xref: /openbmc/linux/fs/btrfs/qgroup.c (revision e3b9f1e8)
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 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1447 		btrfs_warn(fs_info,
1448 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1449 			ret);
1450 		return 0;
1451 	}
1452 
1453 	/*
1454 	 * Here we don't need to get the lock of
1455 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1456 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1457 	 *
1458 	 * So modifying qrecord->old_roots is safe here
1459 	 */
1460 	qrecord->old_roots = old_root;
1461 	return 0;
1462 }
1463 
1464 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1465 		struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1466 		gfp_t gfp_flag)
1467 {
1468 	struct btrfs_qgroup_extent_record *record;
1469 	struct btrfs_delayed_ref_root *delayed_refs;
1470 	int ret;
1471 
1472 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1473 	    || bytenr == 0 || num_bytes == 0)
1474 		return 0;
1475 	if (WARN_ON(trans == NULL))
1476 		return -EINVAL;
1477 	record = kmalloc(sizeof(*record), gfp_flag);
1478 	if (!record)
1479 		return -ENOMEM;
1480 
1481 	delayed_refs = &trans->transaction->delayed_refs;
1482 	record->bytenr = bytenr;
1483 	record->num_bytes = num_bytes;
1484 	record->old_roots = NULL;
1485 
1486 	spin_lock(&delayed_refs->lock);
1487 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1488 	spin_unlock(&delayed_refs->lock);
1489 	if (ret > 0) {
1490 		kfree(record);
1491 		return 0;
1492 	}
1493 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1494 }
1495 
1496 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1497 				  struct btrfs_fs_info *fs_info,
1498 				  struct extent_buffer *eb)
1499 {
1500 	int nr = btrfs_header_nritems(eb);
1501 	int i, extent_type, ret;
1502 	struct btrfs_key key;
1503 	struct btrfs_file_extent_item *fi;
1504 	u64 bytenr, num_bytes;
1505 
1506 	/* We can be called directly from walk_up_proc() */
1507 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1508 		return 0;
1509 
1510 	for (i = 0; i < nr; i++) {
1511 		btrfs_item_key_to_cpu(eb, &key, i);
1512 
1513 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1514 			continue;
1515 
1516 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1517 		/* filter out non qgroup-accountable extents  */
1518 		extent_type = btrfs_file_extent_type(eb, fi);
1519 
1520 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1521 			continue;
1522 
1523 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1524 		if (!bytenr)
1525 			continue;
1526 
1527 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1528 
1529 		ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1530 						num_bytes, GFP_NOFS);
1531 		if (ret)
1532 			return ret;
1533 	}
1534 	cond_resched();
1535 	return 0;
1536 }
1537 
1538 /*
1539  * Walk up the tree from the bottom, freeing leaves and any interior
1540  * nodes which have had all slots visited. If a node (leaf or
1541  * interior) is freed, the node above it will have it's slot
1542  * incremented. The root node will never be freed.
1543  *
1544  * At the end of this function, we should have a path which has all
1545  * slots incremented to the next position for a search. If we need to
1546  * read a new node it will be NULL and the node above it will have the
1547  * correct slot selected for a later read.
1548  *
1549  * If we increment the root nodes slot counter past the number of
1550  * elements, 1 is returned to signal completion of the search.
1551  */
1552 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1553 {
1554 	int level = 0;
1555 	int nr, slot;
1556 	struct extent_buffer *eb;
1557 
1558 	if (root_level == 0)
1559 		return 1;
1560 
1561 	while (level <= root_level) {
1562 		eb = path->nodes[level];
1563 		nr = btrfs_header_nritems(eb);
1564 		path->slots[level]++;
1565 		slot = path->slots[level];
1566 		if (slot >= nr || level == 0) {
1567 			/*
1568 			 * Don't free the root -  we will detect this
1569 			 * condition after our loop and return a
1570 			 * positive value for caller to stop walking the tree.
1571 			 */
1572 			if (level != root_level) {
1573 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1574 				path->locks[level] = 0;
1575 
1576 				free_extent_buffer(eb);
1577 				path->nodes[level] = NULL;
1578 				path->slots[level] = 0;
1579 			}
1580 		} else {
1581 			/*
1582 			 * We have a valid slot to walk back down
1583 			 * from. Stop here so caller can process these
1584 			 * new nodes.
1585 			 */
1586 			break;
1587 		}
1588 
1589 		level++;
1590 	}
1591 
1592 	eb = path->nodes[root_level];
1593 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1594 		return 1;
1595 
1596 	return 0;
1597 }
1598 
1599 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1600 			       struct btrfs_root *root,
1601 			       struct extent_buffer *root_eb,
1602 			       u64 root_gen, int root_level)
1603 {
1604 	struct btrfs_fs_info *fs_info = root->fs_info;
1605 	int ret = 0;
1606 	int level;
1607 	struct extent_buffer *eb = root_eb;
1608 	struct btrfs_path *path = NULL;
1609 
1610 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1611 	BUG_ON(root_eb == NULL);
1612 
1613 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1614 		return 0;
1615 
1616 	if (!extent_buffer_uptodate(root_eb)) {
1617 		ret = btrfs_read_buffer(root_eb, root_gen);
1618 		if (ret)
1619 			goto out;
1620 	}
1621 
1622 	if (root_level == 0) {
1623 		ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1624 		goto out;
1625 	}
1626 
1627 	path = btrfs_alloc_path();
1628 	if (!path)
1629 		return -ENOMEM;
1630 
1631 	/*
1632 	 * Walk down the tree.  Missing extent blocks are filled in as
1633 	 * we go. Metadata is accounted every time we read a new
1634 	 * extent block.
1635 	 *
1636 	 * When we reach a leaf, we account for file extent items in it,
1637 	 * walk back up the tree (adjusting slot pointers as we go)
1638 	 * and restart the search process.
1639 	 */
1640 	extent_buffer_get(root_eb); /* For path */
1641 	path->nodes[root_level] = root_eb;
1642 	path->slots[root_level] = 0;
1643 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1644 walk_down:
1645 	level = root_level;
1646 	while (level >= 0) {
1647 		if (path->nodes[level] == NULL) {
1648 			int parent_slot;
1649 			u64 child_gen;
1650 			u64 child_bytenr;
1651 
1652 			/*
1653 			 * We need to get child blockptr/gen from parent before
1654 			 * we can read it.
