xref: /openbmc/linux/fs/btrfs/qgroup.c (revision 44013260)
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 #include "fs.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
31 #include "tree-checker.h"
32 
33 /*
34  * Helpers to access qgroup reservation
35  *
36  * Callers should ensure the lock context and type are valid
37  */
38 
qgroup_rsv_total(const struct btrfs_qgroup * qgroup)39 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
40 {
41 	u64 ret = 0;
42 	int i;
43 
44 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
45 		ret += qgroup->rsv.values[i];
46 
47 	return ret;
48 }
49 
50 #ifdef CONFIG_BTRFS_DEBUG
qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)51 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
52 {
53 	if (type == BTRFS_QGROUP_RSV_DATA)
54 		return "data";
55 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
56 		return "meta_pertrans";
57 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
58 		return "meta_prealloc";
59 	return NULL;
60 }
61 #endif
62 
qgroup_rsv_add(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup,u64 num_bytes,enum btrfs_qgroup_rsv_type type)63 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
64 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
65 			   enum btrfs_qgroup_rsv_type type)
66 {
67 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
68 	qgroup->rsv.values[type] += num_bytes;
69 }
70 
qgroup_rsv_release(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup,u64 num_bytes,enum btrfs_qgroup_rsv_type type)71 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
72 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
73 			       enum btrfs_qgroup_rsv_type type)
74 {
75 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
76 	if (qgroup->rsv.values[type] >= num_bytes) {
77 		qgroup->rsv.values[type] -= num_bytes;
78 		return;
79 	}
80 #ifdef CONFIG_BTRFS_DEBUG
81 	WARN_RATELIMIT(1,
82 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
83 		qgroup->qgroupid, qgroup_rsv_type_str(type),
84 		qgroup->rsv.values[type], num_bytes);
85 #endif
86 	qgroup->rsv.values[type] = 0;
87 }
88 
qgroup_rsv_add_by_qgroup(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * dest,struct btrfs_qgroup * src)89 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
90 				     struct btrfs_qgroup *dest,
91 				     struct btrfs_qgroup *src)
92 {
93 	int i;
94 
95 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
96 		qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
97 }
98 
qgroup_rsv_release_by_qgroup(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * dest,struct btrfs_qgroup * src)99 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
100 					 struct btrfs_qgroup *dest,
101 					  struct btrfs_qgroup *src)
102 {
103 	int i;
104 
105 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
106 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
107 }
108 
btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup * qg,u64 seq,int mod)109 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
110 					   int mod)
111 {
112 	if (qg->old_refcnt < seq)
113 		qg->old_refcnt = seq;
114 	qg->old_refcnt += mod;
115 }
116 
btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup * qg,u64 seq,int mod)117 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
118 					   int mod)
119 {
120 	if (qg->new_refcnt < seq)
121 		qg->new_refcnt = seq;
122 	qg->new_refcnt += mod;
123 }
124 
btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup * qg,u64 seq)125 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
126 {
127 	if (qg->old_refcnt < seq)
128 		return 0;
129 	return qg->old_refcnt - seq;
130 }
131 
btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup * qg,u64 seq)132 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
133 {
134 	if (qg->new_refcnt < seq)
135 		return 0;
136 	return qg->new_refcnt - seq;
137 }
138 
139 /*
140  * glue structure to represent the relations between qgroups.
141  */
142 struct btrfs_qgroup_list {
143 	struct list_head next_group;
144 	struct list_head next_member;
145 	struct btrfs_qgroup *group;
146 	struct btrfs_qgroup *member;
147 };
148 
qgroup_to_aux(struct btrfs_qgroup * qg)149 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
150 {
151 	return (u64)(uintptr_t)qg;
152 }
153 
unode_aux_to_qgroup(struct ulist_node * n)154 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
155 {
156 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
157 }
158 
159 static int
160 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
161 		   int init_flags);
162 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
163 
164 /* must be called with qgroup_ioctl_lock held */
find_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)165 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
166 					   u64 qgroupid)
167 {
168 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
169 	struct btrfs_qgroup *qgroup;
170 
171 	while (n) {
172 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
173 		if (qgroup->qgroupid < qgroupid)
174 			n = n->rb_left;
175 		else if (qgroup->qgroupid > qgroupid)
176 			n = n->rb_right;
177 		else
178 			return qgroup;
179 	}
180 	return NULL;
181 }
182 
183 /* must be called with qgroup_lock held */
add_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)184 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
185 					  u64 qgroupid)
186 {
187 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
188 	struct rb_node *parent = NULL;
189 	struct btrfs_qgroup *qgroup;
190 
191 	while (*p) {
192 		parent = *p;
193 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
194 
195 		if (qgroup->qgroupid < qgroupid)
196 			p = &(*p)->rb_left;
197 		else if (qgroup->qgroupid > qgroupid)
198 			p = &(*p)->rb_right;
199 		else
200 			return qgroup;
201 	}
202 
203 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
204 	if (!qgroup)
205 		return ERR_PTR(-ENOMEM);
206 
207 	qgroup->qgroupid = qgroupid;
208 	INIT_LIST_HEAD(&qgroup->groups);
209 	INIT_LIST_HEAD(&qgroup->members);
210 	INIT_LIST_HEAD(&qgroup->dirty);
211 	INIT_LIST_HEAD(&qgroup->iterator);
212 
213 	rb_link_node(&qgroup->node, parent, p);
214 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
215 
216 	return qgroup;
217 }
218 
__del_qgroup_rb(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup)219 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
220 			    struct btrfs_qgroup *qgroup)
221 {
222 	struct btrfs_qgroup_list *list;
223 
224 	list_del(&qgroup->dirty);
225 	while (!list_empty(&qgroup->groups)) {
226 		list = list_first_entry(&qgroup->groups,
227 					struct btrfs_qgroup_list, next_group);
228 		list_del(&list->next_group);
229 		list_del(&list->next_member);
230 		kfree(list);
231 	}
232 
233 	while (!list_empty(&qgroup->members)) {
234 		list = list_first_entry(&qgroup->members,
235 					struct btrfs_qgroup_list, next_member);
236 		list_del(&list->next_group);
237 		list_del(&list->next_member);
238 		kfree(list);
239 	}
240 }
241 
242 /* must be called with qgroup_lock held */
del_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)243 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
244 {
245 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
246 
247 	if (!qgroup)
248 		return -ENOENT;
249 
250 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
251 	__del_qgroup_rb(fs_info, qgroup);
252 	return 0;
253 }
254 
255 /*
256  * Add relation specified by two qgroups.
257  *
258  * Must be called with qgroup_lock held.
259  *
260  * Return: 0        on success
261  *         -ENOENT  if one of the qgroups is NULL
262  *         <0       other errors
263  */
__add_relation_rb(struct btrfs_qgroup * member,struct btrfs_qgroup * parent)264 static int __add_relation_rb(struct btrfs_qgroup *member, struct btrfs_qgroup *parent)
265 {
266 	struct btrfs_qgroup_list *list;
267 
268 	if (!member || !parent)
269 		return -ENOENT;
270 
271 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
272 	if (!list)
273 		return -ENOMEM;
274 
275 	list->group = parent;
276 	list->member = member;
277 	list_add_tail(&list->next_group, &member->groups);
278 	list_add_tail(&list->next_member, &parent->members);
279 
280 	return 0;
281 }
282 
283 /*
284  * Add relation specified by two qgroup ids.
285  *
286  * Must be called with qgroup_lock held.
287  *
288  * Return: 0        on success
289  *         -ENOENT  if one of the ids does not exist
290  *         <0       other errors
291  */
add_relation_rb(struct btrfs_fs_info * fs_info,u64 memberid,u64 parentid)292 static int add_relation_rb(struct btrfs_fs_info *fs_info, u64 memberid, u64 parentid)
293 {
294 	struct btrfs_qgroup *member;
295 	struct btrfs_qgroup *parent;
296 
297 	member = find_qgroup_rb(fs_info, memberid);
298 	parent = find_qgroup_rb(fs_info, parentid);
299 
300 	return __add_relation_rb(member, parent);
301 }
302 
303 /* Must be called with qgroup_lock held */
del_relation_rb(struct btrfs_fs_info * fs_info,u64 memberid,u64 parentid)304 static int del_relation_rb(struct btrfs_fs_info *fs_info,
305 			   u64 memberid, u64 parentid)
306 {
307 	struct btrfs_qgroup *member;
308 	struct btrfs_qgroup *parent;
309 	struct btrfs_qgroup_list *list;
310 
311 	member = find_qgroup_rb(fs_info, memberid);
312 	parent = find_qgroup_rb(fs_info, parentid);
313 	if (!member || !parent)
314 		return -ENOENT;
315 
316 	list_for_each_entry(list, &member->groups, next_group) {
317 		if (list->group == parent) {
318 			list_del(&list->next_group);
319 			list_del(&list->next_member);
320 			kfree(list);
321 			return 0;
322 		}
323 	}
324 	return -ENOENT;
325 }
326 
327 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_verify_qgroup_counts(struct btrfs_fs_info * fs_info,u64 qgroupid,u64 rfer,u64 excl)328 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
329 			       u64 rfer, u64 excl)
330 {
331 	struct btrfs_qgroup *qgroup;
332 
333 	qgroup = find_qgroup_rb(fs_info, qgroupid);
334 	if (!qgroup)
335 		return -EINVAL;
336 	if (qgroup->rfer != rfer || qgroup->excl != excl)
337 		return -EINVAL;
338 	return 0;
339 }
340 #endif
341 
qgroup_mark_inconsistent(struct btrfs_fs_info * fs_info)342 static void qgroup_mark_inconsistent(struct btrfs_fs_info *fs_info)
343 {
344 	fs_info->qgroup_flags |= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
345 				  BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
346 				  BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
347 }
348 
349 /*
350  * The full config is read in one go, only called from open_ctree()
351  * It doesn't use any locking, as at this point we're still single-threaded
352  */
btrfs_read_qgroup_config(struct btrfs_fs_info * fs_info)353 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
354 {
355 	struct btrfs_key key;
356 	struct btrfs_key found_key;
357 	struct btrfs_root *quota_root = fs_info->quota_root;
358 	struct btrfs_path *path = NULL;
359 	struct extent_buffer *l;
360 	int slot;
361 	int ret = 0;
362 	u64 flags = 0;
363 	u64 rescan_progress = 0;
364 
365 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
366 		return 0;
367 
368 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
369 	if (!fs_info->qgroup_ulist) {
370 		ret = -ENOMEM;
371 		goto out;
372 	}
373 
374 	path = btrfs_alloc_path();
375 	if (!path) {
376 		ret = -ENOMEM;
377 		goto out;
378 	}
379 
380 	ret = btrfs_sysfs_add_qgroups(fs_info);
381 	if (ret < 0)
382 		goto out;
383 	/* default this to quota off, in case no status key is found */
384 	fs_info->qgroup_flags = 0;
385 
386 	/*
387 	 * pass 1: read status, all qgroup infos and limits
388 	 */
389 	key.objectid = 0;
390 	key.type = 0;
391 	key.offset = 0;
392 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
393 	if (ret)
394 		goto out;
395 
396 	while (1) {
397 		struct btrfs_qgroup *qgroup;
398 
399 		slot = path->slots[0];
400 		l = path->nodes[0];
401 		btrfs_item_key_to_cpu(l, &found_key, slot);
402 
403 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
404 			struct btrfs_qgroup_status_item *ptr;
405 
406 			ptr = btrfs_item_ptr(l, slot,
407 					     struct btrfs_qgroup_status_item);
408 
409 			if (btrfs_qgroup_status_version(l, ptr) !=
410 			    BTRFS_QGROUP_STATUS_VERSION) {
411 				btrfs_err(fs_info,
412 				 "old qgroup version, quota disabled");
413 				goto out;
414 			}
415 			if (btrfs_qgroup_status_generation(l, ptr) !=
416 			    fs_info->generation) {
417 				qgroup_mark_inconsistent(fs_info);
418 				btrfs_err(fs_info,
419 					"qgroup generation mismatch, marked as inconsistent");
420 			}
421 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
422 									  ptr);
423 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
424 			goto next1;
425 		}
426 
427 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
428 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
429 			goto next1;
430 
431 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
432 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
433 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
434 			btrfs_err(fs_info, "inconsistent qgroup config");
435 			qgroup_mark_inconsistent(fs_info);
436 		}
437 		if (!qgroup) {
438 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
439 			if (IS_ERR(qgroup)) {
440 				ret = PTR_ERR(qgroup);
441 				goto out;
442 			}
443 		}
444 		ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
445 		if (ret < 0)
446 			goto out;
447 
448 		switch (found_key.type) {
449 		case BTRFS_QGROUP_INFO_KEY: {
450 			struct btrfs_qgroup_info_item *ptr;
451 
452 			ptr = btrfs_item_ptr(l, slot,
453 					     struct btrfs_qgroup_info_item);
454 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
455 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
456 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
457 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
458 			/* generation currently unused */
459 			break;
460 		}
461 		case BTRFS_QGROUP_LIMIT_KEY: {
462 			struct btrfs_qgroup_limit_item *ptr;
463 
464 			ptr = btrfs_item_ptr(l, slot,
465 					     struct btrfs_qgroup_limit_item);
466 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
467 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
468 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
469 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
470 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
471 			break;
472 		}
473 		}
474 next1:
475 		ret = btrfs_next_item(quota_root, path);
476 		if (ret < 0)
477 			goto out;
478 		if (ret)
479 			break;
480 	}
481 	btrfs_release_path(path);
482 
483 	/*
484 	 * pass 2: read all qgroup relations
485 	 */
486 	key.objectid = 0;
487 	key.type = BTRFS_QGROUP_RELATION_KEY;
488 	key.offset = 0;
489 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
490 	if (ret)
491 		goto out;
492 	while (1) {
493 		slot = path->slots[0];
494 		l = path->nodes[0];
495 		btrfs_item_key_to_cpu(l, &found_key, slot);
496 
497 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
498 			goto next2;
499 
500 		if (found_key.objectid > found_key.offset) {
501 			/* parent <- member, not needed to build config */
502 			/* FIXME should we omit the key completely? */
503 			goto next2;
504 		}
505 
506 		ret = add_relation_rb(fs_info, found_key.objectid,
507 				      found_key.offset);
508 		if (ret == -ENOENT) {
509 			btrfs_warn(fs_info,
510 				"orphan qgroup relation 0x%llx->0x%llx",
511 				found_key.objectid, found_key.offset);
512 			ret = 0;	/* ignore the error */
513 		}
514 		if (ret)
515 			goto out;
516 next2:
517 		ret = btrfs_next_item(quota_root, path);
518 		if (ret < 0)
519 			goto out;
520 		if (ret)
521 			break;
522 	}
523 out:
524 	btrfs_free_path(path);
525 	fs_info->qgroup_flags |= flags;
526 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
527 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
528 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
529 		 ret >= 0)
530 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
531 
532 	if (ret < 0) {
533 		ulist_free(fs_info->qgroup_ulist);
534 		fs_info->qgroup_ulist = NULL;
535 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
536 		btrfs_sysfs_del_qgroups(fs_info);
537 	}
538 
539 	return ret < 0 ? ret : 0;
540 }
541 
542 /*
543  * Called in close_ctree() when quota is still enabled.  This verifies we don't
544  * leak some reserved space.
545  *
546  * Return false if no reserved space is left.
547  * Return true if some reserved space is leaked.
548  */
btrfs_check_quota_leak(struct btrfs_fs_info * fs_info)549 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
550 {
551 	struct rb_node *node;
552 	bool ret = false;
553 
554 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
555 		return ret;
556 	/*
557 	 * Since we're unmounting, there is no race and no need to grab qgroup
558 	 * lock.  And here we don't go post-order to provide a more user
559 	 * friendly sorted result.
560 	 */
561 	for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
562 		struct btrfs_qgroup *qgroup;
563 		int i;
564 
565 		qgroup = rb_entry(node, struct btrfs_qgroup, node);
566 		for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
567 			if (qgroup->rsv.values[i]) {
568 				ret = true;
569 				btrfs_warn(fs_info,
570 		"qgroup %hu/%llu has unreleased space, type %d rsv %llu",
571 				   btrfs_qgroup_level(qgroup->qgroupid),
572 				   btrfs_qgroup_subvolid(qgroup->qgroupid),
573 				   i, qgroup->rsv.values[i]);
574 			}
575 		}
576 	}
577 	return ret;
578 }
579 
580 /*
581  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
582  * first two are in single-threaded paths.And for the third one, we have set
583  * quota_root to be null with qgroup_lock held before, so it is safe to clean
584  * up the in-memory structures without qgroup_lock held.
585  */
btrfs_free_qgroup_config(struct btrfs_fs_info * fs_info)586 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
587 {
588 	struct rb_node *n;
589 	struct btrfs_qgroup *qgroup;
590 
591 	while ((n = rb_first(&fs_info->qgroup_tree))) {
592 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
593 		rb_erase(n, &fs_info->qgroup_tree);
594 		__del_qgroup_rb(fs_info, qgroup);
595 		btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
596 		kfree(qgroup);
597 	}
598 	/*
599 	 * We call btrfs_free_qgroup_config() when unmounting
600 	 * filesystem and disabling quota, so we set qgroup_ulist
601 	 * to be null here to avoid double free.
