xref: /openbmc/linux/fs/btrfs/qgroup.c (revision 4cff79e9)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sizes.h>
15 
16 #include "ctree.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "locking.h"
20 #include "ulist.h"
21 #include "backref.h"
22 #include "extent_io.h"
23 #include "qgroup.h"
24 
25 
26 /* TODO XXX FIXME
27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
28  *  - reorganize keys
29  *  - compressed
30  *  - sync
31  *  - copy also limits on subvol creation
32  *  - limit
33  *  - caches fuer 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 umounting
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,
534 				    struct btrfs_root *quota_root,
535 				    u64 src, u64 dst)
536 {
537 	int ret;
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,
558 				    struct btrfs_root *quota_root,
559 				    u64 src, u64 dst)
560 {
561 	int ret;
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,
657 			   struct btrfs_root *quota_root, u64 qgroupid)
658 {
659 	int ret;
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_root *root,
704 				    struct btrfs_qgroup *qgroup)
705 {
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, 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_root *root,
746 				   struct btrfs_qgroup *qgroup)
747 {
748 	struct btrfs_path *path;
749 	struct btrfs_key key;
750 	struct extent_buffer *l;
751 	struct btrfs_qgroup_info_item *qgroup_info;
752 	int ret;
753 	int slot;
754 
755 	if (btrfs_is_testing(root->fs_info))
756 		return 0;
757 
758 	key.objectid = 0;
759 	key.type = BTRFS_QGROUP_INFO_KEY;
760 	key.offset = qgroup->qgroupid;
761 
762 	path = btrfs_alloc_path();
763 	if (!path)
764 		return -ENOMEM;
765 
766 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
767 	if (ret > 0)
768 		ret = -ENOENT;
769 
770 	if (ret)
771 		goto out;
772 
773 	l = path->nodes[0];
774 	slot = path->slots[0];
775 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
776 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
777 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
778 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
779 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
780 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
781 
782 	btrfs_mark_buffer_dirty(l);
783 
784 out:
785 	btrfs_free_path(path);
786 	return ret;
787 }
788 
789 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
790 				     struct btrfs_fs_info *fs_info,
791 				    struct btrfs_root *root)
792 {
793 	struct btrfs_path *path;
794 	struct btrfs_key key;
795 	struct extent_buffer *l;
796 	struct btrfs_qgroup_status_item *ptr;
797 	int ret;
798 	int slot;
799 
800 	key.objectid = 0;
801 	key.type = BTRFS_QGROUP_STATUS_KEY;
802 	key.offset = 0;
803 
804 	path = btrfs_alloc_path();
805 	if (!path)
806 		return -ENOMEM;
807 
808 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
809 	if (ret > 0)
810 		ret = -ENOENT;
811 
812 	if (ret)
813 		goto out;
814 
815 	l = path->nodes[0];
816 	slot = path->slots[0];
817 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
818 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
819 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
820 	btrfs_set_qgroup_status_rescan(l, ptr,
821 				fs_info->qgroup_rescan_progress.objectid);
822 
823 	btrfs_mark_buffer_dirty(l);
824 
825 out:
826 	btrfs_free_path(path);
827 	return ret;
828 }
829 
830 /*
831  * called with qgroup_lock held
832  */
833 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
834 				  struct btrfs_root *root)
835 {
836 	struct btrfs_path *path;
837 	struct btrfs_key key;
838 	struct extent_buffer *leaf = NULL;
839 	int ret;
840 	int nr = 0;
841 
842 	path = btrfs_alloc_path();
843 	if (!path)
844 		return -ENOMEM;
845 
846 	path->leave_spinning = 1;
847 
848 	key.objectid = 0;
849 	key.offset = 0;
850 	key.type = 0;
851 
852 	while (1) {
853 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
854 		if (ret < 0)
855 			goto out;
856 		leaf = path->nodes[0];
857 		nr = btrfs_header_nritems(leaf);
858 		if (!nr)
859 			break;
860 		/*
861 		 * delete the leaf one by one
862 		 * since the whole tree is going
863 		 * to be deleted.
864 		 */
865 		path->slots[0] = 0;
866 		ret = btrfs_del_items(trans, root, path, 0, nr);
867 		if (ret)
868 			goto out;
869 
870 		btrfs_release_path(path);
871 	}
872 	ret = 0;
873 out:
874 	btrfs_free_path(path);
875 	return ret;
876 }
877 
878 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
879 		       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 	int ret = 0;
890 	int slot;
891 
892 	mutex_lock(&fs_info->qgroup_ioctl_lock);
893 	if (fs_info->quota_root)
894 		goto out;
895 
896 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
897 	if (!fs_info->qgroup_ulist) {
898 		ret = -ENOMEM;
899 		goto out;
900 	}
901 
902 	/*
903 	 * initially create the quota tree
904 	 */
905 	quota_root = btrfs_create_tree(trans, fs_info,
906 				       BTRFS_QUOTA_TREE_OBJECTID);
907 	if (IS_ERR(quota_root)) {
908 		ret =  PTR_ERR(quota_root);
909 		goto out;
910 	}
911 
912 	path = btrfs_alloc_path();
913 	if (!path) {
914 		ret = -ENOMEM;
915 		goto out_free_root;
916 	}
917 
918 	key.objectid = 0;
919 	key.type = BTRFS_QGROUP_STATUS_KEY;
920 	key.offset = 0;
921 
922 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
923 				      sizeof(*ptr));
924 	if (ret)
925 		goto out_free_path;
926 
927 	leaf = path->nodes[0];
928 	ptr = btrfs_item_ptr(leaf, path->slots[0],
929 				 struct btrfs_qgroup_status_item);
930 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
931 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
932 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
933 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
934 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
935 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
936 
937 	btrfs_mark_buffer_dirty(leaf);
938 
939 	key.objectid = 0;
940 	key.type = BTRFS_ROOT_REF_KEY;
941 	key.offset = 0;
942 
943 	btrfs_release_path(path);
944 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
945 	if (ret > 0)
946 		goto out_add_root;
947 	if (ret < 0)
948 		goto out_free_path;
949 
950 
951 	while (1) {
952 		slot = path->slots[0];
953 		leaf = path->nodes[0];
954 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
955 
956 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
957 			ret = add_qgroup_item(trans, quota_root,
958 					      found_key.offset);
959 			if (ret)
960 				goto out_free_path;
961 
962 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
963 			if (IS_ERR(qgroup)) {
964 				ret = PTR_ERR(qgroup);
965 				goto out_free_path;
966 			}
967 		}
968 		ret = btrfs_next_item(tree_root, path);
969 		if (ret < 0)
970 			goto out_free_path;
971 		if (ret)
972 			break;
973 	}
974 
975 out_add_root:
976 	btrfs_release_path(path);
977 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
978 	if (ret)
979 		goto out_free_path;
980 
981 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
982 	if (IS_ERR(qgroup)) {
983 		ret = PTR_ERR(qgroup);
984 		goto out_free_path;
985 	}
986 	spin_lock(&fs_info->qgroup_lock);
987 	fs_info->quota_root = quota_root;
988 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
989 	spin_unlock(&fs_info->qgroup_lock);
990 	ret = qgroup_rescan_init(fs_info, 0, 1);
991 	if (!ret) {
992 	        qgroup_rescan_zero_tracking(fs_info);
993 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
994 	                         &fs_info->qgroup_rescan_work);
995 	}
996 
997 out_free_path:
998 	btrfs_free_path(path);
999 out_free_root:
1000 	if (ret) {
1001 		free_extent_buffer(quota_root->node);
1002 		free_extent_buffer(quota_root->commit_root);
1003 		kfree(quota_root);
1004 	}
1005 out:
1006 	if (ret) {
1007 		ulist_free(fs_info->qgroup_ulist);
1008 		fs_info->qgroup_ulist = NULL;
1009 	}
1010 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1011 	return ret;
1012 }
1013 
1014 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
1015 			struct btrfs_fs_info *fs_info)
1016 {
1017 	struct btrfs_root *quota_root;
1018 	int ret = 0;
1019 
1020 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1021 	if (!fs_info->quota_root)
1022 		goto out;
1023 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1024 	btrfs_qgroup_wait_for_completion(fs_info, false);
1025 	spin_lock(&fs_info->qgroup_lock);
1026 	quota_root = fs_info->quota_root;
1027 	fs_info->quota_root = NULL;
1028 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1029 	spin_unlock(&fs_info->qgroup_lock);
1030 
1031 	btrfs_free_qgroup_config(fs_info);
1032 
1033 	ret = btrfs_clean_quota_tree(trans, quota_root);
1034 	if (ret)
1035 		goto out;
1036 
1037 	ret = btrfs_del_root(trans, fs_info, &quota_root->root_key);
1038 	if (ret)
1039 		goto out;
1040 
1041 	list_del(&quota_root->dirty_list);
1042 
1043 	btrfs_tree_lock(quota_root->node);
1044 	clean_tree_block(fs_info, quota_root->node);
1045 	btrfs_tree_unlock(quota_root->node);
1046 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1047 
1048 	free_extent_buffer(quota_root->node);
1049 	free_extent_buffer(quota_root->commit_root);
1050 	kfree(quota_root);
1051 out:
1052 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1053 	return ret;
1054 }
1055 
1056 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1057 			 struct btrfs_qgroup *qgroup)
1058 {
1059 	if (list_empty(&qgroup->dirty))
1060 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1061 }
1062 
1063 /*
1064  * The easy accounting, we're updating qgroup relationship whose child qgroup
1065  * only has exclusive extents.
1066  *
1067  * In this case, all exclsuive extents will also be exlusive for parent, so
1068  * excl/rfer just get added/removed.
1069  *
1070  * So is qgroup reservation space, which should also be added/removed to
1071  * parent.
1072  * Or when child tries to release reservation space, parent will underflow its
1073  * reservation (for relationship adding case).