1655 			  */
1656 			eb = path->nodes[level + 1];
1657 			parent_slot = path->slots[level + 1];
1658 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1659 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1660 
1661 			eb = read_tree_block(fs_info, child_bytenr, child_gen);
1662 			if (IS_ERR(eb)) {
1663 				ret = PTR_ERR(eb);
1664 				goto out;
1665 			} else if (!extent_buffer_uptodate(eb)) {
1666 				free_extent_buffer(eb);
1667 				ret = -EIO;
1668 				goto out;
1669 			}
1670 
1671 			path->nodes[level] = eb;
1672 			path->slots[level] = 0;
1673 
1674 			btrfs_tree_read_lock(eb);
1675 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1676 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1677 
1678 			ret = btrfs_qgroup_trace_extent(trans, fs_info,
1679 							child_bytenr,
1680 							fs_info->nodesize,
1681 							GFP_NOFS);
1682 			if (ret)
1683 				goto out;
1684 		}
1685 
1686 		if (level == 0) {
1687 			ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1688 							   path->nodes[level]);
1689 			if (ret)
1690 				goto out;
1691 
1692 			/* Nonzero return here means we completed our search */
1693 			ret = adjust_slots_upwards(path, root_level);
1694 			if (ret)
1695 				break;
1696 
1697 			/* Restart search with new slots */
1698 			goto walk_down;
1699 		}
1700 
1701 		level--;
1702 	}
1703 
1704 	ret = 0;
1705 out:
1706 	btrfs_free_path(path);
1707 
1708 	return ret;
1709 }
1710 
1711 #define UPDATE_NEW	0
1712 #define UPDATE_OLD	1
1713 /*
1714  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1715  */
1716 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1717 				struct ulist *roots, struct ulist *tmp,
1718 				struct ulist *qgroups, u64 seq, int update_old)
1719 {
1720 	struct ulist_node *unode;
1721 	struct ulist_iterator uiter;
1722 	struct ulist_node *tmp_unode;
1723 	struct ulist_iterator tmp_uiter;
1724 	struct btrfs_qgroup *qg;
1725 	int ret = 0;
1726 
1727 	if (!roots)
1728 		return 0;
1729 	ULIST_ITER_INIT(&uiter);
1730 	while ((unode = ulist_next(roots, &uiter))) {
1731 		qg = find_qgroup_rb(fs_info, unode->val);
1732 		if (!qg)
1733 			continue;
1734 
1735 		ulist_reinit(tmp);
1736 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1737 				GFP_ATOMIC);
1738 		if (ret < 0)
1739 			return ret;
1740 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1741 		if (ret < 0)
1742 			return ret;
1743 		ULIST_ITER_INIT(&tmp_uiter);
1744 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1745 			struct btrfs_qgroup_list *glist;
1746 
1747 			qg = unode_aux_to_qgroup(tmp_unode);
1748 			if (update_old)
1749 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1750 			else
1751 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1752 			list_for_each_entry(glist, &qg->groups, next_group) {
1753 				ret = ulist_add(qgroups, glist->group->qgroupid,
1754 						qgroup_to_aux(glist->group),
1755 						GFP_ATOMIC);
1756 				if (ret < 0)
1757 					return ret;
1758 				ret = ulist_add(tmp, glist->group->qgroupid,
1759 						qgroup_to_aux(glist->group),
1760 						GFP_ATOMIC);
1761 				if (ret < 0)
1762 					return ret;
1763 			}
1764 		}
1765 	}
1766 	return 0;
1767 }
1768 
1769 /*
1770  * Update qgroup rfer/excl counters.
1771  * Rfer update is easy, codes can explain themselves.
1772  *
1773  * Excl update is tricky, the update is split into 2 part.
1774  * Part 1: Possible exclusive <-> sharing detect:
1775  *	|	A	|	!A	|
1776  *  -------------------------------------
1777  *  B	|	*	|	-	|
1778  *  -------------------------------------
1779  *  !B	|	+	|	**	|
1780  *  -------------------------------------
1781  *
1782  * Conditions:
1783  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
1784  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
1785  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
1786  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
1787  *
1788  * Results:
1789  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
1790  * *: Definitely not changed.		**: Possible unchanged.
1791  *
1792  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1793  *
1794  * To make the logic clear, we first use condition A and B to split
1795  * combination into 4 results.
1796  *
1797  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1798  * only on variant maybe 0.
1799  *
1800  * Lastly, check result **, since there are 2 variants maybe 0, split them
1801  * again(2x2).
1802  * But this time we don't need to consider other things, the codes and logic
1803  * is easy to understand now.
1804  */
1805 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1806 				  struct ulist *qgroups,
1807 				  u64 nr_old_roots,
1808 				  u64 nr_new_roots,
1809 				  u64 num_bytes, u64 seq)
1810 {
1811 	struct ulist_node *unode;
1812 	struct ulist_iterator uiter;
1813 	struct btrfs_qgroup *qg;
1814 	u64 cur_new_count, cur_old_count;
1815 
1816 	ULIST_ITER_INIT(&uiter);
1817 	while ((unode = ulist_next(qgroups, &uiter))) {
1818 		bool dirty = false;
1819 
1820 		qg = unode_aux_to_qgroup(unode);
1821 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1822 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1823 
1824 		trace_qgroup_update_counters(fs_info, qg->qgroupid,
1825 					     cur_old_count, cur_new_count);
1826 
1827 		/* Rfer update part */
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 		if (cur_old_count > 0 && cur_new_count == 0) {
1834 			qg->rfer -= num_bytes;
1835 			qg->rfer_cmpr -= num_bytes;
1836 			dirty = true;
1837 		}
1838 
1839 		/* Excl update part */
1840 		/* Exclusive/none -> shared case */
1841 		if (cur_old_count == nr_old_roots &&
1842 		    cur_new_count < nr_new_roots) {
1843 			/* Exclusive -> shared */
1844 			if (cur_old_count != 0) {
1845 				qg->excl -= num_bytes;
1846 				qg->excl_cmpr -= num_bytes;
1847 				dirty = true;
1848 			}
1849 		}
1850 
1851 		/* Shared -> exclusive/none case */
1852 		if (cur_old_count < nr_old_roots &&
1853 		    cur_new_count == nr_new_roots) {
1854 			/* Shared->exclusive */
1855 			if (cur_new_count != 0) {
1856 				qg->excl += num_bytes;
1857 				qg->excl_cmpr += num_bytes;
1858 				dirty = true;
1859 			}
1860 		}
1861 
1862 		/* Exclusive/none -> exclusive/none case */
1863 		if (cur_old_count == nr_old_roots &&
1864 		    cur_new_count == nr_new_roots) {
1865 			if (cur_old_count == 0) {
1866 				/* None -> exclusive/none */
1867 
1868 				if (cur_new_count != 0) {
1869 					/* None -> exclusive */
1870 					qg->excl += num_bytes;
1871 					qg->excl_cmpr += num_bytes;
1872 					dirty = true;
1873 				}
1874 				/* None -> none, nothing changed */
1875 			} else {
1876 				/* Exclusive -> exclusive/none */
1877 
1878 				if (cur_new_count == 0) {
1879 					/* Exclusive -> none */
1880 					qg->excl -= num_bytes;
1881 					qg->excl_cmpr -= num_bytes;
1882 					dirty = true;
1883 				}
1884 				/* Exclusive -> exclusive, nothing changed */
1885 			}
1886 		}
1887 
1888 		if (dirty)
1889 			qgroup_dirty(fs_info, qg);
1890 	}
1891 	return 0;
1892 }
1893 
1894 /*
1895  * Check if the @roots potentially is a list of fs tree roots
1896  *
1897  * Return 0 for definitely not a fs/subvol tree roots ulist
1898  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
1899  *          one as well)
1900  */
1901 static int maybe_fs_roots(struct ulist *roots)
1902 {
1903 	struct ulist_node *unode;
1904 	struct ulist_iterator uiter;
1905 
1906 	/* Empty one, still possible for fs roots */
1907 	if (!roots || roots->nnodes == 0)
1908 		return 1;
1909 
1910 	ULIST_ITER_INIT(&uiter);
1911 	unode = ulist_next(roots, &uiter);
1912 	if (!unode)
1913 		return 1;
1914 
1915 	/*
1916 	 * If it contains fs tree roots, then it must belong to fs/subvol
1917 	 * trees.
