xref: /openbmc/linux/fs/btrfs/qgroup.c (revision 35267cea)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sched/mm.h>
15 
16 #include "ctree.h"
17 #include "transaction.h"
18 #include "disk-io.h"
19 #include "locking.h"
20 #include "ulist.h"
21 #include "backref.h"
22 #include "extent_io.h"
23 #include "qgroup.h"
24 #include "block-group.h"
25 #include "sysfs.h"
26 #include "tree-mod-log.h"
27 
28 /* TODO XXX FIXME
29  *  - subvol delete -> delete when ref goes to 0? delete limits also?
30  *  - reorganize keys
31  *  - compressed
32  *  - sync
33  *  - copy also limits on subvol creation
34  *  - limit
35  *  - caches for ulists
36  *  - performance benchmarks
37  *  - check all ioctl parameters
38  */
39 
40 /*
41  * Helpers to access qgroup reservation
42  *
43  * Callers should ensure the lock context and type are valid
44  */
45 
46 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
47 {
48 	u64 ret = 0;
49 	int i;
50 
51 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
52 		ret += qgroup->rsv.values[i];
53 
54 	return ret;
55 }
56 
57 #ifdef CONFIG_BTRFS_DEBUG
58 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
59 {
60 	if (type == BTRFS_QGROUP_RSV_DATA)
61 		return "data";
62 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
63 		return "meta_pertrans";
64 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
65 		return "meta_prealloc";
66 	return NULL;
67 }
68 #endif
69 
70 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
71 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
72 			   enum btrfs_qgroup_rsv_type type)
73 {
74 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
75 	qgroup->rsv.values[type] += num_bytes;
76 }
77 
78 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
79 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
80 			       enum btrfs_qgroup_rsv_type type)
81 {
82 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
83 	if (qgroup->rsv.values[type] >= num_bytes) {
84 		qgroup->rsv.values[type] -= num_bytes;
85 		return;
86 	}
87 #ifdef CONFIG_BTRFS_DEBUG
88 	WARN_RATELIMIT(1,
89 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
90 		qgroup->qgroupid, qgroup_rsv_type_str(type),
91 		qgroup->rsv.values[type], num_bytes);
92 #endif
93 	qgroup->rsv.values[type] = 0;
94 }
95 
96 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
97 				     struct btrfs_qgroup *dest,
98 				     struct btrfs_qgroup *src)
99 {
100 	int i;
101 
102 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
103 		qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
104 }
105 
106 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
107 					 struct btrfs_qgroup *dest,
108 					  struct btrfs_qgroup *src)
109 {
110 	int i;
111 
112 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
113 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
114 }
115 
116 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
117 					   int mod)
118 {
119 	if (qg->old_refcnt < seq)
120 		qg->old_refcnt = seq;
121 	qg->old_refcnt += mod;
122 }
123 
124 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
125 					   int mod)
126 {
127 	if (qg->new_refcnt < seq)
128 		qg->new_refcnt = seq;
129 	qg->new_refcnt += mod;
130 }
131 
132 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
133 {
134 	if (qg->old_refcnt < seq)
135 		return 0;
136 	return qg->old_refcnt - seq;
137 }
138 
139 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
140 {
141 	if (qg->new_refcnt < seq)
142 		return 0;
143 	return qg->new_refcnt - seq;
144 }
145 
146 /*
147  * glue structure to represent the relations between qgroups.
148  */
149 struct btrfs_qgroup_list {
150 	struct list_head next_group;
151 	struct list_head next_member;
152 	struct btrfs_qgroup *group;
153 	struct btrfs_qgroup *member;
154 };
155 
156 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
157 {
158 	return (u64)(uintptr_t)qg;
159 }
160 
161 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
162 {
163 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
164 }
165 
166 static int
167 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
168 		   int init_flags);
169 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
170 
171 /* must be called with qgroup_ioctl_lock held */
172 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
173 					   u64 qgroupid)
174 {
175 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
176 	struct btrfs_qgroup *qgroup;
177 
178 	while (n) {
179 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
180 		if (qgroup->qgroupid < qgroupid)
181 			n = n->rb_left;
182 		else if (qgroup->qgroupid > qgroupid)
183 			n = n->rb_right;
184 		else
185 			return qgroup;
186 	}
187 	return NULL;
188 }
189 
190 /* must be called with qgroup_lock held */
191 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
192 					  u64 qgroupid)
193 {
194 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
195 	struct rb_node *parent = NULL;
196 	struct btrfs_qgroup *qgroup;
197 
198 	while (*p) {
199 		parent = *p;
200 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
201 
202 		if (qgroup->qgroupid < qgroupid)
203 			p = &(*p)->rb_left;
204 		else if (qgroup->qgroupid > qgroupid)
205 			p = &(*p)->rb_right;
206 		else
207 			return qgroup;
208 	}
209 
210 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
211 	if (!qgroup)
212 		return ERR_PTR(-ENOMEM);
213 
214 	qgroup->qgroupid = qgroupid;
215 	INIT_LIST_HEAD(&qgroup->groups);
216 	INIT_LIST_HEAD(&qgroup->members);
217 	INIT_LIST_HEAD(&qgroup->dirty);
218 
219 	rb_link_node(&qgroup->node, parent, p);
220 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
221 
222 	return qgroup;
223 }
224 
225 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
226 			    struct btrfs_qgroup *qgroup)
227 {
228 	struct btrfs_qgroup_list *list;
229 
230 	list_del(&qgroup->dirty);
231 	while (!list_empty(&qgroup->groups)) {
232 		list = list_first_entry(&qgroup->groups,
233 					struct btrfs_qgroup_list, next_group);
234 		list_del(&list->next_group);
235 		list_del(&list->next_member);
236 		kfree(list);
237 	}
238 
239 	while (!list_empty(&qgroup->members)) {
240 		list = list_first_entry(&qgroup->members,
241 					struct btrfs_qgroup_list, next_member);
242 		list_del(&list->next_group);
243 		list_del(&list->next_member);
244 		kfree(list);
245 	}
246 }
247 
248 /* must be called with qgroup_lock held */
249 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
250 {
251 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
252 
253 	if (!qgroup)
254 		return -ENOENT;
255 
256 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
257 	__del_qgroup_rb(fs_info, qgroup);
258 	return 0;
259 }
260 
261 /* must be called with qgroup_lock held */
262 static int add_relation_rb(struct btrfs_fs_info *fs_info,
263 			   u64 memberid, u64 parentid)
264 {
265 	struct btrfs_qgroup *member;
266 	struct btrfs_qgroup *parent;
267 	struct btrfs_qgroup_list *list;
268 
269 	member = find_qgroup_rb(fs_info, memberid);
270 	parent = find_qgroup_rb(fs_info, parentid);
271 	if (!member || !parent)
272 		return -ENOENT;
273 
274 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
275 	if (!list)
276 		return -ENOMEM;
277 
278 	list->group = parent;
279 	list->member = member;
280 	list_add_tail(&list->next_group, &member->groups);
281 	list_add_tail(&list->next_member, &parent->members);
282 
283 	return 0;
284 }
285 
286 /* must be called with qgroup_lock held */
287 static int del_relation_rb(struct btrfs_fs_info *fs_info,
288 			   u64 memberid, u64 parentid)
289 {
290 	struct btrfs_qgroup *member;
291 	struct btrfs_qgroup *parent;
292 	struct btrfs_qgroup_list *list;
293 
294 	member = find_qgroup_rb(fs_info, memberid);
295 	parent = find_qgroup_rb(fs_info, parentid);
296 	if (!member || !parent)
297 		return -ENOENT;
298 
299 	list_for_each_entry(list, &member->groups, next_group) {
300 		if (list->group == parent) {
301 			list_del(&list->next_group);
302 			list_del(&list->next_member);
303 			kfree(list);
304 			return 0;
305 		}
306 	}
307 	return -ENOENT;
308 }
309 
310 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
311 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
312 			       u64 rfer, u64 excl)
313 {
314 	struct btrfs_qgroup *qgroup;
315 
316 	qgroup = find_qgroup_rb(fs_info, qgroupid);
317 	if (!qgroup)
318 		return -EINVAL;
319 	if (qgroup->rfer != rfer || qgroup->excl != excl)
320 		return -EINVAL;
321 	return 0;
322 }
323 #endif
324 
325 /*
326  * The full config is read in one go, only called from open_ctree()
327  * It doesn't use any locking, as at this point we're still single-threaded
328  */
329 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
330 {
331 	struct btrfs_key key;
332 	struct btrfs_key found_key;
333 	struct btrfs_root *quota_root = fs_info->quota_root;
334 	struct btrfs_path *path = NULL;
335 	struct extent_buffer *l;
336 	int slot;
337 	int ret = 0;
338 	u64 flags = 0;
339 	u64 rescan_progress = 0;
340 
341 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
342 		return 0;
343 
344 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
345 	if (!fs_info->qgroup_ulist) {
346 		ret = -ENOMEM;
347 		goto out;
348 	}
349 
350 	path = btrfs_alloc_path();
351 	if (!path) {
352 		ret = -ENOMEM;
353 		goto out;
354 	}
355 
356 	ret = btrfs_sysfs_add_qgroups(fs_info);
357 	if (ret < 0)
358 		goto out;
359 	/* default this to quota off, in case no status key is found */
360 	fs_info->qgroup_flags = 0;
361 
362 	/*
363 	 * pass 1: read status, all qgroup infos and limits
364 	 */
365 	key.objectid = 0;
366 	key.type = 0;
367 	key.offset = 0;
368 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
369 	if (ret)
370 		goto out;
371 
372 	while (1) {
373 		struct btrfs_qgroup *qgroup;
374 
375 		slot = path->slots[0];
376 		l = path->nodes[0];
377 		btrfs_item_key_to_cpu(l, &found_key, slot);
378 
379 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
380 			struct btrfs_qgroup_status_item *ptr;
381 
382 			ptr = btrfs_item_ptr(l, slot,
383 					     struct btrfs_qgroup_status_item);
384 
385 			if (btrfs_qgroup_status_version(l, ptr) !=
386 			    BTRFS_QGROUP_STATUS_VERSION) {
387 				btrfs_err(fs_info,
388 				 "old qgroup version, quota disabled");
389 				goto out;
390 			}
391 			if (btrfs_qgroup_status_generation(l, ptr) !=
392 			    fs_info->generation) {
393 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
394 				btrfs_err(fs_info,
395 					"qgroup generation mismatch, marked as inconsistent");
396 			}
397 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
398 									  ptr);
399 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
400 			goto next1;
401 		}
402 
403 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
404 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
405 			goto next1;
406 
407 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
408 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
409 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
410 			btrfs_err(fs_info, "inconsistent qgroup config");
411 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
412 		}
413 		if (!qgroup) {
414 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
415 			if (IS_ERR(qgroup)) {
416 				ret = PTR_ERR(qgroup);
417 				goto out;
418 			}
419 		}
420 		ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
421 		if (ret < 0)
422 			goto out;
423 
424 		switch (found_key.type) {
425 		case BTRFS_QGROUP_INFO_KEY: {
426 			struct btrfs_qgroup_info_item *ptr;
427 
428 			ptr = btrfs_item_ptr(l, slot,
429 					     struct btrfs_qgroup_info_item);
430 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
431 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
432 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
433 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
434 			/* generation currently unused */
435 			break;
436 		}
437 		case BTRFS_QGROUP_LIMIT_KEY: {
438 			struct btrfs_qgroup_limit_item *ptr;
439 
440 			ptr = btrfs_item_ptr(l, slot,
441 					     struct btrfs_qgroup_limit_item);
442 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
443 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
444 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
445 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
446 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
447 			break;
448 		}
449 		}
450 next1:
451 		ret = btrfs_next_item(quota_root, path);
452 		if (ret < 0)
453 			goto out;
454 		if (ret)
455 			break;
456 	}
457 	btrfs_release_path(path);
458 
459 	/*
460 	 * pass 2: read all qgroup relations
461 	 */
462 	key.objectid = 0;
463 	key.type = BTRFS_QGROUP_RELATION_KEY;
464 	key.offset = 0;
465 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
466 	if (ret)
467 		goto out;
468 	while (1) {
469 		slot = path->slots[0];
470 		l = path->nodes[0];
471 		btrfs_item_key_to_cpu(l, &found_key, slot);
472 
473 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
474 			goto next2;
475 
476 		if (found_key.objectid > found_key.offset) {
477 			/* parent <- member, not needed to build config */
478 			/* FIXME should we omit the key completely? */
479 			goto next2;
480 		}
481 
482 		ret = add_relation_rb(fs_info, found_key.objectid,
483 				      found_key.offset);
484 		if (ret == -ENOENT) {
485 			btrfs_warn(fs_info,
486 				"orphan qgroup relation 0x%llx->0x%llx",
487 				found_key.objectid, found_key.offset);
488 			ret = 0;	/* ignore the error */
489 		}
490 		if (ret)
491 			goto out;
492 next2:
493 		ret = btrfs_next_item(quota_root, path);
494 		if (ret < 0)
495 			goto out;
496 		if (ret)
497 			break;
498 	}
499 out:
500 	btrfs_free_path(path);
501 	fs_info->qgroup_flags |= flags;
502 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
503 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
504 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
505 		 ret >= 0)
506 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
507 
508 	if (ret < 0) {
509 		ulist_free(fs_info->qgroup_ulist);
510 		fs_info->qgroup_ulist = NULL;
511 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
512 		btrfs_sysfs_del_qgroups(fs_info);
513 	}
514 
515 	return ret < 0 ? ret : 0;
516 }
517 
518 /*
519  * Called in close_ctree() when quota is still enabled.  This verifies we don't
520  * leak some reserved space.
521  *
522  * Return false if no reserved space is left.
523  * Return true if some reserved space is leaked.
524  */
525 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
526 {
527 	struct rb_node *node;
528 	bool ret = false;
529 
530 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
531 		return ret;
532 	/*
533 	 * Since we're unmounting, there is no race and no need to grab qgroup
534 	 * lock.  And here we don't go post-order to provide a more user
535 	 * friendly sorted result.
536 	 */
537 	for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
538 		struct btrfs_qgroup *qgroup;
539 		int i;
540 
541 		qgroup = rb_entry(node, struct btrfs_qgroup, node);
542 		for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
543 			if (qgroup->rsv.values[i]) {
544 				ret = true;
545 				btrfs_warn(fs_info,
546 		"qgroup %hu/%llu has unreleased space, type %d rsv %llu",
547 				   btrfs_qgroup_level(qgroup->qgroupid),
548 				   btrfs_qgroup_subvolid(qgroup->qgroupid),
549 				   i, qgroup->rsv.values[i]);
550 			}
551 		}
552 	}
553 	return ret;
554 }
555 
556 /*
557  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
558  * first two are in single-threaded paths.And for the third one, we have set
559  * quota_root to be null with qgroup_lock held before, so it is safe to clean
560  * up the in-memory structures without qgroup_lock held.
561  */
562 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
563 {
564 	struct rb_node *n;
565 	struct btrfs_qgroup *qgroup;
566 
567 	while ((n = rb_first(&fs_info->qgroup_tree))) {
568 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
569 		rb_erase(n, &fs_info->qgroup_tree);
570 		__del_qgroup_rb(fs_info, qgroup);
571 		btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
572 		kfree(qgroup);
573 	}
574 	/*
575 	 * We call btrfs_free_qgroup_config() when unmounting
576 	 * filesystem and disabling quota, so we set qgroup_ulist
577 	 * to be null here to avoid double free.