1074  *
1075  * Caller should hold fs_info->qgroup_lock.
1076  */
1077 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1078 				    struct ulist *tmp, u64 ref_root,
1079 				    struct btrfs_qgroup *src, int sign)
1080 {
1081 	struct btrfs_qgroup *qgroup;
1082 	struct btrfs_qgroup_list *glist;
1083 	struct ulist_node *unode;
1084 	struct ulist_iterator uiter;
1085 	u64 num_bytes = src->excl;
1086 	int ret = 0;
1087 
1088 	qgroup = find_qgroup_rb(fs_info, ref_root);
1089 	if (!qgroup)
1090 		goto out;
1091 
1092 	qgroup->rfer += sign * num_bytes;
1093 	qgroup->rfer_cmpr += sign * num_bytes;
1094 
1095 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1096 	qgroup->excl += sign * num_bytes;
1097 	qgroup->excl_cmpr += sign * num_bytes;
1098 
1099 	if (sign > 0)
1100 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1101 	else
1102 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1103 
1104 	qgroup_dirty(fs_info, qgroup);
1105 
1106 	/* Get all of the parent groups that contain this qgroup */
1107 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1108 		ret = ulist_add(tmp, glist->group->qgroupid,
1109 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1110 		if (ret < 0)
1111 			goto out;
1112 	}
1113 
1114 	/* Iterate all of the parents and adjust their reference counts */
1115 	ULIST_ITER_INIT(&uiter);
1116 	while ((unode = ulist_next(tmp, &uiter))) {
1117 		qgroup = unode_aux_to_qgroup(unode);
1118 		qgroup->rfer += sign * num_bytes;
1119 		qgroup->rfer_cmpr += sign * num_bytes;
1120 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1121 		qgroup->excl += sign * num_bytes;
1122 		if (sign > 0)
1123 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1124 		else
1125 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1126 		qgroup->excl_cmpr += sign * num_bytes;
1127 		qgroup_dirty(fs_info, qgroup);
1128 
1129 		/* Add any parents of the parents */
1130 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1131 			ret = ulist_add(tmp, glist->group->qgroupid,
1132 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1133 			if (ret < 0)
1134 				goto out;
1135 		}
1136 	}
1137 	ret = 0;
1138 out:
1139 	return ret;
1140 }
1141 
1142 
1143 /*
1144  * Quick path for updating qgroup with only excl refs.
1145  *
1146  * In that case, just update all parent will be enough.
1147  * Or we needs to do a full rescan.
1148  * Caller should also hold fs_info->qgroup_lock.
1149  *
1150  * Return 0 for quick update, return >0 for need to full rescan
1151  * and mark INCONSISTENT flag.
1152  * Return < 0 for other error.
1153  */
1154 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1155 				   struct ulist *tmp, u64 src, u64 dst,
1156 				   int sign)
1157 {
1158 	struct btrfs_qgroup *qgroup;
1159 	int ret = 1;
1160 	int err = 0;
1161 
1162 	qgroup = find_qgroup_rb(fs_info, src);
1163 	if (!qgroup)
1164 		goto out;
1165 	if (qgroup->excl == qgroup->rfer) {
1166 		ret = 0;
1167 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1168 					       qgroup, sign);
1169 		if (err < 0) {
1170 			ret = err;
1171 			goto out;
1172 		}
1173 	}
1174 out:
1175 	if (ret)
1176 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1177 	return ret;
1178 }
1179 
1180 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1181 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1182 {
1183 	struct btrfs_root *quota_root;
1184 	struct btrfs_qgroup *parent;
1185 	struct btrfs_qgroup *member;
1186 	struct btrfs_qgroup_list *list;
1187 	struct ulist *tmp;
1188 	int ret = 0;
1189 
1190 	/* Check the level of src and dst first */
1191 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1192 		return -EINVAL;
1193 
1194 	tmp = ulist_alloc(GFP_KERNEL);
1195 	if (!tmp)
1196 		return -ENOMEM;
1197 
1198 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1199 	quota_root = fs_info->quota_root;
1200 	if (!quota_root) {
1201 		ret = -EINVAL;
1202 		goto out;
1203 	}
1204 	member = find_qgroup_rb(fs_info, src);
1205 	parent = find_qgroup_rb(fs_info, dst);
1206 	if (!member || !parent) {
1207 		ret = -EINVAL;
1208 		goto out;
1209 	}
1210 
1211 	/* check if such qgroup relation exist firstly */
1212 	list_for_each_entry(list, &member->groups, next_group) {
1213 		if (list->group == parent) {
1214 			ret = -EEXIST;
1215 			goto out;
1216 		}
1217 	}
1218 
1219 	ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1220 	if (ret)
1221 		goto out;
1222 
1223 	ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1224 	if (ret) {
1225 		del_qgroup_relation_item(trans, quota_root, src, dst);
1226 		goto out;
1227 	}
1228 
1229 	spin_lock(&fs_info->qgroup_lock);
1230 	ret = add_relation_rb(fs_info, src, dst);
1231 	if (ret < 0) {
1232 		spin_unlock(&fs_info->qgroup_lock);
1233 		goto out;
1234 	}
1235 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1236 	spin_unlock(&fs_info->qgroup_lock);
1237 out:
1238 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1239 	ulist_free(tmp);
1240 	return ret;
1241 }
1242 
1243 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1244 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1245 {
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 	int err;
1253 
1254 	tmp = ulist_alloc(GFP_KERNEL);
1255 	if (!tmp)
1256 		return -ENOMEM;
1257 
1258 	quota_root = fs_info->quota_root;
1259 	if (!quota_root) {
1260 		ret = -EINVAL;
1261 		goto out;
1262 	}
1263 
1264 	member = find_qgroup_rb(fs_info, src);
1265 	parent = find_qgroup_rb(fs_info, dst);
1266 	if (!member || !parent) {
1267 		ret = -EINVAL;
1268 		goto out;
1269 	}
1270 
1271 	/* check if such qgroup relation exist firstly */
1272 	list_for_each_entry(list, &member->groups, next_group) {
1273 		if (list->group == parent)
1274 			goto exist;
1275 	}
1276 	ret = -ENOENT;
1277 	goto out;
1278 exist:
1279 	ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1280 	err = del_qgroup_relation_item(trans, quota_root, dst, src);
1281 	if (err && !ret)
1282 		ret = err;
1283 
1284 	spin_lock(&fs_info->qgroup_lock);
1285 	del_relation_rb(fs_info, src, dst);
1286 	ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1287 	spin_unlock(&fs_info->qgroup_lock);
1288 out:
1289 	ulist_free(tmp);
1290 	return ret;
1291 }
1292 
1293 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1294 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1295 {
1296 	int ret = 0;
1297 
1298 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1299 	ret = __del_qgroup_relation(trans, fs_info, src, dst);
1300 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1301 
1302 	return ret;
1303 }
1304 
1305 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1306 			struct btrfs_fs_info *fs_info, u64 qgroupid)
1307 {
1308 	struct btrfs_root *quota_root;
1309 	struct btrfs_qgroup *qgroup;
1310 	int ret = 0;
1311 
1312 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1313 	quota_root = fs_info->quota_root;
1314 	if (!quota_root) {
1315 		ret = -EINVAL;
1316 		goto out;
1317 	}
1318 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1319 	if (qgroup) {
1320 		ret = -EEXIST;
1321 		goto out;
1322 	}
1323 
1324 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1325 	if (ret)
1326 		goto out;
1327 
1328 	spin_lock(&fs_info->qgroup_lock);
1329 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1330 	spin_unlock(&fs_info->qgroup_lock);
1331 
1332 	if (IS_ERR(qgroup))
1333 		ret = PTR_ERR(qgroup);
1334 out:
1335 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1336 	return ret;
1337 }
1338 
1339 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1340 			struct btrfs_fs_info *fs_info, u64 qgroupid)
1341 {
1342 	struct btrfs_root *quota_root;
1343 	struct btrfs_qgroup *qgroup;
1344 	struct btrfs_qgroup_list *list;
1345 	int ret = 0;
1346 
1347 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1348 	quota_root = fs_info->quota_root;
1349 	if (!quota_root) {
1350 		ret = -EINVAL;
1351 		goto out;
1352 	}
1353 
1354 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1355 	if (!qgroup) {
1356 		ret = -ENOENT;
1357 		goto out;
1358 	} else {
1359 		/* check if there are no children of this qgroup */
1360 		if (!list_empty(&qgroup->members)) {
1361 			ret = -EBUSY;
1362 			goto out;
1363 		}
1364 	}
1365 	ret = del_qgroup_item(trans, quota_root, qgroupid);
1366 	if (ret && ret != -ENOENT)
1367 		goto out;
1368 
1369 	while (!list_empty(&qgroup->groups)) {
1370 		list = list_first_entry(&qgroup->groups,
1371 					struct btrfs_qgroup_list, next_group);
1372 		ret = __del_qgroup_relation(trans, fs_info,
1373 					   qgroupid,
1374 					   list->group->qgroupid);
1375 		if (ret)
1376 			goto out;
1377 	}
1378 
1379 	spin_lock(&fs_info->qgroup_lock);
1380 	del_qgroup_rb(fs_info, qgroupid);
1381 	spin_unlock(&fs_info->qgroup_lock);
1382 out:
1383 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1384 	return ret;
1385 }
1386 
1387 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1388 		       struct btrfs_fs_info *fs_info, u64 qgroupid,
1389 		       struct btrfs_qgroup_limit *limit)
1390 {
1391 	struct btrfs_root *quota_root;
1392 	struct btrfs_qgroup *qgroup;
1393 	int ret = 0;
1394 	/* Sometimes we would want to clear the limit on this qgroup.
1395 	 * To meet this requirement, we treat the -1 as a special value
1396 	 * which tell kernel to clear the limit on this qgroup.
