xref: /openbmc/linux/fs/btrfs/qgroup.c (revision 4a3fad70)
1 /*
2  * Copyright (C) 2011 STRATO.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
25 #include <linux/workqueue.h>
26 #include <linux/btrfs.h>
27 
28 #include "ctree.h"
29 #include "transaction.h"
30 #include "disk-io.h"
31 #include "locking.h"
32 #include "ulist.h"
33 #include "backref.h"
34 #include "extent_io.h"
35 #include "qgroup.h"
36 
37 
38 /* TODO XXX FIXME
39  *  - subvol delete -> delete when ref goes to 0? delete limits also?
40  *  - reorganize keys
41  *  - compressed
42  *  - sync
43  *  - copy also limits on subvol creation
44  *  - limit
45  *  - caches fuer ulists
46  *  - performance benchmarks
47  *  - check all ioctl parameters
48  */
49 
50 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
51 					   int mod)
52 {
53 	if (qg->old_refcnt < seq)
54 		qg->old_refcnt = seq;
55 	qg->old_refcnt += mod;
56 }
57 
58 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
59 					   int mod)
60 {
61 	if (qg->new_refcnt < seq)
62 		qg->new_refcnt = seq;
63 	qg->new_refcnt += mod;
64 }
65 
66 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
67 {
68 	if (qg->old_refcnt < seq)
69 		return 0;
70 	return qg->old_refcnt - seq;
71 }
72 
73 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
74 {
75 	if (qg->new_refcnt < seq)
76 		return 0;
77 	return qg->new_refcnt - seq;
78 }
79 
80 /*
81  * glue structure to represent the relations between qgroups.
82  */
83 struct btrfs_qgroup_list {
84 	struct list_head next_group;
85 	struct list_head next_member;
86 	struct btrfs_qgroup *group;
87 	struct btrfs_qgroup *member;
88 };
89 
90 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
91 {
92 	return (u64)(uintptr_t)qg;
93 }
94 
95 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
96 {
97 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
98 }
99 
100 static int
101 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
102 		   int init_flags);
103 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
104 
105 /* must be called with qgroup_ioctl_lock held */
106 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
107 					   u64 qgroupid)
108 {
109 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
110 	struct btrfs_qgroup *qgroup;
111 
112 	while (n) {
113 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
114 		if (qgroup->qgroupid < qgroupid)
115 			n = n->rb_left;
116 		else if (qgroup->qgroupid > qgroupid)
117 			n = n->rb_right;
118 		else
119 			return qgroup;
120 	}
121 	return NULL;
122 }
123 
124 /* must be called with qgroup_lock held */
125 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
126 					  u64 qgroupid)
127 {
128 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
129 	struct rb_node *parent = NULL;
130 	struct btrfs_qgroup *qgroup;
131 
132 	while (*p) {
133 		parent = *p;
134 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
135 
136 		if (qgroup->qgroupid < qgroupid)
137 			p = &(*p)->rb_left;
138 		else if (qgroup->qgroupid > qgroupid)
139 			p = &(*p)->rb_right;
140 		else
141 			return qgroup;
142 	}
143 
144 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
145 	if (!qgroup)
146 		return ERR_PTR(-ENOMEM);
147 
148 	qgroup->qgroupid = qgroupid;
149 	INIT_LIST_HEAD(&qgroup->groups);
150 	INIT_LIST_HEAD(&qgroup->members);
151 	INIT_LIST_HEAD(&qgroup->dirty);
152 
153 	rb_link_node(&qgroup->node, parent, p);
154 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
155 
156 	return qgroup;
157 }
158 
159 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
160 {
161 	struct btrfs_qgroup_list *list;
162 
163 	list_del(&qgroup->dirty);
164 	while (!list_empty(&qgroup->groups)) {
165 		list = list_first_entry(&qgroup->groups,
166 					struct btrfs_qgroup_list, next_group);
167 		list_del(&list->next_group);
168 		list_del(&list->next_member);
169 		kfree(list);
170 	}
171 
172 	while (!list_empty(&qgroup->members)) {
173 		list = list_first_entry(&qgroup->members,
174 					struct btrfs_qgroup_list, next_member);
175 		list_del(&list->next_group);
176 		list_del(&list->next_member);
177 		kfree(list);
178 	}
179 	kfree(qgroup);
180 }
181 
182 /* must be called with qgroup_lock held */
183 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
184 {
185 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
186 
187 	if (!qgroup)
188 		return -ENOENT;
189 
190 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
191 	__del_qgroup_rb(qgroup);
192 	return 0;
193 }
194 
195 /* must be called with qgroup_lock held */
196 static int add_relation_rb(struct btrfs_fs_info *fs_info,
197 			   u64 memberid, u64 parentid)
198 {
199 	struct btrfs_qgroup *member;
200 	struct btrfs_qgroup *parent;
201 	struct btrfs_qgroup_list *list;
202 
203 	member = find_qgroup_rb(fs_info, memberid);
204 	parent = find_qgroup_rb(fs_info, parentid);
205 	if (!member || !parent)
206 		return -ENOENT;
207 
208 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
209 	if (!list)
210 		return -ENOMEM;
211 
212 	list->group = parent;
213 	list->member = member;
214 	list_add_tail(&list->next_group, &member->groups);
215 	list_add_tail(&list->next_member, &parent->members);
216 
217 	return 0;
218 }
219 
220 /* must be called with qgroup_lock held */
221 static int del_relation_rb(struct btrfs_fs_info *fs_info,
222 			   u64 memberid, u64 parentid)
223 {
224 	struct btrfs_qgroup *member;
225 	struct btrfs_qgroup *parent;
226 	struct btrfs_qgroup_list *list;
227 
228 	member = find_qgroup_rb(fs_info, memberid);
229 	parent = find_qgroup_rb(fs_info, parentid);
230 	if (!member || !parent)
231 		return -ENOENT;
232 
233 	list_for_each_entry(list, &member->groups, next_group) {
234 		if (list->group == parent) {
235 			list_del(&list->next_group);
236 			list_del(&list->next_member);
237 			kfree(list);
238 			return 0;
239 		}
240 	}
241 	return -ENOENT;
242 }
243 
244 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
245 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
246 			       u64 rfer, u64 excl)
247 {
248 	struct btrfs_qgroup *qgroup;
249 
250 	qgroup = find_qgroup_rb(fs_info, qgroupid);
251 	if (!qgroup)
252 		return -EINVAL;
253 	if (qgroup->rfer != rfer || qgroup->excl != excl)
254 		return -EINVAL;
255 	return 0;
256 }
257 #endif
258 
259 /*
260  * The full config is read in one go, only called from open_ctree()
261  * It doesn't use any locking, as at this point we're still single-threaded
262  */
263 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
264 {
265 	struct btrfs_key key;
266 	struct btrfs_key found_key;
267 	struct btrfs_root *quota_root = fs_info->quota_root;
268 	struct btrfs_path *path = NULL;
269 	struct extent_buffer *l;
270 	int slot;
271 	int ret = 0;
272 	u64 flags = 0;
273 	u64 rescan_progress = 0;
274 
275 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
276 		return 0;
277 
278 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
279 	if (!fs_info->qgroup_ulist) {
280 		ret = -ENOMEM;
281 		goto out;
282 	}
283 
284 	path = btrfs_alloc_path();
285 	if (!path) {
286 		ret = -ENOMEM;
287 		goto out;
288 	}
289 
290 	/* default this to quota off, in case no status key is found */
291 	fs_info->qgroup_flags = 0;
292 
293 	/*
294 	 * pass 1: read status, all qgroup infos and limits
295 	 */
296 	key.objectid = 0;
297 	key.type = 0;
298 	key.offset = 0;
299 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
300 	if (ret)
301 		goto out;
302 
303 	while (1) {
304 		struct btrfs_qgroup *qgroup;
305 
306 		slot = path->slots[0];
307 		l = path->nodes[0];
308 		btrfs_item_key_to_cpu(l, &found_key, slot);
309 
310 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
311 			struct btrfs_qgroup_status_item *ptr;
312 
313 			ptr = btrfs_item_ptr(l, slot,
314 					     struct btrfs_qgroup_status_item);
315 
316 			if (btrfs_qgroup_status_version(l, ptr) !=
317 			    BTRFS_QGROUP_STATUS_VERSION) {
318 				btrfs_err(fs_info,
319 				 "old qgroup version, quota disabled");
320 				goto out;
321 			}
322 			if (btrfs_qgroup_status_generation(l, ptr) !=
323 			    fs_info->generation) {
324 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
325 				btrfs_err(fs_info,
326 					"qgroup generation mismatch, marked as inconsistent");
327 			}
328 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
329 									  ptr);
330 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
331 			goto next1;
332 		}
333 
334 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
335 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
336 			goto next1;
337 
338 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
339 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
340 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
341 			btrfs_err(fs_info, "inconsistent qgroup config");
342 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
343 		}
344 		if (!qgroup) {
345 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
346 			if (IS_ERR(qgroup)) {
347 				ret = PTR_ERR(qgroup);
348 				goto out;
349 			}
350 		}
351 		switch (found_key.type) {
352 		case BTRFS_QGROUP_INFO_KEY: {
353 			struct btrfs_qgroup_info_item *ptr;
354 
355 			ptr = btrfs_item_ptr(l, slot,
356 					     struct btrfs_qgroup_info_item);
357 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
358 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
359 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
360 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
361 			/* generation currently unused */
362 			break;
363 		}
364 		case BTRFS_QGROUP_LIMIT_KEY: {
365 			struct btrfs_qgroup_limit_item *ptr;
366 
367 			ptr = btrfs_item_ptr(l, slot,
368 					     struct btrfs_qgroup_limit_item);
369 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
370 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
371 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
372 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
373 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
374 			break;
375 		}
376 		}
377 next1:
378 		ret = btrfs_next_item(quota_root, path);
379 		if (ret < 0)
380 			goto out;
381 		if (ret)
382 			break;
383 	}
384 	btrfs_release_path(path);
385 
386 	/*
387 	 * pass 2: read all qgroup relations
388 	 */
389 	key.objectid = 0;
390 	key.type = BTRFS_QGROUP_RELATION_KEY;
391 	key.offset = 0;
392 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
393 	if (ret)
394 		goto out;
395 	while (1) {
396 		slot = path->slots[0];
397 		l = path->nodes[0];
398 		btrfs_item_key_to_cpu(l, &found_key, slot);
399 
400 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
401 			goto next2;
402 
403 		if (found_key.objectid > found_key.offset) {
404 			/* parent <- member, not needed to build config */
405 			/* FIXME should we omit the key completely? */
406 			goto next2;
407 		}
408 
409 		ret = add_relation_rb(fs_info, found_key.objectid,
410 				      found_key.offset);
411 		if (ret == -ENOENT) {
412 			btrfs_warn(fs_info,
413 				"orphan qgroup relation 0x%llx->0x%llx",
414 				found_key.objectid, found_key.offset);
415 			ret = 0;	/* ignore the error */
416 		}
417 		if (ret)
418 			goto out;
419 next2:
420 		ret = btrfs_next_item(quota_root, path);
421 		if (ret < 0)
422 			goto out;
423 		if (ret)
424 			break;
425 	}
426 out:
427 	fs_info->qgroup_flags |= flags;
428 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
429 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
430 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
431 		 ret >= 0)
432 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
433 	btrfs_free_path(path);
434 
435 	if (ret < 0) {
436 		ulist_free(fs_info->qgroup_ulist);
437 		fs_info->qgroup_ulist = NULL;
438 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
439 	}
440 
441 	return ret < 0 ? ret : 0;
442 }
443 
444 /*
445  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
446  * first two are in single-threaded paths.And for the third one, we have set
447  * quota_root to be null with qgroup_lock held before, so it is safe to clean
448  * up the in-memory structures without qgroup_lock held.
