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 
27 /* TODO XXX FIXME
28  *  - subvol delete -> delete when ref goes to 0? delete limits also?
29  *  - reorganize keys
30  *  - compressed
31  *  - sync
32  *  - copy also limits on subvol creation
33  *  - limit
34  *  - caches for ulists
35  *  - performance benchmarks
36  *  - check all ioctl parameters
37  */
38 
39 /*
40  * Helpers to access qgroup reservation
41  *
42  * Callers should ensure the lock context and type are valid
43  */
44 
qgroup_rsv_total(const struct btrfs_qgroup * qgroup)45 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
46 {
47 	u64 ret = 0;
48 	int i;
49 
50 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
51 		ret += qgroup->rsv.values[i];
52 
53 	return ret;
54 }
55 
56 #ifdef CONFIG_BTRFS_DEBUG
qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)57 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
58 {
59 	if (type == BTRFS_QGROUP_RSV_DATA)
60 		return "data";
61 	if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
62 		return "meta_pertrans";
63 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
64 		return "meta_prealloc";
65 	return NULL;
66 }
67 #endif
68 
qgroup_rsv_add(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup,u64 num_bytes,enum btrfs_qgroup_rsv_type type)69 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
70 			   struct btrfs_qgroup *qgroup, u64 num_bytes,
71 			   enum btrfs_qgroup_rsv_type type)
72 {
73 	trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
74 	qgroup->rsv.values[type] += num_bytes;
75 }
76 
qgroup_rsv_release(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup,u64 num_bytes,enum btrfs_qgroup_rsv_type type)77 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
78 			       struct btrfs_qgroup *qgroup, u64 num_bytes,
79 			       enum btrfs_qgroup_rsv_type type)
80 {
81 	trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
82 	if (qgroup->rsv.values[type] >= num_bytes) {
83 		qgroup->rsv.values[type] -= num_bytes;
84 		return;
85 	}
86 #ifdef CONFIG_BTRFS_DEBUG
87 	WARN_RATELIMIT(1,
88 		"qgroup %llu %s reserved space underflow, have %llu to free %llu",
89 		qgroup->qgroupid, qgroup_rsv_type_str(type),
90 		qgroup->rsv.values[type], num_bytes);
91 #endif
92 	qgroup->rsv.values[type] = 0;
93 }
94 
qgroup_rsv_add_by_qgroup(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * dest,struct btrfs_qgroup * src)95 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
96 				     struct btrfs_qgroup *dest,
97 				     struct btrfs_qgroup *src)
98 {
99 	int i;
100 
101 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
102 		qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
103 }
104 
qgroup_rsv_release_by_qgroup(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * dest,struct btrfs_qgroup * src)105 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
106 					 struct btrfs_qgroup *dest,
107 					  struct btrfs_qgroup *src)
108 {
109 	int i;
110 
111 	for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
112 		qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
113 }
114 
btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup * qg,u64 seq,int mod)115 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
116 					   int mod)
117 {
118 	if (qg->old_refcnt < seq)
119 		qg->old_refcnt = seq;
120 	qg->old_refcnt += mod;
121 }
122 
btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup * qg,u64 seq,int mod)123 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
124 					   int mod)
125 {
126 	if (qg->new_refcnt < seq)
127 		qg->new_refcnt = seq;
128 	qg->new_refcnt += mod;
129 }
130 
btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup * qg,u64 seq)131 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
132 {
133 	if (qg->old_refcnt < seq)
134 		return 0;
135 	return qg->old_refcnt - seq;
136 }
137 
btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup * qg,u64 seq)138 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
139 {
140 	if (qg->new_refcnt < seq)
141 		return 0;
142 	return qg->new_refcnt - seq;
143 }
144 
145 /*
146  * glue structure to represent the relations between qgroups.
147  */
148 struct btrfs_qgroup_list {
149 	struct list_head next_group;
150 	struct list_head next_member;
151 	struct btrfs_qgroup *group;
152 	struct btrfs_qgroup *member;
153 };
154 
qgroup_to_aux(struct btrfs_qgroup * qg)155 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
156 {
157 	return (u64)(uintptr_t)qg;
158 }
159 
unode_aux_to_qgroup(struct ulist_node * n)160 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
161 {
162 	return (struct btrfs_qgroup *)(uintptr_t)n->aux;
163 }
164 
165 static int
166 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
167 		   int init_flags);
168 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
169 
170 /* must be called with qgroup_ioctl_lock held */
find_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)171 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
172 					   u64 qgroupid)
173 {
174 	struct rb_node *n = fs_info->qgroup_tree.rb_node;
175 	struct btrfs_qgroup *qgroup;
176 
177 	while (n) {
178 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
179 		if (qgroup->qgroupid < qgroupid)
180 			n = n->rb_left;
181 		else if (qgroup->qgroupid > qgroupid)
182 			n = n->rb_right;
183 		else
184 			return qgroup;
185 	}
186 	return NULL;
187 }
188 
189 /* must be called with qgroup_lock held */
add_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)190 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
191 					  u64 qgroupid)
192 {
193 	struct rb_node **p = &fs_info->qgroup_tree.rb_node;
194 	struct rb_node *parent = NULL;
195 	struct btrfs_qgroup *qgroup;
196 
197 	while (*p) {
198 		parent = *p;
199 		qgroup = rb_entry(parent, struct btrfs_qgroup, node);
200 
201 		if (qgroup->qgroupid < qgroupid)
202 			p = &(*p)->rb_left;
203 		else if (qgroup->qgroupid > qgroupid)
204 			p = &(*p)->rb_right;
205 		else
206 			return qgroup;
207 	}
208 
209 	qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
210 	if (!qgroup)
211 		return ERR_PTR(-ENOMEM);
212 
213 	qgroup->qgroupid = qgroupid;
214 	INIT_LIST_HEAD(&qgroup->groups);
215 	INIT_LIST_HEAD(&qgroup->members);
216 	INIT_LIST_HEAD(&qgroup->dirty);
217 
218 	rb_link_node(&qgroup->node, parent, p);
219 	rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
220 
221 	return qgroup;
222 }
223 
__del_qgroup_rb(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup)224 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
225 			    struct btrfs_qgroup *qgroup)
226 {
227 	struct btrfs_qgroup_list *list;
228 
229 	btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
230 	list_del(&qgroup->dirty);
231 	while (!list_empty(&qgroup->groups)) {
232 		list = list_first_entry(&qgroup->groups,
233 					struct btrfs_qgroup_list, next_group);
234 		list_del(&list->next_group);
235 		list_del(&list->next_member);
236 		kfree(list);
237 	}
238 
239 	while (!list_empty(&qgroup->members)) {
240 		list = list_first_entry(&qgroup->members,
241 					struct btrfs_qgroup_list, next_member);
242 		list_del(&list->next_group);
243 		list_del(&list->next_member);
244 		kfree(list);
245 	}
246 	kfree(qgroup);
247 }
248 
249 /* must be called with qgroup_lock held */
del_qgroup_rb(struct btrfs_fs_info * fs_info,u64 qgroupid)250 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
251 {
252 	struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
253 
254 	if (!qgroup)
255 		return -ENOENT;
256 
257 	rb_erase(&qgroup->node, &fs_info->qgroup_tree);
258 	__del_qgroup_rb(fs_info, qgroup);
259 	return 0;
260 }
261 
262 /* must be called with qgroup_lock held */
add_relation_rb(struct btrfs_fs_info * fs_info,u64 memberid,u64 parentid)263 static int add_relation_rb(struct btrfs_fs_info *fs_info,
264 			   u64 memberid, u64 parentid)
265 {
266 	struct btrfs_qgroup *member;
267 	struct btrfs_qgroup *parent;
268 	struct btrfs_qgroup_list *list;
269 
270 	member = find_qgroup_rb(fs_info, memberid);
271 	parent = find_qgroup_rb(fs_info, parentid);
272 	if (!member || !parent)
273 		return -ENOENT;
274 
275 	list = kzalloc(sizeof(*list), GFP_ATOMIC);
276 	if (!list)
277 		return -ENOMEM;
278 
279 	list->group = parent;
280 	list->member = member;
281 	list_add_tail(&list->next_group, &member->groups);
282 	list_add_tail(&list->next_member, &parent->members);
283 
284 	return 0;
285 }
286 
287 /* must be called with qgroup_lock held */
del_relation_rb(struct btrfs_fs_info * fs_info,u64 memberid,u64 parentid)288 static int del_relation_rb(struct btrfs_fs_info *fs_info,
289 			   u64 memberid, u64 parentid)
290 {
291 	struct btrfs_qgroup *member;
292 	struct btrfs_qgroup *parent;
293 	struct btrfs_qgroup_list *list;
294 
295 	member = find_qgroup_rb(fs_info, memberid);
296 	parent = find_qgroup_rb(fs_info, parentid);
297 	if (!member || !parent)
298 		return -ENOENT;
299 
300 	list_for_each_entry(list, &member->groups, next_group) {
301 		if (list->group == parent) {
302 			list_del(&list->next_group);
303 			list_del(&list->next_member);
304 			kfree(list);
305 			return 0;
306 		}
307 	}
308 	return -ENOENT;
309 }
310 
311 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_verify_qgroup_counts(struct btrfs_fs_info * fs_info,u64 qgroupid,u64 rfer,u64 excl)312 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
313 			       u64 rfer, u64 excl)
314 {
315 	struct btrfs_qgroup *qgroup;
316 
317 	qgroup = find_qgroup_rb(fs_info, qgroupid);
318 	if (!qgroup)
319 		return -EINVAL;
320 	if (qgroup->rfer != rfer || qgroup->excl != excl)
321 		return -EINVAL;
322 	return 0;
323 }
324 #endif
325 
326 /*
327  * The full config is read in one go, only called from open_ctree()
328  * It doesn't use any locking, as at this point we're still single-threaded
329  */
btrfs_read_qgroup_config(struct btrfs_fs_info * fs_info)330 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
331 {
332 	struct btrfs_key key;
333 	struct btrfs_key found_key;
334 	struct btrfs_root *quota_root = fs_info->quota_root;
335 	struct btrfs_path *path = NULL;
336 	struct extent_buffer *l;
337 	int slot;
338 	int ret = 0;
339 	u64 flags = 0;
340 	u64 rescan_progress = 0;
341 
342 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
343 		return 0;
344 
345 	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
346 	if (!fs_info->qgroup_ulist) {
347 		ret = -ENOMEM;
348 		goto out;
349 	}
350 
351 	path = btrfs_alloc_path();
352 	if (!path) {
353 		ret = -ENOMEM;
354 		goto out;
355 	}
356 
357 	ret = btrfs_sysfs_add_qgroups(fs_info);
358 	if (ret < 0)
359 		goto out;
360 	/* default this to quota off, in case no status key is found */
361 	fs_info->qgroup_flags = 0;
362 
363 	/*
364 	 * pass 1: read status, all qgroup infos and limits
365 	 */
366 	key.objectid = 0;
367 	key.type = 0;
368 	key.offset = 0;
369 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
370 	if (ret)
371 		goto out;
372 
373 	while (1) {
374 		struct btrfs_qgroup *qgroup;
375 
376 		slot = path->slots[0];
377 		l = path->nodes[0];
378 		btrfs_item_key_to_cpu(l, &found_key, slot);
379 
380 		if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
381 			struct btrfs_qgroup_status_item *ptr;
382 
383 			ptr = btrfs_item_ptr(l, slot,
384 					     struct btrfs_qgroup_status_item);
385 
386 			if (btrfs_qgroup_status_version(l, ptr) !=
387 			    BTRFS_QGROUP_STATUS_VERSION) {
388 				btrfs_err(fs_info,
389 				 "old qgroup version, quota disabled");
390 				goto out;
391 			}
392 			if (btrfs_qgroup_status_generation(l, ptr) !=
393 			    fs_info->generation) {
394 				flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
395 				btrfs_err(fs_info,
396 					"qgroup generation mismatch, marked as inconsistent");
397 			}
398 			fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
399 									  ptr);
400 			rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
401 			goto next1;
402 		}
403 
404 		if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
405 		    found_key.type != BTRFS_QGROUP_LIMIT_KEY)
406 			goto next1;
407 
408 		qgroup = find_qgroup_rb(fs_info, found_key.offset);
409 		if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
410 		    (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
411 			btrfs_err(fs_info, "inconsistent qgroup config");
412 			flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
413 		}
414 		if (!qgroup) {
415 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
416 			if (IS_ERR(qgroup)) {
417 				ret = PTR_ERR(qgroup);
418 				goto out;
419 			}
420 		}
421 		ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
422 		if (ret < 0)
423 			goto out;
424 
425 		switch (found_key.type) {
426 		case BTRFS_QGROUP_INFO_KEY: {
427 			struct btrfs_qgroup_info_item *ptr;
428 
429 			ptr = btrfs_item_ptr(l, slot,
430 					     struct btrfs_qgroup_info_item);
431 			qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
432 			qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
433 			qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
434 			qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
435 			/* generation currently unused */
436 			break;
437 		}
438 		case BTRFS_QGROUP_LIMIT_KEY: {
439 			struct btrfs_qgroup_limit_item *ptr;
440 
441 			ptr = btrfs_item_ptr(l, slot,
442 					     struct btrfs_qgroup_limit_item);
443 			qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
444 			qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
445 			qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
446 			qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
447 			qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
448 			break;
449 		}
450 		}
451 next1:
452 		ret = btrfs_next_item(quota_root, path);
453 		if (ret < 0)
454 			goto out;
455 		if (ret)
456 			break;
457 	}
458 	btrfs_release_path(path);
459 
460 	/*
461 	 * pass 2: read all qgroup relations
462 	 */
463 	key.objectid = 0;
464 	key.type = BTRFS_QGROUP_RELATION_KEY;
465 	key.offset = 0;
466 	ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
467 	if (ret)
468 		goto out;
469 	while (1) {
470 		slot = path->slots[0];
471 		l = path->nodes[0];
472 		btrfs_item_key_to_cpu(l, &found_key, slot);
473 
474 		if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
475 			goto next2;
476 
477 		if (found_key.objectid > found_key.offset) {
478 			/* parent <- member, not needed to build config */
479 			/* FIXME should we omit the key completely? */
480 			goto next2;
481 		}
482 
483 		ret = add_relation_rb(fs_info, found_key.objectid,
484 				      found_key.offset);
485 		if (ret == -ENOENT) {
486 			btrfs_warn(fs_info,
487 				"orphan qgroup relation 0x%llx->0x%llx",
488 				found_key.objectid, found_key.offset);
489 			ret = 0;	/* ignore the error */
490 		}
491 		if (ret)
492 			goto out;
493 next2:
494 		ret = btrfs_next_item(quota_root, path);
495 		if (ret < 0)
496 			goto out;
497 		if (ret)
498 			break;
499 	}
500 out:
501 	btrfs_free_path(path);
502 	fs_info->qgroup_flags |= flags;
503 	if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
504 		clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
505 	else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
506 		 ret >= 0)
507 		ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
508 
509 	if (ret < 0) {
510 		ulist_free(fs_info->qgroup_ulist);
511 		fs_info->qgroup_ulist = NULL;
512 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
513 		btrfs_sysfs_del_qgroups(fs_info);
514 	}
515 
516 	return ret < 0 ? ret : 0;
517 }
518 
519 /*
520  * Called in close_ctree() when quota is still enabled.  This verifies we don't
521  * leak some reserved space.
