xref: /openbmc/linux/fs/btrfs/qgroup.c (revision 4bce6fce)
1 /*
2  * Copyright (C) 2011 STRATO.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
25 #include <linux/workqueue.h>
26 #include <linux/btrfs.h>
27 
28 #include "ctree.h"
29 #include "transaction.h"
30 #include "disk-io.h"
31 #include "locking.h"
32 #include "ulist.h"
33 #include "backref.h"
34 #include "extent_io.h"
35 #include "qgroup.h"
36 
37 /* TODO XXX FIXME
38  *  - subvol delete -> delete when ref goes to 0? delete limits also?
39  *  - reorganize keys
40  *  - compressed
41  *  - sync
42  *  - copy also limits on subvol creation
43  *  - limit
44  *  - caches fuer ulists
45  *  - performance benchmarks
46  *  - check all ioctl parameters
47  */
48 
49 /*
50  * one struct for each qgroup, organized in fs_info->qgroup_tree.
51  */
52 struct btrfs_qgroup {
53 	u64 qgroupid;
54 
55 	/*
56 	 * state
57 	 */
58 	u64 rfer;	/* referenced */
59 	u64 rfer_cmpr;	/* referenced compressed */
60 	u64 excl;	/* exclusive */
61 	u64 excl_cmpr;	/* exclusive compressed */
62 
63 	/*
64 	 * limits
65 	 */
66 	u64 lim_flags;	/* which limits are set */
67 	u64 max_rfer;
68 	u64 max_excl;
69 	u64 rsv_rfer;
70 	u64 rsv_excl;
71 
72 	/*
73 	 * reservation tracking
74 	 */
75 	u64 reserved;
76 
77 	/*
78 	 * lists
79 	 */
80 	struct list_head groups;  /* groups this group is member of */
81 	struct list_head members; /* groups that are members of this group */
82 	struct list_head dirty;   /* dirty groups */
83 	struct rb_node node;	  /* tree of qgroups */
84 
85 	/*
86 	 * temp variables for accounting operations
87 	 */
88 	u64 old_refcnt;
89 	u64 new_refcnt;
90 };
91 
92 /*
93  * glue structure to represent the relations between qgroups.
94  */
95 struct btrfs_qgroup_list {
96 	struct list_head next_group;
97 	struct list_head next_member;
98 	struct btrfs_qgroup *group;
99 	struct btrfs_qgroup *member;
100 };
101 
102 #define ptr_to_u64(x) ((u64)(uintptr_t)x)
103 #define u64_to_ptr(x) ((struct btrfs_qgroup *)(uintptr_t)x)
104 
105 static int
106 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
107 		   int init_flags);
108 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
109 
110 /* must be called with qgroup_ioctl_lock held */
111 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
112 					   u64 qgroupid)
113 {
114 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
115 	struct btrfs_qgroup *qgroup;
116 
117 	while (n) {
118 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
119 		if (qgroup->qgroupid < qgroupid)
120 			n = n->rb_left;
121 		else if (qgroup->qgroupid > qgroupid)
122 			n = n->rb_right;
123 		else
124 			return qgroup;
125 	}
126 	return NULL;
127 }
128 
129 /* must be called with qgroup_lock held */
130 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
131 					  u64 qgroupid)
132 {
133 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
134 	struct rb_node *parent = NULL;
135 	struct btrfs_qgroup *qgroup;
136 
137 	while (*p) {
138 		parent = *p;
139 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
140 
141 		if (qgroup->qgroupid < qgroupid)
142 			p = &(*p)->rb_left;
143 		else if (qgroup->qgroupid > qgroupid)
144 			p = &(*p)->rb_right;
145 		else
146 			return qgroup;
147 	}
148 
149 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
150 	if (!qgroup)
151 		return ERR_PTR(-ENOMEM);
152 
153 	qgroup->qgroupid = qgroupid;
154 	INIT_LIST_HEAD(&qgroup->groups);
155 	INIT_LIST_HEAD(&qgroup->members);
156 	INIT_LIST_HEAD(&qgroup->dirty);
157 
158 	rb_link_node(&qgroup->node, parent, p);
159 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
160 
161 	return qgroup;
162 }
163 
164 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
165 {
166 	struct btrfs_qgroup_list *list;
167 
168 	list_del(&qgroup->dirty);
169 	while (!list_empty(&qgroup->groups)) {
170 		list = list_first_entry(&qgroup->groups,
171 					struct btrfs_qgroup_list, next_group);
172 		list_del(&list->next_group);
173 		list_del(&list->next_member);
174 		kfree(list);
175 	}
176 
177 	while (!list_empty(&qgroup->members)) {
178 		list = list_first_entry(&qgroup->members,
179 					struct btrfs_qgroup_list, next_member);
180 		list_del(&list->next_group);
181 		list_del(&list->next_member);
182 		kfree(list);
183 	}
184 	kfree(qgroup);
185 }
186 
187 /* must be called with qgroup_lock held */
188 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
189 {
190 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
191 
192 	if (!qgroup)
193 		return -ENOENT;
194 
195 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
196 	__del_qgroup_rb(qgroup);
197 	return 0;
198 }
199 
200 /* must be called with qgroup_lock held */
201 static int add_relation_rb(struct btrfs_fs_info *fs_info,
202 			   u64 memberid, u64 parentid)
203 {
204 	struct btrfs_qgroup *member;
205 	struct btrfs_qgroup *parent;
206 	struct btrfs_qgroup_list *list;
207 
208 	member = find_qgroup_rb(fs_info, memberid);
209 	parent = find_qgroup_rb(fs_info, parentid);
210 	if (!member || !parent)
211 		return -ENOENT;
212 
213 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
214 	if (!list)
215 		return -ENOMEM;
216 
217 	list->group = parent;
218 	list->member = member;
219 	list_add_tail(&list->next_group, &member->groups);
220 	list_add_tail(&list->next_member, &parent->members);
221 
222 	return 0;
223 }
224 
225 /* must be called with qgroup_lock held */
226 static int del_relation_rb(struct btrfs_fs_info *fs_info,
227 			   u64 memberid, u64 parentid)
228 {
229 	struct btrfs_qgroup *member;
230 	struct btrfs_qgroup *parent;
231 	struct btrfs_qgroup_list *list;
232 
233 	member = find_qgroup_rb(fs_info, memberid);
234 	parent = find_qgroup_rb(fs_info, parentid);
235 	if (!member || !parent)
236 		return -ENOENT;
237 
238 	list_for_each_entry(list, &member->groups, next_group) {
239 		if (list->group == parent) {
240 			list_del(&list->next_group);
241 			list_del(&list->next_member);
242 			kfree(list);
243 			return 0;
244 		}
245 	}
246 	return -ENOENT;
247 }
248 
249 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
250 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
251 			       u64 rfer, u64 excl)
252 {
253 	struct btrfs_qgroup *qgroup;
254 
255 	qgroup = find_qgroup_rb(fs_info, qgroupid);
256 	if (!qgroup)
257 		return -EINVAL;
258 	if (qgroup->rfer != rfer || qgroup->excl != excl)
259 		return -EINVAL;
260 	return 0;
261 }
262 #endif
263 
264 /*
265  * The full config is read in one go, only called from open_ctree()
266  * It doesn't use any locking, as at this point we're still single-threaded
267  */
268 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
269 {
270 	struct btrfs_key key;
271 	struct btrfs_key found_key;
272 	struct btrfs_root *quota_root = fs_info->quota_root;
273 	struct btrfs_path *path = NULL;
274 	struct extent_buffer *l;
275 	int slot;
276 	int ret = 0;
277 	u64 flags = 0;
278 	u64 rescan_progress = 0;
279 
280 	if (!fs_info->quota_enabled)
281 		return 0;
282 
283 	fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
284 	if (!fs_info->qgroup_ulist) {
285 		ret = -ENOMEM;
286 		goto out;
287 	}
288 
289 	path = btrfs_alloc_path();
290 	if (!path) {
291 		ret = -ENOMEM;
292 		goto out;
293 	}
294 
295 	/* default this to quota off, in case no status key is found */
296 	fs_info->qgroup_flags = 0;
297 
298 	/*
299 	 * pass 1: read status, all qgroup infos and limits
300 	 */
301 	key.objectid = 0;
302 	key.type = 0;
303 	key.offset = 0;
304 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
305 	if (ret)
306 		goto out;
307 
308 	while (1) {
309 		struct btrfs_qgroup *qgroup;
310 
311 		slot = path->slots[0];
312 		l = path->nodes[0];
313 		btrfs_item_key_to_cpu(l, &found_key, slot);
314 
315 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
316 			struct btrfs_qgroup_status_item *ptr;
317 
318 			ptr = btrfs_item_ptr(l, slot,
319 					     struct btrfs_qgroup_status_item);
320 
321 			if (btrfs_qgroup_status_version(l, ptr) !=
322 			    BTRFS_QGROUP_STATUS_VERSION) {
323 				btrfs_err(fs_info,
324 				 "old qgroup version, quota disabled");
325 				goto out;
326 			}
327 			if (btrfs_qgroup_status_generation(l, ptr) !=
328 			    fs_info->generation) {
329 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
330 				btrfs_err(fs_info,
331 					"qgroup generation mismatch, "
332 					"marked as inconsistent");
333 			}
334 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
335 									  ptr);
336 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
337 			goto next1;
338 		}
339 
340 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
341 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
342 			goto next1;
343 
344 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
345 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
346 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
347 			btrfs_err(fs_info, "inconsitent qgroup config");
348 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
349 		}
350 		if (!qgroup) {
351 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
352 			if (IS_ERR(qgroup)) {
353 				ret = PTR_ERR(qgroup);
354 				goto out;
355 			}
356 		}
357 		switch (found_key.type) {
358 		case BTRFS_QGROUP_INFO_KEY: {
359 			struct btrfs_qgroup_info_item *ptr;
360 
361 			ptr = btrfs_item_ptr(l, slot,
362 					     struct btrfs_qgroup_info_item);
363 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
364 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
365 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
366 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
367 			/* generation currently unused */
368 			break;
369 		}
370 		case BTRFS_QGROUP_LIMIT_KEY: {
371 			struct btrfs_qgroup_limit_item *ptr;
372 
373 			ptr = btrfs_item_ptr(l, slot,
374 					     struct btrfs_qgroup_limit_item);
375 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
376 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
377 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
378 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
379 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
380 			break;
381 		}
382 		}
383 next1:
384 		ret = btrfs_next_item(quota_root, path);
385 		if (ret < 0)
386 			goto out;
387 		if (ret)
388 			break;
389 	}
390 	btrfs_release_path(path);
391 
392 	/*
393 	 * pass 2: read all qgroup relations
394 	 */
395 	key.objectid = 0;
396 	key.type = BTRFS_QGROUP_RELATION_KEY;
397 	key.offset = 0;
398 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
399 	if (ret)
400 		goto out;
401 	while (1) {
402 		slot = path->slots[0];
403 		l = path->nodes[0];
404 		btrfs_item_key_to_cpu(l, &found_key, slot);
405 
406 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
407 			goto next2;
408 
409 		if (found_key.objectid > found_key.offset) {
410 			/* parent <- member, not needed to build config */
411 			/* FIXME should we omit the key completely? */
412 			goto next2;
413 		}
414 
415 		ret = add_relation_rb(fs_info, found_key.objectid,
416 				      found_key.offset);
417 		if (ret == -ENOENT) {
418 			btrfs_warn(fs_info,
419 				"orphan qgroup relation 0x%llx->0x%llx",
420 				found_key.objectid, found_key.offset);
421 			ret = 0;	/* ignore the error */
422 		}
423 		if (ret)
424 			goto out;
425 next2:
426 		ret = btrfs_next_item(quota_root, path);
427 		if (ret < 0)
428 			goto out;
429 		if (ret)
430 			break;
431 	}
432 out:
433 	fs_info->qgroup_flags |= flags;
434 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
435 		fs_info->quota_enabled = 0;
436 		fs_info->pending_quota_state = 0;
437 	} else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
438 		   ret >= 0) {
439 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
440 	}
441 	btrfs_free_path(path);
442 
443 	if (ret < 0) {
444 		ulist_free(fs_info->qgroup_ulist);
445 		fs_info->qgroup_ulist = NULL;
446 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
447 	}
448 
449 	return ret < 0 ? ret : 0;
450 }
451 
452 /*
453  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
454  * first two are in single-threaded paths.And for the third one, we have set
455  * quota_root to be null with qgroup_lock held before, so it is safe to clean
456  * up the in-memory structures without qgroup_lock held.
