num_bytes == (u64)-1)
2992 		/*
2993 		 * We're freeing all pertrans rsv, get reserved value from
2994 		 * level 0 qgroup as real num_bytes to free.
2995 		 */
2996 		num_bytes = qgroup->rsv.values[type];
2997 
2998 	ulist_reinit(fs_info->qgroup_ulist);
2999 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3000 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3001 	if (ret < 0)
3002 		goto out;
3003 	ULIST_ITER_INIT(&uiter);
3004 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3005 		struct btrfs_qgroup *qg;
3006 		struct btrfs_qgroup_list *glist;
3007 
3008 		qg = unode_aux_to_qgroup(unode);
3009 
3010 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3011 
3012 		list_for_each_entry(glist, &qg->groups, next_group) {
3013 			ret = ulist_add(fs_info->qgroup_ulist,
3014 					glist->group->qgroupid,
3015 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3016 			if (ret < 0)
3017 				goto out;
3018 		}
3019 	}
3020 
3021 out:
3022 	spin_unlock(&fs_info->qgroup_lock);
3023 }
3024 
3025 /*
3026  * Check if the leaf is the last leaf. Which means all node pointers
3027  * are at their last position.
3028  */
3029 static bool is_last_leaf(struct btrfs_path *path)
3030 {
3031 	int i;
3032 
3033 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3034 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3035 			return false;
3036 	}
3037 	return true;
3038 }
3039 
3040 /*
3041  * returns < 0 on error, 0 when more leafs are to be scanned.
3042  * returns 1 when done.
3043  */
3044 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3045 			      struct btrfs_path *path)
3046 {
3047 	struct btrfs_fs_info *fs_info = trans->fs_info;
3048 	struct btrfs_key found;
3049 	struct extent_buffer *scratch_leaf = NULL;
3050 	struct ulist *roots = NULL;
3051 	u64 num_bytes;
3052 	bool done;
3053 	int slot;
3054 	int ret;
3055 
3056 	mutex_lock(&fs_info->qgroup_rescan_lock);
3057 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3058 					 &fs_info->qgroup_rescan_progress,
3059 					 path, 1, 0);
3060 
3061 	btrfs_debug(fs_info,
3062 		"current progress key (%llu %u %llu), search_slot ret %d",
3063 		fs_info->qgroup_rescan_progress.objectid,
3064 		fs_info->qgroup_rescan_progress.type,
3065 		fs_info->qgroup_rescan_progress.offset, ret);
3066 
3067 	if (ret) {
3068 		/*
3069 		 * The rescan is about to end, we will not be scanning any
3070 		 * further blocks. We cannot unset the RESCAN flag here, because
3071 		 * we want to commit the transaction if everything went well.
3072 		 * To make the live accounting work in this phase, we set our
3073 		 * scan progress pointer such that every real extent objectid
3074 		 * will be smaller.
3075 		 */
3076 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3077 		btrfs_release_path(path);
3078 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3079 		return ret;
3080 	}
3081 	done = is_last_leaf(path);
3082 
3083 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3084 			      btrfs_header_nritems(path->nodes[0]) - 1);
3085 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3086 
3087 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3088 	if (!scratch_leaf) {
3089 		ret = -ENOMEM;
3090 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3091 		goto out;
3092 	}
3093 	slot = path->slots[0];
3094 	btrfs_release_path(path);
3095 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3096 
3097 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3098 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3099 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3100 		    found.type != BTRFS_METADATA_ITEM_KEY)
3101 			continue;
3102 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3103 			num_bytes = fs_info->nodesize;
3104 		else
3105 			num_bytes = found.offset;
3106 
3107 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3108 					   &roots, false);
3109 		if (ret < 0)
3110 			goto out;
3111 		/* For rescan, just pass old_roots as NULL */
3112 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3113 						  num_bytes, NULL, roots);
3114 		if (ret < 0)
3115 			goto out;
3116 	}
3117 out:
3118 	if (scratch_leaf)
3119 		free_extent_buffer(scratch_leaf);
3120 
3121 	if (done && !ret) {
3122 		ret = 1;
3123 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3124 	}
3125 	return ret;
3126 }
3127 
3128 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3129 {
3130 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3131 						     qgroup_rescan_work);
3132 	struct btrfs_path *path;
3133 	struct btrfs_trans_handle *trans = NULL;
3134 	int err = -ENOMEM;
3135 	int ret = 0;
3136 
3137 	path = btrfs_alloc_path();
3138 	if (!path)
3139 		goto out;
3140 	/*
3141 	 * Rescan should only search for commit root, and any later difference
3142 	 * should be recorded by qgroup
3143 	 */
3144 	path->search_commit_root = 1;
3145 	path->skip_locking = 1;
3146 
3147 	err = 0;