578 	 */
579 	ulist_free(fs_info->qgroup_ulist);
580 	fs_info->qgroup_ulist = NULL;
581 	btrfs_sysfs_del_qgroups(fs_info);
582 }
583 
584 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
585 				    u64 dst)
586 {
587 	int ret;
588 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
589 	struct btrfs_path *path;
590 	struct btrfs_key key;
591 
592 	path = btrfs_alloc_path();
593 	if (!path)
594 		return -ENOMEM;
595 
596 	key.objectid = src;
597 	key.type = BTRFS_QGROUP_RELATION_KEY;
598 	key.offset = dst;
599 
600 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
601 
602 	btrfs_mark_buffer_dirty(path->nodes[0]);
603 
604 	btrfs_free_path(path);
605 	return ret;
606 }
607 
608 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
609 				    u64 dst)
610 {
611 	int ret;
612 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
613 	struct btrfs_path *path;
614 	struct btrfs_key key;
615 
616 	path = btrfs_alloc_path();
617 	if (!path)
618 		return -ENOMEM;
619 
620 	key.objectid = src;
621 	key.type = BTRFS_QGROUP_RELATION_KEY;
622 	key.offset = dst;
623 
624 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
625 	if (ret < 0)
626 		goto out;
627 
628 	if (ret > 0) {
629 		ret = -ENOENT;
630 		goto out;
631 	}
632 
633 	ret = btrfs_del_item(trans, quota_root, path);
634 out:
635 	btrfs_free_path(path);
636 	return ret;
637 }
638 
639 static int add_qgroup_item(struct btrfs_trans_handle *trans,
640 			   struct btrfs_root *quota_root, u64 qgroupid)
641 {
642 	int ret;
643 	struct btrfs_path *path;
644 	struct btrfs_qgroup_info_item *qgroup_info;
645 	struct btrfs_qgroup_limit_item *qgroup_limit;
646 	struct extent_buffer *leaf;
647 	struct btrfs_key key;
648 
649 	if (btrfs_is_testing(quota_root->fs_info))
650 		return 0;
651 
652 	path = btrfs_alloc_path();
653 	if (!path)
654 		return -ENOMEM;
655 
656 	key.objectid = 0;
657 	key.type = BTRFS_QGROUP_INFO_KEY;
658 	key.offset = qgroupid;
659 
660 	/*
661 	 * Avoid a transaction abort by catching -EEXIST here. In that
662 	 * case, we proceed by re-initializing the existing structure
663 	 * on disk.
664 	 */
665 
666 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
667 				      sizeof(*qgroup_info));
668 	if (ret && ret != -EEXIST)
669 		goto out;
670 
671 	leaf = path->nodes[0];
672 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
673 				 struct btrfs_qgroup_info_item);
674 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
675 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
676 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
677 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
678 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
679 
680 	btrfs_mark_buffer_dirty(leaf);
681 
682 	btrfs_release_path(path);
683 
684 	key.type = BTRFS_QGROUP_LIMIT_KEY;
685 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
686 				      sizeof(*qgroup_limit));
687 	if (ret && ret != -EEXIST)
688 		goto out;
689 
690 	leaf = path->nodes[0];
691 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
692 				  struct btrfs_qgroup_limit_item);
693 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
694 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
695 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
696 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
697 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
698 
699 	btrfs_mark_buffer_dirty(leaf);
700 
701 	ret = 0;
702 out:
703 	btrfs_free_path(path);
704 	return ret;
705 }
706 
707 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
708 {
709 	int ret;
710 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
711 	struct btrfs_path *path;
712 	struct btrfs_key key;
713 
714 	path = btrfs_alloc_path();
715 	if (!path)
716 		return -ENOMEM;
717 
718 	key.objectid = 0;
719 	key.type = BTRFS_QGROUP_INFO_KEY;
720 	key.offset = qgroupid;
721 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
722 	if (ret < 0)
723 		goto out;
724 
725 	if (ret > 0) {
726 		ret = -ENOENT;
727 		goto out;
728 	}
729 
730 	ret = btrfs_del_item(trans, quota_root, path);
731 	if (ret)
732 		goto out;
733 
734 	btrfs_release_path(path);
735 
736 	key.type = BTRFS_QGROUP_LIMIT_KEY;
737 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
738 	if (ret < 0)
739 		goto out;
740 
741 	if (ret > 0) {
742 		ret = -ENOENT;
743 		goto out;
744 	}
745 
746 	ret = btrfs_del_item(trans, quota_root, path);
747 
748 out:
749 	btrfs_free_path(path);
750 	return ret;
751 }
752 
753 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
754 				    struct btrfs_qgroup *qgroup)
755 {
756 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
757 	struct btrfs_path *path;
758 	struct btrfs_key key;
759 	struct extent_buffer *l;
760 	struct btrfs_qgroup_limit_item *qgroup_limit;
761 	int ret;
762 	int slot;
763 
764 	key.objectid = 0;
765 	key.type = BTRFS_QGROUP_LIMIT_KEY;
766 	key.offset = qgroup->qgroupid;
767 
768 	path = btrfs_alloc_path();
769 	if (!path)
770 		return -ENOMEM;
771 
772 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
773 	if (ret > 0)
774 		ret = -ENOENT;
775 
776 	if (ret)
777 		goto out;
778 
779 	l = path->nodes[0];
780 	slot = path->slots[0];
781 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
782 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
783 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
784 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
785 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
786 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
787 
788 	btrfs_mark_buffer_dirty(l);
789 
790 out:
791 	btrfs_free_path(path);
792 	return ret;
793 }
794 
795 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
796 				   struct btrfs_qgroup *qgroup)
797 {
798 	struct btrfs_fs_info *fs_info = trans->fs_info;
799 	struct btrfs_root *quota_root = fs_info->quota_root;
800 	struct btrfs_path *path;
801 	struct btrfs_key key;
802 	struct extent_buffer *l;
803 	struct btrfs_qgroup_info_item *qgroup_info;
804 	int ret;
805 	int slot;
806 
807 	if (btrfs_is_testing(fs_info))
808 		return 0;
809 
810 	key.objectid = 0;
811 	key.type = BTRFS_QGROUP_INFO_KEY;
812 	key.offset = qgroup->qgroupid;
813 
814 	path = btrfs_alloc_path();
815 	if (!path)
816 		return -ENOMEM;
817 
818 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
819 	if (ret > 0)
820 		ret = -ENOENT;
821 
822 	if (ret)
823 		goto out;
824 
825 	l = path->nodes[0];
826 	slot = path->slots[0];
827 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
828 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
829 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
830 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
831 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
832 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
833 
834 	btrfs_mark_buffer_dirty(l);
835 
836 out:
837 	btrfs_free_path(path);
838 	return ret;
839 }
840 
841 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
842 {
843 	struct btrfs_fs_info *fs_info = trans->fs_info;
844 	struct btrfs_root *quota_root = fs_info->quota_root;
845 	struct btrfs_path *path;
846 	struct btrfs_key key;
847 	struct extent_buffer *l;
848 	struct btrfs_qgroup_status_item *ptr;
849 	int ret;
850 	int slot;
851 
852 	key.objectid = 0;
853 	key.type = BTRFS_QGROUP_STATUS_KEY;
854 	key.offset = 0;
855 
856 	path = btrfs_alloc_path();
857 	if (!path)
858 		return -ENOMEM;
859 
860 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
861 	if (ret > 0)
862 		ret = -ENOENT;
863 
864 	if (ret)
865 		goto out;
866 
867 	l = path->nodes[0];
868 	slot = path->slots[0];
869 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
870 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
871 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
872 	btrfs_set_qgroup_status_rescan(l, ptr,
873 				fs_info->qgroup_rescan_progress.objectid);
874 
875 	btrfs_mark_buffer_dirty(l);
876 
877 out:
878 	btrfs_free_path(path);
879 	return ret;
880 }
881 
882 /*
883  * called with qgroup_lock held
884  */
885 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
886 				  struct btrfs_root *root)
887 {
888 	struct btrfs_path *path;
889 	struct btrfs_key key;
890 	struct extent_buffer *leaf = NULL;
891 	int ret;
892 	int nr = 0;
893 
894 	path = btrfs_alloc_path();
895 	if (!path)
896 		return -ENOMEM;
897 
898 	key.objectid = 0;
899 	key.offset = 0;
900 	key.type = 0;
901 
902 	while (1) {
903 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
904 		if (ret < 0)
905 			goto out;
906 		leaf = path->nodes[0];
907 		nr = btrfs_header_nritems(leaf);
908 		if (!nr)
909 			break;
910 		/*
911 		 * delete the leaf one by one
912 		 * since the whole tree is going
913 		 * to be deleted.
914 		 */
915 		path->slots[0] = 0;
916 		ret = btrfs_del_items(trans, root, path, 0, nr);
917 		if (ret)
918 			goto out;
919 
920 		btrfs_release_path(path);
921 	}
922 	ret = 0;
923 out:
924 	btrfs_free_path(path);
925 	return ret;
926 }
927 
928 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
929 {
930 	struct btrfs_root *quota_root;
931 	struct btrfs_root *tree_root = fs_info->tree_root;
932 	struct btrfs_path *path = NULL;
933 	struct btrfs_qgroup_status_item *ptr;
934 	struct extent_buffer *leaf;
935 	struct btrfs_key key;
936 	struct btrfs_key found_key;
937 	struct btrfs_qgroup *qgroup = NULL;
938 	struct btrfs_trans_handle *trans = NULL;
939 	struct ulist *ulist = NULL;
940 	int ret = 0;
941 	int slot;
942 
943 	mutex_lock(&fs_info->qgroup_ioctl_lock);
944 	if (fs_info->quota_root)
945 		goto out;
946 
947 	ulist = ulist_alloc(GFP_KERNEL);
948 	if (!ulist) {
949 		ret = -ENOMEM;
950 		goto out;
951 	}
952 
953 	ret = btrfs_sysfs_add_qgroups(fs_info);
954 	if (ret < 0)
955 		goto out;
956 
957 	/*
958 	 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
959 	 * avoid lock acquisition inversion problems (reported by lockdep) between
960 	 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
961 	 * start a transaction.
962 	 * After we started the transaction lock qgroup_ioctl_lock again and
963 	 * check if someone else created the quota root in the meanwhile. If so,
964 	 * just return success and release the transaction handle.
965 	 *
966 	 * Also we don't need to worry about someone else calling
967 	 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
968 	 * that function returns 0 (success) when the sysfs entries already exist.
969 	 */
970 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
971 
972 	/*
973 	 * 1 for quota root item
974 	 * 1 for BTRFS_QGROUP_STATUS item
975 	 *
976 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
977 	 * per subvolume. However those are not currently reserved since it
978 	 * would be a lot of overkill.
979 	 */
980 	trans = btrfs_start_transaction(tree_root, 2);
981 
982 	mutex_lock(&fs_info->qgroup_ioctl_lock);
983 	if (IS_ERR(trans)) {
984 		ret = PTR_ERR(trans);
985 		trans = NULL;
986 		goto out;
987 	}
988 
989 	if (fs_info->quota_root)
990 		goto out;
991 
992 	fs_info->qgroup_ulist = ulist;
993 	ulist = NULL;
994 
995 	/*
996 	 * initially create the quota tree
997 	 */
998 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
999 	if (IS_ERR(quota_root)) {
1000 		ret =  PTR_ERR(quota_root);
1001 		btrfs_abort_transaction(trans, ret);
1002 		goto out;
1003 	}
1004 
1005 	path = btrfs_alloc_path();
1006 	if (!path) {
1007 		ret = -ENOMEM;
1008 		btrfs_abort_transaction(trans, ret);
1009 		goto out_free_root;
1010 	}
1011 
1012 	key.objectid = 0;
1013 	key.type = BTRFS_QGROUP_STATUS_KEY;
1014 	key.offset = 0;
1015 
1016 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1017 				      sizeof(*ptr));
1018 	if (ret) {
1019 		btrfs_abort_transaction(trans, ret);
1020 		goto out_free_path;
1021 	}
1022 
1023 	leaf = path->nodes[0];
1024 	ptr = btrfs_item_ptr(leaf, path->slots[0],
1025 				 struct btrfs_qgroup_status_item);
1026 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1027 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1028 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1029 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1030 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1031 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1032 
1033 	btrfs_mark_buffer_dirty(leaf);
1034 
1035 	key.objectid = 0;
1036 	key.type = BTRFS_ROOT_REF_KEY;
1037 	key.offset = 0;
1038 
1039 	btrfs_release_path(path);
1040 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1041 	if (ret > 0)
1042 		goto out_add_root;
1043 	if (ret < 0) {
1044 		btrfs_abort_transaction(trans, ret);
1045 		goto out_free_path;
1046 	}
1047 
1048 	while (1) {
1049 		slot = path->slots[0];
1050 		leaf = path->nodes[0];
1051 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1052 
1053 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
1054 
1055 			/* Release locks on tree_root before we access quota_root */
1056 			btrfs_release_path(path);
1057 
1058 			ret = add_qgroup_item(trans, quota_root,
1059 					      found_key.offset);
1060 			if (ret) {
1061 				btrfs_abort_transaction(trans, ret);
1062 				goto out_free_path;
1063 			}
1064 
1065 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
1066 			if (IS_ERR(qgroup)) {
1067 				ret = PTR_ERR(qgroup);
1068 				btrfs_abort_transaction(trans, ret);
1069 				goto out_free_path;
1070 			}
1071 			ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1072 			if (ret < 0) {
1073 				btrfs_abort_transaction(trans, ret);
1074 				goto out_free_path;
1075 			}
1076 			ret = btrfs_search_slot_for_read(tree_root, &found_key,
1077 							 path, 1, 0);
1078 			if (ret < 0) {
1079 				btrfs_abort_transaction(trans, ret);
1080 				goto out_free_path;
1081 			}
1082 			if (ret > 0) {
1083 				/*
1084 				 * Shouldn't happen, but in case it does we
1085 				 * don't need to do the btrfs_next_item, just
1086 				 * continue.
1087 				 */
1088 				continue;
1089 			}
1090 		}
1091 		ret = btrfs_next_item(tree_root, path);
1092 		if (ret < 0) {
1093 			btrfs_abort_transaction(trans, ret);
1094 			goto out_free_path;
1095 		}
1096 		if (ret)
1097 			break;
1098 	}
1099 
1100 out_add_root:
1101 	btrfs_release_path(path);
1102 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1103 	if (ret) {
1104 		btrfs_abort_transaction(trans, ret);
1105 		goto out_free_path;
1106 	}
1107 
1108 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1109 	if (IS_ERR(qgroup)) {
1110 		ret = PTR_ERR(qgroup);
1111 		btrfs_abort_transaction(trans, ret);
1112 		goto out_free_path;
1113 	}
1114 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1115 	if (ret < 0) {
1116 		btrfs_abort_transaction(trans, ret);
1117 		goto out_free_path;
1118 	}
1119 
1120 	ret = btrfs_commit_transaction(trans);
1121 	trans = NULL;
1122 	if (ret)
1123 		goto out_free_path;
1124 
1125 	/*
1126 	 * Set quota enabled flag after committing the transaction, to avoid
1127 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1128 	 * creation.
1129 	 */
1130 	spin_lock(&fs_info->qgroup_lock);
1131 	fs_info->quota_root = quota_root;
1132 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1133 	spin_unlock(&fs_info->qgroup_lock);
1134 
1135 	ret = qgroup_rescan_init(fs_info, 0, 1);
1136 	if (!ret) {
1137 	        qgroup_rescan_zero_tracking(fs_info);
1138 		fs_info->qgroup_rescan_running = true;
1139 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1140 	                         &fs_info->qgroup_rescan_work);
1141 	}
1142 
1143 out_free_path:
1144 	btrfs_free_path(path);
1145 out_free_root:
1146 	if (ret)
1147 		btrfs_put_root(quota_root);
1148 out:
1149 	if (ret) {
1150 		ulist_free(fs_info->qgroup_ulist);
1151 		fs_info->qgroup_ulist = NULL;
1152 		btrfs_sysfs_del_qgroups(fs_info);
1153 	}
1154 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1155 	if (ret && trans)
1156 		btrfs_end_transaction(trans);
1157 	else if (trans)
1158 		ret = btrfs_end_transaction(trans);
1159 	ulist_free(ulist);
1160 	return ret;
1161 }
1162 
1163 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1164 {
1165 	struct btrfs_root *quota_root;
1166 	struct btrfs_trans_handle *trans = NULL;
1167 	int ret = 0;
1168 
1169 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1170 	if (!fs_info->quota_root)
1171 		goto out;
1172 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1173 
1174 	/*
1175 	 * 1 For the root item
1176 	 *
1177 	 * We should also reserve enough items for the quota tree deletion in
1178 	 * btrfs_clean_quota_tree but this is not done.