457  */
458 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
459 {
460 	struct rb_node *n;
461 	struct btrfs_qgroup *qgroup;
462 
463 	while ((n = rb_first(&fs_info->qgroup_tree))) {
464 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
465 		rb_erase(n, &fs_info->qgroup_tree);
466 		__del_qgroup_rb(qgroup);
467 	}
468 	/*
469 	 * we call btrfs_free_qgroup_config() when umounting
470 	 * filesystem and disabling quota, so we set qgroup_ulit
471 	 * to be null here to avoid double free.
472 	 */
473 	ulist_free(fs_info->qgroup_ulist);
474 	fs_info->qgroup_ulist = NULL;
475 }
476 
477 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
478 				    struct btrfs_root *quota_root,
479 				    u64 src, u64 dst)
480 {
481 	int ret;
482 	struct btrfs_path *path;
483 	struct btrfs_key key;
484 
485 	path = btrfs_alloc_path();
486 	if (!path)
487 		return -ENOMEM;
488 
489 	key.objectid = src;
490 	key.type = BTRFS_QGROUP_RELATION_KEY;
491 	key.offset = dst;
492 
493 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
494 
495 	btrfs_mark_buffer_dirty(path->nodes[0]);
496 
497 	btrfs_free_path(path);
498 	return ret;
499 }
500 
501 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
502 				    struct btrfs_root *quota_root,
503 				    u64 src, u64 dst)
504 {
505 	int ret;
506 	struct btrfs_path *path;
507 	struct btrfs_key key;
508 
509 	path = btrfs_alloc_path();
510 	if (!path)
511 		return -ENOMEM;
512 
513 	key.objectid = src;
514 	key.type = BTRFS_QGROUP_RELATION_KEY;
515 	key.offset = dst;
516 
517 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
518 	if (ret < 0)
519 		goto out;
520 
521 	if (ret > 0) {
522 		ret = -ENOENT;
523 		goto out;
524 	}
525 
526 	ret = btrfs_del_item(trans, quota_root, path);
527 out:
528 	btrfs_free_path(path);
529 	return ret;
530 }
531 
532 static int add_qgroup_item(struct btrfs_trans_handle *trans,
533 			   struct btrfs_root *quota_root, u64 qgroupid)
534 {
535 	int ret;
536 	struct btrfs_path *path;
537 	struct btrfs_qgroup_info_item *qgroup_info;
538 	struct btrfs_qgroup_limit_item *qgroup_limit;
539 	struct extent_buffer *leaf;
540 	struct btrfs_key key;
541 
542 	if (btrfs_test_is_dummy_root(quota_root))
543 		return 0;
544 
545 	path = btrfs_alloc_path();
546 	if (!path)
547 		return -ENOMEM;
548 
549 	key.objectid = 0;
550 	key.type = BTRFS_QGROUP_INFO_KEY;
551 	key.offset = qgroupid;
552 
553 	/*
554 	 * Avoid a transaction abort by catching -EEXIST here. In that
555 	 * case, we proceed by re-initializing the existing structure
556 	 * on disk.
557 	 */
558 
559 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
560 				      sizeof(*qgroup_info));
561 	if (ret && ret != -EEXIST)
562 		goto out;
563 
564 	leaf = path->nodes[0];
565 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
566 				 struct btrfs_qgroup_info_item);
567 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
568 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
569 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
570 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
571 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
572 
573 	btrfs_mark_buffer_dirty(leaf);
574 
575 	btrfs_release_path(path);
576 
577 	key.type = BTRFS_QGROUP_LIMIT_KEY;
578 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
579 				      sizeof(*qgroup_limit));
580 	if (ret && ret != -EEXIST)
581 		goto out;
582 
583 	leaf = path->nodes[0];
584 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
585 				  struct btrfs_qgroup_limit_item);
586 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
587 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
588 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
589 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
590 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
591 
592 	btrfs_mark_buffer_dirty(leaf);
593 
594 	ret = 0;
595 out:
596 	btrfs_free_path(path);
597 	return ret;
598 }
599 
600 static int del_qgroup_item(struct btrfs_trans_handle *trans,
601 			   struct btrfs_root *quota_root, u64 qgroupid)
602 {
603 	int ret;
604 	struct btrfs_path *path;
605 	struct btrfs_key key;
606 
607 	path = btrfs_alloc_path();
608 	if (!path)
609 		return -ENOMEM;
610 
611 	key.objectid = 0;
612 	key.type = BTRFS_QGROUP_INFO_KEY;
613 	key.offset = qgroupid;
614 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
615 	if (ret < 0)
616 		goto out;
617 
618 	if (ret > 0) {
619 		ret = -ENOENT;
620 		goto out;
621 	}
622 
623 	ret = btrfs_del_item(trans, quota_root, path);
624 	if (ret)
625 		goto out;
626 
627 	btrfs_release_path(path);
628 
629 	key.type = BTRFS_QGROUP_LIMIT_KEY;
630 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
631 	if (ret < 0)
632 		goto out;
633 
634 	if (ret > 0) {
635 		ret = -ENOENT;
636 		goto out;
637 	}
638 
639 	ret = btrfs_del_item(trans, quota_root, path);
640 
641 out:
642 	btrfs_free_path(path);
643 	return ret;
644 }
645 
646 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
647 				    struct btrfs_root *root,
648 				    struct btrfs_qgroup *qgroup)
649 {
650 	struct btrfs_path *path;
651 	struct btrfs_key key;
652 	struct extent_buffer *l;
653 	struct btrfs_qgroup_limit_item *qgroup_limit;
654 	int ret;
655 	int slot;
656 
657 	key.objectid = 0;
658 	key.type = BTRFS_QGROUP_LIMIT_KEY;
659 	key.offset = qgroup->qgroupid;
660 
661 	path = btrfs_alloc_path();
662 	if (!path)
663 		return -ENOMEM;
664 
665 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
666 	if (ret > 0)
667 		ret = -ENOENT;
668 
669 	if (ret)
670 		goto out;
671 
672 	l = path->nodes[0];
673 	slot = path->slots[0];
674 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
675 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
676 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
677 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
678 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
679 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
680 
681 	btrfs_mark_buffer_dirty(l);
682 
683 out:
684 	btrfs_free_path(path);
685 	return ret;
686 }
687 
688 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
689 				   struct btrfs_root *root,
690 				   struct btrfs_qgroup *qgroup)
691 {
692 	struct btrfs_path *path;
693 	struct btrfs_key key;
694 	struct extent_buffer *l;
695 	struct btrfs_qgroup_info_item *qgroup_info;
696 	int ret;
697 	int slot;
698 
699 	if (btrfs_test_is_dummy_root(root))
700 		return 0;
701 
702 	key.objectid = 0;
703 	key.type = BTRFS_QGROUP_INFO_KEY;
704 	key.offset = qgroup->qgroupid;
705 
706 	path = btrfs_alloc_path();
707 	if (!path)
708 		return -ENOMEM;
709 
710 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
711 	if (ret > 0)
712 		ret = -ENOENT;
713 
714 	if (ret)
715 		goto out;
716 
717 	l = path->nodes[0];
718 	slot = path->slots[0];
719 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
720 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
721 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
722 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
723 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
724 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
725 
726 	btrfs_mark_buffer_dirty(l);
727 
728 out:
729 	btrfs_free_path(path);
730 	return ret;
731 }
732 
733 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
734 				     struct btrfs_fs_info *fs_info,
735 				    struct btrfs_root *root)
736 {
737 	struct btrfs_path *path;
738 	struct btrfs_key key;
739 	struct extent_buffer *l;
740 	struct btrfs_qgroup_status_item *ptr;
741 	int ret;
742 	int slot;
743 
744 	key.objectid = 0;
745 	key.type = BTRFS_QGROUP_STATUS_KEY;
746 	key.offset = 0;
747 
748 	path = btrfs_alloc_path();
749 	if (!path)
750 		return -ENOMEM;
751 
752 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
753 	if (ret > 0)
754 		ret = -ENOENT;
755 
756 	if (ret)
757 		goto out;
758 
759 	l = path->nodes[0];
760 	slot = path->slots[0];
761 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
762 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
763 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
764 	btrfs_set_qgroup_status_rescan(l, ptr,
765 				fs_info->qgroup_rescan_progress.objectid);
766 
767 	btrfs_mark_buffer_dirty(l);
768 
769 out:
770 	btrfs_free_path(path);
771 	return ret;
772 }
773 
774 /*
775  * called with qgroup_lock held
776  */
777 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
778 				  struct btrfs_root *root)
779 {
780 	struct btrfs_path *path;
781 	struct btrfs_key key;
782 	struct extent_buffer *leaf = NULL;
783 	int ret;
784 	int nr = 0;
785 
786 	path = btrfs_alloc_path();
787 	if (!path)
788 		return -ENOMEM;
789 
790 	path->leave_spinning = 1;
791 
792 	key.objectid = 0;
793 	key.offset = 0;
794 	key.type = 0;
795 
796 	while (1) {
797 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
798 		if (ret < 0)
799 			goto out;
800 		leaf = path->nodes[0];
801 		nr = btrfs_header_nritems(leaf);
802 		if (!nr)
803 			break;
804 		/*
805 		 * delete the leaf one by one
806 		 * since the whole tree is going
807 		 * to be deleted.