449  */
450 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
451 {
452 	struct rb_node *n;
453 	struct btrfs_qgroup *qgroup;
454 
455 	while ((n = rb_first(&fs_info->qgroup_tree))) {
456 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
457 		rb_erase(n, &fs_info->qgroup_tree);
458 		__del_qgroup_rb(qgroup);
459 	}
460 	/*
461 	 * we call btrfs_free_qgroup_config() when umounting
462 	 * filesystem and disabling quota, so we set qgroup_ulist
463 	 * to be null here to avoid double free.
464 	 */
465 	ulist_free(fs_info->qgroup_ulist);
466 	fs_info->qgroup_ulist = NULL;
467 }
468 
469 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
470 				    struct btrfs_root *quota_root,
471 				    u64 src, u64 dst)
472 {
473 	int ret;
474 	struct btrfs_path *path;
475 	struct btrfs_key key;
476 
477 	path = btrfs_alloc_path();
478 	if (!path)
479 		return -ENOMEM;
480 
481 	key.objectid = src;
482 	key.type = BTRFS_QGROUP_RELATION_KEY;
483 	key.offset = dst;
484 
485 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
486 
487 	btrfs_mark_buffer_dirty(path->nodes[0]);
488 
489 	btrfs_free_path(path);
490 	return ret;
491 }
492 
493 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
494 				    struct btrfs_root *quota_root,
495 				    u64 src, u64 dst)
496 {
497 	int ret;
498 	struct btrfs_path *path;
499 	struct btrfs_key key;
500 
501 	path = btrfs_alloc_path();
502 	if (!path)
503 		return -ENOMEM;
504 
505 	key.objectid = src;
506 	key.type = BTRFS_QGROUP_RELATION_KEY;
507 	key.offset = dst;
508 
509 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
510 	if (ret < 0)
511 		goto out;
512 
513 	if (ret > 0) {
514 		ret = -ENOENT;
515 		goto out;
516 	}
517 
518 	ret = btrfs_del_item(trans, quota_root, path);
519 out:
520 	btrfs_free_path(path);
521 	return ret;
522 }
523 
524 static int add_qgroup_item(struct btrfs_trans_handle *trans,
525 			   struct btrfs_root *quota_root, u64 qgroupid)
526 {
527 	int ret;
528 	struct btrfs_path *path;
529 	struct btrfs_qgroup_info_item *qgroup_info;
530 	struct btrfs_qgroup_limit_item *qgroup_limit;
531 	struct extent_buffer *leaf;
532 	struct btrfs_key key;
533 
534 	if (btrfs_is_testing(quota_root->fs_info))
535 		return 0;
536 
537 	path = btrfs_alloc_path();
538 	if (!path)
539 		return -ENOMEM;
540 
541 	key.objectid = 0;
542 	key.type = BTRFS_QGROUP_INFO_KEY;
543 	key.offset = qgroupid;
544 
545 	/*
546 	 * Avoid a transaction abort by catching -EEXIST here. In that
547 	 * case, we proceed by re-initializing the existing structure
548 	 * on disk.
549 	 */
550 
551 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
552 				      sizeof(*qgroup_info));
553 	if (ret && ret != -EEXIST)
554 		goto out;
555 
556 	leaf = path->nodes[0];
557 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
558 				 struct btrfs_qgroup_info_item);
559 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
560 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
561 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
562 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
563 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
564 
565 	btrfs_mark_buffer_dirty(leaf);
566 
567 	btrfs_release_path(path);
568 
569 	key.type = BTRFS_QGROUP_LIMIT_KEY;
570 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
571 				      sizeof(*qgroup_limit));
572 	if (ret && ret != -EEXIST)
573 		goto out;
574 
575 	leaf = path->nodes[0];
576 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
577 				  struct btrfs_qgroup_limit_item);
578 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
579 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
580 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
581 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
582 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
583 
584 	btrfs_mark_buffer_dirty(leaf);
585 
586 	ret = 0;
587 out:
588 	btrfs_free_path(path);
589 	return ret;
590 }
591 
592 static int del_qgroup_item(struct btrfs_trans_handle *trans,
593 			   struct btrfs_root *quota_root, u64 qgroupid)
594 {
595 	int ret;
596 	struct btrfs_path *path;
597 	struct btrfs_key key;
598 
599 	path = btrfs_alloc_path();
600 	if (!path)
601 		return -ENOMEM;
602 
603 	key.objectid = 0;
604 	key.type = BTRFS_QGROUP_INFO_KEY;
605 	key.offset = qgroupid;
606 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
607 	if (ret < 0)
608 		goto out;
609 
610 	if (ret > 0) {
611 		ret = -ENOENT;
612 		goto out;
613 	}
614 
615 	ret = btrfs_del_item(trans, quota_root, path);
616 	if (ret)
617 		goto out;
618 
619 	btrfs_release_path(path);
620 
621 	key.type = BTRFS_QGROUP_LIMIT_KEY;
622 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
623 	if (ret < 0)
624 		goto out;
625 
626 	if (ret > 0) {
627 		ret = -ENOENT;
628 		goto out;
629 	}
630 
631 	ret = btrfs_del_item(trans, quota_root, path);
632 
633 out:
634 	btrfs_free_path(path);
635 	return ret;
636 }
637 
638 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
639 				    struct btrfs_root *root,
640 				    struct btrfs_qgroup *qgroup)
641 {
642 	struct btrfs_path *path;
643 	struct btrfs_key key;
644 	struct extent_buffer *l;
645 	struct btrfs_qgroup_limit_item *qgroup_limit;
646 	int ret;
647 	int slot;
648 
649 	key.objectid = 0;
650 	key.type = BTRFS_QGROUP_LIMIT_KEY;
651 	key.offset = qgroup->qgroupid;
652 
653 	path = btrfs_alloc_path();
654 	if (!path)
655 		return -ENOMEM;
656 
657 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
658 	if (ret > 0)
659 		ret = -ENOENT;
660 
661 	if (ret)
662 		goto out;
663 
664 	l = path->nodes[0];
665 	slot = path->slots[0];
666 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
667 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
668 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
669 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
670 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
671 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
672 
673 	btrfs_mark_buffer_dirty(l);
674 
675 out:
676 	btrfs_free_path(path);
677 	return ret;
678 }
679 
680 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
681 				   struct btrfs_root *root,
682 				   struct btrfs_qgroup *qgroup)
683 {
684 	struct btrfs_path *path;
685 	struct btrfs_key key;
686 	struct extent_buffer *l;
687 	struct btrfs_qgroup_info_item *qgroup_info;
688 	int ret;
689 	int slot;
690 
691 	if (btrfs_is_testing(root->fs_info))
692 		return 0;
693 
694 	key.objectid = 0;
695 	key.type = BTRFS_QGROUP_INFO_KEY;
696 	key.offset = qgroup->qgroupid;
697 
698 	path = btrfs_alloc_path();
699 	if (!path)
700 		return -ENOMEM;
701 
702 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
703 	if (ret > 0)
704 		ret = -ENOENT;
705 
706 	if (ret)
707 		goto out;
708 
709 	l = path->nodes[0];
710 	slot = path->slots[0];
711 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
712 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
713 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
714 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
715 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
716 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
717 
718 	btrfs_mark_buffer_dirty(l);
719 
720 out:
721 	btrfs_free_path(path);
722 	return ret;
723 }
724 
725 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
726 				     struct btrfs_fs_info *fs_info,
727 				    struct btrfs_root *root)
728 {
729 	struct btrfs_path *path;
730 	struct btrfs_key key;
731 	struct extent_buffer *l;
732 	struct btrfs_qgroup_status_item *ptr;
733 	int ret;
734 	int slot;
735 
736 	key.objectid = 0;
737 	key.type = BTRFS_QGROUP_STATUS_KEY;
738 	key.offset = 0;
739 
740 	path = btrfs_alloc_path();
741 	if (!path)
742 		return -ENOMEM;
743 
744 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
745 	if (ret > 0)
746 		ret = -ENOENT;
747 
748 	if (ret)
749 		goto out;
750 
751 	l = path->nodes[0];
752 	slot = path->slots[0];
753 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
754 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
755 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
756 	btrfs_set_qgroup_status_rescan(l, ptr,
757 				fs_info->qgroup_rescan_progress.objectid);
758 
759 	btrfs_mark_buffer_dirty(l);
760 
761 out:
762 	btrfs_free_path(path);
763 	return ret;
764 }
765 
766 /*
767  * called with qgroup_lock held
768  */
769 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
770 				  struct btrfs_root *root)
771 {
772 	struct btrfs_path *path;
773 	struct btrfs_key key;
774 	struct extent_buffer *leaf = NULL;
775 	int ret;
776 	int nr = 0;
777 
778 	path = btrfs_alloc_path();
779 	if (!path)
780 		return -ENOMEM;
781 
782 	path->leave_spinning = 1;
783 
784 	key.objectid = 0;
785 	key.offset = 0;
786 	key.type = 0;
787 
788 	while (1) {
789 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
790 		if (ret < 0)
791 			goto out;
792 		leaf = path->nodes[0];
793 		nr = btrfs_header_nritems(leaf);
794 		if (!nr)
795 			break;
796 		/*
797 		 * delete the leaf one by one
798 		 * since the whole tree is going
799 		 * to be deleted.