522  *
523  * Return false if no reserved space is left.
524  * Return true if some reserved space is leaked.
525  */
btrfs_check_quota_leak(struct btrfs_fs_info * fs_info)526 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
527 {
528 	struct rb_node *node;
529 	bool ret = false;
530 
531 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
532 		return ret;
533 	/*
534 	 * Since we're unmounting, there is no race and no need to grab qgroup
535 	 * lock.  And here we don't go post-order to provide a more user
536 	 * friendly sorted result.
537 	 */
538 	for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
539 		struct btrfs_qgroup *qgroup;
540 		int i;
541 
542 		qgroup = rb_entry(node, struct btrfs_qgroup, node);
543 		for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
544 			if (qgroup->rsv.values[i]) {
545 				ret = true;
546 				btrfs_warn(fs_info,
547 		"qgroup %hu/%llu has unreleased space, type %d rsv %llu",
548 				   btrfs_qgroup_level(qgroup->qgroupid),
549 				   btrfs_qgroup_subvolid(qgroup->qgroupid),
550 				   i, qgroup->rsv.values[i]);
551 			}
552 		}
553 	}
554 	return ret;
555 }
556 
557 /*
558  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
559  * first two are in single-threaded paths.And for the third one, we have set
560  * quota_root to be null with qgroup_lock held before, so it is safe to clean
561  * up the in-memory structures without qgroup_lock held.
562  */
btrfs_free_qgroup_config(struct btrfs_fs_info * fs_info)563 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
564 {
565 	struct rb_node *n;
566 	struct btrfs_qgroup *qgroup;
567 
568 	while ((n = rb_first(&fs_info->qgroup_tree))) {
569 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
570 		rb_erase(n, &fs_info->qgroup_tree);
571 		__del_qgroup_rb(fs_info, qgroup);
572 	}
573 	/*
574 	 * We call btrfs_free_qgroup_config() when unmounting
575 	 * filesystem and disabling quota, so we set qgroup_ulist
576 	 * to be null here to avoid double free.
577 	 */
578 	ulist_free(fs_info->qgroup_ulist);
579 	fs_info->qgroup_ulist = NULL;
580 	btrfs_sysfs_del_qgroups(fs_info);
581 }
582 
add_qgroup_relation_item(struct btrfs_trans_handle * trans,u64 src,u64 dst)583 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
584 				    u64 dst)
585 {
586 	int ret;
587 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
588 	struct btrfs_path *path;
589 	struct btrfs_key key;
590 
591 	path = btrfs_alloc_path();
592 	if (!path)
593 		return -ENOMEM;
594 
595 	key.objectid = src;
596 	key.type = BTRFS_QGROUP_RELATION_KEY;
597 	key.offset = dst;
598 
599 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
600 
601 	btrfs_mark_buffer_dirty(path->nodes[0]);
602 
603 	btrfs_free_path(path);
604 	return ret;
605 }
606 
del_qgroup_relation_item(struct btrfs_trans_handle * trans,u64 src,u64 dst)607 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
608 				    u64 dst)
609 {
610 	int ret;
611 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
612 	struct btrfs_path *path;
613 	struct btrfs_key key;
614 
615 	path = btrfs_alloc_path();
616 	if (!path)
617 		return -ENOMEM;
618 
619 	key.objectid = src;
620 	key.type = BTRFS_QGROUP_RELATION_KEY;
621 	key.offset = dst;
622 
623 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
624 	if (ret < 0)
625 		goto out;
626 
627 	if (ret > 0) {
628 		ret = -ENOENT;
629 		goto out;
630 	}
631 
632 	ret = btrfs_del_item(trans, quota_root, path);
633 out:
634 	btrfs_free_path(path);
635 	return ret;
636 }
637 
add_qgroup_item(struct btrfs_trans_handle * trans,struct btrfs_root * quota_root,u64 qgroupid)638 static int add_qgroup_item(struct btrfs_trans_handle *trans,
639 			   struct btrfs_root *quota_root, u64 qgroupid)
640 {
641 	int ret;
642 	struct btrfs_path *path;
643 	struct btrfs_qgroup_info_item *qgroup_info;
644 	struct btrfs_qgroup_limit_item *qgroup_limit;
645 	struct extent_buffer *leaf;
646 	struct btrfs_key key;
647 
648 	if (btrfs_is_testing(quota_root->fs_info))
649 		return 0;
650 
651 	path = btrfs_alloc_path();
652 	if (!path)
653 		return -ENOMEM;
654 
655 	key.objectid = 0;
656 	key.type = BTRFS_QGROUP_INFO_KEY;
657 	key.offset = qgroupid;
658 
659 	/*
660 	 * Avoid a transaction abort by catching -EEXIST here. In that
661 	 * case, we proceed by re-initializing the existing structure
662 	 * on disk.
663 	 */
664 
665 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
666 				      sizeof(*qgroup_info));
667 	if (ret && ret != -EEXIST)
668 		goto out;
669 
670 	leaf = path->nodes[0];
671 	qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
672 				 struct btrfs_qgroup_info_item);
673 	btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
674 	btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
675 	btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
676 	btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
677 	btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
678 
679 	btrfs_mark_buffer_dirty(leaf);
680 
681 	btrfs_release_path(path);
682 
683 	key.type = BTRFS_QGROUP_LIMIT_KEY;
684 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
685 				      sizeof(*qgroup_limit));
686 	if (ret && ret != -EEXIST)
687 		goto out;
688 
689 	leaf = path->nodes[0];
690 	qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
691 				  struct btrfs_qgroup_limit_item);
692 	btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
693 	btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
694 	btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
695 	btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
696 	btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
697 
698 	btrfs_mark_buffer_dirty(leaf);
699 
700 	ret = 0;
701 out:
702 	btrfs_free_path(path);
703 	return ret;
704 }
705 
del_qgroup_item(struct btrfs_trans_handle * trans,u64 qgroupid)706 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
707 {
708 	int ret;
709 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
710 	struct btrfs_path *path;
711 	struct btrfs_key key;
712 
713 	path = btrfs_alloc_path();
714 	if (!path)
715 		return -ENOMEM;
716 
717 	key.objectid = 0;
718 	key.type = BTRFS_QGROUP_INFO_KEY;
719 	key.offset = qgroupid;
720 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
721 	if (ret < 0)
722 		goto out;
723 
724 	if (ret > 0) {
725 		ret = -ENOENT;
726 		goto out;
727 	}
728 
729 	ret = btrfs_del_item(trans, quota_root, path);
730 	if (ret)
731 		goto out;
732 
733 	btrfs_release_path(path);
734 
735 	key.type = BTRFS_QGROUP_LIMIT_KEY;
736 	ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
737 	if (ret < 0)
738 		goto out;
739 
740 	if (ret > 0) {
741 		ret = -ENOENT;
742 		goto out;
743 	}
744 
745 	ret = btrfs_del_item(trans, quota_root, path);
746 
747 out:
748 	btrfs_free_path(path);
749 	return ret;
750 }
751 
update_qgroup_limit_item(struct btrfs_trans_handle * trans,struct btrfs_qgroup * qgroup)752 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
753 				    struct btrfs_qgroup *qgroup)
754 {
755 	struct btrfs_root *quota_root = trans->fs_info->quota_root;
756 	struct btrfs_path *path;
757 	struct btrfs_key key;
758 	struct extent_buffer *l;
759 	struct btrfs_qgroup_limit_item *qgroup_limit;
760 	int ret;
761 	int slot;
762 
763 	key.objectid = 0;
764 	key.type = BTRFS_QGROUP_LIMIT_KEY;
765 	key.offset = qgroup->qgroupid;
766 
767 	path = btrfs_alloc_path();
768 	if (!path)
769 		return -ENOMEM;
770 
771 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
772 	if (ret > 0)
773 		ret = -ENOENT;
774 
775 	if (ret)
776 		goto out;
777 
778 	l = path->nodes[0];
779 	slot = path->slots[0];
780 	qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
781 	btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
782 	btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
783 	btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
784 	btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
785 	btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
786 
787 	btrfs_mark_buffer_dirty(l);
788 
789 out:
790 	btrfs_free_path(path);
791 	return ret;
792 }
793 
update_qgroup_info_item(struct btrfs_trans_handle * trans,struct btrfs_qgroup * qgroup)794 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
795 				   struct btrfs_qgroup *qgroup)
796 {
797 	struct btrfs_fs_info *fs_info = trans->fs_info;
798 	struct btrfs_root *quota_root = fs_info->quota_root;
799 	struct btrfs_path *path;
800 	struct btrfs_key key;
801 	struct extent_buffer *l;
802 	struct btrfs_qgroup_info_item *qgroup_info;
803 	int ret;
804 	int slot;
805 
806 	if (btrfs_is_testing(fs_info))
807 		return 0;
808 
809 	key.objectid = 0;
810 	key.type = BTRFS_QGROUP_INFO_KEY;
811 	key.offset = qgroup->qgroupid;
812 
813 	path = btrfs_alloc_path();
814 	if (!path)
815 		return -ENOMEM;
816 
817 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
818 	if (ret > 0)
819 		ret = -ENOENT;
820 
821 	if (ret)
822 		goto out;
823 
824 	l = path->nodes[0];
825 	slot = path->slots[0];
826 	qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
827 	btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
828 	btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
829 	btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
830 	btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
831 	btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
832 
833 	btrfs_mark_buffer_dirty(l);
834 
835 out:
836 	btrfs_free_path(path);
837 	return ret;
838 }
839 
update_qgroup_status_item(struct btrfs_trans_handle * trans)840 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
841 {
842 	struct btrfs_fs_info *fs_info = trans->fs_info;
843 	struct btrfs_root *quota_root = fs_info->quota_root;
844 	struct btrfs_path *path;
845 	struct btrfs_key key;
846 	struct extent_buffer *l;
847 	struct btrfs_qgroup_status_item *ptr;
848 	int ret;
849 	int slot;
850 
851 	key.objectid = 0;
852 	key.type = BTRFS_QGROUP_STATUS_KEY;
853 	key.offset = 0;
854 
855 	path = btrfs_alloc_path();
856 	if (!path)
857 		return -ENOMEM;
858 
859 	ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
860 	if (ret > 0)
861 		ret = -ENOENT;
862 
863 	if (ret)
864 		goto out;
865 
866 	l = path->nodes[0];
867 	slot = path->slots[0];
868 	ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
869 	btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
870 	btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
871 	btrfs_set_qgroup_status_rescan(l, ptr,
872 				fs_info->qgroup_rescan_progress.objectid);
873 
874 	btrfs_mark_buffer_dirty(l);
875 
876 out:
877 	btrfs_free_path(path);
878 	return ret;
879 }
880 
881 /*
882  * called with qgroup_lock held
883  */
btrfs_clean_quota_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)884 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
885 				  struct btrfs_root *root)
886 {
887 	struct btrfs_path *path;
888 	struct btrfs_key key;
889 	struct extent_buffer *leaf = NULL;
890 	int ret;
891 	int nr = 0;
892 
893 	path = btrfs_alloc_path();
894 	if (!path)
895 		return -ENOMEM;
896 
897 	path->leave_spinning = 1;
898 
899 	key.objectid = 0;
900 	key.offset = 0;
901 	key.type = 0;
902 
903 	while (1) {
904 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
905 		if (ret < 0)
906 			goto out;
907 		leaf = path->nodes[0];
908 		nr = btrfs_header_nritems(leaf);
909 		if (!nr)
910 			break;
911 		/*
912 		 * delete the leaf one by one
913 		 * since the whole tree is going
914 		 * to be deleted.
915 		 */
916 		path->slots[0] = 0;
917 		ret = btrfs_del_items(trans, root, path, 0, nr);
918 		if (ret)
919 			goto out;
920 
921 		btrfs_release_path(path);
922 	}
923 	ret = 0;
924 out:
925 	btrfs_free_path(path);
926 	return ret;
927 }
928 
btrfs_quota_enable(struct btrfs_fs_info * fs_info)929 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
930 {
931 	struct btrfs_root *quota_root;
932 	struct btrfs_root *tree_root = fs_info->tree_root;
933 	struct btrfs_path *path = NULL;
934 	struct btrfs_qgroup_status_item *ptr;
935 	struct extent_buffer *leaf;
936 	struct btrfs_key key;
937 	struct btrfs_key found_key;
938 	struct btrfs_qgroup *qgroup = NULL;
939 	struct btrfs_trans_handle *trans = NULL;
940 	struct ulist *ulist = NULL;
941 	int ret = 0;
942 	int slot;
943 
944 	mutex_lock(&fs_info->qgroup_ioctl_lock);
945 	if (fs_info->quota_root)
946 		goto out;
947 
948 	ulist = ulist_alloc(GFP_KERNEL);
949 	if (!ulist) {
950 		ret = -ENOMEM;
951 		goto out;
952 	}
953 
954 	ret = btrfs_sysfs_add_qgroups(fs_info);
955 	if (ret < 0)
956 		goto out;
957 
958 	/*
959 	 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
960 	 * avoid lock acquisition inversion problems (reported by lockdep) between
961 	 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
962 	 * start a transaction.