808 		 */
809 		path->slots[0] = 0;
810 		ret = btrfs_del_items(trans, root, path, 0, nr);
811 		if (ret)
812 			goto out;
813 
814 		btrfs_release_path(path);
815 	}
816 	ret = 0;
817 out:
818 	root->fs_info->pending_quota_state = 0;
819 	btrfs_free_path(path);
820 	return ret;
821 }
822 
823 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
824 		       struct btrfs_fs_info *fs_info)
825 {
826 	struct btrfs_root *quota_root;
827 	struct btrfs_root *tree_root = fs_info->tree_root;
828 	struct btrfs_path *path = NULL;
829 	struct btrfs_qgroup_status_item *ptr;
830 	struct extent_buffer *leaf;
831 	struct btrfs_key key;
832 	struct btrfs_key found_key;
833 	struct btrfs_qgroup *qgroup = NULL;
834 	int ret = 0;
835 	int slot;
836 
837 	mutex_lock(&fs_info->qgroup_ioctl_lock);
838 	if (fs_info->quota_root) {
839 		fs_info->pending_quota_state = 1;
840 		goto out;
841 	}
842 
843 	fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
844 	if (!fs_info->qgroup_ulist) {
845 		ret = -ENOMEM;
846 		goto out;
847 	}
848 
849 	/*
850 	 * initially create the quota tree
851 	 */
852 	quota_root = btrfs_create_tree(trans, fs_info,
853 				       BTRFS_QUOTA_TREE_OBJECTID);
854 	if (IS_ERR(quota_root)) {
855 		ret =  PTR_ERR(quota_root);
856 		goto out;
857 	}
858 
859 	path = btrfs_alloc_path();
860 	if (!path) {
861 		ret = -ENOMEM;
862 		goto out_free_root;
863 	}
864 
865 	key.objectid = 0;
866 	key.type = BTRFS_QGROUP_STATUS_KEY;
867 	key.offset = 0;
868 
869 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
870 				      sizeof(*ptr));
871 	if (ret)
872 		goto out_free_path;
873 
874 	leaf = path->nodes[0];
875 	ptr = btrfs_item_ptr(leaf, path->slots[0],
876 				 struct btrfs_qgroup_status_item);
877 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
878 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
879 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
880 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
881 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
882 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
883 
884 	btrfs_mark_buffer_dirty(leaf);
885 
886 	key.objectid = 0;
887 	key.type = BTRFS_ROOT_REF_KEY;
888 	key.offset = 0;
889 
890 	btrfs_release_path(path);
891 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
892 	if (ret > 0)
893 		goto out_add_root;
894 	if (ret < 0)
895 		goto out_free_path;
896 
897 
898 	while (1) {
899 		slot = path->slots[0];
900 		leaf = path->nodes[0];
901 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
902 
903 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
904 			ret = add_qgroup_item(trans, quota_root,
905 					      found_key.offset);
906 			if (ret)
907 				goto out_free_path;
908 
909 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
910 			if (IS_ERR(qgroup)) {
911 				ret = PTR_ERR(qgroup);
912 				goto out_free_path;
913 			}
914 		}
915 		ret = btrfs_next_item(tree_root, path);
916 		if (ret < 0)
917 			goto out_free_path;
918 		if (ret)
919 			break;
920 	}
921 
922 out_add_root:
923 	btrfs_release_path(path);
924 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
925 	if (ret)
926 		goto out_free_path;
927 
928 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
929 	if (IS_ERR(qgroup)) {
930 		ret = PTR_ERR(qgroup);
931 		goto out_free_path;
932 	}
933 	spin_lock(&fs_info->qgroup_lock);
934 	fs_info->quota_root = quota_root;
935 	fs_info->pending_quota_state = 1;
936 	spin_unlock(&fs_info->qgroup_lock);
937 out_free_path:
938 	btrfs_free_path(path);
939 out_free_root:
940 	if (ret) {
941 		free_extent_buffer(quota_root->node);
942 		free_extent_buffer(quota_root->commit_root);
943 		kfree(quota_root);
944 	}
945 out:
946 	if (ret) {
947 		ulist_free(fs_info->qgroup_ulist);
948 		fs_info->qgroup_ulist = NULL;
949 	}
950 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
951 	return ret;
952 }
953 
954 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
955 			struct btrfs_fs_info *fs_info)
956 {
957 	struct btrfs_root *tree_root = fs_info->tree_root;
958 	struct btrfs_root *quota_root;
959 	int ret = 0;
960 
961 	mutex_lock(&fs_info->qgroup_ioctl_lock);
962 	if (!fs_info->quota_root)
963 		goto out;
964 	spin_lock(&fs_info->qgroup_lock);
965 	fs_info->quota_enabled = 0;
966 	fs_info->pending_quota_state = 0;
967 	quota_root = fs_info->quota_root;
968 	fs_info->quota_root = NULL;
969 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
970 	spin_unlock(&fs_info->qgroup_lock);
971 
972 	btrfs_free_qgroup_config(fs_info);
973 
974 	ret = btrfs_clean_quota_tree(trans, quota_root);
975 	if (ret)
976 		goto out;
977 
978 	ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
979 	if (ret)
980 		goto out;
981 
982 	list_del(&quota_root->dirty_list);
983 
984 	btrfs_tree_lock(quota_root->node);
985 	clean_tree_block(trans, tree_root->fs_info, quota_root->node);
986 	btrfs_tree_unlock(quota_root->node);
987 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
988 
989 	free_extent_buffer(quota_root->node);
990 	free_extent_buffer(quota_root->commit_root);
991 	kfree(quota_root);
992 out:
993 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
994 	return ret;
995 }
996 
997 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
998 			 struct btrfs_qgroup *qgroup)
999 {
1000 	if (list_empty(&qgroup->dirty))
1001 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1002 }
1003 
1004 /*
1005  * The easy accounting, if we are adding/removing the only ref for an extent
1006  * then this qgroup and all of the parent qgroups get their refrence and
1007  * exclusive counts adjusted.
1008  *
1009  * Caller should hold fs_info->qgroup_lock.
1010  */
1011 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1012 				    struct ulist *tmp, u64 ref_root,
1013 				    u64 num_bytes, int sign)
1014 {
1015 	struct btrfs_qgroup *qgroup;
1016 	struct btrfs_qgroup_list *glist;
1017 	struct ulist_node *unode;
1018 	struct ulist_iterator uiter;
1019 	int ret = 0;
1020 
1021 	qgroup = find_qgroup_rb(fs_info, ref_root);
1022 	if (!qgroup)
1023 		goto out;
1024 
1025 	qgroup->rfer += sign * num_bytes;
1026 	qgroup->rfer_cmpr += sign * num_bytes;
1027 
1028 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1029 	qgroup->excl += sign * num_bytes;
1030 	qgroup->excl_cmpr += sign * num_bytes;
1031 	if (sign > 0)
1032 		qgroup->reserved -= num_bytes;
1033 
1034 	qgroup_dirty(fs_info, qgroup);
1035 
1036 	/* Get all of the parent groups that contain this qgroup */
1037 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1038 		ret = ulist_add(tmp, glist->group->qgroupid,
1039 				ptr_to_u64(glist->group), GFP_ATOMIC);
1040 		if (ret < 0)
1041 			goto out;
1042 	}
1043 
1044 	/* Iterate all of the parents and adjust their reference counts */
1045 	ULIST_ITER_INIT(&uiter);
1046 	while ((unode = ulist_next(tmp, &uiter))) {
1047 		qgroup = u64_to_ptr(unode->aux);
1048 		qgroup->rfer += sign * num_bytes;
1049 		qgroup->rfer_cmpr += sign * num_bytes;
1050 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1051 		qgroup->excl += sign * num_bytes;
1052 		if (sign > 0)
1053 			qgroup->reserved -= num_bytes;
1054 		qgroup->excl_cmpr += sign * num_bytes;
1055 		qgroup_dirty(fs_info, qgroup);
1056 
1057 		/* Add any parents of the parents */
1058 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1059 			ret = ulist_add(tmp, glist->group->qgroupid,
1060 					ptr_to_u64(glist->group), GFP_ATOMIC);
1061 			if (ret < 0)
1062 				goto out;
1063 		}
1064 	}
1065 	ret = 0;
1066 out:
1067 	return ret;
1068 }
1069 
1070 
1071 /*
1072  * Quick path for updating qgroup with only excl refs.
1073  *
1074  * In that case, just update all parent will be enough.
1075  * Or we needs to do a full rescan.
1076  * Caller should also hold fs_info->qgroup_lock.
1077  *
1078  * Return 0 for quick update, return >0 for need to full rescan
1079  * and mark INCONSISTENT flag.
1080  * Return < 0 for other error.
1081  */
1082 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1083 				   struct ulist *tmp, u64 src, u64 dst,
1084 				   int sign)
1085 {
1086 	struct btrfs_qgroup *qgroup;
1087 	int ret = 1;
1088 	int err = 0;
1089 
1090 	qgroup = find_qgroup_rb(fs_info, src);
1091 	if (!qgroup)
1092 		goto out;
1093 	if (qgroup->excl == qgroup->rfer) {
1094 		ret = 0;
1095 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1096 					       qgroup->excl, sign);
1097 		if (err < 0) {
1098 			ret = err;
1099 			goto out;
1100 		}
1101 	}
1102 out:
1103 	if (ret)
1104 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1105 	return ret;
1106 }
1107 
1108 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1109 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1110 {
1111 	struct btrfs_root *quota_root;
1112 	struct btrfs_qgroup *parent;
1113 	struct btrfs_qgroup *member;
1114 	struct btrfs_qgroup_list *list;
1115 	struct ulist *tmp;
1116 	int ret = 0;
1117 
1118 	tmp = ulist_alloc(GFP_NOFS);
1119 	if (!tmp)
1120 		return -ENOMEM;
1121 
1122 	/* Check the level of src and dst first */
1123 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1124 		return -EINVAL;
1125 
1126 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1127 	quota_root = fs_info->quota_root;
1128 	if (!quota_root) {
1129 		ret = -EINVAL;
1130 		goto out;
1131 	}
1132 	member = find_qgroup_rb(fs_info, src);
1133 	parent = find_qgroup_rb(fs_info, dst);
1134 	if (!member || !parent) {
1135 		ret = -EINVAL;
1136 		goto out;
1137 	}
1138 
1139 	/* check if such qgroup relation exist firstly */
1140 	list_for_each_entry(list, &member->groups, next_group) {
1141 		if (list->group == parent) {
1142 			ret = -EEXIST;
1143 			goto out;
1144 		}
1145 	}
1146 
1147 	ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1148 	if (ret)
1149 		goto out;
1150 
1151 	ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1152 	if (ret) {
1153 		del_qgroup_relation_item(trans, quota_root, src, dst);
1154 		goto out;
1155 	}
1156 
1157 	spin_lock(&fs_info->qgroup_lock);
1158 	ret = add_relation_rb(quota_root->fs_info, src, dst);
1159 	if (ret < 0) {
1160 		spin_unlock(&fs_info->qgroup_lock);
1161 		goto out;
1162 	}
1163 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1164 	spin_unlock(&fs_info->qgroup_lock);
1165 out:
1166 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1167 	ulist_free(tmp);
1168 	return ret;
1169 }
1170 
1171 int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1172 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1173 {
1174 	struct btrfs_root *quota_root;
1175 	struct btrfs_qgroup *parent;
1176 	struct btrfs_qgroup *member;
1177 	struct btrfs_qgroup_list *list;
1178 	struct ulist *tmp;
1179 	int ret = 0;
1180 	int err;
1181 
1182 	tmp = ulist_alloc(GFP_NOFS);
1183 	if (!tmp)
1184 		return -ENOMEM;
1185 
1186 	quota_root = fs_info->quota_root;
1187 	if (!quota_root) {
1188 		ret = -EINVAL;
1189 		goto out;
1190 	}
1191 
1192 	member = find_qgroup_rb(fs_info, src);
1193 	parent = find_qgroup_rb(fs_info, dst);
1194 	if (!member || !parent) {
1195 		ret = -EINVAL;
1196 		goto out;
1197 	}
1198 
1199 	/* check if such qgroup relation exist firstly */
1200 	list_for_each_entry(list, &member->groups, next_group) {
1201 		if (list->group == parent)
1202 			goto exist;
1203 	}
1204 	ret = -ENOENT;
1205 	goto out;
1206 exist:
1207 	ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1208 	err = del_qgroup_relation_item(trans, quota_root, dst, src);
1209 	if (err && !ret)
1210 		ret = err;
1211 
1212 	spin_lock(&fs_info->qgroup_lock);
1213 	del_relation_rb(fs_info, src, dst);
1214 	ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1215 	spin_unlock(&fs_info->qgroup_lock);
1216 out:
1217 	ulist_free(tmp);
1218 	return ret;
1219 }
1220 
1221 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1222 			      struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1223 {
1224 	int ret = 0;
1225 
1226 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1227 	ret = __del_qgroup_relation(trans, fs_info, src, dst);
1228 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1229 
1230 	return ret;
1231 }
1232 
1233 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1234 			struct btrfs_fs_info *fs_info, u64 qgroupid)
1235 {
1236 	struct btrfs_root *quota_root;
1237 	struct btrfs_qgroup *qgroup;
1238 	int ret = 0;
1239 
1240 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1241 	quota_root = fs_info->quota_root;
1242 	if (!quota_root) {
1243 		ret = -EINVAL;
1244 		goto out;
1245 	}
1246 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1247 	if (qgroup) {
1248 		ret = -EEXIST;
1249 		goto out;
1250 	}
1251 
1252 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1253 	if (ret)
1254 		goto out;
1255 
1256 	spin_lock(&fs_info->qgroup_lock);
1257 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1258 	spin_unlock(&fs_info->qgroup_lock);
1259 
1260 	if (IS_ERR(qgroup))
1261 		ret = PTR_ERR(qgroup);
1262 out:
1263 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1264 	return ret;
1265 }
1266 
1267 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1268 			struct btrfs_fs_info *fs_info, u64 qgroupid)
1269 {
1270 	struct btrfs_root *quota_root;
1271 	struct btrfs_qgroup *qgroup;
1272 	struct btrfs_qgroup_list *list;
1273 	int ret = 0;
1274 
1275 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1276 	quota_root = fs_info->quota_root;
1277 	if (!quota_root) {
1278 		ret = -EINVAL;
1279 		goto out;
1280 	}
1281 
1282 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1283 	if (!qgroup) {
1284 		ret = -ENOENT;
1285 		goto out;
1286 	} else {
1287 		/* check if there are no children of this qgroup */
1288 		if (!list_empty(&qgroup->members)) {
1289 			ret = -EBUSY;
1290 			goto out;
1291 		}
1292 	}
1293 	ret = del_qgroup_item(trans, quota_root, qgroupid);
1294 
1295 	while (!list_empty(&qgroup->groups)) {
1296 		list = list_first_entry(&qgroup->groups,
1297 					struct btrfs_qgroup_list, next_group);
1298 		ret = __del_qgroup_relation(trans, fs_info,
1299 					   qgroupid,
1300 					   list->group->qgroupid);
1301 		if (ret)
1302 			goto out;
1303 	}
1304 
1305 	spin_lock(&fs_info->qgroup_lock);
1306 	del_qgroup_rb(quota_root->fs_info, qgroupid);
1307 	spin_unlock(&fs_info->qgroup_lock);
1308 out:
1309 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1310 	return ret;
1311 }
1312 
1313 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1314 		       struct btrfs_fs_info *fs_info, u64 qgroupid,
1315 		       struct btrfs_qgroup_limit *limit)
1316 {
1317 	struct btrfs_root *quota_root;
1318 	struct btrfs_qgroup *qgroup;
1319 	int ret = 0;
1320 
1321 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1322 	quota_root = fs_info->quota_root;
1323 	if (!quota_root) {
1324 		ret = -EINVAL;
1325 		goto out;
1326 	}
1327 
1328 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1329 	if (!qgroup) {
1330 		ret = -ENOENT;
1331 		goto out;
1332 	}
1333 
1334 	spin_lock(&fs_info->qgroup_lock);
1335 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER)
1336 		qgroup->max_rfer = limit->max_rfer;
1337 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
1338 		qgroup->max_excl = limit->max_excl;
1339 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER)
1340 		qgroup->rsv_rfer = limit->rsv_rfer;
1341 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL)
1342 		qgroup->rsv_excl = limit->rsv_excl;
1343 	qgroup->lim_flags |= limit->flags;
1344 
1345 	spin_unlock(&fs_info->qgroup_lock);
1346 
1347 	ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1348 	if (ret) {
1349 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1350 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1351 		       qgroupid);
1352 	}
1353 
1354 out:
1355 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1356 	return ret;
1357 }
1358 
1359 static int comp_oper_exist(struct btrfs_qgroup_operation *oper1,
1360 			   struct btrfs_qgroup_operation *oper2)
1361 {
1362 	/*
1363 	 * Ignore seq and type here, we're looking for any operation
1364 	 * at all related to this extent on that root.