602 	 */
603 	ulist_free(fs_info->qgroup_ulist);
604 	fs_info->qgroup_ulist = NULL;
605 	btrfs_sysfs_del_qgroups(fs_info);
606 }
607 
add_qgroup_relation_item(struct btrfs_trans_handle * trans,u64 src,u64 dst)608 static int add_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_insert_empty_item(trans, quota_root, path, &key, 0);
625 
626 	btrfs_mark_buffer_dirty(trans, path->nodes[0]);
627 
628 	btrfs_free_path(path);
629 	return ret;
630 }
631 
del_qgroup_relation_item(struct btrfs_trans_handle * trans,u64 src,u64 dst)632 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
633 				    u64 dst)
634 {
635 	int ret;
636 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
637 	struct btrfs_path *path;
638 	struct btrfs_key key;
639 
640 	path = btrfs_alloc_path();
641 	if (!path)
642 		return -ENOMEM;
643 
644 	key.objectid = src;
645 	key.type = BTRFS_QGROUP_RELATION_KEY;
646 	key.offset = dst;
647 
648 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
649 	if (ret < 0)
650 		goto out;
651 
652 	if (ret > 0) {
653 		ret = -ENOENT;
654 		goto out;
655 	}
656 
657 	ret = btrfs_del_item(trans, quota_root, path);
658 out:
659 	btrfs_free_path(path);
660 	return ret;
661 }
662 
add_qgroup_item(struct btrfs_trans_handle * trans,struct btrfs_root * quota_root,u64 qgroupid)663 static int add_qgroup_item(struct btrfs_trans_handle *trans,
664 			   struct btrfs_root *quota_root, u64 qgroupid)
665 {
666 	int ret;
667 	struct btrfs_path *path;
668 	struct btrfs_qgroup_info_item *qgroup_info;
669 	struct btrfs_qgroup_limit_item *qgroup_limit;
670 	struct extent_buffer *leaf;
671 	struct btrfs_key key;
672 
673 	if (btrfs_is_testing(quota_root->fs_info))
674 		return 0;
675 
676 	path = btrfs_alloc_path();
677 	if (!path)
678 		return -ENOMEM;
679 
680 	key.objectid = 0;
681 	key.type = BTRFS_QGROUP_INFO_KEY;
682 	key.offset = qgroupid;
683 
684 	/*
685 	 * Avoid a transaction abort by catching -EEXIST here. In that
686 	 * case, we proceed by re-initializing the existing structure
687 	 * on disk.
688 	 */
689 
690 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
691 				      sizeof(*qgroup_info));
692 	if (ret && ret != -EEXIST)
693 		goto out;
694 
695 	leaf = path->nodes[0];
696 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
697 				 struct btrfs_qgroup_info_item);
698 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
699 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
700 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
701 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
702 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
703 
704 	btrfs_mark_buffer_dirty(trans, leaf);
705 
706 	btrfs_release_path(path);
707 
708 	key.type = BTRFS_QGROUP_LIMIT_KEY;
709 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
710 				      sizeof(*qgroup_limit));
711 	if (ret && ret != -EEXIST)
712 		goto out;
713 
714 	leaf = path->nodes[0];
715 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
716 				  struct btrfs_qgroup_limit_item);
717 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
718 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
719 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
720 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
721 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
722 
723 	btrfs_mark_buffer_dirty(trans, leaf);
724 
725 	ret = 0;
726 out:
727 	btrfs_free_path(path);
728 	return ret;
729 }
730 
del_qgroup_item(struct btrfs_trans_handle * trans,u64 qgroupid)731 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
732 {
733 	int ret;
734 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
735 	struct btrfs_path *path;
736 	struct btrfs_key key;
737 
738 	path = btrfs_alloc_path();
739 	if (!path)
740 		return -ENOMEM;
741 
742 	key.objectid = 0;
743 	key.type = BTRFS_QGROUP_INFO_KEY;
744 	key.offset = qgroupid;
745 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
746 	if (ret < 0)
747 		goto out;
748 
749 	if (ret > 0) {
750 		ret = -ENOENT;
751 		goto out;
752 	}
753 
754 	ret = btrfs_del_item(trans, quota_root, path);
755 	if (ret)
756 		goto out;
757 
758 	btrfs_release_path(path);
759 
760 	key.type = BTRFS_QGROUP_LIMIT_KEY;
761 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
762 	if (ret < 0)
763 		goto out;
764 
765 	if (ret > 0) {
766 		ret = -ENOENT;
767 		goto out;
768 	}
769 
770 	ret = btrfs_del_item(trans, quota_root, path);
771 
772 out:
773 	btrfs_free_path(path);
774 	return ret;
775 }
776 
update_qgroup_limit_item(struct btrfs_trans_handle * trans,struct btrfs_qgroup * qgroup)777 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
778 				    struct btrfs_qgroup *qgroup)
779 {
780 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
781 	struct btrfs_path *path;
782 	struct btrfs_key key;
783 	struct extent_buffer *l;
784 	struct btrfs_qgroup_limit_item *qgroup_limit;
785 	int ret;
786 	int slot;
787 
788 	key.objectid = 0;
789 	key.type = BTRFS_QGROUP_LIMIT_KEY;
790 	key.offset = qgroup->qgroupid;
791 
792 	path = btrfs_alloc_path();
793 	if (!path)
794 		return -ENOMEM;
795 
796 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
797 	if (ret > 0)
798 		ret = -ENOENT;
799 
800 	if (ret)
801 		goto out;
802 
803 	l = path->nodes[0];
804 	slot = path->slots[0];
805 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
806 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
807 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
808 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
809 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
810 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
811 
812 	btrfs_mark_buffer_dirty(trans, l);
813 
814 out:
815 	btrfs_free_path(path);
816 	return ret;
817 }
818 
update_qgroup_info_item(struct btrfs_trans_handle * trans,struct btrfs_qgroup * qgroup)819 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
820 				   struct btrfs_qgroup *qgroup)
821 {
822 	struct btrfs_fs_info *fs_info = trans->fs_info;
823 	struct btrfs_root *quota_root = fs_info->quota_root;
824 	struct btrfs_path *path;
825 	struct btrfs_key key;
826 	struct extent_buffer *l;
827 	struct btrfs_qgroup_info_item *qgroup_info;
828 	int ret;
829 	int slot;
830 
831 	if (btrfs_is_testing(fs_info))
832 		return 0;
833 
834 	key.objectid = 0;
835 	key.type = BTRFS_QGROUP_INFO_KEY;
836 	key.offset = qgroup->qgroupid;
837 
838 	path = btrfs_alloc_path();
839 	if (!path)
840 		return -ENOMEM;
841 
842 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
843 	if (ret > 0)
844 		ret = -ENOENT;
845 
846 	if (ret)
847 		goto out;
848 
849 	l = path->nodes[0];
850 	slot = path->slots[0];
851 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
852 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
853 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
854 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
855 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
856 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
857 
858 	btrfs_mark_buffer_dirty(trans, l);
859 
860 out:
861 	btrfs_free_path(path);
862 	return ret;
863 }
864 
update_qgroup_status_item(struct btrfs_trans_handle * trans)865 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
866 {
867 	struct btrfs_fs_info *fs_info = trans->fs_info;
868 	struct btrfs_root *quota_root = fs_info->quota_root;
869 	struct btrfs_path *path;
870 	struct btrfs_key key;
871 	struct extent_buffer *l;
872 	struct btrfs_qgroup_status_item *ptr;
873 	int ret;
874 	int slot;
875 
876 	key.objectid = 0;
877 	key.type = BTRFS_QGROUP_STATUS_KEY;
878 	key.offset = 0;
879 
880 	path = btrfs_alloc_path();
881 	if (!path)
882 		return -ENOMEM;
883 
884 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
885 	if (ret > 0)
886 		ret = -ENOENT;
887 
888 	if (ret)
889 		goto out;
890 
891 	l = path->nodes[0];
892 	slot = path->slots[0];
893 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
894 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags &
895 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
896 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
897 	btrfs_set_qgroup_status_rescan(l, ptr,
898 				fs_info->qgroup_rescan_progress.objectid);
899 
900 	btrfs_mark_buffer_dirty(trans, l);
901 
902 out:
903 	btrfs_free_path(path);
904 	return ret;
905 }
906 
907 /*
908  * called with qgroup_lock held
909  */
btrfs_clean_quota_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)910 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
911 				  struct btrfs_root *root)
912 {
913 	struct btrfs_path *path;
914 	struct btrfs_key key;
915 	struct extent_buffer *leaf = NULL;
916 	int ret;
917 	int nr = 0;
918 
919 	path = btrfs_alloc_path();
920 	if (!path)
921 		return -ENOMEM;
922 
923 	key.objectid = 0;
924 	key.offset = 0;
925 	key.type = 0;
926 
927 	while (1) {
928 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
929 		if (ret < 0)
930 			goto out;
931 		leaf = path->nodes[0];
932 		nr = btrfs_header_nritems(leaf);
933 		if (!nr)
934 			break;
935 		/*
936 		 * delete the leaf one by one
937 		 * since the whole tree is going
938 		 * to be deleted.
939 		 */
940 		path->slots[0] = 0;
941 		ret = btrfs_del_items(trans, root, path, 0, nr);
942 		if (ret)
943 			goto out;
944 
945 		btrfs_release_path(path);
946 	}
947 	ret = 0;
948 out:
949 	btrfs_free_path(path);
950 	return ret;
951 }
952 
btrfs_quota_enable(struct btrfs_fs_info * fs_info)953 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
954 {
955 	struct btrfs_root *quota_root;
956 	struct btrfs_root *tree_root = fs_info->tree_root;
957 	struct btrfs_path *path = NULL;
958 	struct btrfs_qgroup_status_item *ptr;
959 	struct extent_buffer *leaf;
960 	struct btrfs_key key;
961 	struct btrfs_key found_key;
962 	struct btrfs_qgroup *qgroup = NULL;
963 	struct btrfs_trans_handle *trans = NULL;
964 	struct ulist *ulist = NULL;
965 	int ret = 0;
966 	int slot;
967 
968 	/*
969 	 * We need to have subvol_sem write locked, to prevent races between
970 	 * concurrent tasks trying to enable quotas, because we will unlock
971 	 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
972 	 * and before setting BTRFS_FS_QUOTA_ENABLED.
973 	 */
974 	lockdep_assert_held_write(&fs_info->subvol_sem);
975 
976 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
977 		btrfs_err(fs_info,
978 			  "qgroups are currently unsupported in extent tree v2");
979 		return -EINVAL;
980 	}
981 
982 	mutex_lock(&fs_info->qgroup_ioctl_lock);
983 	if (fs_info->quota_root)
984 		goto out;
985 
986 	ulist = ulist_alloc(GFP_KERNEL);
987 	if (!ulist) {
988 		ret = -ENOMEM;
989 		goto out;
990 	}
991 
992 	ret = btrfs_sysfs_add_qgroups(fs_info);
993 	if (ret < 0)
994 		goto out;
995 
996 	/*
997 	 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
998 	 * avoid lock acquisition inversion problems (reported by lockdep) between
999 	 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
1000 	 * start a transaction.
1001 	 * After we started the transaction lock qgroup_ioctl_lock again and
1002 	 * check if someone else created the quota root in the meanwhile. If so,
1003 	 * just return success and release the transaction handle.
1004 	 *
1005 	 * Also we don't need to worry about someone else calling
1006 	 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
1007 	 * that function returns 0 (success) when the sysfs entries already exist.
1008 	 */
1009 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1010 
1011 	/*
1012 	 * 1 for quota root item
1013 	 * 1 for BTRFS_QGROUP_STATUS item
1014 	 *
1015 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1016 	 * per subvolume. However those are not currently reserved since it
1017 	 * would be a lot of overkill.
1018 	 */
1019 	trans = btrfs_start_transaction(tree_root, 2);
1020 
1021 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1022 	if (IS_ERR(trans)) {
1023 		ret = PTR_ERR(trans);
1024 		trans = NULL;
1025 		goto out;
1026 	}
1027 
1028 	if (fs_info->quota_root)
1029 		goto out;
1030 
1031 	fs_info->qgroup_ulist = ulist;
1032 	ulist = NULL;
1033 
1034 	/*
1035 	 * initially create the quota tree
1036 	 */
1037 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1038 	if (IS_ERR(quota_root)) {
1039 		ret =  PTR_ERR(quota_root);
1040 		btrfs_abort_transaction(trans, ret);
1041 		goto out;
1042 	}
1043 
1044 	path = btrfs_alloc_path();
1045 	if (!path) {
1046 		ret = -ENOMEM;
1047 		btrfs_abort_transaction(trans, ret);
1048 		goto out_free_root;
1049 	}
1050 
1051 	key.objectid = 0;
1052 	key.type = BTRFS_QGROUP_STATUS_KEY;
1053 	key.offset = 0;
1054 
1055 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1056 				      sizeof(*ptr));
1057 	if (ret) {
1058 		btrfs_abort_transaction(trans, ret);
1059 		goto out_free_path;
1060 	}
1061 
1062 	leaf = path->nodes[0];
1063 	ptr = btrfs_item_ptr(leaf, path->slots[0],
1064 				 struct btrfs_qgroup_status_item);
1065 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1066 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1067 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1068 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1069 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
1070 				      BTRFS_QGROUP_STATUS_FLAGS_MASK);
1071 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1072 
1073 	btrfs_mark_buffer_dirty(trans, leaf);
1074 
1075 	key.objectid = 0;
1076 	key.type = BTRFS_ROOT_REF_KEY;
1077 	key.offset = 0;
1078 
1079 	btrfs_release_path(path);
1080 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1081 	if (ret > 0)
1082 		goto out_add_root;
1083 	if (ret < 0) {
1084 		btrfs_abort_transaction(trans, ret);
1085 		goto out_free_path;
1086 	}
1087 
1088 	while (1) {
1089 		slot = path->slots[0];
1090 		leaf = path->nodes[0];
1091 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1092 
1093 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
1094 
1095 			/* Release locks on tree_root before we access quota_root */
1096 			btrfs_release_path(path);
1097 
1098 			ret = add_qgroup_item(trans, quota_root,
1099 					      found_key.offset);
1100 			if (ret) {
1101 				btrfs_abort_transaction(trans, ret);
1102 				goto out_free_path;
1103 			}
1104 
1105 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
1106 			if (IS_ERR(qgroup)) {
1107 				ret = PTR_ERR(qgroup);
1108 				btrfs_abort_transaction(trans, ret);
1109 				goto out_free_path;
1110 			}
1111 			ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1112 			if (ret < 0) {
1113 				btrfs_abort_transaction(trans, ret);
1114 				goto out_free_path;
1115 			}
1116 			ret = btrfs_search_slot_for_read(tree_root, &found_key,
1117 							 path, 1, 0);
1118 			if (ret < 0) {
1119 				btrfs_abort_transaction(trans, ret);
1120 				goto out_free_path;
1121 			}
1122 			if (ret > 0) {
1123 				/*
1124 				 * Shouldn't happen, but in case it does we
1125 				 * don't need to do the btrfs_next_item, just
1126 				 * continue.
1127 				 */
1128 				continue;
1129 			}
1130 		}
1131 		ret = btrfs_next_item(tree_root, path);
1132 		if (ret < 0) {
1133 			btrfs_abort_transaction(trans, ret);
1134 			goto out_free_path;
1135 		}
1136 		if (ret)
1137 			break;
1138 	}
1139 
1140 out_add_root:
1141 	btrfs_release_path(path);
1142 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1143 	if (ret) {
1144 		btrfs_abort_transaction(trans, ret);
1145 		goto out_free_path;
1146 	}
1147 
1148 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1149 	if (IS_ERR(qgroup)) {
1150 		ret = PTR_ERR(qgroup);
1151 		btrfs_abort_transaction(trans, ret);
1152 		goto out_free_path;
1153 	}
1154 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1155 	if (ret < 0) {
1156 		btrfs_abort_transaction(trans, ret);
1157 		goto out_free_path;
1158 	}
1159 
1160 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1161 	/*
1162 	 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1163 	 * a deadlock with tasks concurrently doing other qgroup operations, such
1164 	 * adding/removing qgroups or adding/deleting qgroup relations for example,
1165 	 * because all qgroup operations first start or join a transaction and then
1166 	 * lock the qgroup_ioctl_lock mutex.
1167 	 * We are safe from a concurrent task trying to enable quotas, by calling
1168 	 * this function, since we are serialized by fs_info->subvol_sem.
1169 	 */
1170 	ret = btrfs_commit_transaction(trans);
1171 	trans = NULL;
1172 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1173 	if (ret)
1174 		goto out_free_path;
1175 
1176 	/*
1177 	 * Set quota enabled flag after committing the transaction, to avoid
1178 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1179 	 * creation.
1180 	 */
1181 	spin_lock(&fs_info->qgroup_lock);
1182 	fs_info->quota_root = quota_root;
1183 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1184 	spin_unlock(&fs_info->qgroup_lock);
1185 
1186 	ret = qgroup_rescan_init(fs_info, 0, 1);
1187 	if (!ret) {
1188 	        qgroup_rescan_zero_tracking(fs_info);
1189 		fs_info->qgroup_rescan_running = true;
1190 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1191 	                         &fs_info->qgroup_rescan_work);
1192 	} else {
1193 		/*
1194 		 * We have set both BTRFS_FS_QUOTA_ENABLED and
1195 		 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1196 		 * -EINPROGRESS. That can happen because someone started the
1197 		 * rescan worker by calling quota rescan ioctl before we
1198 		 * attempted to initialize the rescan worker. Failure due to
1199 		 * quotas disabled in the meanwhile is not possible, because
1200 		 * we are holding a write lock on fs_info->subvol_sem, which
1201 		 * is also acquired when disabling quotas.
1202 		 * Ignore such error, and any other error would need to undo
1203 		 * everything we did in the transaction we just committed.
1204 		 */
1205 		ASSERT(ret == -EINPROGRESS);
1206 		ret = 0;
1207 	}
1208 
1209 out_free_path:
1210 	btrfs_free_path(path);
1211 out_free_root:
1212 	if (ret)
1213 		btrfs_put_root(quota_root);
1214 out:
1215 	if (ret) {
1216 		ulist_free(fs_info->qgroup_ulist);
1217 		fs_info->qgroup_ulist = NULL;
1218 		btrfs_sysfs_del_qgroups(fs_info);
1219 	}
1220 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1221 	if (ret && trans)
1222 		btrfs_end_transaction(trans);
1223 	else if (trans)
1224 		ret = btrfs_end_transaction(trans);
1225 	ulist_free(ulist);
1226 	return ret;
1227 }
1228 
btrfs_quota_disable(struct btrfs_fs_info * fs_info)1229 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1230 {
1231 	struct btrfs_root *quota_root = NULL;
1232 	struct btrfs_trans_handle *trans = NULL;
1233 	int ret = 0;
1234 
1235 	/*
1236 	 * We need to have subvol_sem write locked to prevent races with
1237 	 * snapshot creation.