963 	 * After we started the transaction lock qgroup_ioctl_lock again and
964 	 * check if someone else created the quota root in the meanwhile. If so,
965 	 * just return success and release the transaction handle.
966 	 *
967 	 * Also we don't need to worry about someone else calling
968 	 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
969 	 * that function returns 0 (success) when the sysfs entries already exist.
970 	 */
971 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
972 
973 	/*
974 	 * 1 for quota root item
975 	 * 1 for BTRFS_QGROUP_STATUS item
976 	 *
977 	 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
978 	 * per subvolume. However those are not currently reserved since it
979 	 * would be a lot of overkill.
980 	 */
981 	trans = btrfs_start_transaction(tree_root, 2);
982 
983 	mutex_lock(&fs_info->qgroup_ioctl_lock);
984 	if (IS_ERR(trans)) {
985 		ret = PTR_ERR(trans);
986 		trans = NULL;
987 		goto out;
988 	}
989 
990 	if (fs_info->quota_root)
991 		goto out;
992 
993 	fs_info->qgroup_ulist = ulist;
994 	ulist = NULL;
995 
996 	/*
997 	 * initially create the quota tree
998 	 */
999 	quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1000 	if (IS_ERR(quota_root)) {
1001 		ret =  PTR_ERR(quota_root);
1002 		btrfs_abort_transaction(trans, ret);
1003 		goto out;
1004 	}
1005 
1006 	path = btrfs_alloc_path();
1007 	if (!path) {
1008 		ret = -ENOMEM;
1009 		btrfs_abort_transaction(trans, ret);
1010 		goto out_free_root;
1011 	}
1012 
1013 	key.objectid = 0;
1014 	key.type = BTRFS_QGROUP_STATUS_KEY;
1015 	key.offset = 0;
1016 
1017 	ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1018 				      sizeof(*ptr));
1019 	if (ret) {
1020 		btrfs_abort_transaction(trans, ret);
1021 		goto out_free_path;
1022 	}
1023 
1024 	leaf = path->nodes[0];
1025 	ptr = btrfs_item_ptr(leaf, path->slots[0],
1026 				 struct btrfs_qgroup_status_item);
1027 	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1028 	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1029 	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1030 				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1031 	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1032 	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1033 
1034 	btrfs_mark_buffer_dirty(leaf);
1035 
1036 	key.objectid = 0;
1037 	key.type = BTRFS_ROOT_REF_KEY;
1038 	key.offset = 0;
1039 
1040 	btrfs_release_path(path);
1041 	ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1042 	if (ret > 0)
1043 		goto out_add_root;
1044 	if (ret < 0) {
1045 		btrfs_abort_transaction(trans, ret);
1046 		goto out_free_path;
1047 	}
1048 
1049 	while (1) {
1050 		slot = path->slots[0];
1051 		leaf = path->nodes[0];
1052 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1053 
1054 		if (found_key.type == BTRFS_ROOT_REF_KEY) {
1055 
1056 			/* Release locks on tree_root before we access quota_root */
1057 			btrfs_release_path(path);
1058 
1059 			ret = add_qgroup_item(trans, quota_root,
1060 					      found_key.offset);
1061 			if (ret) {
1062 				btrfs_abort_transaction(trans, ret);
1063 				goto out_free_path;
1064 			}
1065 
1066 			qgroup = add_qgroup_rb(fs_info, found_key.offset);
1067 			if (IS_ERR(qgroup)) {
1068 				ret = PTR_ERR(qgroup);
1069 				btrfs_abort_transaction(trans, ret);
1070 				goto out_free_path;
1071 			}
1072 			ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1073 			if (ret < 0) {
1074 				btrfs_abort_transaction(trans, ret);
1075 				goto out_free_path;
1076 			}
1077 			ret = btrfs_search_slot_for_read(tree_root, &found_key,
1078 							 path, 1, 0);
1079 			if (ret < 0) {
1080 				btrfs_abort_transaction(trans, ret);
1081 				goto out_free_path;
1082 			}
1083 			if (ret > 0) {
1084 				/*
1085 				 * Shouldn't happen, but in case it does we
1086 				 * don't need to do the btrfs_next_item, just
1087 				 * continue.
1088 				 */
1089 				continue;
1090 			}
1091 		}
1092 		ret = btrfs_next_item(tree_root, path);
1093 		if (ret < 0) {
1094 			btrfs_abort_transaction(trans, ret);
1095 			goto out_free_path;
1096 		}
1097 		if (ret)
1098 			break;
1099 	}
1100 
1101 out_add_root:
1102 	btrfs_release_path(path);
1103 	ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1104 	if (ret) {
1105 		btrfs_abort_transaction(trans, ret);
1106 		goto out_free_path;
1107 	}
1108 
1109 	qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1110 	if (IS_ERR(qgroup)) {
1111 		ret = PTR_ERR(qgroup);
1112 		btrfs_abort_transaction(trans, ret);
1113 		goto out_free_path;
1114 	}
1115 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1116 	if (ret < 0) {
1117 		btrfs_abort_transaction(trans, ret);
1118 		goto out_free_path;
1119 	}
1120 
1121 	ret = btrfs_commit_transaction(trans);
1122 	trans = NULL;
1123 	if (ret)
1124 		goto out_free_path;
1125 
1126 	/*
1127 	 * Set quota enabled flag after committing the transaction, to avoid
1128 	 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1129 	 * creation.
1130 	 */
1131 	spin_lock(&fs_info->qgroup_lock);
1132 	fs_info->quota_root = quota_root;
1133 	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1134 	spin_unlock(&fs_info->qgroup_lock);
1135 
1136 	ret = qgroup_rescan_init(fs_info, 0, 1);
1137 	if (!ret) {
1138 	        qgroup_rescan_zero_tracking(fs_info);
1139 		fs_info->qgroup_rescan_running = true;
1140 	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
1141 	                         &fs_info->qgroup_rescan_work);
1142 	}
1143 
1144 out_free_path:
1145 	btrfs_free_path(path);
1146 out_free_root:
1147 	if (ret)
1148 		btrfs_put_root(quota_root);
1149 out:
1150 	if (ret) {
1151 		ulist_free(fs_info->qgroup_ulist);
1152 		fs_info->qgroup_ulist = NULL;
1153 		btrfs_sysfs_del_qgroups(fs_info);
1154 	}
1155 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1156 	if (ret && trans)
1157 		btrfs_end_transaction(trans);
1158 	else if (trans)
1159 		ret = btrfs_end_transaction(trans);
1160 	ulist_free(ulist);
1161 	return ret;
1162 }
1163 
btrfs_quota_disable(struct btrfs_fs_info * fs_info)1164 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1165 {
1166 	struct btrfs_root *quota_root;
1167 	struct btrfs_trans_handle *trans = NULL;
1168 	int ret = 0;
1169 
1170 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1171 	if (!fs_info->quota_root)
1172 		goto out;
1173 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1174 
1175 	/*
1176 	 * 1 For the root item
1177 	 *
1178 	 * We should also reserve enough items for the quota tree deletion in
1179 	 * btrfs_clean_quota_tree but this is not done.
1180 	 *
1181 	 * Also, we must always start a transaction without holding the mutex
1182 	 * qgroup_ioctl_lock, see btrfs_quota_enable().
1183 	 */
1184 	trans = btrfs_start_transaction(fs_info->tree_root, 1);
1185 
1186 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1187 	if (IS_ERR(trans)) {
1188 		ret = PTR_ERR(trans);
1189 		trans = NULL;
1190 		goto out;
1191 	}
1192 
1193 	if (!fs_info->quota_root)
1194 		goto out;
1195 
1196 	clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1197 	btrfs_qgroup_wait_for_completion(fs_info, false);
1198 	spin_lock(&fs_info->qgroup_lock);
1199 	quota_root = fs_info->quota_root;
1200 	fs_info->quota_root = NULL;
1201 	fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1202 	spin_unlock(&fs_info->qgroup_lock);
1203 
1204 	btrfs_free_qgroup_config(fs_info);
1205 
1206 	ret = btrfs_clean_quota_tree(trans, quota_root);
1207 	if (ret) {
1208 		btrfs_abort_transaction(trans, ret);
1209 		goto out;
1210 	}
1211 
1212 	ret = btrfs_del_root(trans, &quota_root->root_key);
1213 	if (ret) {
1214 		btrfs_abort_transaction(trans, ret);
1215 		goto out;
1216 	}
1217 
1218 	list_del(&quota_root->dirty_list);
1219 
1220 	btrfs_tree_lock(quota_root->node);
1221 	btrfs_clean_tree_block(quota_root->node);
1222 	btrfs_tree_unlock(quota_root->node);
1223 	btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1224 
1225 	btrfs_put_root(quota_root);
1226 
1227 out:
1228 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1229 	if (ret && trans)
1230 		btrfs_end_transaction(trans);
1231 	else if (trans)
1232 		ret = btrfs_end_transaction(trans);
1233 
1234 	return ret;
1235 }
1236 
qgroup_dirty(struct btrfs_fs_info * fs_info,struct btrfs_qgroup * qgroup)1237 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1238 			 struct btrfs_qgroup *qgroup)
1239 {
1240 	if (list_empty(&qgroup->dirty))
1241 		list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1242 }
1243 
1244 /*
1245  * The easy accounting, we're updating qgroup relationship whose child qgroup
1246  * only has exclusive extents.
1247  *
1248  * In this case, all exclusive extents will also be exclusive for parent, so
1249  * excl/rfer just get added/removed.
1250  *
1251  * So is qgroup reservation space, which should also be added/removed to
1252  * parent.
1253  * Or when child tries to release reservation space, parent will underflow its
1254  * reservation (for relationship adding case).
1255  *
1256  * Caller should hold fs_info->qgroup_lock.
1257  */
__qgroup_excl_accounting(struct btrfs_fs_info * fs_info,struct ulist * tmp,u64 ref_root,struct btrfs_qgroup * src,int sign)1258 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1259 				    struct ulist *tmp, u64 ref_root,
1260 				    struct btrfs_qgroup *src, int sign)
1261 {
1262 	struct btrfs_qgroup *qgroup;
1263 	struct btrfs_qgroup_list *glist;
1264 	struct ulist_node *unode;
1265 	struct ulist_iterator uiter;
1266 	u64 num_bytes = src->excl;
1267 	int ret = 0;
1268 
1269 	qgroup = find_qgroup_rb(fs_info, ref_root);
1270 	if (!qgroup)
1271 		goto out;
1272 
1273 	qgroup->rfer += sign * num_bytes;
1274 	qgroup->rfer_cmpr += sign * num_bytes;
1275 
1276 	WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1277 	qgroup->excl += sign * num_bytes;
1278 	qgroup->excl_cmpr += sign * num_bytes;
1279 
1280 	if (sign > 0)
1281 		qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1282 	else
1283 		qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1284 
1285 	qgroup_dirty(fs_info, qgroup);
1286 
1287 	/* Get all of the parent groups that contain this qgroup */
1288 	list_for_each_entry(glist, &qgroup->groups, next_group) {
1289 		ret = ulist_add(tmp, glist->group->qgroupid,
1290 				qgroup_to_aux(glist->group), GFP_ATOMIC);
1291 		if (ret < 0)
1292 			goto out;
1293 	}
1294 
1295 	/* Iterate all of the parents and adjust their reference counts */
1296 	ULIST_ITER_INIT(&uiter);
1297 	while ((unode = ulist_next(tmp, &uiter))) {
1298 		qgroup = unode_aux_to_qgroup(unode);
1299 		qgroup->rfer += sign * num_bytes;
1300 		qgroup->rfer_cmpr += sign * num_bytes;
1301 		WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1302 		qgroup->excl += sign * num_bytes;
1303 		if (sign > 0)
1304 			qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1305 		else
1306 			qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1307 		qgroup->excl_cmpr += sign * num_bytes;
1308 		qgroup_dirty(fs_info, qgroup);
1309 
1310 		/* Add any parents of the parents */
1311 		list_for_each_entry(glist, &qgroup->groups, next_group) {
1312 			ret = ulist_add(tmp, glist->group->qgroupid,
1313 					qgroup_to_aux(glist->group), GFP_ATOMIC);
1314 			if (ret < 0)
1315 				goto out;
1316 		}
1317 	}
1318 	ret = 0;
1319 out:
1320 	return ret;
1321 }
1322 
1323 
1324 /*
1325  * Quick path for updating qgroup with only excl refs.
1326  *
1327  * In that case, just update all parent will be enough.
1328  * Or we needs to do a full rescan.
1329  * Caller should also hold fs_info->qgroup_lock.
1330  *
1331  * Return 0 for quick update, return >0 for need to full rescan
1332  * and mark INCONSISTENT flag.
1333  * Return < 0 for other error.