1179 	 *
1180 	 * Also, we must always start a transaction without holding the mutex
1181 	 * qgroup_ioctl_lock, see btrfs_quota_enable().
1182 	 */
1183 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1184 
1185 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1186 	if (IS_ERR(trans)) {
1187 		ret = PTR_ERR(trans);
1188 		trans = NULL;
1189 		goto out;
1190 	}
1191 
1192 	if (!fs_info->quota_root)
1193 		goto out;
1194 
1195 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1196 	btrfs_qgroup_wait_for_completion(fs_info, false);
1197 	spin_lock(&fs_info->qgroup_lock);
1198 	quota_root = fs_info->quota_root;
1199 	fs_info->quota_root = NULL;
1200 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1201 	spin_unlock(&fs_info->qgroup_lock);
1202 
1203 	btrfs_free_qgroup_config(fs_info);
1204 
1205 	ret = btrfs_clean_quota_tree(trans, quota_root);
1206 	if (ret) {
1207 		btrfs_abort_transaction(trans, ret);
1208 		goto out;
1209 	}
1210 
1211 	ret = btrfs_del_root(trans, &quota_root->root_key);
1212 	if (ret) {
1213 		btrfs_abort_transaction(trans, ret);
1214 		goto out;
1215 	}
1216 
1217 	list_del(&quota_root->dirty_list);
1218 
1219 	btrfs_tree_lock(quota_root->node);
1220 	btrfs_clean_tree_block(quota_root->node);
1221 	btrfs_tree_unlock(quota_root->node);
1222 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1223 
1224 	btrfs_put_root(quota_root);
1225 
1226 out:
1227 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1228 	if (ret && trans)
1229 		btrfs_end_transaction(trans);
1230 	else if (trans)
1231 		ret = btrfs_end_transaction(trans);
1232 
1233 	return ret;
1234 }
1235 
1236 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1237 			 struct btrfs_qgroup *qgroup)
1238 {
1239 	if (list_empty(&qgroup->dirty))
1240 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1241 }
1242 
1243 /*
1244  * The easy accounting, we're updating qgroup relationship whose child qgroup
1245  * only has exclusive extents.
1246  *
1247  * In this case, all exclusive extents will also be exclusive for parent, so
1248  * excl/rfer just get added/removed.
1249  *
1250  * So is qgroup reservation space, which should also be added/removed to
1251  * parent.
1252  * Or when child tries to release reservation space, parent will underflow its
1253  * reservation (for relationship adding case).
1254  *
1255  * Caller should hold fs_info->qgroup_lock.
1256  */
1257 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1258 				    struct ulist *tmp, u64 ref_root,
1259 				    struct btrfs_qgroup *src, int sign)
1260 {
1261 	struct btrfs_qgroup *qgroup;
1262 	struct btrfs_qgroup_list *glist;
1263 	struct ulist_node *unode;
1264 	struct ulist_iterator uiter;
1265 	u64 num_bytes = src->excl;
1266 	int ret = 0;
1267 
1268 	qgroup = find_qgroup_rb(fs_info, ref_root);
1269 	if (!qgroup)
1270 		goto out;
1271 
1272 	qgroup->rfer += sign * num_bytes;
1273 	qgroup->rfer_cmpr += sign * num_bytes;
1274 
1275 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1276 	qgroup->excl += sign * num_bytes;
1277 	qgroup->excl_cmpr += sign * num_bytes;
1278 
1279 	if (sign > 0)
1280 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1281 	else
1282 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1283 
1284 	qgroup_dirty(fs_info, qgroup);
1285 
1286 	/* Get all of the parent groups that contain this qgroup */
1287 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1288 		ret = ulist_add(tmp, glist->group->qgroupid,
1289 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1290 		if (ret < 0)
1291 			goto out;
1292 	}
1293 
1294 	/* Iterate all of the parents and adjust their reference counts */
1295 	ULIST_ITER_INIT(&uiter);
1296 	while ((unode = ulist_next(tmp, &uiter))) {
1297 		qgroup = unode_aux_to_qgroup(unode);
1298 		qgroup->rfer += sign * num_bytes;
1299 		qgroup->rfer_cmpr += sign * num_bytes;
1300 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1301 		qgroup->excl += sign * num_bytes;
1302 		if (sign > 0)
1303 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1304 		else
1305 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1306 		qgroup->excl_cmpr += sign * num_bytes;
1307 		qgroup_dirty(fs_info, qgroup);
1308 
1309 		/* Add any parents of the parents */
1310 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1311 			ret = ulist_add(tmp, glist->group->qgroupid,
1312 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1313 			if (ret < 0)
1314 				goto out;
1315 		}
1316 	}
1317 	ret = 0;
1318 out:
1319 	return ret;
1320 }
1321 
1322 
1323 /*
1324  * Quick path for updating qgroup with only excl refs.
1325  *
1326  * In that case, just update all parent will be enough.
1327  * Or we needs to do a full rescan.
1328  * Caller should also hold fs_info->qgroup_lock.
1329  *
1330  * Return 0 for quick update, return >0 for need to full rescan
1331  * and mark INCONSISTENT flag.
1332  * Return < 0 for other error.
1333  */
1334 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1335 				   struct ulist *tmp, u64 src, u64 dst,
1336 				   int sign)
1337 {
1338 	struct btrfs_qgroup *qgroup;
1339 	int ret = 1;
1340 	int err = 0;
1341 
1342 	qgroup = find_qgroup_rb(fs_info, src);
1343 	if (!qgroup)
1344 		goto out;
1345 	if (qgroup->excl == qgroup->rfer) {
1346 		ret = 0;
1347 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1348 					       qgroup, sign);
1349 		if (err < 0) {
1350 			ret = err;
1351 			goto out;
1352 		}
1353 	}
1354 out:
1355 	if (ret)
1356 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1357 	return ret;
1358 }
1359 
1360 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1361 			      u64 dst)
1362 {
1363 	struct btrfs_fs_info *fs_info = trans->fs_info;
1364 	struct btrfs_qgroup *parent;
1365 	struct btrfs_qgroup *member;
1366 	struct btrfs_qgroup_list *list;
1367 	struct ulist *tmp;
1368 	unsigned int nofs_flag;
1369 	int ret = 0;
1370 
1371 	/* Check the level of src and dst first */
1372 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1373 		return -EINVAL;
1374 
1375 	/* We hold a transaction handle open, must do a NOFS allocation. */
1376 	nofs_flag = memalloc_nofs_save();
1377 	tmp = ulist_alloc(GFP_KERNEL);
1378 	memalloc_nofs_restore(nofs_flag);
1379 	if (!tmp)
1380 		return -ENOMEM;
1381 
1382 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1383 	if (!fs_info->quota_root) {
1384 		ret = -ENOTCONN;
1385 		goto out;
1386 	}
1387 	member = find_qgroup_rb(fs_info, src);
1388 	parent = find_qgroup_rb(fs_info, dst);
1389 	if (!member || !parent) {
1390 		ret = -EINVAL;
1391 		goto out;
1392 	}
1393 
1394 	/* check if such qgroup relation exist firstly */
1395 	list_for_each_entry(list, &member->groups, next_group) {
1396 		if (list->group == parent) {
1397 			ret = -EEXIST;
1398 			goto out;
1399 		}
1400 	}
1401 
1402 	ret = add_qgroup_relation_item(trans, src, dst);
1403 	if (ret)
1404 		goto out;
1405 
1406 	ret = add_qgroup_relation_item(trans, dst, src);
1407 	if (ret) {
1408 		del_qgroup_relation_item(trans, src, dst);
1409 		goto out;
1410 	}
1411 
1412 	spin_lock(&fs_info->qgroup_lock);
1413 	ret = add_relation_rb(fs_info, src, dst);
1414 	if (ret < 0) {
1415 		spin_unlock(&fs_info->qgroup_lock);
1416 		goto out;
1417 	}
1418 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1419 	spin_unlock(&fs_info->qgroup_lock);
1420 out:
1421 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1422 	ulist_free(tmp);
1423 	return ret;
1424 }
1425 
1426 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1427 				 u64 dst)
1428 {
1429 	struct btrfs_fs_info *fs_info = trans->fs_info;
1430 	struct btrfs_qgroup *parent;
1431 	struct btrfs_qgroup *member;
1432 	struct btrfs_qgroup_list *list;
1433 	struct ulist *tmp;
1434 	bool found = false;
1435 	unsigned int nofs_flag;
1436 	int ret = 0;
1437 	int ret2;
1438 
1439 	/* We hold a transaction handle open, must do a NOFS allocation. */
1440 	nofs_flag = memalloc_nofs_save();
1441 	tmp = ulist_alloc(GFP_KERNEL);
1442 	memalloc_nofs_restore(nofs_flag);
1443 	if (!tmp)
1444 		return -ENOMEM;
1445 
1446 	if (!fs_info->quota_root) {
1447 		ret = -ENOTCONN;
1448 		goto out;
1449 	}
1450 
1451 	member = find_qgroup_rb(fs_info, src);
1452 	parent = find_qgroup_rb(fs_info, dst);
1453 	/*
1454 	 * The parent/member pair doesn't exist, then try to delete the dead
1455 	 * relation items only.
1456 	 */
1457 	if (!member || !parent)
1458 		goto delete_item;
1459 
1460 	/* check if such qgroup relation exist firstly */
1461 	list_for_each_entry(list, &member->groups, next_group) {
1462 		if (list->group == parent) {
1463 			found = true;
1464 			break;
1465 		}
1466 	}
1467 
1468 delete_item:
1469 	ret = del_qgroup_relation_item(trans, src, dst);
1470 	if (ret < 0 && ret != -ENOENT)
1471 		goto out;
1472 	ret2 = del_qgroup_relation_item(trans, dst, src);
1473 	if (ret2 < 0 && ret2 != -ENOENT)
1474 		goto out;
1475 
1476 	/* At least one deletion succeeded, return 0 */
1477 	if (!ret || !ret2)
1478 		ret = 0;
1479 
1480 	if (found) {
1481 		spin_lock(&fs_info->qgroup_lock);
1482 		del_relation_rb(fs_info, src, dst);
1483 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1484 		spin_unlock(&fs_info->qgroup_lock);
1485 	}
1486 out:
1487 	ulist_free(tmp);
1488 	return ret;
1489 }
1490 
1491 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1492 			      u64 dst)
1493 {
1494 	struct btrfs_fs_info *fs_info = trans->fs_info;
1495 	int ret = 0;
1496 
1497 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1498 	ret = __del_qgroup_relation(trans, src, dst);
1499 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1500 
1501 	return ret;
1502 }
1503 
1504 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1505 {
1506 	struct btrfs_fs_info *fs_info = trans->fs_info;
1507 	struct btrfs_root *quota_root;
1508 	struct btrfs_qgroup *qgroup;
1509 	int ret = 0;
1510 
1511 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1512 	if (!fs_info->quota_root) {
1513 		ret = -ENOTCONN;
1514 		goto out;
1515 	}
1516 	quota_root = fs_info->quota_root;
1517 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1518 	if (qgroup) {
1519 		ret = -EEXIST;
1520 		goto out;
1521 	}
1522 
1523 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1524 	if (ret)
1525 		goto out;
1526 
1527 	spin_lock(&fs_info->qgroup_lock);
1528 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1529 	spin_unlock(&fs_info->qgroup_lock);
1530 
1531 	if (IS_ERR(qgroup)) {
1532 		ret = PTR_ERR(qgroup);
1533 		goto out;
1534 	}
1535 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1536 out:
1537 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1538 	return ret;
1539 }
1540 
1541 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1542 {
1543 	struct btrfs_fs_info *fs_info = trans->fs_info;
1544 	struct btrfs_qgroup *qgroup;
1545 	struct btrfs_qgroup_list *list;
1546 	int ret = 0;
1547 
1548 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1549 	if (!fs_info->quota_root) {
1550 		ret = -ENOTCONN;
1551 		goto out;
1552 	}
1553 
1554 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1555 	if (!qgroup) {
1556 		ret = -ENOENT;
1557 		goto out;
1558 	}
1559 
1560 	/* Check if there are no children of this qgroup */
1561 	if (!list_empty(&qgroup->members)) {
1562 		ret = -EBUSY;
1563 		goto out;
1564 	}
1565 
1566 	ret = del_qgroup_item(trans, qgroupid);
1567 	if (ret && ret != -ENOENT)
1568 		goto out;
1569 
1570 	while (!list_empty(&qgroup->groups)) {
1571 		list = list_first_entry(&qgroup->groups,
1572 					struct btrfs_qgroup_list, next_group);
1573 		ret = __del_qgroup_relation(trans, qgroupid,
1574 					    list->group->qgroupid);
1575 		if (ret)
1576 			goto out;
1577 	}
1578 
1579 	spin_lock(&fs_info->qgroup_lock);
1580 	del_qgroup_rb(fs_info, qgroupid);
1581 	spin_unlock(&fs_info->qgroup_lock);
1582 
1583 	/*
1584 	 * Remove the qgroup from sysfs now without holding the qgroup_lock
1585 	 * spinlock, since the sysfs_remove_group() function needs to take
1586 	 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1587 	 */
1588 	btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1589 	kfree(qgroup);
1590 out:
1591 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1592 	return ret;
1593 }
1594 
1595 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1596 		       struct btrfs_qgroup_limit *limit)
1597 {
1598 	struct btrfs_fs_info *fs_info = trans->fs_info;
1599 	struct btrfs_qgroup *qgroup;
1600 	int ret = 0;
1601 	/* Sometimes we would want to clear the limit on this qgroup.
1602 	 * To meet this requirement, we treat the -1 as a special value
1603 	 * which tell kernel to clear the limit on this qgroup.
1604 	 */
1605 	const u64 CLEAR_VALUE = -1;
1606 
1607 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1608 	if (!fs_info->quota_root) {
1609 		ret = -ENOTCONN;
1610 		goto out;
1611 	}
1612 
1613 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1614 	if (!qgroup) {
1615 		ret = -ENOENT;
1616 		goto out;
1617 	}
1618 
1619 	spin_lock(&fs_info->qgroup_lock);
1620 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1621 		if (limit->max_rfer == CLEAR_VALUE) {
1622 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1623 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1624 			qgroup->max_rfer = 0;
1625 		} else {
1626 			qgroup->max_rfer = limit->max_rfer;
1627 		}
1628 	}
1629 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1630 		if (limit->max_excl == CLEAR_VALUE) {
1631 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1632 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1633 			qgroup->max_excl = 0;
1634 		} else {
1635 			qgroup->max_excl = limit->max_excl;
1636 		}
1637 	}
1638 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1639 		if (limit->rsv_rfer == CLEAR_VALUE) {
1640 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1641 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1642 			qgroup->rsv_rfer = 0;
1643 		} else {
1644 			qgroup->rsv_rfer = limit->rsv_rfer;
1645 		}
1646 	}
1647 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1648 		if (limit->rsv_excl == CLEAR_VALUE) {
1649 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1650 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1651 			qgroup->rsv_excl = 0;
1652 		} else {
1653 			qgroup->rsv_excl = limit->rsv_excl;
1654 		}
1655 	}
1656 	qgroup->lim_flags |= limit->flags;
1657 
1658 	spin_unlock(&fs_info->qgroup_lock);
1659 
1660 	ret = update_qgroup_limit_item(trans, qgroup);
1661 	if (ret) {
1662 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1663 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1664 		       qgroupid);
1665 	}
1666 
1667 out:
1668 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1669 	return ret;
1670 }
1671 
1672 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1673 				struct btrfs_delayed_ref_root *delayed_refs,
1674 				struct btrfs_qgroup_extent_record *record)
1675 {
1676 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1677 	struct rb_node *parent_node = NULL;
1678 	struct btrfs_qgroup_extent_record *entry;
1679 	u64 bytenr = record->bytenr;
1680 
1681 	lockdep_assert_held(&delayed_refs->lock);
1682 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1683 
1684 	while (*p) {
1685 		parent_node = *p;
1686 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1687 				 node);
1688 		if (bytenr < entry->bytenr) {
1689 			p = &(*p)->rb_left;
1690 		} else if (bytenr > entry->bytenr) {
1691 			p = &(*p)->rb_right;
1692 		} else {
1693 			if (record->data_rsv && !entry->data_rsv) {
1694 				entry->data_rsv = record->data_rsv;
1695 				entry->data_rsv_refroot =
1696 					record->data_rsv_refroot;
1697 			}
1698 			return 1;
1699 		}
1700 	}
1701 
1702 	rb_link_node(&record->node, parent_node, p);
1703 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1704 	return 0;
1705 }
1706 
1707 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1708 				   struct btrfs_qgroup_extent_record *qrecord)
1709 {
1710 	struct ulist *old_root;
1711 	u64 bytenr = qrecord->bytenr;
1712 	int ret;
1713 
1714 	/*
1715 	 * We are always called in a context where we are already holding a
1716 	 * transaction handle. Often we are called when adding a data delayed
1717 	 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1718 	 * in which case we will be holding a write lock on extent buffer from a
1719 	 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1720 	 * acquire fs_info->commit_root_sem, because that is a higher level lock
1721 	 * that must be acquired before locking any extent buffers.