1238 	 */
1239 	lockdep_assert_held_write(&fs_info->subvol_sem);
1240 
1241 	/*
1242 	 * Lock the cleaner mutex to prevent races with concurrent relocation,
1243 	 * because relocation may be building backrefs for blocks of the quota
1244 	 * root while we are deleting the root. This is like dropping fs roots
1245 	 * of deleted snapshots/subvolumes, we need the same protection.
1246 	 *
1247 	 * This also prevents races between concurrent tasks trying to disable
1248 	 * quotas, because we will unlock and relock qgroup_ioctl_lock across
1249 	 * BTRFS_FS_QUOTA_ENABLED changes.
1250 	 */
1251 	mutex_lock(&fs_info->cleaner_mutex);
1252 
1253 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1254 	if (!fs_info->quota_root)
1255 		goto out;
1256 
1257 	/*
1258 	 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1259 	 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1260 	 * to lock that mutex while holding a transaction handle and the rescan
1261 	 * worker needs to commit a transaction.
1262 	 */
1263 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1264 
1265 	/*
1266 	 * Request qgroup rescan worker to complete and wait for it. This wait
1267 	 * must be done before transaction start for quota disable since it may
1268 	 * deadlock with transaction by the qgroup rescan worker.
1269 	 */
1270 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1271 	btrfs_qgroup_wait_for_completion(fs_info, false);
1272 
1273 	/*
1274 	 * 1 For the root item
1275 	 *
1276 	 * We should also reserve enough items for the quota tree deletion in
1277 	 * btrfs_clean_quota_tree but this is not done.
1278 	 *
1279 	 * Also, we must always start a transaction without holding the mutex
1280 	 * qgroup_ioctl_lock, see btrfs_quota_enable().
1281 	 */
1282 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1283 
1284 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1285 	if (IS_ERR(trans)) {
1286 		ret = PTR_ERR(trans);
1287 		trans = NULL;
1288 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1289 		goto out;
1290 	}
1291 
1292 	if (!fs_info->quota_root)
1293 		goto out;
1294 
1295 	spin_lock(&fs_info->qgroup_lock);
1296 	quota_root = fs_info->quota_root;
1297 	fs_info->quota_root = NULL;
1298 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1299 	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1300 	spin_unlock(&fs_info->qgroup_lock);
1301 
1302 	btrfs_free_qgroup_config(fs_info);
1303 
1304 	ret = btrfs_clean_quota_tree(trans, quota_root);
1305 	if (ret) {
1306 		btrfs_abort_transaction(trans, ret);
1307 		goto out;
1308 	}
1309 
1310 	ret = btrfs_del_root(trans, &quota_root->root_key);
1311 	if (ret) {
1312 		btrfs_abort_transaction(trans, ret);
1313 		goto out;
1314 	}
1315 
1316 	spin_lock(&fs_info->trans_lock);
1317 	list_del(&quota_root->dirty_list);
1318 	spin_unlock(&fs_info->trans_lock);
1319 
1320 	btrfs_tree_lock(quota_root->node);
1321 	btrfs_clear_buffer_dirty(trans, quota_root->node);
1322 	btrfs_tree_unlock(quota_root->node);
1323 	btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1324 			      quota_root->node, 0, 1);
1325 
1326 
1327 out:
1328 	btrfs_put_root(quota_root);
1329 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1330 	if (ret && trans)
1331 		btrfs_end_transaction(trans);
1332 	else if (trans)
1333 		ret = btrfs_end_transaction(trans);
1334 	mutex_unlock(&fs_info->cleaner_mutex);
1335 
1336 	return ret;
1337 }
1338 
qgroup_dirty(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup)1339 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1340 			 struct btrfs_qgroup *qgroup)
1341 {
1342 	if (list_empty(&qgroup->dirty))
1343 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1344 }
1345 
qgroup_iterator_add(struct list_head * head,struct btrfs_qgroup * qgroup)1346 static void qgroup_iterator_add(struct list_head *head, struct btrfs_qgroup *qgroup)
1347 {
1348 	if (!list_empty(&qgroup->iterator))
1349 		return;
1350 
1351 	list_add_tail(&qgroup->iterator, head);
1352 }
1353 
qgroup_iterator_clean(struct list_head * head)1354 static void qgroup_iterator_clean(struct list_head *head)
1355 {
1356 	while (!list_empty(head)) {
1357 		struct btrfs_qgroup *qgroup;
1358 
1359 		qgroup = list_first_entry(head, struct btrfs_qgroup, iterator);
1360 		list_del_init(&qgroup->iterator);
1361 	}
1362 }
1363 
1364 /*
1365  * The easy accounting, we're updating qgroup relationship whose child qgroup
1366  * only has exclusive extents.
1367  *
1368  * In this case, all exclusive extents will also be exclusive for parent, so
1369  * excl/rfer just get added/removed.
1370  *
1371  * So is qgroup reservation space, which should also be added/removed to
1372  * parent.
1373  * Or when child tries to release reservation space, parent will underflow its
1374  * reservation (for relationship adding case).
1375  *
1376  * Caller should hold fs_info->qgroup_lock.
1377  */
__qgroup_excl_accounting(struct btrfs_fs_info * fs_info,struct ulist * tmp,u64 ref_root,struct btrfs_qgroup * src,int sign)1378 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1379 				    struct ulist *tmp, u64 ref_root,
1380 				    struct btrfs_qgroup *src, int sign)
1381 {
1382 	struct btrfs_qgroup *qgroup;
1383 	struct btrfs_qgroup_list *glist;
1384 	struct ulist_node *unode;
1385 	struct ulist_iterator uiter;
1386 	u64 num_bytes = src->excl;
1387 	int ret = 0;
1388 
1389 	qgroup = find_qgroup_rb(fs_info, ref_root);
1390 	if (!qgroup)
1391 		goto out;
1392 
1393 	qgroup->rfer += sign * num_bytes;
1394 	qgroup->rfer_cmpr += sign * num_bytes;
1395 
1396 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1397 	qgroup->excl += sign * num_bytes;
1398 	qgroup->excl_cmpr += sign * num_bytes;
1399 
1400 	if (sign > 0)
1401 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1402 	else
1403 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1404 
1405 	qgroup_dirty(fs_info, qgroup);
1406 
1407 	/* Get all of the parent groups that contain this qgroup */
1408 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1409 		ret = ulist_add(tmp, glist->group->qgroupid,
1410 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1411 		if (ret < 0)
1412 			goto out;
1413 	}
1414 
1415 	/* Iterate all of the parents and adjust their reference counts */
1416 	ULIST_ITER_INIT(&uiter);
1417 	while ((unode = ulist_next(tmp, &uiter))) {
1418 		qgroup = unode_aux_to_qgroup(unode);
1419 		qgroup->rfer += sign * num_bytes;
1420 		qgroup->rfer_cmpr += sign * num_bytes;
1421 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1422 		qgroup->excl += sign * num_bytes;
1423 		if (sign > 0)
1424 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1425 		else
1426 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1427 		qgroup->excl_cmpr += sign * num_bytes;
1428 		qgroup_dirty(fs_info, qgroup);
1429 
1430 		/* Add any parents of the parents */
1431 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1432 			ret = ulist_add(tmp, glist->group->qgroupid,
1433 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1434 			if (ret < 0)
1435 				goto out;
1436 		}
1437 	}
1438 	ret = 0;
1439 out:
1440 	return ret;
1441 }
1442 
1443 
1444 /*
1445  * Quick path for updating qgroup with only excl refs.
1446  *
1447  * In that case, just update all parent will be enough.
1448  * Or we needs to do a full rescan.
1449  * Caller should also hold fs_info->qgroup_lock.
1450  *
1451  * Return 0 for quick update, return >0 for need to full rescan
1452  * and mark INCONSISTENT flag.
1453  * Return < 0 for other error.
1454  */
quick_update_accounting(struct btrfs_fs_info * fs_info,struct ulist * tmp,u64 src,u64 dst,int sign)1455 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1456 				   struct ulist *tmp, u64 src, u64 dst,
1457 				   int sign)
1458 {
1459 	struct btrfs_qgroup *qgroup;
1460 	int ret = 1;
1461 	int err = 0;
1462 
1463 	qgroup = find_qgroup_rb(fs_info, src);
1464 	if (!qgroup)
1465 		goto out;
1466 	if (qgroup->excl == qgroup->rfer) {
1467 		ret = 0;
1468 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1469 					       qgroup, sign);
1470 		if (err < 0) {
1471 			ret = err;
1472 			goto out;
1473 		}
1474 	}
1475 out:
1476 	if (ret)
1477 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1478 	return ret;
1479 }
1480 
btrfs_add_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1481 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1482 			      u64 dst)
1483 {
1484 	struct btrfs_fs_info *fs_info = trans->fs_info;
1485 	struct btrfs_qgroup *parent;
1486 	struct btrfs_qgroup *member;
1487 	struct btrfs_qgroup_list *list;
1488 	struct ulist *tmp;
1489 	unsigned int nofs_flag;
1490 	int ret = 0;
1491 
1492 	/* Check the level of src and dst first */
1493 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1494 		return -EINVAL;
1495 
1496 	/* We hold a transaction handle open, must do a NOFS allocation. */
1497 	nofs_flag = memalloc_nofs_save();
1498 	tmp = ulist_alloc(GFP_KERNEL);
1499 	memalloc_nofs_restore(nofs_flag);
1500 	if (!tmp)
1501 		return -ENOMEM;
1502 
1503 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1504 	if (!fs_info->quota_root) {
1505 		ret = -ENOTCONN;
1506 		goto out;
1507 	}
1508 	member = find_qgroup_rb(fs_info, src);
1509 	parent = find_qgroup_rb(fs_info, dst);
1510 	if (!member || !parent) {
1511 		ret = -EINVAL;
1512 		goto out;
1513 	}
1514 
1515 	/* check if such qgroup relation exist firstly */
1516 	list_for_each_entry(list, &member->groups, next_group) {
1517 		if (list->group == parent) {
1518 			ret = -EEXIST;
1519 			goto out;
1520 		}
1521 	}
1522 
1523 	ret = add_qgroup_relation_item(trans, src, dst);
1524 	if (ret)
1525 		goto out;
1526 
1527 	ret = add_qgroup_relation_item(trans, dst, src);
1528 	if (ret) {
1529 		del_qgroup_relation_item(trans, src, dst);
1530 		goto out;
1531 	}
1532 
1533 	spin_lock(&fs_info->qgroup_lock);
1534 	ret = __add_relation_rb(member, parent);
1535 	if (ret < 0) {
1536 		spin_unlock(&fs_info->qgroup_lock);
1537 		goto out;
1538 	}
1539 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1540 	spin_unlock(&fs_info->qgroup_lock);
1541 out:
1542 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1543 	ulist_free(tmp);
1544 	return ret;
1545 }
1546 
__del_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1547 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1548 				 u64 dst)
1549 {
1550 	struct btrfs_fs_info *fs_info = trans->fs_info;
1551 	struct btrfs_qgroup *parent;
1552 	struct btrfs_qgroup *member;
1553 	struct btrfs_qgroup_list *list;
1554 	struct ulist *tmp;
1555 	bool found = false;
1556 	unsigned int nofs_flag;
1557 	int ret = 0;
1558 	int ret2;
1559 
1560 	/* We hold a transaction handle open, must do a NOFS allocation. */
1561 	nofs_flag = memalloc_nofs_save();
1562 	tmp = ulist_alloc(GFP_KERNEL);
1563 	memalloc_nofs_restore(nofs_flag);
1564 	if (!tmp)
1565 		return -ENOMEM;
1566 
1567 	if (!fs_info->quota_root) {
1568 		ret = -ENOTCONN;
1569 		goto out;
1570 	}
1571 
1572 	member = find_qgroup_rb(fs_info, src);
1573 	parent = find_qgroup_rb(fs_info, dst);
1574 	/*
1575 	 * The parent/member pair doesn't exist, then try to delete the dead
1576 	 * relation items only.
1577 	 */
1578 	if (!member || !parent)
1579 		goto delete_item;
1580 
1581 	/* check if such qgroup relation exist firstly */
1582 	list_for_each_entry(list, &member->groups, next_group) {
1583 		if (list->group == parent) {
1584 			found = true;
1585 			break;
1586 		}
1587 	}
1588 
1589 delete_item:
1590 	ret = del_qgroup_relation_item(trans, src, dst);
1591 	if (ret < 0 && ret != -ENOENT)
1592 		goto out;
1593 	ret2 = del_qgroup_relation_item(trans, dst, src);
1594 	if (ret2 < 0 && ret2 != -ENOENT)
1595 		goto out;
1596 
1597 	/* At least one deletion succeeded, return 0 */
1598 	if (!ret || !ret2)
1599 		ret = 0;
1600 
1601 	if (found) {
1602 		spin_lock(&fs_info->qgroup_lock);
1603 		del_relation_rb(fs_info, src, dst);
1604 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1605 		spin_unlock(&fs_info->qgroup_lock);
1606 	}
1607 out:
1608 	ulist_free(tmp);
1609 	return ret;
1610 }
1611 
btrfs_del_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1612 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1613 			      u64 dst)
1614 {
1615 	struct btrfs_fs_info *fs_info = trans->fs_info;
1616 	int ret = 0;
1617 
1618 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1619 	ret = __del_qgroup_relation(trans, src, dst);
1620 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1621 
1622 	return ret;
1623 }
1624 
btrfs_create_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)1625 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1626 {
1627 	struct btrfs_fs_info *fs_info = trans->fs_info;
1628 	struct btrfs_root *quota_root;
1629 	struct btrfs_qgroup *qgroup;
1630 	int ret = 0;
1631 
1632 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1633 	if (!fs_info->quota_root) {
1634 		ret = -ENOTCONN;
1635 		goto out;
1636 	}
1637 	quota_root = fs_info->quota_root;
1638 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1639 	if (qgroup) {
1640 		ret = -EEXIST;
1641 		goto out;
1642 	}
1643 
1644 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1645 	if (ret)
1646 		goto out;
1647 
1648 	spin_lock(&fs_info->qgroup_lock);
1649 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1650 	spin_unlock(&fs_info->qgroup_lock);
1651 
1652 	if (IS_ERR(qgroup)) {
1653 		ret = PTR_ERR(qgroup);
1654 		goto out;
1655 	}
1656 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1657 out:
1658 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1659 	return ret;
1660 }
1661 
qgroup_has_usage(struct btrfs_qgroup * qgroup)1662 static bool qgroup_has_usage(struct btrfs_qgroup *qgroup)
1663 {
1664 	return (qgroup->rfer > 0 || qgroup->rfer_cmpr > 0 ||
1665 		qgroup->excl > 0 || qgroup->excl_cmpr > 0 ||
1666 		qgroup->rsv.values[BTRFS_QGROUP_RSV_DATA] > 0 ||
1667 		qgroup->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC] > 0 ||
1668 		qgroup->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > 0);
1669 }
1670 
btrfs_remove_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)1671 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1672 {
1673 	struct btrfs_fs_info *fs_info = trans->fs_info;
1674 	struct btrfs_qgroup *qgroup;
1675 	struct btrfs_qgroup_list *list;
1676 	int ret = 0;
1677 
1678 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1679 	if (!fs_info->quota_root) {
1680 		ret = -ENOTCONN;
1681 		goto out;
1682 	}
1683 
1684 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1685 	if (!qgroup) {
1686 		ret = -ENOENT;
1687 		goto out;
1688 	}
1689 
1690 	if (is_fstree(qgroupid) && qgroup_has_usage(qgroup)) {
1691 		ret = -EBUSY;
1692 		goto out;
1693 	}
1694 
1695 	/* Check if there are no children of this qgroup */
1696 	if (!list_empty(&qgroup->members)) {
1697 		ret = -EBUSY;
1698 		goto out;
1699 	}
1700 
1701 	ret = del_qgroup_item(trans, qgroupid);
1702 	if (ret && ret != -ENOENT)
1703 		goto out;
1704 
1705 	while (!list_empty(&qgroup->groups)) {
1706 		list = list_first_entry(&qgroup->groups,
1707 					struct btrfs_qgroup_list, next_group);
1708 		ret = __del_qgroup_relation(trans, qgroupid,
1709 					    list->group->qgroupid);
1710 		if (ret)
1711 			goto out;
1712 	}
1713 
1714 	spin_lock(&fs_info->qgroup_lock);
1715 	del_qgroup_rb(fs_info, qgroupid);
1716 	spin_unlock(&fs_info->qgroup_lock);
1717 
1718 	/*
1719 	 * Remove the qgroup from sysfs now without holding the qgroup_lock
1720 	 * spinlock, since the sysfs_remove_group() function needs to take
1721 	 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1722 	 */
1723 	btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1724 	kfree(qgroup);
1725 out:
1726 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1727 	return ret;
1728 }
1729 
btrfs_limit_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid,struct btrfs_qgroup_limit * limit)1730 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1731 		       struct btrfs_qgroup_limit *limit)
1732 {
1733 	struct btrfs_fs_info *fs_info = trans->fs_info;
1734 	struct btrfs_qgroup *qgroup;
1735 	int ret = 0;
1736 	/* Sometimes we would want to clear the limit on this qgroup.
1737 	 * To meet this requirement, we treat the -1 as a special value
1738 	 * which tell kernel to clear the limit on this qgroup.