1334  */
quick_update_accounting(struct btrfs_fs_info * fs_info,struct ulist * tmp,u64 src,u64 dst,int sign)1335 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1336 				   struct ulist *tmp, u64 src, u64 dst,
1337 				   int sign)
1338 {
1339 	struct btrfs_qgroup *qgroup;
1340 	int ret = 1;
1341 	int err = 0;
1342 
1343 	qgroup = find_qgroup_rb(fs_info, src);
1344 	if (!qgroup)
1345 		goto out;
1346 	if (qgroup->excl == qgroup->rfer) {
1347 		ret = 0;
1348 		err = __qgroup_excl_accounting(fs_info, tmp, dst,
1349 					       qgroup, sign);
1350 		if (err < 0) {
1351 			ret = err;
1352 			goto out;
1353 		}
1354 	}
1355 out:
1356 	if (ret)
1357 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1358 	return ret;
1359 }
1360 
btrfs_add_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1361 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1362 			      u64 dst)
1363 {
1364 	struct btrfs_fs_info *fs_info = trans->fs_info;
1365 	struct btrfs_qgroup *parent;
1366 	struct btrfs_qgroup *member;
1367 	struct btrfs_qgroup_list *list;
1368 	struct ulist *tmp;
1369 	unsigned int nofs_flag;
1370 	int ret = 0;
1371 
1372 	/* Check the level of src and dst first */
1373 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1374 		return -EINVAL;
1375 
1376 	/* We hold a transaction handle open, must do a NOFS allocation. */
1377 	nofs_flag = memalloc_nofs_save();
1378 	tmp = ulist_alloc(GFP_KERNEL);
1379 	memalloc_nofs_restore(nofs_flag);
1380 	if (!tmp)
1381 		return -ENOMEM;
1382 
1383 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1384 	if (!fs_info->quota_root) {
1385 		ret = -ENOTCONN;
1386 		goto out;
1387 	}
1388 	member = find_qgroup_rb(fs_info, src);
1389 	parent = find_qgroup_rb(fs_info, dst);
1390 	if (!member || !parent) {
1391 		ret = -EINVAL;
1392 		goto out;
1393 	}
1394 
1395 	/* check if such qgroup relation exist firstly */
1396 	list_for_each_entry(list, &member->groups, next_group) {
1397 		if (list->group == parent) {
1398 			ret = -EEXIST;
1399 			goto out;
1400 		}
1401 	}
1402 
1403 	ret = add_qgroup_relation_item(trans, src, dst);
1404 	if (ret)
1405 		goto out;
1406 
1407 	ret = add_qgroup_relation_item(trans, dst, src);
1408 	if (ret) {
1409 		del_qgroup_relation_item(trans, src, dst);
1410 		goto out;
1411 	}
1412 
1413 	spin_lock(&fs_info->qgroup_lock);
1414 	ret = add_relation_rb(fs_info, src, dst);
1415 	if (ret < 0) {
1416 		spin_unlock(&fs_info->qgroup_lock);
1417 		goto out;
1418 	}
1419 	ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1420 	spin_unlock(&fs_info->qgroup_lock);
1421 out:
1422 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1423 	ulist_free(tmp);
1424 	return ret;
1425 }
1426 
__del_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1427 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1428 				 u64 dst)
1429 {
1430 	struct btrfs_fs_info *fs_info = trans->fs_info;
1431 	struct btrfs_qgroup *parent;
1432 	struct btrfs_qgroup *member;
1433 	struct btrfs_qgroup_list *list;
1434 	struct ulist *tmp;
1435 	bool found = false;
1436 	unsigned int nofs_flag;
1437 	int ret = 0;
1438 	int ret2;
1439 
1440 	/* We hold a transaction handle open, must do a NOFS allocation. */
1441 	nofs_flag = memalloc_nofs_save();
1442 	tmp = ulist_alloc(GFP_KERNEL);
1443 	memalloc_nofs_restore(nofs_flag);
1444 	if (!tmp)
1445 		return -ENOMEM;
1446 
1447 	if (!fs_info->quota_root) {
1448 		ret = -ENOTCONN;
1449 		goto out;
1450 	}
1451 
1452 	member = find_qgroup_rb(fs_info, src);
1453 	parent = find_qgroup_rb(fs_info, dst);
1454 	/*
1455 	 * The parent/member pair doesn't exist, then try to delete the dead
1456 	 * relation items only.
1457 	 */
1458 	if (!member || !parent)
1459 		goto delete_item;
1460 
1461 	/* check if such qgroup relation exist firstly */
1462 	list_for_each_entry(list, &member->groups, next_group) {
1463 		if (list->group == parent) {
1464 			found = true;
1465 			break;
1466 		}
1467 	}
1468 
1469 delete_item:
1470 	ret = del_qgroup_relation_item(trans, src, dst);
1471 	if (ret < 0 && ret != -ENOENT)
1472 		goto out;
1473 	ret2 = del_qgroup_relation_item(trans, dst, src);
1474 	if (ret2 < 0 && ret2 != -ENOENT)
1475 		goto out;
1476 
1477 	/* At least one deletion succeeded, return 0 */
1478 	if (!ret || !ret2)
1479 		ret = 0;
1480 
1481 	if (found) {
1482 		spin_lock(&fs_info->qgroup_lock);
1483 		del_relation_rb(fs_info, src, dst);
1484 		ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1485 		spin_unlock(&fs_info->qgroup_lock);
1486 	}
1487 out:
1488 	ulist_free(tmp);
1489 	return ret;
1490 }
1491 
btrfs_del_qgroup_relation(struct btrfs_trans_handle * trans,u64 src,u64 dst)1492 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1493 			      u64 dst)
1494 {
1495 	struct btrfs_fs_info *fs_info = trans->fs_info;
1496 	int ret = 0;
1497 
1498 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1499 	ret = __del_qgroup_relation(trans, src, dst);
1500 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1501 
1502 	return ret;
1503 }
1504 
btrfs_create_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)1505 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1506 {
1507 	struct btrfs_fs_info *fs_info = trans->fs_info;
1508 	struct btrfs_root *quota_root;
1509 	struct btrfs_qgroup *qgroup;
1510 	int ret = 0;
1511 
1512 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1513 	if (!fs_info->quota_root) {
1514 		ret = -ENOTCONN;
1515 		goto out;
1516 	}
1517 	quota_root = fs_info->quota_root;
1518 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1519 	if (qgroup) {
1520 		ret = -EEXIST;
1521 		goto out;
1522 	}
1523 
1524 	ret = add_qgroup_item(trans, quota_root, qgroupid);
1525 	if (ret)
1526 		goto out;
1527 
1528 	spin_lock(&fs_info->qgroup_lock);
1529 	qgroup = add_qgroup_rb(fs_info, qgroupid);
1530 	spin_unlock(&fs_info->qgroup_lock);
1531 
1532 	if (IS_ERR(qgroup)) {
1533 		ret = PTR_ERR(qgroup);
1534 		goto out;
1535 	}
1536 	ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1537 out:
1538 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1539 	return ret;
1540 }
1541 
btrfs_remove_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)1542 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1543 {
1544 	struct btrfs_fs_info *fs_info = trans->fs_info;
1545 	struct btrfs_qgroup *qgroup;
1546 	struct btrfs_qgroup_list *list;
1547 	int ret = 0;
1548 
1549 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1550 	if (!fs_info->quota_root) {
1551 		ret = -ENOTCONN;
1552 		goto out;
1553 	}
1554 
1555 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1556 	if (!qgroup) {
1557 		ret = -ENOENT;
1558 		goto out;
1559 	}
1560 
1561 	/* Check if there are no children of this qgroup */
1562 	if (!list_empty(&qgroup->members)) {
1563 		ret = -EBUSY;
1564 		goto out;
1565 	}
1566 
1567 	ret = del_qgroup_item(trans, qgroupid);
1568 	if (ret && ret != -ENOENT)
1569 		goto out;
1570 
1571 	while (!list_empty(&qgroup->groups)) {
1572 		list = list_first_entry(&qgroup->groups,
1573 					struct btrfs_qgroup_list, next_group);
1574 		ret = __del_qgroup_relation(trans, qgroupid,
1575 					    list->group->qgroupid);
1576 		if (ret)
1577 			goto out;
1578 	}
1579 
1580 	spin_lock(&fs_info->qgroup_lock);
1581 	del_qgroup_rb(fs_info, qgroupid);
1582 	spin_unlock(&fs_info->qgroup_lock);
1583 out:
1584 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1585 	return ret;
1586 }
1587 
btrfs_limit_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid,struct btrfs_qgroup_limit * limit)1588 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1589 		       struct btrfs_qgroup_limit *limit)
1590 {
1591 	struct btrfs_fs_info *fs_info = trans->fs_info;
1592 	struct btrfs_qgroup *qgroup;
1593 	int ret = 0;
1594 	/* Sometimes we would want to clear the limit on this qgroup.
1595 	 * To meet this requirement, we treat the -1 as a special value
1596 	 * which tell kernel to clear the limit on this qgroup.
1597 	 */
1598 	const u64 CLEAR_VALUE = -1;
1599 
1600 	mutex_lock(&fs_info->qgroup_ioctl_lock);
1601 	if (!fs_info->quota_root) {
1602 		ret = -ENOTCONN;
1603 		goto out;
1604 	}
1605 
1606 	qgroup = find_qgroup_rb(fs_info, qgroupid);
1607 	if (!qgroup) {
1608 		ret = -ENOENT;
1609 		goto out;
1610 	}
1611 
1612 	spin_lock(&fs_info->qgroup_lock);
1613 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1614 		if (limit->max_rfer == CLEAR_VALUE) {
1615 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1616 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1617 			qgroup->max_rfer = 0;
1618 		} else {
1619 			qgroup->max_rfer = limit->max_rfer;
1620 		}
1621 	}
1622 	if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1623 		if (limit->max_excl == CLEAR_VALUE) {
1624 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1625 			limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1626 			qgroup->max_excl = 0;
1627 		} else {
1628 			qgroup->max_excl = limit->max_excl;
1629 		}
1630 	}
1631 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1632 		if (limit->rsv_rfer == CLEAR_VALUE) {
1633 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1634 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1635 			qgroup->rsv_rfer = 0;
1636 		} else {
1637 			qgroup->rsv_rfer = limit->rsv_rfer;
1638 		}
1639 	}
1640 	if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1641 		if (limit->rsv_excl == CLEAR_VALUE) {
1642 			qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1643 			limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1644 			qgroup->rsv_excl = 0;
1645 		} else {
1646 			qgroup->rsv_excl = limit->rsv_excl;
1647 		}
1648 	}
1649 	qgroup->lim_flags |= limit->flags;
1650 
1651 	spin_unlock(&fs_info->qgroup_lock);
1652 
1653 	ret = update_qgroup_limit_item(trans, qgroup);
1654 	if (ret) {
1655 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1656 		btrfs_info(fs_info, "unable to update quota limit for %llu",
1657 		       qgroupid);
1658 	}
1659 
1660 out:
1661 	mutex_unlock(&fs_info->qgroup_ioctl_lock);
1662 	return ret;
1663 }
1664 
btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info * fs_info,struct btrfs_delayed_ref_root * delayed_refs,struct btrfs_qgroup_extent_record * record)1665 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1666 				struct btrfs_delayed_ref_root *delayed_refs,
1667 				struct btrfs_qgroup_extent_record *record)
1668 {
1669 	struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1670 	struct rb_node *parent_node = NULL;
1671 	struct btrfs_qgroup_extent_record *entry;
1672 	u64 bytenr = record->bytenr;
1673 
1674 	lockdep_assert_held(&delayed_refs->lock);
1675 	trace_btrfs_qgroup_trace_extent(fs_info, record);
1676 
1677 	while (*p) {
1678 		parent_node = *p;
1679 		entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1680 				 node);
1681 		if (bytenr < entry->bytenr) {
1682 			p = &(*p)->rb_left;
1683 		} else if (bytenr > entry->bytenr) {
1684 			p = &(*p)->rb_right;
1685 		} else {
1686 			if (record->data_rsv && !entry->data_rsv) {
1687 				entry->data_rsv = record->data_rsv;
1688 				entry->data_rsv_refroot =
1689 					record->data_rsv_refroot;
1690 			}
1691 			return 1;
1692 		}
1693 	}
1694 
1695 	rb_link_node(&record->node, parent_node, p);
1696 	rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1697 	return 0;
1698 }
1699 
btrfs_qgroup_trace_extent_post(struct btrfs_fs_info * fs_info,struct btrfs_qgroup_extent_record * qrecord)1700 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1701 				   struct btrfs_qgroup_extent_record *qrecord)
1702 {
1703 	struct ulist *old_root;
1704 	u64 bytenr = qrecord->bytenr;
1705 	int ret;
1706 
1707 	ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1708 	if (ret < 0) {
1709 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1710 		btrfs_warn(fs_info,
1711 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1712 			ret);
1713 		return 0;
1714 	}
1715 
1716 	/*
1717 	 * Here we don't need to get the lock of
1718 	 * trans->transaction->delayed_refs, since inserted qrecord won't
1719 	 * be deleted, only qrecord->node may be modified (new qrecord insert)
1720 	 *
1721 	 * So modifying qrecord->old_roots is safe here
1722 	 */
1723 	qrecord->old_roots = old_root;
1724 	return 0;
1725 }
1726 
btrfs_qgroup_trace_extent(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes,gfp_t gfp_flag)1727 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1728 			      u64 num_bytes, gfp_t gfp_flag)
1729 {
1730 	struct btrfs_fs_info *fs_info = trans->fs_info;
1731 	struct btrfs_qgroup_extent_record *record;
1732 	struct btrfs_delayed_ref_root *delayed_refs;
1733 	int ret;
1734 
1735 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1736 	    || bytenr == 0 || num_bytes == 0)
1737 		return 0;
1738 	record = kzalloc(sizeof(*record), gfp_flag);
1739 	if (!record)
1740 		return -ENOMEM;
1741 
1742 	delayed_refs = &trans->transaction->delayed_refs;
1743 	record->bytenr = bytenr;
1744 	record->num_bytes = num_bytes;
1745 	record->old_roots = NULL;
1746 
1747 	spin_lock(&delayed_refs->lock);
1748 	ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1749 	spin_unlock(&delayed_refs->lock);
1750 	if (ret > 0) {
1751 		kfree(record);
1752 		return 0;
1753 	}
1754 	return btrfs_qgroup_trace_extent_post(fs_info, record);
1755 }
1756 
btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle * trans,struct extent_buffer * eb)1757 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1758 				  struct extent_buffer *eb)
1759 {
1760 	struct btrfs_fs_info *fs_info = trans->fs_info;
1761 	int nr = btrfs_header_nritems(eb);
1762 	int i, extent_type, ret;
1763 	struct btrfs_key key;
1764 	struct btrfs_file_extent_item *fi;
1765 	u64 bytenr, num_bytes;
1766 
1767 	/* We can be called directly from walk_up_proc() */
1768 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1769 		return 0;
1770 
1771 	for (i = 0; i < nr; i++) {
1772 		btrfs_item_key_to_cpu(eb, &key, i);
1773 
1774 		if (key.type != BTRFS_EXTENT_DATA_KEY)
1775 			continue;
1776 
1777 		fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1778 		/* filter out non qgroup-accountable extents  */
1779 		extent_type = btrfs_file_extent_type(eb, fi);
1780 
1781 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1782 			continue;
1783 
1784 		bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1785 		if (!bytenr)
1786 			continue;
1787 
1788 		num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1789 
1790 		ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1791 						GFP_NOFS);
1792 		if (ret)
1793 			return ret;
1794 	}
1795 	cond_resched();
1796 	return 0;
1797 }
1798 
1799 /*
1800  * Walk up the tree from the bottom, freeing leaves and any interior
1801  * nodes which have had all slots visited. If a node (leaf or
1802  * interior) is freed, the node above it will have it's slot
1803  * incremented. The root node will never be freed.