1722 	 *
1723 	 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1724 	 * but we can't pass it a non-NULL transaction handle, because otherwise
1725 	 * it would not use commit roots and would lock extent buffers, causing
1726 	 * a deadlock if it ends up trying to read lock the same extent buffer
1727 	 * that was previously write locked at btrfs_truncate_inode_items().
1728 	 *
1729 	 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1730 	 * explicitly tell it to not acquire the commit_root_sem - if we are
1731 	 * holding a transaction handle we don't need its protection.
1732 	 */
1733 	ASSERT(trans != NULL);
1734 
1735 	ret = btrfs_find_all_roots(NULL, trans->fs_info, bytenr, 0, &old_root,
1736 				   false, true);
1737 	if (ret < 0) {
1738 		trans->fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1739 		btrfs_warn(trans->fs_info,
1740 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1741 			ret);
1742 		return 0;
1743 	}
1744 
1745 	/*
1746 	 * Here we don't need to get the lock of
1747 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1748 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1749 	 *
1750 	 * So modifying qrecord->old_roots is safe here
1751 	 */
1752 	qrecord->old_roots = old_root;
1753 	return 0;
1754 }
1755 
1756 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1757 			      u64 num_bytes, gfp_t gfp_flag)
1758 {
1759 	struct btrfs_fs_info *fs_info = trans->fs_info;
1760 	struct btrfs_qgroup_extent_record *record;
1761 	struct btrfs_delayed_ref_root *delayed_refs;
1762 	int ret;
1763 
1764 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1765 	    || bytenr == 0 || num_bytes == 0)
1766 		return 0;
1767 	record = kzalloc(sizeof(*record), gfp_flag);
1768 	if (!record)
1769 		return -ENOMEM;
1770 
1771 	delayed_refs = &trans->transaction->delayed_refs;
1772 	record->bytenr = bytenr;
1773 	record->num_bytes = num_bytes;
1774 	record->old_roots = NULL;
1775 
1776 	spin_lock(&delayed_refs->lock);
1777 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1778 	spin_unlock(&delayed_refs->lock);
1779 	if (ret > 0) {
1780 		kfree(record);
1781 		return 0;
1782 	}
1783 	return btrfs_qgroup_trace_extent_post(trans, record);
1784 }
1785 
1786 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1787 				  struct extent_buffer *eb)
1788 {
1789 	struct btrfs_fs_info *fs_info = trans->fs_info;
1790 	int nr = btrfs_header_nritems(eb);
1791 	int i, extent_type, ret;
1792 	struct btrfs_key key;
1793 	struct btrfs_file_extent_item *fi;
1794 	u64 bytenr, num_bytes;
1795 
1796 	/* We can be called directly from walk_up_proc() */
1797 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1798 		return 0;
1799 
1800 	for (i = 0; i < nr; i++) {
1801 		btrfs_item_key_to_cpu(eb, &key, i);
1802 
1803 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1804 			continue;
1805 
1806 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1807 		/* filter out non qgroup-accountable extents  */
1808 		extent_type = btrfs_file_extent_type(eb, fi);
1809 
1810 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1811 			continue;
1812 
1813 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1814 		if (!bytenr)
1815 			continue;
1816 
1817 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1818 
1819 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1820 						GFP_NOFS);
1821 		if (ret)
1822 			return ret;
1823 	}
1824 	cond_resched();
1825 	return 0;
1826 }
1827 
1828 /*
1829  * Walk up the tree from the bottom, freeing leaves and any interior
1830  * nodes which have had all slots visited. If a node (leaf or
1831  * interior) is freed, the node above it will have it's slot
1832  * incremented. The root node will never be freed.
1833  *
1834  * At the end of this function, we should have a path which has all
1835  * slots incremented to the next position for a search. If we need to
1836  * read a new node it will be NULL and the node above it will have the
1837  * correct slot selected for a later read.
1838  *
1839  * If we increment the root nodes slot counter past the number of
1840  * elements, 1 is returned to signal completion of the search.
1841  */
1842 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1843 {
1844 	int level = 0;
1845 	int nr, slot;
1846 	struct extent_buffer *eb;
1847 
1848 	if (root_level == 0)
1849 		return 1;
1850 
1851 	while (level <= root_level) {
1852 		eb = path->nodes[level];
1853 		nr = btrfs_header_nritems(eb);
1854 		path->slots[level]++;
1855 		slot = path->slots[level];
1856 		if (slot >= nr || level == 0) {
1857 			/*
1858 			 * Don't free the root -  we will detect this
1859 			 * condition after our loop and return a
1860 			 * positive value for caller to stop walking the tree.
1861 			 */
1862 			if (level != root_level) {
1863 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1864 				path->locks[level] = 0;
1865 
1866 				free_extent_buffer(eb);
1867 				path->nodes[level] = NULL;
1868 				path->slots[level] = 0;
1869 			}
1870 		} else {
1871 			/*
1872 			 * We have a valid slot to walk back down
1873 			 * from. Stop here so caller can process these
1874 			 * new nodes.
1875 			 */
1876 			break;
1877 		}
1878 
1879 		level++;
1880 	}
1881 
1882 	eb = path->nodes[root_level];
1883 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1884 		return 1;
1885 
1886 	return 0;
1887 }
1888 
1889 /*
1890  * Helper function to trace a subtree tree block swap.
1891  *
1892  * The swap will happen in highest tree block, but there may be a lot of
1893  * tree blocks involved.
1894  *
1895  * For example:
1896  *  OO = Old tree blocks
1897  *  NN = New tree blocks allocated during balance
1898  *
1899  *           File tree (257)                  Reloc tree for 257
1900  * L2              OO                                NN
1901  *               /    \                            /    \
1902  * L1          OO      OO (a)                    OO      NN (a)
1903  *            / \     / \                       / \     / \
1904  * L0       OO   OO OO   OO                   OO   OO NN   NN
1905  *                  (b)  (c)                          (b)  (c)
1906  *
1907  * When calling qgroup_trace_extent_swap(), we will pass:
1908  * @src_eb = OO(a)
1909  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1910  * @dst_level = 0
1911  * @root_level = 1
1912  *
1913  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1914  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1915  *
1916  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1917  *
1918  * 1) Tree search from @src_eb
1919  *    It should acts as a simplified btrfs_search_slot().
1920  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1921  *    (first key).
1922  *
1923  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1924  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1925  *    They should be marked during previous (@dst_level = 1) iteration.
1926  *
1927  * 3) Mark file extents in leaves dirty
1928  *    We don't have good way to pick out new file extents only.
1929  *    So we still follow the old method by scanning all file extents in
1930  *    the leave.
1931  *
1932  * This function can free us from keeping two paths, thus later we only need
1933  * to care about how to iterate all new tree blocks in reloc tree.
1934  */
1935 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1936 				    struct extent_buffer *src_eb,
1937 				    struct btrfs_path *dst_path,
1938 				    int dst_level, int root_level,
1939 				    bool trace_leaf)
1940 {
1941 	struct btrfs_key key;
1942 	struct btrfs_path *src_path;
1943 	struct btrfs_fs_info *fs_info = trans->fs_info;
1944 	u32 nodesize = fs_info->nodesize;
1945 	int cur_level = root_level;
1946 	int ret;
1947 
1948 	BUG_ON(dst_level > root_level);
1949 	/* Level mismatch */
1950 	if (btrfs_header_level(src_eb) != root_level)
1951 		return -EINVAL;
1952 
1953 	src_path = btrfs_alloc_path();
1954 	if (!src_path) {
1955 		ret = -ENOMEM;
1956 		goto out;
1957 	}
1958 
1959 	if (dst_level)
1960 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1961 	else
1962 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1963 
1964 	/* For src_path */
1965 	atomic_inc(&src_eb->refs);
1966 	src_path->nodes[root_level] = src_eb;
1967 	src_path->slots[root_level] = dst_path->slots[root_level];
1968 	src_path->locks[root_level] = 0;
1969 
1970 	/* A simplified version of btrfs_search_slot() */
1971 	while (cur_level >= dst_level) {
1972 		struct btrfs_key src_key;
1973 		struct btrfs_key dst_key;
1974 
1975 		if (src_path->nodes[cur_level] == NULL) {
1976 			struct extent_buffer *eb;
1977 			int parent_slot;
1978 
1979 			eb = src_path->nodes[cur_level + 1];
1980 			parent_slot = src_path->slots[cur_level + 1];
1981 
1982 			eb = btrfs_read_node_slot(eb, parent_slot);
1983 			if (IS_ERR(eb)) {
1984 				ret = PTR_ERR(eb);
1985 				goto out;
1986 			}
1987 
1988 			src_path->nodes[cur_level] = eb;
1989 
1990 			btrfs_tree_read_lock(eb);
1991 			src_path->locks[cur_level] = BTRFS_READ_LOCK;
1992 		}
1993 
1994 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1995 		if (cur_level) {
1996 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1997 					&dst_key, dst_path->slots[cur_level]);
1998 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1999 					&src_key, src_path->slots[cur_level]);
2000 		} else {
2001 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2002 					&dst_key, dst_path->slots[cur_level]);
2003 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2004 					&src_key, src_path->slots[cur_level]);
2005 		}
2006 		/* Content mismatch, something went wrong */
2007 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2008 			ret = -ENOENT;
2009 			goto out;
2010 		}
2011 		cur_level--;
2012 	}
2013 
2014 	/*
2015 	 * Now both @dst_path and @src_path have been populated, record the tree
2016 	 * blocks for qgroup accounting.
2017 	 */
2018 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2019 			nodesize, GFP_NOFS);
2020 	if (ret < 0)
2021 		goto out;
2022 	ret = btrfs_qgroup_trace_extent(trans,
2023 			dst_path->nodes[dst_level]->start,
2024 			nodesize, GFP_NOFS);
2025 	if (ret < 0)
2026 		goto out;
2027 
2028 	/* Record leaf file extents */
2029 	if (dst_level == 0 && trace_leaf) {
2030 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2031 		if (ret < 0)
2032 			goto out;
2033 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2034 	}
2035 out:
2036 	btrfs_free_path(src_path);
2037 	return ret;
2038 }
2039 
2040 /*
2041  * Helper function to do recursive generation-aware depth-first search, to
2042  * locate all new tree blocks in a subtree of reloc tree.
2043  *
2044  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2045  *         reloc tree
2046  * L2         NN (a)
2047  *          /    \
2048  * L1    OO        NN (b)
2049  *      /  \      /  \
2050  * L0  OO  OO    OO  NN
2051  *               (c) (d)
2052  * If we pass:
2053  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2054  * @cur_level = 1
2055  * @root_level = 1
2056  *
2057  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2058  * above tree blocks along with their counter parts in file tree.
2059  * While during search, old tree blocks OO(c) will be skipped as tree block swap
2060  * won't affect OO(c).
2061  */
2062 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2063 					   struct extent_buffer *src_eb,
2064 					   struct btrfs_path *dst_path,
2065 					   int cur_level, int root_level,
2066 					   u64 last_snapshot, bool trace_leaf)
2067 {
2068 	struct btrfs_fs_info *fs_info = trans->fs_info;
2069 	struct extent_buffer *eb;
2070 	bool need_cleanup = false;
2071 	int ret = 0;
2072 	int i;
2073 
2074 	/* Level sanity check */
2075 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2076 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2077 	    root_level < cur_level) {
2078 		btrfs_err_rl(fs_info,
2079 			"%s: bad levels, cur_level=%d root_level=%d",
2080 			__func__, cur_level, root_level);
2081 		return -EUCLEAN;
2082 	}
2083 
2084 	/* Read the tree block if needed */
2085 	if (dst_path->nodes[cur_level] == NULL) {
2086 		int parent_slot;
2087 		u64 child_gen;
2088 
2089 		/*
2090 		 * dst_path->nodes[root_level] must be initialized before
2091 		 * calling this function.
2092 		 */
2093 		if (cur_level == root_level) {
2094 			btrfs_err_rl(fs_info,
2095 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2096 				__func__, root_level, root_level, cur_level);
2097 			return -EUCLEAN;
2098 		}
2099 
2100 		/*
2101 		 * We need to get child blockptr/gen from parent before we can
2102 		 * read it.
2103 		  */
2104 		eb = dst_path->nodes[cur_level + 1];
2105 		parent_slot = dst_path->slots[cur_level + 1];
2106 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2107 
2108 		/* This node is old, no need to trace */
2109 		if (child_gen < last_snapshot)
2110 			goto out;
2111 
2112 		eb = btrfs_read_node_slot(eb, parent_slot);
2113 		if (IS_ERR(eb)) {
2114 			ret = PTR_ERR(eb);
2115 			goto out;
2116 		}
2117 
2118 		dst_path->nodes[cur_level] = eb;
2119 		dst_path->slots[cur_level] = 0;
2120 
2121 		btrfs_tree_read_lock(eb);
2122 		dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2123 		need_cleanup = true;
2124 	}
2125 
2126 	/* Now record this tree block and its counter part for qgroups */
2127 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2128 				       root_level, trace_leaf);
2129 	if (ret < 0)
2130 		goto cleanup;
2131 
2132 	eb = dst_path->nodes[cur_level];
2133 
2134 	if (cur_level > 0) {
2135 		/* Iterate all child tree blocks */
2136 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2137 			/* Skip old tree blocks as they won't be swapped */
2138 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2139 				continue;
2140 			dst_path->slots[cur_level] = i;
2141 
2142 			/* Recursive call (at most 7 times) */
2143 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2144 					dst_path, cur_level - 1, root_level,
2145 					last_snapshot, trace_leaf);
2146 			if (ret < 0)
2147 				goto cleanup;
2148 		}
2149 	}
2150 
2151 cleanup:
2152 	if (need_cleanup) {
2153 		/* Clean up */
2154 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2155 				     dst_path->locks[cur_level]);
2156 		free_extent_buffer(dst_path->nodes[cur_level]);
2157 		dst_path->nodes[cur_level] = NULL;
2158 		dst_path->slots[cur_level] = 0;
2159 		dst_path->locks[cur_level] = 0;
2160 	}
2161 out:
2162 	return ret;
2163 }
2164 
2165 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2166 				struct extent_buffer *src_eb,
2167 				struct extent_buffer *dst_eb,
2168 				u64 last_snapshot, bool trace_leaf)
2169 {
2170 	struct btrfs_fs_info *fs_info = trans->fs_info;
2171 	struct btrfs_path *dst_path = NULL;
2172 	int level;
2173 	int ret;
2174 
2175 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2176 		return 0;
2177 
2178 	/* Wrong parameter order */
2179 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2180 		btrfs_err_rl(fs_info,
2181 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2182 			     btrfs_header_generation(src_eb),
2183 			     btrfs_header_generation(dst_eb));
2184 		return -EUCLEAN;
2185 	}
2186 
2187 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2188 		ret = -EIO;
2189 		goto out;
2190 	}
2191 
2192 	level = btrfs_header_level(dst_eb);
2193 	dst_path = btrfs_alloc_path();
2194 	if (!dst_path) {
2195 		ret = -ENOMEM;
2196 		goto out;
2197 	}
2198 	/* For dst_path */
2199 	atomic_inc(&dst_eb->refs);
2200 	dst_path->nodes[level] = dst_eb;
2201 	dst_path->slots[level] = 0;
2202 	dst_path->locks[level] = 0;
2203 
2204 	/* Do the generation aware breadth-first search */
2205 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2206 					      level, last_snapshot, trace_leaf);
2207 	if (ret < 0)
2208 		goto out;
2209 	ret = 0;
2210 
2211 out:
2212 	btrfs_free_path(dst_path);
2213 	if (ret < 0)
2214 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2215 	return ret;
2216 }
2217 
2218 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2219 			       struct extent_buffer *root_eb,
2220 			       u64 root_gen, int root_level)
2221 {
2222 	struct btrfs_fs_info *fs_info = trans->fs_info;
2223 	int ret = 0;
2224 	int level;
2225 	struct extent_buffer *eb = root_eb;
2226 	struct btrfs_path *path = NULL;
2227 
2228 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2229 	BUG_ON(root_eb == NULL);
2230 
2231 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2232 		return 0;
2233 
2234 	if (!extent_buffer_uptodate(root_eb)) {
2235 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2236 		if (ret)
2237 			goto out;
2238 	}
2239 
2240 	if (root_level == 0) {
2241 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2242 		goto out;
2243 	}
2244 
2245 	path = btrfs_alloc_path();
2246 	if (!path)
2247 		return -ENOMEM;
2248 
2249 	/*
2250 	 * Walk down the tree.  Missing extent blocks are filled in as
2251 	 * we go. Metadata is accounted every time we read a new
2252 	 * extent block.