1397 	 */
1398 	const u64 CLEAR_VALUE = -1;
1399 
1400 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1401 	quota_root = fs_info->quota_root;
1402 	if (!quota_root) {
1403 		ret = -EINVAL;
1404 		goto out;
1405 	}
1406 
1407 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1408 	if (!qgroup) {
1409 		ret = -ENOENT;
1410 		goto out;
1411 	}
1412 
1413 	spin_lock(&fs_info->qgroup_lock);
1414 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1415 		if (limit->max_rfer == CLEAR_VALUE) {
1416 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1417 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1418 			qgroup->max_rfer = 0;
1419 		} else {
1420 			qgroup->max_rfer = limit->max_rfer;
1421 		}
1422 	}
1423 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1424 		if (limit->max_excl == CLEAR_VALUE) {
1425 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1426 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1427 			qgroup->max_excl = 0;
1428 		} else {
1429 			qgroup->max_excl = limit->max_excl;
1430 		}
1431 	}
1432 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1433 		if (limit->rsv_rfer == CLEAR_VALUE) {
1434 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1435 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1436 			qgroup->rsv_rfer = 0;
1437 		} else {
1438 			qgroup->rsv_rfer = limit->rsv_rfer;
1439 		}
1440 	}
1441 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1442 		if (limit->rsv_excl == CLEAR_VALUE) {
1443 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1444 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1445 			qgroup->rsv_excl = 0;
1446 		} else {
1447 			qgroup->rsv_excl = limit->rsv_excl;
1448 		}
1449 	}
1450 	qgroup->lim_flags |= limit->flags;
1451 
1452 	spin_unlock(&fs_info->qgroup_lock);
1453 
1454 	ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1455 	if (ret) {
1456 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1457 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1458 		       qgroupid);
1459 	}
1460 
1461 out:
1462 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1463 	return ret;
1464 }
1465 
1466 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1467 				struct btrfs_delayed_ref_root *delayed_refs,
1468 				struct btrfs_qgroup_extent_record *record)
1469 {
1470 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1471 	struct rb_node *parent_node = NULL;
1472 	struct btrfs_qgroup_extent_record *entry;
1473 	u64 bytenr = record->bytenr;
1474 
1475 	lockdep_assert_held(&delayed_refs->lock);
1476 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1477 
1478 	while (*p) {
1479 		parent_node = *p;
1480 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1481 				 node);
1482 		if (bytenr < entry->bytenr)
1483 			p = &(*p)->rb_left;
1484 		else if (bytenr > entry->bytenr)
1485 			p = &(*p)->rb_right;
1486 		else
1487 			return 1;
1488 	}
1489 
1490 	rb_link_node(&record->node, parent_node, p);
1491 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1492 	return 0;
1493 }
1494 
1495 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1496 				   struct btrfs_qgroup_extent_record *qrecord)
1497 {
1498 	struct ulist *old_root;
1499 	u64 bytenr = qrecord->bytenr;
1500 	int ret;
1501 
1502 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1503 	if (ret < 0) {
1504 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1505 		btrfs_warn(fs_info,
1506 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1507 			ret);
1508 		return 0;
1509 	}
1510 
1511 	/*
1512 	 * Here we don't need to get the lock of
1513 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1514 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1515 	 *
1516 	 * So modifying qrecord->old_roots is safe here
1517 	 */
1518 	qrecord->old_roots = old_root;
1519 	return 0;
1520 }
1521 
1522 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1523 		struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1524 		gfp_t gfp_flag)
1525 {
1526 	struct btrfs_qgroup_extent_record *record;
1527 	struct btrfs_delayed_ref_root *delayed_refs;
1528 	int ret;
1529 
1530 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1531 	    || bytenr == 0 || num_bytes == 0)
1532 		return 0;
1533 	if (WARN_ON(trans == NULL))
1534 		return -EINVAL;
1535 	record = kmalloc(sizeof(*record), gfp_flag);
1536 	if (!record)
1537 		return -ENOMEM;
1538 
1539 	delayed_refs = &trans->transaction->delayed_refs;
1540 	record->bytenr = bytenr;
1541 	record->num_bytes = num_bytes;
1542 	record->old_roots = NULL;
1543 
1544 	spin_lock(&delayed_refs->lock);
1545 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1546 	spin_unlock(&delayed_refs->lock);
1547 	if (ret > 0) {
1548 		kfree(record);
1549 		return 0;
1550 	}
1551 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1552 }
1553 
1554 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1555 				  struct btrfs_fs_info *fs_info,
1556 				  struct extent_buffer *eb)
1557 {
1558 	int nr = btrfs_header_nritems(eb);
1559 	int i, extent_type, ret;
1560 	struct btrfs_key key;
1561 	struct btrfs_file_extent_item *fi;
1562 	u64 bytenr, num_bytes;
1563 
1564 	/* We can be called directly from walk_up_proc() */
1565 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1566 		return 0;
1567 
1568 	for (i = 0; i < nr; i++) {
1569 		btrfs_item_key_to_cpu(eb, &key, i);
1570 
1571 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1572 			continue;
1573 
1574 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1575 		/* filter out non qgroup-accountable extents  */
1576 		extent_type = btrfs_file_extent_type(eb, fi);
1577 
1578 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1579 			continue;
1580 
1581 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1582 		if (!bytenr)
1583 			continue;
1584 
1585 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1586 
1587 		ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1588 						num_bytes, GFP_NOFS);
1589 		if (ret)
1590 			return ret;
1591 	}
1592 	cond_resched();
1593 	return 0;
1594 }
1595 
1596 /*
1597  * Walk up the tree from the bottom, freeing leaves and any interior
1598  * nodes which have had all slots visited. If a node (leaf or
1599  * interior) is freed, the node above it will have it's slot
1600  * incremented. The root node will never be freed.
1601  *
1602  * At the end of this function, we should have a path which has all
1603  * slots incremented to the next position for a search. If we need to
1604  * read a new node it will be NULL and the node above it will have the
1605  * correct slot selected for a later read.
1606  *
1607  * If we increment the root nodes slot counter past the number of
1608  * elements, 1 is returned to signal completion of the search.
1609  */
1610 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1611 {
1612 	int level = 0;
1613 	int nr, slot;
1614 	struct extent_buffer *eb;
1615 
1616 	if (root_level == 0)
1617 		return 1;
1618 
1619 	while (level <= root_level) {
1620 		eb = path->nodes[level];
1621 		nr = btrfs_header_nritems(eb);
1622 		path->slots[level]++;
1623 		slot = path->slots[level];
1624 		if (slot >= nr || level == 0) {
1625 			/*
1626 			 * Don't free the root -  we will detect this
1627 			 * condition after our loop and return a
1628 			 * positive value for caller to stop walking the tree.
1629 			 */
1630 			if (level != root_level) {
1631 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1632 				path->locks[level] = 0;
1633 
1634 				free_extent_buffer(eb);
1635 				path->nodes[level] = NULL;
1636 				path->slots[level] = 0;
1637 			}
1638 		} else {
1639 			/*
1640 			 * We have a valid slot to walk back down
1641 			 * from. Stop here so caller can process these
1642 			 * new nodes.
1643 			 */
1644 			break;
1645 		}
1646 
1647 		level++;
1648 	}
1649 
1650 	eb = path->nodes[root_level];
1651 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1652 		return 1;
1653 
1654 	return 0;
1655 }
1656 
1657 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1658 			       struct btrfs_root *root,
1659 			       struct extent_buffer *root_eb,
1660 			       u64 root_gen, int root_level)
1661 {
1662 	struct btrfs_fs_info *fs_info = root->fs_info;
1663 	int ret = 0;
1664 	int level;
1665 	struct extent_buffer *eb = root_eb;
1666 	struct btrfs_path *path = NULL;
1667 
1668 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1669 	BUG_ON(root_eb == NULL);
1670 
1671 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1672 		return 0;
1673 
1674 	if (!extent_buffer_uptodate(root_eb)) {
1675 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1676 		if (ret)
1677 			goto out;
1678 	}
1679 
1680 	if (root_level == 0) {
1681 		ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1682 		goto out;
1683 	}
1684 
1685 	path = btrfs_alloc_path();
1686 	if (!path)
1687 		return -ENOMEM;
1688 
1689 	/*
1690 	 * Walk down the tree.  Missing extent blocks are filled in as
1691 	 * we go. Metadata is accounted every time we read a new
1692 	 * extent block.
1693 	 *
1694 	 * When we reach a leaf, we account for file extent items in it,
1695 	 * walk back up the tree (adjusting slot pointers as we go)
1696 	 * and restart the search process.
1697 	 */
1698 	extent_buffer_get(root_eb); /* For path */
1699 	path->nodes[root_level] = root_eb;
1700 	path->slots[root_level] = 0;
1701 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1702 walk_down:
1703 	level = root_level;
1704 	while (level >= 0) {
1705 		if (path->nodes[level] == NULL) {
1706 			struct btrfs_key first_key;
1707 			int parent_slot;
1708 			u64 child_gen;
1709 			u64 child_bytenr;
1710 
1711 			/*
1712 			 * We need to get child blockptr/gen from parent before
1713 			 * we can read it.