1804  *
1805  * At the end of this function, we should have a path which has all
1806  * slots incremented to the next position for a search. If we need to
1807  * read a new node it will be NULL and the node above it will have the
1808  * correct slot selected for a later read.
1809  *
1810  * If we increment the root nodes slot counter past the number of
1811  * elements, 1 is returned to signal completion of the search.
1812  */
adjust_slots_upwards(struct btrfs_path * path,int root_level)1813 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1814 {
1815 	int level = 0;
1816 	int nr, slot;
1817 	struct extent_buffer *eb;
1818 
1819 	if (root_level == 0)
1820 		return 1;
1821 
1822 	while (level <= root_level) {
1823 		eb = path->nodes[level];
1824 		nr = btrfs_header_nritems(eb);
1825 		path->slots[level]++;
1826 		slot = path->slots[level];
1827 		if (slot >= nr || level == 0) {
1828 			/*
1829 			 * Don't free the root -  we will detect this
1830 			 * condition after our loop and return a
1831 			 * positive value for caller to stop walking the tree.
1832 			 */
1833 			if (level != root_level) {
1834 				btrfs_tree_unlock_rw(eb, path->locks[level]);
1835 				path->locks[level] = 0;
1836 
1837 				free_extent_buffer(eb);
1838 				path->nodes[level] = NULL;
1839 				path->slots[level] = 0;
1840 			}
1841 		} else {
1842 			/*
1843 			 * We have a valid slot to walk back down
1844 			 * from. Stop here so caller can process these
1845 			 * new nodes.
1846 			 */
1847 			break;
1848 		}
1849 
1850 		level++;
1851 	}
1852 
1853 	eb = path->nodes[root_level];
1854 	if (path->slots[root_level] >= btrfs_header_nritems(eb))
1855 		return 1;
1856 
1857 	return 0;
1858 }
1859 
1860 /*
1861  * Helper function to trace a subtree tree block swap.
1862  *
1863  * The swap will happen in highest tree block, but there may be a lot of
1864  * tree blocks involved.
1865  *
1866  * For example:
1867  *  OO = Old tree blocks
1868  *  NN = New tree blocks allocated during balance
1869  *
1870  *           File tree (257)                  Reloc tree for 257
1871  * L2              OO                                NN
1872  *               /    \                            /    \
1873  * L1          OO      OO (a)                    OO      NN (a)
1874  *            / \     / \                       / \     / \
1875  * L0       OO   OO OO   OO                   OO   OO NN   NN
1876  *                  (b)  (c)                          (b)  (c)
1877  *
1878  * When calling qgroup_trace_extent_swap(), we will pass:
1879  * @src_eb = OO(a)
1880  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1881  * @dst_level = 0
1882  * @root_level = 1
1883  *
1884  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1885  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1886  *
1887  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1888  *
1889  * 1) Tree search from @src_eb
1890  *    It should acts as a simplified btrfs_search_slot().
1891  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1892  *    (first key).
1893  *
1894  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1895  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1896  *    They should be marked during previous (@dst_level = 1) iteration.
1897  *
1898  * 3) Mark file extents in leaves dirty
1899  *    We don't have good way to pick out new file extents only.
1900  *    So we still follow the old method by scanning all file extents in
1901  *    the leave.
1902  *
1903  * This function can free us from keeping two paths, thus later we only need
1904  * to care about how to iterate all new tree blocks in reloc tree.
1905  */
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)1906 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1907 				    struct extent_buffer *src_eb,
1908 				    struct btrfs_path *dst_path,
1909 				    int dst_level, int root_level,
1910 				    bool trace_leaf)
1911 {
1912 	struct btrfs_key key;
1913 	struct btrfs_path *src_path;
1914 	struct btrfs_fs_info *fs_info = trans->fs_info;
1915 	u32 nodesize = fs_info->nodesize;
1916 	int cur_level = root_level;
1917 	int ret;
1918 
1919 	BUG_ON(dst_level > root_level);
1920 	/* Level mismatch */
1921 	if (btrfs_header_level(src_eb) != root_level)
1922 		return -EINVAL;
1923 
1924 	src_path = btrfs_alloc_path();
1925 	if (!src_path) {
1926 		ret = -ENOMEM;
1927 		goto out;
1928 	}
1929 
1930 	if (dst_level)
1931 		btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1932 	else
1933 		btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1934 
1935 	/* For src_path */
1936 	atomic_inc(&src_eb->refs);
1937 	src_path->nodes[root_level] = src_eb;
1938 	src_path->slots[root_level] = dst_path->slots[root_level];
1939 	src_path->locks[root_level] = 0;
1940 
1941 	/* A simplified version of btrfs_search_slot() */
1942 	while (cur_level >= dst_level) {
1943 		struct btrfs_key src_key;
1944 		struct btrfs_key dst_key;
1945 
1946 		if (src_path->nodes[cur_level] == NULL) {
1947 			struct btrfs_key first_key;
1948 			struct extent_buffer *eb;
1949 			int parent_slot;
1950 			u64 child_gen;
1951 			u64 child_bytenr;
1952 
1953 			eb = src_path->nodes[cur_level + 1];
1954 			parent_slot = src_path->slots[cur_level + 1];
1955 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1956 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1957 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1958 
1959 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
1960 					     cur_level, &first_key);
1961 			if (IS_ERR(eb)) {
1962 				ret = PTR_ERR(eb);
1963 				goto out;
1964 			} else if (!extent_buffer_uptodate(eb)) {
1965 				free_extent_buffer(eb);
1966 				ret = -EIO;
1967 				goto out;
1968 			}
1969 
1970 			src_path->nodes[cur_level] = eb;
1971 
1972 			btrfs_tree_read_lock(eb);
1973 			btrfs_set_lock_blocking_read(eb);
1974 			src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1975 		}
1976 
1977 		src_path->slots[cur_level] = dst_path->slots[cur_level];
1978 		if (cur_level) {
1979 			btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1980 					&dst_key, dst_path->slots[cur_level]);
1981 			btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1982 					&src_key, src_path->slots[cur_level]);
1983 		} else {
1984 			btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1985 					&dst_key, dst_path->slots[cur_level]);
1986 			btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1987 					&src_key, src_path->slots[cur_level]);
1988 		}
1989 		/* Content mismatch, something went wrong */
1990 		if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1991 			ret = -ENOENT;
1992 			goto out;
1993 		}
1994 		cur_level--;
1995 	}
1996 
1997 	/*
1998 	 * Now both @dst_path and @src_path have been populated, record the tree
1999 	 * blocks for qgroup accounting.
2000 	 */
2001 	ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2002 			nodesize, GFP_NOFS);
2003 	if (ret < 0)
2004 		goto out;
2005 	ret = btrfs_qgroup_trace_extent(trans,
2006 			dst_path->nodes[dst_level]->start,
2007 			nodesize, GFP_NOFS);
2008 	if (ret < 0)
2009 		goto out;
2010 
2011 	/* Record leaf file extents */
2012 	if (dst_level == 0 && trace_leaf) {
2013 		ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2014 		if (ret < 0)
2015 			goto out;
2016 		ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2017 	}
2018 out:
2019 	btrfs_free_path(src_path);
2020 	return ret;
2021 }
2022 
2023 /*
2024  * Helper function to do recursive generation-aware depth-first search, to
2025  * locate all new tree blocks in a subtree of reloc tree.
2026  *
2027  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2028  *         reloc tree
2029  * L2         NN (a)
2030  *          /    \
2031  * L1    OO        NN (b)
2032  *      /  \      /  \
2033  * L0  OO  OO    OO  NN
2034  *               (c) (d)
2035  * If we pass:
2036  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2037  * @cur_level = 1
2038  * @root_level = 1
2039  *
2040  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2041  * above tree blocks along with their counter parts in file tree.
2042  * While during search, old tree blocks OO(c) will be skipped as tree block swap
2043  * won't affect OO(c).
2044  */
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)2045 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2046 					   struct extent_buffer *src_eb,
2047 					   struct btrfs_path *dst_path,
2048 					   int cur_level, int root_level,
2049 					   u64 last_snapshot, bool trace_leaf)
2050 {
2051 	struct btrfs_fs_info *fs_info = trans->fs_info;
2052 	struct extent_buffer *eb;
2053 	bool need_cleanup = false;
2054 	int ret = 0;
2055 	int i;
2056 
2057 	/* Level sanity check */
2058 	if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2059 	    root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2060 	    root_level < cur_level) {
2061 		btrfs_err_rl(fs_info,
2062 			"%s: bad levels, cur_level=%d root_level=%d",
2063 			__func__, cur_level, root_level);
2064 		return -EUCLEAN;
2065 	}
2066 
2067 	/* Read the tree block if needed */
2068 	if (dst_path->nodes[cur_level] == NULL) {
2069 		struct btrfs_key first_key;
2070 		int parent_slot;
2071 		u64 child_gen;
2072 		u64 child_bytenr;
2073 
2074 		/*
2075 		 * dst_path->nodes[root_level] must be initialized before
2076 		 * calling this function.
2077 		 */
2078 		if (cur_level == root_level) {
2079 			btrfs_err_rl(fs_info,
2080 	"%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2081 				__func__, root_level, root_level, cur_level);
2082 			return -EUCLEAN;
2083 		}
2084 
2085 		/*
2086 		 * We need to get child blockptr/gen from parent before we can
2087 		 * read it.
2088 		  */
2089 		eb = dst_path->nodes[cur_level + 1];
2090 		parent_slot = dst_path->slots[cur_level + 1];
2091 		child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2092 		child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2093 		btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2094 
2095 		/* This node is old, no need to trace */
2096 		if (child_gen < last_snapshot)
2097 			goto out;
2098 
2099 		eb = read_tree_block(fs_info, child_bytenr, child_gen,
2100 				     cur_level, &first_key);
2101 		if (IS_ERR(eb)) {
2102 			ret = PTR_ERR(eb);
2103 			goto out;
2104 		} else if (!extent_buffer_uptodate(eb)) {
2105 			free_extent_buffer(eb);
2106 			ret = -EIO;
2107 			goto out;
2108 		}
2109 
2110 		dst_path->nodes[cur_level] = eb;
2111 		dst_path->slots[cur_level] = 0;
2112 
2113 		btrfs_tree_read_lock(eb);
2114 		btrfs_set_lock_blocking_read(eb);
2115 		dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
2116 		need_cleanup = true;
2117 	}
2118 
2119 	/* Now record this tree block and its counter part for qgroups */
2120 	ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2121 				       root_level, trace_leaf);
2122 	if (ret < 0)
2123 		goto cleanup;
2124 
2125 	eb = dst_path->nodes[cur_level];
2126 
2127 	if (cur_level > 0) {
2128 		/* Iterate all child tree blocks */
2129 		for (i = 0; i < btrfs_header_nritems(eb); i++) {
2130 			/* Skip old tree blocks as they won't be swapped */
2131 			if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2132 				continue;
2133 			dst_path->slots[cur_level] = i;
2134 
2135 			/* Recursive call (at most 7 times) */
2136 			ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2137 					dst_path, cur_level - 1, root_level,
2138 					last_snapshot, trace_leaf);
2139 			if (ret < 0)
2140 				goto cleanup;
2141 		}
2142 	}
2143 
2144 cleanup:
2145 	if (need_cleanup) {
2146 		/* Clean up */
2147 		btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2148 				     dst_path->locks[cur_level]);
2149 		free_extent_buffer(dst_path->nodes[cur_level]);
2150 		dst_path->nodes[cur_level] = NULL;
2151 		dst_path->slots[cur_level] = 0;
2152 		dst_path->locks[cur_level] = 0;
2153 	}
2154 out:
2155 	return ret;
2156 }
2157 
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)2158 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2159 				struct extent_buffer *src_eb,
2160 				struct extent_buffer *dst_eb,
2161 				u64 last_snapshot, bool trace_leaf)
2162 {
2163 	struct btrfs_fs_info *fs_info = trans->fs_info;
2164 	struct btrfs_path *dst_path = NULL;
2165 	int level;
2166 	int ret;
2167 
2168 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2169 		return 0;
2170 
2171 	/* Wrong parameter order */
2172 	if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2173 		btrfs_err_rl(fs_info,
2174 		"%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2175 			     btrfs_header_generation(src_eb),
2176 			     btrfs_header_generation(dst_eb));
2177 		return -EUCLEAN;
2178 	}
2179 
2180 	if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2181 		ret = -EIO;
2182 		goto out;
2183 	}
2184 
2185 	level = btrfs_header_level(dst_eb);
2186 	dst_path = btrfs_alloc_path();
2187 	if (!dst_path) {
2188 		ret = -ENOMEM;
2189 		goto out;
2190 	}
2191 	/* For dst_path */
2192 	atomic_inc(&dst_eb->refs);
2193 	dst_path->nodes[level] = dst_eb;
2194 	dst_path->slots[level] = 0;
2195 	dst_path->locks[level] = 0;
2196 
2197 	/* Do the generation aware breadth-first search */
2198 	ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2199 					      level, last_snapshot, trace_leaf);
2200 	if (ret < 0)
2201 		goto out;
2202 	ret = 0;
2203 
2204 out:
2205 	btrfs_free_path(dst_path);
2206 	if (ret < 0)
2207 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2208 	return ret;
2209 }
2210 
btrfs_qgroup_trace_subtree(struct btrfs_trans_handle * trans,struct extent_buffer * root_eb,u64 root_gen,int root_level)2211 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2212 			       struct extent_buffer *root_eb,
2213 			       u64 root_gen, int root_level)
2214 {
2215 	struct btrfs_fs_info *fs_info = trans->fs_info;
2216 	int ret = 0;
2217 	int level;
2218 	struct extent_buffer *eb = root_eb;
2219 	struct btrfs_path *path = NULL;
2220 
2221 	BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2222 	BUG_ON(root_eb == NULL);
2223 
2224 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2225 		return 0;
2226 
2227 	if (!extent_buffer_uptodate(root_eb)) {
2228 		ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2229 		if (ret)
2230 			goto out;
2231 	}
2232 
2233 	if (root_level == 0) {
2234 		ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2235 		goto out;
2236 	}
2237 
2238 	path = btrfs_alloc_path();
2239 	if (!path)
2240 		return -ENOMEM;
2241 
2242 	/*
2243 	 * Walk down the tree.  Missing extent blocks are filled in as
2244 	 * we go. Metadata is accounted every time we read a new
2245 	 * extent block.