3148 	while (!err && !btrfs_fs_closing(fs_info)) {
3149 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3150 		if (IS_ERR(trans)) {
3151 			err = PTR_ERR(trans);
3152 			break;
3153 		}
3154 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3155 			err = -EINTR;
3156 		} else {
3157 			err = qgroup_rescan_leaf(trans, path);
3158 		}
3159 		if (err > 0)
3160 			btrfs_commit_transaction(trans);
3161 		else
3162 			btrfs_end_transaction(trans);
3163 	}
3164 
3165 out:
3166 	btrfs_free_path(path);
3167 
3168 	mutex_lock(&fs_info->qgroup_rescan_lock);
3169 	if (err > 0 &&
3170 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3171 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3172 	} else if (err < 0) {
3173 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3174 	}
3175 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3176 
3177 	/*
3178 	 * only update status, since the previous part has already updated the
3179 	 * qgroup info.
3180 	 */
3181 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3182 	if (IS_ERR(trans)) {
3183 		err = PTR_ERR(trans);
3184 		trans = NULL;
3185 		btrfs_err(fs_info,
3186 			  "fail to start transaction for status update: %d",
3187 			  err);
3188 	}
3189 
3190 	mutex_lock(&fs_info->qgroup_rescan_lock);
3191 	if (!btrfs_fs_closing(fs_info))
3192 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3193 	if (trans) {
3194 		ret = update_qgroup_status_item(trans);
3195 		if (ret < 0) {
3196 			err = ret;
3197 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3198 				  err);
3199 		}
3200 	}
3201 	fs_info->qgroup_rescan_running = false;
3202 	complete_all(&fs_info->qgroup_rescan_completion);
3203 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3204 
3205 	if (!trans)
3206 		return;
3207 
3208 	btrfs_end_transaction(trans);
3209 
3210 	if (btrfs_fs_closing(fs_info)) {
3211 		btrfs_info(fs_info, "qgroup scan paused");
3212 	} else if (err >= 0) {
3213 		btrfs_info(fs_info, "qgroup scan completed%s",
3214 			err > 0 ? " (inconsistency flag cleared)" : "");
3215 	} else {
3216 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3217 	}
3218 }
3219 
3220 /*
3221  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3222  * memory required for the rescan context.
3223  */
3224 static int
3225 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3226 		   int init_flags)
3227 {
3228 	int ret = 0;
3229 
3230 	if (!init_flags) {
3231 		/* we're resuming qgroup rescan at mount time */
3232 		if (!(fs_info->qgroup_flags &
3233 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3234 			btrfs_warn(fs_info,
3235 			"qgroup rescan init failed, qgroup rescan is not queued");
3236 			ret = -EINVAL;
3237 		} else if (!(fs_info->qgroup_flags &
3238 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3239 			btrfs_warn(fs_info,
3240 			"qgroup rescan init failed, qgroup is not enabled");
3241 			ret = -EINVAL;
3242 		}
3243 
3244 		if (ret)
3245 			return ret;
3246 	}
3247 
3248 	mutex_lock(&fs_info->qgroup_rescan_lock);
3249 	spin_lock(&fs_info->qgroup_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 			spin_unlock(&fs_info->qgroup_lock);
3265 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3266 			return ret;
3267 		}
3268 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3269 	}
3270 
3271 	memset(&fs_info->qgroup_rescan_progress, 0,
3272 		sizeof(fs_info->qgroup_rescan_progress));
3273 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3274 	init_completion(&fs_info->qgroup_rescan_completion);
3275 	fs_info->qgroup_rescan_running = true;
3276 
3277 	spin_unlock(&fs_info->qgroup_lock);
3278 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3279 
3280 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3281 			btrfs_qgroup_rescan_worker, NULL, NULL);
3282 	return 0;
3283 }
3284 
3285 static void
3286 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3287 {
3288 	struct rb_node *n;
3289 	struct btrfs_qgroup *qgroup;
3290 
3291 	spin_lock(&fs_info->qgroup_lock);
3292 	/* clear all current qgroup tracking information */
3293 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3294 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3295 		qgroup->rfer = 0;
3296 		qgroup->rfer_cmpr = 0;
3297 		qgroup->excl = 0;
3298 		qgroup->excl_cmpr = 0;
3299 		qgroup_dirty(fs_info, qgroup);
3300 	}
3301 	spin_unlock(&fs_info->qgroup_lock);
3302 }
3303 
3304 int
3305 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3306 {
3307 	int ret = 0;
3308 	struct btrfs_trans_handle *trans;
3309 
3310 	ret = qgroup_rescan_init(fs_info, 0, 1);
3311 	if (ret)
3312 		return ret;
3313 
3314 	/*
3315 	 * We have set the rescan_progress to 0, which means no more
3316 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3317 	 * However, btrfs_qgroup_account_ref may be right after its call
3318 	 * to btrfs_find_all_roots, in which case it would still do the
3319 	 * accounting.