1918 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
1919 	 */
1920 	return is_fstree(unode->val);
1921 }
1922 
1923 int
1924 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1925 			    struct btrfs_fs_info *fs_info,
1926 			    u64 bytenr, u64 num_bytes,
1927 			    struct ulist *old_roots, struct ulist *new_roots)
1928 {
1929 	struct ulist *qgroups = NULL;
1930 	struct ulist *tmp = NULL;
1931 	u64 seq;
1932 	u64 nr_new_roots = 0;
1933 	u64 nr_old_roots = 0;
1934 	int ret = 0;
1935 
1936 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1937 		return 0;
1938 
1939 	if (new_roots) {
1940 		if (!maybe_fs_roots(new_roots))
1941 			goto out_free;
1942 		nr_new_roots = new_roots->nnodes;
1943 	}
1944 	if (old_roots) {
1945 		if (!maybe_fs_roots(old_roots))
1946 			goto out_free;
1947 		nr_old_roots = old_roots->nnodes;
1948 	}
1949 
1950 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
1951 	if (nr_old_roots == 0 && nr_new_roots == 0)
1952 		goto out_free;
1953 
1954 	BUG_ON(!fs_info->quota_root);
1955 
1956 	trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1957 					  nr_old_roots, nr_new_roots);
1958 
1959 	qgroups = ulist_alloc(GFP_NOFS);
1960 	if (!qgroups) {
1961 		ret = -ENOMEM;
1962 		goto out_free;
1963 	}
1964 	tmp = ulist_alloc(GFP_NOFS);
1965 	if (!tmp) {
1966 		ret = -ENOMEM;
1967 		goto out_free;
1968 	}
1969 
1970 	mutex_lock(&fs_info->qgroup_rescan_lock);
1971 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1972 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1973 			mutex_unlock(&fs_info->qgroup_rescan_lock);
1974 			ret = 0;
1975 			goto out_free;
1976 		}
1977 	}
1978 	mutex_unlock(&fs_info->qgroup_rescan_lock);
1979 
1980 	spin_lock(&fs_info->qgroup_lock);
1981 	seq = fs_info->qgroup_seq;
1982 
1983 	/* Update old refcnts using old_roots */
1984 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1985 				   UPDATE_OLD);
1986 	if (ret < 0)
1987 		goto out;
1988 
1989 	/* Update new refcnts using new_roots */
1990 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1991 				   UPDATE_NEW);
1992 	if (ret < 0)
1993 		goto out;
1994 
1995 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1996 			       num_bytes, seq);
1997 
1998 	/*
1999 	 * Bump qgroup_seq to avoid seq overlap
2000 	 */
2001 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2002 out:
2003 	spin_unlock(&fs_info->qgroup_lock);
2004 out_free:
2005 	ulist_free(tmp);
2006 	ulist_free(qgroups);
2007 	ulist_free(old_roots);
2008 	ulist_free(new_roots);
2009 	return ret;
2010 }
2011 
2012 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
2013 				 struct btrfs_fs_info *fs_info)
2014 {
2015 	struct btrfs_qgroup_extent_record *record;
2016 	struct btrfs_delayed_ref_root *delayed_refs;
2017 	struct ulist *new_roots = NULL;
2018 	struct rb_node *node;
2019 	u64 qgroup_to_skip;
2020 	int ret = 0;
2021 
2022 	delayed_refs = &trans->transaction->delayed_refs;
2023 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2024 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2025 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2026 				  node);
2027 
2028 		trace_btrfs_qgroup_account_extents(fs_info, record);
2029 
2030 		if (!ret) {
2031 			/*
2032 			 * Old roots should be searched when inserting qgroup
2033 			 * extent record
2034 			 */
2035 			if (WARN_ON(!record->old_roots)) {
2036 				/* Search commit root to find old_roots */
2037 				ret = btrfs_find_all_roots(NULL, fs_info,
2038 						record->bytenr, 0,
2039 						&record->old_roots, false);
2040 				if (ret < 0)
2041 					goto cleanup;
2042 			}
2043 
2044 			/*
2045 			 * Use SEQ_LAST as time_seq to do special search, which
2046 			 * doesn't lock tree or delayed_refs and search current
2047 			 * root. It's safe inside commit_transaction().
2048 			 */
2049 			ret = btrfs_find_all_roots(trans, fs_info,
2050 				record->bytenr, SEQ_LAST, &new_roots, false);
2051 			if (ret < 0)
2052 				goto cleanup;
2053 			if (qgroup_to_skip) {
2054 				ulist_del(new_roots, qgroup_to_skip, 0);
2055 				ulist_del(record->old_roots, qgroup_to_skip,
2056 					  0);
2057 			}
2058 			ret = btrfs_qgroup_account_extent(trans, fs_info,
2059 					record->bytenr, record->num_bytes,
2060 					record->old_roots, new_roots);
2061 			record->old_roots = NULL;
2062 			new_roots = NULL;
2063 		}
2064 cleanup:
2065 		ulist_free(record->old_roots);
2066 		ulist_free(new_roots);
2067 		new_roots = NULL;
2068 		rb_erase(node, &delayed_refs->dirty_extent_root);
2069 		kfree(record);
2070 
2071 	}
2072 	return ret;
2073 }
2074 
2075 /*
2076  * called from commit_transaction. Writes all changed qgroups to disk.