2246 	 *
2247 	 * When we reach a leaf, we account for file extent items in it,
2248 	 * walk back up the tree (adjusting slot pointers as we go)
2249 	 * and restart the search process.
2250 	 */
2251 	atomic_inc(&root_eb->refs);	/* For path */
2252 	path->nodes[root_level] = root_eb;
2253 	path->slots[root_level] = 0;
2254 	path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2255 walk_down:
2256 	level = root_level;
2257 	while (level >= 0) {
2258 		if (path->nodes[level] == NULL) {
2259 			struct btrfs_key first_key;
2260 			int parent_slot;
2261 			u64 child_gen;
2262 			u64 child_bytenr;
2263 
2264 			/*
2265 			 * We need to get child blockptr/gen from parent before
2266 			 * we can read it.
2267 			  */
2268 			eb = path->nodes[level + 1];
2269 			parent_slot = path->slots[level + 1];
2270 			child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2271 			child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2272 			btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2273 
2274 			eb = read_tree_block(fs_info, child_bytenr, child_gen,
2275 					     level, &first_key);
2276 			if (IS_ERR(eb)) {
2277 				ret = PTR_ERR(eb);
2278 				goto out;
2279 			} else if (!extent_buffer_uptodate(eb)) {
2280 				free_extent_buffer(eb);
2281 				ret = -EIO;
2282 				goto out;
2283 			}
2284 
2285 			path->nodes[level] = eb;
2286 			path->slots[level] = 0;
2287 
2288 			btrfs_tree_read_lock(eb);
2289 			btrfs_set_lock_blocking_read(eb);
2290 			path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2291 
2292 			ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2293 							fs_info->nodesize,
2294 							GFP_NOFS);
2295 			if (ret)
2296 				goto out;
2297 		}
2298 
2299 		if (level == 0) {
2300 			ret = btrfs_qgroup_trace_leaf_items(trans,
2301 							    path->nodes[level]);
2302 			if (ret)
2303 				goto out;
2304 
2305 			/* Nonzero return here means we completed our search */
2306 			ret = adjust_slots_upwards(path, root_level);
2307 			if (ret)
2308 				break;
2309 
2310 			/* Restart search with new slots */
2311 			goto walk_down;
2312 		}
2313 
2314 		level--;
2315 	}
2316 
2317 	ret = 0;
2318 out:
2319 	btrfs_free_path(path);
2320 
2321 	return ret;
2322 }
2323 
2324 #define UPDATE_NEW	0
2325 #define UPDATE_OLD	1
2326 /*
2327  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2328  */
qgroup_update_refcnt(struct btrfs_fs_info * fs_info,struct ulist * roots,struct ulist * tmp,struct ulist * qgroups,u64 seq,int update_old)2329 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2330 				struct ulist *roots, struct ulist *tmp,
2331 				struct ulist *qgroups, u64 seq, int update_old)
2332 {
2333 	struct ulist_node *unode;
2334 	struct ulist_iterator uiter;
2335 	struct ulist_node *tmp_unode;
2336 	struct ulist_iterator tmp_uiter;
2337 	struct btrfs_qgroup *qg;
2338 	int ret = 0;
2339 
2340 	if (!roots)
2341 		return 0;
2342 	ULIST_ITER_INIT(&uiter);
2343 	while ((unode = ulist_next(roots, &uiter))) {
2344 		qg = find_qgroup_rb(fs_info, unode->val);
2345 		if (!qg)
2346 			continue;
2347 
2348 		ulist_reinit(tmp);
2349 		ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2350 				GFP_ATOMIC);
2351 		if (ret < 0)
2352 			return ret;
2353 		ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2354 		if (ret < 0)
2355 			return ret;
2356 		ULIST_ITER_INIT(&tmp_uiter);
2357 		while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2358 			struct btrfs_qgroup_list *glist;
2359 
2360 			qg = unode_aux_to_qgroup(tmp_unode);
2361 			if (update_old)
2362 				btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2363 			else
2364 				btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2365 			list_for_each_entry(glist, &qg->groups, next_group) {
2366 				ret = ulist_add(qgroups, glist->group->qgroupid,
2367 						qgroup_to_aux(glist->group),
2368 						GFP_ATOMIC);
2369 				if (ret < 0)
2370 					return ret;
2371 				ret = ulist_add(tmp, glist->group->qgroupid,
2372 						qgroup_to_aux(glist->group),
2373 						GFP_ATOMIC);
2374 				if (ret < 0)
2375 					return ret;
2376 			}
2377 		}
2378 	}
2379 	return 0;
2380 }
2381 
2382 /*
2383  * Update qgroup rfer/excl counters.
2384  * Rfer update is easy, codes can explain themselves.
2385  *
2386  * Excl update is tricky, the update is split into 2 parts.
2387  * Part 1: Possible exclusive <-> sharing detect:
2388  *	|	A	|	!A	|
2389  *  -------------------------------------
2390  *  B	|	*	|	-	|
2391  *  -------------------------------------
2392  *  !B	|	+	|	**	|
2393  *  -------------------------------------
2394  *
2395  * Conditions:
2396  * A:	cur_old_roots < nr_old_roots	(not exclusive before)
2397  * !A:	cur_old_roots == nr_old_roots	(possible exclusive before)
2398  * B:	cur_new_roots < nr_new_roots	(not exclusive now)
2399  * !B:	cur_new_roots == nr_new_roots	(possible exclusive now)
2400  *
2401  * Results:
2402  * +: Possible sharing -> exclusive	-: Possible exclusive -> sharing
2403  * *: Definitely not changed.		**: Possible unchanged.
2404  *
2405  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2406  *
2407  * To make the logic clear, we first use condition A and B to split
2408  * combination into 4 results.
2409  *
2410  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2411  * only on variant maybe 0.
2412  *
2413  * Lastly, check result **, since there are 2 variants maybe 0, split them
2414  * again(2x2).
2415  * But this time we don't need to consider other things, the codes and logic
2416  * is easy to understand now.
2417  */
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)2418 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2419 				  struct ulist *qgroups,
2420 				  u64 nr_old_roots,
2421 				  u64 nr_new_roots,
2422 				  u64 num_bytes, u64 seq)
2423 {
2424 	struct ulist_node *unode;
2425 	struct ulist_iterator uiter;
2426 	struct btrfs_qgroup *qg;
2427 	u64 cur_new_count, cur_old_count;
2428 
2429 	ULIST_ITER_INIT(&uiter);
2430 	while ((unode = ulist_next(qgroups, &uiter))) {
2431 		bool dirty = false;
2432 
2433 		qg = unode_aux_to_qgroup(unode);
2434 		cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2435 		cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2436 
2437 		trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2438 					     cur_new_count);
2439 
2440 		/* Rfer update part */
2441 		if (cur_old_count == 0 && cur_new_count > 0) {
2442 			qg->rfer += num_bytes;
2443 			qg->rfer_cmpr += num_bytes;
2444 			dirty = true;
2445 		}
2446 		if (cur_old_count > 0 && cur_new_count == 0) {
2447 			qg->rfer -= num_bytes;
2448 			qg->rfer_cmpr -= num_bytes;
2449 			dirty = true;
2450 		}
2451 
2452 		/* Excl update part */
2453 		/* Exclusive/none -> shared case */
2454 		if (cur_old_count == nr_old_roots &&
2455 		    cur_new_count < nr_new_roots) {
2456 			/* Exclusive -> shared */
2457 			if (cur_old_count != 0) {
2458 				qg->excl -= num_bytes;
2459 				qg->excl_cmpr -= num_bytes;
2460 				dirty = true;
2461 			}
2462 		}
2463 
2464 		/* Shared -> exclusive/none case */
2465 		if (cur_old_count < nr_old_roots &&
2466 		    cur_new_count == nr_new_roots) {
2467 			/* Shared->exclusive */
2468 			if (cur_new_count != 0) {
2469 				qg->excl += num_bytes;
2470 				qg->excl_cmpr += num_bytes;
2471 				dirty = true;
2472 			}
2473 		}
2474 
2475 		/* Exclusive/none -> exclusive/none case */
2476 		if (cur_old_count == nr_old_roots &&
2477 		    cur_new_count == nr_new_roots) {
2478 			if (cur_old_count == 0) {
2479 				/* None -> exclusive/none */
2480 
2481 				if (cur_new_count != 0) {
2482 					/* None -> exclusive */
2483 					qg->excl += num_bytes;
2484 					qg->excl_cmpr += num_bytes;
2485 					dirty = true;
2486 				}
2487 				/* None -> none, nothing changed */
2488 			} else {
2489 				/* Exclusive -> exclusive/none */
2490 
2491 				if (cur_new_count == 0) {
2492 					/* Exclusive -> none */
2493 					qg->excl -= num_bytes;
2494 					qg->excl_cmpr -= num_bytes;
2495 					dirty = true;
2496 				}
2497 				/* Exclusive -> exclusive, nothing changed */
2498 			}
2499 		}
2500 
2501 		if (dirty)
2502 			qgroup_dirty(fs_info, qg);
2503 	}
2504 	return 0;
2505 }
2506 
2507 /*
2508  * Check if the @roots potentially is a list of fs tree roots
2509  *
2510  * Return 0 for definitely not a fs/subvol tree roots ulist
2511  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2512  *          one as well)
2513  */
maybe_fs_roots(struct ulist * roots)2514 static int maybe_fs_roots(struct ulist *roots)
2515 {
2516 	struct ulist_node *unode;
2517 	struct ulist_iterator uiter;
2518 
2519 	/* Empty one, still possible for fs roots */
2520 	if (!roots || roots->nnodes == 0)
2521 		return 1;
2522 
2523 	ULIST_ITER_INIT(&uiter);
2524 	unode = ulist_next(roots, &uiter);
2525 	if (!unode)
2526 		return 1;
2527 
2528 	/*
2529 	 * If it contains fs tree roots, then it must belong to fs/subvol
2530 	 * trees.
2531 	 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2532 	 */
2533 	return is_fstree(unode->val);
2534 }
2535 
btrfs_qgroup_account_extent(struct btrfs_trans_handle * trans,u64 bytenr,u64 num_bytes,struct ulist * old_roots,struct ulist * new_roots)2536 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2537 				u64 num_bytes, struct ulist *old_roots,
2538 				struct ulist *new_roots)
2539 {
2540 	struct btrfs_fs_info *fs_info = trans->fs_info;
2541 	struct ulist *qgroups = NULL;
2542 	struct ulist *tmp = NULL;
2543 	u64 seq;
2544 	u64 nr_new_roots = 0;
2545 	u64 nr_old_roots = 0;
2546 	int ret = 0;
2547 
2548 	/*
2549 	 * If quotas get disabled meanwhile, the resouces need to be freed and
2550 	 * we can't just exit here.
2551 	 */
2552 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2553 		goto out_free;
2554 
2555 	if (new_roots) {
2556 		if (!maybe_fs_roots(new_roots))
2557 			goto out_free;
2558 		nr_new_roots = new_roots->nnodes;
2559 	}
2560 	if (old_roots) {
2561 		if (!maybe_fs_roots(old_roots))
2562 			goto out_free;
2563 		nr_old_roots = old_roots->nnodes;
2564 	}
2565 
2566 	/* Quick exit, either not fs tree roots, or won't affect any qgroup */
2567 	if (nr_old_roots == 0 && nr_new_roots == 0)
2568 		goto out_free;
2569 
2570 	BUG_ON(!fs_info->quota_root);
2571 
2572 	trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2573 					num_bytes, nr_old_roots, nr_new_roots);
2574 
2575 	qgroups = ulist_alloc(GFP_NOFS);
2576 	if (!qgroups) {
2577 		ret = -ENOMEM;
2578 		goto out_free;
2579 	}
2580 	tmp = ulist_alloc(GFP_NOFS);
2581 	if (!tmp) {
2582 		ret = -ENOMEM;
2583 		goto out_free;
2584 	}
2585 
2586 	mutex_lock(&fs_info->qgroup_rescan_lock);
2587 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2588 		if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2589 			mutex_unlock(&fs_info->qgroup_rescan_lock);
2590 			ret = 0;
2591 			goto out_free;
2592 		}
2593 	}
2594 	mutex_unlock(&fs_info->qgroup_rescan_lock);
2595 
2596 	spin_lock(&fs_info->qgroup_lock);
2597 	seq = fs_info->qgroup_seq;
2598 
2599 	/* Update old refcnts using old_roots */
2600 	ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2601 				   UPDATE_OLD);
2602 	if (ret < 0)
2603 		goto out;
2604 
2605 	/* Update new refcnts using new_roots */
2606 	ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2607 				   UPDATE_NEW);
2608 	if (ret < 0)
2609 		goto out;
2610 
2611 	qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2612 			       num_bytes, seq);
2613 
2614 	/*
2615 	 * Bump qgroup_seq to avoid seq overlap
2616 	 */
2617 	fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2618 out:
2619 	spin_unlock(&fs_info->qgroup_lock);
2620 out_free:
2621 	ulist_free(tmp);
2622 	ulist_free(qgroups);
2623 	ulist_free(old_roots);
2624 	ulist_free(new_roots);
2625 	return ret;
2626 }
2627 
btrfs_qgroup_account_extents(struct btrfs_trans_handle * trans)2628 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2629 {
2630 	struct btrfs_fs_info *fs_info = trans->fs_info;
2631 	struct btrfs_qgroup_extent_record *record;
2632 	struct btrfs_delayed_ref_root *delayed_refs;
2633 	struct ulist *new_roots = NULL;
2634 	struct rb_node *node;
2635 	u64 num_dirty_extents = 0;
2636 	u64 qgroup_to_skip;
2637 	int ret = 0;
2638 
2639 	delayed_refs = &trans->transaction->delayed_refs;
2640 	qgroup_to_skip = delayed_refs->qgroup_to_skip;
2641 	while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2642 		record = rb_entry(node, struct btrfs_qgroup_extent_record,
2643 				  node);
2644 
2645 		num_dirty_extents++;
2646 		trace_btrfs_qgroup_account_extents(fs_info, record);
2647 
2648 		if (!ret) {
2649 			/*
2650 			 * Old roots should be searched when inserting qgroup
2651 			 * extent record
2652 			 */
2653 			if (WARN_ON(!record->old_roots)) {
2654 				/* Search commit root to find old_roots */
2655 				ret = btrfs_find_all_roots(NULL, fs_info,
2656 						record->bytenr, 0,
2657 						&record->old_roots, false);
2658 				if (ret < 0)
2659 					goto cleanup;
2660 			}
2661 
2662 			/* Free the reserved data space */
2663 			btrfs_qgroup_free_refroot(fs_info,
2664 					record->data_rsv_refroot,
2665 					record->data_rsv,
2666 					BTRFS_QGROUP_RSV_DATA);
2667 			/*
2668 			 * Use SEQ_LAST as time_seq to do special search, which
2669 			 * doesn't lock tree or delayed_refs and search current
2670 			 * root. It's safe inside commit_transaction().