1739 	 */
1740 	const u64 CLEAR_VALUE = -1;
1741 
1742 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1743 	if (!fs_info->quota_root) {
1744 		ret = -ENOTCONN;
1745 		goto out;
1746 	}
1747 
1748 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1749 	if (!qgroup) {
1750 		ret = -ENOENT;
1751 		goto out;
1752 	}
1753 
1754 	spin_lock(&fs_info->qgroup_lock);
1755 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1756 		if (limit->max_rfer == CLEAR_VALUE) {
1757 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1758 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1759 			qgroup->max_rfer = 0;
1760 		} else {
1761 			qgroup->max_rfer = limit->max_rfer;
1762 		}
1763 	}
1764 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1765 		if (limit->max_excl == CLEAR_VALUE) {
1766 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1767 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1768 			qgroup->max_excl = 0;
1769 		} else {
1770 			qgroup->max_excl = limit->max_excl;
1771 		}
1772 	}
1773 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1774 		if (limit->rsv_rfer == CLEAR_VALUE) {
1775 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1776 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1777 			qgroup->rsv_rfer = 0;
1778 		} else {
1779 			qgroup->rsv_rfer = limit->rsv_rfer;
1780 		}
1781 	}
1782 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1783 		if (limit->rsv_excl == CLEAR_VALUE) {
1784 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1785 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1786 			qgroup->rsv_excl = 0;
1787 		} else {
1788 			qgroup->rsv_excl = limit->rsv_excl;
1789 		}
1790 	}
1791 	qgroup->lim_flags |= limit->flags;
1792 
1793 	spin_unlock(&fs_info->qgroup_lock);
1794 
1795 	ret = update_qgroup_limit_item(trans, qgroup);
1796 	if (ret) {
1797 		qgroup_mark_inconsistent(fs_info);
1798 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1799 		       qgroupid);
1800 	}
1801 
1802 out:
1803 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1804 	return ret;
1805 }
1806 
btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info * fs_info,struct btrfs_delayed_ref_root * delayed_refs,struct btrfs_qgroup_extent_record * record)1807 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1808 				struct btrfs_delayed_ref_root *delayed_refs,
1809 				struct btrfs_qgroup_extent_record *record)
1810 {
1811 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1812 	struct rb_node *parent_node = NULL;
1813 	struct btrfs_qgroup_extent_record *entry;
1814 	u64 bytenr = record->bytenr;
1815 
1816 	lockdep_assert_held(&delayed_refs->lock);
1817 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1818 
1819 	while (*p) {
1820 		parent_node = *p;
1821 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1822 				 node);
1823 		if (bytenr < entry->bytenr) {
1824 			p = &(*p)->rb_left;
1825 		} else if (bytenr > entry->bytenr) {
1826 			p = &(*p)->rb_right;
1827 		} else {
1828 			if (record->data_rsv && !entry->data_rsv) {
1829 				entry->data_rsv = record->data_rsv;
1830 				entry->data_rsv_refroot =
1831 					record->data_rsv_refroot;
1832 			}
1833 			return 1;
1834 		}
1835 	}
1836 
1837 	rb_link_node(&record->node, parent_node, p);
1838 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1839 	return 0;
1840 }
1841 
btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle * trans,struct btrfs_qgroup_extent_record * qrecord)1842 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1843 				   struct btrfs_qgroup_extent_record *qrecord)
1844 {
1845 	struct btrfs_backref_walk_ctx ctx = { 0 };
1846 	int ret;
1847 
1848 	/*
1849 	 * We are always called in a context where we are already holding a
1850 	 * transaction handle. Often we are called when adding a data delayed
1851 	 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1852 	 * in which case we will be holding a write lock on extent buffer from a
1853 	 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1854 	 * acquire fs_info->commit_root_sem, because that is a higher level lock
1855 	 * that must be acquired before locking any extent buffers.
1856 	 *
1857 	 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1858 	 * but we can't pass it a non-NULL transaction handle, because otherwise
1859 	 * it would not use commit roots and would lock extent buffers, causing
1860 	 * a deadlock if it ends up trying to read lock the same extent buffer
1861 	 * that was previously write locked at btrfs_truncate_inode_items().
1862 	 *
1863 	 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1864 	 * explicitly tell it to not acquire the commit_root_sem - if we are
1865 	 * holding a transaction handle we don't need its protection.
1866 	 */
1867 	ASSERT(trans != NULL);
1868 
1869 	if (trans->fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
1870 		return 0;
1871 
1872 	ctx.bytenr = qrecord->bytenr;
1873 	ctx.fs_info = trans->fs_info;
1874 
1875 	ret = btrfs_find_all_roots(&ctx, true);
1876 	if (ret < 0) {
1877 		qgroup_mark_inconsistent(trans->fs_info);
1878 		btrfs_warn(trans->fs_info,
1879 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1880 			ret);
1881 		return 0;
1882 	}
1883 
1884 	/*
1885 	 * Here we don't need to get the lock of
1886 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1887 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1888 	 *
1889 	 * So modifying qrecord->old_roots is safe here
1890 	 */
1891 	qrecord->old_roots = ctx.roots;
1892 	return 0;
1893 }
1894 
btrfs_qgroup_trace_extent(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes)1895 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1896 			      u64 num_bytes)
1897 {
1898 	struct btrfs_fs_info *fs_info = trans->fs_info;
1899 	struct btrfs_qgroup_extent_record *record;
1900 	struct btrfs_delayed_ref_root *delayed_refs;
1901 	int ret;
1902 
1903 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1904 	    || bytenr == 0 || num_bytes == 0)
1905 		return 0;
1906 	record = kzalloc(sizeof(*record), GFP_NOFS);
1907 	if (!record)
1908 		return -ENOMEM;
1909 
1910 	delayed_refs = &trans->transaction->delayed_refs;
1911 	record->bytenr = bytenr;
1912 	record->num_bytes = num_bytes;
1913 	record->old_roots = NULL;
1914 
1915 	spin_lock(&delayed_refs->lock);
1916 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1917 	spin_unlock(&delayed_refs->lock);
1918 	if (ret > 0) {
1919 		kfree(record);
1920 		return 0;
1921 	}
1922 	return btrfs_qgroup_trace_extent_post(trans, record);
1923 }
1924 
btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle * trans,struct extent_buffer * eb)1925 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1926 				  struct extent_buffer *eb)
1927 {
1928 	struct btrfs_fs_info *fs_info = trans->fs_info;
1929 	int nr = btrfs_header_nritems(eb);
1930 	int i, extent_type, ret;
1931 	struct btrfs_key key;
1932 	struct btrfs_file_extent_item *fi;
1933 	u64 bytenr, num_bytes;
1934 
1935 	/* We can be called directly from walk_up_proc() */
1936 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1937 		return 0;
1938 
1939 	for (i = 0; i < nr; i++) {
1940 		btrfs_item_key_to_cpu(eb, &key, i);
1941 
1942 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1943 			continue;
1944 
1945 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1946 		/* filter out non qgroup-accountable extents  */
1947 		extent_type = btrfs_file_extent_type(eb, fi);
1948 
1949 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1950 			continue;
1951 
1952 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1953 		if (!bytenr)
1954 			continue;
1955 
1956 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1957 
1958 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes);
1959 		if (ret)
1960 			return ret;
1961 	}
1962 	cond_resched();
1963 	return 0;
1964 }
1965 
1966 /*
1967  * Walk up the tree from the bottom, freeing leaves and any interior
1968  * nodes which have had all slots visited. If a node (leaf or
1969  * interior) is freed, the node above it will have it's slot
1970  * incremented. The root node will never be freed.
1971  *
1972  * At the end of this function, we should have a path which has all
1973  * slots incremented to the next position for a search. If we need to
1974  * read a new node it will be NULL and the node above it will have the
1975  * correct slot selected for a later read.
1976  *
1977  * If we increment the root nodes slot counter past the number of
1978  * elements, 1 is returned to signal completion of the search.
1979  */
adjust_slots_upwards(struct btrfs_path * path,int root_level)1980 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1981 {
1982 	int level = 0;
1983 	int nr, slot;
1984 	struct extent_buffer *eb;
1985 
1986 	if (root_level == 0)
1987 		return 1;
1988 
1989 	while (level <= root_level) {
1990 		eb = path->nodes[level];
1991 		nr = btrfs_header_nritems(eb);
1992 		path->slots[level]++;
1993 		slot = path->slots[level];
1994 		if (slot >= nr || level == 0) {
1995 			/*
1996 			 * Don't free the root -  we will detect this
1997 			 * condition after our loop and return a
1998 			 * positive value for caller to stop walking the tree.
1999 			 */
2000 			if (level != root_level) {
2001 				btrfs_tree_unlock_rw(eb, path->locks[level]);
2002 				path->locks[level] = 0;
2003 
2004 				free_extent_buffer(eb);
2005 				path->nodes[level] = NULL;
2006 				path->slots[level] = 0;
2007 			}
2008 		} else {
2009 			/*
2010 			 * We have a valid slot to walk back down
2011 			 * from. Stop here so caller can process these
2012 			 * new nodes.
2013 			 */
2014 			break;
2015 		}
2016 
2017 		level++;
2018 	}
2019 
2020 	eb = path->nodes[root_level];
2021 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
2022 		return 1;
2023 
2024 	return 0;
2025 }
2026 
2027 /*
2028  * Helper function to trace a subtree tree block swap.
2029  *
2030  * The swap will happen in highest tree block, but there may be a lot of
2031  * tree blocks involved.
2032  *
2033  * For example:
2034  *  OO = Old tree blocks
2035  *  NN = New tree blocks allocated during balance
2036  *
2037  *           File tree (257)                  Reloc tree for 257
2038  * L2              OO                                NN
2039  *               /    \                            /    \
2040  * L1          OO      OO (a)                    OO      NN (a)
2041  *            / \     / \                       / \     / \
2042  * L0       OO   OO OO   OO                   OO   OO NN   NN
2043  *                  (b)  (c)                          (b)  (c)
2044  *
2045  * When calling qgroup_trace_extent_swap(), we will pass:
2046  * @src_eb = OO(a)
2047  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
2048  * @dst_level = 0
2049  * @root_level = 1
2050  *
2051  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
2052  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
2053  *
2054  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
2055  *
2056  * 1) Tree search from @src_eb
2057  *    It should acts as a simplified btrfs_search_slot().
2058  *    The key for search can be extracted from @dst_path->nodes[dst_level]
2059  *    (first key).
2060  *
2061  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
2062  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
2063  *    They should be marked during previous (@dst_level = 1) iteration.
2064  *
2065  * 3) Mark file extents in leaves dirty
2066  *    We don't have good way to pick out new file extents only.
2067  *    So we still follow the old method by scanning all file extents in
2068  *    the leave.
2069  *
2070  * This function can free us from keeping two paths, thus later we only need
2071  * to care about how to iterate all new tree blocks in reloc tree.
2072  */
qgroup_trace_extent_swap(struct btrfs_trans_handle * trans,struct extent_buffer * src_eb,struct btrfs_path * dst_path,int dst_level,int root_level,bool trace_leaf)2073 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
2074 				    struct extent_buffer *src_eb,
2075 				    struct btrfs_path *dst_path,
2076 				    int dst_level, int root_level,
2077 				    bool trace_leaf)
2078 {
2079 	struct btrfs_key key;
2080 	struct btrfs_path *src_path;
2081 	struct btrfs_fs_info *fs_info = trans->fs_info;
2082 	u32 nodesize = fs_info->nodesize;
2083 	int cur_level = root_level;
2084 	int ret;
2085 
2086 	BUG_ON(dst_level > root_level);
2087 	/* Level mismatch */
2088 	if (btrfs_header_level(src_eb) != root_level)
2089 		return -EINVAL;
2090 
2091 	src_path = btrfs_alloc_path();
2092 	if (!src_path) {
2093 		ret = -ENOMEM;
2094 		goto out;
2095 	}
2096 
2097 	if (dst_level)
2098 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2099 	else
2100 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2101 
2102 	/* For src_path */
2103 	atomic_inc(&src_eb->refs);
2104 	src_path->nodes[root_level] = src_eb;
2105 	src_path->slots[root_level] = dst_path->slots[root_level];
2106 	src_path->locks[root_level] = 0;
2107 
2108 	/* A simplified version of btrfs_search_slot() */
2109 	while (cur_level >= dst_level) {
2110 		struct btrfs_key src_key;
2111 		struct btrfs_key dst_key;
2112 
2113 		if (src_path->nodes[cur_level] == NULL) {
2114 			struct extent_buffer *eb;
2115 			int parent_slot;
2116 
2117 			eb = src_path->nodes[cur_level + 1];
2118 			parent_slot = src_path->slots[cur_level + 1];
2119 
2120 			eb = btrfs_read_node_slot(eb, parent_slot);
2121 			if (IS_ERR(eb)) {
2122 				ret = PTR_ERR(eb);
2123 				goto out;
2124 			}
2125 
2126 			src_path->nodes[cur_level] = eb;
2127 
2128 			btrfs_tree_read_lock(eb);
2129 			src_path->locks[cur_level] = BTRFS_READ_LOCK;
2130 		}
2131 
2132 		src_path->slots[cur_level] = dst_path->slots[cur_level];
2133 		if (cur_level) {
2134 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2135 					&dst_key, dst_path->slots[cur_level]);
2136 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2137 					&src_key, src_path->slots[cur_level]);
2138 		} else {
2139 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2140 					&dst_key, dst_path->slots[cur_level]);
2141 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2142 					&src_key, src_path->slots[cur_level]);
2143 		}
2144 		/* Content mismatch, something went wrong */
2145 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2146 			ret = -ENOENT;
2147 			goto out;
2148 		}
2149 		cur_level--;
2150 	}
2151 
2152 	/*
2153 	 * Now both @dst_path and @src_path have been populated, record the tree
2154 	 * blocks for qgroup accounting.
2155 	 */
2156 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2157 					nodesize);
2158 	if (ret < 0)
2159 		goto out;
2160 	ret = btrfs_qgroup_trace_extent(trans, dst_path->nodes[dst_level]->start,
2161 					nodesize);
2162 	if (ret < 0)
2163 		goto out;
2164 
2165 	/* Record leaf file extents */
2166 	if (dst_level == 0 && trace_leaf) {
2167 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2168 		if (ret < 0)
2169 			goto out;
2170 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2171 	}
2172 out:
2173 	btrfs_free_path(src_path);
2174 	return ret;
2175 }
2176 
2177 /*
2178  * Helper function to do recursive generation-aware depth-first search, to
2179  * locate all new tree blocks in a subtree of reloc tree.
2180  *
2181  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2182  *         reloc tree
2183  * L2         NN (a)
2184  *          /    \
2185  * L1    OO        NN (b)
2186  *      /  \      /  \
2187  * L0  OO  OO    OO  NN
2188  *               (c) (d)
2189  * If we pass:
2190  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2191  * @cur_level = 1
2192  * @root_level = 1
2193  *
2194  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2195  * above tree blocks along with their counter parts in file tree.
2196  * While during search, old tree blocks OO(c) will be skipped as tree block swap
2197  * won't affect OO(c).
2198  */
qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle * trans,struct extent_buffer * src_eb,struct btrfs_path * dst_path,int cur_level,int root_level,u64 last_snapshot,bool trace_leaf)2199 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2200 					   struct extent_buffer *src_eb,
2201 					   struct btrfs_path *dst_path,
2202 					   int cur_level, int root_level,
2203 					   u64 last_snapshot, bool trace_leaf)
2204 {
2205 	struct btrfs_fs_info *fs_info = trans->fs_info;
2206 	struct extent_buffer *eb;
2207 	bool need_cleanup = false;
2208 	int ret = 0;
2209 	int i;
2210 
2211 	/* Level sanity check */
2212 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2213 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2214 	    root_level < cur_level) {
2215 		btrfs_err_rl(fs_info,
2216 			"%s: bad levels, cur_level=%d root_level=%d",
2217 			__func__, cur_level, root_level);
2218 		return -EUCLEAN;
2219 	}
2220 
2221 	/* Read the tree block if needed */
2222 	if (dst_path->nodes[cur_level] == NULL) {
2223 		int parent_slot;
2224 		u64 child_gen;
2225 
2226 		/*
2227 		 * dst_path->nodes[root_level] must be initialized before
2228 		 * calling this function.
2229 		 */
2230 		if (cur_level == root_level) {
2231 			btrfs_err_rl(fs_info,
2232 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2233 				__func__, root_level, root_level, cur_level);
2234 			return -EUCLEAN;
2235 		}
2236 
2237 		/*
2238 		 * We need to get child blockptr/gen from parent before we can
2239 		 * read it.
2240 		  */
2241 		eb = dst_path->nodes[cur_level + 1];
2242 		parent_slot = dst_path->slots[cur_level + 1];
2243 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2244 
2245 		/* This node is old, no need to trace */
2246 		if (child_gen < last_snapshot)
2247 			goto out;
2248 
2249 		eb = btrfs_read_node_slot(eb, parent_slot);
2250 		if (IS_ERR(eb)) {
2251 			ret = PTR_ERR(eb);
2252 			goto out;
2253 		}
2254 
2255 		dst_path->nodes[cur_level] = eb;
2256 		dst_path->slots[cur_level] = 0;
2257 
2258 		btrfs_tree_read_lock(eb);
2259 		dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2260 		need_cleanup = true;
2261 	}
2262 
2263 	/* Now record this tree block and its counter part for qgroups */
2264 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2265 				       root_level, trace_leaf);
2266 	if (ret < 0)
2267 		goto cleanup;
2268 
2269 	eb = dst_path->nodes[cur_level];
2270 
2271 	if (cur_level > 0) {
2272 		/* Iterate all child tree blocks */
2273 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2274 			/* Skip old tree blocks as they won't be swapped */
2275 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2276 				continue;
2277 			dst_path->slots[cur_level] = i;
2278 
2279 			/* Recursive call (at most 7 times) */
2280 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2281 					dst_path, cur_level - 1, root_level,
2282 					last_snapshot, trace_leaf);
2283 			if (ret < 0)
2284 				goto cleanup;
2285 		}
2286 	}
2287 
2288 cleanup:
2289 	if (need_cleanup) {
2290 		/* Clean up */
2291 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2292 				     dst_path->locks[cur_level]);
2293 		free_extent_buffer(dst_path->nodes[cur_level]);
2294 		dst_path->nodes[cur_level] = NULL;
2295 		dst_path->slots[cur_level] = 0;
2296 		dst_path->locks[cur_level] = 0;
2297 	}
2298 out:
2299 	return ret;
2300 }
2301 
qgroup_trace_subtree_swap(struct btrfs_trans_handle * trans,struct extent_buffer * src_eb,struct extent_buffer * dst_eb,u64 last_snapshot,bool trace_leaf)2302 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2303 				struct extent_buffer *src_eb,
2304 				struct extent_buffer *dst_eb,
2305 				u64 last_snapshot, bool trace_leaf)
2306 {
2307 	struct btrfs_fs_info *fs_info = trans->fs_info;
2308 	struct btrfs_path *dst_path = NULL;
2309 	int level;
2310 	int ret;
2311 
2312 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2313 		return 0;
2314 
2315 	/* Wrong parameter order */
2316 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2317 		btrfs_err_rl(fs_info,
2318 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2319 			     btrfs_header_generation(src_eb),
2320 			     btrfs_header_generation(dst_eb));
2321 		return -EUCLEAN;
2322 	}
2323 
2324 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2325 		ret = -EIO;
2326 		goto out;
2327 	}
2328 
2329 	level = btrfs_header_level(dst_eb);
2330 	dst_path = btrfs_alloc_path();
2331 	if (!dst_path) {
2332 		ret = -ENOMEM;
2333 		goto out;
2334 	}
2335 	/* For dst_path */
2336 	atomic_inc(&dst_eb->refs);
2337 	dst_path->nodes[level] = dst_eb;
2338 	dst_path->slots[level] = 0;
2339 	dst_path->locks[level] = 0;
2340 
2341 	/* Do the generation aware breadth-first search */
2342 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2343 					      level, last_snapshot, trace_leaf);
2344 	if (ret < 0)
2345 		goto out;
2346 	ret = 0;
2347 
2348 out:
2349 	btrfs_free_path(dst_path);
2350 	if (ret < 0)
2351 		qgroup_mark_inconsistent(fs_info);
2352 	return ret;
2353 }
2354 
btrfs_qgroup_trace_subtree(struct btrfs_trans_handle * trans,struct extent_buffer * root_eb,u64 root_gen,int root_level)2355 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2356 			       struct extent_buffer *root_eb,
2357 			       u64 root_gen, int root_level)
2358 {
2359 	struct btrfs_fs_info *fs_info = trans->fs_info;
2360 	int ret = 0;
2361 	int level;
2362 	u8 drop_subptree_thres;
2363 	struct extent_buffer *eb = root_eb;
2364 	struct btrfs_path *path = NULL;
2365 
2366 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2367 	BUG_ON(root_eb == NULL);
2368 
2369 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2370 		return 0;
2371 
2372 	spin_lock(&fs_info->qgroup_lock);
2373 	drop_subptree_thres = fs_info->qgroup_drop_subtree_thres;
2374 	spin_unlock(&fs_info->qgroup_lock);
2375 
2376 	/*
2377 	 * This function only gets called for snapshot drop, if we hit a high
2378 	 * node here, it means we are going to change ownership for quite a lot
2379 	 * of extents, which will greatly slow down btrfs_commit_transaction().