2253 	 *
2254 	 * When we reach a leaf, we account for file extent items in it,
2255 	 * walk back up the tree (adjusting slot pointers as we go)
2256 	 * and restart the search process.
2257 	 */
2258 	atomic_inc(&root_eb->refs);	/* For path */
2259 	path->nodes[root_level] = root_eb;
2260 	path->slots[root_level] = 0;
2261 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2262 walk_down:
2263 	level = root_level;
2264 	while (level >= 0) {
2265 		if (path->nodes[level] == NULL) {
2266 			int parent_slot;
2267 			u64 child_bytenr;
2268 
2269 			/*
2270 			 * We need to get child blockptr from parent before we
2271 			 * can read it.
2272 			  */
2273 			eb = path->nodes[level + 1];
2274 			parent_slot = path->slots[level + 1];
2275 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2276 
2277 			eb = btrfs_read_node_slot(eb, parent_slot);
2278 			if (IS_ERR(eb)) {
2279 				ret = PTR_ERR(eb);
2280 				goto out;
2281 			}
2282 
2283 			path->nodes[level] = eb;
2284 			path->slots[level] = 0;
2285 
2286 			btrfs_tree_read_lock(eb);
2287 			path->locks[level] = BTRFS_READ_LOCK;
2288 
2289 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2290 							fs_info->nodesize,
2291 							GFP_NOFS);
2292 			if (ret)
2293 				goto out;
2294 		}
2295 
2296 		if (level == 0) {
2297 			ret = btrfs_qgroup_trace_leaf_items(trans,
2298 							    path->nodes[level]);
2299 			if (ret)
2300 				goto out;
2301 
2302 			/* Nonzero return here means we completed our search */
2303 			ret = adjust_slots_upwards(path, root_level);
2304 			if (ret)
2305 				break;
2306 
2307 			/* Restart search with new slots */
2308 			goto walk_down;
2309 		}
2310 
2311 		level--;
2312 	}
2313 
2314 	ret = 0;
2315 out:
2316 	btrfs_free_path(path);
2317 
2318 	return ret;
2319 }
2320 
2321 #define UPDATE_NEW	0
2322 #define UPDATE_OLD	1
2323 /*
2324  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2325  */
2326 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2327 				struct ulist *roots, struct ulist *tmp,
2328 				struct ulist *qgroups, u64 seq, int update_old)
2329 {
2330 	struct ulist_node *unode;
2331 	struct ulist_iterator uiter;
2332 	struct ulist_node *tmp_unode;
2333 	struct ulist_iterator tmp_uiter;
2334 	struct btrfs_qgroup *qg;
2335 	int ret = 0;
2336 
2337 	if (!roots)
2338 		return 0;
2339 	ULIST_ITER_INIT(&uiter);
2340 	while ((unode = ulist_next(roots, &uiter))) {
2341 		qg = find_qgroup_rb(fs_info, unode->val);
2342 		if (!qg)
2343 			continue;
2344 
2345 		ulist_reinit(tmp);
2346 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2347 				GFP_ATOMIC);
2348 		if (ret < 0)
2349 			return ret;
2350 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2351 		if (ret < 0)
2352 			return ret;
2353 		ULIST_ITER_INIT(&tmp_uiter);
2354 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2355 			struct btrfs_qgroup_list *glist;
2356 
2357 			qg = unode_aux_to_qgroup(tmp_unode);
2358 			if (update_old)
2359 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2360 			else
2361 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2362 			list_for_each_entry(glist, &qg->groups, next_group) {
2363 				ret = ulist_add(qgroups, glist->group->qgroupid,
2364 						qgroup_to_aux(glist->group),
2365 						GFP_ATOMIC);
2366 				if (ret < 0)
2367 					return ret;
2368 				ret = ulist_add(tmp, glist->group->qgroupid,
2369 						qgroup_to_aux(glist->group),
2370 						GFP_ATOMIC);
2371 				if (ret < 0)
2372 					return ret;
2373 			}
2374 		}
2375 	}
2376 	return 0;
2377 }
2378 
2379 /*
2380  * Update qgroup rfer/excl counters.
2381  * Rfer update is easy, codes can explain themselves.
2382  *
2383  * Excl update is tricky, the update is split into 2 parts.
2384  * Part 1: Possible exclusive <-> sharing detect:
2385  *	|	A	|	!A	|
2386  *  -------------------------------------
2387  *  B	|	*	|	-	|
2388  *  -------------------------------------
2389  *  !B	|	+	|	**	|
2390  *  -------------------------------------
2391  *
2392  * Conditions:
2393  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2394  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2395  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2396  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2397  *
2398  * Results:
2399  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2400  * *: Definitely not changed.		**: Possible unchanged.
2401  *
2402  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2403  *
2404  * To make the logic clear, we first use condition A and B to split
2405  * combination into 4 results.
2406  *
2407  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2408  * only on variant maybe 0.
2409  *
2410  * Lastly, check result **, since there are 2 variants maybe 0, split them
2411  * again(2x2).
2412  * But this time we don't need to consider other things, the codes and logic
2413  * is easy to understand now.
2414  */
2415 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2416 				  struct ulist *qgroups,
2417 				  u64 nr_old_roots,
2418 				  u64 nr_new_roots,
2419 				  u64 num_bytes, u64 seq)
2420 {
2421 	struct ulist_node *unode;
2422 	struct ulist_iterator uiter;
2423 	struct btrfs_qgroup *qg;
2424 	u64 cur_new_count, cur_old_count;
2425 
2426 	ULIST_ITER_INIT(&uiter);
2427 	while ((unode = ulist_next(qgroups, &uiter))) {
2428 		bool dirty = false;
2429 
2430 		qg = unode_aux_to_qgroup(unode);
2431 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2432 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2433 
2434 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2435 					     cur_new_count);
2436 
2437 		/* Rfer update part */
2438 		if (cur_old_count == 0 && cur_new_count > 0) {
2439 			qg->rfer += num_bytes;
2440 			qg->rfer_cmpr += num_bytes;
2441 			dirty = true;
2442 		}
2443 		if (cur_old_count > 0 && cur_new_count == 0) {
2444 			qg->rfer -= num_bytes;
2445 			qg->rfer_cmpr -= num_bytes;
2446 			dirty = true;
2447 		}
2448 
2449 		/* Excl update part */
2450 		/* Exclusive/none -> shared case */
2451 		if (cur_old_count == nr_old_roots &&
2452 		    cur_new_count < nr_new_roots) {
2453 			/* Exclusive -> shared */
2454 			if (cur_old_count != 0) {
2455 				qg->excl -= num_bytes;
2456 				qg->excl_cmpr -= num_bytes;
2457 				dirty = true;
2458 			}
2459 		}
2460 
2461 		/* Shared -> exclusive/none case */
2462 		if (cur_old_count < nr_old_roots &&
2463 		    cur_new_count == nr_new_roots) {
2464 			/* Shared->exclusive */
2465 			if (cur_new_count != 0) {
2466 				qg->excl += num_bytes;
2467 				qg->excl_cmpr += num_bytes;
2468 				dirty = true;
2469 			}
2470 		}
2471 
2472 		/* Exclusive/none -> exclusive/none case */
2473 		if (cur_old_count == nr_old_roots &&
2474 		    cur_new_count == nr_new_roots) {
2475 			if (cur_old_count == 0) {
2476 				/* None -> exclusive/none */
2477 
2478 				if (cur_new_count != 0) {
2479 					/* None -> exclusive */
2480 					qg->excl += num_bytes;
2481 					qg->excl_cmpr += num_bytes;
2482 					dirty = true;
2483 				}
2484 				/* None -> none, nothing changed */
2485 			} else {
2486 				/* Exclusive -> exclusive/none */
2487 
2488 				if (cur_new_count == 0) {
2489 					/* Exclusive -> none */
2490 					qg->excl -= num_bytes;
2491 					qg->excl_cmpr -= num_bytes;
2492 					dirty = true;
2493 				}
2494 				/* Exclusive -> exclusive, nothing changed */
2495 			}
2496 		}
2497 
2498 		if (dirty)
2499 			qgroup_dirty(fs_info, qg);
2500 	}
2501 	return 0;
2502 }
2503 
2504 /*
2505  * Check if the @roots potentially is a list of fs tree roots
2506  *
2507  * Return 0 for definitely not a fs/subvol tree roots ulist
2508  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2509  *          one as well)
2510  */
2511 static int maybe_fs_roots(struct ulist *roots)
2512 {
2513 	struct ulist_node *unode;
2514 	struct ulist_iterator uiter;
2515 
2516 	/* Empty one, still possible for fs roots */
2517 	if (!roots || roots->nnodes == 0)
2518 		return 1;
2519 
2520 	ULIST_ITER_INIT(&uiter);
2521 	unode = ulist_next(roots, &uiter);
2522 	if (!unode)
2523 		return 1;
2524 
2525 	/*
2526 	 * If it contains fs tree roots, then it must belong to fs/subvol
2527 	 * trees.
2528 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2529 	 */
2530 	return is_fstree(unode->val);
2531 }
2532 
2533 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2534 				u64 num_bytes, struct ulist *old_roots,
2535 				struct ulist *new_roots)
2536 {
2537 	struct btrfs_fs_info *fs_info = trans->fs_info;
2538 	struct ulist *qgroups = NULL;
2539 	struct ulist *tmp = NULL;
2540 	u64 seq;
2541 	u64 nr_new_roots = 0;
2542 	u64 nr_old_roots = 0;
2543 	int ret = 0;
2544 
2545 	/*
2546 	 * If quotas get disabled meanwhile, the resources need to be freed and
2547 	 * we can't just exit here.
2548 	 */
2549 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2550 		goto out_free;
2551 
2552 	if (new_roots) {
2553 		if (!maybe_fs_roots(new_roots))
2554 			goto out_free;
2555 		nr_new_roots = new_roots->nnodes;
2556 	}
2557 	if (old_roots) {
2558 		if (!maybe_fs_roots(old_roots))
2559 			goto out_free;
2560 		nr_old_roots = old_roots->nnodes;
2561 	}
2562 
2563 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2564 	if (nr_old_roots == 0 && nr_new_roots == 0)
2565 		goto out_free;
2566 
2567 	BUG_ON(!fs_info->quota_root);
2568 
2569 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2570 					num_bytes, nr_old_roots, nr_new_roots);
2571 
2572 	qgroups = ulist_alloc(GFP_NOFS);
2573 	if (!qgroups) {
2574 		ret = -ENOMEM;
2575 		goto out_free;
2576 	}
2577 	tmp = ulist_alloc(GFP_NOFS);
2578 	if (!tmp) {
2579 		ret = -ENOMEM;
2580 		goto out_free;
2581 	}
2582 
2583 	mutex_lock(&fs_info->qgroup_rescan_lock);
2584 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2585 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2586 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2587 			ret = 0;
2588 			goto out_free;
2589 		}
2590 	}
2591 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2592 
2593 	spin_lock(&fs_info->qgroup_lock);
2594 	seq = fs_info->qgroup_seq;
2595 
2596 	/* Update old refcnts using old_roots */
2597 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2598 				   UPDATE_OLD);
2599 	if (ret < 0)
2600 		goto out;
2601 
2602 	/* Update new refcnts using new_roots */
2603 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2604 				   UPDATE_NEW);
2605 	if (ret < 0)
2606 		goto out;
2607 
2608 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2609 			       num_bytes, seq);
2610 
2611 	/*
2612 	 * Bump qgroup_seq to avoid seq overlap
2613 	 */
2614 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2615 out:
2616 	spin_unlock(&fs_info->qgroup_lock);
2617 out_free:
2618 	ulist_free(tmp);
2619 	ulist_free(qgroups);
2620 	ulist_free(old_roots);
2621 	ulist_free(new_roots);
2622 	return ret;
2623 }
2624 
2625 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2626 {
2627 	struct btrfs_fs_info *fs_info = trans->fs_info;
2628 	struct btrfs_qgroup_extent_record *record;
2629 	struct btrfs_delayed_ref_root *delayed_refs;
2630 	struct ulist *new_roots = NULL;
2631 	struct rb_node *node;
2632 	u64 num_dirty_extents = 0;
2633 	u64 qgroup_to_skip;
2634 	int ret = 0;
2635 
2636 	delayed_refs = &trans->transaction->delayed_refs;
2637 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2638 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2639 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2640 				  node);
2641 
2642 		num_dirty_extents++;
2643 		trace_btrfs_qgroup_account_extents(fs_info, record);
2644 
2645 		if (!ret) {
2646 			/*
2647 			 * Old roots should be searched when inserting qgroup
2648 			 * extent record
2649 			 */
2650 			if (WARN_ON(!record->old_roots)) {
2651 				/* Search commit root to find old_roots */
2652 				ret = btrfs_find_all_roots(NULL, fs_info,
2653 						record->bytenr, 0,
2654 						&record->old_roots, false, false);
2655 				if (ret < 0)
2656 					goto cleanup;
2657 			}
2658 
2659 			/* Free the reserved data space */
2660 			btrfs_qgroup_free_refroot(fs_info,
2661 					record->data_rsv_refroot,
2662 					record->data_rsv,
2663 					BTRFS_QGROUP_RSV_DATA);
2664 			/*
2665 			 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2666 			 * which doesn't lock tree or delayed_refs and search
2667 			 * current root. It's safe inside commit_transaction().
2668 			 */
2669 			ret = btrfs_find_all_roots(trans, fs_info,
2670 			   record->bytenr, BTRFS_SEQ_LAST, &new_roots, false, false);
2671 			if (ret < 0)
2672 				goto cleanup;
2673 			if (qgroup_to_skip) {
2674 				ulist_del(new_roots, qgroup_to_skip, 0);
2675 				ulist_del(record->old_roots, qgroup_to_skip,
2676 					  0);
2677 			}
2678 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2679 							  record->num_bytes,
2680 							  record->old_roots,
2681 							  new_roots);
2682 			record->old_roots = NULL;
2683 			new_roots = NULL;
2684 		}
2685 cleanup:
2686 		ulist_free(record->old_roots);
2687 		ulist_free(new_roots);
2688 		new_roots = NULL;
2689 		rb_erase(node, &delayed_refs->dirty_extent_root);
2690 		kfree(record);
2691 
2692 	}
2693 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2694 				       num_dirty_extents);
2695 	return ret;
2696 }
2697 
2698 /*
2699  * called from commit_transaction. Writes all changed qgroups to disk.