1714 			  */
1715 			eb = path->nodes[level + 1];
1716 			parent_slot = path->slots[level + 1];
1717 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1718 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1719 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1720 
1721 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1722 					     level, &first_key);
1723 			if (IS_ERR(eb)) {
1724 				ret = PTR_ERR(eb);
1725 				goto out;
1726 			} else if (!extent_buffer_uptodate(eb)) {
1727 				free_extent_buffer(eb);
1728 				ret = -EIO;
1729 				goto out;
1730 			}
1731 
1732 			path->nodes[level] = eb;
1733 			path->slots[level] = 0;
1734 
1735 			btrfs_tree_read_lock(eb);
1736 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1737 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1738 
1739 			ret = btrfs_qgroup_trace_extent(trans, fs_info,
1740 							child_bytenr,
1741 							fs_info->nodesize,
1742 							GFP_NOFS);
1743 			if (ret)
1744 				goto out;
1745 		}
1746 
1747 		if (level == 0) {
1748 			ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1749 							   path->nodes[level]);
1750 			if (ret)
1751 				goto out;
1752 
1753 			/* Nonzero return here means we completed our search */
1754 			ret = adjust_slots_upwards(path, root_level);
1755 			if (ret)
1756 				break;
1757 
1758 			/* Restart search with new slots */
1759 			goto walk_down;
1760 		}
1761 
1762 		level--;
1763 	}
1764 
1765 	ret = 0;
1766 out:
1767 	btrfs_free_path(path);
1768 
1769 	return ret;
1770 }
1771 
1772 #define UPDATE_NEW	0
1773 #define UPDATE_OLD	1
1774 /*
1775  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1776  */
1777 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1778 				struct ulist *roots, struct ulist *tmp,
1779 				struct ulist *qgroups, u64 seq, int update_old)
1780 {
1781 	struct ulist_node *unode;
1782 	struct ulist_iterator uiter;
1783 	struct ulist_node *tmp_unode;
1784 	struct ulist_iterator tmp_uiter;
1785 	struct btrfs_qgroup *qg;
1786 	int ret = 0;
1787 
1788 	if (!roots)
1789 		return 0;
1790 	ULIST_ITER_INIT(&uiter);
1791 	while ((unode = ulist_next(roots, &uiter))) {
1792 		qg = find_qgroup_rb(fs_info, unode->val);
1793 		if (!qg)
1794 			continue;
1795 
1796 		ulist_reinit(tmp);
1797 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1798 				GFP_ATOMIC);
1799 		if (ret < 0)
1800 			return ret;
1801 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1802 		if (ret < 0)
1803 			return ret;
1804 		ULIST_ITER_INIT(&tmp_uiter);
1805 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1806 			struct btrfs_qgroup_list *glist;
1807 
1808 			qg = unode_aux_to_qgroup(tmp_unode);
1809 			if (update_old)
1810 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1811 			else
1812 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1813 			list_for_each_entry(glist, &qg->groups, next_group) {
1814 				ret = ulist_add(qgroups, glist->group->qgroupid,
1815 						qgroup_to_aux(glist->group),
1816 						GFP_ATOMIC);
1817 				if (ret < 0)
1818 					return ret;
1819 				ret = ulist_add(tmp, glist->group->qgroupid,
1820 						qgroup_to_aux(glist->group),
1821 						GFP_ATOMIC);
1822 				if (ret < 0)
1823 					return ret;
1824 			}
1825 		}
1826 	}
1827 	return 0;
1828 }
1829 
1830 /*
1831  * Update qgroup rfer/excl counters.
1832  * Rfer update is easy, codes can explain themselves.
1833  *
1834  * Excl update is tricky, the update is split into 2 part.
1835  * Part 1: Possible exclusive <-> sharing detect:
1836  *	|	A	|	!A	|
1837  *  -------------------------------------
1838  *  B	|	*	|	-	|
1839  *  -------------------------------------
1840  *  !B	|	+	|	**	|
1841  *  -------------------------------------
1842  *
1843  * Conditions:
1844  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
1845  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
1846  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
1847  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
1848  *
1849  * Results:
1850  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
1851  * *: Definitely not changed.		**: Possible unchanged.
1852  *
1853  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1854  *
1855  * To make the logic clear, we first use condition A and B to split
1856  * combination into 4 results.
1857  *
1858  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1859  * only on variant maybe 0.
1860  *
1861  * Lastly, check result **, since there are 2 variants maybe 0, split them
1862  * again(2x2).
1863  * But this time we don't need to consider other things, the codes and logic
1864  * is easy to understand now.
1865  */
1866 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1867 				  struct ulist *qgroups,
1868 				  u64 nr_old_roots,
1869 				  u64 nr_new_roots,
1870 				  u64 num_bytes, u64 seq)
1871 {
1872 	struct ulist_node *unode;
1873 	struct ulist_iterator uiter;
1874 	struct btrfs_qgroup *qg;
1875 	u64 cur_new_count, cur_old_count;
1876 
1877 	ULIST_ITER_INIT(&uiter);
1878 	while ((unode = ulist_next(qgroups, &uiter))) {
1879 		bool dirty = false;
1880 
1881 		qg = unode_aux_to_qgroup(unode);
1882 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1883 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1884 
1885 		trace_qgroup_update_counters(fs_info, qg->qgroupid,
1886 					     cur_old_count, cur_new_count);
1887 
1888 		/* Rfer update part */
1889 		if (cur_old_count == 0 && cur_new_count > 0) {
1890 			qg->rfer += num_bytes;
1891 			qg->rfer_cmpr += num_bytes;
1892 			dirty = true;
1893 		}
1894 		if (cur_old_count > 0 && cur_new_count == 0) {
1895 			qg->rfer -= num_bytes;
1896 			qg->rfer_cmpr -= num_bytes;
1897 			dirty = true;
1898 		}
1899 
1900 		/* Excl update part */
1901 		/* Exclusive/none -> shared case */
1902 		if (cur_old_count == nr_old_roots &&
1903 		    cur_new_count < nr_new_roots) {
1904 			/* Exclusive -> shared */
1905 			if (cur_old_count != 0) {
1906 				qg->excl -= num_bytes;
1907 				qg->excl_cmpr -= num_bytes;
1908 				dirty = true;
1909 			}
1910 		}
1911 
1912 		/* Shared -> exclusive/none case */
1913 		if (cur_old_count < nr_old_roots &&
1914 		    cur_new_count == nr_new_roots) {
1915 			/* Shared->exclusive */
1916 			if (cur_new_count != 0) {
1917 				qg->excl += num_bytes;
1918 				qg->excl_cmpr += num_bytes;
1919 				dirty = true;
1920 			}
1921 		}
1922 
1923 		/* Exclusive/none -> exclusive/none case */
1924 		if (cur_old_count == nr_old_roots &&
1925 		    cur_new_count == nr_new_roots) {
1926 			if (cur_old_count == 0) {
1927 				/* None -> exclusive/none */
1928 
1929 				if (cur_new_count != 0) {
1930 					/* None -> exclusive */
1931 					qg->excl += num_bytes;
1932 					qg->excl_cmpr += num_bytes;
1933 					dirty = true;
1934 				}
1935 				/* None -> none, nothing changed */
1936 			} else {
1937 				/* Exclusive -> exclusive/none */
1938 
1939 				if (cur_new_count == 0) {
1940 					/* Exclusive -> none */
1941 					qg->excl -= num_bytes;
1942 					qg->excl_cmpr -= num_bytes;
1943 					dirty = true;
1944 				}
1945 				/* Exclusive -> exclusive, nothing changed */
1946 			}
1947 		}
1948 
1949 		if (dirty)
1950 			qgroup_dirty(fs_info, qg);
1951 	}
1952 	return 0;
1953 }
1954 
1955 /*
1956  * Check if the @roots potentially is a list of fs tree roots
1957  *
1958  * Return 0 for definitely not a fs/subvol tree roots ulist
1959  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
1960  *          one as well)
1961  */
1962 static int maybe_fs_roots(struct ulist *roots)
1963 {
1964 	struct ulist_node *unode;
1965 	struct ulist_iterator uiter;
1966 
1967 	/* Empty one, still possible for fs roots */
1968 	if (!roots || roots->nnodes == 0)
1969 		return 1;
1970 
1971 	ULIST_ITER_INIT(&uiter);
1972 	unode = ulist_next(roots, &uiter);
1973 	if (!unode)
1974 		return 1;
1975 
1976 	/*
1977 	 * If it contains fs tree roots, then it must belong to fs/subvol
1978 	 * trees.