2380 	 *
2381 	 * So here if we find a high tree here, we just skip the accounting and
2382 	 * mark qgroup inconsistent.
2383 	 */
2384 	if (root_level >= drop_subptree_thres) {
2385 		qgroup_mark_inconsistent(fs_info);
2386 		return 0;
2387 	}
2388 
2389 	if (!extent_buffer_uptodate(root_eb)) {
2390 		struct btrfs_tree_parent_check check = {
2391 			.has_first_key = false,
2392 			.transid = root_gen,
2393 			.level = root_level
2394 		};
2395 
2396 		ret = btrfs_read_extent_buffer(root_eb, &check);
2397 		if (ret)
2398 			goto out;
2399 	}
2400 
2401 	if (root_level == 0) {
2402 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2403 		goto out;
2404 	}
2405 
2406 	path = btrfs_alloc_path();
2407 	if (!path)
2408 		return -ENOMEM;
2409 
2410 	/*
2411 	 * Walk down the tree.  Missing extent blocks are filled in as
2412 	 * we go. Metadata is accounted every time we read a new
2413 	 * extent block.
2414 	 *
2415 	 * When we reach a leaf, we account for file extent items in it,
2416 	 * walk back up the tree (adjusting slot pointers as we go)
2417 	 * and restart the search process.
2418 	 */
2419 	atomic_inc(&root_eb->refs);	/* For path */
2420 	path->nodes[root_level] = root_eb;
2421 	path->slots[root_level] = 0;
2422 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2423 walk_down:
2424 	level = root_level;
2425 	while (level >= 0) {
2426 		if (path->nodes[level] == NULL) {
2427 			int parent_slot;
2428 			u64 child_bytenr;
2429 
2430 			/*
2431 			 * We need to get child blockptr from parent before we
2432 			 * can read it.
2433 			  */
2434 			eb = path->nodes[level + 1];
2435 			parent_slot = path->slots[level + 1];
2436 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2437 
2438 			eb = btrfs_read_node_slot(eb, parent_slot);
2439 			if (IS_ERR(eb)) {
2440 				ret = PTR_ERR(eb);
2441 				goto out;
2442 			}
2443 
2444 			path->nodes[level] = eb;
2445 			path->slots[level] = 0;
2446 
2447 			btrfs_tree_read_lock(eb);
2448 			path->locks[level] = BTRFS_READ_LOCK;
2449 
2450 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2451 							fs_info->nodesize);
2452 			if (ret)
2453 				goto out;
2454 		}
2455 
2456 		if (level == 0) {
2457 			ret = btrfs_qgroup_trace_leaf_items(trans,
2458 							    path->nodes[level]);
2459 			if (ret)
2460 				goto out;
2461 
2462 			/* Nonzero return here means we completed our search */
2463 			ret = adjust_slots_upwards(path, root_level);
2464 			if (ret)
2465 				break;
2466 
2467 			/* Restart search with new slots */
2468 			goto walk_down;
2469 		}
2470 
2471 		level--;
2472 	}
2473 
2474 	ret = 0;
2475 out:
2476 	btrfs_free_path(path);
2477 
2478 	return ret;
2479 }
2480 
2481 #define UPDATE_NEW	0
2482 #define UPDATE_OLD	1
2483 /*
2484  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2485  */
qgroup_update_refcnt(struct btrfs_fs_info * fs_info,struct ulist * roots,struct ulist * tmp,struct ulist * qgroups,u64 seq,int update_old)2486 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2487 				struct ulist *roots, struct ulist *tmp,
2488 				struct ulist *qgroups, u64 seq, int update_old)
2489 {
2490 	struct ulist_node *unode;
2491 	struct ulist_iterator uiter;
2492 	struct ulist_node *tmp_unode;
2493 	struct ulist_iterator tmp_uiter;
2494 	struct btrfs_qgroup *qg;
2495 	int ret = 0;
2496 
2497 	if (!roots)
2498 		return 0;
2499 	ULIST_ITER_INIT(&uiter);
2500 	while ((unode = ulist_next(roots, &uiter))) {
2501 		qg = find_qgroup_rb(fs_info, unode->val);
2502 		if (!qg)
2503 			continue;
2504 
2505 		ulist_reinit(tmp);
2506 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2507 				GFP_ATOMIC);
2508 		if (ret < 0)
2509 			return ret;
2510 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2511 		if (ret < 0)
2512 			return ret;
2513 		ULIST_ITER_INIT(&tmp_uiter);
2514 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2515 			struct btrfs_qgroup_list *glist;
2516 
2517 			qg = unode_aux_to_qgroup(tmp_unode);
2518 			if (update_old)
2519 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2520 			else
2521 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2522 			list_for_each_entry(glist, &qg->groups, next_group) {
2523 				ret = ulist_add(qgroups, glist->group->qgroupid,
2524 						qgroup_to_aux(glist->group),
2525 						GFP_ATOMIC);
2526 				if (ret < 0)
2527 					return ret;
2528 				ret = ulist_add(tmp, glist->group->qgroupid,
2529 						qgroup_to_aux(glist->group),
2530 						GFP_ATOMIC);
2531 				if (ret < 0)
2532 					return ret;
2533 			}
2534 		}
2535 	}
2536 	return 0;
2537 }
2538 
2539 /*
2540  * Update qgroup rfer/excl counters.
2541  * Rfer update is easy, codes can explain themselves.
2542  *
2543  * Excl update is tricky, the update is split into 2 parts.
2544  * Part 1: Possible exclusive <-> sharing detect:
2545  *	|	A	|	!A	|
2546  *  -------------------------------------
2547  *  B	|	*	|	-	|
2548  *  -------------------------------------
2549  *  !B	|	+	|	**	|
2550  *  -------------------------------------
2551  *
2552  * Conditions:
2553  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2554  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2555  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2556  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2557  *
2558  * Results:
2559  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2560  * *: Definitely not changed.		**: Possible unchanged.
2561  *
2562  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2563  *
2564  * To make the logic clear, we first use condition A and B to split
2565  * combination into 4 results.
2566  *
2567  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2568  * only on variant maybe 0.
2569  *
2570  * Lastly, check result **, since there are 2 variants maybe 0, split them
2571  * again(2x2).
2572  * But this time we don't need to consider other things, the codes and logic
2573  * is easy to understand now.
2574  */
qgroup_update_counters(struct btrfs_fs_info * fs_info,struct ulist * qgroups,u64 nr_old_roots,u64 nr_new_roots,u64 num_bytes,u64 seq)2575 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2576 				  struct ulist *qgroups,
2577 				  u64 nr_old_roots,
2578 				  u64 nr_new_roots,
2579 				  u64 num_bytes, u64 seq)
2580 {
2581 	struct ulist_node *unode;
2582 	struct ulist_iterator uiter;
2583 	struct btrfs_qgroup *qg;
2584 	u64 cur_new_count, cur_old_count;
2585 
2586 	ULIST_ITER_INIT(&uiter);
2587 	while ((unode = ulist_next(qgroups, &uiter))) {
2588 		bool dirty = false;
2589 
2590 		qg = unode_aux_to_qgroup(unode);
2591 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2592 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2593 
2594 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2595 					     cur_new_count);
2596 
2597 		/* Rfer update part */
2598 		if (cur_old_count == 0 && cur_new_count > 0) {
2599 			qg->rfer += num_bytes;
2600 			qg->rfer_cmpr += num_bytes;
2601 			dirty = true;
2602 		}
2603 		if (cur_old_count > 0 && cur_new_count == 0) {
2604 			qg->rfer -= num_bytes;
2605 			qg->rfer_cmpr -= num_bytes;
2606 			dirty = true;
2607 		}
2608 
2609 		/* Excl update part */
2610 		/* Exclusive/none -> shared case */
2611 		if (cur_old_count == nr_old_roots &&
2612 		    cur_new_count < nr_new_roots) {
2613 			/* Exclusive -> shared */
2614 			if (cur_old_count != 0) {
2615 				qg->excl -= num_bytes;
2616 				qg->excl_cmpr -= num_bytes;
2617 				dirty = true;
2618 			}
2619 		}
2620 
2621 		/* Shared -> exclusive/none case */
2622 		if (cur_old_count < nr_old_roots &&
2623 		    cur_new_count == nr_new_roots) {
2624 			/* Shared->exclusive */
2625 			if (cur_new_count != 0) {
2626 				qg->excl += num_bytes;
2627 				qg->excl_cmpr += num_bytes;
2628 				dirty = true;
2629 			}
2630 		}
2631 
2632 		/* Exclusive/none -> exclusive/none case */
2633 		if (cur_old_count == nr_old_roots &&
2634 		    cur_new_count == nr_new_roots) {
2635 			if (cur_old_count == 0) {
2636 				/* None -> exclusive/none */
2637 
2638 				if (cur_new_count != 0) {
2639 					/* None -> exclusive */
2640 					qg->excl += num_bytes;
2641 					qg->excl_cmpr += num_bytes;
2642 					dirty = true;
2643 				}
2644 				/* None -> none, nothing changed */
2645 			} else {
2646 				/* Exclusive -> exclusive/none */
2647 
2648 				if (cur_new_count == 0) {
2649 					/* Exclusive -> none */
2650 					qg->excl -= num_bytes;
2651 					qg->excl_cmpr -= num_bytes;
2652 					dirty = true;
2653 				}
2654 				/* Exclusive -> exclusive, nothing changed */
2655 			}
2656 		}
2657 
2658 		if (dirty)
2659 			qgroup_dirty(fs_info, qg);
2660 	}
2661 	return 0;
2662 }
2663 
2664 /*
2665  * Check if the @roots potentially is a list of fs tree roots
2666  *
2667  * Return 0 for definitely not a fs/subvol tree roots ulist
2668  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2669  *          one as well)
2670  */
maybe_fs_roots(struct ulist * roots)2671 static int maybe_fs_roots(struct ulist *roots)
2672 {
2673 	struct ulist_node *unode;
2674 	struct ulist_iterator uiter;
2675 
2676 	/* Empty one, still possible for fs roots */
2677 	if (!roots || roots->nnodes == 0)
2678 		return 1;
2679 
2680 	ULIST_ITER_INIT(&uiter);
2681 	unode = ulist_next(roots, &uiter);
2682 	if (!unode)
2683 		return 1;
2684 
2685 	/*
2686 	 * If it contains fs tree roots, then it must belong to fs/subvol
2687 	 * trees.
2688 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2689 	 */
2690 	return is_fstree(unode->val);
2691 }
2692 
btrfs_qgroup_account_extent(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes,struct ulist * old_roots,struct ulist * new_roots)2693 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2694 				u64 num_bytes, struct ulist *old_roots,
2695 				struct ulist *new_roots)
2696 {
2697 	struct btrfs_fs_info *fs_info = trans->fs_info;
2698 	struct ulist *qgroups = NULL;
2699 	struct ulist *tmp = NULL;
2700 	u64 seq;
2701 	u64 nr_new_roots = 0;
2702 	u64 nr_old_roots = 0;
2703 	int ret = 0;
2704 
2705 	/*
2706 	 * If quotas get disabled meanwhile, the resources need to be freed and
2707 	 * we can't just exit here.
2708 	 */
2709 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2710 	    fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
2711 		goto out_free;
2712 
2713 	if (new_roots) {
2714 		if (!maybe_fs_roots(new_roots))
2715 			goto out_free;
2716 		nr_new_roots = new_roots->nnodes;
2717 	}
2718 	if (old_roots) {
2719 		if (!maybe_fs_roots(old_roots))
2720 			goto out_free;
2721 		nr_old_roots = old_roots->nnodes;
2722 	}
2723 
2724 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2725 	if (nr_old_roots == 0 && nr_new_roots == 0)
2726 		goto out_free;
2727 
2728 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2729 					num_bytes, nr_old_roots, nr_new_roots);
2730 
2731 	qgroups = ulist_alloc(GFP_NOFS);
2732 	if (!qgroups) {
2733 		ret = -ENOMEM;
2734 		goto out_free;
2735 	}
2736 	tmp = ulist_alloc(GFP_NOFS);
2737 	if (!tmp) {
2738 		ret = -ENOMEM;
2739 		goto out_free;
2740 	}
2741 
2742 	mutex_lock(&fs_info->qgroup_rescan_lock);
2743 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2744 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2745 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2746 			ret = 0;
2747 			goto out_free;
2748 		}
2749 	}
2750 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2751 
2752 	spin_lock(&fs_info->qgroup_lock);
2753 	seq = fs_info->qgroup_seq;
2754 
2755 	/* Update old refcnts using old_roots */
2756 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2757 				   UPDATE_OLD);
2758 	if (ret < 0)
2759 		goto out;
2760 
2761 	/* Update new refcnts using new_roots */
2762 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2763 				   UPDATE_NEW);
2764 	if (ret < 0)
2765 		goto out;
2766 
2767 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2768 			       num_bytes, seq);
2769 
2770 	/*
2771 	 * Bump qgroup_seq to avoid seq overlap
2772 	 */
2773 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2774 out:
2775 	spin_unlock(&fs_info->qgroup_lock);
2776 out_free:
2777 	ulist_free(tmp);
2778 	ulist_free(qgroups);
2779 	ulist_free(old_roots);
2780 	ulist_free(new_roots);
2781 	return ret;
2782 }
2783 
btrfs_qgroup_account_extents(struct btrfs_trans_handle * trans)2784 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2785 {
2786 	struct btrfs_fs_info *fs_info = trans->fs_info;
2787 	struct btrfs_qgroup_extent_record *record;
2788 	struct btrfs_delayed_ref_root *delayed_refs;
2789 	struct ulist *new_roots = NULL;
2790 	struct rb_node *node;
2791 	u64 num_dirty_extents = 0;
2792 	u64 qgroup_to_skip;
2793 	int ret = 0;
2794 
2795 	delayed_refs = &trans->transaction->delayed_refs;
2796 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2797 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2798 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2799 				  node);
2800 
2801 		num_dirty_extents++;
2802 		trace_btrfs_qgroup_account_extents(fs_info, record);
2803 
2804 		if (!ret && !(fs_info->qgroup_flags &
2805 			      BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)) {
2806 			struct btrfs_backref_walk_ctx ctx = { 0 };
2807 
2808 			ctx.bytenr = record->bytenr;
2809 			ctx.fs_info = fs_info;
2810 
2811 			/*
2812 			 * Old roots should be searched when inserting qgroup
2813 			 * extent record.
2814 			 *
2815 			 * But for INCONSISTENT (NO_ACCOUNTING) -> rescan case,
2816 			 * we may have some record inserted during
2817 			 * NO_ACCOUNTING (thus no old_roots populated), but
2818 			 * later we start rescan, which clears NO_ACCOUNTING,
2819 			 * leaving some inserted records without old_roots
2820 			 * populated.
2821 			 *
2822 			 * Those cases are rare and should not cause too much
2823 			 * time spent during commit_transaction().
2824 			 */
2825 			if (!record->old_roots) {
2826 				/* Search commit root to find old_roots */
2827 				ret = btrfs_find_all_roots(&ctx, false);
2828 				if (ret < 0)
2829 					goto cleanup;
2830 				record->old_roots = ctx.roots;
2831 				ctx.roots = NULL;
2832 			}
2833 
2834 			/*
2835 			 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2836 			 * which doesn't lock tree or delayed_refs and search
2837 			 * current root. It's safe inside commit_transaction().