3320 	 * To solve this, we're committing the transaction, which will
3321 	 * ensure we run all delayed refs and only after that, we are
3322 	 * going to clear all tracking information for a clean start.
3323 	 */
3324 
3325 	trans = btrfs_join_transaction(fs_info->fs_root);
3326 	if (IS_ERR(trans)) {
3327 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3328 		return PTR_ERR(trans);
3329 	}
3330 	ret = btrfs_commit_transaction(trans);
3331 	if (ret) {
3332 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3333 		return ret;
3334 	}
3335 
3336 	qgroup_rescan_zero_tracking(fs_info);
3337 
3338 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3339 			 &fs_info->qgroup_rescan_work);
3340 
3341 	return 0;
3342 }
3343 
3344 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3345 				     bool interruptible)
3346 {
3347 	int running;
3348 	int ret = 0;
3349 
3350 	mutex_lock(&fs_info->qgroup_rescan_lock);
3351 	spin_lock(&fs_info->qgroup_lock);
3352 	running = fs_info->qgroup_rescan_running;
3353 	spin_unlock(&fs_info->qgroup_lock);
3354 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3355 
3356 	if (!running)
3357 		return 0;
3358 
3359 	if (interruptible)
3360 		ret = wait_for_completion_interruptible(
3361 					&fs_info->qgroup_rescan_completion);
3362 	else
3363 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3364 
3365 	return ret;
3366 }
3367 
3368 /*
3369  * this is only called from open_ctree where we're still single threaded, thus
3370  * locking is omitted here.
3371  */
3372 void
3373 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3374 {
3375 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3376 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3377 				 &fs_info->qgroup_rescan_work);
3378 }
3379 
3380 /*
3381  * Reserve qgroup space for range [start, start + len).
3382  *
3383  * This function will either reserve space from related qgroups or doing
3384  * nothing if the range is already reserved.
3385  *
3386  * Return 0 for successful reserve
3387  * Return <0 for error (including -EQUOT)
3388  *
3389  * NOTE: this function may sleep for memory allocation.
3390  *       if btrfs_qgroup_reserve_data() is called multiple times with
3391  *       same @reserved, caller must ensure when error happens it's OK
3392  *       to free *ALL* reserved space.