2077  */
2078 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2079 		      struct btrfs_fs_info *fs_info)
2080 {
2081 	struct btrfs_root *quota_root = fs_info->quota_root;
2082 	int ret = 0;
2083 	int start_rescan_worker = 0;
2084 
2085 	if (!quota_root)
2086 		goto out;
2087 
2088 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2089 	    test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2090 		start_rescan_worker = 1;
2091 
2092 	if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2093 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2094 
2095 	spin_lock(&fs_info->qgroup_lock);
2096 	while (!list_empty(&fs_info->dirty_qgroups)) {
2097 		struct btrfs_qgroup *qgroup;
2098 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2099 					  struct btrfs_qgroup, dirty);
2100 		list_del_init(&qgroup->dirty);
2101 		spin_unlock(&fs_info->qgroup_lock);
2102 		ret = update_qgroup_info_item(trans, quota_root, qgroup);
2103 		if (ret)
2104 			fs_info->qgroup_flags |=
2105 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2106 		ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2107 		if (ret)
2108 			fs_info->qgroup_flags |=
2109 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2110 		spin_lock(&fs_info->qgroup_lock);
2111 	}
2112 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2113 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2114 	else
2115 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2116 	spin_unlock(&fs_info->qgroup_lock);
2117 
2118 	ret = update_qgroup_status_item(trans, fs_info, quota_root);
2119 	if (ret)
2120 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2121 
2122 	if (!ret && start_rescan_worker) {
2123 		ret = qgroup_rescan_init(fs_info, 0, 1);
2124 		if (!ret) {
2125 			qgroup_rescan_zero_tracking(fs_info);
2126 			btrfs_queue_work(fs_info->qgroup_rescan_workers,
2127 					 &fs_info->qgroup_rescan_work);
2128 		}
2129 		ret = 0;
2130 	}
2131 
2132 out:
2133 
2134 	return ret;
2135 }
2136 
2137 /*
2138  * Copy the accounting information between qgroups. This is necessary
2139  * when a snapshot or a subvolume is created. Throwing an error will
2140  * cause a transaction abort so we take extra care here to only error
2141  * when a readonly fs is a reasonable outcome.
2142  */
2143 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2144 			 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2145 			 struct btrfs_qgroup_inherit *inherit)
2146 {
2147 	int ret = 0;
2148 	int i;
2149 	u64 *i_qgroups;
2150 	struct btrfs_root *quota_root = fs_info->quota_root;
2151 	struct btrfs_qgroup *srcgroup;
2152 	struct btrfs_qgroup *dstgroup;
2153 	u32 level_size = 0;
2154 	u64 nums;
2155 
2156 	mutex_lock(&fs_info->qgroup_ioctl_lock);
2157 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2158 		goto out;
2159 
2160 	if (!quota_root) {
2161 		ret = -EINVAL;
2162 		goto out;
2163 	}
2164 
2165 	if (inherit) {
2166 		i_qgroups = (u64 *)(inherit + 1);
2167 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2168 		       2 * inherit->num_excl_copies;
2169 		for (i = 0; i < nums; ++i) {
2170 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2171 
2172 			/*
2173 			 * Zero out invalid groups so we can ignore
2174 			 * them later.
2175 			 */
2176 			if (!srcgroup ||
2177 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2178 				*i_qgroups = 0ULL;
2179 
2180 			++i_qgroups;
2181 		}
2182 	}
2183 
2184 	/*
2185 	 * create a tracking group for the subvol itself
2186 	 */
2187 	ret = add_qgroup_item(trans, quota_root, objectid);
2188 	if (ret)
2189 		goto out;
2190 
2191 	if (srcid) {
2192 		struct btrfs_root *srcroot;
2193 		struct btrfs_key srckey;
2194 
2195 		srckey.objectid = srcid;
2196 		srckey.type = BTRFS_ROOT_ITEM_KEY;
2197 		srckey.offset = (u64)-1;
2198 		srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2199 		if (IS_ERR(srcroot)) {
2200 			ret = PTR_ERR(srcroot);
2201 			goto out;
2202 		}
2203 
2204 		level_size = fs_info->nodesize;
2205 	}
2206 
2207 	/*
2208 	 * add qgroup to all inherited groups
2209 	 */
2210 	if (inherit) {
2211 		i_qgroups = (u64 *)(inherit + 1);
2212 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2213 			if (*i_qgroups == 0)
2214 				continue;
2215 			ret = add_qgroup_relation_item(trans, quota_root,
2216 						       objectid, *i_qgroups);
2217 			if (ret && ret != -EEXIST)
2218 				goto out;
2219 			ret = add_qgroup_relation_item(trans, quota_root,
2220 						       *i_qgroups, objectid);
2221 			if (ret && ret != -EEXIST)
2222 				goto out;
2223 		}
2224 		ret = 0;
2225 	}
2226 
2227 
2228 	spin_lock(&fs_info->qgroup_lock);
2229 
2230 	dstgroup = add_qgroup_rb(fs_info, objectid);
2231 	if (IS_ERR(dstgroup)) {
2232 		ret = PTR_ERR(dstgroup);
2233 		goto unlock;
2234 	}
2235 
2236 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2237 		dstgroup->lim_flags = inherit->lim.flags;
2238 		dstgroup->max_rfer = inherit->lim.max_rfer;
2239 		dstgroup->max_excl = inherit->lim.max_excl;
2240 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2241 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2242 
2243 		ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2244 		if (ret) {
2245 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2246 			btrfs_info(fs_info,
2247 				   "unable to update quota limit for %llu",
2248 				   dstgroup->qgroupid);
2249 			goto unlock;
2250 		}
2251 	}
2252 
2253 	if (srcid) {
2254 		srcgroup = find_qgroup_rb(fs_info, srcid);
2255 		if (!srcgroup)
2256 			goto unlock;
2257 
2258 		/*
2259 		 * We call inherit after we clone the root in order to make sure
2260 		 * our counts don't go crazy, so at this point the only
2261 		 * difference between the two roots should be the root node.