800 		 */
801 		path->slots[0] = 0;
802 		ret = btrfs_del_items(trans, root, path, 0, nr);
803 		if (ret)
804 			goto out;
805 
806 		btrfs_release_path(path);
807 	}
808 	ret = 0;
809 out:
810 	btrfs_free_path(path);
811 	return ret;
812 }
813 
814 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
815 		       struct btrfs_fs_info *fs_info)
816 {
817 	struct btrfs_root *quota_root;
818 	struct btrfs_root *tree_root = fs_info->tree_root;
819 	struct btrfs_path *path = NULL;
820 	struct btrfs_qgroup_status_item *ptr;
821 	struct extent_buffer *leaf;
822 	struct btrfs_key key;
823 	struct btrfs_key found_key;
824 	struct btrfs_qgroup *qgroup = NULL;
825 	int ret = 0;
826 	int slot;
827 
828 	mutex_lock(&fs_info->qgroup_ioctl_lock);
829 	if (fs_info->quota_root) {
830 		set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
831 		goto out;
832 	}
833 
834 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
835 	if (!fs_info->qgroup_ulist) {
836 		ret = -ENOMEM;
837 		goto out;
838 	}
839 
840 	/*
841 	 * initially create the quota tree
842 	 */
843 	quota_root = btrfs_create_tree(trans, fs_info,
844 				       BTRFS_QUOTA_TREE_OBJECTID);
845 	if (IS_ERR(quota_root)) {
846 		ret =  PTR_ERR(quota_root);
847 		goto out;
848 	}
849 
850 	path = btrfs_alloc_path();
851 	if (!path) {
852 		ret = -ENOMEM;
853 		goto out_free_root;
854 	}
855 
856 	key.objectid = 0;
857 	key.type = BTRFS_QGROUP_STATUS_KEY;
858 	key.offset = 0;
859 
860 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
861 				      sizeof(*ptr));
862 	if (ret)
863 		goto out_free_path;
864 
865 	leaf = path->nodes[0];
866 	ptr = btrfs_item_ptr(leaf, path->slots[0],
867 				 struct btrfs_qgroup_status_item);
868 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
869 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
870 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
871 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
872 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
873 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
874 
875 	btrfs_mark_buffer_dirty(leaf);
876 
877 	key.objectid = 0;
878 	key.type = BTRFS_ROOT_REF_KEY;
879 	key.offset = 0;
880 
881 	btrfs_release_path(path);
882 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
883 	if (ret > 0)
884 		goto out_add_root;
885 	if (ret < 0)
886 		goto out_free_path;
887 
888 
889 	while (1) {
890 		slot = path->slots[0];
891 		leaf = path->nodes[0];
892 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
893 
894 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
895 			ret = add_qgroup_item(trans, quota_root,
896 					      found_key.offset);
897 			if (ret)
898 				goto out_free_path;
899 
900 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
901 			if (IS_ERR(qgroup)) {
902 				ret = PTR_ERR(qgroup);
903 				goto out_free_path;
904 			}
905 		}
906 		ret = btrfs_next_item(tree_root, path);
907 		if (ret < 0)
908 			goto out_free_path;
909 		if (ret)
910 			break;
911 	}
912 
913 out_add_root:
914 	btrfs_release_path(path);
915 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
916 	if (ret)
917 		goto out_free_path;
918 
919 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
920 	if (IS_ERR(qgroup)) {
921 		ret = PTR_ERR(qgroup);
922 		goto out_free_path;
923 	}
924 	spin_lock(&fs_info->qgroup_lock);
925 	fs_info->quota_root = quota_root;
926 	set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
927 	spin_unlock(&fs_info->qgroup_lock);
928 out_free_path:
929 	btrfs_free_path(path);
930 out_free_root:
931 	if (ret) {
932 		free_extent_buffer(quota_root->node);
933 		free_extent_buffer(quota_root->commit_root);
934 		kfree(quota_root);
935 	}
936 out:
937 	if (ret) {
938 		ulist_free(fs_info->qgroup_ulist);
939 		fs_info->qgroup_ulist = NULL;
940 	}
941 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
942 	return ret;
943 }
944 
945 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
946 			struct btrfs_fs_info *fs_info)
947 {
948 	struct btrfs_root *quota_root;
949 	int ret = 0;
950 
951 	mutex_lock(&fs_info->qgroup_ioctl_lock);
952 	if (!fs_info->quota_root)
953 		goto out;
954 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
955 	btrfs_qgroup_wait_for_completion(fs_info, false);
956 	spin_lock(&fs_info->qgroup_lock);
957 	quota_root = fs_info->quota_root;
958 	fs_info->quota_root = NULL;
959 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
960 	spin_unlock(&fs_info->qgroup_lock);
961 
962 	btrfs_free_qgroup_config(fs_info);
963 
964 	ret = btrfs_clean_quota_tree(trans, quota_root);
965 	if (ret)
966 		goto out;
967 
968 	ret = btrfs_del_root(trans, fs_info, &quota_root->root_key);
969 	if (ret)
970 		goto out;
971 
972 	list_del(&quota_root->dirty_list);
973 
974 	btrfs_tree_lock(quota_root->node);
975 	clean_tree_block(fs_info, quota_root->node);
976 	btrfs_tree_unlock(quota_root->node);
977 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
978 
979 	free_extent_buffer(quota_root->node);
980 	free_extent_buffer(quota_root->commit_root);
981 	kfree(quota_root);
982 out:
983 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
984 	return ret;
985 }
986 
987 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
988 			 struct btrfs_qgroup *qgroup)
989 {
990 	if (list_empty(&qgroup->dirty))
991 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
992 }
993 
994 static void report_reserved_underflow(struct btrfs_fs_info *fs_info,
995 				      struct btrfs_qgroup *qgroup,
996 				      u64 num_bytes)
997 {
998 #ifdef CONFIG_BTRFS_DEBUG
999 	WARN_ON(qgroup->reserved < num_bytes);
1000 	btrfs_debug(fs_info,
1001 		"qgroup %llu reserved space underflow, have: %llu, to free: %llu",
1002 		qgroup->qgroupid, qgroup->reserved, num_bytes);
1003 #endif
1004 	qgroup->reserved = 0;
1005 }
1006 /*
1007  * The easy accounting, if we are adding/removing the only ref for an extent
1008  * then this qgroup and all of the parent qgroups get their reference and
1009  * exclusive counts adjusted.
1010  *
1011  * Caller should hold fs_info->qgroup_lock.
1012  */
1013 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1014 				    struct ulist *tmp, u64 ref_root,
1015 				    u64 num_bytes, int sign)
1016 {
1017 	struct btrfs_qgroup *qgroup;
1018 	struct btrfs_qgroup_list *glist;
1019 	struct ulist_node *unode;
1020 	struct ulist_iterator uiter;
1021 	int ret = 0;
1022 
1023 	qgroup = find_qgroup_rb(fs_info, ref_root);
1024 	if (!qgroup)
1025 		goto out;
1026 
1027 	qgroup->rfer += sign * num_bytes;
1028 	qgroup->rfer_cmpr += sign * num_bytes;
1029 
1030 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1031 	qgroup->excl += sign * num_bytes;
1032 	qgroup->excl_cmpr += sign * num_bytes;
1033 	if (sign > 0) {
1034 		trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes);
1035 		if (qgroup->reserved < num_bytes)
1036 			report_reserved_underflow(fs_info, qgroup, num_bytes);
1037 		else
1038 			qgroup->reserved -= num_bytes;
1039 	}
1040 
1041 	qgroup_dirty(fs_info, qgroup);
1042 
1043 	/* Get all of the parent groups that contain this qgroup */
1044 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1045 		ret = ulist_add(tmp, glist->group->qgroupid,
1046 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1047 		if (ret < 0)
1048 			goto out;
1049 	}
1050 
1051 	/* Iterate all of the parents and adjust their reference counts */
1052 	ULIST_ITER_INIT(&uiter);
1053 	while ((unode = ulist_next(tmp, &uiter))) {
1054 		qgroup = unode_aux_to_qgroup(unode);
1055 		qgroup->rfer += sign * num_bytes;
1056 		qgroup->rfer_cmpr += sign * num_bytes;
1057 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1058 		qgroup->excl += sign * num_bytes;
1059 		if (sign > 0) {
1060 			trace_qgroup_update_reserve(fs_info, qgroup,
1061 						    -(s64)num_bytes);
1062 			if (qgroup->reserved < num_bytes)
1063 				report_reserved_underflow(fs_info, qgroup,
1064 							  num_bytes);
1065 			else
1066 				qgroup->reserved -= num_bytes;
1067 		}
1068 		qgroup->excl_cmpr += sign * num_bytes;
1069 		qgroup_dirty(fs_info, qgroup);
1070 
1071 		/* Add any parents of the parents */
1072 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1073 			ret = ulist_add(tmp, glist->group->qgroupid,
1074 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1075 			if (ret < 0)
1076 				goto out;
1077 		}
1078 	}
1079 	ret = 0;
1080 out:
1081 	return ret;
1082 }
1083 
1084 
1085 /*
1086  * Quick path for updating qgroup with only excl refs.
1087  *
1088  * In that case, just update all parent will be enough.
1089  * Or we needs to do a full rescan.
1090  * Caller should also hold fs_info->qgroup_lock.
1091  *
1092  * Return 0 for quick update, return >0 for need to full rescan
1093  * and mark INCONSISTENT flag.
1094  * Return < 0 for other error.
1095  */
1096 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1097 				   struct ulist *tmp, u64 src, u64 dst,
1098 				   int sign)
1099 {
1100 	struct btrfs_qgroup *qgroup;
1101 	int ret = 1;
1102 	int err = 0;
1103 
1104 	qgroup = find_qgroup_rb(fs_info, src);
1105 	if (!qgroup)
1106 		goto out;
1107 	if (qgroup->excl == qgroup->rfer) {
1108 		ret = 0;
1109 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1110 					       qgroup->excl, sign);
1111 		if (err < 0) {
1112 			ret = err;
1113 			goto out;
1114 		}
1115 	}
1116 out:
1117 	if (ret)
1118 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1119 	return ret;
1120 }
1121 
1122 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1123 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1124 {
1125 	struct btrfs_root *quota_root;
1126 	struct btrfs_qgroup *parent;
1127 	struct btrfs_qgroup *member;
1128 	struct btrfs_qgroup_list *list;
1129 	struct ulist *tmp;
1130 	int ret = 0;
1131 
1132 	/* Check the level of src and dst first */
1133 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1134 		return -EINVAL;
1135 
1136 	tmp = ulist_alloc(GFP_KERNEL);
1137 	if (!tmp)
1138 		return -ENOMEM;
1139 
1140 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1141 	quota_root = fs_info->quota_root;
1142 	if (!quota_root) {
1143 		ret = -EINVAL;
1144 		goto out;
1145 	}
1146 	member = find_qgroup_rb(fs_info, src);
1147 	parent = find_qgroup_rb(fs_info, dst);
1148 	if (!member || !parent) {
1149 		ret = -EINVAL;
1150 		goto out;
1151 	}
1152 
1153 	/* check if such qgroup relation exist firstly */
1154 	list_for_each_entry(list, &member->groups, next_group) {
1155 		if (list->group == parent) {
1156 			ret = -EEXIST;
1157 			goto out;
1158 		}
1159 	}
1160 
1161 	ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1162 	if (ret)
1163 		goto out;
1164 
1165 	ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1166 	if (ret) {
1167 		del_qgroup_relation_item(trans, quota_root, src, dst);
1168 		goto out;
1169 	}
1170 
1171 	spin_lock(&fs_info->qgroup_lock);
1172 	ret = add_relation_rb(fs_info, src, dst);
1173 	if (ret < 0) {
1174 		spin_unlock(&fs_info->qgroup_lock);
1175 		goto out;
1176 	}
1177 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1178 	spin_unlock(&fs_info->qgroup_lock);
1179 out:
1180 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1181 	ulist_free(tmp);
1182 	return ret;
1183 }
1184 
1185 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1186 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1187 {
1188 	struct btrfs_root *quota_root;
1189 	struct btrfs_qgroup *parent;
1190 	struct btrfs_qgroup *member;
1191 	struct btrfs_qgroup_list *list;
1192 	struct ulist *tmp;
1193 	int ret = 0;
1194 	int err;
1195 
1196 	tmp = ulist_alloc(GFP_KERNEL);
1197 	if (!tmp)
1198 		return -ENOMEM;
1199 
1200 	quota_root = fs_info->quota_root;
1201 	if (!quota_root) {
1202 		ret = -EINVAL;
1203 		goto out;
1204 	}
1205 
1206 	member = find_qgroup_rb(fs_info, src);
1207 	parent = find_qgroup_rb(fs_info, dst);
1208 	if (!member || !parent) {
1209 		ret = -EINVAL;
1210 		goto out;
1211 	}
1212 
1213 	/* check if such qgroup relation exist firstly */
1214 	list_for_each_entry(list, &member->groups, next_group) {
1215 		if (list->group == parent)
1216 			goto exist;
1217 	}
1218 	ret = -ENOENT;
1219 	goto out;
1220 exist:
1221 	ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1222 	err = del_qgroup_relation_item(trans, quota_root, dst, src);
1223 	if (err && !ret)
1224 		ret = err;
1225 
1226 	spin_lock(&fs_info->qgroup_lock);
1227 	del_relation_rb(fs_info, src, dst);
1228 	ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1229 	spin_unlock(&fs_info->qgroup_lock);
1230 out:
1231 	ulist_free(tmp);
1232 	return ret;
1233 }
1234 
1235 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1236 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1237 {
1238 	int ret = 0;
1239 
1240 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1241 	ret = __del_qgroup_relation(trans, fs_info, src, dst);
1242 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1243 
1244 	return ret;
1245 }
1246 
1247 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1248 			struct btrfs_fs_info *fs_info, u64 qgroupid)
1249 {
1250 	struct btrfs_root *quota_root;
1251 	struct btrfs_qgroup *qgroup;
1252 	int ret = 0;
1253 
1254 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1255 	quota_root = fs_info->quota_root;
1256 	if (!quota_root) {
1257 		ret = -EINVAL;
1258 		goto out;
1259 	}
1260 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1261 	if (qgroup) {
1262 		ret = -EEXIST;
1263 		goto out;
1264 	}
1265 
1266 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1267 	if (ret)
1268 		goto out;
1269 
1270 	spin_lock(&fs_info->qgroup_lock);
1271 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1272 	spin_unlock(&fs_info->qgroup_lock);
1273 
1274 	if (IS_ERR(qgroup))
1275 		ret = PTR_ERR(qgroup);
1276 out:
1277 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1278 	return ret;
1279 }
1280 
1281 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1282 			struct btrfs_fs_info *fs_info, u64 qgroupid)
1283 {
1284 	struct btrfs_root *quota_root;
1285 	struct btrfs_qgroup *qgroup;
1286 	struct btrfs_qgroup_list *list;
1287 	int ret = 0;
1288 
1289 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1290 	quota_root = fs_info->quota_root;
1291 	if (!quota_root) {
1292 		ret = -EINVAL;
1293 		goto out;
1294 	}
1295 
1296 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1297 	if (!qgroup) {
1298 		ret = -ENOENT;
1299 		goto out;
1300 	} else {
1301 		/* check if there are no children of this qgroup */
1302 		if (!list_empty(&qgroup->members)) {
1303 			ret = -EBUSY;
1304 			goto out;
1305 		}
1306 	}
1307 	ret = del_qgroup_item(trans, quota_root, qgroupid);
1308 	if (ret && ret != -ENOENT)
1309 		goto out;
1310 
1311 	while (!list_empty(&qgroup->groups)) {
1312 		list = list_first_entry(&qgroup->groups,
1313 					struct btrfs_qgroup_list, next_group);
1314 		ret = __del_qgroup_relation(trans, fs_info,
1315 					   qgroupid,
1316 					   list->group->qgroupid);
1317 		if (ret)
1318 			goto out;
1319 	}
1320 
1321 	spin_lock(&fs_info->qgroup_lock);
1322 	del_qgroup_rb(fs_info, qgroupid);
1323 	spin_unlock(&fs_info->qgroup_lock);
1324 out:
1325 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1326 	return ret;
1327 }
1328 
1329 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1330 		       struct btrfs_fs_info *fs_info, u64 qgroupid,
1331 		       struct btrfs_qgroup_limit *limit)
1332 {
1333 	struct btrfs_root *quota_root;
1334 	struct btrfs_qgroup *qgroup;
1335 	int ret = 0;
1336 	/* Sometimes we would want to clear the limit on this qgroup.