3393  */
3394 int btrfs_qgroup_reserve_data(struct inode *inode,
3395 			struct extent_changeset **reserved_ret, u64 start,
3396 			u64 len)
3397 {
3398 	struct btrfs_root *root = BTRFS_I(inode)->root;
3399 	struct ulist_node *unode;
3400 	struct ulist_iterator uiter;
3401 	struct extent_changeset *reserved;
3402 	u64 orig_reserved;
3403 	u64 to_reserve;
3404 	int ret;
3405 
3406 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3407 	    !is_fstree(root->root_key.objectid) || len == 0)
3408 		return 0;
3409 
3410 	/* @reserved parameter is mandatory for qgroup */
3411 	if (WARN_ON(!reserved_ret))
3412 		return -EINVAL;
3413 	if (!*reserved_ret) {
3414 		*reserved_ret = extent_changeset_alloc();
3415 		if (!*reserved_ret)
3416 			return -ENOMEM;
3417 	}
3418 	reserved = *reserved_ret;
3419 	/* Record already reserved space */
3420 	orig_reserved = reserved->bytes_changed;
3421 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3422 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3423 
3424 	/* Newly reserved space */
3425 	to_reserve = reserved->bytes_changed - orig_reserved;
3426 	trace_btrfs_qgroup_reserve_data(inode, start, len,
3427 					to_reserve, QGROUP_RESERVE);
3428 	if (ret < 0)
3429 		goto cleanup;
3430 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3431 	if (ret < 0)
3432 		goto cleanup;
3433 
3434 	return ret;
3435 
3436 cleanup:
3437 	/* cleanup *ALL* already reserved ranges */
3438 	ULIST_ITER_INIT(&uiter);
3439 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3440 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3441 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3442 	/* Also free data bytes of already reserved one */
3443 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3444 				  orig_reserved, BTRFS_QGROUP_RSV_DATA);
3445 	extent_changeset_release(reserved);
3446 	return ret;
3447 }
3448 
3449 /* Free ranges specified by @reserved, normally in error path */
3450 static int qgroup_free_reserved_data(struct inode *inode,
3451 			struct extent_changeset *reserved, u64 start, u64 len)
3452 {
3453 	struct btrfs_root *root = BTRFS_I(inode)->root;
3454 	struct ulist_node *unode;
3455 	struct ulist_iterator uiter;
3456 	struct extent_changeset changeset;
3457 	int freed = 0;
3458 	int ret;
3459 
3460 	extent_changeset_init(&changeset);
3461 	len = round_up(start + len, root->fs_info->sectorsize);
3462 	start = round_down(start, root->fs_info->sectorsize);
3463 
3464 	ULIST_ITER_INIT(&uiter);
3465 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3466 		u64 range_start = unode->val;
3467 		/* unode->aux is the inclusive end */
3468 		u64 range_len = unode->aux - range_start + 1;
3469 		u64 free_start;
3470 		u64 free_len;
3471 
3472 		extent_changeset_release(&changeset);
3473 
3474 		/* Only free range in range [start, start + len) */
3475 		if (range_start >= start + len ||
3476 		    range_start + range_len <= start)
3477 			continue;
3478 		free_start = max(range_start, start);
3479 		free_len = min(start + len, range_start + range_len) -
3480 			   free_start;
3481 		/*
3482 		 * TODO: To also modify reserved->ranges_reserved to reflect
3483 		 * the modification.
3484 		 *
3485 		 * However as long as we free qgroup reserved according to
3486 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3487 		 * So not need to rush.
3488 		 */
3489 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree,
3490 				free_start, free_start + free_len - 1,
3491 				EXTENT_QGROUP_RESERVED, &changeset);
3492 		if (ret < 0)
3493 			goto out;
3494 		freed += changeset.bytes_changed;
3495 	}
3496 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3497 				  BTRFS_QGROUP_RSV_DATA);
3498 	ret = freed;
3499 out:
3500 	extent_changeset_release(&changeset);
3501 	return ret;
3502 }
3503 
3504 static int __btrfs_qgroup_release_data(struct inode *inode,
3505 			struct extent_changeset *reserved, u64 start, u64 len,
3506 			int free)
3507 {
3508 	struct extent_changeset changeset;
3509 	int trace_op = QGROUP_RELEASE;
3510 	int ret;
3511 
3512 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3513 		      &BTRFS_I(inode)->root->fs_info->flags))
3514 		return 0;
3515 
3516 	/* In release case, we shouldn't have @reserved */
3517 	WARN_ON(!free && reserved);
3518 	if (free && reserved)
3519 		return qgroup_free_reserved_data(inode, reserved, start, len);
3520 	extent_changeset_init(&changeset);
3521 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3522 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3523 	if (ret < 0)
3524 		goto out;
3525 
3526 	if (free)
3527 		trace_op = QGROUP_FREE;
3528 	trace_btrfs_qgroup_release_data(inode, start, len,
3529 					changeset.bytes_changed, trace_op);
3530 	if (free)
3531 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3532 				BTRFS_I(inode)->root->root_key.objectid,
3533 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3534 	ret = changeset.bytes_changed;
3535 out:
3536 	extent_changeset_release(&changeset);
3537 	return ret;
3538 }
3539 
3540 /*
3541  * Free a reserved space range from io_tree and related qgroups
3542  *
3543  * Should be called when a range of pages get invalidated before reaching disk.