1365 	 */
1366 	if (oper1->bytenr < oper2->bytenr)
1367 		return -1;
1368 	if (oper1->bytenr > oper2->bytenr)
1369 		return 1;
1370 	if (oper1->ref_root < oper2->ref_root)
1371 		return -1;
1372 	if (oper1->ref_root > oper2->ref_root)
1373 		return 1;
1374 	return 0;
1375 }
1376 
1377 static int qgroup_oper_exists(struct btrfs_fs_info *fs_info,
1378 			      struct btrfs_qgroup_operation *oper)
1379 {
1380 	struct rb_node *n;
1381 	struct btrfs_qgroup_operation *cur;
1382 	int cmp;
1383 
1384 	spin_lock(&fs_info->qgroup_op_lock);
1385 	n = fs_info->qgroup_op_tree.rb_node;
1386 	while (n) {
1387 		cur = rb_entry(n, struct btrfs_qgroup_operation, n);
1388 		cmp = comp_oper_exist(cur, oper);
1389 		if (cmp < 0) {
1390 			n = n->rb_right;
1391 		} else if (cmp) {
1392 			n = n->rb_left;
1393 		} else {
1394 			spin_unlock(&fs_info->qgroup_op_lock);
1395 			return -EEXIST;
1396 		}
1397 	}
1398 	spin_unlock(&fs_info->qgroup_op_lock);
1399 	return 0;
1400 }
1401 
1402 static int comp_oper(struct btrfs_qgroup_operation *oper1,
1403 		     struct btrfs_qgroup_operation *oper2)
1404 {
1405 	if (oper1->bytenr < oper2->bytenr)
1406 		return -1;
1407 	if (oper1->bytenr > oper2->bytenr)
1408 		return 1;
1409 	if (oper1->ref_root < oper2->ref_root)
1410 		return -1;
1411 	if (oper1->ref_root > oper2->ref_root)
1412 		return 1;
1413 	if (oper1->seq < oper2->seq)
1414 		return -1;
1415 	if (oper1->seq > oper2->seq)
1416 		return 1;
1417 	if (oper1->type < oper2->type)
1418 		return -1;
1419 	if (oper1->type > oper2->type)
1420 		return 1;
1421 	return 0;
1422 }
1423 
1424 static int insert_qgroup_oper(struct btrfs_fs_info *fs_info,
1425 			      struct btrfs_qgroup_operation *oper)
1426 {
1427 	struct rb_node **p;
1428 	struct rb_node *parent = NULL;
1429 	struct btrfs_qgroup_operation *cur;
1430 	int cmp;
1431 
1432 	spin_lock(&fs_info->qgroup_op_lock);
1433 	p = &fs_info->qgroup_op_tree.rb_node;
1434 	while (*p) {
1435 		parent = *p;
1436 		cur = rb_entry(parent, struct btrfs_qgroup_operation, n);
1437 		cmp = comp_oper(cur, oper);
1438 		if (cmp < 0) {
1439 			p = &(*p)->rb_right;
1440 		} else if (cmp) {
1441 			p = &(*p)->rb_left;
1442 		} else {
1443 			spin_unlock(&fs_info->qgroup_op_lock);
1444 			return -EEXIST;
1445 		}
1446 	}
1447 	rb_link_node(&oper->n, parent, p);
1448 	rb_insert_color(&oper->n, &fs_info->qgroup_op_tree);
1449 	spin_unlock(&fs_info->qgroup_op_lock);
1450 	return 0;
1451 }
1452 
1453 /*
1454  * Record a quota operation for processing later on.
1455  * @trans: the transaction we are adding the delayed op to.
1456  * @fs_info: the fs_info for this fs.
1457  * @ref_root: the root of the reference we are acting on,
1458  * @bytenr: the bytenr we are acting on.
1459  * @num_bytes: the number of bytes in the reference.
1460  * @type: the type of operation this is.
1461  * @mod_seq: do we need to get a sequence number for looking up roots.
1462  *
1463  * We just add it to our trans qgroup_ref_list and carry on and process these
1464  * operations in order at some later point.  If the reference root isn't a fs
1465  * root then we don't bother with doing anything.
1466  *
1467  * MUST BE HOLDING THE REF LOCK.
1468  */
1469 int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1470 			    struct btrfs_fs_info *fs_info, u64 ref_root,
1471 			    u64 bytenr, u64 num_bytes,
1472 			    enum btrfs_qgroup_operation_type type, int mod_seq)
1473 {
1474 	struct btrfs_qgroup_operation *oper;
1475 	int ret;
1476 
1477 	if (!is_fstree(ref_root) || !fs_info->quota_enabled)
1478 		return 0;
1479 
1480 	oper = kmalloc(sizeof(*oper), GFP_NOFS);
1481 	if (!oper)
1482 		return -ENOMEM;
1483 
1484 	oper->ref_root = ref_root;
1485 	oper->bytenr = bytenr;
1486 	oper->num_bytes = num_bytes;
1487 	oper->type = type;
1488 	oper->seq = atomic_inc_return(&fs_info->qgroup_op_seq);
1489 	INIT_LIST_HEAD(&oper->elem.list);
1490 	oper->elem.seq = 0;
1491 
1492 	trace_btrfs_qgroup_record_ref(oper);
1493 
1494 	if (type == BTRFS_QGROUP_OPER_SUB_SUBTREE) {
1495 		/*
1496 		 * If any operation for this bytenr/ref_root combo
1497 		 * exists, then we know it's not exclusively owned and
1498 		 * shouldn't be queued up.
1499 		 *
1500 		 * This also catches the case where we have a cloned
1501 		 * extent that gets queued up multiple times during
1502 		 * drop snapshot.
1503 		 */
1504 		if (qgroup_oper_exists(fs_info, oper)) {
1505 			kfree(oper);
1506 			return 0;
1507 		}
1508 	}
1509 
1510 	ret = insert_qgroup_oper(fs_info, oper);
1511 	if (ret) {
1512 		/* Shouldn't happen so have an assert for developers */
1513 		ASSERT(0);
1514 		kfree(oper);
1515 		return ret;
1516 	}
1517 	list_add_tail(&oper->list, &trans->qgroup_ref_list);
1518 
1519 	if (mod_seq)
1520 		btrfs_get_tree_mod_seq(fs_info, &oper->elem);
1521 
1522 	return 0;
1523 }
1524 
1525 static int qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1526 				  struct btrfs_qgroup_operation *oper)
1527 {
1528 	struct ulist *tmp;
1529 	int sign = 0;
1530 	int ret = 0;
1531 
1532 	tmp = ulist_alloc(GFP_NOFS);
1533 	if (!tmp)
1534 		return -ENOMEM;
1535 
1536 	spin_lock(&fs_info->qgroup_lock);
1537 	if (!fs_info->quota_root)
1538 		goto out;
1539 
1540 	switch (oper->type) {
1541 	case BTRFS_QGROUP_OPER_ADD_EXCL:
1542 		sign = 1;
1543 		break;
1544 	case BTRFS_QGROUP_OPER_SUB_EXCL:
1545 		sign = -1;
1546 		break;
1547 	default:
1548 		ASSERT(0);
1549 	}
1550 	ret = __qgroup_excl_accounting(fs_info, tmp, oper->ref_root,
1551 				       oper->num_bytes, sign);
1552 out:
1553 	spin_unlock(&fs_info->qgroup_lock);
1554 	ulist_free(tmp);
1555 	return ret;
1556 }
1557 
1558 /*
1559  * Walk all of the roots that pointed to our bytenr and adjust their refcnts as
1560  * properly.
1561  */
1562 static int qgroup_calc_old_refcnt(struct btrfs_fs_info *fs_info,
1563 				  u64 root_to_skip, struct ulist *tmp,
1564 				  struct ulist *roots, struct ulist *qgroups,
1565 				  u64 seq, int *old_roots, int rescan)
1566 {
1567 	struct ulist_node *unode;
1568 	struct ulist_iterator uiter;
1569 	struct ulist_node *tmp_unode;
1570 	struct ulist_iterator tmp_uiter;
1571 	struct btrfs_qgroup *qg;
1572 	int ret;
1573 
1574 	ULIST_ITER_INIT(&uiter);
1575 	while ((unode = ulist_next(roots, &uiter))) {
1576 		/* We don't count our current root here */
1577 		if (unode->val == root_to_skip)
1578 			continue;
1579 		qg = find_qgroup_rb(fs_info, unode->val);
1580 		if (!qg)
1581 			continue;
1582 		/*
1583 		 * We could have a pending removal of this same ref so we may
1584 		 * not have actually found our ref root when doing
1585 		 * btrfs_find_all_roots, so we need to keep track of how many
1586 		 * old roots we find in case we removed ours and added a
1587 		 * different one at the same time.  I don't think this could
1588 		 * happen in practice but that sort of thinking leads to pain
1589 		 * and suffering and to the dark side.
1590 		 */
1591 		(*old_roots)++;
1592 
1593 		ulist_reinit(tmp);
1594 		ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg),
1595 				GFP_ATOMIC);
1596 		if (ret < 0)
1597 			return ret;
1598 		ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg), GFP_ATOMIC);
1599 		if (ret < 0)
1600 			return ret;
1601 		ULIST_ITER_INIT(&tmp_uiter);
1602 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1603 			struct btrfs_qgroup_list *glist;
1604 
1605 			qg = u64_to_ptr(tmp_unode->aux);
1606 			/*
1607 			 * We use this sequence number to keep from having to
1608 			 * run the whole list and 0 out the refcnt every time.
1609 			 * We basically use sequnce as the known 0 count and
1610 			 * then add 1 everytime we see a qgroup.  This is how we
1611 			 * get how many of the roots actually point up to the
1612 			 * upper level qgroups in order to determine exclusive
1613 			 * counts.
1614 			 *
1615 			 * For rescan we want to set old_refcnt to seq so our
1616 			 * exclusive calculations end up correct.