1337 	 * To meet this requirement, we treat the -1 as a special value
1338 	 * which tell kernel to clear the limit on this qgroup.
1339 	 */
1340 	const u64 CLEAR_VALUE = -1;
1341 
1342 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1343 	quota_root = fs_info->quota_root;
1344 	if (!quota_root) {
1345 		ret = -EINVAL;
1346 		goto out;
1347 	}
1348 
1349 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1350 	if (!qgroup) {
1351 		ret = -ENOENT;
1352 		goto out;
1353 	}
1354 
1355 	spin_lock(&fs_info->qgroup_lock);
1356 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1357 		if (limit->max_rfer == CLEAR_VALUE) {
1358 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1359 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1360 			qgroup->max_rfer = 0;
1361 		} else {
1362 			qgroup->max_rfer = limit->max_rfer;
1363 		}
1364 	}
1365 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1366 		if (limit->max_excl == CLEAR_VALUE) {
1367 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1368 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1369 			qgroup->max_excl = 0;
1370 		} else {
1371 			qgroup->max_excl = limit->max_excl;
1372 		}
1373 	}
1374 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1375 		if (limit->rsv_rfer == CLEAR_VALUE) {
1376 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1377 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1378 			qgroup->rsv_rfer = 0;
1379 		} else {
1380 			qgroup->rsv_rfer = limit->rsv_rfer;
1381 		}
1382 	}
1383 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1384 		if (limit->rsv_excl == CLEAR_VALUE) {
1385 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1386 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1387 			qgroup->rsv_excl = 0;
1388 		} else {
1389 			qgroup->rsv_excl = limit->rsv_excl;
1390 		}
1391 	}
1392 	qgroup->lim_flags |= limit->flags;
1393 
1394 	spin_unlock(&fs_info->qgroup_lock);
1395 
1396 	ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1397 	if (ret) {
1398 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1399 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1400 		       qgroupid);
1401 	}
1402 
1403 out:
1404 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1405 	return ret;
1406 }
1407 
1408 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1409 				struct btrfs_delayed_ref_root *delayed_refs,
1410 				struct btrfs_qgroup_extent_record *record)
1411 {
1412 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1413 	struct rb_node *parent_node = NULL;
1414 	struct btrfs_qgroup_extent_record *entry;
1415 	u64 bytenr = record->bytenr;
1416 
1417 	assert_spin_locked(&delayed_refs->lock);
1418 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1419 
1420 	while (*p) {
1421 		parent_node = *p;
1422 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1423 				 node);
1424 		if (bytenr < entry->bytenr)
1425 			p = &(*p)->rb_left;
1426 		else if (bytenr > entry->bytenr)
1427 			p = &(*p)->rb_right;
1428 		else
1429 			return 1;
1430 	}
1431 
1432 	rb_link_node(&record->node, parent_node, p);
1433 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1434 	return 0;
1435 }
1436 
1437 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1438 				   struct btrfs_qgroup_extent_record *qrecord)
1439 {
1440 	struct ulist *old_root;
1441 	u64 bytenr = qrecord->bytenr;
1442 	int ret;
1443 
1444 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1445 	if (ret < 0)
1446 		return ret;
1447 
1448 	/*
1449 	 * Here we don't need to get the lock of
1450 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1451 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1452 	 *
1453 	 * So modifying qrecord->old_roots is safe here
1454 	 */
1455 	qrecord->old_roots = old_root;
1456 	return 0;
1457 }
1458 
1459 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1460 		struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1461 		gfp_t gfp_flag)
1462 {
1463 	struct btrfs_qgroup_extent_record *record;
1464 	struct btrfs_delayed_ref_root *delayed_refs;
1465 	int ret;
1466 
1467 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1468 	    || bytenr == 0 || num_bytes == 0)
1469 		return 0;
1470 	if (WARN_ON(trans == NULL))
1471 		return -EINVAL;
1472 	record = kmalloc(sizeof(*record), gfp_flag);
1473 	if (!record)
1474 		return -ENOMEM;
1475 
1476 	delayed_refs = &trans->transaction->delayed_refs;
1477 	record->bytenr = bytenr;
1478 	record->num_bytes = num_bytes;
1479 	record->old_roots = NULL;
1480 
1481 	spin_lock(&delayed_refs->lock);
1482 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1483 	spin_unlock(&delayed_refs->lock);
1484 	if (ret > 0) {
1485 		kfree(record);
1486 		return 0;
1487 	}
1488 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1489 }
1490 
1491 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1492 				  struct btrfs_fs_info *fs_info,
1493 				  struct extent_buffer *eb)
1494 {
1495 	int nr = btrfs_header_nritems(eb);
1496 	int i, extent_type, ret;
1497 	struct btrfs_key key;
1498 	struct btrfs_file_extent_item *fi;
1499 	u64 bytenr, num_bytes;
1500 
1501 	/* We can be called directly from walk_up_proc() */
1502 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1503 		return 0;
1504 
1505 	for (i = 0; i < nr; i++) {
1506 		btrfs_item_key_to_cpu(eb, &key, i);
1507 
1508 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1509 			continue;
1510 
1511 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1512 		/* filter out non qgroup-accountable extents  */
1513 		extent_type = btrfs_file_extent_type(eb, fi);
1514 
1515 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1516 			continue;
1517 
1518 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1519 		if (!bytenr)
1520 			continue;
1521 
1522 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1523 
1524 		ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1525 						num_bytes, GFP_NOFS);
1526 		if (ret)
1527 			return ret;
1528 	}
1529 	cond_resched();
1530 	return 0;
1531 }
1532 
1533 /*
1534  * Walk up the tree from the bottom, freeing leaves and any interior
1535  * nodes which have had all slots visited. If a node (leaf or
1536  * interior) is freed, the node above it will have it's slot
1537  * incremented. The root node will never be freed.
1538  *
1539  * At the end of this function, we should have a path which has all
1540  * slots incremented to the next position for a search. If we need to
1541  * read a new node it will be NULL and the node above it will have the
1542  * correct slot selected for a later read.
1543  *
1544  * If we increment the root nodes slot counter past the number of
1545  * elements, 1 is returned to signal completion of the search.
1546  */
1547 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1548 {
1549 	int level = 0;
1550 	int nr, slot;
1551 	struct extent_buffer *eb;
1552 
1553 	if (root_level == 0)
1554 		return 1;
1555 
1556 	while (level <= root_level) {
1557 		eb = path->nodes[level];
1558 		nr = btrfs_header_nritems(eb);
1559 		path->slots[level]++;
1560 		slot = path->slots[level];
1561 		if (slot >= nr || level == 0) {
1562 			/*
1563 			 * Don't free the root -  we will detect this
1564 			 * condition after our loop and return a
1565 			 * positive value for caller to stop walking the tree.
1566 			 */
1567 			if (level != root_level) {
1568 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1569 				path->locks[level] = 0;
1570 
1571 				free_extent_buffer(eb);
1572 				path->nodes[level] = NULL;
1573 				path->slots[level] = 0;
1574 			}
1575 		} else {
1576 			/*
1577 			 * We have a valid slot to walk back down
1578 			 * from. Stop here so caller can process these
1579 			 * new nodes.
1580 			 */
1581 			break;
1582 		}
1583 
1584 		level++;
1585 	}
1586 
1587 	eb = path->nodes[root_level];
1588 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1589 		return 1;
1590 
1591 	return 0;
1592 }
1593 
1594 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1595 			       struct btrfs_root *root,
1596 			       struct extent_buffer *root_eb,
1597 			       u64 root_gen, int root_level)
1598 {
1599 	struct btrfs_fs_info *fs_info = root->fs_info;
1600 	int ret = 0;
1601 	int level;
1602 	struct extent_buffer *eb = root_eb;
1603 	struct btrfs_path *path = NULL;
1604 
1605 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1606 	BUG_ON(root_eb == NULL);
1607 
1608 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1609 		return 0;
1610 
1611 	if (!extent_buffer_uptodate(root_eb)) {
1612 		ret = btrfs_read_buffer(root_eb, root_gen);
1613 		if (ret)
1614 			goto out;
1615 	}
1616 
1617 	if (root_level == 0) {
1618 		ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1619 		goto out;
1620 	}
1621 
1622 	path = btrfs_alloc_path();
1623 	if (!path)
1624 		return -ENOMEM;
1625 
1626 	/*
1627 	 * Walk down the tree.  Missing extent blocks are filled in as
1628 	 * we go. Metadata is accounted every time we read a new
1629 	 * extent block.
1630 	 *
1631 	 * When we reach a leaf, we account for file extent items in it,
1632 	 * walk back up the tree (adjusting slot pointers as we go)
1633 	 * and restart the search process.
1634 	 */
1635 	extent_buffer_get(root_eb); /* For path */
1636 	path->nodes[root_level] = root_eb;
1637 	path->slots[root_level] = 0;
1638 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1639 walk_down:
1640 	level = root_level;
1641 	while (level >= 0) {
1642 		if (path->nodes[level] == NULL) {
1643 			int parent_slot;
1644 			u64 child_gen;
1645 			u64 child_bytenr;
1646 
1647 			/*
1648 			 * We need to get child blockptr/gen from parent before
1649 			 * we can read it.