2700  */
2701 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2702 {
2703 	struct btrfs_fs_info *fs_info = trans->fs_info;
2704 	int ret = 0;
2705 
2706 	if (!fs_info->quota_root)
2707 		return ret;
2708 
2709 	spin_lock(&fs_info->qgroup_lock);
2710 	while (!list_empty(&fs_info->dirty_qgroups)) {
2711 		struct btrfs_qgroup *qgroup;
2712 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2713 					  struct btrfs_qgroup, dirty);
2714 		list_del_init(&qgroup->dirty);
2715 		spin_unlock(&fs_info->qgroup_lock);
2716 		ret = update_qgroup_info_item(trans, qgroup);
2717 		if (ret)
2718 			fs_info->qgroup_flags |=
2719 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2720 		ret = update_qgroup_limit_item(trans, qgroup);
2721 		if (ret)
2722 			fs_info->qgroup_flags |=
2723 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2724 		spin_lock(&fs_info->qgroup_lock);
2725 	}
2726 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2727 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2728 	else
2729 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2730 	spin_unlock(&fs_info->qgroup_lock);
2731 
2732 	ret = update_qgroup_status_item(trans);
2733 	if (ret)
2734 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2735 
2736 	return ret;
2737 }
2738 
2739 /*
2740  * Copy the accounting information between qgroups. This is necessary
2741  * when a snapshot or a subvolume is created. Throwing an error will
2742  * cause a transaction abort so we take extra care here to only error
2743  * when a readonly fs is a reasonable outcome.
2744  */
2745 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2746 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2747 {
2748 	int ret = 0;
2749 	int i;
2750 	u64 *i_qgroups;
2751 	bool committing = false;
2752 	struct btrfs_fs_info *fs_info = trans->fs_info;
2753 	struct btrfs_root *quota_root;
2754 	struct btrfs_qgroup *srcgroup;
2755 	struct btrfs_qgroup *dstgroup;
2756 	bool need_rescan = false;
2757 	u32 level_size = 0;
2758 	u64 nums;
2759 
2760 	/*
2761 	 * There are only two callers of this function.
2762 	 *
2763 	 * One in create_subvol() in the ioctl context, which needs to hold
2764 	 * the qgroup_ioctl_lock.
2765 	 *
2766 	 * The other one in create_pending_snapshot() where no other qgroup
2767 	 * code can modify the fs as they all need to either start a new trans
2768 	 * or hold a trans handler, thus we don't need to hold
2769 	 * qgroup_ioctl_lock.
2770 	 * This would avoid long and complex lock chain and make lockdep happy.
2771 	 */
2772 	spin_lock(&fs_info->trans_lock);
2773 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2774 		committing = true;
2775 	spin_unlock(&fs_info->trans_lock);
2776 
2777 	if (!committing)
2778 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2779 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2780 		goto out;
2781 
2782 	quota_root = fs_info->quota_root;
2783 	if (!quota_root) {
2784 		ret = -EINVAL;
2785 		goto out;
2786 	}
2787 
2788 	if (inherit) {
2789 		i_qgroups = (u64 *)(inherit + 1);
2790 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2791 		       2 * inherit->num_excl_copies;
2792 		for (i = 0; i < nums; ++i) {
2793 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2794 
2795 			/*
2796 			 * Zero out invalid groups so we can ignore
2797 			 * them later.
2798 			 */
2799 			if (!srcgroup ||
2800 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2801 				*i_qgroups = 0ULL;
2802 
2803 			++i_qgroups;
2804 		}
2805 	}
2806 
2807 	/*
2808 	 * create a tracking group for the subvol itself
2809 	 */
2810 	ret = add_qgroup_item(trans, quota_root, objectid);
2811 	if (ret)
2812 		goto out;
2813 
2814 	/*
2815 	 * add qgroup to all inherited groups
2816 	 */
2817 	if (inherit) {
2818 		i_qgroups = (u64 *)(inherit + 1);
2819 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2820 			if (*i_qgroups == 0)
2821 				continue;
2822 			ret = add_qgroup_relation_item(trans, objectid,
2823 						       *i_qgroups);
2824 			if (ret && ret != -EEXIST)
2825 				goto out;
2826 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2827 						       objectid);
2828 			if (ret && ret != -EEXIST)
2829 				goto out;
2830 		}
2831 		ret = 0;
2832 	}
2833 
2834 
2835 	spin_lock(&fs_info->qgroup_lock);
2836 
2837 	dstgroup = add_qgroup_rb(fs_info, objectid);
2838 	if (IS_ERR(dstgroup)) {
2839 		ret = PTR_ERR(dstgroup);
2840 		goto unlock;
2841 	}
2842 
2843 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2844 		dstgroup->lim_flags = inherit->lim.flags;
2845 		dstgroup->max_rfer = inherit->lim.max_rfer;
2846 		dstgroup->max_excl = inherit->lim.max_excl;
2847 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2848 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2849 
2850 		ret = update_qgroup_limit_item(trans, dstgroup);
2851 		if (ret) {
2852 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2853 			btrfs_info(fs_info,
2854 				   "unable to update quota limit for %llu",
2855 				   dstgroup->qgroupid);
2856 			goto unlock;
2857 		}
2858 	}
2859 
2860 	if (srcid) {
2861 		srcgroup = find_qgroup_rb(fs_info, srcid);
2862 		if (!srcgroup)
2863 			goto unlock;
2864 
2865 		/*
2866 		 * We call inherit after we clone the root in order to make sure
2867 		 * our counts don't go crazy, so at this point the only
2868 		 * difference between the two roots should be the root node.
2869 		 */
2870 		level_size = fs_info->nodesize;
2871 		dstgroup->rfer = srcgroup->rfer;
2872 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2873 		dstgroup->excl = level_size;
2874 		dstgroup->excl_cmpr = level_size;
2875 		srcgroup->excl = level_size;
2876 		srcgroup->excl_cmpr = level_size;
2877 
2878 		/* inherit the limit info */
2879 		dstgroup->lim_flags = srcgroup->lim_flags;
2880 		dstgroup->max_rfer = srcgroup->max_rfer;
2881 		dstgroup->max_excl = srcgroup->max_excl;
2882 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2883 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2884 
2885 		qgroup_dirty(fs_info, dstgroup);
2886 		qgroup_dirty(fs_info, srcgroup);
2887 	}
2888 
2889 	if (!inherit)
2890 		goto unlock;
2891 
2892 	i_qgroups = (u64 *)(inherit + 1);
2893 	for (i = 0; i < inherit->num_qgroups; ++i) {
2894 		if (*i_qgroups) {
2895 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2896 			if (ret)
2897 				goto unlock;
2898 		}
2899 		++i_qgroups;
2900 
2901 		/*
2902 		 * If we're doing a snapshot, and adding the snapshot to a new
2903 		 * qgroup, the numbers are guaranteed to be incorrect.
2904 		 */
2905 		if (srcid)
2906 			need_rescan = true;
2907 	}
2908 
2909 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2910 		struct btrfs_qgroup *src;
2911 		struct btrfs_qgroup *dst;
2912 
2913 		if (!i_qgroups[0] || !i_qgroups[1])
2914 			continue;
2915 
2916 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2917 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2918 
2919 		if (!src || !dst) {
2920 			ret = -EINVAL;
2921 			goto unlock;
2922 		}
2923 
2924 		dst->rfer = src->rfer - level_size;
2925 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2926 
2927 		/* Manually tweaking numbers certainly needs a rescan */
2928 		need_rescan = true;
2929 	}
2930 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2931 		struct btrfs_qgroup *src;
2932 		struct btrfs_qgroup *dst;
2933 
2934 		if (!i_qgroups[0] || !i_qgroups[1])
2935 			continue;
2936 
2937 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2938 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2939 
2940 		if (!src || !dst) {
2941 			ret = -EINVAL;
2942 			goto unlock;
2943 		}
2944 
2945 		dst->excl = src->excl + level_size;
2946 		dst->excl_cmpr = src->excl_cmpr + level_size;
2947 		need_rescan = true;
2948 	}
2949 
2950 unlock:
2951 	spin_unlock(&fs_info->qgroup_lock);
2952 	if (!ret)
2953 		ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
2954 out:
2955 	if (!committing)
2956 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
2957 	if (need_rescan)
2958 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2959 	return ret;
2960 }
2961 
2962 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2963 {
2964 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2965 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2966 		return false;
2967 
2968 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2969 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2970 		return false;
2971 
2972 	return true;
2973 }
2974 
2975 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2976 			  enum btrfs_qgroup_rsv_type type)
2977 {
2978 	struct btrfs_qgroup *qgroup;
2979 	struct btrfs_fs_info *fs_info = root->fs_info;
2980 	u64 ref_root = root->root_key.objectid;
2981 	int ret = 0;
2982 	struct ulist_node *unode;
2983 	struct ulist_iterator uiter;
2984 
2985 	if (!is_fstree(ref_root))
2986 		return 0;
2987 
2988 	if (num_bytes == 0)
2989 		return 0;
2990 
2991 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2992 	    capable(CAP_SYS_RESOURCE))
2993 		enforce = false;
2994 
2995 	spin_lock(&fs_info->qgroup_lock);
2996 	if (!fs_info->quota_root)
2997 		goto out;
2998 
2999 	qgroup = find_qgroup_rb(fs_info, ref_root);
3000 	if (!qgroup)
3001 		goto out;
3002 
3003 	/*
3004 	 * in a first step, we check all affected qgroups if any limits would
3005 	 * be exceeded
3006 	 */
3007 	ulist_reinit(fs_info->qgroup_ulist);
3008 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3009 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3010 	if (ret < 0)
3011 		goto out;
3012 	ULIST_ITER_INIT(&uiter);
3013 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3014 		struct btrfs_qgroup *qg;
3015 		struct btrfs_qgroup_list *glist;
3016 
3017 		qg = unode_aux_to_qgroup(unode);
3018 
3019 		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3020 			ret = -EDQUOT;
3021 			goto out;
3022 		}
3023 
3024 		list_for_each_entry(glist, &qg->groups, next_group) {
3025 			ret = ulist_add(fs_info->qgroup_ulist,
3026 					glist->group->qgroupid,
3027 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3028 			if (ret < 0)
3029 				goto out;
3030 		}
3031 	}
3032 	ret = 0;
3033 	/*
3034 	 * no limits exceeded, now record the reservation into all qgroups
3035 	 */
3036 	ULIST_ITER_INIT(&uiter);
3037 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3038 		struct btrfs_qgroup *qg;
3039 
3040 		qg = unode_aux_to_qgroup(unode);
3041 
3042 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
3043 	}
3044 
3045 out:
3046 	spin_unlock(&fs_info->qgroup_lock);
3047 	return ret;
3048 }
3049 
3050 /*
3051  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3052  * qgroup).
3053  *
3054  * Will handle all higher level qgroup too.
3055  *
3056  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3057  * This special case is only used for META_PERTRANS type.
3058  */
3059 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3060 			       u64 ref_root, u64 num_bytes,
3061 			       enum btrfs_qgroup_rsv_type type)
3062 {
3063 	struct btrfs_qgroup *qgroup;
3064 	struct ulist_node *unode;
3065 	struct ulist_iterator uiter;
3066 	int ret = 0;
3067 
3068 	if (!is_fstree(ref_root))
3069 		return;
3070 
3071 	if (num_bytes == 0)
3072 		return;
3073 
3074 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3075 		WARN(1, "%s: Invalid type to free", __func__);
3076 		return;
3077 	}
3078 	spin_lock(&fs_info->qgroup_lock);
3079 
3080 	if (!fs_info->quota_root)
3081 		goto out;
3082 
3083 	qgroup = find_qgroup_rb(fs_info, ref_root);
3084 	if (!qgroup)
3085 		goto out;
3086 
3087 	if (num_bytes == (u64)-1)
3088 		/*
3089 		 * We're freeing all pertrans rsv, get reserved value from
3090 		 * level 0 qgroup as real num_bytes to free.
3091 		 */
3092 		num_bytes = qgroup->rsv.values[type];
3093 
3094 	ulist_reinit(fs_info->qgroup_ulist);
3095 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3096 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3097 	if (ret < 0)
3098 		goto out;
3099 	ULIST_ITER_INIT(&uiter);
3100 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3101 		struct btrfs_qgroup *qg;
3102 		struct btrfs_qgroup_list *glist;
3103 
3104 		qg = unode_aux_to_qgroup(unode);
3105 
3106 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3107 
3108 		list_for_each_entry(glist, &qg->groups, next_group) {
3109 			ret = ulist_add(fs_info->qgroup_ulist,
3110 					glist->group->qgroupid,
3111 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3112 			if (ret < 0)
3113 				goto out;
3114 		}
3115 	}
3116 
3117 out:
3118 	spin_unlock(&fs_info->qgroup_lock);
3119 }
3120 
3121 /*
3122  * Check if the leaf is the last leaf. Which means all node pointers
3123  * are at their last position.
3124  */
3125 static bool is_last_leaf(struct btrfs_path *path)
3126 {
3127 	int i;
3128 
3129 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3130 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3131 			return false;
3132 	}
3133 	return true;
3134 }
3135 
3136 /*
3137  * returns < 0 on error, 0 when more leafs are to be scanned.
3138  * returns 1 when done.
3139  */
3140 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3141 			      struct btrfs_path *path)
3142 {
3143 	struct btrfs_fs_info *fs_info = trans->fs_info;
3144 	struct btrfs_key found;
3145 	struct extent_buffer *scratch_leaf = NULL;
3146 	struct ulist *roots = NULL;
3147 	u64 num_bytes;
3148 	bool done;
3149 	int slot;
3150 	int ret;
3151 
3152 	mutex_lock(&fs_info->qgroup_rescan_lock);
3153 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3154 					 &fs_info->qgroup_rescan_progress,
3155 					 path, 1, 0);
3156 
3157 	btrfs_debug(fs_info,
3158 		"current progress key (%llu %u %llu), search_slot ret %d",
3159 		fs_info->qgroup_rescan_progress.objectid,
3160 		fs_info->qgroup_rescan_progress.type,
3161 		fs_info->qgroup_rescan_progress.offset, ret);
3162 
3163 	if (ret) {
3164 		/*
3165 		 * The rescan is about to end, we will not be scanning any
3166 		 * further blocks. We cannot unset the RESCAN flag here, because
3167 		 * we want to commit the transaction if everything went well.
3168 		 * To make the live accounting work in this phase, we set our
3169 		 * scan progress pointer such that every real extent objectid
3170 		 * will be smaller.