1979 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
1980 	 */
1981 	return is_fstree(unode->val);
1982 }
1983 
1984 int
1985 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1986 			    struct btrfs_fs_info *fs_info,
1987 			    u64 bytenr, u64 num_bytes,
1988 			    struct ulist *old_roots, struct ulist *new_roots)
1989 {
1990 	struct ulist *qgroups = NULL;
1991 	struct ulist *tmp = NULL;
1992 	u64 seq;
1993 	u64 nr_new_roots = 0;
1994 	u64 nr_old_roots = 0;
1995 	int ret = 0;
1996 
1997 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1998 		return 0;
1999 
2000 	if (new_roots) {
2001 		if (!maybe_fs_roots(new_roots))
2002 			goto out_free;
2003 		nr_new_roots = new_roots->nnodes;
2004 	}
2005 	if (old_roots) {
2006 		if (!maybe_fs_roots(old_roots))
2007 			goto out_free;
2008 		nr_old_roots = old_roots->nnodes;
2009 	}
2010 
2011 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2012 	if (nr_old_roots == 0 && nr_new_roots == 0)
2013 		goto out_free;
2014 
2015 	BUG_ON(!fs_info->quota_root);
2016 
2017 	trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
2018 					  nr_old_roots, nr_new_roots);
2019 
2020 	qgroups = ulist_alloc(GFP_NOFS);
2021 	if (!qgroups) {
2022 		ret = -ENOMEM;
2023 		goto out_free;
2024 	}
2025 	tmp = ulist_alloc(GFP_NOFS);
2026 	if (!tmp) {
2027 		ret = -ENOMEM;
2028 		goto out_free;
2029 	}
2030 
2031 	mutex_lock(&fs_info->qgroup_rescan_lock);
2032 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2033 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2034 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2035 			ret = 0;
2036 			goto out_free;
2037 		}
2038 	}
2039 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2040 
2041 	spin_lock(&fs_info->qgroup_lock);
2042 	seq = fs_info->qgroup_seq;
2043 
2044 	/* Update old refcnts using old_roots */
2045 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2046 				   UPDATE_OLD);
2047 	if (ret < 0)
2048 		goto out;
2049 
2050 	/* Update new refcnts using new_roots */
2051 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2052 				   UPDATE_NEW);
2053 	if (ret < 0)
2054 		goto out;
2055 
2056 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2057 			       num_bytes, seq);
2058 
2059 	/*
2060 	 * Bump qgroup_seq to avoid seq overlap
2061 	 */
2062 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2063 out:
2064 	spin_unlock(&fs_info->qgroup_lock);
2065 out_free:
2066 	ulist_free(tmp);
2067 	ulist_free(qgroups);
2068 	ulist_free(old_roots);
2069 	ulist_free(new_roots);
2070 	return ret;
2071 }
2072 
2073 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2074 {
2075 	struct btrfs_fs_info *fs_info = trans->fs_info;
2076 	struct btrfs_qgroup_extent_record *record;
2077 	struct btrfs_delayed_ref_root *delayed_refs;
2078 	struct ulist *new_roots = NULL;
2079 	struct rb_node *node;
2080 	u64 qgroup_to_skip;
2081 	int ret = 0;
2082 
2083 	delayed_refs = &trans->transaction->delayed_refs;
2084 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2085 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2086 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2087 				  node);
2088 
2089 		trace_btrfs_qgroup_account_extents(fs_info, record);
2090 
2091 		if (!ret) {
2092 			/*
2093 			 * Old roots should be searched when inserting qgroup
2094 			 * extent record
2095 			 */
2096 			if (WARN_ON(!record->old_roots)) {
2097 				/* Search commit root to find old_roots */
2098 				ret = btrfs_find_all_roots(NULL, fs_info,
2099 						record->bytenr, 0,
2100 						&record->old_roots, false);
2101 				if (ret < 0)
2102 					goto cleanup;
2103 			}
2104 
2105 			/*
2106 			 * Use SEQ_LAST as time_seq to do special search, which
2107 			 * doesn't lock tree or delayed_refs and search current
2108 			 * root. It's safe inside commit_transaction().
2109 			 */
2110 			ret = btrfs_find_all_roots(trans, fs_info,
2111 				record->bytenr, SEQ_LAST, &new_roots, false);
2112 			if (ret < 0)
2113 				goto cleanup;
2114 			if (qgroup_to_skip) {
2115 				ulist_del(new_roots, qgroup_to_skip, 0);
2116 				ulist_del(record->old_roots, qgroup_to_skip,
2117 					  0);
2118 			}
2119 			ret = btrfs_qgroup_account_extent(trans, fs_info,
2120 					record->bytenr, record->num_bytes,
2121 					record->old_roots, new_roots);
2122 			record->old_roots = NULL;
2123 			new_roots = NULL;
2124 		}
2125 cleanup:
2126 		ulist_free(record->old_roots);
2127 		ulist_free(new_roots);
2128 		new_roots = NULL;
2129 		rb_erase(node, &delayed_refs->dirty_extent_root);
2130 		kfree(record);
2131 
2132 	}
2133 	return ret;
2134 }
2135 
2136 /*
2137  * called from commit_transaction. Writes all changed qgroups to disk.
2138  */
2139 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2140 		      struct btrfs_fs_info *fs_info)
2141 {
2142 	struct btrfs_root *quota_root = fs_info->quota_root;
2143 	int ret = 0;
2144 
2145 	if (!quota_root)
2146 		return ret;
2147 
2148 	spin_lock(&fs_info->qgroup_lock);
2149 	while (!list_empty(&fs_info->dirty_qgroups)) {
2150 		struct btrfs_qgroup *qgroup;
2151 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2152 					  struct btrfs_qgroup, dirty);
2153 		list_del_init(&qgroup->dirty);
2154 		spin_unlock(&fs_info->qgroup_lock);
2155 		ret = update_qgroup_info_item(trans, quota_root, qgroup);
2156 		if (ret)
2157 			fs_info->qgroup_flags |=
2158 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2159 		ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2160 		if (ret)
2161 			fs_info->qgroup_flags |=
2162 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2163 		spin_lock(&fs_info->qgroup_lock);
2164 	}
2165 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2166 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2167 	else
2168 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2169 	spin_unlock(&fs_info->qgroup_lock);
2170 
2171 	ret = update_qgroup_status_item(trans, fs_info, quota_root);
2172 	if (ret)
2173 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2174 
2175 	return ret;
2176 }
2177 
2178 /*
2179  * Copy the accounting information between qgroups. This is necessary
2180  * when a snapshot or a subvolume is created. Throwing an error will
2181  * cause a transaction abort so we take extra care here to only error
2182  * when a readonly fs is a reasonable outcome.
2183  */
2184 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2185 			 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2186 			 struct btrfs_qgroup_inherit *inherit)
2187 {
2188 	int ret = 0;
2189 	int i;
2190 	u64 *i_qgroups;
2191 	struct btrfs_root *quota_root = fs_info->quota_root;
2192 	struct btrfs_qgroup *srcgroup;
2193 	struct btrfs_qgroup *dstgroup;
2194 	u32 level_size = 0;
2195 	u64 nums;
2196 
2197 	mutex_lock(&fs_info->qgroup_ioctl_lock);
2198 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2199 		goto out;
2200 
2201 	if (!quota_root) {
2202 		ret = -EINVAL;
2203 		goto out;
2204 	}
2205 
2206 	if (inherit) {
2207 		i_qgroups = (u64 *)(inherit + 1);
2208 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2209 		       2 * inherit->num_excl_copies;
2210 		for (i = 0; i < nums; ++i) {
2211 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2212 
2213 			/*
2214 			 * Zero out invalid groups so we can ignore
2215 			 * them later.
2216 			 */
2217 			if (!srcgroup ||
2218 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2219 				*i_qgroups = 0ULL;
2220 
2221 			++i_qgroups;
2222 		}
2223 	}
2224 
2225 	/*
2226 	 * create a tracking group for the subvol itself
2227 	 */
2228 	ret = add_qgroup_item(trans, quota_root, objectid);
2229 	if (ret)
2230 		goto out;
2231 
2232 	if (srcid) {
2233 		struct btrfs_root *srcroot;
2234 		struct btrfs_key srckey;
2235 
2236 		srckey.objectid = srcid;
2237 		srckey.type = BTRFS_ROOT_ITEM_KEY;
2238 		srckey.offset = (u64)-1;
2239 		srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2240 		if (IS_ERR(srcroot)) {
2241 			ret = PTR_ERR(srcroot);
2242 			goto out;
2243 		}
2244 
2245 		level_size = fs_info->nodesize;
2246 	}
2247 
2248 	/*
2249 	 * add qgroup to all inherited groups
2250 	 */
2251 	if (inherit) {
2252 		i_qgroups = (u64 *)(inherit + 1);
2253 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2254 			if (*i_qgroups == 0)
2255 				continue;
2256 			ret = add_qgroup_relation_item(trans, quota_root,
2257 						       objectid, *i_qgroups);
2258 			if (ret && ret != -EEXIST)
2259 				goto out;
2260 			ret = add_qgroup_relation_item(trans, quota_root,
2261 						       *i_qgroups, objectid);
2262 			if (ret && ret != -EEXIST)
2263 				goto out;
2264 		}
2265 		ret = 0;
2266 	}
2267 
2268 
2269 	spin_lock(&fs_info->qgroup_lock);
2270 
2271 	dstgroup = add_qgroup_rb(fs_info, objectid);
2272 	if (IS_ERR(dstgroup)) {
2273 		ret = PTR_ERR(dstgroup);
2274 		goto unlock;
2275 	}
2276 
2277 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2278 		dstgroup->lim_flags = inherit->lim.flags;
2279 		dstgroup->max_rfer = inherit->lim.max_rfer;
2280 		dstgroup->max_excl = inherit->lim.max_excl;
2281 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2282 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2283 
2284 		ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2285 		if (ret) {
2286 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2287 			btrfs_info(fs_info,
2288 				   "unable to update quota limit for %llu",
2289 				   dstgroup->qgroupid);
2290 			goto unlock;
2291 		}
2292 	}
2293 
2294 	if (srcid) {
2295 		srcgroup = find_qgroup_rb(fs_info, srcid);
2296 		if (!srcgroup)
2297 			goto unlock;
2298 
2299 		/*
2300 		 * We call inherit after we clone the root in order to make sure
2301 		 * our counts don't go crazy, so at this point the only
2302 		 * difference between the two roots should be the root node.
2303 		 */
2304 		dstgroup->rfer = srcgroup->rfer;
2305 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2306 		dstgroup->excl = level_size;
2307 		dstgroup->excl_cmpr = level_size;
2308 		srcgroup->excl = level_size;
2309 		srcgroup->excl_cmpr = level_size;
2310 
2311 		/* inherit the limit info */
2312 		dstgroup->lim_flags = srcgroup->lim_flags;
2313 		dstgroup->max_rfer = srcgroup->max_rfer;
2314 		dstgroup->max_excl = srcgroup->max_excl;
2315 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2316 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2317 
2318 		qgroup_dirty(fs_info, dstgroup);
2319 		qgroup_dirty(fs_info, srcgroup);
2320 	}
2321 
2322 	if (!inherit)
2323 		goto unlock;
2324 
2325 	i_qgroups = (u64 *)(inherit + 1);
2326 	for (i = 0; i < inherit->num_qgroups; ++i) {
2327 		if (*i_qgroups) {
2328 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2329 			if (ret)
2330 				goto unlock;
2331 		}
2332 		++i_qgroups;
2333 	}
2334 
2335 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2336 		struct btrfs_qgroup *src;
2337 		struct btrfs_qgroup *dst;
2338 
2339 		if (!i_qgroups[0] || !i_qgroups[1])
2340 			continue;
2341 
2342 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2343 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2344 
2345 		if (!src || !dst) {
2346 			ret = -EINVAL;
2347 			goto unlock;
2348 		}
2349 
2350 		dst->rfer = src->rfer - level_size;
2351 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2352 	}
2353 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2354 		struct btrfs_qgroup *src;
2355 		struct btrfs_qgroup *dst;
2356 
2357 		if (!i_qgroups[0] || !i_qgroups[1])
2358 			continue;
2359 
2360 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2361 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2362 
2363 		if (!src || !dst) {
2364 			ret = -EINVAL;
2365 			goto unlock;
2366 		}
2367 
2368 		dst->excl = src->excl + level_size;
2369 		dst->excl_cmpr = src->excl_cmpr + level_size;
2370 	}
2371 
2372 unlock:
2373 	spin_unlock(&fs_info->qgroup_lock);
2374 out:
2375 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
2376 	return ret;
2377 }
2378 
2379 /*
2380  * Two limits to commit transaction in advance.