1650 			  */
1651 			eb = path->nodes[level + 1];
1652 			parent_slot = path->slots[level + 1];
1653 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1654 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1655 
1656 			eb = read_tree_block(fs_info, child_bytenr, child_gen);
1657 			if (IS_ERR(eb)) {
1658 				ret = PTR_ERR(eb);
1659 				goto out;
1660 			} else if (!extent_buffer_uptodate(eb)) {
1661 				free_extent_buffer(eb);
1662 				ret = -EIO;
1663 				goto out;
1664 			}
1665 
1666 			path->nodes[level] = eb;
1667 			path->slots[level] = 0;
1668 
1669 			btrfs_tree_read_lock(eb);
1670 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1671 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1672 
1673 			ret = btrfs_qgroup_trace_extent(trans, fs_info,
1674 							child_bytenr,
1675 							fs_info->nodesize,
1676 							GFP_NOFS);
1677 			if (ret)
1678 				goto out;
1679 		}
1680 
1681 		if (level == 0) {
1682 			ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1683 							   path->nodes[level]);
1684 			if (ret)
1685 				goto out;
1686 
1687 			/* Nonzero return here means we completed our search */
1688 			ret = adjust_slots_upwards(path, root_level);
1689 			if (ret)
1690 				break;
1691 
1692 			/* Restart search with new slots */
1693 			goto walk_down;
1694 		}
1695 
1696 		level--;
1697 	}
1698 
1699 	ret = 0;
1700 out:
1701 	btrfs_free_path(path);
1702 
1703 	return ret;
1704 }
1705 
1706 #define UPDATE_NEW	0
1707 #define UPDATE_OLD	1
1708 /*
1709  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1710  */
1711 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1712 				struct ulist *roots, struct ulist *tmp,
1713 				struct ulist *qgroups, u64 seq, int update_old)
1714 {
1715 	struct ulist_node *unode;
1716 	struct ulist_iterator uiter;
1717 	struct ulist_node *tmp_unode;
1718 	struct ulist_iterator tmp_uiter;
1719 	struct btrfs_qgroup *qg;
1720 	int ret = 0;
1721 
1722 	if (!roots)
1723 		return 0;
1724 	ULIST_ITER_INIT(&uiter);
1725 	while ((unode = ulist_next(roots, &uiter))) {
1726 		qg = find_qgroup_rb(fs_info, unode->val);
1727 		if (!qg)
1728 			continue;
1729 
1730 		ulist_reinit(tmp);
1731 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1732 				GFP_ATOMIC);
1733 		if (ret < 0)
1734 			return ret;
1735 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1736 		if (ret < 0)
1737 			return ret;
1738 		ULIST_ITER_INIT(&tmp_uiter);
1739 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1740 			struct btrfs_qgroup_list *glist;
1741 
1742 			qg = unode_aux_to_qgroup(tmp_unode);
1743 			if (update_old)
1744 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1745 			else
1746 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1747 			list_for_each_entry(glist, &qg->groups, next_group) {
1748 				ret = ulist_add(qgroups, glist->group->qgroupid,
1749 						qgroup_to_aux(glist->group),
1750 						GFP_ATOMIC);
1751 				if (ret < 0)
1752 					return ret;
1753 				ret = ulist_add(tmp, glist->group->qgroupid,
1754 						qgroup_to_aux(glist->group),
1755 						GFP_ATOMIC);
1756 				if (ret < 0)
1757 					return ret;
1758 			}
1759 		}
1760 	}
1761 	return 0;
1762 }
1763 
1764 /*
1765  * Update qgroup rfer/excl counters.
1766  * Rfer update is easy, codes can explain themselves.
1767  *
1768  * Excl update is tricky, the update is split into 2 part.
1769  * Part 1: Possible exclusive <-> sharing detect:
1770  *	|	A	|	!A	|
1771  *  -------------------------------------
1772  *  B	|	*	|	-	|
1773  *  -------------------------------------
1774  *  !B	|	+	|	**	|
1775  *  -------------------------------------
1776  *
1777  * Conditions:
1778  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
1779  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
1780  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
1781  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
1782  *
1783  * Results:
1784  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
1785  * *: Definitely not changed.		**: Possible unchanged.
1786  *
1787  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1788  *
1789  * To make the logic clear, we first use condition A and B to split
1790  * combination into 4 results.
1791  *
1792  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1793  * only on variant maybe 0.
1794  *
1795  * Lastly, check result **, since there are 2 variants maybe 0, split them
1796  * again(2x2).
1797  * But this time we don't need to consider other things, the codes and logic
1798  * is easy to understand now.
1799  */
1800 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1801 				  struct ulist *qgroups,
1802 				  u64 nr_old_roots,
1803 				  u64 nr_new_roots,
1804 				  u64 num_bytes, u64 seq)
1805 {
1806 	struct ulist_node *unode;
1807 	struct ulist_iterator uiter;
1808 	struct btrfs_qgroup *qg;
1809 	u64 cur_new_count, cur_old_count;
1810 
1811 	ULIST_ITER_INIT(&uiter);
1812 	while ((unode = ulist_next(qgroups, &uiter))) {
1813 		bool dirty = false;
1814 
1815 		qg = unode_aux_to_qgroup(unode);
1816 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1817 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1818 
1819 		trace_qgroup_update_counters(fs_info, qg->qgroupid,
1820 					     cur_old_count, cur_new_count);
1821 
1822 		/* Rfer update part */
1823 		if (cur_old_count == 0 && cur_new_count > 0) {
1824 			qg->rfer += num_bytes;
1825 			qg->rfer_cmpr += num_bytes;
1826 			dirty = true;
1827 		}
1828 		if (cur_old_count > 0 && cur_new_count == 0) {
1829 			qg->rfer -= num_bytes;
1830 			qg->rfer_cmpr -= num_bytes;
1831 			dirty = true;
1832 		}
1833 
1834 		/* Excl update part */
1835 		/* Exclusive/none -> shared case */
1836 		if (cur_old_count == nr_old_roots &&
1837 		    cur_new_count < nr_new_roots) {
1838 			/* Exclusive -> shared */
1839 			if (cur_old_count != 0) {
1840 				qg->excl -= num_bytes;
1841 				qg->excl_cmpr -= num_bytes;
1842 				dirty = true;
1843 			}
1844 		}
1845 
1846 		/* Shared -> exclusive/none case */
1847 		if (cur_old_count < nr_old_roots &&
1848 		    cur_new_count == nr_new_roots) {
1849 			/* Shared->exclusive */
1850 			if (cur_new_count != 0) {
1851 				qg->excl += num_bytes;
1852 				qg->excl_cmpr += num_bytes;
1853 				dirty = true;
1854 			}
1855 		}
1856 
1857 		/* Exclusive/none -> exclusive/none case */
1858 		if (cur_old_count == nr_old_roots &&
1859 		    cur_new_count == nr_new_roots) {
1860 			if (cur_old_count == 0) {
1861 				/* None -> exclusive/none */
1862 
1863 				if (cur_new_count != 0) {
1864 					/* None -> exclusive */
1865 					qg->excl += num_bytes;
1866 					qg->excl_cmpr += num_bytes;
1867 					dirty = true;
1868 				}
1869 				/* None -> none, nothing changed */
1870 			} else {
1871 				/* Exclusive -> exclusive/none */
1872 
1873 				if (cur_new_count == 0) {
1874 					/* Exclusive -> none */
1875 					qg->excl -= num_bytes;
1876 					qg->excl_cmpr -= num_bytes;
1877 					dirty = true;
1878 				}
1879 				/* Exclusive -> exclusive, nothing changed */
1880 			}
1881 		}
1882 
1883 		if (dirty)
1884 			qgroup_dirty(fs_info, qg);
1885 	}
1886 	return 0;
1887 }
1888 
1889 /*
1890  * Check if the @roots potentially is a list of fs tree roots
1891  *
1892  * Return 0 for definitely not a fs/subvol tree roots ulist
1893  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
1894  *          one as well)
1895  */
1896 static int maybe_fs_roots(struct ulist *roots)
1897 {
1898 	struct ulist_node *unode;
1899 	struct ulist_iterator uiter;
1900 
1901 	/* Empty one, still possible for fs roots */
1902 	if (!roots || roots->nnodes == 0)
1903 		return 1;
1904 
1905 	ULIST_ITER_INIT(&uiter);
1906 	unode = ulist_next(roots, &uiter);
1907 	if (!unode)
1908 		return 1;
1909 
1910 	/*
1911 	 * If it contains fs tree roots, then it must belong to fs/subvol
1912 	 * trees.
1913 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
1914 	 */
1915 	return is_fstree(unode->val);
1916 }
1917 
1918 int
1919 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1920 			    struct btrfs_fs_info *fs_info,
1921 			    u64 bytenr, u64 num_bytes,
1922 			    struct ulist *old_roots, struct ulist *new_roots)
1923 {
1924 	struct ulist *qgroups = NULL;
1925 	struct ulist *tmp = NULL;
1926 	u64 seq;
1927 	u64 nr_new_roots = 0;
1928 	u64 nr_old_roots = 0;
1929 	int ret = 0;
1930 
1931 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1932 		return 0;
1933 
1934 	if (new_roots) {
1935 		if (!maybe_fs_roots(new_roots))
1936 			goto out_free;
1937 		nr_new_roots = new_roots->nnodes;
1938 	}
1939 	if (old_roots) {
1940 		if (!maybe_fs_roots(old_roots))
1941 			goto out_free;
1942 		nr_old_roots = old_roots->nnodes;
1943 	}
1944 
1945 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
1946 	if (nr_old_roots == 0 && nr_new_roots == 0)
1947 		goto out_free;
1948 
1949 	BUG_ON(!fs_info->quota_root);
1950 
1951 	trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1952 					  nr_old_roots, nr_new_roots);
1953 
1954 	qgroups = ulist_alloc(GFP_NOFS);
1955 	if (!qgroups) {
1956 		ret = -ENOMEM;
1957 		goto out_free;
1958 	}
1959 	tmp = ulist_alloc(GFP_NOFS);
1960 	if (!tmp) {
1961 		ret = -ENOMEM;
1962 		goto out_free;
1963 	}
1964 
1965 	mutex_lock(&fs_info->qgroup_rescan_lock);
1966 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1967 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1968 			mutex_unlock(&fs_info->qgroup_rescan_lock);
1969 			ret = 0;
1970 			goto out_free;
1971 		}
1972 	}
1973 	mutex_unlock(&fs_info->qgroup_rescan_lock);
1974 
1975 	spin_lock(&fs_info->qgroup_lock);
1976 	seq = fs_info->qgroup_seq;
1977 
1978 	/* Update old refcnts using old_roots */
1979 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1980 				   UPDATE_OLD);
1981 	if (ret < 0)
1982 		goto out;
1983 
1984 	/* Update new refcnts using new_roots */
1985 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1986 				   UPDATE_NEW);
1987 	if (ret < 0)
1988 		goto out;
1989 
1990 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1991 			       num_bytes, seq);
1992 
1993 	/*
1994 	 * Bump qgroup_seq to avoid seq overlap
1995 	 */
1996 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1997 out:
1998 	spin_unlock(&fs_info->qgroup_lock);
1999 out_free:
2000 	ulist_free(tmp);
2001 	ulist_free(qgroups);
2002 	ulist_free(old_roots);
2003 	ulist_free(new_roots);
2004 	return ret;
2005 }
2006 
2007 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
2008 				 struct btrfs_fs_info *fs_info)
2009 {
2010 	struct btrfs_qgroup_extent_record *record;
2011 	struct btrfs_delayed_ref_root *delayed_refs;
2012 	struct ulist *new_roots = NULL;
2013 	struct rb_node *node;
2014 	u64 qgroup_to_skip;
2015 	int ret = 0;
2016 
2017 	delayed_refs = &trans->transaction->delayed_refs;
2018 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2019 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2020 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2021 				  node);
2022 
2023 		trace_btrfs_qgroup_account_extents(fs_info, record);
2024 
2025 		if (!ret) {
2026 			/*
2027 			 * Old roots should be searched when inserting qgroup
2028 			 * extent record
2029 			 */
2030 			if (WARN_ON(!record->old_roots)) {
2031 				/* Search commit root to find old_roots */
2032 				ret = btrfs_find_all_roots(NULL, fs_info,
2033 						record->bytenr, 0,
2034 						&record->old_roots, false);
2035 				if (ret < 0)
2036 					goto cleanup;
2037 			}
2038 
2039 			/*
2040 			 * Use SEQ_LAST as time_seq to do special search, which
2041 			 * doesn't lock tree or delayed_refs and search current
2042 			 * root. It's safe inside commit_transaction().