3544  * Or for error cleanup case.
3545  * if @reserved is given, only reserved range in [@start, @start + @len) will
3546  * be freed.
3547  *
3548  * For data written to disk, use btrfs_qgroup_release_data().
3549  *
3550  * NOTE: This function may sleep for memory allocation.
3551  */
3552 int btrfs_qgroup_free_data(struct inode *inode,
3553 			struct extent_changeset *reserved, u64 start, u64 len)
3554 {
3555 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3556 }
3557 
3558 /*
3559  * Release a reserved space range from io_tree only.
3560  *
3561  * Should be called when a range of pages get written to disk and corresponding
3562  * FILE_EXTENT is inserted into corresponding root.
3563  *
3564  * Since new qgroup accounting framework will only update qgroup numbers at
3565  * commit_transaction() time, its reserved space shouldn't be freed from
3566  * related qgroups.
3567  *
3568  * But we should release the range from io_tree, to allow further write to be
3569  * COWed.
3570  *
3571  * NOTE: This function may sleep for memory allocation.
3572  */
3573 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3574 {
3575 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3576 }
3577 
3578 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3579 			      enum btrfs_qgroup_rsv_type type)
3580 {
3581 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3582 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3583 		return;
3584 	if (num_bytes == 0)
3585 		return;
3586 
3587 	spin_lock(&root->qgroup_meta_rsv_lock);
3588 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3589 		root->qgroup_meta_rsv_prealloc += num_bytes;
3590 	else
3591 		root->qgroup_meta_rsv_pertrans += num_bytes;
3592 	spin_unlock(&root->qgroup_meta_rsv_lock);
3593 }
3594 
3595 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3596 			     enum btrfs_qgroup_rsv_type type)
3597 {
3598 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3599 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3600 		return 0;
3601 	if (num_bytes == 0)
3602 		return 0;
3603 
3604 	spin_lock(&root->qgroup_meta_rsv_lock);
3605 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3606 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3607 				  num_bytes);
3608 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3609 	} else {
3610 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3611 				  num_bytes);
3612 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3613 	}
3614 	spin_unlock(&root->qgroup_meta_rsv_lock);
3615 	return num_bytes;
3616 }
3617 
3618 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3619 				enum btrfs_qgroup_rsv_type type, bool enforce)
3620 {
3621 	struct btrfs_fs_info *fs_info = root->fs_info;
3622 	int ret;
3623 
3624 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3625 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3626 		return 0;
3627 
3628 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3629 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3630 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3631 	if (ret < 0)
3632 		return ret;
3633 	/*
3634 	 * Record what we have reserved into root.
3635 	 *
3636 	 * To avoid quota disabled->enabled underflow.
3637 	 * In that case, we may try to free space we haven't reserved
3638 	 * (since quota was disabled), so record what we reserved into root.
3639 	 * And ensure later release won't underflow this number.
3640 	 */
3641 	add_root_meta_rsv(root, num_bytes, type);
3642 	return ret;
3643 }
3644 
3645 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3646 {
3647 	struct btrfs_fs_info *fs_info = root->fs_info;
3648 
3649 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3650 	    !is_fstree(root->root_key.objectid))
3651 		return;
3652 
3653 	/* TODO: Update trace point to handle such free */
3654 	trace_qgroup_meta_free_all_pertrans(root);
3655 	/* Special value -1 means to free all reserved space */
3656 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3657 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3658 }
3659 
3660 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3661 			      enum btrfs_qgroup_rsv_type type)
3662 {
3663 	struct btrfs_fs_info *fs_info = root->fs_info;
3664 
3665 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3666 	    !is_fstree(root->root_key.objectid))
3667 		return;
3668 
3669 	/*
3670 	 * reservation for META_PREALLOC can happen before quota is enabled,
3671 	 * which can lead to underflow.
3672 	 * Here ensure we will only free what we really have reserved.
3673 	 */
3674 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3675 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3676 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3677 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3678 				  num_bytes, type);
3679 }
3680 
3681 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3682 				int num_bytes)
3683 {
3684 	struct btrfs_root *quota_root = fs_info->quota_root;
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 (!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