2262 		 */
2263 		dstgroup->rfer = srcgroup->rfer;
2264 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2265 		dstgroup->excl = level_size;
2266 		dstgroup->excl_cmpr = level_size;
2267 		srcgroup->excl = level_size;
2268 		srcgroup->excl_cmpr = level_size;
2269 
2270 		/* inherit the limit info */
2271 		dstgroup->lim_flags = srcgroup->lim_flags;
2272 		dstgroup->max_rfer = srcgroup->max_rfer;
2273 		dstgroup->max_excl = srcgroup->max_excl;
2274 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2275 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2276 
2277 		qgroup_dirty(fs_info, dstgroup);
2278 		qgroup_dirty(fs_info, srcgroup);
2279 	}
2280 
2281 	if (!inherit)
2282 		goto unlock;
2283 
2284 	i_qgroups = (u64 *)(inherit + 1);
2285 	for (i = 0; i < inherit->num_qgroups; ++i) {
2286 		if (*i_qgroups) {
2287 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2288 			if (ret)
2289 				goto unlock;
2290 		}
2291 		++i_qgroups;
2292 	}
2293 
2294 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2295 		struct btrfs_qgroup *src;
2296 		struct btrfs_qgroup *dst;
2297 
2298 		if (!i_qgroups[0] || !i_qgroups[1])
2299 			continue;
2300 
2301 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2302 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2303 
2304 		if (!src || !dst) {
2305 			ret = -EINVAL;
2306 			goto unlock;
2307 		}
2308 
2309 		dst->rfer = src->rfer - level_size;
2310 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2311 	}
2312 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2313 		struct btrfs_qgroup *src;
2314 		struct btrfs_qgroup *dst;
2315 
2316 		if (!i_qgroups[0] || !i_qgroups[1])
2317 			continue;
2318 
2319 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2320 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2321 
2322 		if (!src || !dst) {
2323 			ret = -EINVAL;
2324 			goto unlock;
2325 		}
2326 
2327 		dst->excl = src->excl + level_size;
2328 		dst->excl_cmpr = src->excl_cmpr + level_size;
2329 	}
2330 
2331 unlock:
2332 	spin_unlock(&fs_info->qgroup_lock);
2333 out:
2334 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
2335 	return ret;
2336 }
2337 
2338 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2339 {
2340 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2341 	    qg->reserved + (s64)qg->rfer + num_bytes > qg->max_rfer)
2342 		return false;
2343 
2344 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2345 	    qg->reserved + (s64)qg->excl + num_bytes > qg->max_excl)
2346 		return false;
2347 
2348 	return true;
2349 }
2350 
2351 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce)
2352 {
2353 	struct btrfs_root *quota_root;
2354 	struct btrfs_qgroup *qgroup;
2355 	struct btrfs_fs_info *fs_info = root->fs_info;
2356 	u64 ref_root = root->root_key.objectid;
2357 	int ret = 0;
2358 	int retried = 0;
2359 	struct ulist_node *unode;
2360 	struct ulist_iterator uiter;
2361 
2362 	if (!is_fstree(ref_root))
2363 		return 0;
2364 
2365 	if (num_bytes == 0)
2366 		return 0;
2367 
2368 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2369 	    capable(CAP_SYS_RESOURCE))
2370 		enforce = false;
2371 
2372 retry:
2373 	spin_lock(&fs_info->qgroup_lock);
2374 	quota_root = fs_info->quota_root;
2375 	if (!quota_root)
2376 		goto out;
2377 
2378 	qgroup = find_qgroup_rb(fs_info, ref_root);
2379 	if (!qgroup)
2380 		goto out;
2381 
2382 	/*
2383 	 * in a first step, we check all affected qgroups if any limits would
2384 	 * be exceeded
2385 	 */
2386 	ulist_reinit(fs_info->qgroup_ulist);
2387 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2388 			(uintptr_t)qgroup, GFP_ATOMIC);
2389 	if (ret < 0)
2390 		goto out;
2391 	ULIST_ITER_INIT(&uiter);
2392 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2393 		struct btrfs_qgroup *qg;
2394 		struct btrfs_qgroup_list *glist;
2395 
2396 		qg = unode_aux_to_qgroup(unode);
2397 
2398 		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2399 			/*
2400 			 * Commit the tree and retry, since we may have
2401 			 * deletions which would free up space.
2402 			 */
2403 			if (!retried && qg->reserved > 0) {
2404 				struct btrfs_trans_handle *trans;
2405 
2406 				spin_unlock(&fs_info->qgroup_lock);
2407 				ret = btrfs_start_delalloc_inodes(root, 0);
2408 				if (ret)
2409 					return ret;
2410 				btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
2411 				trans = btrfs_join_transaction(root);
2412 				if (IS_ERR(trans))
2413 					return PTR_ERR(trans);
2414 				ret = btrfs_commit_transaction(trans);
2415 				if (ret)
2416 					return ret;
2417 				retried++;
2418 				goto retry;
2419 			}
2420 			ret = -EDQUOT;
2421 			goto out;
2422 		}
2423 
2424 		list_for_each_entry(glist, &qg->groups, next_group) {
2425 			ret = ulist_add(fs_info->qgroup_ulist,
2426 					glist->group->qgroupid,
2427 					(uintptr_t)glist->group, GFP_ATOMIC);
2428 			if (ret < 0)
2429 				goto out;
2430 		}
2431 	}
2432 	ret = 0;
2433 	/*
2434 	 * no limits exceeded, now record the reservation into all qgroups
2435 	 */
2436 	ULIST_ITER_INIT(&uiter);
2437 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2438 		struct btrfs_qgroup *qg;
2439 
2440 		qg = unode_aux_to_qgroup(unode);
2441 
2442 		trace_qgroup_update_reserve(fs_info, qg, num_bytes);
2443 		qg->reserved += num_bytes;
2444 	}
2445 
2446 out:
2447 	spin_unlock(&fs_info->qgroup_lock);
2448 	return ret;
2449 }
2450 
2451 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2452 			       u64 ref_root, u64 num_bytes)
2453 {
2454 	struct btrfs_root *quota_root;
2455 	struct btrfs_qgroup *qgroup;
2456 	struct ulist_node *unode;
2457 	struct ulist_iterator uiter;
2458 	int ret = 0;
2459 
2460 	if (!is_fstree(ref_root))
2461 		return;
2462 
2463 	if (num_bytes == 0)
2464 		return;
2465 
2466 	spin_lock(&fs_info->qgroup_lock);
2467 
2468 	quota_root = fs_info->quota_root;
2469 	if (!quota_root)
2470 		goto out;
2471 
2472 	qgroup = find_qgroup_rb(fs_info, ref_root);
2473 	if (!qgroup)
2474 		goto out;
2475 
2476 	ulist_reinit(fs_info->qgroup_ulist);
2477 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2478 			(uintptr_t)qgroup, GFP_ATOMIC);
2479 	if (ret < 0)
2480 		goto out;
2481 	ULIST_ITER_INIT(&uiter);
2482 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2483 		struct btrfs_qgroup *qg;
2484 		struct btrfs_qgroup_list *glist;
2485 
2486 		qg = unode_aux_to_qgroup(unode);
2487 
2488 		trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes);
2489 		if (qg->reserved < num_bytes)
2490 			report_reserved_underflow(fs_info, qg, num_bytes);
2491 		else
2492 			qg->reserved -= num_bytes;
2493 
2494 		list_for_each_entry(glist, &qg->groups, next_group) {
2495 			ret = ulist_add(fs_info->qgroup_ulist,
2496 					glist->group->qgroupid,
2497 					(uintptr_t)glist->group, GFP_ATOMIC);
2498 			if (ret < 0)
2499 				goto out;
2500 		}
2501 	}
2502 
2503 out:
2504 	spin_unlock(&fs_info->qgroup_lock);
2505 }
2506 
2507 /*
2508  * returns < 0 on error, 0 when more leafs are to be scanned.