2838 			 */
2839 			ctx.trans = trans;
2840 			ctx.time_seq = BTRFS_SEQ_LAST;
2841 			ret = btrfs_find_all_roots(&ctx, false);
2842 			if (ret < 0)
2843 				goto cleanup;
2844 			new_roots = ctx.roots;
2845 			if (qgroup_to_skip) {
2846 				ulist_del(new_roots, qgroup_to_skip, 0);
2847 				ulist_del(record->old_roots, qgroup_to_skip,
2848 					  0);
2849 			}
2850 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2851 							  record->num_bytes,
2852 							  record->old_roots,
2853 							  new_roots);
2854 			record->old_roots = NULL;
2855 			new_roots = NULL;
2856 		}
2857 		/* Free the reserved data space */
2858 		btrfs_qgroup_free_refroot(fs_info,
2859 				record->data_rsv_refroot,
2860 				record->data_rsv,
2861 				BTRFS_QGROUP_RSV_DATA);
2862 cleanup:
2863 		ulist_free(record->old_roots);
2864 		ulist_free(new_roots);
2865 		new_roots = NULL;
2866 		rb_erase(node, &delayed_refs->dirty_extent_root);
2867 		kfree(record);
2868 
2869 	}
2870 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2871 				       num_dirty_extents);
2872 	return ret;
2873 }
2874 
2875 /*
2876  * Writes all changed qgroups to disk.
2877  * Called by the transaction commit path and the qgroup assign ioctl.
2878  */
btrfs_run_qgroups(struct btrfs_trans_handle * trans)2879 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2880 {
2881 	struct btrfs_fs_info *fs_info = trans->fs_info;
2882 	int ret = 0;
2883 
2884 	/*
2885 	 * In case we are called from the qgroup assign ioctl, assert that we
2886 	 * are holding the qgroup_ioctl_lock, otherwise we can race with a quota
2887 	 * disable operation (ioctl) and access a freed quota root.
2888 	 */
2889 	if (trans->transaction->state != TRANS_STATE_COMMIT_DOING)
2890 		lockdep_assert_held(&fs_info->qgroup_ioctl_lock);
2891 
2892 	if (!fs_info->quota_root)
2893 		return ret;
2894 
2895 	spin_lock(&fs_info->qgroup_lock);
2896 	while (!list_empty(&fs_info->dirty_qgroups)) {
2897 		struct btrfs_qgroup *qgroup;
2898 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2899 					  struct btrfs_qgroup, dirty);
2900 		list_del_init(&qgroup->dirty);
2901 		spin_unlock(&fs_info->qgroup_lock);
2902 		ret = update_qgroup_info_item(trans, qgroup);
2903 		if (ret)
2904 			qgroup_mark_inconsistent(fs_info);
2905 		ret = update_qgroup_limit_item(trans, qgroup);
2906 		if (ret)
2907 			qgroup_mark_inconsistent(fs_info);
2908 		spin_lock(&fs_info->qgroup_lock);
2909 	}
2910 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2911 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2912 	else
2913 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2914 	spin_unlock(&fs_info->qgroup_lock);
2915 
2916 	ret = update_qgroup_status_item(trans);
2917 	if (ret)
2918 		qgroup_mark_inconsistent(fs_info);
2919 
2920 	return ret;
2921 }
2922 
2923 /*
2924  * Copy the accounting information between qgroups. This is necessary
2925  * when a snapshot or a subvolume is created. Throwing an error will
2926  * cause a transaction abort so we take extra care here to only error
2927  * when a readonly fs is a reasonable outcome.
2928  */
btrfs_qgroup_inherit(struct btrfs_trans_handle * trans,u64 srcid,u64 objectid,struct btrfs_qgroup_inherit * inherit)2929 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2930 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2931 {
2932 	int ret = 0;
2933 	int i;
2934 	u64 *i_qgroups;
2935 	bool committing = false;
2936 	struct btrfs_fs_info *fs_info = trans->fs_info;
2937 	struct btrfs_root *quota_root;
2938 	struct btrfs_qgroup *srcgroup;
2939 	struct btrfs_qgroup *dstgroup;
2940 	bool need_rescan = false;
2941 	u32 level_size = 0;
2942 	u64 nums;
2943 
2944 	/*
2945 	 * There are only two callers of this function.
2946 	 *
2947 	 * One in create_subvol() in the ioctl context, which needs to hold
2948 	 * the qgroup_ioctl_lock.
2949 	 *
2950 	 * The other one in create_pending_snapshot() where no other qgroup
2951 	 * code can modify the fs as they all need to either start a new trans
2952 	 * or hold a trans handler, thus we don't need to hold
2953 	 * qgroup_ioctl_lock.
2954 	 * This would avoid long and complex lock chain and make lockdep happy.
2955 	 */
2956 	spin_lock(&fs_info->trans_lock);
2957 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2958 		committing = true;
2959 	spin_unlock(&fs_info->trans_lock);
2960 
2961 	if (!committing)
2962 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2963 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2964 		goto out;
2965 
2966 	quota_root = fs_info->quota_root;
2967 	if (!quota_root) {
2968 		ret = -EINVAL;
2969 		goto out;
2970 	}
2971 
2972 	if (inherit) {
2973 		i_qgroups = (u64 *)(inherit + 1);
2974 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2975 		       2 * inherit->num_excl_copies;
2976 		for (i = 0; i < nums; ++i) {
2977 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2978 
2979 			/*
2980 			 * Zero out invalid groups so we can ignore
2981 			 * them later.
2982 			 */
2983 			if (!srcgroup ||
2984 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2985 				*i_qgroups = 0ULL;
2986 
2987 			++i_qgroups;
2988 		}
2989 	}
2990 
2991 	/*
2992 	 * create a tracking group for the subvol itself
2993 	 */
2994 	ret = add_qgroup_item(trans, quota_root, objectid);
2995 	if (ret)
2996 		goto out;
2997 
2998 	/*
2999 	 * add qgroup to all inherited groups
3000 	 */
3001 	if (inherit) {
3002 		i_qgroups = (u64 *)(inherit + 1);
3003 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
3004 			if (*i_qgroups == 0)
3005 				continue;
3006 			ret = add_qgroup_relation_item(trans, objectid,
3007 						       *i_qgroups);
3008 			if (ret && ret != -EEXIST)
3009 				goto out;
3010 			ret = add_qgroup_relation_item(trans, *i_qgroups,
3011 						       objectid);
3012 			if (ret && ret != -EEXIST)
3013 				goto out;
3014 		}
3015 		ret = 0;
3016 	}
3017 
3018 
3019 	spin_lock(&fs_info->qgroup_lock);
3020 
3021 	dstgroup = add_qgroup_rb(fs_info, objectid);
3022 	if (IS_ERR(dstgroup)) {
3023 		ret = PTR_ERR(dstgroup);
3024 		goto unlock;
3025 	}
3026 
3027 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
3028 		dstgroup->lim_flags = inherit->lim.flags;
3029 		dstgroup->max_rfer = inherit->lim.max_rfer;
3030 		dstgroup->max_excl = inherit->lim.max_excl;
3031 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
3032 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
3033 
3034 		qgroup_dirty(fs_info, dstgroup);
3035 	}
3036 
3037 	if (srcid) {
3038 		srcgroup = find_qgroup_rb(fs_info, srcid);
3039 		if (!srcgroup)
3040 			goto unlock;
3041 
3042 		/*
3043 		 * We call inherit after we clone the root in order to make sure
3044 		 * our counts don't go crazy, so at this point the only
3045 		 * difference between the two roots should be the root node.
3046 		 */
3047 		level_size = fs_info->nodesize;
3048 		dstgroup->rfer = srcgroup->rfer;
3049 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
3050 		dstgroup->excl = level_size;
3051 		dstgroup->excl_cmpr = level_size;
3052 		srcgroup->excl = level_size;
3053 		srcgroup->excl_cmpr = level_size;
3054 
3055 		/* inherit the limit info */
3056 		dstgroup->lim_flags = srcgroup->lim_flags;
3057 		dstgroup->max_rfer = srcgroup->max_rfer;
3058 		dstgroup->max_excl = srcgroup->max_excl;
3059 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
3060 		dstgroup->rsv_excl = srcgroup->rsv_excl;
3061 
3062 		qgroup_dirty(fs_info, dstgroup);
3063 		qgroup_dirty(fs_info, srcgroup);
3064 	}
3065 
3066 	if (!inherit)
3067 		goto unlock;
3068 
3069 	i_qgroups = (u64 *)(inherit + 1);
3070 	for (i = 0; i < inherit->num_qgroups; ++i) {
3071 		if (*i_qgroups) {
3072 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
3073 			if (ret)
3074 				goto unlock;
3075 		}
3076 		++i_qgroups;
3077 
3078 		/*
3079 		 * If we're doing a snapshot, and adding the snapshot to a new
3080 		 * qgroup, the numbers are guaranteed to be incorrect.
3081 		 */
3082 		if (srcid)
3083 			need_rescan = true;
3084 	}
3085 
3086 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
3087 		struct btrfs_qgroup *src;
3088 		struct btrfs_qgroup *dst;
3089 
3090 		if (!i_qgroups[0] || !i_qgroups[1])
3091 			continue;
3092 
3093 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
3094 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3095 
3096 		if (!src || !dst) {
3097 			ret = -EINVAL;
3098 			goto unlock;
3099 		}
3100 
3101 		dst->rfer = src->rfer - level_size;
3102 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
3103 
3104 		/* Manually tweaking numbers certainly needs a rescan */
3105 		need_rescan = true;
3106 	}
3107 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
3108 		struct btrfs_qgroup *src;
3109 		struct btrfs_qgroup *dst;
3110 
3111 		if (!i_qgroups[0] || !i_qgroups[1])
3112 			continue;
3113 
3114 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
3115 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3116 
3117 		if (!src || !dst) {
3118 			ret = -EINVAL;
3119 			goto unlock;
3120 		}
3121 
3122 		dst->excl = src->excl + level_size;
3123 		dst->excl_cmpr = src->excl_cmpr + level_size;
3124 		need_rescan = true;
3125 	}
3126 
3127 unlock:
3128 	spin_unlock(&fs_info->qgroup_lock);
3129 	if (!ret)
3130 		ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3131 out:
3132 	if (!committing)
3133 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
3134 	if (need_rescan)
3135 		qgroup_mark_inconsistent(fs_info);
3136 	return ret;
3137 }
3138 
qgroup_check_limits(const struct btrfs_qgroup * qg,u64 num_bytes)3139 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3140 {
3141 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3142 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3143 		return false;
3144 
3145 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3146 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3147 		return false;
3148 
3149 	return true;
3150 }
3151 
qgroup_reserve(struct btrfs_root * root,u64 num_bytes,bool enforce,enum btrfs_qgroup_rsv_type type)3152 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3153 			  enum btrfs_qgroup_rsv_type type)
3154 {
3155 	struct btrfs_qgroup *qgroup;
3156 	struct btrfs_fs_info *fs_info = root->fs_info;
3157 	u64 ref_root = root->root_key.objectid;
3158 	int ret = 0;
3159 	LIST_HEAD(qgroup_list);
3160 
3161 	if (!is_fstree(ref_root))
3162 		return 0;
3163 
3164 	if (num_bytes == 0)
3165 		return 0;
3166 
3167 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3168 	    capable(CAP_SYS_RESOURCE))
3169 		enforce = false;
3170 
3171 	spin_lock(&fs_info->qgroup_lock);
3172 	if (!fs_info->quota_root)
3173 		goto out;
3174 
3175 	qgroup = find_qgroup_rb(fs_info, ref_root);
3176 	if (!qgroup)
3177 		goto out;
3178 
3179 	qgroup_iterator_add(&qgroup_list, qgroup);
3180 	list_for_each_entry(qgroup, &qgroup_list, iterator) {
3181 		struct btrfs_qgroup_list *glist;
3182 
3183 		if (enforce && !qgroup_check_limits(qgroup, num_bytes)) {
3184 			ret = -EDQUOT;
3185 			goto out;
3186 		}
3187 
3188 		list_for_each_entry(glist, &qgroup->groups, next_group)
3189 			qgroup_iterator_add(&qgroup_list, glist->group);
3190 	}
3191 
3192 	ret = 0;
3193 	/*
3194 	 * no limits exceeded, now record the reservation into all qgroups
3195 	 */
3196 	list_for_each_entry(qgroup, &qgroup_list, iterator)
3197 		qgroup_rsv_add(fs_info, qgroup, num_bytes, type);
3198 
3199 out:
3200 	qgroup_iterator_clean(&qgroup_list);
3201 	spin_unlock(&fs_info->qgroup_lock);
3202 	return ret;
3203 }
3204 
3205 /*
3206  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3207  * qgroup).
3208  *
3209  * Will handle all higher level qgroup too.
3210  *
3211  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3212  * This special case is only used for META_PERTRANS type.
3213  */
btrfs_qgroup_free_refroot(struct btrfs_fs_info * fs_info,u64 ref_root,u64 num_bytes,enum btrfs_qgroup_rsv_type type)3214 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3215 			       u64 ref_root, u64 num_bytes,
3216 			       enum btrfs_qgroup_rsv_type type)
3217 {
3218 	struct btrfs_qgroup *qgroup;
3219 	struct ulist_node *unode;
3220 	struct ulist_iterator uiter;
3221 	int ret = 0;
3222 
3223 	if (!is_fstree(ref_root))
3224 		return;
3225 
3226 	if (num_bytes == 0)
3227 		return;
3228 
3229 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3230 		WARN(1, "%s: Invalid type to free", __func__);
3231 		return;
3232 	}
3233 	spin_lock(&fs_info->qgroup_lock);
3234 
3235 	if (!fs_info->quota_root)
3236 		goto out;
3237 
3238 	qgroup = find_qgroup_rb(fs_info, ref_root);
3239 	if (!qgroup)
3240 		goto out;
3241 
3242 	if (num_bytes == (u64)-1)
3243 		/*
3244 		 * We're freeing all pertrans rsv, get reserved value from
3245 		 * level 0 qgroup as real num_bytes to free.
3246 		 */
3247 		num_bytes = qgroup->rsv.values[type];
3248 
3249 	ulist_reinit(fs_info->qgroup_ulist);
3250 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3251 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3252 	if (ret < 0)
3253 		goto out;
3254 	ULIST_ITER_INIT(&uiter);
3255 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3256 		struct btrfs_qgroup *qg;
3257 		struct btrfs_qgroup_list *glist;
3258 
3259 		qg = unode_aux_to_qgroup(unode);
3260 
3261 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3262 
3263 		list_for_each_entry(glist, &qg->groups, next_group) {
3264 			ret = ulist_add(fs_info->qgroup_ulist,
3265 					glist->group->qgroupid,
3266 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3267 			if (ret < 0)
3268 				goto out;
3269 		}
3270 	}
3271 
3272 out:
3273 	spin_unlock(&fs_info->qgroup_lock);
3274 }
3275 
3276 /*
3277  * Check if the leaf is the last leaf. Which means all node pointers
3278  * are at their last position.
3279  */
is_last_leaf(struct btrfs_path * path)3280 static bool is_last_leaf(struct btrfs_path *path)
3281 {
3282 	int i;
3283 
3284 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3285 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3286 			return false;
3287 	}
3288 	return true;
3289 }
3290 
3291 /*
3292  * returns < 0 on error, 0 when more leafs are to be scanned.
3293  * returns 1 when done.
3294  */
qgroup_rescan_leaf(struct btrfs_trans_handle * trans,struct btrfs_path * path)3295 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3296 			      struct btrfs_path *path)
3297 {
3298 	struct btrfs_fs_info *fs_info = trans->fs_info;
3299 	struct btrfs_root *extent_root;
3300 	struct btrfs_key found;
3301 	struct extent_buffer *scratch_leaf = NULL;
3302 	u64 num_bytes;
3303 	bool done;
3304 	int slot;
3305 	int ret;
3306 
3307 	mutex_lock(&fs_info->qgroup_rescan_lock);
3308 	extent_root = btrfs_extent_root(fs_info,
3309 				fs_info->qgroup_rescan_progress.objectid);
3310 	ret = btrfs_search_slot_for_read(extent_root,
3311 					 &fs_info->qgroup_rescan_progress,
3312 					 path, 1, 0);
3313 
3314 	btrfs_debug(fs_info,
3315 		"current progress key (%llu %u %llu), search_slot ret %d",
3316 		fs_info->qgroup_rescan_progress.objectid,
3317 		fs_info->qgroup_rescan_progress.type,
3318 		fs_info->qgroup_rescan_progress.offset, ret);
3319 
3320 	if (ret) {
3321 		/*
3322 		 * The rescan is about to end, we will not be scanning any
3323 		 * further blocks. We cannot unset the RESCAN flag here, because
3324 		 * we want to commit the transaction if everything went well.
3325 		 * To make the live accounting work in this phase, we set our
3326 		 * scan progress pointer such that every real extent objectid
3327 		 * will be smaller.
3328 		 */
3329 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3330 		btrfs_release_path(path);
3331 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3332 		return ret;
3333 	}
3334 	done = is_last_leaf(path);
3335 
3336 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3337 			      btrfs_header_nritems(path->nodes[0]) - 1);
3338 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3339 
3340 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3341 	if (!scratch_leaf) {
3342 		ret = -ENOMEM;
3343 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3344 		goto out;
3345 	}
3346 	slot = path->slots[0];
3347 	btrfs_release_path(path);
3348 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3349 
3350 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3351 		struct btrfs_backref_walk_ctx ctx = { 0 };
3352 
3353 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3354 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3355 		    found.type != BTRFS_METADATA_ITEM_KEY)
3356 			continue;
3357 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3358 			num_bytes = fs_info->nodesize;
3359 		else
3360 			num_bytes = found.offset;
3361 
3362 		ctx.bytenr = found.objectid;
3363 		ctx.fs_info = fs_info;
3364 
3365 		ret = btrfs_find_all_roots(&ctx, false);
3366 		if (ret < 0)
3367 			goto out;
3368 		/* For rescan, just pass old_roots as NULL */
3369 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3370 						  num_bytes, NULL, ctx.roots);
3371 		if (ret < 0)
3372 			goto out;
3373 	}
3374 out:
3375 	if (scratch_leaf)
3376 		free_extent_buffer(scratch_leaf);
3377 
3378 	if (done && !ret) {
3379 		ret = 1;
3380 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3381 	}
3382 	return ret;
3383 }
3384 
rescan_should_stop(struct btrfs_fs_info * fs_info)3385 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3386 {
3387 	return btrfs_fs_closing(fs_info) ||
3388 		test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3389 		!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3390 			  fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3391 }
3392 
btrfs_qgroup_rescan_worker(struct btrfs_work * work)3393 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3394 {
3395 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3396 						     qgroup_rescan_work);
3397 	struct btrfs_path *path;
3398 	struct btrfs_trans_handle *trans = NULL;
3399 	int err = -ENOMEM;
3400 	int ret = 0;
3401 	bool stopped = false;
3402 	bool did_leaf_rescans = false;
3403 
3404 	path = btrfs_alloc_path();
3405 	if (!path)
3406 		goto out;
3407 	/*
3408 	 * Rescan should only search for commit root, and any later difference
3409 	 * should be recorded by qgroup
3410 	 */
3411 	path->search_commit_root = 1;
3412 	path->skip_locking = 1;
3413 
3414 	err = 0;
3415 	while (!err && !(stopped = rescan_should_stop(fs_info))) {
3416 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3417 		if (IS_ERR(trans)) {
3418 			err = PTR_ERR(trans);
3419 			break;
3420 		}
3421 
3422 		err = qgroup_rescan_leaf(trans, path);
3423 		did_leaf_rescans = true;
3424 
3425 		if (err > 0)
3426 			btrfs_commit_transaction(trans);
3427 		else
3428 			btrfs_end_transaction(trans);
3429 	}
3430 
3431 out:
3432 	btrfs_free_path(path);
3433 
3434 	mutex_lock(&fs_info->qgroup_rescan_lock);
3435 	if (err > 0 &&
3436 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3437 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3438 	} else if (err < 0 || stopped) {
3439 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3440 	}
3441 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3442 
3443 	/*
3444 	 * Only update status, since the previous part has already updated the
3445 	 * qgroup info, and only if we did any actual work. This also prevents
3446 	 * race with a concurrent quota disable, which has already set
3447 	 * fs_info->quota_root to NULL and cleared BTRFS_FS_QUOTA_ENABLED at
3448 	 * btrfs_quota_disable().