2381  *
2382  * For RATIO, it will be 1/RATIO of the remaining limit
2383  * (excluding data and prealloc meta) as threshold.
2384  * For SIZE, it will be in byte unit as threshold.
2385  */
2386 #define QGROUP_PERTRANS_RATIO		32
2387 #define QGROUP_PERTRANS_SIZE		SZ_32M
2388 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2389 				const struct btrfs_qgroup *qg, u64 num_bytes)
2390 {
2391 	u64 limit;
2392 	u64 threshold;
2393 
2394 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2395 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2396 		return false;
2397 
2398 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2399 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2400 		return false;
2401 
2402 	/*
2403 	 * Even if we passed the check, it's better to check if reservation
2404 	 * for meta_pertrans is pushing us near limit.
2405 	 * If there is too much pertrans reservation or it's near the limit,
2406 	 * let's try commit transaction to free some, using transaction_kthread
2407 	 */
2408 	if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2409 			      BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2410 		if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2411 			limit = qg->max_excl;
2412 		else
2413 			limit = qg->max_rfer;
2414 		threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2415 			    qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2416 			    QGROUP_PERTRANS_RATIO;
2417 		threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2418 
2419 		/*
2420 		 * Use transaction_kthread to commit transaction, so we no
2421 		 * longer need to bother nested transaction nor lock context.
2422 		 */
2423 		if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2424 			btrfs_commit_transaction_locksafe(fs_info);
2425 	}
2426 
2427 	return true;
2428 }
2429 
2430 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2431 			  enum btrfs_qgroup_rsv_type type)
2432 {
2433 	struct btrfs_root *quota_root;
2434 	struct btrfs_qgroup *qgroup;
2435 	struct btrfs_fs_info *fs_info = root->fs_info;
2436 	u64 ref_root = root->root_key.objectid;
2437 	int ret = 0;
2438 	struct ulist_node *unode;
2439 	struct ulist_iterator uiter;
2440 
2441 	if (!is_fstree(ref_root))
2442 		return 0;
2443 
2444 	if (num_bytes == 0)
2445 		return 0;
2446 
2447 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2448 	    capable(CAP_SYS_RESOURCE))
2449 		enforce = false;
2450 
2451 	spin_lock(&fs_info->qgroup_lock);
2452 	quota_root = fs_info->quota_root;
2453 	if (!quota_root)
2454 		goto out;
2455 
2456 	qgroup = find_qgroup_rb(fs_info, ref_root);
2457 	if (!qgroup)
2458 		goto out;
2459 
2460 	/*
2461 	 * in a first step, we check all affected qgroups if any limits would
2462 	 * be exceeded
2463 	 */
2464 	ulist_reinit(fs_info->qgroup_ulist);
2465 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2466 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2467 	if (ret < 0)
2468 		goto out;
2469 	ULIST_ITER_INIT(&uiter);
2470 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2471 		struct btrfs_qgroup *qg;
2472 		struct btrfs_qgroup_list *glist;
2473 
2474 		qg = unode_aux_to_qgroup(unode);
2475 
2476 		if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2477 			ret = -EDQUOT;
2478 			goto out;
2479 		}
2480 
2481 		list_for_each_entry(glist, &qg->groups, next_group) {
2482 			ret = ulist_add(fs_info->qgroup_ulist,
2483 					glist->group->qgroupid,
2484 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2485 			if (ret < 0)
2486 				goto out;
2487 		}
2488 	}
2489 	ret = 0;
2490 	/*
2491 	 * no limits exceeded, now record the reservation into all qgroups
2492 	 */
2493 	ULIST_ITER_INIT(&uiter);
2494 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2495 		struct btrfs_qgroup *qg;
2496 
2497 		qg = unode_aux_to_qgroup(unode);
2498 
2499 		trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2500 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
2501 	}
2502 
2503 out:
2504 	spin_unlock(&fs_info->qgroup_lock);
2505 	return ret;
2506 }
2507 
2508 /*
2509  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2510  * qgroup).
2511  *
2512  * Will handle all higher level qgroup too.
2513  *
2514  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2515  * This special case is only used for META_PERTRANS type.
2516  */
2517 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2518 			       u64 ref_root, u64 num_bytes,
2519 			       enum btrfs_qgroup_rsv_type type)
2520 {
2521 	struct btrfs_root *quota_root;
2522 	struct btrfs_qgroup *qgroup;
2523 	struct ulist_node *unode;
2524 	struct ulist_iterator uiter;
2525 	int ret = 0;
2526 
2527 	if (!is_fstree(ref_root))
2528 		return;
2529 
2530 	if (num_bytes == 0)
2531 		return;
2532 
2533 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2534 		WARN(1, "%s: Invalid type to free", __func__);
2535 		return;
2536 	}
2537 	spin_lock(&fs_info->qgroup_lock);
2538 
2539 	quota_root = fs_info->quota_root;
2540 	if (!quota_root)
2541 		goto out;
2542 
2543 	qgroup = find_qgroup_rb(fs_info, ref_root);
2544 	if (!qgroup)
2545 		goto out;
2546 
2547 	if (num_bytes == (u64)-1)
2548 		/*
2549 		 * We're freeing all pertrans rsv, get reserved value from
2550 		 * level 0 qgroup as real num_bytes to free.
2551 		 */
2552 		num_bytes = qgroup->rsv.values[type];
2553 
2554 	ulist_reinit(fs_info->qgroup_ulist);
2555 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2556 			qgroup_to_aux(qgroup), GFP_ATOMIC);
2557 	if (ret < 0)
2558 		goto out;
2559 	ULIST_ITER_INIT(&uiter);
2560 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2561 		struct btrfs_qgroup *qg;
2562 		struct btrfs_qgroup_list *glist;
2563 
2564 		qg = unode_aux_to_qgroup(unode);
2565 
2566 		trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2567 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
2568 
2569 		list_for_each_entry(glist, &qg->groups, next_group) {
2570 			ret = ulist_add(fs_info->qgroup_ulist,
2571 					glist->group->qgroupid,
2572 					qgroup_to_aux(glist->group), GFP_ATOMIC);
2573 			if (ret < 0)
2574 				goto out;
2575 		}
2576 	}
2577 
2578 out:
2579 	spin_unlock(&fs_info->qgroup_lock);
2580 }
2581 
2582 /*
2583  * returns < 0 on error, 0 when more leafs are to be scanned.
2584  * returns 1 when done.
2585  */
2586 static int
2587 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2588 		   struct btrfs_trans_handle *trans)
2589 {
2590 	struct btrfs_key found;
2591 	struct extent_buffer *scratch_leaf = NULL;
2592 	struct ulist *roots = NULL;
2593 	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2594 	u64 num_bytes;
2595 	int slot;
2596 	int ret;
2597 
2598 	mutex_lock(&fs_info->qgroup_rescan_lock);
2599 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
2600 					 &fs_info->qgroup_rescan_progress,
2601 					 path, 1, 0);
2602 
2603 	btrfs_debug(fs_info,
2604 		"current progress key (%llu %u %llu), search_slot ret %d",
2605 		fs_info->qgroup_rescan_progress.objectid,
2606 		fs_info->qgroup_rescan_progress.type,
2607 		fs_info->qgroup_rescan_progress.offset, ret);
2608 
2609 	if (ret) {
2610 		/*
2611 		 * The rescan is about to end, we will not be scanning any
2612 		 * further blocks. We cannot unset the RESCAN flag here, because
2613 		 * we want to commit the transaction if everything went well.
2614 		 * To make the live accounting work in this phase, we set our
2615 		 * scan progress pointer such that every real extent objectid
2616 		 * will be smaller.