2509  * returns 1 when done.
2510  */
2511 static int
2512 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2513 		   struct btrfs_trans_handle *trans)
2514 {
2515 	struct btrfs_key found;
2516 	struct extent_buffer *scratch_leaf = NULL;
2517 	struct ulist *roots = NULL;
2518 	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2519 	u64 num_bytes;
2520 	int slot;
2521 	int ret;
2522 
2523 	mutex_lock(&fs_info->qgroup_rescan_lock);
2524 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
2525 					 &fs_info->qgroup_rescan_progress,
2526 					 path, 1, 0);
2527 
2528 	btrfs_debug(fs_info,
2529 		"current progress key (%llu %u %llu), search_slot ret %d",
2530 		fs_info->qgroup_rescan_progress.objectid,
2531 		fs_info->qgroup_rescan_progress.type,
2532 		fs_info->qgroup_rescan_progress.offset, ret);
2533 
2534 	if (ret) {
2535 		/*
2536 		 * The rescan is about to end, we will not be scanning any
2537 		 * further blocks. We cannot unset the RESCAN flag here, because
2538 		 * we want to commit the transaction if everything went well.
2539 		 * To make the live accounting work in this phase, we set our
2540 		 * scan progress pointer such that every real extent objectid
2541 		 * will be smaller.
2542 		 */
2543 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2544 		btrfs_release_path(path);
2545 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2546 		return ret;
2547 	}
2548 
2549 	btrfs_item_key_to_cpu(path->nodes[0], &found,
2550 			      btrfs_header_nritems(path->nodes[0]) - 1);
2551 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2552 
2553 	btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2554 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2555 	if (!scratch_leaf) {
2556 		ret = -ENOMEM;
2557 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2558 		goto out;
2559 	}
2560 	extent_buffer_get(scratch_leaf);
2561 	btrfs_tree_read_lock(scratch_leaf);
2562 	btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2563 	slot = path->slots[0];
2564 	btrfs_release_path(path);
2565 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2566 
2567 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2568 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2569 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2570 		    found.type != BTRFS_METADATA_ITEM_KEY)
2571 			continue;
2572 		if (found.type == BTRFS_METADATA_ITEM_KEY)
2573 			num_bytes = fs_info->nodesize;
2574 		else
2575 			num_bytes = found.offset;
2576 
2577 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2578 					   &roots, false);
2579 		if (ret < 0)
2580 			goto out;
2581 		/* For rescan, just pass old_roots as NULL */
2582 		ret = btrfs_qgroup_account_extent(trans, fs_info,
2583 				found.objectid, num_bytes, NULL, roots);
2584 		if (ret < 0)
2585 			goto out;
2586 	}
2587 out:
2588 	if (scratch_leaf) {
2589 		btrfs_tree_read_unlock_blocking(scratch_leaf);
2590 		free_extent_buffer(scratch_leaf);
2591 	}
2592 	btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2593 
2594 	return ret;
2595 }
2596 
2597 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2598 {
2599 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2600 						     qgroup_rescan_work);
2601 	struct btrfs_path *path;
2602 	struct btrfs_trans_handle *trans = NULL;
2603 	int err = -ENOMEM;
2604 	int ret = 0;
2605 
2606 	path = btrfs_alloc_path();
2607 	if (!path)
2608 		goto out;
2609 
2610 	err = 0;
2611 	while (!err && !btrfs_fs_closing(fs_info)) {
2612 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
2613 		if (IS_ERR(trans)) {
2614 			err = PTR_ERR(trans);
2615 			break;
2616 		}
2617 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2618 			err = -EINTR;
2619 		} else {
2620 			err = qgroup_rescan_leaf(fs_info, path, trans);
2621 		}
2622 		if (err > 0)
2623 			btrfs_commit_transaction(trans);
2624 		else
2625 			btrfs_end_transaction(trans);
2626 	}
2627 
2628 out:
2629 	btrfs_free_path(path);
2630 
2631 	mutex_lock(&fs_info->qgroup_rescan_lock);
2632 	if (!btrfs_fs_closing(fs_info))
2633 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2634 
2635 	if (err > 0 &&
2636 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2637 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2638 	} else if (err < 0) {
2639 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2640 	}
2641 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2642 
2643 	/*
2644 	 * only update status, since the previous part has already updated the
2645 	 * qgroup info.
2646 	 */
2647 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
2648 	if (IS_ERR(trans)) {
2649 		err = PTR_ERR(trans);
2650 		btrfs_err(fs_info,
2651 			  "fail to start transaction for status update: %d",
2652 			  err);
2653 		goto done;
2654 	}
2655 	ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2656 	if (ret < 0) {
2657 		err = ret;
2658 		btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2659 	}
2660 	btrfs_end_transaction(trans);
2661 
2662 	if (btrfs_fs_closing(fs_info)) {
2663 		btrfs_info(fs_info, "qgroup scan paused");
2664 	} else if (err >= 0) {
2665 		btrfs_info(fs_info, "qgroup scan completed%s",
2666 			err > 0 ? " (inconsistency flag cleared)" : "");
2667 	} else {
2668 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
2669 	}
2670 
2671 done:
2672 	mutex_lock(&fs_info->qgroup_rescan_lock);
2673 	fs_info->qgroup_rescan_running = false;
2674 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2675 	complete_all(&fs_info->qgroup_rescan_completion);
2676 }
2677 
2678 /*
2679  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2680  * memory required for the rescan context.