3449 	 */
3450 	if (did_leaf_rescans) {
3451 		trans = btrfs_start_transaction(fs_info->quota_root, 1);
3452 		if (IS_ERR(trans)) {
3453 			err = PTR_ERR(trans);
3454 			trans = NULL;
3455 			btrfs_err(fs_info,
3456 				  "fail to start transaction for status update: %d",
3457 				  err);
3458 		}
3459 	} else {
3460 		trans = NULL;
3461 	}
3462 
3463 	mutex_lock(&fs_info->qgroup_rescan_lock);
3464 	if (!stopped ||
3465 	    fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN)
3466 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3467 	if (trans) {
3468 		ret = update_qgroup_status_item(trans);
3469 		if (ret < 0) {
3470 			err = ret;
3471 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3472 				  err);
3473 		}
3474 	}
3475 	fs_info->qgroup_rescan_running = false;
3476 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3477 	complete_all(&fs_info->qgroup_rescan_completion);
3478 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3479 
3480 	if (!trans)
3481 		return;
3482 
3483 	btrfs_end_transaction(trans);
3484 
3485 	if (stopped) {
3486 		btrfs_info(fs_info, "qgroup scan paused");
3487 	} else if (fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN) {
3488 		btrfs_info(fs_info, "qgroup scan cancelled");
3489 	} else if (err >= 0) {
3490 		btrfs_info(fs_info, "qgroup scan completed%s",
3491 			err > 0 ? " (inconsistency flag cleared)" : "");
3492 	} else {
3493 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3494 	}
3495 }
3496 
3497 /*
3498  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3499  * memory required for the rescan context.
3500  */
3501 static int
qgroup_rescan_init(struct btrfs_fs_info * fs_info,u64 progress_objectid,int init_flags)3502 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3503 		   int init_flags)
3504 {
3505 	int ret = 0;
3506 
3507 	if (!init_flags) {
3508 		/* we're resuming qgroup rescan at mount time */
3509 		if (!(fs_info->qgroup_flags &
3510 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3511 			btrfs_warn(fs_info,
3512 			"qgroup rescan init failed, qgroup rescan is not queued");
3513 			ret = -EINVAL;
3514 		} else if (!(fs_info->qgroup_flags &
3515 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3516 			btrfs_warn(fs_info,
3517 			"qgroup rescan init failed, qgroup is not enabled");
3518 			ret = -EINVAL;
3519 		}
3520 
3521 		if (ret)
3522 			return ret;
3523 	}
3524 
3525 	mutex_lock(&fs_info->qgroup_rescan_lock);
3526 
3527 	if (init_flags) {
3528 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3529 			btrfs_warn(fs_info,
3530 				   "qgroup rescan is already in progress");
3531 			ret = -EINPROGRESS;
3532 		} else if (!(fs_info->qgroup_flags &
3533 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3534 			btrfs_warn(fs_info,
3535 			"qgroup rescan init failed, qgroup is not enabled");
3536 			ret = -EINVAL;
3537 		} else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3538 			/* Quota disable is in progress */
3539 			ret = -EBUSY;
3540 		}
3541 
3542 		if (ret) {
3543 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3544 			return ret;
3545 		}
3546 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3547 	}
3548 
3549 	memset(&fs_info->qgroup_rescan_progress, 0,
3550 		sizeof(fs_info->qgroup_rescan_progress));
3551 	fs_info->qgroup_flags &= ~(BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
3552 				   BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
3553 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3554 	init_completion(&fs_info->qgroup_rescan_completion);
3555 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3556 
3557 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3558 			btrfs_qgroup_rescan_worker, NULL, NULL);
3559 	return 0;
3560 }
3561 
3562 static void
qgroup_rescan_zero_tracking(struct btrfs_fs_info * fs_info)3563 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3564 {
3565 	struct rb_node *n;
3566 	struct btrfs_qgroup *qgroup;
3567 
3568 	spin_lock(&fs_info->qgroup_lock);
3569 	/* clear all current qgroup tracking information */
3570 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3571 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3572 		qgroup->rfer = 0;
3573 		qgroup->rfer_cmpr = 0;
3574 		qgroup->excl = 0;
3575 		qgroup->excl_cmpr = 0;
3576 		qgroup_dirty(fs_info, qgroup);
3577 	}
3578 	spin_unlock(&fs_info->qgroup_lock);
3579 }
3580 
3581 int
btrfs_qgroup_rescan(struct btrfs_fs_info * fs_info)3582 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3583 {
3584 	int ret = 0;
3585 	struct btrfs_trans_handle *trans;
3586 
3587 	ret = qgroup_rescan_init(fs_info, 0, 1);
3588 	if (ret)
3589 		return ret;
3590 
3591 	/*
3592 	 * We have set the rescan_progress to 0, which means no more
3593 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3594 	 * However, btrfs_qgroup_account_ref may be right after its call
3595 	 * to btrfs_find_all_roots, in which case it would still do the
3596 	 * accounting.
3597 	 * To solve this, we're committing the transaction, which will
3598 	 * ensure we run all delayed refs and only after that, we are
3599 	 * going to clear all tracking information for a clean start.
3600 	 */
3601 
3602 	trans = btrfs_attach_transaction_barrier(fs_info->fs_root);
3603 	if (IS_ERR(trans) && trans != ERR_PTR(-ENOENT)) {
3604 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3605 		return PTR_ERR(trans);
3606 	} else if (trans != ERR_PTR(-ENOENT)) {
3607 		ret = btrfs_commit_transaction(trans);
3608 		if (ret) {
3609 			fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3610 			return ret;
3611 		}
3612 	}
3613 
3614 	qgroup_rescan_zero_tracking(fs_info);
3615 
3616 	mutex_lock(&fs_info->qgroup_rescan_lock);
3617 	fs_info->qgroup_rescan_running = true;
3618 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3619 			 &fs_info->qgroup_rescan_work);
3620 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3621 
3622 	return 0;
3623 }
3624 
btrfs_qgroup_wait_for_completion(struct btrfs_fs_info * fs_info,bool interruptible)3625 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3626 				     bool interruptible)
3627 {
3628 	int running;
3629 	int ret = 0;
3630 
3631 	mutex_lock(&fs_info->qgroup_rescan_lock);
3632 	running = fs_info->qgroup_rescan_running;
3633 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3634 
3635 	if (!running)
3636 		return 0;
3637 
3638 	if (interruptible)
3639 		ret = wait_for_completion_interruptible(
3640 					&fs_info->qgroup_rescan_completion);
3641 	else
3642 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3643 
3644 	return ret;
3645 }
3646 
3647 /*
3648  * this is only called from open_ctree where we're still single threaded, thus
3649  * locking is omitted here.
3650  */
3651 void
btrfs_qgroup_rescan_resume(struct btrfs_fs_info * fs_info)3652 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3653 {
3654 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3655 		mutex_lock(&fs_info->qgroup_rescan_lock);
3656 		fs_info->qgroup_rescan_running = true;
3657 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3658 				 &fs_info->qgroup_rescan_work);
3659 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3660 	}
3661 }
3662 
3663 #define rbtree_iterate_from_safe(node, next, start)				\
3664        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3665 
qgroup_unreserve_range(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len)3666 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3667 				  struct extent_changeset *reserved, u64 start,
3668 				  u64 len)
3669 {
3670 	struct rb_node *node;
3671 	struct rb_node *next;
3672 	struct ulist_node *entry;
3673 	int ret = 0;
3674 
3675 	node = reserved->range_changed.root.rb_node;
3676 	if (!node)
3677 		return 0;
3678 	while (node) {
3679 		entry = rb_entry(node, struct ulist_node, rb_node);
3680 		if (entry->val < start)
3681 			node = node->rb_right;
3682 		else
3683 			node = node->rb_left;
3684 	}
3685 
3686 	if (entry->val > start && rb_prev(&entry->rb_node))
3687 		entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3688 				 rb_node);
3689 
3690 	rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3691 		u64 entry_start;
3692 		u64 entry_end;
3693 		u64 entry_len;
3694 		int clear_ret;
3695 
3696 		entry = rb_entry(node, struct ulist_node, rb_node);
3697 		entry_start = entry->val;
3698 		entry_end = entry->aux;
3699 		entry_len = entry_end - entry_start + 1;
3700 
3701 		if (entry_start >= start + len)
3702 			break;
3703 		if (entry_start + entry_len <= start)
3704 			continue;
3705 		/*
3706 		 * Now the entry is in [start, start + len), revert the
3707 		 * EXTENT_QGROUP_RESERVED bit.
3708 		 */
3709 		clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3710 					      entry_end, EXTENT_QGROUP_RESERVED);
3711 		if (!ret && clear_ret < 0)
3712 			ret = clear_ret;
3713 
3714 		ulist_del(&reserved->range_changed, entry->val, entry->aux);
3715 		if (likely(reserved->bytes_changed >= entry_len)) {
3716 			reserved->bytes_changed -= entry_len;
3717 		} else {
3718 			WARN_ON(1);
3719 			reserved->bytes_changed = 0;
3720 		}
3721 	}
3722 
3723 	return ret;
3724 }
3725 
3726 /*
3727  * Try to free some space for qgroup.
3728  *
3729  * For qgroup, there are only 3 ways to free qgroup space:
3730  * - Flush nodatacow write
3731  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3732  *   In theory, we should only flush nodatacow inodes, but it's not yet
3733  *   possible, so we need to flush the whole root.
3734  *
3735  * - Wait for ordered extents
3736  *   When ordered extents are finished, their reserved metadata is finally
3737  *   converted to per_trans status, which can be freed by later commit
3738  *   transaction.
3739  *
3740  * - Commit transaction
3741  *   This would free the meta_per_trans space.
3742  *   In theory this shouldn't provide much space, but any more qgroup space
3743  *   is needed.
3744  */
try_flush_qgroup(struct btrfs_root * root)3745 static int try_flush_qgroup(struct btrfs_root *root)
3746 {
3747 	struct btrfs_trans_handle *trans;
3748 	int ret;
3749 
3750 	/* Can't hold an open transaction or we run the risk of deadlocking. */
3751 	ASSERT(current->journal_info == NULL);
3752 	if (WARN_ON(current->journal_info))
3753 		return 0;
3754 
3755 	/*
3756 	 * We don't want to run flush again and again, so if there is a running
3757 	 * one, we won't try to start a new flush, but exit directly.
3758 	 */
3759 	if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3760 		wait_event(root->qgroup_flush_wait,
3761 			!test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3762 		return 0;
3763 	}
3764 
3765 	btrfs_run_delayed_iputs(root->fs_info);
3766 	btrfs_wait_on_delayed_iputs(root->fs_info);
3767 	ret = btrfs_start_delalloc_snapshot(root, true);
3768 	if (ret < 0)
3769 		goto out;
3770 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3771 
3772 	trans = btrfs_attach_transaction_barrier(root);
3773 	if (IS_ERR(trans)) {
3774 		ret = PTR_ERR(trans);
3775 		if (ret == -ENOENT)
3776 			ret = 0;
3777 		goto out;
3778 	}
3779 
3780 	ret = btrfs_commit_transaction(trans);
3781 out:
3782 	clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3783 	wake_up(&root->qgroup_flush_wait);
3784 	return ret;
3785 }
3786 
qgroup_reserve_data(struct btrfs_inode * inode,struct extent_changeset ** reserved_ret,u64 start,u64 len)3787 static int qgroup_reserve_data(struct btrfs_inode *inode,
3788 			struct extent_changeset **reserved_ret, u64 start,
3789 			u64 len)
3790 {
3791 	struct btrfs_root *root = inode->root;
3792 	struct extent_changeset *reserved;
3793 	bool new_reserved = false;
3794 	u64 orig_reserved;
3795 	u64 to_reserve;
3796 	int ret;
3797 
3798 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3799 	    !is_fstree(root->root_key.objectid) || len == 0)
3800 		return 0;
3801 
3802 	/* @reserved parameter is mandatory for qgroup */
3803 	if (WARN_ON(!reserved_ret))
3804 		return -EINVAL;
3805 	if (!*reserved_ret) {
3806 		new_reserved = true;
3807 		*reserved_ret = extent_changeset_alloc();
3808 		if (!*reserved_ret)
3809 			return -ENOMEM;
3810 	}
3811 	reserved = *reserved_ret;
3812 	/* Record already reserved space */
3813 	orig_reserved = reserved->bytes_changed;
3814 	ret = set_record_extent_bits(&inode->io_tree, start,
3815 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3816 
3817 	/* Newly reserved space */
3818 	to_reserve = reserved->bytes_changed - orig_reserved;
3819 	trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3820 					to_reserve, QGROUP_RESERVE);
3821 	if (ret < 0)
3822 		goto out;
3823 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3824 	if (ret < 0)
3825 		goto cleanup;
3826 
3827 	return ret;
3828 
3829 cleanup:
3830 	qgroup_unreserve_range(inode, reserved, start, len);
3831 out:
3832 	if (new_reserved) {
3833 		extent_changeset_free(reserved);
3834 		*reserved_ret = NULL;
3835 	}
3836 	return ret;
3837 }
3838 
3839 /*
3840  * Reserve qgroup space for range [start, start + len).
3841  *
3842  * This function will either reserve space from related qgroups or do nothing
3843  * if the range is already reserved.
3844  *
3845  * Return 0 for successful reservation
3846  * Return <0 for error (including -EQUOT)
3847  *
3848  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3849  *	 commit transaction. So caller should not hold any dirty page locked.
3850  */
btrfs_qgroup_reserve_data(struct btrfs_inode * inode,struct extent_changeset ** reserved_ret,u64 start,u64 len)3851 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3852 			struct extent_changeset **reserved_ret, u64 start,
3853 			u64 len)
3854 {
3855 	int ret;
3856 
3857 	ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3858 	if (ret <= 0 && ret != -EDQUOT)
3859 		return ret;
3860 
3861 	ret = try_flush_qgroup(inode->root);
3862 	if (ret < 0)
3863 		return ret;
3864 	return qgroup_reserve_data(inode, reserved_ret, start, len);
3865 }
3866 
3867 /* Free ranges specified by @reserved, normally in error path */
qgroup_free_reserved_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,u64 * freed_ret)3868 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3869 				     struct extent_changeset *reserved,
3870 				     u64 start, u64 len, u64 *freed_ret)
3871 {
3872 	struct btrfs_root *root = inode->root;
3873 	struct ulist_node *unode;
3874 	struct ulist_iterator uiter;
3875 	struct extent_changeset changeset;
3876 	u64 freed = 0;
3877 	int ret;
3878 
3879 	extent_changeset_init(&changeset);
3880 	len = round_up(start + len, root->fs_info->sectorsize);
3881 	start = round_down(start, root->fs_info->sectorsize);
3882 
3883 	ULIST_ITER_INIT(&uiter);
3884 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3885 		u64 range_start = unode->val;
3886 		/* unode->aux is the inclusive end */
3887 		u64 range_len = unode->aux - range_start + 1;
3888 		u64 free_start;
3889 		u64 free_len;
3890 
3891 		extent_changeset_release(&changeset);
3892 
3893 		/* Only free range in range [start, start + len) */
3894 		if (range_start >= start + len ||
3895 		    range_start + range_len <= start)
3896 			continue;
3897 		free_start = max(range_start, start);
3898 		free_len = min(start + len, range_start + range_len) -
3899 			   free_start;
3900 		/*
3901 		 * TODO: To also modify reserved->ranges_reserved to reflect
3902 		 * the modification.
3903 		 *
3904 		 * However as long as we free qgroup reserved according to
3905 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3906 		 * So not need to rush.