1617 			 */
1618 			if (rescan)
1619 				qg->old_refcnt = seq;
1620 			else if (qg->old_refcnt < seq)
1621 				qg->old_refcnt = seq + 1;
1622 			else
1623 				qg->old_refcnt++;
1624 
1625 			if (qg->new_refcnt < seq)
1626 				qg->new_refcnt = seq + 1;
1627 			else
1628 				qg->new_refcnt++;
1629 			list_for_each_entry(glist, &qg->groups, next_group) {
1630 				ret = ulist_add(qgroups, glist->group->qgroupid,
1631 						ptr_to_u64(glist->group),
1632 						GFP_ATOMIC);
1633 				if (ret < 0)
1634 					return ret;
1635 				ret = ulist_add(tmp, glist->group->qgroupid,
1636 						ptr_to_u64(glist->group),
1637 						GFP_ATOMIC);
1638 				if (ret < 0)
1639 					return ret;
1640 			}
1641 		}
1642 	}
1643 	return 0;
1644 }
1645 
1646 /*
1647  * We need to walk forward in our operation tree and account for any roots that
1648  * were deleted after we made this operation.
1649  */
1650 static int qgroup_account_deleted_refs(struct btrfs_fs_info *fs_info,
1651 				       struct btrfs_qgroup_operation *oper,
1652 				       struct ulist *tmp,
1653 				       struct ulist *qgroups, u64 seq,
1654 				       int *old_roots)
1655 {
1656 	struct ulist_node *unode;
1657 	struct ulist_iterator uiter;
1658 	struct btrfs_qgroup *qg;
1659 	struct btrfs_qgroup_operation *tmp_oper;
1660 	struct rb_node *n;
1661 	int ret;
1662 
1663 	ulist_reinit(tmp);
1664 
1665 	/*
1666 	 * We only walk forward in the tree since we're only interested in
1667 	 * removals that happened _after_  our operation.
1668 	 */
1669 	spin_lock(&fs_info->qgroup_op_lock);
1670 	n = rb_next(&oper->n);
1671 	spin_unlock(&fs_info->qgroup_op_lock);
1672 	if (!n)
1673 		return 0;
1674 	tmp_oper = rb_entry(n, struct btrfs_qgroup_operation, n);
1675 	while (tmp_oper->bytenr == oper->bytenr) {
1676 		/*
1677 		 * If it's not a removal we don't care, additions work out
1678 		 * properly with our refcnt tracking.
1679 		 */
1680 		if (tmp_oper->type != BTRFS_QGROUP_OPER_SUB_SHARED &&
1681 		    tmp_oper->type != BTRFS_QGROUP_OPER_SUB_EXCL)
1682 			goto next;
1683 		qg = find_qgroup_rb(fs_info, tmp_oper->ref_root);
1684 		if (!qg)
1685 			goto next;
1686 		ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg),
1687 				GFP_ATOMIC);
1688 		if (ret) {
1689 			if (ret < 0)
1690 				return ret;
1691 			/*
1692 			 * We only want to increase old_roots if this qgroup is
1693 			 * not already in the list of qgroups.  If it is already
1694 			 * there then that means it must have been re-added or
1695 			 * the delete will be discarded because we had an
1696 			 * existing ref that we haven't looked up yet.  In this
1697 			 * case we don't want to increase old_roots.  So if ret
1698 			 * == 1 then we know that this is the first time we've
1699 			 * seen this qgroup and we can bump the old_roots.
1700 			 */
1701 			(*old_roots)++;
1702 			ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg),
1703 					GFP_ATOMIC);
1704 			if (ret < 0)
1705 				return ret;
1706 		}
1707 next:
1708 		spin_lock(&fs_info->qgroup_op_lock);
1709 		n = rb_next(&tmp_oper->n);
1710 		spin_unlock(&fs_info->qgroup_op_lock);
1711 		if (!n)
1712 			break;
1713 		tmp_oper = rb_entry(n, struct btrfs_qgroup_operation, n);
1714 	}
1715 
1716 	/* Ok now process the qgroups we found */
1717 	ULIST_ITER_INIT(&uiter);
1718 	while ((unode = ulist_next(tmp, &uiter))) {
1719 		struct btrfs_qgroup_list *glist;
1720 
1721 		qg = u64_to_ptr(unode->aux);
1722 		if (qg->old_refcnt < seq)
1723 			qg->old_refcnt = seq + 1;
1724 		else
1725 			qg->old_refcnt++;
1726 		if (qg->new_refcnt < seq)
1727 			qg->new_refcnt = seq + 1;
1728 		else
1729 			qg->new_refcnt++;
1730 		list_for_each_entry(glist, &qg->groups, next_group) {
1731 			ret = ulist_add(qgroups, glist->group->qgroupid,
1732 					ptr_to_u64(glist->group), GFP_ATOMIC);
1733 			if (ret < 0)
1734 				return ret;
1735 			ret = ulist_add(tmp, glist->group->qgroupid,
1736 					ptr_to_u64(glist->group), GFP_ATOMIC);
1737 			if (ret < 0)
1738 				return ret;
1739 		}
1740 	}
1741 	return 0;
1742 }
1743 
1744 /* Add refcnt for the newly added reference. */
1745 static int qgroup_calc_new_refcnt(struct btrfs_fs_info *fs_info,
1746 				  struct btrfs_qgroup_operation *oper,
1747 				  struct btrfs_qgroup *qgroup,
1748 				  struct ulist *tmp, struct ulist *qgroups,
1749 				  u64 seq)
1750 {
1751 	struct ulist_node *unode;
1752 	struct ulist_iterator uiter;
1753 	struct btrfs_qgroup *qg;
1754 	int ret;
1755 
1756 	ulist_reinit(tmp);
1757 	ret = ulist_add(qgroups, qgroup->qgroupid, ptr_to_u64(qgroup),
1758 			GFP_ATOMIC);
1759 	if (ret < 0)
1760 		return ret;
1761 	ret = ulist_add(tmp, qgroup->qgroupid, ptr_to_u64(qgroup),
1762 			GFP_ATOMIC);
1763 	if (ret < 0)
1764 		return ret;
1765 	ULIST_ITER_INIT(&uiter);
1766 	while ((unode = ulist_next(tmp, &uiter))) {
1767 		struct btrfs_qgroup_list *glist;
1768 
1769 		qg = u64_to_ptr(unode->aux);
1770 		if (oper->type == BTRFS_QGROUP_OPER_ADD_SHARED) {
1771 			if (qg->new_refcnt < seq)
1772 				qg->new_refcnt = seq + 1;
1773 			else
1774 				qg->new_refcnt++;
1775 		} else {
1776 			if (qg->old_refcnt < seq)
1777 				qg->old_refcnt = seq + 1;
1778 			else
1779 				qg->old_refcnt++;
1780 		}
1781 		list_for_each_entry(glist, &qg->groups, next_group) {
1782 			ret = ulist_add(tmp, glist->group->qgroupid,
1783 					ptr_to_u64(glist->group), GFP_ATOMIC);
1784 			if (ret < 0)
1785 				return ret;
1786 			ret = ulist_add(qgroups, glist->group->qgroupid,
1787 					ptr_to_u64(glist->group), GFP_ATOMIC);
1788 			if (ret < 0)
1789 				return ret;
1790 		}
1791 	}
1792 	return 0;
1793 }
1794 
1795 /*
1796  * This adjusts the counters for all referenced qgroups if need be.
1797  */
1798 static int qgroup_adjust_counters(struct btrfs_fs_info *fs_info,
1799 				  u64 root_to_skip, u64 num_bytes,
1800 				  struct ulist *qgroups, u64 seq,
1801 				  int old_roots, int new_roots, int rescan)
1802 {
1803 	struct ulist_node *unode;
1804 	struct ulist_iterator uiter;
1805 	struct btrfs_qgroup *qg;
1806 	u64 cur_new_count, cur_old_count;
1807 
1808 	ULIST_ITER_INIT(&uiter);
1809 	while ((unode = ulist_next(qgroups, &uiter))) {
1810 		bool dirty = false;
1811 
1812 		qg = u64_to_ptr(unode->aux);
1813 		/*
1814 		 * Wasn't referenced before but is now, add to the reference
1815 		 * counters.
1816 		 */
1817 		if (qg->old_refcnt <= seq && qg->new_refcnt > seq) {
1818 			qg->rfer += num_bytes;
1819 			qg->rfer_cmpr += num_bytes;
1820 			dirty = true;
1821 		}
1822 
1823 		/*
1824 		 * Was referenced before but isn't now, subtract from the
1825 		 * reference counters.
1826 		 */
1827 		if (qg->old_refcnt > seq && qg->new_refcnt <= seq) {
1828 			qg->rfer -= num_bytes;
1829 			qg->rfer_cmpr -= num_bytes;
1830 			dirty = true;
1831 		}
1832 
1833 		if (qg->old_refcnt < seq)
1834 			cur_old_count = 0;
1835 		else
1836 			cur_old_count = qg->old_refcnt - seq;
1837 		if (qg->new_refcnt < seq)
1838 			cur_new_count = 0;
1839 		else
1840 			cur_new_count = qg->new_refcnt - seq;
1841 
1842 		/*
1843 		 * If our refcount was the same as the roots previously but our
1844 		 * new count isn't the same as the number of roots now then we
1845 		 * went from having a exclusive reference on this range to not.
1846 		 */
1847 		if (old_roots && cur_old_count == old_roots &&
1848 		    (cur_new_count != new_roots || new_roots == 0)) {
1849 			WARN_ON(cur_new_count != new_roots && new_roots == 0);
1850 			qg->excl -= num_bytes;
1851 			qg->excl_cmpr -= num_bytes;
1852 			dirty = true;
1853 		}
1854 
1855 		/*
1856 		 * If we didn't reference all the roots before but now we do we
1857 		 * have an exclusive reference to this range.
1858 		 */
1859 		if ((!old_roots || (old_roots && cur_old_count != old_roots))
1860 		    && cur_new_count == new_roots) {
1861 			qg->excl += num_bytes;
1862 			qg->excl_cmpr += num_bytes;
1863 			dirty = true;
1864 		}
1865 
1866 		if (dirty)
1867 			qgroup_dirty(fs_info, qg);
1868 	}
1869 	return 0;
1870 }
1871 
1872 /*
1873  * If we removed a data extent and there were other references for that bytenr
1874  * then we need to lookup all referenced roots to make sure we still don't
1875  * reference this bytenr.  If we do then we can just discard this operation.
1876  */
1877 static int check_existing_refs(struct btrfs_trans_handle *trans,
1878 			       struct btrfs_fs_info *fs_info,
1879 			       struct btrfs_qgroup_operation *oper)
1880 {
1881 	struct ulist *roots = NULL;
1882 	struct ulist_node *unode;
1883 	struct ulist_iterator uiter;
1884 	int ret = 0;
1885 
1886 	ret = btrfs_find_all_roots(trans, fs_info, oper->bytenr,
1887 				   oper->elem.seq, &roots);
1888 	if (ret < 0)
1889 		return ret;
1890 	ret = 0;
1891 
1892 	ULIST_ITER_INIT(&uiter);
1893 	while ((unode = ulist_next(roots, &uiter))) {
1894 		if (unode->val == oper->ref_root) {
1895 			ret = 1;
1896 			break;
1897 		}
1898 	}
1899 	ulist_free(roots);
1900 	btrfs_put_tree_mod_seq(fs_info, &oper->elem);
1901 
1902 	return ret;
1903 }
1904 
1905 /*
1906  * If we share a reference across multiple roots then we may need to adjust
1907  * various qgroups referenced and exclusive counters.  The basic premise is this
1908  *
1909  * 1) We have seq to represent a 0 count.  Instead of looping through all of the
1910  * qgroups and resetting their refcount to 0 we just constantly bump this
1911  * sequence number to act as the base reference count.  This means that if
1912  * anybody is equal to or below this sequence they were never referenced.  We
1913  * jack this sequence up by the number of roots we found each time in order to
1914  * make sure we don't have any overlap.
1915  *
1916  * 2) We first search all the roots that reference the area _except_ the root
1917  * we're acting on currently.  This makes up the old_refcnt of all the qgroups
1918  * before.
1919  *
1920  * 3) We walk all of the qgroups referenced by the root we are currently acting
1921  * on, and will either adjust old_refcnt in the case of a removal or the
1922  * new_refcnt in the case of an addition.