2617 		 */
2618 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2619 		btrfs_release_path(path);
2620 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2621 		return ret;
2622 	}
2623 
2624 	btrfs_item_key_to_cpu(path->nodes[0], &found,
2625 			      btrfs_header_nritems(path->nodes[0]) - 1);
2626 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2627 
2628 	btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2629 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2630 	if (!scratch_leaf) {
2631 		ret = -ENOMEM;
2632 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2633 		goto out;
2634 	}
2635 	extent_buffer_get(scratch_leaf);
2636 	btrfs_tree_read_lock(scratch_leaf);
2637 	btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2638 	slot = path->slots[0];
2639 	btrfs_release_path(path);
2640 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2641 
2642 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2643 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2644 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2645 		    found.type != BTRFS_METADATA_ITEM_KEY)
2646 			continue;
2647 		if (found.type == BTRFS_METADATA_ITEM_KEY)
2648 			num_bytes = fs_info->nodesize;
2649 		else
2650 			num_bytes = found.offset;
2651 
2652 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2653 					   &roots, false);
2654 		if (ret < 0)
2655 			goto out;
2656 		/* For rescan, just pass old_roots as NULL */
2657 		ret = btrfs_qgroup_account_extent(trans, fs_info,
2658 				found.objectid, num_bytes, NULL, roots);
2659 		if (ret < 0)
2660 			goto out;
2661 	}
2662 out:
2663 	if (scratch_leaf) {
2664 		btrfs_tree_read_unlock_blocking(scratch_leaf);
2665 		free_extent_buffer(scratch_leaf);
2666 	}
2667 	btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2668 
2669 	return ret;
2670 }
2671 
2672 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2673 {
2674 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2675 						     qgroup_rescan_work);
2676 	struct btrfs_path *path;
2677 	struct btrfs_trans_handle *trans = NULL;
2678 	int err = -ENOMEM;
2679 	int ret = 0;
2680 
2681 	path = btrfs_alloc_path();
2682 	if (!path)
2683 		goto out;
2684 
2685 	err = 0;
2686 	while (!err && !btrfs_fs_closing(fs_info)) {
2687 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
2688 		if (IS_ERR(trans)) {
2689 			err = PTR_ERR(trans);
2690 			break;
2691 		}
2692 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2693 			err = -EINTR;
2694 		} else {
2695 			err = qgroup_rescan_leaf(fs_info, path, trans);
2696 		}
2697 		if (err > 0)
2698 			btrfs_commit_transaction(trans);
2699 		else
2700 			btrfs_end_transaction(trans);
2701 	}
2702 
2703 out:
2704 	btrfs_free_path(path);
2705 
2706 	mutex_lock(&fs_info->qgroup_rescan_lock);
2707 	if (!btrfs_fs_closing(fs_info))
2708 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2709 
2710 	if (err > 0 &&
2711 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2712 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2713 	} else if (err < 0) {
2714 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2715 	}
2716 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2717 
2718 	/*
2719 	 * only update status, since the previous part has already updated the
2720 	 * qgroup info.
2721 	 */
2722 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
2723 	if (IS_ERR(trans)) {
2724 		err = PTR_ERR(trans);
2725 		btrfs_err(fs_info,
2726 			  "fail to start transaction for status update: %d",
2727 			  err);
2728 		goto done;
2729 	}
2730 	ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2731 	if (ret < 0) {
2732 		err = ret;
2733 		btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2734 	}
2735 	btrfs_end_transaction(trans);
2736 
2737 	if (btrfs_fs_closing(fs_info)) {
2738 		btrfs_info(fs_info, "qgroup scan paused");
2739 	} else if (err >= 0) {
2740 		btrfs_info(fs_info, "qgroup scan completed%s",
2741 			err > 0 ? " (inconsistency flag cleared)" : "");
2742 	} else {
2743 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
2744 	}
2745 
2746 done:
2747 	mutex_lock(&fs_info->qgroup_rescan_lock);
2748 	fs_info->qgroup_rescan_running = false;
2749 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2750 	complete_all(&fs_info->qgroup_rescan_completion);
2751 }
2752 
2753 /*
2754  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2755  * memory required for the rescan context.
2756  */
2757 static int
2758 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2759 		   int init_flags)
2760 {
2761 	int ret = 0;
2762 
2763 	if (!init_flags &&
2764 	    (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2765 	     !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2766 		ret = -EINVAL;
2767 		goto err;
2768 	}
2769 
2770 	mutex_lock(&fs_info->qgroup_rescan_lock);
2771 	spin_lock(&fs_info->qgroup_lock);
2772 
2773 	if (init_flags) {
2774 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2775 			ret = -EINPROGRESS;
2776 		else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2777 			ret = -EINVAL;
2778 
2779 		if (ret) {
2780 			spin_unlock(&fs_info->qgroup_lock);
2781 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2782 			goto err;
2783 		}
2784 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2785 	}
2786 
2787 	memset(&fs_info->qgroup_rescan_progress, 0,
2788 		sizeof(fs_info->qgroup_rescan_progress));
2789 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2790 	init_completion(&fs_info->qgroup_rescan_completion);
2791 	fs_info->qgroup_rescan_running = true;
2792 
2793 	spin_unlock(&fs_info->qgroup_lock);
2794 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2795 
2796 	memset(&fs_info->qgroup_rescan_work, 0,
2797 	       sizeof(fs_info->qgroup_rescan_work));
2798 	btrfs_init_work(&fs_info->qgroup_rescan_work,
2799 			btrfs_qgroup_rescan_helper,
2800 			btrfs_qgroup_rescan_worker, NULL, NULL);
2801 
2802 	if (ret) {
2803 err:
2804 		btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2805 		return ret;
2806 	}
2807 
2808 	return 0;
2809 }
2810 
2811 static void
2812 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2813 {
2814 	struct rb_node *n;
2815 	struct btrfs_qgroup *qgroup;
2816 
2817 	spin_lock(&fs_info->qgroup_lock);
2818 	/* clear all current qgroup tracking information */
2819 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2820 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
2821 		qgroup->rfer = 0;
2822 		qgroup->rfer_cmpr = 0;
2823 		qgroup->excl = 0;
2824 		qgroup->excl_cmpr = 0;
2825 	}
2826 	spin_unlock(&fs_info->qgroup_lock);
2827 }
2828 
2829 int
2830 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2831 {
2832 	int ret = 0;
2833 	struct btrfs_trans_handle *trans;
2834 
2835 	ret = qgroup_rescan_init(fs_info, 0, 1);
2836 	if (ret)
2837 		return ret;
2838 
2839 	/*
2840 	 * We have set the rescan_progress to 0, which means no more
2841 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2842 	 * However, btrfs_qgroup_account_ref may be right after its call
2843 	 * to btrfs_find_all_roots, in which case it would still do the
2844 	 * accounting.
2845 	 * To solve this, we're committing the transaction, which will
2846 	 * ensure we run all delayed refs and only after that, we are
2847 	 * going to clear all tracking information for a clean start.
2848 	 */
2849 
2850 	trans = btrfs_join_transaction(fs_info->fs_root);
2851 	if (IS_ERR(trans)) {
2852 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2853 		return PTR_ERR(trans);
2854 	}
2855 	ret = btrfs_commit_transaction(trans);
2856 	if (ret) {
2857 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2858 		return ret;
2859 	}
2860 
2861 	qgroup_rescan_zero_tracking(fs_info);
2862 
2863 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
2864 			 &fs_info->qgroup_rescan_work);
2865 
2866 	return 0;
2867 }
2868 
2869 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2870 				     bool interruptible)
2871 {
2872 	int running;
2873 	int ret = 0;
2874 
2875 	mutex_lock(&fs_info->qgroup_rescan_lock);
2876 	spin_lock(&fs_info->qgroup_lock);
2877 	running = fs_info->qgroup_rescan_running;
2878 	spin_unlock(&fs_info->qgroup_lock);
2879 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2880 
2881 	if (!running)
2882 		return 0;
2883 
2884 	if (interruptible)
2885 		ret = wait_for_completion_interruptible(
2886 					&fs_info->qgroup_rescan_completion);
2887 	else
2888 		wait_for_completion(&fs_info->qgroup_rescan_completion);
2889 
2890 	return ret;
2891 }
2892 
2893 /*
2894  * this is only called from open_ctree where we're still single threaded, thus
2895  * locking is omitted here.
2896  */
2897 void
2898 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2899 {
2900 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2901 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
2902 				 &fs_info->qgroup_rescan_work);
2903 }
2904 
2905 /*
2906  * Reserve qgroup space for range [start, start + len).
2907  *
2908  * This function will either reserve space from related qgroups or doing
2909  * nothing if the range is already reserved.
2910  *
2911  * Return 0 for successful reserve
2912  * Return <0 for error (including -EQUOT)
2913  *
2914  * NOTE: this function may sleep for memory allocation.
2915  *       if btrfs_qgroup_reserve_data() is called multiple times with
2916  *       same @reserved, caller must ensure when error happens it's OK
2917  *       to free *ALL* reserved space.
2918  */
2919 int btrfs_qgroup_reserve_data(struct inode *inode,
2920 			struct extent_changeset **reserved_ret, u64 start,
2921 			u64 len)
2922 {
2923 	struct btrfs_root *root = BTRFS_I(inode)->root;
2924 	struct ulist_node *unode;
2925 	struct ulist_iterator uiter;
2926 	struct extent_changeset *reserved;
2927 	u64 orig_reserved;
2928 	u64 to_reserve;
2929 	int ret;
2930 
2931 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2932 	    !is_fstree(root->objectid) || len == 0)
2933 		return 0;
2934 
2935 	/* @reserved parameter is mandatory for qgroup */
2936 	if (WARN_ON(!reserved_ret))
2937 		return -EINVAL;
2938 	if (!*reserved_ret) {
2939 		*reserved_ret = extent_changeset_alloc();
2940 		if (!*reserved_ret)
2941 			return -ENOMEM;
2942 	}
2943 	reserved = *reserved_ret;
2944 	/* Record already reserved space */
2945 	orig_reserved = reserved->bytes_changed;
2946 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2947 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
2948 
2949 	/* Newly reserved space */
2950 	to_reserve = reserved->bytes_changed - orig_reserved;
2951 	trace_btrfs_qgroup_reserve_data(inode, start, len,
2952 					to_reserve, QGROUP_RESERVE);
2953 	if (ret < 0)
2954 		goto cleanup;
2955 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
2956 	if (ret < 0)
2957 		goto cleanup;
2958 
2959 	return ret;
2960 
2961 cleanup:
2962 	/* cleanup *ALL* already reserved ranges */
2963 	ULIST_ITER_INIT(&uiter);
2964 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
2965 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2966 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
2967 	extent_changeset_release(reserved);
2968 	return ret;
2969 }
2970 
2971 /* Free ranges specified by @reserved, normally in error path */
2972 static int qgroup_free_reserved_data(struct inode *inode,
2973 			struct extent_changeset *reserved, u64 start, u64 len)
2974 {
2975 	struct btrfs_root *root = BTRFS_I(inode)->root;
2976 	struct ulist_node *unode;
2977 	struct ulist_iterator uiter;
2978 	struct extent_changeset changeset;
2979 	int freed = 0;
2980 	int ret;
2981 
2982 	extent_changeset_init(&changeset);
2983 	len = round_up(start + len, root->fs_info->sectorsize);
2984 	start = round_down(start, root->fs_info->sectorsize);
2985 
2986 	ULIST_ITER_INIT(&uiter);
2987 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
2988 		u64 range_start = unode->val;
2989 		/* unode->aux is the inclusive end */
2990 		u64 range_len = unode->aux - range_start + 1;
2991 		u64 free_start;
2992 		u64 free_len;
2993 
2994 		extent_changeset_release(&changeset);
2995 
2996 		/* Only free range in range [start, start + len) */
2997 		if (range_start >= start + len ||
2998 		    range_start + range_len <= start)
2999 			continue;
3000 		free_start = max(range_start, start);
3001 		free_len = min(start + len, range_start + range_len) -
3002 			   free_start;
3003 		/*
3004 		 * TODO: To also modify reserved->ranges_reserved to reflect
3005 		 * the modification.