3907 		 */
3908 		ret = clear_record_extent_bits(&inode->io_tree, free_start,
3909 				free_start + free_len - 1,
3910 				EXTENT_QGROUP_RESERVED, &changeset);
3911 		if (ret < 0)
3912 			goto out;
3913 		freed += changeset.bytes_changed;
3914 	}
3915 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3916 				  BTRFS_QGROUP_RSV_DATA);
3917 	if (freed_ret)
3918 		*freed_ret = freed;
3919 	ret = 0;
3920 out:
3921 	extent_changeset_release(&changeset);
3922 	return ret;
3923 }
3924 
__btrfs_qgroup_release_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,u64 * released,int free)3925 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3926 			struct extent_changeset *reserved, u64 start, u64 len,
3927 			u64 *released, int free)
3928 {
3929 	struct extent_changeset changeset;
3930 	int trace_op = QGROUP_RELEASE;
3931 	int ret;
3932 
3933 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3934 		return 0;
3935 
3936 	/* In release case, we shouldn't have @reserved */
3937 	WARN_ON(!free && reserved);
3938 	if (free && reserved)
3939 		return qgroup_free_reserved_data(inode, reserved, start, len, released);
3940 	extent_changeset_init(&changeset);
3941 	ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3942 				       EXTENT_QGROUP_RESERVED, &changeset);
3943 	if (ret < 0)
3944 		goto out;
3945 
3946 	if (free)
3947 		trace_op = QGROUP_FREE;
3948 	trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3949 					changeset.bytes_changed, trace_op);
3950 	if (free)
3951 		btrfs_qgroup_free_refroot(inode->root->fs_info,
3952 				inode->root->root_key.objectid,
3953 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3954 	if (released)
3955 		*released = changeset.bytes_changed;
3956 out:
3957 	extent_changeset_release(&changeset);
3958 	return ret;
3959 }
3960 
3961 /*
3962  * Free a reserved space range from io_tree and related qgroups
3963  *
3964  * Should be called when a range of pages get invalidated before reaching disk.
3965  * Or for error cleanup case.
3966  * if @reserved is given, only reserved range in [@start, @start + @len) will
3967  * be freed.
3968  *
3969  * For data written to disk, use btrfs_qgroup_release_data().
3970  *
3971  * NOTE: This function may sleep for memory allocation.
3972  */
btrfs_qgroup_free_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,u64 * freed)3973 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3974 			   struct extent_changeset *reserved,
3975 			   u64 start, u64 len, u64 *freed)
3976 {
3977 	return __btrfs_qgroup_release_data(inode, reserved, start, len, freed, 1);
3978 }
3979 
3980 /*
3981  * Release a reserved space range from io_tree only.
3982  *
3983  * Should be called when a range of pages get written to disk and corresponding
3984  * FILE_EXTENT is inserted into corresponding root.
3985  *
3986  * Since new qgroup accounting framework will only update qgroup numbers at
3987  * commit_transaction() time, its reserved space shouldn't be freed from
3988  * related qgroups.
3989  *
3990  * But we should release the range from io_tree, to allow further write to be
3991  * COWed.
3992  *
3993  * NOTE: This function may sleep for memory allocation.
3994  */
btrfs_qgroup_release_data(struct btrfs_inode * inode,u64 start,u64 len,u64 * released)3995 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released)
3996 {
3997 	return __btrfs_qgroup_release_data(inode, NULL, start, len, released, 0);
3998 }
3999 
add_root_meta_rsv(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)4000 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
4001 			      enum btrfs_qgroup_rsv_type type)
4002 {
4003 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
4004 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
4005 		return;
4006 	if (num_bytes == 0)
4007 		return;
4008 
4009 	spin_lock(&root->qgroup_meta_rsv_lock);
4010 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
4011 		root->qgroup_meta_rsv_prealloc += num_bytes;
4012 	else
4013 		root->qgroup_meta_rsv_pertrans += num_bytes;
4014 	spin_unlock(&root->qgroup_meta_rsv_lock);
4015 }
4016 
sub_root_meta_rsv(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)4017 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
4018 			     enum btrfs_qgroup_rsv_type type)
4019 {
4020 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
4021 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
4022 		return 0;
4023 	if (num_bytes == 0)
4024 		return 0;
4025 
4026 	spin_lock(&root->qgroup_meta_rsv_lock);
4027 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
4028 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
4029 				  num_bytes);
4030 		root->qgroup_meta_rsv_prealloc -= num_bytes;
4031 	} else {
4032 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
4033 				  num_bytes);
4034 		root->qgroup_meta_rsv_pertrans -= num_bytes;
4035 	}
4036 	spin_unlock(&root->qgroup_meta_rsv_lock);
4037 	return num_bytes;
4038 }
4039 
btrfs_qgroup_reserve_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type,bool enforce)4040 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4041 			      enum btrfs_qgroup_rsv_type type, bool enforce)
4042 {
4043 	struct btrfs_fs_info *fs_info = root->fs_info;
4044 	int ret;
4045 
4046 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4047 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
4048 		return 0;
4049 
4050 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4051 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
4052 	ret = qgroup_reserve(root, num_bytes, enforce, type);
4053 	if (ret < 0)
4054 		return ret;
4055 	/*
4056 	 * Record what we have reserved into root.
4057 	 *
4058 	 * To avoid quota disabled->enabled underflow.
4059 	 * In that case, we may try to free space we haven't reserved
4060 	 * (since quota was disabled), so record what we reserved into root.
4061 	 * And ensure later release won't underflow this number.
4062 	 */
4063 	add_root_meta_rsv(root, num_bytes, type);
4064 	return ret;
4065 }
4066 
__btrfs_qgroup_reserve_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type,bool enforce,bool noflush)4067 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4068 				enum btrfs_qgroup_rsv_type type, bool enforce,
4069 				bool noflush)
4070 {
4071 	int ret;
4072 
4073 	ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4074 	if ((ret <= 0 && ret != -EDQUOT) || noflush)
4075 		return ret;
4076 
4077 	ret = try_flush_qgroup(root);
4078 	if (ret < 0)
4079 		return ret;
4080 	return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4081 }
4082 
btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root * root)4083 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
4084 {
4085 	struct btrfs_fs_info *fs_info = root->fs_info;
4086 
4087 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4088 	    !is_fstree(root->root_key.objectid))
4089 		return;
4090 
4091 	/* TODO: Update trace point to handle such free */
4092 	trace_qgroup_meta_free_all_pertrans(root);
4093 	/* Special value -1 means to free all reserved space */
4094 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
4095 				  BTRFS_QGROUP_RSV_META_PERTRANS);
4096 }
4097 
__btrfs_qgroup_free_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)4098 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
4099 			      enum btrfs_qgroup_rsv_type type)
4100 {
4101 	struct btrfs_fs_info *fs_info = root->fs_info;
4102 
4103 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4104 	    !is_fstree(root->root_key.objectid))
4105 		return;
4106 
4107 	/*
4108 	 * reservation for META_PREALLOC can happen before quota is enabled,
4109 	 * which can lead to underflow.
4110 	 * Here ensure we will only free what we really have reserved.
4111 	 */
4112 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
4113 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4114 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
4115 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
4116 				  num_bytes, type);
4117 }
4118 
qgroup_convert_meta(struct btrfs_fs_info * fs_info,u64 ref_root,int num_bytes)4119 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
4120 				int num_bytes)
4121 {
4122 	struct btrfs_qgroup *qgroup;
4123 	LIST_HEAD(qgroup_list);
4124 
4125 	if (num_bytes == 0)
4126 		return;
4127 	if (!fs_info->quota_root)
4128 		return;
4129 
4130 	spin_lock(&fs_info->qgroup_lock);
4131 	qgroup = find_qgroup_rb(fs_info, ref_root);
4132 	if (!qgroup)
4133 		goto out;
4134 
4135 	qgroup_iterator_add(&qgroup_list, qgroup);
4136 	list_for_each_entry(qgroup, &qgroup_list, iterator) {
4137 		struct btrfs_qgroup_list *glist;
4138 
4139 		qgroup_rsv_release(fs_info, qgroup, num_bytes,
4140 				BTRFS_QGROUP_RSV_META_PREALLOC);
4141 		if (!sb_rdonly(fs_info->sb))
4142 			qgroup_rsv_add(fs_info, qgroup, num_bytes,
4143 				       BTRFS_QGROUP_RSV_META_PERTRANS);
4144 
4145 		list_for_each_entry(glist, &qgroup->groups, next_group)
4146 			qgroup_iterator_add(&qgroup_list, glist->group);
4147 	}
4148 out:
4149 	qgroup_iterator_clean(&qgroup_list);
4150 	spin_unlock(&fs_info->qgroup_lock);
4151 }
4152 
btrfs_qgroup_convert_reserved_meta(struct btrfs_root * root,int num_bytes)4153 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4154 {
4155 	struct btrfs_fs_info *fs_info = root->fs_info;
4156 
4157 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4158 	    !is_fstree(root->root_key.objectid))
4159 		return;
4160 	/* Same as btrfs_qgroup_free_meta_prealloc() */
4161 	num_bytes = sub_root_meta_rsv(root, num_bytes,
4162 				      BTRFS_QGROUP_RSV_META_PREALLOC);
4163 	trace_qgroup_meta_convert(root, num_bytes);
4164 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4165 	if (!sb_rdonly(fs_info->sb))
4166 		add_root_meta_rsv(root, num_bytes, BTRFS_QGROUP_RSV_META_PERTRANS);
4167 }
4168 
4169 /*
4170  * Check qgroup reserved space leaking, normally at destroy inode
4171  * time
4172  */
btrfs_qgroup_check_reserved_leak(struct btrfs_inode * inode)4173 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4174 {
4175 	struct extent_changeset changeset;
4176 	struct ulist_node *unode;
4177 	struct ulist_iterator iter;
4178 	int ret;
4179 
4180 	extent_changeset_init(&changeset);
4181 	ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4182 			EXTENT_QGROUP_RESERVED, &changeset);
4183 
4184 	WARN_ON(ret < 0);
4185 	if (WARN_ON(changeset.bytes_changed)) {
4186 		ULIST_ITER_INIT(&iter);
4187 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4188 			btrfs_warn(inode->root->fs_info,
4189 		"leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4190 				btrfs_ino(inode), unode->val, unode->aux);
4191 		}
4192 		btrfs_qgroup_free_refroot(inode->root->fs_info,
4193 				inode->root->root_key.objectid,
4194 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4195 
4196 	}
4197 	extent_changeset_release(&changeset);
4198 }
4199 
btrfs_qgroup_init_swapped_blocks(struct btrfs_qgroup_swapped_blocks * swapped_blocks)4200 void btrfs_qgroup_init_swapped_blocks(
4201 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4202 {
4203 	int i;
4204 
4205 	spin_lock_init(&swapped_blocks->lock);
4206 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4207 		swapped_blocks->blocks[i] = RB_ROOT;
4208 	swapped_blocks->swapped = false;
4209 }
4210 
4211 /*
4212  * Delete all swapped blocks record of @root.
4213  * Every record here means we skipped a full subtree scan for qgroup.
4214  *
4215  * Gets called when committing one transaction.
4216  */
btrfs_qgroup_clean_swapped_blocks(struct btrfs_root * root)4217 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4218 {
4219 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4220 	int i;
4221 
4222 	swapped_blocks = &root->swapped_blocks;
4223 
4224 	spin_lock(&swapped_blocks->lock);
4225 	if (!swapped_blocks->swapped)
4226 		goto out;
4227 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4228 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
4229 		struct btrfs_qgroup_swapped_block *entry;
4230 		struct btrfs_qgroup_swapped_block *next;
4231 
4232 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4233 						     node)
4234 			kfree(entry);
4235 		swapped_blocks->blocks[i] = RB_ROOT;
4236 	}
4237 	swapped_blocks->swapped = false;
4238 out:
4239 	spin_unlock(&swapped_blocks->lock);
4240 }
4241 
4242 /*
4243  * Add subtree roots record into @subvol_root.
4244  *
4245  * @subvol_root:	tree root of the subvolume tree get swapped
4246  * @bg:			block group under balance
4247  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
4248  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
4249  *			BOTH POINTERS ARE BEFORE TREE SWAP
4250  * @last_snapshot:	last snapshot generation of the subvolume tree
4251  */
btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle * trans,struct btrfs_root * subvol_root,struct btrfs_block_group * bg,struct extent_buffer * subvol_parent,int subvol_slot,struct extent_buffer * reloc_parent,int reloc_slot,u64 last_snapshot)4252 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4253 		struct btrfs_root *subvol_root,
4254 		struct btrfs_block_group *bg,
4255 		struct extent_buffer *subvol_parent, int subvol_slot,
4256 		struct extent_buffer *reloc_parent, int reloc_slot,
4257 		u64 last_snapshot)
4258 {
4259 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4260 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4261 	struct btrfs_qgroup_swapped_block *block;
4262 	struct rb_node **cur;
4263 	struct rb_node *parent = NULL;
4264 	int level = btrfs_header_level(subvol_parent) - 1;
4265 	int ret = 0;
4266 
4267 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4268 		return 0;
4269 
4270 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4271 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4272 		btrfs_err_rl(fs_info,
4273 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4274 			__func__,
4275 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4276 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4277 		return -EUCLEAN;
4278 	}
4279 
4280 	block = kmalloc(sizeof(*block), GFP_NOFS);
4281 	if (!block) {
4282 		ret = -ENOMEM;
4283 		goto out;
4284 	}
4285 
4286 	/*
4287 	 * @reloc_parent/slot is still before swap, while @block is going to
4288 	 * record the bytenr after swap, so we do the swap here.
4289 	 */
4290 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4291 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4292 							     reloc_slot);
4293 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4294 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4295 							    subvol_slot);
4296 	block->last_snapshot = last_snapshot;
4297 	block->level = level;
4298 
4299 	/*
4300 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4301 	 * no one else can modify tree blocks thus we qgroup will not change
4302 	 * no matter the value of trace_leaf.
4303 	 */
4304 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4305 		block->trace_leaf = true;
4306 	else
4307 		block->trace_leaf = false;
4308 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4309 
4310 	/* Insert @block into @blocks */
4311 	spin_lock(&blocks->lock);
4312 	cur = &blocks->blocks[level].rb_node;
4313 	while (*cur) {
4314 		struct btrfs_qgroup_swapped_block *entry;
4315 
4316 		parent = *cur;
4317 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4318 				 node);
4319 
4320 		if (entry->subvol_bytenr < block->subvol_bytenr) {
4321 			cur = &(*cur)->rb_left;
4322 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
4323 			cur = &(*cur)->rb_right;
4324 		} else {
4325 			if (entry->subvol_generation !=
4326 					block->subvol_generation ||
4327 			    entry->reloc_bytenr != block->reloc_bytenr ||
4328 			    entry->reloc_generation !=
4329 					block->reloc_generation) {
4330 				/*
4331 				 * Duplicated but mismatch entry found.
4332 				 * Shouldn't happen.
4333 				 *
4334 				 * Marking qgroup inconsistent should be enough
4335 				 * for end users.
4336 				 */
4337 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4338 				ret = -EEXIST;
4339 			}
4340 			kfree(block);
4341 			goto out_unlock;
4342 		}
4343 	}
4344 	rb_link_node(&block->node, parent, cur);
4345 	rb_insert_color(&block->node, &blocks->blocks[level]);
4346 	blocks->swapped = true;
4347 out_unlock:
4348 	spin_unlock(&blocks->lock);
4349 out:
4350 	if (ret < 0)
4351 		qgroup_mark_inconsistent(fs_info);
4352 	return ret;
4353 }
4354 
4355 /*
4356  * Check if the tree block is a subtree root, and if so do the needed
4357  * delayed subtree trace for qgroup.
4358  *
4359  * This is called during btrfs_cow_block().
4360  */
btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * subvol_eb)4361 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4362 					 struct btrfs_root *root,
4363 					 struct extent_buffer *subvol_eb)
4364 {
4365 	struct btrfs_fs_info *fs_info = root->fs_info;
4366 	struct btrfs_tree_parent_check check = { 0 };
4367 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4368 	struct btrfs_qgroup_swapped_block *block;
4369 	struct extent_buffer *reloc_eb = NULL;
4370 	struct rb_node *node;
4371 	bool found = false;
4372 	bool swapped = false;
4373 	int level = btrfs_header_level(subvol_eb);
4374 	int ret = 0;
4375 	int i;
4376 
4377 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4378 		return 0;
4379 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4380 		return 0;
4381 
4382 	spin_lock(&blocks->lock);
4383 	if (!blocks->swapped) {
4384 		spin_unlock(&blocks->lock);
4385 		return 0;
4386 	}
4387 	node = blocks->blocks[level].rb_node;
4388 
4389 	while (node) {
4390 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4391 		if (block->subvol_bytenr < subvol_eb->start) {
4392 			node = node->rb_left;
4393 		} else if (block->subvol_bytenr > subvol_eb->start) {
4394 			node = node->rb_right;
4395 		} else {
4396 			found = true;
4397 			break;
4398 		}
4399 	}
4400 	if (!found) {
4401 		spin_unlock(&blocks->lock);
4402 		goto out;
4403 	}
4404 	/* Found one, remove it from @blocks first and update blocks->swapped */
4405 	rb_erase(&block->node, &blocks->blocks[level]);
4406 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4407 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4408 			swapped = true;
4409 			break;
4410 		}
4411 	}
4412 	blocks->swapped = swapped;
4413 	spin_unlock(&blocks->lock);
4414 
4415 	check.level = block->level;
4416 	check.transid = block->reloc_generation;
4417 	check.has_first_key = true;
4418 	memcpy(&check.first_key, &block->first_key, sizeof(check.first_key));
4419 
4420 	/* Read out reloc subtree root */
4421 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, &check);
4422 	if (IS_ERR(reloc_eb)) {
4423 		ret = PTR_ERR(reloc_eb);
4424 		reloc_eb = NULL;
4425 		goto free_out;
4426 	}
4427 	if (!extent_buffer_uptodate(reloc_eb)) {
4428 		ret = -EIO;
4429 		goto free_out;
4430 	}
4431 
4432 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4433 			block->last_snapshot, block->trace_leaf);
4434 free_out:
4435 	kfree(block);
4436 	free_extent_buffer(reloc_eb);
4437 out:
4438 	if (ret < 0) {
4439 		btrfs_err_rl(fs_info,
4440 			     "failed to account subtree at bytenr %llu: %d",
4441 			     subvol_eb->start, ret);
4442 		qgroup_mark_inconsistent(fs_info);
4443 	}
4444 	return ret;
4445 }
4446 
btrfs_qgroup_destroy_extent_records(struct btrfs_transaction * trans)4447 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4448 {
4449 	struct btrfs_qgroup_extent_record *entry;
4450 	struct btrfs_qgroup_extent_record *next;
4451 	struct rb_root *root;
4452 
4453 	root = &trans->delayed_refs.dirty_extent_root;
4454 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4455 		ulist_free(entry->old_roots);
4456 		kfree(entry);
4457 	}
4458 	*root = RB_ROOT;
4459 }
4460