2671 			 */
2672 			ret = btrfs_find_all_roots(trans, fs_info,
2673 				record->bytenr, SEQ_LAST, &new_roots, false);
2674 			if (ret < 0)
2675 				goto cleanup;
2676 			if (qgroup_to_skip) {
2677 				ulist_del(new_roots, qgroup_to_skip, 0);
2678 				ulist_del(record->old_roots, qgroup_to_skip,
2679 					  0);
2680 			}
2681 			ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2682 							  record->num_bytes,
2683 							  record->old_roots,
2684 							  new_roots);
2685 			record->old_roots = NULL;
2686 			new_roots = NULL;
2687 		}
2688 cleanup:
2689 		ulist_free(record->old_roots);
2690 		ulist_free(new_roots);
2691 		new_roots = NULL;
2692 		rb_erase(node, &delayed_refs->dirty_extent_root);
2693 		kfree(record);
2694 
2695 	}
2696 	trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2697 				       num_dirty_extents);
2698 	return ret;
2699 }
2700 
2701 /*
2702  * called from commit_transaction. Writes all changed qgroups to disk.
2703  */
btrfs_run_qgroups(struct btrfs_trans_handle * trans)2704 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2705 {
2706 	struct btrfs_fs_info *fs_info = trans->fs_info;
2707 	int ret = 0;
2708 
2709 	if (!fs_info->quota_root)
2710 		return ret;
2711 
2712 	spin_lock(&fs_info->qgroup_lock);
2713 	while (!list_empty(&fs_info->dirty_qgroups)) {
2714 		struct btrfs_qgroup *qgroup;
2715 		qgroup = list_first_entry(&fs_info->dirty_qgroups,
2716 					  struct btrfs_qgroup, dirty);
2717 		list_del_init(&qgroup->dirty);
2718 		spin_unlock(&fs_info->qgroup_lock);
2719 		ret = update_qgroup_info_item(trans, qgroup);
2720 		if (ret)
2721 			fs_info->qgroup_flags |=
2722 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2723 		ret = update_qgroup_limit_item(trans, qgroup);
2724 		if (ret)
2725 			fs_info->qgroup_flags |=
2726 					BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2727 		spin_lock(&fs_info->qgroup_lock);
2728 	}
2729 	if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2730 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2731 	else
2732 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2733 	spin_unlock(&fs_info->qgroup_lock);
2734 
2735 	ret = update_qgroup_status_item(trans);
2736 	if (ret)
2737 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2738 
2739 	return ret;
2740 }
2741 
2742 /*
2743  * Copy the accounting information between qgroups. This is necessary
2744  * when a snapshot or a subvolume is created. Throwing an error will
2745  * cause a transaction abort so we take extra care here to only error
2746  * when a readonly fs is a reasonable outcome.
2747  */
btrfs_qgroup_inherit(struct btrfs_trans_handle * trans,u64 srcid,u64 objectid,struct btrfs_qgroup_inherit * inherit)2748 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2749 			 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2750 {
2751 	int ret = 0;
2752 	int i;
2753 	u64 *i_qgroups;
2754 	bool committing = false;
2755 	struct btrfs_fs_info *fs_info = trans->fs_info;
2756 	struct btrfs_root *quota_root;
2757 	struct btrfs_qgroup *srcgroup;
2758 	struct btrfs_qgroup *dstgroup;
2759 	bool need_rescan = false;
2760 	u32 level_size = 0;
2761 	u64 nums;
2762 
2763 	/*
2764 	 * There are only two callers of this function.
2765 	 *
2766 	 * One in create_subvol() in the ioctl context, which needs to hold
2767 	 * the qgroup_ioctl_lock.
2768 	 *
2769 	 * The other one in create_pending_snapshot() where no other qgroup
2770 	 * code can modify the fs as they all need to either start a new trans
2771 	 * or hold a trans handler, thus we don't need to hold
2772 	 * qgroup_ioctl_lock.
2773 	 * This would avoid long and complex lock chain and make lockdep happy.
2774 	 */
2775 	spin_lock(&fs_info->trans_lock);
2776 	if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2777 		committing = true;
2778 	spin_unlock(&fs_info->trans_lock);
2779 
2780 	if (!committing)
2781 		mutex_lock(&fs_info->qgroup_ioctl_lock);
2782 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2783 		goto out;
2784 
2785 	quota_root = fs_info->quota_root;
2786 	if (!quota_root) {
2787 		ret = -EINVAL;
2788 		goto out;
2789 	}
2790 
2791 	if (inherit) {
2792 		i_qgroups = (u64 *)(inherit + 1);
2793 		nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2794 		       2 * inherit->num_excl_copies;
2795 		for (i = 0; i < nums; ++i) {
2796 			srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2797 
2798 			/*
2799 			 * Zero out invalid groups so we can ignore
2800 			 * them later.
2801 			 */
2802 			if (!srcgroup ||
2803 			    ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2804 				*i_qgroups = 0ULL;
2805 
2806 			++i_qgroups;
2807 		}
2808 	}
2809 
2810 	/*
2811 	 * create a tracking group for the subvol itself
2812 	 */
2813 	ret = add_qgroup_item(trans, quota_root, objectid);
2814 	if (ret)
2815 		goto out;
2816 
2817 	/*
2818 	 * add qgroup to all inherited groups
2819 	 */
2820 	if (inherit) {
2821 		i_qgroups = (u64 *)(inherit + 1);
2822 		for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2823 			if (*i_qgroups == 0)
2824 				continue;
2825 			ret = add_qgroup_relation_item(trans, objectid,
2826 						       *i_qgroups);
2827 			if (ret && ret != -EEXIST)
2828 				goto out;
2829 			ret = add_qgroup_relation_item(trans, *i_qgroups,
2830 						       objectid);
2831 			if (ret && ret != -EEXIST)
2832 				goto out;
2833 		}
2834 		ret = 0;
2835 	}
2836 
2837 
2838 	spin_lock(&fs_info->qgroup_lock);
2839 
2840 	dstgroup = add_qgroup_rb(fs_info, objectid);
2841 	if (IS_ERR(dstgroup)) {
2842 		ret = PTR_ERR(dstgroup);
2843 		goto unlock;
2844 	}
2845 
2846 	if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2847 		dstgroup->lim_flags = inherit->lim.flags;
2848 		dstgroup->max_rfer = inherit->lim.max_rfer;
2849 		dstgroup->max_excl = inherit->lim.max_excl;
2850 		dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2851 		dstgroup->rsv_excl = inherit->lim.rsv_excl;
2852 
2853 		ret = update_qgroup_limit_item(trans, dstgroup);
2854 		if (ret) {
2855 			fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2856 			btrfs_info(fs_info,
2857 				   "unable to update quota limit for %llu",
2858 				   dstgroup->qgroupid);
2859 			goto unlock;
2860 		}
2861 	}
2862 
2863 	if (srcid) {
2864 		srcgroup = find_qgroup_rb(fs_info, srcid);
2865 		if (!srcgroup)
2866 			goto unlock;
2867 
2868 		/*
2869 		 * We call inherit after we clone the root in order to make sure
2870 		 * our counts don't go crazy, so at this point the only
2871 		 * difference between the two roots should be the root node.
2872 		 */
2873 		level_size = fs_info->nodesize;
2874 		dstgroup->rfer = srcgroup->rfer;
2875 		dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2876 		dstgroup->excl = level_size;
2877 		dstgroup->excl_cmpr = level_size;
2878 		srcgroup->excl = level_size;
2879 		srcgroup->excl_cmpr = level_size;
2880 
2881 		/* inherit the limit info */
2882 		dstgroup->lim_flags = srcgroup->lim_flags;
2883 		dstgroup->max_rfer = srcgroup->max_rfer;
2884 		dstgroup->max_excl = srcgroup->max_excl;
2885 		dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2886 		dstgroup->rsv_excl = srcgroup->rsv_excl;
2887 
2888 		qgroup_dirty(fs_info, dstgroup);
2889 		qgroup_dirty(fs_info, srcgroup);
2890 	}
2891 
2892 	if (!inherit)
2893 		goto unlock;
2894 
2895 	i_qgroups = (u64 *)(inherit + 1);
2896 	for (i = 0; i < inherit->num_qgroups; ++i) {
2897 		if (*i_qgroups) {
2898 			ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2899 			if (ret)
2900 				goto unlock;
2901 		}
2902 		++i_qgroups;
2903 
2904 		/*
2905 		 * If we're doing a snapshot, and adding the snapshot to a new
2906 		 * qgroup, the numbers are guaranteed to be incorrect.
2907 		 */
2908 		if (srcid)
2909 			need_rescan = true;
2910 	}
2911 
2912 	for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2913 		struct btrfs_qgroup *src;
2914 		struct btrfs_qgroup *dst;
2915 
2916 		if (!i_qgroups[0] || !i_qgroups[1])
2917 			continue;
2918 
2919 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2920 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2921 
2922 		if (!src || !dst) {
2923 			ret = -EINVAL;
2924 			goto unlock;
2925 		}
2926 
2927 		dst->rfer = src->rfer - level_size;
2928 		dst->rfer_cmpr = src->rfer_cmpr - level_size;
2929 
2930 		/* Manually tweaking numbers certainly needs a rescan */
2931 		need_rescan = true;
2932 	}
2933 	for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2934 		struct btrfs_qgroup *src;
2935 		struct btrfs_qgroup *dst;
2936 
2937 		if (!i_qgroups[0] || !i_qgroups[1])
2938 			continue;
2939 
2940 		src = find_qgroup_rb(fs_info, i_qgroups[0]);
2941 		dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2942 
2943 		if (!src || !dst) {
2944 			ret = -EINVAL;
2945 			goto unlock;
2946 		}
2947 
2948 		dst->excl = src->excl + level_size;
2949 		dst->excl_cmpr = src->excl_cmpr + level_size;
2950 		need_rescan = true;
2951 	}
2952 
2953 unlock:
2954 	spin_unlock(&fs_info->qgroup_lock);
2955 	if (!ret)
2956 		ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
2957 out:
2958 	if (!committing)
2959 		mutex_unlock(&fs_info->qgroup_ioctl_lock);
2960 	if (need_rescan)
2961 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2962 	return ret;
2963 }
2964 
qgroup_check_limits(const struct btrfs_qgroup * qg,u64 num_bytes)2965 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2966 {
2967 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2968 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2969 		return false;
2970 
2971 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2972 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2973 		return false;
2974 
2975 	return true;
2976 }
2977 
qgroup_reserve(struct btrfs_root * root,u64 num_bytes,bool enforce,enum btrfs_qgroup_rsv_type type)2978 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2979 			  enum btrfs_qgroup_rsv_type type)
2980 {
2981 	struct btrfs_qgroup *qgroup;
2982 	struct btrfs_fs_info *fs_info = root->fs_info;
2983 	u64 ref_root = root->root_key.objectid;
2984 	int ret = 0;
2985 	struct ulist_node *unode;
2986 	struct ulist_iterator uiter;
2987 
2988 	if (!is_fstree(ref_root))
2989 		return 0;
2990 
2991 	if (num_bytes == 0)
2992 		return 0;
2993 
2994 	if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2995 	    capable(CAP_SYS_RESOURCE))
2996 		enforce = false;
2997 
2998 	spin_lock(&fs_info->qgroup_lock);
2999 	if (!fs_info->quota_root)
3000 		goto out;
3001 
3002 	qgroup = find_qgroup_rb(fs_info, ref_root);
3003 	if (!qgroup)
3004 		goto out;
3005 
3006 	/*
3007 	 * in a first step, we check all affected qgroups if any limits would
3008 	 * be exceeded
3009 	 */
3010 	ulist_reinit(fs_info->qgroup_ulist);
3011 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3012 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3013 	if (ret < 0)
3014 		goto out;
3015 	ULIST_ITER_INIT(&uiter);
3016 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3017 		struct btrfs_qgroup *qg;
3018 		struct btrfs_qgroup_list *glist;
3019 
3020 		qg = unode_aux_to_qgroup(unode);
3021 
3022 		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3023 			ret = -EDQUOT;
3024 			goto out;
3025 		}
3026 
3027 		list_for_each_entry(glist, &qg->groups, next_group) {
3028 			ret = ulist_add(fs_info->qgroup_ulist,
3029 					glist->group->qgroupid,
3030 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3031 			if (ret < 0)
3032 				goto out;
3033 		}
3034 	}
3035 	ret = 0;
3036 	/*
3037 	 * no limits exceeded, now record the reservation into all qgroups
3038 	 */
3039 	ULIST_ITER_INIT(&uiter);
3040 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3041 		struct btrfs_qgroup *qg;
3042 
3043 		qg = unode_aux_to_qgroup(unode);
3044 
3045 		qgroup_rsv_add(fs_info, qg, num_bytes, type);
3046 	}
3047 
3048 out:
3049 	spin_unlock(&fs_info->qgroup_lock);
3050 	return ret;
3051 }
3052 
3053 /*
3054  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
3055  * qgroup).
3056  *
3057  * Will handle all higher level qgroup too.
3058  *
3059  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3060  * This special case is only used for META_PERTRANS type.