3006 		 *
3007 		 * However as long as we free qgroup reserved according to
3008 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3009 		 * So not need to rush.
3010 		 */
3011 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3012 				free_start, free_start + free_len - 1,
3013 				EXTENT_QGROUP_RESERVED, &changeset);
3014 		if (ret < 0)
3015 			goto out;
3016 		freed += changeset.bytes_changed;
3017 	}
3018 	btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
3019 				  BTRFS_QGROUP_RSV_DATA);
3020 	ret = freed;
3021 out:
3022 	extent_changeset_release(&changeset);
3023 	return ret;
3024 }
3025 
3026 static int __btrfs_qgroup_release_data(struct inode *inode,
3027 			struct extent_changeset *reserved, u64 start, u64 len,
3028 			int free)
3029 {
3030 	struct extent_changeset changeset;
3031 	int trace_op = QGROUP_RELEASE;
3032 	int ret;
3033 
3034 	/* In release case, we shouldn't have @reserved */
3035 	WARN_ON(!free && reserved);
3036 	if (free && reserved)
3037 		return qgroup_free_reserved_data(inode, reserved, start, len);
3038 	extent_changeset_init(&changeset);
3039 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3040 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3041 	if (ret < 0)
3042 		goto out;
3043 
3044 	if (free)
3045 		trace_op = QGROUP_FREE;
3046 	trace_btrfs_qgroup_release_data(inode, start, len,
3047 					changeset.bytes_changed, trace_op);
3048 	if (free)
3049 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3050 				BTRFS_I(inode)->root->objectid,
3051 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3052 	ret = changeset.bytes_changed;
3053 out:
3054 	extent_changeset_release(&changeset);
3055 	return ret;
3056 }
3057 
3058 /*
3059  * Free a reserved space range from io_tree and related qgroups
3060  *
3061  * Should be called when a range of pages get invalidated before reaching disk.
3062  * Or for error cleanup case.
3063  * if @reserved is given, only reserved range in [@start, @start + @len) will
3064  * be freed.
3065  *
3066  * For data written to disk, use btrfs_qgroup_release_data().
3067  *
3068  * NOTE: This function may sleep for memory allocation.
3069  */
3070 int btrfs_qgroup_free_data(struct inode *inode,
3071 			struct extent_changeset *reserved, u64 start, u64 len)
3072 {
3073 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3074 }
3075 
3076 /*
3077  * Release a reserved space range from io_tree only.
3078  *
3079  * Should be called when a range of pages get written to disk and corresponding
3080  * FILE_EXTENT is inserted into corresponding root.
3081  *
3082  * Since new qgroup accounting framework will only update qgroup numbers at
3083  * commit_transaction() time, its reserved space shouldn't be freed from
3084  * related qgroups.
3085  *
3086  * But we should release the range from io_tree, to allow further write to be
3087  * COWed.
3088  *
3089  * NOTE: This function may sleep for memory allocation.
3090  */
3091 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3092 {
3093 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3094 }
3095 
3096 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3097 			      enum btrfs_qgroup_rsv_type type)
3098 {
3099 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3100 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3101 		return;
3102 	if (num_bytes == 0)
3103 		return;
3104 
3105 	spin_lock(&root->qgroup_meta_rsv_lock);
3106 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3107 		root->qgroup_meta_rsv_prealloc += num_bytes;
3108 	else
3109 		root->qgroup_meta_rsv_pertrans += num_bytes;
3110 	spin_unlock(&root->qgroup_meta_rsv_lock);
3111 }
3112 
3113 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3114 			     enum btrfs_qgroup_rsv_type type)
3115 {
3116 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3117 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3118 		return 0;
3119 	if (num_bytes == 0)
3120 		return 0;
3121 
3122 	spin_lock(&root->qgroup_meta_rsv_lock);
3123 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3124 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3125 				  num_bytes);
3126 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3127 	} else {
3128 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3129 				  num_bytes);
3130 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3131 	}
3132 	spin_unlock(&root->qgroup_meta_rsv_lock);
3133 	return num_bytes;
3134 }
3135 
3136 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3137 				enum btrfs_qgroup_rsv_type type, bool enforce)
3138 {
3139 	struct btrfs_fs_info *fs_info = root->fs_info;
3140 	int ret;
3141 
3142 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3143 	    !is_fstree(root->objectid) || num_bytes == 0)
3144 		return 0;
3145 
3146 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3147 	trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3148 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3149 	if (ret < 0)
3150 		return ret;
3151 	/*
3152 	 * Record what we have reserved into root.
3153 	 *
3154 	 * To avoid quota disabled->enabled underflow.
3155 	 * In that case, we may try to free space we haven't reserved
3156 	 * (since quota was disabled), so record what we reserved into root.
3157 	 * And ensure later release won't underflow this number.
3158 	 */
3159 	add_root_meta_rsv(root, num_bytes, type);
3160 	return ret;
3161 }
3162 
3163 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3164 {
3165 	struct btrfs_fs_info *fs_info = root->fs_info;
3166 
3167 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3168 	    !is_fstree(root->objectid))
3169 		return;
3170 
3171 	/* TODO: Update trace point to handle such free */
3172 	trace_qgroup_meta_free_all_pertrans(root);
3173 	/* Special value -1 means to free all reserved space */
3174 	btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3175 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3176 }
3177 
3178 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3179 			      enum btrfs_qgroup_rsv_type type)
3180 {
3181 	struct btrfs_fs_info *fs_info = root->fs_info;
3182 
3183 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3184 	    !is_fstree(root->objectid))
3185 		return;
3186 
3187 	/*
3188 	 * reservation for META_PREALLOC can happen before quota is enabled,
3189 	 * which can lead to underflow.
3190 	 * Here ensure we will only free what we really have reserved.
3191 	 */
3192 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3193 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3194 	trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3195 	btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3196 }
3197 
3198 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3199 				int num_bytes)
3200 {
3201 	struct btrfs_root *quota_root = fs_info->quota_root;
3202 	struct btrfs_qgroup *qgroup;
3203 	struct ulist_node *unode;
3204 	struct ulist_iterator uiter;
3205 	int ret = 0;
3206 
3207 	if (num_bytes == 0)
3208 		return;
3209 	if (!quota_root)
3210 		return;
3211 
3212 	spin_lock(&fs_info->qgroup_lock);
3213 	qgroup = find_qgroup_rb(fs_info, ref_root);
3214 	if (!qgroup)
3215 		goto out;
3216 	ulist_reinit(fs_info->qgroup_ulist);
3217 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3218 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3219 	if (ret < 0)
3220 		goto out;
3221 	ULIST_ITER_INIT(&uiter);
3222 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3223 		struct btrfs_qgroup *qg;
3224 		struct btrfs_qgroup_list *glist;
3225 
3226 		qg = unode_aux_to_qgroup(unode);
3227 
3228 		qgroup_rsv_release(fs_info, qg, num_bytes,
3229 				BTRFS_QGROUP_RSV_META_PREALLOC);
3230 		qgroup_rsv_add(fs_info, qg, num_bytes,
3231 				BTRFS_QGROUP_RSV_META_PERTRANS);
3232 		list_for_each_entry(glist, &qg->groups, next_group) {
3233 			ret = ulist_add(fs_info->qgroup_ulist,
3234 					glist->group->qgroupid,
3235 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3236 			if (ret < 0)
3237 				goto out;
3238 		}
3239 	}
3240 out:
3241 	spin_unlock(&fs_info->qgroup_lock);
3242 }
3243 
3244 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3245 {
3246 	struct btrfs_fs_info *fs_info = root->fs_info;
3247 
3248 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3249 	    !is_fstree(root->objectid))
3250 		return;
3251 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3252 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3253 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3254 	trace_qgroup_meta_convert(root, num_bytes);
3255 	qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3256 }
3257 
3258 /*
3259  * Check qgroup reserved space leaking, normally at destroy inode
3260  * time
3261  */
3262 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3263 {
3264 	struct extent_changeset changeset;
3265 	struct ulist_node *unode;
3266 	struct ulist_iterator iter;
3267 	int ret;
3268 
3269 	extent_changeset_init(&changeset);
3270 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3271 			EXTENT_QGROUP_RESERVED, &changeset);
3272 
3273 	WARN_ON(ret < 0);
3274 	if (WARN_ON(changeset.bytes_changed)) {
3275 		ULIST_ITER_INIT(&iter);
3276 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3277 			btrfs_warn(BTRFS_I(inode)->root->fs_info,
3278 				"leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3279 				inode->i_ino, unode->val, unode->aux);
3280 		}
3281 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3282 				BTRFS_I(inode)->root->objectid,
3283 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3284 
3285 	}
3286 	extent_changeset_release(&changeset);
3287 }
3288