3171 		 */
3172 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3173 		btrfs_release_path(path);
3174 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3175 		return ret;
3176 	}
3177 	done = is_last_leaf(path);
3178 
3179 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3180 			      btrfs_header_nritems(path->nodes[0]) - 1);
3181 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3182 
3183 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3184 	if (!scratch_leaf) {
3185 		ret = -ENOMEM;
3186 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3187 		goto out;
3188 	}
3189 	slot = path->slots[0];
3190 	btrfs_release_path(path);
3191 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3192 
3193 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3194 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3195 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3196 		    found.type != BTRFS_METADATA_ITEM_KEY)
3197 			continue;
3198 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3199 			num_bytes = fs_info->nodesize;
3200 		else
3201 			num_bytes = found.offset;
3202 
3203 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3204 					   &roots, false, false);
3205 		if (ret < 0)
3206 			goto out;
3207 		/* For rescan, just pass old_roots as NULL */
3208 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3209 						  num_bytes, NULL, roots);
3210 		if (ret < 0)
3211 			goto out;
3212 	}
3213 out:
3214 	if (scratch_leaf)
3215 		free_extent_buffer(scratch_leaf);
3216 
3217 	if (done && !ret) {
3218 		ret = 1;
3219 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3220 	}
3221 	return ret;
3222 }
3223 
3224 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3225 {
3226 	return btrfs_fs_closing(fs_info) ||
3227 		test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
3228 }
3229 
3230 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3231 {
3232 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3233 						     qgroup_rescan_work);
3234 	struct btrfs_path *path;
3235 	struct btrfs_trans_handle *trans = NULL;
3236 	int err = -ENOMEM;
3237 	int ret = 0;
3238 	bool stopped = false;
3239 
3240 	path = btrfs_alloc_path();
3241 	if (!path)
3242 		goto out;
3243 	/*
3244 	 * Rescan should only search for commit root, and any later difference
3245 	 * should be recorded by qgroup
3246 	 */
3247 	path->search_commit_root = 1;
3248 	path->skip_locking = 1;
3249 
3250 	err = 0;
3251 	while (!err && !(stopped = rescan_should_stop(fs_info))) {
3252 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3253 		if (IS_ERR(trans)) {
3254 			err = PTR_ERR(trans);
3255 			break;
3256 		}
3257 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3258 			err = -EINTR;
3259 		} else {
3260 			err = qgroup_rescan_leaf(trans, path);
3261 		}
3262 		if (err > 0)
3263 			btrfs_commit_transaction(trans);
3264 		else
3265 			btrfs_end_transaction(trans);
3266 	}
3267 
3268 out:
3269 	btrfs_free_path(path);
3270 
3271 	mutex_lock(&fs_info->qgroup_rescan_lock);
3272 	if (err > 0 &&
3273 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3274 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3275 	} else if (err < 0) {
3276 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3277 	}
3278 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3279 
3280 	/*
3281 	 * only update status, since the previous part has already updated the
3282 	 * qgroup info.
3283 	 */
3284 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3285 	if (IS_ERR(trans)) {
3286 		err = PTR_ERR(trans);
3287 		trans = NULL;
3288 		btrfs_err(fs_info,
3289 			  "fail to start transaction for status update: %d",
3290 			  err);
3291 	}
3292 
3293 	mutex_lock(&fs_info->qgroup_rescan_lock);
3294 	if (!stopped)
3295 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3296 	if (trans) {
3297 		ret = update_qgroup_status_item(trans);
3298 		if (ret < 0) {
3299 			err = ret;
3300 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3301 				  err);
3302 		}
3303 	}
3304 	fs_info->qgroup_rescan_running = false;
3305 	complete_all(&fs_info->qgroup_rescan_completion);
3306 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3307 
3308 	if (!trans)
3309 		return;
3310 
3311 	btrfs_end_transaction(trans);
3312 
3313 	if (stopped) {
3314 		btrfs_info(fs_info, "qgroup scan paused");
3315 	} else if (err >= 0) {
3316 		btrfs_info(fs_info, "qgroup scan completed%s",
3317 			err > 0 ? " (inconsistency flag cleared)" : "");
3318 	} else {
3319 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3320 	}
3321 }
3322 
3323 /*
3324  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3325  * memory required for the rescan context.
3326  */
3327 static int
3328 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3329 		   int init_flags)
3330 {
3331 	int ret = 0;
3332 
3333 	if (!init_flags) {
3334 		/* we're resuming qgroup rescan at mount time */
3335 		if (!(fs_info->qgroup_flags &
3336 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3337 			btrfs_warn(fs_info,
3338 			"qgroup rescan init failed, qgroup rescan is not queued");
3339 			ret = -EINVAL;
3340 		} else if (!(fs_info->qgroup_flags &
3341 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3342 			btrfs_warn(fs_info,
3343 			"qgroup rescan init failed, qgroup is not enabled");
3344 			ret = -EINVAL;
3345 		}
3346 
3347 		if (ret)
3348 			return ret;
3349 	}
3350 
3351 	mutex_lock(&fs_info->qgroup_rescan_lock);
3352 
3353 	if (init_flags) {
3354 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3355 			btrfs_warn(fs_info,
3356 				   "qgroup rescan is already in progress");
3357 			ret = -EINPROGRESS;
3358 		} else if (!(fs_info->qgroup_flags &
3359 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3360 			btrfs_warn(fs_info,
3361 			"qgroup rescan init failed, qgroup is not enabled");
3362 			ret = -EINVAL;
3363 		}
3364 
3365 		if (ret) {
3366 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3367 			return ret;
3368 		}
3369 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3370 	}
3371 
3372 	memset(&fs_info->qgroup_rescan_progress, 0,
3373 		sizeof(fs_info->qgroup_rescan_progress));
3374 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3375 	init_completion(&fs_info->qgroup_rescan_completion);
3376 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3377 
3378 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3379 			btrfs_qgroup_rescan_worker, NULL, NULL);
3380 	return 0;
3381 }
3382 
3383 static void
3384 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3385 {
3386 	struct rb_node *n;
3387 	struct btrfs_qgroup *qgroup;
3388 
3389 	spin_lock(&fs_info->qgroup_lock);
3390 	/* clear all current qgroup tracking information */
3391 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3392 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3393 		qgroup->rfer = 0;
3394 		qgroup->rfer_cmpr = 0;
3395 		qgroup->excl = 0;
3396 		qgroup->excl_cmpr = 0;
3397 		qgroup_dirty(fs_info, qgroup);
3398 	}
3399 	spin_unlock(&fs_info->qgroup_lock);
3400 }
3401 
3402 int
3403 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3404 {
3405 	int ret = 0;
3406 	struct btrfs_trans_handle *trans;
3407 
3408 	ret = qgroup_rescan_init(fs_info, 0, 1);
3409 	if (ret)
3410 		return ret;
3411 
3412 	/*
3413 	 * We have set the rescan_progress to 0, which means no more
3414 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3415 	 * However, btrfs_qgroup_account_ref may be right after its call
3416 	 * to btrfs_find_all_roots, in which case it would still do the
3417 	 * accounting.
3418 	 * To solve this, we're committing the transaction, which will
3419 	 * ensure we run all delayed refs and only after that, we are
3420 	 * going to clear all tracking information for a clean start.
3421 	 */
3422 
3423 	trans = btrfs_join_transaction(fs_info->fs_root);
3424 	if (IS_ERR(trans)) {
3425 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3426 		return PTR_ERR(trans);
3427 	}
3428 	ret = btrfs_commit_transaction(trans);
3429 	if (ret) {
3430 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3431 		return ret;
3432 	}
3433 
3434 	qgroup_rescan_zero_tracking(fs_info);
3435 
3436 	mutex_lock(&fs_info->qgroup_rescan_lock);
3437 	fs_info->qgroup_rescan_running = true;
3438 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3439 			 &fs_info->qgroup_rescan_work);
3440 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3441 
3442 	return 0;
3443 }
3444 
3445 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3446 				     bool interruptible)
3447 {
3448 	int running;
3449 	int ret = 0;
3450 
3451 	mutex_lock(&fs_info->qgroup_rescan_lock);
3452 	running = fs_info->qgroup_rescan_running;
3453 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3454 
3455 	if (!running)
3456 		return 0;
3457 
3458 	if (interruptible)
3459 		ret = wait_for_completion_interruptible(
3460 					&fs_info->qgroup_rescan_completion);
3461 	else
3462 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3463 
3464 	return ret;
3465 }
3466 
3467 /*
3468  * this is only called from open_ctree where we're still single threaded, thus
3469  * locking is omitted here.
3470  */
3471 void
3472 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3473 {
3474 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3475 		mutex_lock(&fs_info->qgroup_rescan_lock);
3476 		fs_info->qgroup_rescan_running = true;
3477 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3478 				 &fs_info->qgroup_rescan_work);
3479 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3480 	}
3481 }
3482 
3483 #define rbtree_iterate_from_safe(node, next, start)				\
3484        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3485 
3486 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3487 				  struct extent_changeset *reserved, u64 start,
3488 				  u64 len)
3489 {
3490 	struct rb_node *node;
3491 	struct rb_node *next;
3492 	struct ulist_node *entry;
3493 	int ret = 0;
3494 
3495 	node = reserved->range_changed.root.rb_node;
3496 	if (!node)
3497 		return 0;
3498 	while (node) {
3499 		entry = rb_entry(node, struct ulist_node, rb_node);
3500 		if (entry->val < start)
3501 			node = node->rb_right;
3502 		else
3503 			node = node->rb_left;
3504 	}
3505 
3506 	if (entry->val > start && rb_prev(&entry->rb_node))
3507 		entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3508 				 rb_node);
3509 
3510 	rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3511 		u64 entry_start;
3512 		u64 entry_end;
3513 		u64 entry_len;
3514 		int clear_ret;
3515 
3516 		entry = rb_entry(node, struct ulist_node, rb_node);
3517 		entry_start = entry->val;
3518 		entry_end = entry->aux;
3519 		entry_len = entry_end - entry_start + 1;
3520 
3521 		if (entry_start >= start + len)
3522 			break;
3523 		if (entry_start + entry_len <= start)
3524 			continue;
3525 		/*
3526 		 * Now the entry is in [start, start + len), revert the
3527 		 * EXTENT_QGROUP_RESERVED bit.
3528 		 */
3529 		clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3530 					      entry_end, EXTENT_QGROUP_RESERVED);
3531 		if (!ret && clear_ret < 0)
3532 			ret = clear_ret;
3533 
3534 		ulist_del(&reserved->range_changed, entry->val, entry->aux);
3535 		if (likely(reserved->bytes_changed >= entry_len)) {
3536 			reserved->bytes_changed -= entry_len;
3537 		} else {
3538 			WARN_ON(1);
3539 			reserved->bytes_changed = 0;
3540 		}
3541 	}
3542 
3543 	return ret;
3544 }
3545 
3546 /*
3547  * Try to free some space for qgroup.
3548  *
3549  * For qgroup, there are only 3 ways to free qgroup space:
3550  * - Flush nodatacow write
3551  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3552  *   In theory, we should only flush nodatacow inodes, but it's not yet
3553  *   possible, so we need to flush the whole root.
3554  *
3555  * - Wait for ordered extents
3556  *   When ordered extents are finished, their reserved metadata is finally
3557  *   converted to per_trans status, which can be freed by later commit
3558  *   transaction.
3559  *
3560  * - Commit transaction
3561  *   This would free the meta_per_trans space.
3562  *   In theory this shouldn't provide much space, but any more qgroup space
3563  *   is needed.
3564  */
3565 static int try_flush_qgroup(struct btrfs_root *root)
3566 {
3567 	struct btrfs_trans_handle *trans;
3568 	int ret;
3569 
3570 	/* Can't hold an open transaction or we run the risk of deadlocking. */
3571 	ASSERT(current->journal_info == NULL);
3572 	if (WARN_ON(current->journal_info))
3573 		return 0;
3574 
3575 	/*
3576 	 * We don't want to run flush again and again, so if there is a running
3577 	 * one, we won't try to start a new flush, but exit directly.
3578 	 */
3579 	if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3580 		wait_event(root->qgroup_flush_wait,
3581 			!test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3582 		return 0;
3583 	}
3584 
3585 	ret = btrfs_start_delalloc_snapshot(root, true);
3586 	if (ret < 0)
3587 		goto out;
3588 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3589 
3590 	trans = btrfs_join_transaction(root);
3591 	if (IS_ERR(trans)) {
3592 		ret = PTR_ERR(trans);
3593 		goto out;
3594 	}
3595 
3596 	ret = btrfs_commit_transaction(trans);
3597 out:
3598 	clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3599 	wake_up(&root->qgroup_flush_wait);
3600 	return ret;
3601 }
3602 
3603 static int qgroup_reserve_data(struct btrfs_inode *inode,
3604 			struct extent_changeset **reserved_ret, u64 start,
3605 			u64 len)
3606 {
3607 	struct btrfs_root *root = inode->root;
3608 	struct extent_changeset *reserved;
3609 	bool new_reserved = false;
3610 	u64 orig_reserved;
3611 	u64 to_reserve;
3612 	int ret;
3613 
3614 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3615 	    !is_fstree(root->root_key.objectid) || len == 0)
3616 		return 0;
3617 
3618 	/* @reserved parameter is mandatory for qgroup */
3619 	if (WARN_ON(!reserved_ret))
3620 		return -EINVAL;
3621 	if (!*reserved_ret) {
3622 		new_reserved = true;
3623 		*reserved_ret = extent_changeset_alloc();
3624 		if (!*reserved_ret)
3625 			return -ENOMEM;
3626 	}
3627 	reserved = *reserved_ret;
3628 	/* Record already reserved space */
3629 	orig_reserved = reserved->bytes_changed;
3630 	ret = set_record_extent_bits(&inode->io_tree, start,
3631 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3632 
3633 	/* Newly reserved space */
3634 	to_reserve = reserved->bytes_changed - orig_reserved;
3635 	trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3636 					to_reserve, QGROUP_RESERVE);
3637 	if (ret < 0)
3638 		goto out;
3639 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3640 	if (ret < 0)
3641 		goto cleanup;
3642 
3643 	return ret;
3644 
3645 cleanup:
3646 	qgroup_unreserve_range(inode, reserved, start, len);
3647 out:
3648 	if (new_reserved) {
3649 		extent_changeset_free(reserved);
3650 		*reserved_ret = NULL;
3651 	}
3652 	return ret;
3653 }
3654 
3655 /*
3656  * Reserve qgroup space for range [start, start + len).
3657  *
3658  * This function will either reserve space from related qgroups or do nothing
3659  * if the range is already reserved.
3660  *
3661  * Return 0 for successful reservation
3662  * Return <0 for error (including -EQUOT)
3663  *
3664  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3665  *	 commit transaction. So caller should not hold any dirty page locked.
3666  */
3667 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3668 			struct extent_changeset **reserved_ret, u64 start,
3669 			u64 len)
3670 {
3671 	int ret;
3672 
3673 	ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3674 	if (ret <= 0 && ret != -EDQUOT)
3675 		return ret;
3676 
3677 	ret = try_flush_qgroup(inode->root);
3678 	if (ret < 0)
3679 		return ret;
3680 	return qgroup_reserve_data(inode, reserved_ret, start, len);
3681 }
3682 
3683 /* Free ranges specified by @reserved, normally in error path */
3684 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3685 			struct extent_changeset *reserved, u64 start, u64 len)
3686 {
3687 	struct btrfs_root *root = inode->root;
3688 	struct ulist_node *unode;
3689 	struct ulist_iterator uiter;
3690 	struct extent_changeset changeset;
3691 	int freed = 0;
3692 	int ret;
3693 
3694 	extent_changeset_init(&changeset);
3695 	len = round_up(start + len, root->fs_info->sectorsize);
3696 	start = round_down(start, root->fs_info->sectorsize);
3697 
3698 	ULIST_ITER_INIT(&uiter);
3699 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3700 		u64 range_start = unode->val;
3701 		/* unode->aux is the inclusive end */
3702 		u64 range_len = unode->aux - range_start + 1;
3703 		u64 free_start;
3704 		u64 free_len;
3705 
3706 		extent_changeset_release(&changeset);
3707 
3708 		/* Only free range in range [start, start + len) */
3709 		if (range_start >= start + len ||
3710 		    range_start + range_len <= start)
3711 			continue;
3712 		free_start = max(range_start, start);
3713 		free_len = min(start + len, range_start + range_len) -
3714 			   free_start;
3715 		/*
3716 		 * TODO: To also modify reserved->ranges_reserved to reflect
3717 		 * the modification.
3718 		 *
3719 		 * However as long as we free qgroup reserved according to
3720 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3721 		 * So not need to rush.