2043 			 */
2044 			ret = btrfs_find_all_roots(trans, fs_info,
2045 				record->bytenr, SEQ_LAST, &new_roots, false);
2046 			if (ret < 0)
2047 				goto cleanup;
2048 			if (qgroup_to_skip) {
2049 				ulist_del(new_roots, qgroup_to_skip, 0);
2050 				ulist_del(record->old_roots, qgroup_to_skip,
2051 					  0);
2052 			}
2053 			ret = btrfs_qgroup_account_extent(trans, fs_info,
2054 					record->bytenr, record->num_bytes,
2055 					record->old_roots, new_roots);
2056 			record->old_roots = NULL;
2057 			new_roots = NULL;
2058 		}
2059 cleanup:
2060 		ulist_free(record->old_roots);
2061 		ulist_free(new_roots);
2062 		new_roots = NULL;
2063 		rb_erase(node, &delayed_refs->dirty_extent_root);
2064 		kfree(record);
2065 
2066 	}
2067 	return ret;
2068 }
2069 
2070 /*
2071  * called from commit_transaction. Writes all changed qgroups to disk.
2072  */
2073 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2074 		      struct btrfs_fs_info *fs_info)
2075 {
2076 	struct btrfs_root *quota_root = fs_info->quota_root;
2077 	int ret = 0;
2078 	int start_rescan_worker = 0;
2079 
2080 	if (!quota_root)
2081 		goto out;
2082 
2083 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2084 	    test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2085 		start_rescan_worker = 1;
2086 
2087 	if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2088 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2089 
2090 	spin_lock(&fs_info->qgroup_lock);
2091 	while (!list_empty(&fs_info->dirty_qgroups)) {
2092 		struct btrfs_qgroup *qgroup;
2093 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2094 					  struct btrfs_qgroup, dirty);
2095 		list_del_init(&qgroup->dirty);
2096 		spin_unlock(&fs_info->qgroup_lock);
2097 		ret = update_qgroup_info_item(trans, quota_root, qgroup);
2098 		if (ret)
2099 			fs_info->qgroup_flags |=
2100 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2101 		ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2102 		if (ret)
2103 			fs_info->qgroup_flags |=
2104 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2105 		spin_lock(&fs_info->qgroup_lock);
2106 	}
2107 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2108 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2109 	else
2110 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2111 	spin_unlock(&fs_info->qgroup_lock);
2112 
2113 	ret = update_qgroup_status_item(trans, fs_info, quota_root);
2114 	if (ret)
2115 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2116 
2117 	if (!ret && start_rescan_worker) {
2118 		ret = qgroup_rescan_init(fs_info, 0, 1);
2119 		if (!ret) {
2120 			qgroup_rescan_zero_tracking(fs_info);
2121 			btrfs_queue_work(fs_info->qgroup_rescan_workers,
2122 					 &fs_info->qgroup_rescan_work);
2123 		}
2124 		ret = 0;
2125 	}
2126 
2127 out:
2128 
2129 	return ret;
2130 }
2131 
2132 /*
2133  * Copy the accounting information between qgroups. This is necessary
2134  * when a snapshot or a subvolume is created. Throwing an error will
2135  * cause a transaction abort so we take extra care here to only error
2136  * when a readonly fs is a reasonable outcome.
2137  */
2138 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2139 			 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2140 			 struct btrfs_qgroup_inherit *inherit)
2141 {
2142 	int ret = 0;
2143 	int i;
2144 	u64 *i_qgroups;
2145 	struct btrfs_root *quota_root = fs_info->quota_root;
2146 	struct btrfs_qgroup *srcgroup;
2147 	struct btrfs_qgroup *dstgroup;
2148 	u32 level_size = 0;
2149 	u64 nums;
2150 
2151 	mutex_lock(&fs_info->qgroup_ioctl_lock);
2152 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2153 		goto out;
2154 
2155 	if (!quota_root) {
2156 		ret = -EINVAL;
2157 		goto out;
2158 	}
2159 
2160 	if (inherit) {
2161 		i_qgroups = (u64 *)(inherit + 1);
2162 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2163 		       2 * inherit->num_excl_copies;
2164 		for (i = 0; i < nums; ++i) {
2165 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2166 
2167 			/*
2168 			 * Zero out invalid groups so we can ignore
2169 			 * them later.
2170 			 */
2171 			if (!srcgroup ||
2172 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2173 				*i_qgroups = 0ULL;
2174 
2175 			++i_qgroups;
2176 		}
2177 	}
2178 
2179 	/*
2180 	 * create a tracking group for the subvol itself
2181 	 */
2182 	ret = add_qgroup_item(trans, quota_root, objectid);
2183 	if (ret)
2184 		goto out;
2185 
2186 	if (srcid) {
2187 		struct btrfs_root *srcroot;
2188 		struct btrfs_key srckey;
2189 
2190 		srckey.objectid = srcid;
2191 		srckey.type = BTRFS_ROOT_ITEM_KEY;
2192 		srckey.offset = (u64)-1;
2193 		srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2194 		if (IS_ERR(srcroot)) {
2195 			ret = PTR_ERR(srcroot);
2196 			goto out;
2197 		}
2198 
2199 		level_size = fs_info->nodesize;
2200 	}
2201 
2202 	/*
2203 	 * add qgroup to all inherited groups
2204 	 */
2205 	if (inherit) {
2206 		i_qgroups = (u64 *)(inherit + 1);
2207 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2208 			if (*i_qgroups == 0)
2209 				continue;
2210 			ret = add_qgroup_relation_item(trans, quota_root,
2211 						       objectid, *i_qgroups);
2212 			if (ret && ret != -EEXIST)
2213 				goto out;
2214 			ret = add_qgroup_relation_item(trans, quota_root,
2215 						       *i_qgroups, objectid);
2216 			if (ret && ret != -EEXIST)
2217 				goto out;
2218 		}
2219 		ret = 0;
2220 	}
2221 
2222 
2223 	spin_lock(&fs_info->qgroup_lock);
2224 
2225 	dstgroup = add_qgroup_rb(fs_info, objectid);
2226 	if (IS_ERR(dstgroup)) {
2227 		ret = PTR_ERR(dstgroup);
2228 		goto unlock;
2229 	}
2230 
2231 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2232 		dstgroup->lim_flags = inherit->lim.flags;
2233 		dstgroup->max_rfer = inherit->lim.max_rfer;
2234 		dstgroup->max_excl = inherit->lim.max_excl;
2235 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2236 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2237 
2238 		ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2239 		if (ret) {
2240 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2241 			btrfs_info(fs_info,
2242 				   "unable to update quota limit for %llu",
2243 				   dstgroup->qgroupid);
2244 			goto unlock;
2245 		}
2246 	}
2247 
2248 	if (srcid) {
2249 		srcgroup = find_qgroup_rb(fs_info, srcid);
2250 		if (!srcgroup)
2251 			goto unlock;
2252 
2253 		/*
2254 		 * We call inherit after we clone the root in order to make sure
2255 		 * our counts don't go crazy, so at this point the only
2256 		 * difference between the two roots should be the root node.
2257 		 */
2258 		dstgroup->rfer = srcgroup->rfer;
2259 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2260 		dstgroup->excl = level_size;
2261 		dstgroup->excl_cmpr = level_size;
2262 		srcgroup->excl = level_size;
2263 		srcgroup->excl_cmpr = level_size;
2264 
2265 		/* inherit the limit info */
2266 		dstgroup->lim_flags = srcgroup->lim_flags;
2267 		dstgroup->max_rfer = srcgroup->max_rfer;
2268 		dstgroup->max_excl = srcgroup->max_excl;
2269 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2270 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2271 
2272 		qgroup_dirty(fs_info, dstgroup);
2273 		qgroup_dirty(fs_info, srcgroup);
2274 	}
2275 
2276 	if (!inherit)
2277 		goto unlock;
2278 
2279 	i_qgroups = (u64 *)(inherit + 1);
2280 	for (i = 0; i < inherit->num_qgroups; ++i) {
2281 		if (*i_qgroups) {
2282 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2283 			if (ret)
2284 				goto unlock;
2285 		}
2286 		++i_qgroups;
2287 	}
2288 
2289 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2290 		struct btrfs_qgroup *src;
2291 		struct btrfs_qgroup *dst;
2292 
2293 		if (!i_qgroups[0] || !i_qgroups[1])
2294 			continue;
2295 
2296 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2297 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2298 
2299 		if (!src || !dst) {
2300 			ret = -EINVAL;
2301 			goto unlock;
2302 		}
2303 
2304 		dst->rfer = src->rfer - level_size;
2305 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2306 	}
2307 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2308 		struct btrfs_qgroup *src;
2309 		struct btrfs_qgroup *dst;
2310 
2311 		if (!i_qgroups[0] || !i_qgroups[1])
2312 			continue;
2313 
2314 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2315 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2316 
2317 		if (!src || !dst) {
2318 			ret = -EINVAL;
2319 			goto unlock;
2320 		}
2321 
2322 		dst->excl = src->excl + level_size;
2323 		dst->excl_cmpr = src->excl_cmpr + level_size;
2324 	}
2325 
2326 unlock:
2327 	spin_unlock(&fs_info->qgroup_lock);
2328 out:
2329 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
2330 	return ret;
2331 }
2332 
2333 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2334 {
2335 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2336 	    qg->reserved + (s64)qg->rfer + num_bytes > qg->max_rfer)
2337 		return false;
2338 
2339 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2340 	    qg->reserved + (s64)qg->excl + num_bytes > qg->max_excl)
2341 		return false;
2342 
2343 	return true;
2344 }
2345 
2346 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce)
2347 {
2348 	struct btrfs_root *quota_root;
2349 	struct btrfs_qgroup *qgroup;
2350 	struct btrfs_fs_info *fs_info = root->fs_info;
2351 	u64 ref_root = root->root_key.objectid;
2352 	int ret = 0;
2353 	int retried = 0;
2354 	struct ulist_node *unode;
2355 	struct ulist_iterator uiter;
2356 
2357 	if (!is_fstree(ref_root))
2358 		return 0;
2359 
2360 	if (num_bytes == 0)
2361 		return 0;
2362 
2363 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2364 	    capable(CAP_SYS_RESOURCE))
2365 		enforce = false;
2366 
2367 retry:
2368 	spin_lock(&fs_info->qgroup_lock);
2369 	quota_root = fs_info->quota_root;
2370 	if (!quota_root)
2371 		goto out;
2372 
2373 	qgroup = find_qgroup_rb(fs_info, ref_root);
2374 	if (!qgroup)
2375 		goto out;
2376 
2377 	/*
2378 	 * in a first step, we check all affected qgroups if any limits would
2379 	 * be exceeded
2380 	 */
2381 	ulist_reinit(fs_info->qgroup_ulist);
2382 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2383 			(uintptr_t)qgroup, GFP_ATOMIC);
2384 	if (ret < 0)
2385 		goto out;
2386 	ULIST_ITER_INIT(&uiter);
2387 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2388 		struct btrfs_qgroup *qg;
2389 		struct btrfs_qgroup_list *glist;
2390 
2391 		qg = unode_aux_to_qgroup(unode);
2392 
2393 		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2394 			/*
2395 			 * Commit the tree and retry, since we may have
2396 			 * deletions which would free up space.