2681  */
2682 static int
2683 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2684 		   int init_flags)
2685 {
2686 	int ret = 0;
2687 
2688 	if (!init_flags &&
2689 	    (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2690 	     !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2691 		ret = -EINVAL;
2692 		goto err;
2693 	}
2694 
2695 	mutex_lock(&fs_info->qgroup_rescan_lock);
2696 	spin_lock(&fs_info->qgroup_lock);
2697 
2698 	if (init_flags) {
2699 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2700 			ret = -EINPROGRESS;
2701 		else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2702 			ret = -EINVAL;
2703 
2704 		if (ret) {
2705 			spin_unlock(&fs_info->qgroup_lock);
2706 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2707 			goto err;
2708 		}
2709 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2710 	}
2711 
2712 	memset(&fs_info->qgroup_rescan_progress, 0,
2713 		sizeof(fs_info->qgroup_rescan_progress));
2714 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2715 	init_completion(&fs_info->qgroup_rescan_completion);
2716 	fs_info->qgroup_rescan_running = true;
2717 
2718 	spin_unlock(&fs_info->qgroup_lock);
2719 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2720 
2721 	memset(&fs_info->qgroup_rescan_work, 0,
2722 	       sizeof(fs_info->qgroup_rescan_work));
2723 	btrfs_init_work(&fs_info->qgroup_rescan_work,
2724 			btrfs_qgroup_rescan_helper,
2725 			btrfs_qgroup_rescan_worker, NULL, NULL);
2726 
2727 	if (ret) {
2728 err:
2729 		btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2730 		return ret;
2731 	}
2732 
2733 	return 0;
2734 }
2735 
2736 static void
2737 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2738 {
2739 	struct rb_node *n;
2740 	struct btrfs_qgroup *qgroup;
2741 
2742 	spin_lock(&fs_info->qgroup_lock);
2743 	/* clear all current qgroup tracking information */
2744 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2745 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
2746 		qgroup->rfer = 0;
2747 		qgroup->rfer_cmpr = 0;
2748 		qgroup->excl = 0;
2749 		qgroup->excl_cmpr = 0;
2750 	}
2751 	spin_unlock(&fs_info->qgroup_lock);
2752 }
2753 
2754 int
2755 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2756 {
2757 	int ret = 0;
2758 	struct btrfs_trans_handle *trans;
2759 
2760 	ret = qgroup_rescan_init(fs_info, 0, 1);
2761 	if (ret)
2762 		return ret;
2763 
2764 	/*
2765 	 * We have set the rescan_progress to 0, which means no more
2766 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2767 	 * However, btrfs_qgroup_account_ref may be right after its call
2768 	 * to btrfs_find_all_roots, in which case it would still do the
2769 	 * accounting.
2770 	 * To solve this, we're committing the transaction, which will
2771 	 * ensure we run all delayed refs and only after that, we are
2772 	 * going to clear all tracking information for a clean start.
2773 	 */
2774 
2775 	trans = btrfs_join_transaction(fs_info->fs_root);
2776 	if (IS_ERR(trans)) {
2777 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2778 		return PTR_ERR(trans);
2779 	}
2780 	ret = btrfs_commit_transaction(trans);
2781 	if (ret) {
2782 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2783 		return ret;
2784 	}
2785 
2786 	qgroup_rescan_zero_tracking(fs_info);
2787 
2788 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
2789 			 &fs_info->qgroup_rescan_work);
2790 
2791 	return 0;
2792 }
2793 
2794 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2795 				     bool interruptible)
2796 {
2797 	int running;
2798 	int ret = 0;
2799 
2800 	mutex_lock(&fs_info->qgroup_rescan_lock);
2801 	spin_lock(&fs_info->qgroup_lock);
2802 	running = fs_info->qgroup_rescan_running;
2803 	spin_unlock(&fs_info->qgroup_lock);
2804 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2805 
2806 	if (!running)
2807 		return 0;
2808 
2809 	if (interruptible)
2810 		ret = wait_for_completion_interruptible(
2811 					&fs_info->qgroup_rescan_completion);
2812 	else
2813 		wait_for_completion(&fs_info->qgroup_rescan_completion);
2814 
2815 	return ret;
2816 }
2817 
2818 /*
2819  * this is only called from open_ctree where we're still single threaded, thus
2820  * locking is omitted here.
2821  */
2822 void
2823 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2824 {
2825 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2826 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
2827 				 &fs_info->qgroup_rescan_work);
2828 }
2829 
2830 /*
2831  * Reserve qgroup space for range [start, start + len).
2832  *
2833  * This function will either reserve space from related qgroups or doing
2834  * nothing if the range is already reserved.
2835  *
2836  * Return 0 for successful reserve
2837  * Return <0 for error (including -EQUOT)
2838  *
2839  * NOTE: this function may sleep for memory allocation.
2840  *       if btrfs_qgroup_reserve_data() is called multiple times with
2841  *       same @reserved, caller must ensure when error happens it's OK
2842  *       to free *ALL* reserved space.
2843  */
2844 int btrfs_qgroup_reserve_data(struct inode *inode,
2845 			struct extent_changeset **reserved_ret, u64 start,
2846 			u64 len)
2847 {
2848 	struct btrfs_root *root = BTRFS_I(inode)->root;
2849 	struct ulist_node *unode;
2850 	struct ulist_iterator uiter;
2851 	struct extent_changeset *reserved;
2852 	u64 orig_reserved;
2853 	u64 to_reserve;
2854 	int ret;
2855 
2856 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2857 	    !is_fstree(root->objectid) || len == 0)
2858 		return 0;
2859 
2860 	/* @reserved parameter is mandatory for qgroup */
2861 	if (WARN_ON(!reserved_ret))
2862 		return -EINVAL;
2863 	if (!*reserved_ret) {
2864 		*reserved_ret = extent_changeset_alloc();
2865 		if (!*reserved_ret)
2866 			return -ENOMEM;
2867 	}
2868 	reserved = *reserved_ret;
2869 	/* Record already reserved space */
2870 	orig_reserved = reserved->bytes_changed;
2871 	ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2872 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
2873 
2874 	/* Newly reserved space */
2875 	to_reserve = reserved->bytes_changed - orig_reserved;
2876 	trace_btrfs_qgroup_reserve_data(inode, start, len,
2877 					to_reserve, QGROUP_RESERVE);
2878 	if (ret < 0)
2879 		goto cleanup;
2880 	ret = qgroup_reserve(root, to_reserve, true);
2881 	if (ret < 0)
2882 		goto cleanup;
2883 
2884 	return ret;
2885 
2886 cleanup:
2887 	/* cleanup *ALL* already reserved ranges */
2888 	ULIST_ITER_INIT(&uiter);
2889 	while ((unode = ulist_next(&reserved->range_changed, &uiter)))
2890 		clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2891 				 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
2892 	extent_changeset_release(reserved);
2893 	return ret;
2894 }
2895 
2896 /* Free ranges specified by @reserved, normally in error path */
2897 static int qgroup_free_reserved_data(struct inode *inode,
2898 			struct extent_changeset *reserved, u64 start, u64 len)
2899 {
2900 	struct btrfs_root *root = BTRFS_I(inode)->root;
2901 	struct ulist_node *unode;
2902 	struct ulist_iterator uiter;
2903 	struct extent_changeset changeset;
2904 	int freed = 0;
2905 	int ret;
2906 
2907 	extent_changeset_init(&changeset);
2908 	len = round_up(start + len, root->fs_info->sectorsize);
2909 	start = round_down(start, root->fs_info->sectorsize);
2910 
2911 	ULIST_ITER_INIT(&uiter);
2912 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
2913 		u64 range_start = unode->val;
2914 		/* unode->aux is the inclusive end */
2915 		u64 range_len = unode->aux - range_start + 1;
2916 		u64 free_start;
2917 		u64 free_len;
2918 
2919 		extent_changeset_release(&changeset);
2920 
2921 		/* Only free range in range [start, start + len) */
2922 		if (range_start >= start + len ||
2923 		    range_start + range_len <= start)
2924 			continue;
2925 		free_start = max(range_start, start);
2926 		free_len = min(start + len, range_start + range_len) -
2927 			   free_start;
2928 		/*
2929 		 * TODO: To also modify reserved->ranges_reserved to reflect
2930 		 * the modification.