3722 		 */
3723 		ret = clear_record_extent_bits(&inode->io_tree, free_start,
3724 				free_start + free_len - 1,
3725 				EXTENT_QGROUP_RESERVED, &changeset);
3726 		if (ret < 0)
3727 			goto out;
3728 		freed += changeset.bytes_changed;
3729 	}
3730 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3731 				  BTRFS_QGROUP_RSV_DATA);
3732 	ret = freed;
3733 out:
3734 	extent_changeset_release(&changeset);
3735 	return ret;
3736 }
3737 
3738 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3739 			struct extent_changeset *reserved, u64 start, u64 len,
3740 			int free)
3741 {
3742 	struct extent_changeset changeset;
3743 	int trace_op = QGROUP_RELEASE;
3744 	int ret;
3745 
3746 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3747 		return 0;
3748 
3749 	/* In release case, we shouldn't have @reserved */
3750 	WARN_ON(!free && reserved);
3751 	if (free && reserved)
3752 		return qgroup_free_reserved_data(inode, reserved, start, len);
3753 	extent_changeset_init(&changeset);
3754 	ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3755 				       EXTENT_QGROUP_RESERVED, &changeset);
3756 	if (ret < 0)
3757 		goto out;
3758 
3759 	if (free)
3760 		trace_op = QGROUP_FREE;
3761 	trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3762 					changeset.bytes_changed, trace_op);
3763 	if (free)
3764 		btrfs_qgroup_free_refroot(inode->root->fs_info,
3765 				inode->root->root_key.objectid,
3766 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3767 	ret = changeset.bytes_changed;
3768 out:
3769 	extent_changeset_release(&changeset);
3770 	return ret;
3771 }
3772 
3773 /*
3774  * Free a reserved space range from io_tree and related qgroups
3775  *
3776  * Should be called when a range of pages get invalidated before reaching disk.
3777  * Or for error cleanup case.
3778  * if @reserved is given, only reserved range in [@start, @start + @len) will
3779  * be freed.
3780  *
3781  * For data written to disk, use btrfs_qgroup_release_data().
3782  *
3783  * NOTE: This function may sleep for memory allocation.
3784  */
3785 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3786 			struct extent_changeset *reserved, u64 start, u64 len)
3787 {
3788 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3789 }
3790 
3791 /*
3792  * Release a reserved space range from io_tree only.
3793  *
3794  * Should be called when a range of pages get written to disk and corresponding
3795  * FILE_EXTENT is inserted into corresponding root.
3796  *
3797  * Since new qgroup accounting framework will only update qgroup numbers at
3798  * commit_transaction() time, its reserved space shouldn't be freed from
3799  * related qgroups.
3800  *
3801  * But we should release the range from io_tree, to allow further write to be
3802  * COWed.
3803  *
3804  * NOTE: This function may sleep for memory allocation.
3805  */
3806 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3807 {
3808 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3809 }
3810 
3811 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3812 			      enum btrfs_qgroup_rsv_type type)
3813 {
3814 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3815 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3816 		return;
3817 	if (num_bytes == 0)
3818 		return;
3819 
3820 	spin_lock(&root->qgroup_meta_rsv_lock);
3821 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3822 		root->qgroup_meta_rsv_prealloc += num_bytes;
3823 	else
3824 		root->qgroup_meta_rsv_pertrans += num_bytes;
3825 	spin_unlock(&root->qgroup_meta_rsv_lock);
3826 }
3827 
3828 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3829 			     enum btrfs_qgroup_rsv_type type)
3830 {
3831 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3832 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3833 		return 0;
3834 	if (num_bytes == 0)
3835 		return 0;
3836 
3837 	spin_lock(&root->qgroup_meta_rsv_lock);
3838 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3839 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3840 				  num_bytes);
3841 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3842 	} else {
3843 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3844 				  num_bytes);
3845 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3846 	}
3847 	spin_unlock(&root->qgroup_meta_rsv_lock);
3848 	return num_bytes;
3849 }
3850 
3851 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3852 			      enum btrfs_qgroup_rsv_type type, bool enforce)
3853 {
3854 	struct btrfs_fs_info *fs_info = root->fs_info;
3855 	int ret;
3856 
3857 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3858 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3859 		return 0;
3860 
3861 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3862 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3863 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3864 	if (ret < 0)
3865 		return ret;
3866 	/*
3867 	 * Record what we have reserved into root.
3868 	 *
3869 	 * To avoid quota disabled->enabled underflow.
3870 	 * In that case, we may try to free space we haven't reserved
3871 	 * (since quota was disabled), so record what we reserved into root.
3872 	 * And ensure later release won't underflow this number.
3873 	 */
3874 	add_root_meta_rsv(root, num_bytes, type);
3875 	return ret;
3876 }
3877 
3878 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3879 				enum btrfs_qgroup_rsv_type type, bool enforce)
3880 {
3881 	int ret;
3882 
3883 	ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3884 	if (ret <= 0 && ret != -EDQUOT)
3885 		return ret;
3886 
3887 	ret = try_flush_qgroup(root);
3888 	if (ret < 0)
3889 		return ret;
3890 	return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3891 }
3892 
3893 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3894 {
3895 	struct btrfs_fs_info *fs_info = root->fs_info;
3896 
3897 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3898 	    !is_fstree(root->root_key.objectid))
3899 		return;
3900 
3901 	/* TODO: Update trace point to handle such free */
3902 	trace_qgroup_meta_free_all_pertrans(root);
3903 	/* Special value -1 means to free all reserved space */
3904 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3905 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3906 }
3907 
3908 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3909 			      enum btrfs_qgroup_rsv_type type)
3910 {
3911 	struct btrfs_fs_info *fs_info = root->fs_info;
3912 
3913 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3914 	    !is_fstree(root->root_key.objectid))
3915 		return;
3916 
3917 	/*
3918 	 * reservation for META_PREALLOC can happen before quota is enabled,
3919 	 * which can lead to underflow.
3920 	 * Here ensure we will only free what we really have reserved.
3921 	 */
3922 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3923 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3924 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3925 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3926 				  num_bytes, type);
3927 }
3928 
3929 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3930 				int num_bytes)
3931 {
3932 	struct btrfs_qgroup *qgroup;
3933 	struct ulist_node *unode;
3934 	struct ulist_iterator uiter;
3935 	int ret = 0;
3936 
3937 	if (num_bytes == 0)
3938 		return;
3939 	if (!fs_info->quota_root)
3940 		return;
3941 
3942 	spin_lock(&fs_info->qgroup_lock);
3943 	qgroup = find_qgroup_rb(fs_info, ref_root);
3944 	if (!qgroup)
3945 		goto out;
3946 	ulist_reinit(fs_info->qgroup_ulist);
3947 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3948 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3949 	if (ret < 0)
3950 		goto out;
3951 	ULIST_ITER_INIT(&uiter);
3952 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3953 		struct btrfs_qgroup *qg;
3954 		struct btrfs_qgroup_list *glist;
3955 
3956 		qg = unode_aux_to_qgroup(unode);
3957 
3958 		qgroup_rsv_release(fs_info, qg, num_bytes,
3959 				BTRFS_QGROUP_RSV_META_PREALLOC);
3960 		qgroup_rsv_add(fs_info, qg, num_bytes,
3961 				BTRFS_QGROUP_RSV_META_PERTRANS);
3962 		list_for_each_entry(glist, &qg->groups, next_group) {
3963 			ret = ulist_add(fs_info->qgroup_ulist,
3964 					glist->group->qgroupid,
3965 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3966 			if (ret < 0)
3967 				goto out;
3968 		}
3969 	}
3970 out:
3971 	spin_unlock(&fs_info->qgroup_lock);
3972 }
3973 
3974 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3975 {
3976 	struct btrfs_fs_info *fs_info = root->fs_info;
3977 
3978 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3979 	    !is_fstree(root->root_key.objectid))
3980 		return;
3981 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3982 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3983 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3984 	trace_qgroup_meta_convert(root, num_bytes);
3985 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3986 }
3987 
3988 /*
3989  * Check qgroup reserved space leaking, normally at destroy inode
3990  * time
3991  */
3992 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
3993 {
3994 	struct extent_changeset changeset;
3995 	struct ulist_node *unode;
3996 	struct ulist_iterator iter;
3997 	int ret;
3998 
3999 	extent_changeset_init(&changeset);
4000 	ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4001 			EXTENT_QGROUP_RESERVED, &changeset);
4002 
4003 	WARN_ON(ret < 0);
4004 	if (WARN_ON(changeset.bytes_changed)) {
4005 		ULIST_ITER_INIT(&iter);
4006 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4007 			btrfs_warn(inode->root->fs_info,
4008 		"leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4009 				btrfs_ino(inode), unode->val, unode->aux);
4010 		}
4011 		btrfs_qgroup_free_refroot(inode->root->fs_info,
4012 				inode->root->root_key.objectid,
4013 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4014 
4015 	}
4016 	extent_changeset_release(&changeset);
4017 }
4018 
4019 void btrfs_qgroup_init_swapped_blocks(
4020 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4021 {
4022 	int i;
4023 
4024 	spin_lock_init(&swapped_blocks->lock);
4025 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4026 		swapped_blocks->blocks[i] = RB_ROOT;
4027 	swapped_blocks->swapped = false;
4028 }
4029 
4030 /*
4031  * Delete all swapped blocks record of @root.
4032  * Every record here means we skipped a full subtree scan for qgroup.
4033  *
4034  * Gets called when committing one transaction.
4035  */
4036 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4037 {
4038 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4039 	int i;
4040 
4041 	swapped_blocks = &root->swapped_blocks;
4042 
4043 	spin_lock(&swapped_blocks->lock);
4044 	if (!swapped_blocks->swapped)
4045 		goto out;
4046 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4047 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
4048 		struct btrfs_qgroup_swapped_block *entry;
4049 		struct btrfs_qgroup_swapped_block *next;
4050 
4051 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4052 						     node)
4053 			kfree(entry);
4054 		swapped_blocks->blocks[i] = RB_ROOT;
4055 	}
4056 	swapped_blocks->swapped = false;
4057 out:
4058 	spin_unlock(&swapped_blocks->lock);
4059 }
4060 
4061 /*
4062  * Add subtree roots record into @subvol_root.
4063  *
4064  * @subvol_root:	tree root of the subvolume tree get swapped
4065  * @bg:			block group under balance
4066  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
4067  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
4068  *			BOTH POINTERS ARE BEFORE TREE SWAP
4069  * @last_snapshot:	last snapshot generation of the subvolume tree
4070  */
4071 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4072 		struct btrfs_root *subvol_root,
4073 		struct btrfs_block_group *bg,
4074 		struct extent_buffer *subvol_parent, int subvol_slot,
4075 		struct extent_buffer *reloc_parent, int reloc_slot,
4076 		u64 last_snapshot)
4077 {
4078 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4079 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4080 	struct btrfs_qgroup_swapped_block *block;
4081 	struct rb_node **cur;
4082 	struct rb_node *parent = NULL;
4083 	int level = btrfs_header_level(subvol_parent) - 1;
4084 	int ret = 0;
4085 
4086 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4087 		return 0;
4088 
4089 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4090 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4091 		btrfs_err_rl(fs_info,
4092 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4093 			__func__,
4094 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4095 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4096 		return -EUCLEAN;
4097 	}
4098 
4099 	block = kmalloc(sizeof(*block), GFP_NOFS);
4100 	if (!block) {
4101 		ret = -ENOMEM;
4102 		goto out;
4103 	}
4104 
4105 	/*
4106 	 * @reloc_parent/slot is still before swap, while @block is going to
4107 	 * record the bytenr after swap, so we do the swap here.
4108 	 */
4109 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4110 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4111 							     reloc_slot);
4112 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4113 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4114 							    subvol_slot);
4115 	block->last_snapshot = last_snapshot;
4116 	block->level = level;
4117 
4118 	/*
4119 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4120 	 * no one else can modify tree blocks thus we qgroup will not change
4121 	 * no matter the value of trace_leaf.
4122 	 */
4123 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4124 		block->trace_leaf = true;
4125 	else
4126 		block->trace_leaf = false;
4127 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4128 
4129 	/* Insert @block into @blocks */
4130 	spin_lock(&blocks->lock);
4131 	cur = &blocks->blocks[level].rb_node;
4132 	while (*cur) {
4133 		struct btrfs_qgroup_swapped_block *entry;
4134 
4135 		parent = *cur;
4136 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4137 				 node);
4138 
4139 		if (entry->subvol_bytenr < block->subvol_bytenr) {
4140 			cur = &(*cur)->rb_left;
4141 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
4142 			cur = &(*cur)->rb_right;
4143 		} else {
4144 			if (entry->subvol_generation !=
4145 					block->subvol_generation ||
4146 			    entry->reloc_bytenr != block->reloc_bytenr ||
4147 			    entry->reloc_generation !=
4148 					block->reloc_generation) {
4149 				/*
4150 				 * Duplicated but mismatch entry found.
4151 				 * Shouldn't happen.
4152 				 *
4153 				 * Marking qgroup inconsistent should be enough
4154 				 * for end users.
4155 				 */
4156 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4157 				ret = -EEXIST;
4158 			}
4159 			kfree(block);
4160 			goto out_unlock;
4161 		}
4162 	}
4163 	rb_link_node(&block->node, parent, cur);
4164 	rb_insert_color(&block->node, &blocks->blocks[level]);
4165 	blocks->swapped = true;
4166 out_unlock:
4167 	spin_unlock(&blocks->lock);
4168 out:
4169 	if (ret < 0)
4170 		fs_info->qgroup_flags |=
4171 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4172 	return ret;
4173 }
4174 
4175 /*
4176  * Check if the tree block is a subtree root, and if so do the needed
4177  * delayed subtree trace for qgroup.
4178  *
4179  * This is called during btrfs_cow_block().
4180  */
4181 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4182 					 struct btrfs_root *root,
4183 					 struct extent_buffer *subvol_eb)
4184 {
4185 	struct btrfs_fs_info *fs_info = root->fs_info;
4186 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4187 	struct btrfs_qgroup_swapped_block *block;
4188 	struct extent_buffer *reloc_eb = NULL;
4189 	struct rb_node *node;
4190 	bool found = false;
4191 	bool swapped = false;
4192 	int level = btrfs_header_level(subvol_eb);
4193 	int ret = 0;
4194 	int i;
4195 
4196 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4197 		return 0;
4198 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4199 		return 0;
4200 
4201 	spin_lock(&blocks->lock);
4202 	if (!blocks->swapped) {
4203 		spin_unlock(&blocks->lock);
4204 		return 0;
4205 	}
4206 	node = blocks->blocks[level].rb_node;
4207 
4208 	while (node) {
4209 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4210 		if (block->subvol_bytenr < subvol_eb->start) {
4211 			node = node->rb_left;
4212 		} else if (block->subvol_bytenr > subvol_eb->start) {
4213 			node = node->rb_right;
4214 		} else {
4215 			found = true;
4216 			break;
4217 		}
4218 	}
4219 	if (!found) {
4220 		spin_unlock(&blocks->lock);
4221 		goto out;
4222 	}
4223 	/* Found one, remove it from @blocks first and update blocks->swapped */
4224 	rb_erase(&block->node, &blocks->blocks[level]);
4225 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4226 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4227 			swapped = true;
4228 			break;
4229 		}
4230 	}
4231 	blocks->swapped = swapped;
4232 	spin_unlock(&blocks->lock);
4233 
4234 	/* Read out reloc subtree root */
4235 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, 0,
4236 				   block->reloc_generation, block->level,
4237 				   &block->first_key);
4238 	if (IS_ERR(reloc_eb)) {
4239 		ret = PTR_ERR(reloc_eb);
4240 		reloc_eb = NULL;
4241 		goto free_out;
4242 	}
4243 	if (!extent_buffer_uptodate(reloc_eb)) {
4244 		ret = -EIO;
4245 		goto free_out;
4246 	}
4247 
4248 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4249 			block->last_snapshot, block->trace_leaf);
4250 free_out:
4251 	kfree(block);
4252 	free_extent_buffer(reloc_eb);
4253 out:
4254 	if (ret < 0) {
4255 		btrfs_err_rl(fs_info,
4256 			     "failed to account subtree at bytenr %llu: %d",
4257 			     subvol_eb->start, ret);
4258 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4259 	}
4260 	return ret;
4261 }
4262 
4263 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4264 {
4265 	struct btrfs_qgroup_extent_record *entry;
4266 	struct btrfs_qgroup_extent_record *next;
4267 	struct rb_root *root;
4268 
4269 	root = &trans->delayed_refs.dirty_extent_root;
4270 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4271 		ulist_free(entry->old_roots);
4272 		kfree(entry);
4273 	}
4274 }
4275