2397 			 */
2398 			if (!retried && qg->reserved > 0) {
2399 				struct btrfs_trans_handle *trans;
2400 
2401 				spin_unlock(&fs_info->qgroup_lock);
2402 				ret = btrfs_start_delalloc_inodes(root, 0);
2403 				if (ret)
2404 					return ret;
2405 				btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
2406 				trans = btrfs_join_transaction(root);
2407 				if (IS_ERR(trans))
2408 					return PTR_ERR(trans);
2409 				ret = btrfs_commit_transaction(trans);
2410 				if (ret)
2411 					return ret;
2412 				retried++;
2413 				goto retry;
2414 			}
2415 			ret = -EDQUOT;
2416 			goto out;
2417 		}
2418 
2419 		list_for_each_entry(glist, &qg->groups, next_group) {
2420 			ret = ulist_add(fs_info->qgroup_ulist,
2421 					glist->group->qgroupid,
2422 					(uintptr_t)glist->group, GFP_ATOMIC);
2423 			if (ret < 0)
2424 				goto out;
2425 		}
2426 	}
2427 	ret = 0;
2428 	/*
2429 	 * no limits exceeded, now record the reservation into all qgroups
2430 	 */
2431 	ULIST_ITER_INIT(&uiter);
2432 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2433 		struct btrfs_qgroup *qg;
2434 
2435 		qg = unode_aux_to_qgroup(unode);
2436 
2437 		trace_qgroup_update_reserve(fs_info, qg, num_bytes);
2438 		qg->reserved += num_bytes;
2439 	}
2440 
2441 out:
2442 	spin_unlock(&fs_info->qgroup_lock);
2443 	return ret;
2444 }
2445 
2446 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2447 			       u64 ref_root, u64 num_bytes)
2448 {
2449 	struct btrfs_root *quota_root;
2450 	struct btrfs_qgroup *qgroup;
2451 	struct ulist_node *unode;
2452 	struct ulist_iterator uiter;
2453 	int ret = 0;
2454 
2455 	if (!is_fstree(ref_root))
2456 		return;
2457 
2458 	if (num_bytes == 0)
2459 		return;
2460 
2461 	spin_lock(&fs_info->qgroup_lock);
2462 
2463 	quota_root = fs_info->quota_root;
2464 	if (!quota_root)
2465 		goto out;
2466 
2467 	qgroup = find_qgroup_rb(fs_info, ref_root);
2468 	if (!qgroup)
2469 		goto out;
2470 
2471 	ulist_reinit(fs_info->qgroup_ulist);
2472 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2473 			(uintptr_t)qgroup, GFP_ATOMIC);
2474 	if (ret < 0)
2475 		goto out;
2476 	ULIST_ITER_INIT(&uiter);
2477 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2478 		struct btrfs_qgroup *qg;
2479 		struct btrfs_qgroup_list *glist;
2480 
2481 		qg = unode_aux_to_qgroup(unode);
2482 
2483 		trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes);
2484 		if (qg->reserved < num_bytes)
2485 			report_reserved_underflow(fs_info, qg, num_bytes);
2486 		else
2487 			qg->reserved -= num_bytes;
2488 
2489 		list_for_each_entry(glist, &qg->groups, next_group) {
2490 			ret = ulist_add(fs_info->qgroup_ulist,
2491 					glist->group->qgroupid,
2492 					(uintptr_t)glist->group, GFP_ATOMIC);
2493 			if (ret < 0)
2494 				goto out;
2495 		}
2496 	}
2497 
2498 out:
2499 	spin_unlock(&fs_info->qgroup_lock);
2500 }
2501 
2502 /*
2503  * returns < 0 on error, 0 when more leafs are to be scanned.
2504  * returns 1 when done.
2505  */
2506 static int
2507 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2508 		   struct btrfs_trans_handle *trans)
2509 {
2510 	struct btrfs_key found;
2511 	struct extent_buffer *scratch_leaf = NULL;
2512 	struct ulist *roots = NULL;
2513 	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2514 	u64 num_bytes;
2515 	int slot;
2516 	int ret;
2517 
2518 	mutex_lock(&fs_info->qgroup_rescan_lock);
2519 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
2520 					 &fs_info->qgroup_rescan_progress,
2521 					 path, 1, 0);
2522 
2523 	btrfs_debug(fs_info,
2524 		"current progress key (%llu %u %llu), search_slot ret %d",
2525 		fs_info->qgroup_rescan_progress.objectid,
2526 		fs_info->qgroup_rescan_progress.type,
2527 		fs_info->qgroup_rescan_progress.offset, ret);
2528 
2529 	if (ret) {
2530 		/*
2531 		 * The rescan is about to end, we will not be scanning any
2532 		 * further blocks. We cannot unset the RESCAN flag here, because
2533 		 * we want to commit the transaction if everything went well.
2534 		 * To make the live accounting work in this phase, we set our
2535 		 * scan progress pointer such that every real extent objectid
2536 		 * will be smaller.
2537 		 */
2538 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2539 		btrfs_release_path(path);
2540 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2541 		return ret;
2542 	}
2543 
2544 	btrfs_item_key_to_cpu(path->nodes[0], &found,
2545 			      btrfs_header_nritems(path->nodes[0]) - 1);
2546 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2547 
2548 	btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2549 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2550 	if (!scratch_leaf) {
2551 		ret = -ENOMEM;
2552 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2553 		goto out;
2554 	}
2555 	extent_buffer_get(scratch_leaf);
2556 	btrfs_tree_read_lock(scratch_leaf);
2557 	btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2558 	slot = path->slots[0];
2559 	btrfs_release_path(path);
2560 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2561 
2562 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2563 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2564 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2565 		    found.type != BTRFS_METADATA_ITEM_KEY)
2566 			continue;
2567 		if (found.type == BTRFS_METADATA_ITEM_KEY)
2568 			num_bytes = fs_info->nodesize;
2569 		else
2570 			num_bytes = found.offset;
2571 
2572 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2573 					   &roots, false);
2574 		if (ret < 0)
2575 			goto out;
2576 		/* For rescan, just pass old_roots as NULL */
2577 		ret = btrfs_qgroup_account_extent(trans, fs_info,
2578 				found.objectid, num_bytes, NULL, roots);
2579 		if (ret < 0)
2580 			goto out;
2581 	}
2582 out:
2583 	if (scratch_leaf) {
2584 		btrfs_tree_read_unlock_blocking(scratch_leaf);
2585 		free_extent_buffer(scratch_leaf);
2586 	}
2587 	btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2588 
2589 	return ret;
2590 }
2591 
2592 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2593 {
2594 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2595 						     qgroup_rescan_work);
2596 	struct btrfs_path *path;
2597 	struct btrfs_trans_handle *trans = NULL;
2598 	int err = -ENOMEM;
2599 	int ret = 0;
2600 
2601 	path = btrfs_alloc_path();
2602 	if (!path)
2603 		goto out;
2604 
2605 	err = 0;
2606 	while (!err && !btrfs_fs_closing(fs_info)) {
2607 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
2608 		if (IS_ERR(trans)) {
2609 			err = PTR_ERR(trans);
2610 			break;
2611 		}
2612 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2613 			err = -EINTR;
2614 		} else {
2615 			err = qgroup_rescan_leaf(fs_info, path, trans);
2616 		}
2617 		if (err > 0)
2618 			btrfs_commit_transaction(trans);
2619 		else
2620 			btrfs_end_transaction(trans);
2621 	}
2622 
2623 out:
2624 	btrfs_free_path(path);
2625 
2626 	mutex_lock(&fs_info->qgroup_rescan_lock);
2627 	if (!btrfs_fs_closing(fs_info))
2628 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2629 
2630 	if (err > 0 &&
2631 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2632 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2633 	} else if (err < 0) {
2634 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2635 	}
2636 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2637 
2638 	/*
2639 	 * only update status, since the previous part has already updated the
2640 	 * qgroup info.
2641 	 */
2642 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
2643 	if (IS_ERR(trans)) {
2644 		err = PTR_ERR(trans);
2645 		btrfs_err(fs_info,
2646 			  "fail to start transaction for status update: %d",
2647 			  err);
2648 		goto done;
2649 	}
2650 	ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2651 	if (ret < 0) {
2652 		err = ret;
2653 		btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2654 	}
2655 	btrfs_end_transaction(trans);
2656 
2657 	if (btrfs_fs_closing(fs_info)) {
2658 		btrfs_info(fs_info, "qgroup scan paused");
2659 	} else if (err >= 0) {
2660 		btrfs_info(fs_info, "qgroup scan completed%s",
2661 			err > 0 ? " (inconsistency flag cleared)" : "");
2662 	} else {
2663 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
2664 	}
2665 
2666 done:
2667 	mutex_lock(&fs_info->qgroup_rescan_lock);
2668 	fs_info->qgroup_rescan_running = false;
2669 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2670 	complete_all(&fs_info->qgroup_rescan_completion);
2671 }
2672 
2673 /*
2674  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2675  * memory required for the rescan context.
2676  */
2677 static int
2678 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2679 		   int init_flags)
2680 {
2681 	int ret = 0;
2682 
2683 	if (!init_flags &&
2684 	    (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2685 	     !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2686 		ret = -EINVAL;
2687 		goto err;
2688 	}
2689 
2690 	mutex_lock(&fs_info->qgroup_rescan_lock);
2691 	spin_lock(&fs_info->qgroup_lock);
2692 
2693 	if (init_flags) {
2694 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2695 			ret = -EINPROGRESS;
2696 		else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2697 			ret = -EINVAL;
2698 
2699 		if (ret) {
2700 			spin_unlock(&fs_info->qgroup_lock);
2701 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2702 			goto err;
2703 		}
2704 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2705 	}
2706 
2707 	memset(&fs_info->qgroup_rescan_progress, 0,
2708 		sizeof(fs_info->qgroup_rescan_progress));
2709 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2710 	init_completion(&fs_info->qgroup_rescan_completion);
2711 	fs_info->qgroup_rescan_running = true;
2712 
2713 	spin_unlock(&fs_info->qgroup_lock);
2714 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2715 
2716 	memset(&fs_info->qgroup_rescan_work, 0,
2717 	       sizeof(fs_info->qgroup_rescan_work));
2718 	btrfs_init_work(&fs_info->qgroup_rescan_work,
2719 			btrfs_qgroup_rescan_helper,
2720 			btrfs_qgroup_rescan_worker, NULL, NULL);
2721 
2722 	if (ret) {
2723 err:
2724 		btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2725 		return ret;
2726 	}
2727 
2728 	return 0;
2729 }
2730 
2731 static void
2732 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2733 {
2734 	struct rb_node *n;
2735 	struct btrfs_qgroup *qgroup;
2736 
2737 	spin_lock(&fs_info->qgroup_lock);
2738 	/* clear all current qgroup tracking information */
2739 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2740 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
2741 		qgroup->rfer = 0;
2742 		qgroup->rfer_cmpr = 0;
2743 		qgroup->excl = 0;
2744 		qgroup->excl_cmpr = 0;
2745 	}
2746 	spin_unlock(&fs_info->qgroup_lock);
2747 }
2748 
2749 int
2750 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2751 {
2752 	int ret = 0;
2753 	struct btrfs_trans_handle *trans;
2754 
2755 	ret = qgroup_rescan_init(fs_info, 0, 1);
2756 	if (ret)
2757 		return ret;
2758 
2759 	/*
2760 	 * We have set the rescan_progress to 0, which means no more
2761 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2762 	 * However, btrfs_qgroup_account_ref may be right after its call
2763 	 * to btrfs_find_all_roots, in which case it would still do the
2764 	 * accounting.