1923  *
1924  * 4) Finally we walk all the qgroups that are referenced by this range
1925  * including the root we are acting on currently.  We will adjust the counters
1926  * based on the number of roots we had and will have after this operation.
1927  *
1928  * Take this example as an illustration
1929  *
1930  *			[qgroup 1/0]
1931  *		     /         |          \
1932  *		[qg 0/0]   [qg 0/1]	[qg 0/2]
1933  *		   \          |            /
1934  *		  [	   extent	    ]
1935  *
1936  * Say we are adding a reference that is covered by qg 0/0.  The first step
1937  * would give a refcnt of 1 to qg 0/1 and 0/2 and a refcnt of 2 to qg 1/0 with
1938  * old_roots being 2.  Because it is adding new_roots will be 1.  We then go
1939  * through qg 0/0 which will get the new_refcnt set to 1 and add 1 to qg 1/0's
1940  * new_refcnt, bringing it to 3.  We then walk through all of the qgroups, we
1941  * notice that the old refcnt for qg 0/0 < the new refcnt, so we added a
1942  * reference and thus must add the size to the referenced bytes.  Everything
1943  * else is the same so nothing else changes.
1944  */
1945 static int qgroup_shared_accounting(struct btrfs_trans_handle *trans,
1946 				    struct btrfs_fs_info *fs_info,
1947 				    struct btrfs_qgroup_operation *oper)
1948 {
1949 	struct ulist *roots = NULL;
1950 	struct ulist *qgroups, *tmp;
1951 	struct btrfs_qgroup *qgroup;
1952 	struct seq_list elem = SEQ_LIST_INIT(elem);
1953 	u64 seq;
1954 	int old_roots = 0;
1955 	int new_roots = 0;
1956 	int ret = 0;
1957 
1958 	if (oper->elem.seq) {
1959 		ret = check_existing_refs(trans, fs_info, oper);
1960 		if (ret < 0)
1961 			return ret;
1962 		if (ret)
1963 			return 0;
1964 	}
1965 
1966 	qgroups = ulist_alloc(GFP_NOFS);
1967 	if (!qgroups)
1968 		return -ENOMEM;
1969 
1970 	tmp = ulist_alloc(GFP_NOFS);
1971 	if (!tmp) {
1972 		ulist_free(qgroups);
1973 		return -ENOMEM;
1974 	}
1975 
1976 	btrfs_get_tree_mod_seq(fs_info, &elem);
1977 	ret = btrfs_find_all_roots(trans, fs_info, oper->bytenr, elem.seq,
1978 				   &roots);
1979 	btrfs_put_tree_mod_seq(fs_info, &elem);
1980 	if (ret < 0) {
1981 		ulist_free(qgroups);
1982 		ulist_free(tmp);
1983 		return ret;
1984 	}
1985 	spin_lock(&fs_info->qgroup_lock);
1986 	qgroup = find_qgroup_rb(fs_info, oper->ref_root);
1987 	if (!qgroup)
1988 		goto out;
1989 	seq = fs_info->qgroup_seq;
1990 
1991 	/*
1992 	 * So roots is the list of all the roots currently pointing at the
1993 	 * bytenr, including the ref we are adding if we are adding, or not if
1994 	 * we are removing a ref.  So we pass in the ref_root to skip that root
1995 	 * in our calculations.  We set old_refnct and new_refcnt cause who the
1996 	 * hell knows what everything looked like before, and it doesn't matter
1997 	 * except...
1998 	 */
1999 	ret = qgroup_calc_old_refcnt(fs_info, oper->ref_root, tmp, roots, qgroups,
2000 				     seq, &old_roots, 0);
2001 	if (ret < 0)
2002 		goto out;
2003 
2004 	/*
2005 	 * Now adjust the refcounts of the qgroups that care about this
2006 	 * reference, either the old_count in the case of removal or new_count
2007 	 * in the case of an addition.
2008 	 */
2009 	ret = qgroup_calc_new_refcnt(fs_info, oper, qgroup, tmp, qgroups,
2010 				     seq);
2011 	if (ret < 0)
2012 		goto out;
2013 
2014 	/*
2015 	 * ...in the case of removals.  If we had a removal before we got around
2016 	 * to processing this operation then we need to find that guy and count
2017 	 * his references as if they really existed so we don't end up screwing
2018 	 * up the exclusive counts.  Then whenever we go to process the delete
2019 	 * everything will be grand and we can account for whatever exclusive
2020 	 * changes need to be made there.  We also have to pass in old_roots so
2021 	 * we have an accurate count of the roots as it pertains to this
2022 	 * operations view of the world.
2023 	 */
2024 	ret = qgroup_account_deleted_refs(fs_info, oper, tmp, qgroups, seq,
2025 					  &old_roots);
2026 	if (ret < 0)
2027 		goto out;
2028 
2029 	/*
2030 	 * We are adding our root, need to adjust up the number of roots,
2031 	 * otherwise old_roots is the number of roots we want.
2032 	 */
2033 	if (oper->type == BTRFS_QGROUP_OPER_ADD_SHARED) {
2034 		new_roots = old_roots + 1;
2035 	} else {
2036 		new_roots = old_roots;
2037 		old_roots++;
2038 	}
2039 	fs_info->qgroup_seq += old_roots + 1;
2040 
2041 
2042 	/*
2043 	 * And now the magic happens, bless Arne for having a pretty elegant
2044 	 * solution for this.
2045 	 */
2046 	qgroup_adjust_counters(fs_info, oper->ref_root, oper->num_bytes,
2047 			       qgroups, seq, old_roots, new_roots, 0);
2048 out:
2049 	spin_unlock(&fs_info->qgroup_lock);
2050 	ulist_free(qgroups);
2051 	ulist_free(roots);
2052 	ulist_free(tmp);
2053 	return ret;
2054 }
2055 
2056 /*
2057  * Process a reference to a shared subtree. This type of operation is
2058  * queued during snapshot removal when we encounter extents which are
2059  * shared between more than one root.
2060  */
2061 static int qgroup_subtree_accounting(struct btrfs_trans_handle *trans,
2062 				     struct btrfs_fs_info *fs_info,
2063 				     struct btrfs_qgroup_operation *oper)
2064 {
2065 	struct ulist *roots = NULL;
2066 	struct ulist_node *unode;
2067 	struct ulist_iterator uiter;
2068 	struct btrfs_qgroup_list *glist;
2069 	struct ulist *parents;
2070 	int ret = 0;
2071 	int err;
2072 	struct btrfs_qgroup *qg;
2073 	u64 root_obj = 0;
2074 	struct seq_list elem = SEQ_LIST_INIT(elem);
2075 
2076 	parents = ulist_alloc(GFP_NOFS);
2077 	if (!parents)
2078 		return -ENOMEM;
2079 
2080 	btrfs_get_tree_mod_seq(fs_info, &elem);
2081 	ret = btrfs_find_all_roots(trans, fs_info, oper->bytenr,
2082 				   elem.seq, &roots);
2083 	btrfs_put_tree_mod_seq(fs_info, &elem);
2084 	if (ret < 0)
2085 		goto out;
2086 
2087 	if (roots->nnodes != 1)
2088 		goto out;
2089 
2090 	ULIST_ITER_INIT(&uiter);
2091 	unode = ulist_next(roots, &uiter); /* Only want 1 so no need to loop */
2092 	/*
2093 	 * If we find our ref root then that means all refs
2094 	 * this extent has to the root have not yet been
2095 	 * deleted. In that case, we do nothing and let the
2096 	 * last ref for this bytenr drive our update.
2097 	 *
2098 	 * This can happen for example if an extent is
2099 	 * referenced multiple times in a snapshot (clone,
2100 	 * etc). If we are in the middle of snapshot removal,
2101 	 * queued updates for such an extent will find the
2102 	 * root if we have not yet finished removing the
2103 	 * snapshot.
2104 	 */
2105 	if (unode->val == oper->ref_root)
2106 		goto out;
2107 
2108 	root_obj = unode->val;
2109 	BUG_ON(!root_obj);
2110 
2111 	spin_lock(&fs_info->qgroup_lock);
2112 	qg = find_qgroup_rb(fs_info, root_obj);
2113 	if (!qg)
2114 		goto out_unlock;
2115 
2116 	qg->excl += oper->num_bytes;
2117 	qg->excl_cmpr += oper->num_bytes;
2118 	qgroup_dirty(fs_info, qg);
2119 
2120 	/*
2121 	 * Adjust counts for parent groups. First we find all
2122 	 * parents, then in the 2nd loop we do the adjustment
2123 	 * while adding parents of the parents to our ulist.
2124 	 */
2125 	list_for_each_entry(glist, &qg->groups, next_group) {
2126 		err = ulist_add(parents, glist->group->qgroupid,
2127 				ptr_to_u64(glist->group), GFP_ATOMIC);
2128 		if (err < 0) {
2129 			ret = err;
2130 			goto out_unlock;
2131 		}
2132 	}
2133 
2134 	ULIST_ITER_INIT(&uiter);
2135 	while ((unode = ulist_next(parents, &uiter))) {
2136 		qg = u64_to_ptr(unode->aux);
2137 		qg->excl += oper->num_bytes;
2138 		qg->excl_cmpr += oper->num_bytes;
2139 		qgroup_dirty(fs_info, qg);
2140 
2141 		/* Add any parents of the parents */
2142 		list_for_each_entry(glist, &qg->groups, next_group) {
2143 			err = ulist_add(parents, glist->group->qgroupid,
2144 					ptr_to_u64(glist->group), GFP_ATOMIC);
2145 			if (err < 0) {
2146 				ret = err;
2147 				goto out_unlock;
2148 			}
2149 		}
2150 	}
2151 
2152 out_unlock:
2153 	spin_unlock(&fs_info->qgroup_lock);
2154 
2155 out:
2156 	ulist_free(roots);
2157 	ulist_free(parents);
2158 	return ret;
2159 }
2160 
2161 /*
2162  * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
2163  * from the fs. First, all roots referencing the extent are searched, and
2164  * then the space is accounted accordingly to the different roots. The
2165  * accounting algorithm works in 3 steps documented inline.
2166  */
2167 static int btrfs_qgroup_account(struct btrfs_trans_handle *trans,
2168 				struct btrfs_fs_info *fs_info,
2169 				struct btrfs_qgroup_operation *oper)
2170 {
2171 	int ret = 0;
2172 
2173 	if (!fs_info->quota_enabled)
2174 		return 0;
2175 
2176 	BUG_ON(!fs_info->quota_root);
2177 
2178 	mutex_lock(&fs_info->qgroup_rescan_lock);
2179 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2180 		if (fs_info->qgroup_rescan_progress.objectid <= oper->bytenr) {
2181 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2182 			return 0;
2183 		}
2184 	}
2185 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2186 
2187 	ASSERT(is_fstree(oper->ref_root));
2188 
2189 	trace_btrfs_qgroup_account(oper);
2190 
2191 	switch (oper->type) {
2192 	case BTRFS_QGROUP_OPER_ADD_EXCL:
2193 	case BTRFS_QGROUP_OPER_SUB_EXCL:
2194 		ret = qgroup_excl_accounting(fs_info, oper);
2195 		break;
2196 	case BTRFS_QGROUP_OPER_ADD_SHARED:
2197 	case BTRFS_QGROUP_OPER_SUB_SHARED:
2198 		ret = qgroup_shared_accounting(trans, fs_info, oper);
2199 		break;
2200 	case BTRFS_QGROUP_OPER_SUB_SUBTREE:
2201 		ret = qgroup_subtree_accounting(trans, fs_info, oper);
2202 		break;
2203 	default:
2204 		ASSERT(0);
2205 	}
2206 	return ret;
2207 }
2208 
2209 /*
2210  * Needs to be called everytime we run delayed refs, even if there is an error
2211  * in order to cleanup outstanding operations.