2931 		 *
2932 		 * However as long as we free qgroup reserved according to
2933 		 * EXTENT_QGROUP_RESERVED, we won't double free.
2934 		 * So not need to rush.
2935 		 */
2936 		ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
2937 				free_start, free_start + free_len - 1,
2938 				EXTENT_QGROUP_RESERVED, &changeset);
2939 		if (ret < 0)
2940 			goto out;
2941 		freed += changeset.bytes_changed;
2942 	}
2943 	btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed);
2944 	ret = freed;
2945 out:
2946 	extent_changeset_release(&changeset);
2947 	return ret;
2948 }
2949 
2950 static int __btrfs_qgroup_release_data(struct inode *inode,
2951 			struct extent_changeset *reserved, u64 start, u64 len,
2952 			int free)
2953 {
2954 	struct extent_changeset changeset;
2955 	int trace_op = QGROUP_RELEASE;
2956 	int ret;
2957 
2958 	/* In release case, we shouldn't have @reserved */
2959 	WARN_ON(!free && reserved);
2960 	if (free && reserved)
2961 		return qgroup_free_reserved_data(inode, reserved, start, len);
2962 	extent_changeset_init(&changeset);
2963 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2964 			start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2965 	if (ret < 0)
2966 		goto out;
2967 
2968 	if (free)
2969 		trace_op = QGROUP_FREE;
2970 	trace_btrfs_qgroup_release_data(inode, start, len,
2971 					changeset.bytes_changed, trace_op);
2972 	if (free)
2973 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
2974 				BTRFS_I(inode)->root->objectid,
2975 				changeset.bytes_changed);
2976 	ret = changeset.bytes_changed;
2977 out:
2978 	extent_changeset_release(&changeset);
2979 	return ret;
2980 }
2981 
2982 /*
2983  * Free a reserved space range from io_tree and related qgroups
2984  *
2985  * Should be called when a range of pages get invalidated before reaching disk.
2986  * Or for error cleanup case.
2987  * if @reserved is given, only reserved range in [@start, @start + @len) will
2988  * be freed.
2989  *
2990  * For data written to disk, use btrfs_qgroup_release_data().
2991  *
2992  * NOTE: This function may sleep for memory allocation.
2993  */
2994 int btrfs_qgroup_free_data(struct inode *inode,
2995 			struct extent_changeset *reserved, u64 start, u64 len)
2996 {
2997 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
2998 }
2999 
3000 /*
3001  * Release a reserved space range from io_tree only.
3002  *
3003  * Should be called when a range of pages get written to disk and corresponding
3004  * FILE_EXTENT is inserted into corresponding root.
3005  *
3006  * Since new qgroup accounting framework will only update qgroup numbers at
3007  * commit_transaction() time, its reserved space shouldn't be freed from
3008  * related qgroups.
3009  *
3010  * But we should release the range from io_tree, to allow further write to be
3011  * COWed.
3012  *
3013  * NOTE: This function may sleep for memory allocation.
3014  */
3015 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3016 {
3017 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3018 }
3019 
3020 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3021 			      bool enforce)
3022 {
3023 	struct btrfs_fs_info *fs_info = root->fs_info;
3024 	int ret;
3025 
3026 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3027 	    !is_fstree(root->objectid) || num_bytes == 0)
3028 		return 0;
3029 
3030 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3031 	trace_qgroup_meta_reserve(root, (s64)num_bytes);
3032 	ret = qgroup_reserve(root, num_bytes, enforce);
3033 	if (ret < 0)
3034 		return ret;
3035 	atomic64_add(num_bytes, &root->qgroup_meta_rsv);
3036 	return ret;
3037 }
3038 
3039 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
3040 {
3041 	struct btrfs_fs_info *fs_info = root->fs_info;
3042 	u64 reserved;
3043 
3044 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3045 	    !is_fstree(root->objectid))
3046 		return;
3047 
3048 	reserved = atomic64_xchg(&root->qgroup_meta_rsv, 0);
3049 	if (reserved == 0)
3050 		return;
3051 	trace_qgroup_meta_reserve(root, -(s64)reserved);
3052 	btrfs_qgroup_free_refroot(fs_info, root->objectid, reserved);
3053 }
3054 
3055 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
3056 {
3057 	struct btrfs_fs_info *fs_info = root->fs_info;
3058 
3059 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3060 	    !is_fstree(root->objectid))
3061 		return;
3062 
3063 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3064 	WARN_ON(atomic64_read(&root->qgroup_meta_rsv) < num_bytes);
3065 	atomic64_sub(num_bytes, &root->qgroup_meta_rsv);
3066 	trace_qgroup_meta_reserve(root, -(s64)num_bytes);
3067 	btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes);
3068 }
3069 
3070 /*
3071  * Check qgroup reserved space leaking, normally at destroy inode
3072  * time
3073  */
3074 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3075 {
3076 	struct extent_changeset changeset;
3077 	struct ulist_node *unode;
3078 	struct ulist_iterator iter;
3079 	int ret;
3080 
3081 	extent_changeset_init(&changeset);
3082 	ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3083 			EXTENT_QGROUP_RESERVED, &changeset);
3084 
3085 	WARN_ON(ret < 0);
3086 	if (WARN_ON(changeset.bytes_changed)) {
3087 		ULIST_ITER_INIT(&iter);
3088 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3089 			btrfs_warn(BTRFS_I(inode)->root->fs_info,
3090 				"leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3091 				inode->i_ino, unode->val, unode->aux);
3092 		}
3093 		btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3094 				BTRFS_I(inode)->root->objectid,
3095 				changeset.bytes_changed);
3096 
3097 	}
3098 	extent_changeset_release(&changeset);
3099 }
3100