2765 	 * To solve this, we're committing the transaction, which will
2766 	 * ensure we run all delayed refs and only after that, we are
2767 	 * going to clear all tracking information for a clean start.
2768 	 */
2769 
2770 	trans = btrfs_join_transaction(fs_info->fs_root);
2771 	if (IS_ERR(trans)) {
2772 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2773 		return PTR_ERR(trans);
2774 	}
2775 	ret = btrfs_commit_transaction(trans);
2776 	if (ret) {
2777 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2778 		return ret;
2779 	}
2780 
2781 	qgroup_rescan_zero_tracking(fs_info);
2782 
2783 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
2784 			 &fs_info->qgroup_rescan_work);
2785 
2786 	return 0;
2787 }
2788 
2789 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2790 				     bool interruptible)
2791 {
2792 	int running;
2793 	int ret = 0;
2794 
2795 	mutex_lock(&fs_info->qgroup_rescan_lock);
2796 	spin_lock(&fs_info->qgroup_lock);
2797 	running = fs_info->qgroup_rescan_running;
2798 	spin_unlock(&fs_info->qgroup_lock);
2799 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2800 
2801 	if (!running)
2802 		return 0;
2803 
2804 	if (interruptible)
2805 		ret = wait_for_completion_interruptible(
2806 					&fs_info->qgroup_rescan_completion);
2807 	else
2808 		wait_for_completion(&fs_info->qgroup_rescan_completion);
2809 
2810 	return ret;
2811 }
2812 
2813 /*
2814  * this is only called from open_ctree where we're still single threaded, thus
2815  * locking is omitted here.
2816  */
2817 void
2818 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2819 {
2820 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2821 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
2822 				 &fs_info->qgroup_rescan_work);
2823 }
2824 
2825 /*
2826  * Reserve qgroup space for range [start, start + len).
2827  *
2828  * This function will either reserve space from related qgroups or doing
2829  * nothing if the range is already reserved.
2830  *
2831  * Return 0 for successful reserve
2832  * Return <0 for error (including -EQUOT)
2833  *
2834  * NOTE: this function may sleep for memory allocation.
2835  *       if btrfs_qgroup_reserve_data() is called multiple times with
2836  *       same @reserved, caller must ensure when error happens it's OK
2837  *       to free *ALL* reserved space.
2838  */
2839 int btrfs_qgroup_reserve_data(struct inode *inode,
2840 			struct extent_changeset **reserved_ret, u64 start,
2841 			u64 len)
2842 {
2843 	struct btrfs_root *root = BTRFS_I(inode)->root;
2844 	struct ulist_node *unode;
2845 	struct ulist_iterator uiter;
2846 	struct extent_changeset *reserved;
2847 	u64 orig_reserved;
2848 	u64 to_reserve;
2849 	int ret;
2850 
2851 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2852 	    !is_fstree(root->objectid) || len == 0)
2853 		return 0;
2854 
2855 	/* @reserved parameter is mandatory for qgroup */
2856 	if (WARN_ON(!reserved_ret))
2857 		return -EINVAL;
2858 	if (!*reserved_ret) {
2859 		*reserved_ret = extent_changeset_alloc();
2860 		if (!*reserved_ret)
2861 			return -ENOMEM;
2862 	}
2863 	reserved = *reserved_ret;
2864 	/* Record already reserved space */
2865 	orig_reserved = reserved->bytes_changed;
2866 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2867 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
2868 
2869 	/* Newly reserved space */
2870 	to_reserve = reserved->bytes_changed - orig_reserved;
2871 	trace_btrfs_qgroup_reserve_data(inode, start, len,
2872 					to_reserve, QGROUP_RESERVE);
2873 	if (ret < 0)
2874 		goto cleanup;
2875 	ret = qgroup_reserve(root, to_reserve, true);
2876 	if (ret < 0)
2877 		goto cleanup;
2878 
2879 	return ret;
2880 
2881 cleanup:
2882 	/* cleanup *ALL* already reserved ranges */
2883 	ULIST_ITER_INIT(&uiter);
2884 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
2885 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2886 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2887 				 GFP_NOFS);
2888 	extent_changeset_release(reserved);
2889 	return ret;
2890 }
2891 
2892 /* Free ranges specified by @reserved, normally in error path */
2893 static int qgroup_free_reserved_data(struct inode *inode,
2894 			struct extent_changeset *reserved, u64 start, u64 len)
2895 {
2896 	struct btrfs_root *root = BTRFS_I(inode)->root;
2897 	struct ulist_node *unode;
2898 	struct ulist_iterator uiter;
2899 	struct extent_changeset changeset;
2900 	int freed = 0;
2901 	int ret;
2902 
2903 	extent_changeset_init(&changeset);
2904 	len = round_up(start + len, root->fs_info->sectorsize);
2905 	start = round_down(start, root->fs_info->sectorsize);
2906 
2907 	ULIST_ITER_INIT(&uiter);
2908 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
2909 		u64 range_start = unode->val;
2910 		/* unode->aux is the inclusive end */
2911 		u64 range_len = unode->aux - range_start + 1;
2912 		u64 free_start;
2913 		u64 free_len;
2914 
2915 		extent_changeset_release(&changeset);
2916 
2917 		/* Only free range in range [start, start + len) */
2918 		if (range_start >= start + len ||
2919 		    range_start + range_len <= start)
2920 			continue;
2921 		free_start = max(range_start, start);
2922 		free_len = min(start + len, range_start + range_len) -
2923 			   free_start;
2924 		/*
2925 		 * TODO: To also modify reserved->ranges_reserved to reflect
2926 		 * the modification.
2927 		 *
2928 		 * However as long as we free qgroup reserved according to
2929 		 * EXTENT_QGROUP_RESERVED, we won't double free.
2930 		 * So not need to rush.
2931 		 */
2932 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
2933 				free_start, free_start + free_len - 1,
2934 				EXTENT_QGROUP_RESERVED, &changeset);
2935 		if (ret < 0)
2936 			goto out;
2937 		freed += changeset.bytes_changed;
2938 	}
2939 	btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed);
2940 	ret = freed;
2941 out:
2942 	extent_changeset_release(&changeset);
2943 	return ret;
2944 }
2945 
2946 static int __btrfs_qgroup_release_data(struct inode *inode,
2947 			struct extent_changeset *reserved, u64 start, u64 len,
2948 			int free)
2949 {
2950 	struct extent_changeset changeset;
2951 	int trace_op = QGROUP_RELEASE;
2952 	int ret;
2953 
2954 	/* In release case, we shouldn't have @reserved */
2955 	WARN_ON(!free && reserved);
2956 	if (free && reserved)
2957 		return qgroup_free_reserved_data(inode, reserved, start, len);
2958 	extent_changeset_init(&changeset);
2959 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2960 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2961 	if (ret < 0)
2962 		goto out;
2963 
2964 	if (free)
2965 		trace_op = QGROUP_FREE;
2966 	trace_btrfs_qgroup_release_data(inode, start, len,
2967 					changeset.bytes_changed, trace_op);
2968 	if (free)
2969 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
2970 				BTRFS_I(inode)->root->objectid,
2971 				changeset.bytes_changed);
2972 	ret = changeset.bytes_changed;
2973 out:
2974 	extent_changeset_release(&changeset);
2975 	return ret;
2976 }
2977 
2978 /*
2979  * Free a reserved space range from io_tree and related qgroups
2980  *
2981  * Should be called when a range of pages get invalidated before reaching disk.
2982  * Or for error cleanup case.
2983  * if @reserved is given, only reserved range in [@start, @start + @len) will
2984  * be freed.
2985  *
2986  * For data written to disk, use btrfs_qgroup_release_data().
2987  *
2988  * NOTE: This function may sleep for memory allocation.
2989  */
2990 int btrfs_qgroup_free_data(struct inode *inode,
2991 			struct extent_changeset *reserved, u64 start, u64 len)
2992 {
2993 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
2994 }
2995 
2996 /*
2997  * Release a reserved space range from io_tree only.
2998  *
2999  * Should be called when a range of pages get written to disk and corresponding
3000  * FILE_EXTENT is inserted into corresponding root.
3001  *
3002  * Since new qgroup accounting framework will only update qgroup numbers at
3003  * commit_transaction() time, its reserved space shouldn't be freed from
3004  * related qgroups.
3005  *
3006  * But we should release the range from io_tree, to allow further write to be
3007  * COWed.
3008  *
3009  * NOTE: This function may sleep for memory allocation.
3010  */
3011 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3012 {
3013 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3014 }
3015 
3016 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3017 			      bool enforce)
3018 {
3019 	struct btrfs_fs_info *fs_info = root->fs_info;
3020 	int ret;
3021 
3022 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3023 	    !is_fstree(root->objectid) || num_bytes == 0)
3024 		return 0;
3025 
3026 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3027 	trace_qgroup_meta_reserve(root, (s64)num_bytes);
3028 	ret = qgroup_reserve(root, num_bytes, enforce);
3029 	if (ret < 0)
3030 		return ret;
3031 	atomic64_add(num_bytes, &root->qgroup_meta_rsv);
3032 	return ret;
3033 }
3034 
3035 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
3036 {
3037 	struct btrfs_fs_info *fs_info = root->fs_info;
3038 	u64 reserved;
3039 
3040 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3041 	    !is_fstree(root->objectid))
3042 		return;
3043 
3044 	reserved = atomic64_xchg(&root->qgroup_meta_rsv, 0);
3045 	if (reserved == 0)
3046 		return;
3047 	trace_qgroup_meta_reserve(root, -(s64)reserved);
3048 	btrfs_qgroup_free_refroot(fs_info, root->objectid, reserved);
3049 }
3050 
3051 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
3052 {
3053 	struct btrfs_fs_info *fs_info = root->fs_info;
3054 
3055 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3056 	    !is_fstree(root->objectid))
3057 		return;
3058 
3059 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3060 	WARN_ON(atomic64_read(&root->qgroup_meta_rsv) < num_bytes);
3061 	atomic64_sub(num_bytes, &root->qgroup_meta_rsv);
3062 	trace_qgroup_meta_reserve(root, -(s64)num_bytes);
3063 	btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes);
3064 }
3065 
3066 /*
3067  * Check qgroup reserved space leaking, normally at destroy inode
3068  * time
3069  */
3070 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3071 {
3072 	struct extent_changeset changeset;
3073 	struct ulist_node *unode;
3074 	struct ulist_iterator iter;
3075 	int ret;
3076 
3077 	extent_changeset_init(&changeset);
3078 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3079 			EXTENT_QGROUP_RESERVED, &changeset);
3080 
3081 	WARN_ON(ret < 0);
3082 	if (WARN_ON(changeset.bytes_changed)) {
3083 		ULIST_ITER_INIT(&iter);
3084 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3085 			btrfs_warn(BTRFS_I(inode)->root->fs_info,
3086 				"leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3087 				inode->i_ino, unode->val, unode->aux);
3088 		}
3089 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3090 				BTRFS_I(inode)->root->objectid,
3091 				changeset.bytes_changed);
3092 
3093 	}
3094 	extent_changeset_release(&changeset);
3095 }
3096