2212  */
2213 int btrfs_delayed_qgroup_accounting(struct btrfs_trans_handle *trans,
2214 				    struct btrfs_fs_info *fs_info)
2215 {
2216 	struct btrfs_qgroup_operation *oper;
2217 	int ret = 0;
2218 
2219 	while (!list_empty(&trans->qgroup_ref_list)) {
2220 		oper = list_first_entry(&trans->qgroup_ref_list,
2221 					struct btrfs_qgroup_operation, list);
2222 		list_del_init(&oper->list);
2223 		if (!ret || !trans->aborted)
2224 			ret = btrfs_qgroup_account(trans, fs_info, oper);
2225 		spin_lock(&fs_info->qgroup_op_lock);
2226 		rb_erase(&oper->n, &fs_info->qgroup_op_tree);
2227 		spin_unlock(&fs_info->qgroup_op_lock);
2228 		btrfs_put_tree_mod_seq(fs_info, &oper->elem);
2229 		kfree(oper);
2230 	}
2231 	return ret;
2232 }
2233 
2234 /*
2235  * called from commit_transaction. Writes all changed qgroups to disk.
2236  */
2237 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2238 		      struct btrfs_fs_info *fs_info)
2239 {
2240 	struct btrfs_root *quota_root = fs_info->quota_root;
2241 	int ret = 0;
2242 	int start_rescan_worker = 0;
2243 
2244 	if (!quota_root)
2245 		goto out;
2246 
2247 	if (!fs_info->quota_enabled && fs_info->pending_quota_state)
2248 		start_rescan_worker = 1;
2249 
2250 	fs_info->quota_enabled = fs_info->pending_quota_state;
2251 
2252 	spin_lock(&fs_info->qgroup_lock);
2253 	while (!list_empty(&fs_info->dirty_qgroups)) {
2254 		struct btrfs_qgroup *qgroup;
2255 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2256 					  struct btrfs_qgroup, dirty);
2257 		list_del_init(&qgroup->dirty);
2258 		spin_unlock(&fs_info->qgroup_lock);
2259 		ret = update_qgroup_info_item(trans, quota_root, qgroup);
2260 		if (ret)
2261 			fs_info->qgroup_flags |=
2262 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2263 		ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2264 		if (ret)
2265 			fs_info->qgroup_flags |=
2266 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2267 		spin_lock(&fs_info->qgroup_lock);
2268 	}
2269 	if (fs_info->quota_enabled)
2270 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2271 	else
2272 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2273 	spin_unlock(&fs_info->qgroup_lock);
2274 
2275 	ret = update_qgroup_status_item(trans, fs_info, quota_root);
2276 	if (ret)
2277 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2278 
2279 	if (!ret && start_rescan_worker) {
2280 		ret = qgroup_rescan_init(fs_info, 0, 1);
2281 		if (!ret) {
2282 			qgroup_rescan_zero_tracking(fs_info);
2283 			btrfs_queue_work(fs_info->qgroup_rescan_workers,
2284 					 &fs_info->qgroup_rescan_work);
2285 		}
2286 		ret = 0;
2287 	}
2288 
2289 out:
2290 
2291 	return ret;
2292 }
2293 
2294 /*
2295  * copy the acounting information between qgroups. This is necessary when a
2296  * snapshot or a subvolume is created
2297  */
2298 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2299 			 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2300 			 struct btrfs_qgroup_inherit *inherit)
2301 {
2302 	int ret = 0;
2303 	int i;
2304 	u64 *i_qgroups;
2305 	struct btrfs_root *quota_root = fs_info->quota_root;
2306 	struct btrfs_qgroup *srcgroup;
2307 	struct btrfs_qgroup *dstgroup;
2308 	u32 level_size = 0;
2309 	u64 nums;
2310 
2311 	mutex_lock(&fs_info->qgroup_ioctl_lock);
2312 	if (!fs_info->quota_enabled)
2313 		goto out;
2314 
2315 	if (!quota_root) {
2316 		ret = -EINVAL;
2317 		goto out;
2318 	}
2319 
2320 	if (inherit) {
2321 		i_qgroups = (u64 *)(inherit + 1);
2322 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2323 		       2 * inherit->num_excl_copies;
2324 		for (i = 0; i < nums; ++i) {
2325 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2326 			if (!srcgroup) {
2327 				ret = -EINVAL;
2328 				goto out;
2329 			}
2330 
2331 			if ((srcgroup->qgroupid >> 48) <= (objectid >> 48)) {
2332 				ret = -EINVAL;
2333 				goto out;
2334 			}
2335 			++i_qgroups;
2336 		}
2337 	}
2338 
2339 	/*
2340 	 * create a tracking group for the subvol itself
2341 	 */
2342 	ret = add_qgroup_item(trans, quota_root, objectid);
2343 	if (ret)
2344 		goto out;
2345 
2346 	if (srcid) {
2347 		struct btrfs_root *srcroot;
2348 		struct btrfs_key srckey;
2349 
2350 		srckey.objectid = srcid;
2351 		srckey.type = BTRFS_ROOT_ITEM_KEY;
2352 		srckey.offset = (u64)-1;
2353 		srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2354 		if (IS_ERR(srcroot)) {
2355 			ret = PTR_ERR(srcroot);
2356 			goto out;
2357 		}
2358 
2359 		rcu_read_lock();
2360 		level_size = srcroot->nodesize;
2361 		rcu_read_unlock();
2362 	}
2363 
2364 	/*
2365 	 * add qgroup to all inherited groups
2366 	 */
2367 	if (inherit) {
2368 		i_qgroups = (u64 *)(inherit + 1);
2369 		for (i = 0; i < inherit->num_qgroups; ++i) {
2370 			ret = add_qgroup_relation_item(trans, quota_root,
2371 						       objectid, *i_qgroups);
2372 			if (ret)
2373 				goto out;
2374 			ret = add_qgroup_relation_item(trans, quota_root,
2375 						       *i_qgroups, objectid);
2376 			if (ret)
2377 				goto out;
2378 			++i_qgroups;
2379 		}
2380 	}
2381 
2382 
2383 	spin_lock(&fs_info->qgroup_lock);
2384 
2385 	dstgroup = add_qgroup_rb(fs_info, objectid);
2386 	if (IS_ERR(dstgroup)) {
2387 		ret = PTR_ERR(dstgroup);
2388 		goto unlock;
2389 	}
2390 
2391 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2392 		dstgroup->lim_flags = inherit->lim.flags;
2393 		dstgroup->max_rfer = inherit->lim.max_rfer;
2394 		dstgroup->max_excl = inherit->lim.max_excl;
2395 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2396 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2397 
2398 		ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2399 		if (ret) {
2400 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2401 			btrfs_info(fs_info, "unable to update quota limit for %llu",
2402 			       dstgroup->qgroupid);
2403 			goto unlock;
2404 		}
2405 	}
2406 
2407 	if (srcid) {
2408 		srcgroup = find_qgroup_rb(fs_info, srcid);
2409 		if (!srcgroup)
2410 			goto unlock;
2411 
2412 		/*
2413 		 * We call inherit after we clone the root in order to make sure
2414 		 * our counts don't go crazy, so at this point the only
2415 		 * difference between the two roots should be the root node.
2416 		 */
2417 		dstgroup->rfer = srcgroup->rfer;
2418 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2419 		dstgroup->excl = level_size;
2420 		dstgroup->excl_cmpr = level_size;
2421 		srcgroup->excl = level_size;
2422 		srcgroup->excl_cmpr = level_size;
2423 
2424 		/* inherit the limit info */
2425 		dstgroup->lim_flags = srcgroup->lim_flags;
2426 		dstgroup->max_rfer = srcgroup->max_rfer;
2427 		dstgroup->max_excl = srcgroup->max_excl;
2428 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2429 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2430 
2431 		qgroup_dirty(fs_info, dstgroup);
2432 		qgroup_dirty(fs_info, srcgroup);
2433 	}
2434 
2435 	if (!inherit)
2436 		goto unlock;
2437 
2438 	i_qgroups = (u64 *)(inherit + 1);
2439 	for (i = 0; i < inherit->num_qgroups; ++i) {
2440 		ret = add_relation_rb(quota_root->fs_info, objectid,
2441 				      *i_qgroups);
2442 		if (ret)
2443 			goto unlock;
2444 		++i_qgroups;
2445 	}
2446 
2447 	for (i = 0; i <  inherit->num_ref_copies; ++i) {
2448 		struct btrfs_qgroup *src;
2449 		struct btrfs_qgroup *dst;
2450 
2451 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2452 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2453 
2454 		if (!src || !dst) {
2455 			ret = -EINVAL;
2456 			goto unlock;
2457 		}
2458 
2459 		dst->rfer = src->rfer - level_size;
2460 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2461 		i_qgroups += 2;
2462 	}
2463 	for (i = 0; i <  inherit->num_excl_copies; ++i) {
2464 		struct btrfs_qgroup *src;
2465 		struct btrfs_qgroup *dst;
2466 
2467 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2468 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2469 
2470 		if (!src || !dst) {
2471 			ret = -EINVAL;
2472 			goto unlock;
2473 		}
2474 
2475 		dst->excl = src->excl + level_size;
2476 		dst->excl_cmpr = src->excl_cmpr + level_size;
2477 		i_qgroups += 2;
2478 	}
2479 
2480 unlock:
2481 	spin_unlock(&fs_info->qgroup_lock);
2482 out:
2483 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
2484 	return ret;
2485 }
2486 
2487 int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2488 {
2489 	struct btrfs_root *quota_root;
2490 	struct btrfs_qgroup *qgroup;
2491 	struct btrfs_fs_info *fs_info = root->fs_info;
2492 	u64 ref_root = root->root_key.objectid;
2493 	int ret = 0;
2494 	struct ulist_node *unode;
2495 	struct ulist_iterator uiter;
2496 
2497 	if (!is_fstree(ref_root))
2498 		return 0;
2499 
2500 	if (num_bytes == 0)
2501 		return 0;
2502 
2503 	spin_lock(&fs_info->qgroup_lock);
2504 	quota_root = fs_info->quota_root;
2505 	if (!quota_root)
2506 		goto out;
2507 
2508 	qgroup = find_qgroup_rb(fs_info, ref_root);
2509 	if (!qgroup)
2510 		goto out;
2511 
2512 	/*
2513 	 * in a first step, we check all affected qgroups if any limits would
2514 	 * be exceeded
2515 	 */
2516 	ulist_reinit(fs_info->qgroup_ulist);
2517 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2518 			(uintptr_t)qgroup, GFP_ATOMIC);
2519 	if (ret < 0)
2520 		goto out;
2521 	ULIST_ITER_INIT(&uiter);
2522 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2523 		struct btrfs_qgroup *qg;
2524 		struct btrfs_qgroup_list *glist;
2525 
2526 		qg = u64_to_ptr(unode->aux);
2527 
2528 		if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2529 		    qg->reserved + (s64)qg->rfer + num_bytes >
2530 		    qg->max_rfer) {
2531 			ret = -EDQUOT;
2532 			goto out;
2533 		}
2534 
2535 		if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2536 		    qg->reserved + (s64)qg->excl + num_bytes >
2537 		    qg->max_excl) {
2538 			ret = -EDQUOT;
2539 			goto out;
2540 		}
2541 
2542 		list_for_each_entry(glist, &qg->groups, next_group) {
2543 			ret = ulist_add(fs_info->qgroup_ulist,
2544 					glist->group->qgroupid,
2545 					(uintptr_t)glist->group, GFP_ATOMIC);
2546 			if (ret < 0)
2547 				goto out;
2548 		}
2549 	}
2550 	ret = 0;
2551 	/*
2552 	 * no limits exceeded, now record the reservation into all qgroups
2553 	 */
2554 	ULIST_ITER_INIT(&uiter);
2555 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2556 		struct btrfs_qgroup *qg;
2557 
2558 		qg = u64_to_ptr(unode->aux);
2559 
2560 		qg->reserved += num_bytes;
2561 	}
2562 
2563 out:
2564 	spin_unlock(&fs_info->qgroup_lock);
2565 	return ret;
2566 }
2567 
2568 void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
2569 {
2570 	struct btrfs_root *quota_root;
2571 	struct btrfs_qgroup *qgroup;
2572 	struct btrfs_fs_info *fs_info = root->fs_info;
2573 	struct ulist_node *unode;
2574 	struct ulist_iterator uiter;
2575 	u64 ref_root = root->root_key.objectid;
2576 	int ret = 0;
2577 
2578 	if (!is_fstree(ref_root))
2579 		return;
2580 
2581 	if (num_bytes == 0)
2582 		return;
2583 
2584 	spin_lock(&fs_info->qgroup_lock);
2585 
2586 	quota_root = fs_info->quota_root;
2587 	if (!quota_root)
2588 		goto out;
2589 
2590 	qgroup = find_qgroup_rb(fs_info, ref_root);
2591 	if (!qgroup)
2592 		goto out;
2593 
2594 	ulist_reinit(fs_info->qgroup_ulist);
2595 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2596 			(uintptr_t)qgroup, GFP_ATOMIC);
2597 	if (ret < 0)
2598 		goto out;
2599 	ULIST_ITER_INIT(&uiter);
2600 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2601 		struct btrfs_qgroup *qg;
2602 		struct btrfs_qgroup_list *glist;
2603 
2604 		qg = u64_to_ptr(unode->aux);
2605 
2606 		qg->reserved -= num_bytes;
2607 
2608 		list_for_each_entry(glist, &qg->groups, next_group) {
2609 			ret = ulist_add(fs_info->qgroup_ulist,
2610 					glist->group->qgroupid,
2611 					(uintptr_t)glist->group, GFP_ATOMIC);
2612 			if (ret < 0)
2613 				goto out;
2614 		}
2615 	}
2616 
2617 out:
2618 	spin_unlock(&fs_info->qgroup_lock);
2619 }
2620 
2621 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2622 {
2623 	if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2624 		return;
2625 	btrfs_err(trans->root->fs_info,
2626 		"qgroups not uptodate in trans handle %p:  list is%s empty, "
2627 		"seq is %#x.%x",
2628 		trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2629 		(u32)(trans->delayed_ref_elem.seq >> 32),
2630 		(u32)trans->delayed_ref_elem.seq);
2631 	BUG();
2632 }
2633 
2634 /*
2635  * returns < 0 on error, 0 when more leafs are to be scanned.
2636  * returns 1 when done.
2637  */
2638 static int
2639 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2640 		   struct btrfs_trans_handle *trans, struct ulist *qgroups,
2641 		   struct ulist *tmp, struct extent_buffer *scratch_leaf)
2642 {
2643 	struct btrfs_key found;
2644 	struct ulist *roots = NULL;
2645 	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2646 	u64 num_bytes;
2647 	u64 seq;
2648 	int new_roots;
2649 	int slot;
2650 	int ret;
2651 
2652 	path->leave_spinning = 1;
2653 	mutex_lock(&fs_info->qgroup_rescan_lock);
2654 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
2655 					 &fs_info->qgroup_rescan_progress,
2656 					 path, 1, 0);
2657 
2658 	pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2659 		 fs_info->qgroup_rescan_progress.objectid,
2660 		 fs_info->qgroup_rescan_progress.type,
2661 		 fs_info->qgroup_rescan_progress.offset, ret);
2662 
2663 	if (ret) {
2664 		/*
2665 		 * The rescan is about to end, we will not be scanning any
2666 		 * further blocks. We cannot unset the RESCAN flag here, because
2667 		 * we want to commit the transaction if everything went well.
2668 		 * To make the live accounting work in this phase, we set our
2669 		 * scan progress pointer such that every real extent objectid
2670 		 * will be smaller.
2671 		 */
2672 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2673 		btrfs_release_path(path);
2674 		mutex_unlock(&fs_info->qgroup_rescan_lock);
2675 		return ret;
2676 	}
2677 
2678 	btrfs_item_key_to_cpu(path->nodes[0], &found,
2679 			      btrfs_header_nritems(path->nodes[0]) - 1);
2680 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2681 
2682 	btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2683 	memcpy(scratch_leaf, path->nodes[0], sizeof(*scratch_leaf));
2684 	slot = path->slots[0];
2685 	btrfs_release_path(path);
2686 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2687 
2688 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2689 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2690 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2691 		    found.type != BTRFS_METADATA_ITEM_KEY)
2692 			continue;
2693 		if (found.type == BTRFS_METADATA_ITEM_KEY)
2694 			num_bytes = fs_info->extent_root->nodesize;
2695 		else
2696 			num_bytes = found.offset;
2697 
2698 		ulist_reinit(qgroups);
2699 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2700 					   &roots);
2701 		if (ret < 0)
2702 			goto out;
2703 		spin_lock(&fs_info->qgroup_lock);
2704 		seq = fs_info->qgroup_seq;
2705 		fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
2706 
2707 		new_roots = 0;
2708 		ret = qgroup_calc_old_refcnt(fs_info, 0, tmp, roots, qgroups,
2709 					     seq, &new_roots, 1);
2710 		if (ret < 0) {
2711 			spin_unlock(&fs_info->qgroup_lock);
2712 			ulist_free(roots);
2713 			goto out;
2714 		}
2715 
2716 		ret = qgroup_adjust_counters(fs_info, 0, num_bytes, qgroups,
2717 					     seq, 0, new_roots, 1);
2718 		if (ret < 0) {
2719 			spin_unlock(&fs_info->qgroup_lock);
2720 			ulist_free(roots);
2721 			goto out;
2722 		}
2723 		spin_unlock(&fs_info->qgroup_lock);
2724 		ulist_free(roots);
2725 	}
2726 out:
2727 	btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2728 
2729 	return ret;
2730 }
2731 
2732 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2733 {
2734 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2735 						     qgroup_rescan_work);
2736 	struct btrfs_path *path;
2737 	struct btrfs_trans_handle *trans = NULL;
2738 	struct ulist *tmp = NULL, *qgroups = NULL;
2739 	struct extent_buffer *scratch_leaf = NULL;
2740 	int err = -ENOMEM;
2741 	int ret = 0;
2742 
2743 	path = btrfs_alloc_path();
2744 	if (!path)
2745 		goto out;
2746 	qgroups = ulist_alloc(GFP_NOFS);
2747 	if (!qgroups)
2748 		goto out;
2749 	tmp = ulist_alloc(GFP_NOFS);
2750 	if (!tmp)
2751 		goto out;
2752 	scratch_leaf = kmalloc(sizeof(*scratch_leaf), GFP_NOFS);
2753 	if (!scratch_leaf)
2754 		goto out;
2755 
2756 	err = 0;
2757 	while (!err) {
2758 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
2759 		if (IS_ERR(trans)) {
2760 			err = PTR_ERR(trans);
2761 			break;
2762 		}
2763 		if (!fs_info->quota_enabled) {
2764 			err = -EINTR;
2765 		} else {
2766 			err = qgroup_rescan_leaf(fs_info, path, trans,
2767 						 qgroups, tmp, scratch_leaf);
2768 		}
2769 		if (err > 0)
2770 			btrfs_commit_transaction(trans, fs_info->fs_root);
2771 		else
2772 			btrfs_end_transaction(trans, fs_info->fs_root);
2773 	}
2774 
2775 out:
2776 	kfree(scratch_leaf);
2777 	ulist_free(qgroups);
2778 	ulist_free(tmp);
2779 	btrfs_free_path(path);
2780 
2781 	mutex_lock(&fs_info->qgroup_rescan_lock);
2782 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2783 
2784 	if (err > 0 &&
2785 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2786 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2787 	} else if (err < 0) {
2788 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2789 	}
2790 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2791 
2792 	/*
2793 	 * only update status, since the previous part has alreay updated the
2794 	 * qgroup info.
2795 	 */
2796 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
2797 	if (IS_ERR(trans)) {
2798 		err = PTR_ERR(trans);
2799 		btrfs_err(fs_info,
2800 			  "fail to start transaction for status update: %d\n",
2801 			  err);
2802 		goto done;
2803 	}
2804 	ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2805 	if (ret < 0) {
2806 		err = ret;
2807 		btrfs_err(fs_info, "fail to update qgroup status: %d\n", err);
2808 	}
2809 	btrfs_end_transaction(trans, fs_info->quota_root);
2810 
2811 	if (err >= 0) {
2812 		btrfs_info(fs_info, "qgroup scan completed%s",
2813 			err > 0 ? " (inconsistency flag cleared)" : "");
2814 	} else {
2815 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
2816 	}
2817 
2818 done:
2819 	complete_all(&fs_info->qgroup_rescan_completion);
2820 }
2821 
2822 /*
2823  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2824  * memory required for the rescan context.
2825  */
2826 static int
2827 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2828 		   int init_flags)
2829 {
2830 	int ret = 0;
2831 
2832 	if (!init_flags &&
2833 	    (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2834 	     !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2835 		ret = -EINVAL;
2836 		goto err;
2837 	}
2838 
2839 	mutex_lock(&fs_info->qgroup_rescan_lock);
2840 	spin_lock(&fs_info->qgroup_lock);
2841 
2842 	if (init_flags) {
2843 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2844 			ret = -EINPROGRESS;
2845 		else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2846 			ret = -EINVAL;
2847 
2848 		if (ret) {
2849 			spin_unlock(&fs_info->qgroup_lock);
2850 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2851 			goto err;
2852 		}
2853 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2854 	}
2855 
2856 	memset(&fs_info->qgroup_rescan_progress, 0,
2857 		sizeof(fs_info->qgroup_rescan_progress));
2858 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2859 
2860 	spin_unlock(&fs_info->qgroup_lock);
2861 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2862 
2863 	init_completion(&fs_info->qgroup_rescan_completion);
2864 
2865 	memset(&fs_info->qgroup_rescan_work, 0,
2866 	       sizeof(fs_info->qgroup_rescan_work));
2867 	btrfs_init_work(&fs_info->qgroup_rescan_work,
2868 			btrfs_qgroup_rescan_helper,
2869 			btrfs_qgroup_rescan_worker, NULL, NULL);
2870 
2871 	if (ret) {
2872 err:
2873 		btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2874 		return ret;
2875 	}
2876 
2877 	return 0;
2878 }
2879 
2880 static void
2881 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2882 {
2883 	struct rb_node *n;
2884 	struct btrfs_qgroup *qgroup;
2885 
2886 	spin_lock(&fs_info->qgroup_lock);
2887 	/* clear all current qgroup tracking information */
2888 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2889 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
2890 		qgroup->rfer = 0;
2891 		qgroup->rfer_cmpr = 0;
2892 		qgroup->excl = 0;
2893 		qgroup->excl_cmpr = 0;
2894 	}
2895 	spin_unlock(&fs_info->qgroup_lock);
2896 }
2897 
2898 int
2899 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2900 {
2901 	int ret = 0;
2902 	struct btrfs_trans_handle *trans;
2903 
2904 	ret = qgroup_rescan_init(fs_info, 0, 1);
2905 	if (ret)
2906 		return ret;
2907 
2908 	/*
2909 	 * We have set the rescan_progress to 0, which means no more
2910 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2911 	 * However, btrfs_qgroup_account_ref may be right after its call
2912 	 * to btrfs_find_all_roots, in which case it would still do the
2913 	 * accounting.
2914 	 * To solve this, we're committing the transaction, which will
2915 	 * ensure we run all delayed refs and only after that, we are
2916 	 * going to clear all tracking information for a clean start.
2917 	 */
2918 
2919 	trans = btrfs_join_transaction(fs_info->fs_root);
2920 	if (IS_ERR(trans)) {
2921 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2922 		return PTR_ERR(trans);
2923 	}
2924 	ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2925 	if (ret) {
2926 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2927 		return ret;
2928 	}
2929 
2930 	qgroup_rescan_zero_tracking(fs_info);
2931 
2932 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
2933 			 &fs_info->qgroup_rescan_work);
2934 
2935 	return 0;
2936 }
2937 
2938 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info)
2939 {
2940 	int running;
2941 	int ret = 0;
2942 
2943 	mutex_lock(&fs_info->qgroup_rescan_lock);
2944 	spin_lock(&fs_info->qgroup_lock);
2945 	running = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2946 	spin_unlock(&fs_info->qgroup_lock);
2947 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2948 
2949 	if (running)
2950 		ret = wait_for_completion_interruptible(
2951 					&fs_info->qgroup_rescan_completion);
2952 
2953 	return ret;
2954 }
2955 
2956 /*
2957  * this is only called from open_ctree where we're still single threaded, thus
2958  * locking is omitted here.
2959  */
2960 void
2961 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2962 {
2963 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2964 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
2965 				 &fs_info->qgroup_rescan_work);
2966 }
2967