3061  */
btrfs_qgroup_free_refroot(struct btrfs_fs_info * fs_info,u64 ref_root,u64 num_bytes,enum btrfs_qgroup_rsv_type type)3062 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3063 			       u64 ref_root, u64 num_bytes,
3064 			       enum btrfs_qgroup_rsv_type type)
3065 {
3066 	struct btrfs_qgroup *qgroup;
3067 	struct ulist_node *unode;
3068 	struct ulist_iterator uiter;
3069 	int ret = 0;
3070 
3071 	if (!is_fstree(ref_root))
3072 		return;
3073 
3074 	if (num_bytes == 0)
3075 		return;
3076 
3077 	if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3078 		WARN(1, "%s: Invalid type to free", __func__);
3079 		return;
3080 	}
3081 	spin_lock(&fs_info->qgroup_lock);
3082 
3083 	if (!fs_info->quota_root)
3084 		goto out;
3085 
3086 	qgroup = find_qgroup_rb(fs_info, ref_root);
3087 	if (!qgroup)
3088 		goto out;
3089 
3090 	if (num_bytes == (u64)-1)
3091 		/*
3092 		 * We're freeing all pertrans rsv, get reserved value from
3093 		 * level 0 qgroup as real num_bytes to free.
3094 		 */
3095 		num_bytes = qgroup->rsv.values[type];
3096 
3097 	ulist_reinit(fs_info->qgroup_ulist);
3098 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3099 			qgroup_to_aux(qgroup), GFP_ATOMIC);
3100 	if (ret < 0)
3101 		goto out;
3102 	ULIST_ITER_INIT(&uiter);
3103 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3104 		struct btrfs_qgroup *qg;
3105 		struct btrfs_qgroup_list *glist;
3106 
3107 		qg = unode_aux_to_qgroup(unode);
3108 
3109 		qgroup_rsv_release(fs_info, qg, num_bytes, type);
3110 
3111 		list_for_each_entry(glist, &qg->groups, next_group) {
3112 			ret = ulist_add(fs_info->qgroup_ulist,
3113 					glist->group->qgroupid,
3114 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3115 			if (ret < 0)
3116 				goto out;
3117 		}
3118 	}
3119 
3120 out:
3121 	spin_unlock(&fs_info->qgroup_lock);
3122 }
3123 
3124 /*
3125  * Check if the leaf is the last leaf. Which means all node pointers
3126  * are at their last position.
3127  */
is_last_leaf(struct btrfs_path * path)3128 static bool is_last_leaf(struct btrfs_path *path)
3129 {
3130 	int i;
3131 
3132 	for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3133 		if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3134 			return false;
3135 	}
3136 	return true;
3137 }
3138 
3139 /*
3140  * returns < 0 on error, 0 when more leafs are to be scanned.
3141  * returns 1 when done.
3142  */
qgroup_rescan_leaf(struct btrfs_trans_handle * trans,struct btrfs_path * path)3143 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3144 			      struct btrfs_path *path)
3145 {
3146 	struct btrfs_fs_info *fs_info = trans->fs_info;
3147 	struct btrfs_key found;
3148 	struct extent_buffer *scratch_leaf = NULL;
3149 	struct ulist *roots = NULL;
3150 	u64 num_bytes;
3151 	bool done;
3152 	int slot;
3153 	int ret;
3154 
3155 	mutex_lock(&fs_info->qgroup_rescan_lock);
3156 	ret = btrfs_search_slot_for_read(fs_info->extent_root,
3157 					 &fs_info->qgroup_rescan_progress,
3158 					 path, 1, 0);
3159 
3160 	btrfs_debug(fs_info,
3161 		"current progress key (%llu %u %llu), search_slot ret %d",
3162 		fs_info->qgroup_rescan_progress.objectid,
3163 		fs_info->qgroup_rescan_progress.type,
3164 		fs_info->qgroup_rescan_progress.offset, ret);
3165 
3166 	if (ret) {
3167 		/*
3168 		 * The rescan is about to end, we will not be scanning any
3169 		 * further blocks. We cannot unset the RESCAN flag here, because
3170 		 * we want to commit the transaction if everything went well.
3171 		 * To make the live accounting work in this phase, we set our
3172 		 * scan progress pointer such that every real extent objectid
3173 		 * will be smaller.
3174 		 */
3175 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3176 		btrfs_release_path(path);
3177 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3178 		return ret;
3179 	}
3180 	done = is_last_leaf(path);
3181 
3182 	btrfs_item_key_to_cpu(path->nodes[0], &found,
3183 			      btrfs_header_nritems(path->nodes[0]) - 1);
3184 	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3185 
3186 	scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3187 	if (!scratch_leaf) {
3188 		ret = -ENOMEM;
3189 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3190 		goto out;
3191 	}
3192 	slot = path->slots[0];
3193 	btrfs_release_path(path);
3194 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3195 
3196 	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3197 		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3198 		if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3199 		    found.type != BTRFS_METADATA_ITEM_KEY)
3200 			continue;
3201 		if (found.type == BTRFS_METADATA_ITEM_KEY)
3202 			num_bytes = fs_info->nodesize;
3203 		else
3204 			num_bytes = found.offset;
3205 
3206 		ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3207 					   &roots, false);
3208 		if (ret < 0)
3209 			goto out;
3210 		/* For rescan, just pass old_roots as NULL */
3211 		ret = btrfs_qgroup_account_extent(trans, found.objectid,
3212 						  num_bytes, NULL, roots);
3213 		if (ret < 0)
3214 			goto out;
3215 	}
3216 out:
3217 	if (scratch_leaf)
3218 		free_extent_buffer(scratch_leaf);
3219 
3220 	if (done && !ret) {
3221 		ret = 1;
3222 		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3223 	}
3224 	return ret;
3225 }
3226 
btrfs_qgroup_rescan_worker(struct btrfs_work * work)3227 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3228 {
3229 	struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3230 						     qgroup_rescan_work);
3231 	struct btrfs_path *path;
3232 	struct btrfs_trans_handle *trans = NULL;
3233 	int err = -ENOMEM;
3234 	int ret = 0;
3235 
3236 	path = btrfs_alloc_path();
3237 	if (!path)
3238 		goto out;
3239 	/*
3240 	 * Rescan should only search for commit root, and any later difference
3241 	 * should be recorded by qgroup
3242 	 */
3243 	path->search_commit_root = 1;
3244 	path->skip_locking = 1;
3245 
3246 	err = 0;
3247 	while (!err && !btrfs_fs_closing(fs_info)) {
3248 		trans = btrfs_start_transaction(fs_info->fs_root, 0);
3249 		if (IS_ERR(trans)) {
3250 			err = PTR_ERR(trans);
3251 			break;
3252 		}
3253 		if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3254 			err = -EINTR;
3255 		} else {
3256 			err = qgroup_rescan_leaf(trans, path);
3257 		}
3258 		if (err > 0)
3259 			btrfs_commit_transaction(trans);
3260 		else
3261 			btrfs_end_transaction(trans);
3262 	}
3263 
3264 out:
3265 	btrfs_free_path(path);
3266 
3267 	mutex_lock(&fs_info->qgroup_rescan_lock);
3268 	if (err > 0 &&
3269 	    fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3270 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3271 	} else if (err < 0) {
3272 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3273 	}
3274 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3275 
3276 	/*
3277 	 * only update status, since the previous part has already updated the
3278 	 * qgroup info.
3279 	 */
3280 	trans = btrfs_start_transaction(fs_info->quota_root, 1);
3281 	if (IS_ERR(trans)) {
3282 		err = PTR_ERR(trans);
3283 		trans = NULL;
3284 		btrfs_err(fs_info,
3285 			  "fail to start transaction for status update: %d",
3286 			  err);
3287 	}
3288 
3289 	mutex_lock(&fs_info->qgroup_rescan_lock);
3290 	if (!btrfs_fs_closing(fs_info))
3291 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3292 	if (trans) {
3293 		ret = update_qgroup_status_item(trans);
3294 		if (ret < 0) {
3295 			err = ret;
3296 			btrfs_err(fs_info, "fail to update qgroup status: %d",
3297 				  err);
3298 		}
3299 	}
3300 	fs_info->qgroup_rescan_running = false;
3301 	complete_all(&fs_info->qgroup_rescan_completion);
3302 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3303 
3304 	if (!trans)
3305 		return;
3306 
3307 	btrfs_end_transaction(trans);
3308 
3309 	if (btrfs_fs_closing(fs_info)) {
3310 		btrfs_info(fs_info, "qgroup scan paused");
3311 	} else if (err >= 0) {
3312 		btrfs_info(fs_info, "qgroup scan completed%s",
3313 			err > 0 ? " (inconsistency flag cleared)" : "");
3314 	} else {
3315 		btrfs_err(fs_info, "qgroup scan failed with %d", err);
3316 	}
3317 }
3318 
3319 /*
3320  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3321  * memory required for the rescan context.
3322  */
3323 static int
qgroup_rescan_init(struct btrfs_fs_info * fs_info,u64 progress_objectid,int init_flags)3324 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3325 		   int init_flags)
3326 {
3327 	int ret = 0;
3328 
3329 	if (!init_flags) {
3330 		/* we're resuming qgroup rescan at mount time */
3331 		if (!(fs_info->qgroup_flags &
3332 		      BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3333 			btrfs_warn(fs_info,
3334 			"qgroup rescan init failed, qgroup rescan is not queued");
3335 			ret = -EINVAL;
3336 		} else if (!(fs_info->qgroup_flags &
3337 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3338 			btrfs_warn(fs_info,
3339 			"qgroup rescan init failed, qgroup is not enabled");
3340 			ret = -EINVAL;
3341 		}
3342 
3343 		if (ret)
3344 			return ret;
3345 	}
3346 
3347 	mutex_lock(&fs_info->qgroup_rescan_lock);
3348 
3349 	if (init_flags) {
3350 		if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3351 			btrfs_warn(fs_info,
3352 				   "qgroup rescan is already in progress");
3353 			ret = -EINPROGRESS;
3354 		} else if (!(fs_info->qgroup_flags &
3355 			     BTRFS_QGROUP_STATUS_FLAG_ON)) {
3356 			btrfs_warn(fs_info,
3357 			"qgroup rescan init failed, qgroup is not enabled");
3358 			ret = -EINVAL;
3359 		}
3360 
3361 		if (ret) {
3362 			mutex_unlock(&fs_info->qgroup_rescan_lock);
3363 			return ret;
3364 		}
3365 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3366 	}
3367 
3368 	memset(&fs_info->qgroup_rescan_progress, 0,
3369 		sizeof(fs_info->qgroup_rescan_progress));
3370 	fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3371 	init_completion(&fs_info->qgroup_rescan_completion);
3372 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3373 
3374 	btrfs_init_work(&fs_info->qgroup_rescan_work,
3375 			btrfs_qgroup_rescan_worker, NULL, NULL);
3376 	return 0;
3377 }
3378 
3379 static void
qgroup_rescan_zero_tracking(struct btrfs_fs_info * fs_info)3380 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3381 {
3382 	struct rb_node *n;
3383 	struct btrfs_qgroup *qgroup;
3384 
3385 	spin_lock(&fs_info->qgroup_lock);
3386 	/* clear all current qgroup tracking information */
3387 	for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3388 		qgroup = rb_entry(n, struct btrfs_qgroup, node);
3389 		qgroup->rfer = 0;
3390 		qgroup->rfer_cmpr = 0;
3391 		qgroup->excl = 0;
3392 		qgroup->excl_cmpr = 0;
3393 		qgroup_dirty(fs_info, qgroup);
3394 	}
3395 	spin_unlock(&fs_info->qgroup_lock);
3396 }
3397 
3398 int
btrfs_qgroup_rescan(struct btrfs_fs_info * fs_info)3399 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3400 {
3401 	int ret = 0;
3402 	struct btrfs_trans_handle *trans;
3403 
3404 	ret = qgroup_rescan_init(fs_info, 0, 1);
3405 	if (ret)
3406 		return ret;
3407 
3408 	/*
3409 	 * We have set the rescan_progress to 0, which means no more
3410 	 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3411 	 * However, btrfs_qgroup_account_ref may be right after its call
3412 	 * to btrfs_find_all_roots, in which case it would still do the
3413 	 * accounting.
3414 	 * To solve this, we're committing the transaction, which will
3415 	 * ensure we run all delayed refs and only after that, we are
3416 	 * going to clear all tracking information for a clean start.
3417 	 */
3418 
3419 	trans = btrfs_join_transaction(fs_info->fs_root);
3420 	if (IS_ERR(trans)) {
3421 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3422 		return PTR_ERR(trans);
3423 	}
3424 	ret = btrfs_commit_transaction(trans);
3425 	if (ret) {
3426 		fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3427 		return ret;
3428 	}
3429 
3430 	qgroup_rescan_zero_tracking(fs_info);
3431 
3432 	mutex_lock(&fs_info->qgroup_rescan_lock);
3433 	fs_info->qgroup_rescan_running = true;
3434 	btrfs_queue_work(fs_info->qgroup_rescan_workers,
3435 			 &fs_info->qgroup_rescan_work);
3436 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3437 
3438 	return 0;
3439 }
3440 
btrfs_qgroup_wait_for_completion(struct btrfs_fs_info * fs_info,bool interruptible)3441 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3442 				     bool interruptible)
3443 {
3444 	int running;
3445 	int ret = 0;
3446 
3447 	mutex_lock(&fs_info->qgroup_rescan_lock);
3448 	running = fs_info->qgroup_rescan_running;
3449 	mutex_unlock(&fs_info->qgroup_rescan_lock);
3450 
3451 	if (!running)
3452 		return 0;
3453 
3454 	if (interruptible)
3455 		ret = wait_for_completion_interruptible(
3456 					&fs_info->qgroup_rescan_completion);
3457 	else
3458 		wait_for_completion(&fs_info->qgroup_rescan_completion);
3459 
3460 	return ret;
3461 }
3462 
3463 /*
3464  * this is only called from open_ctree where we're still single threaded, thus
3465  * locking is omitted here.
3466  */
3467 void
btrfs_qgroup_rescan_resume(struct btrfs_fs_info * fs_info)3468 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3469 {
3470 	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3471 		mutex_lock(&fs_info->qgroup_rescan_lock);
3472 		fs_info->qgroup_rescan_running = true;
3473 		btrfs_queue_work(fs_info->qgroup_rescan_workers,
3474 				 &fs_info->qgroup_rescan_work);
3475 		mutex_unlock(&fs_info->qgroup_rescan_lock);
3476 	}
3477 }
3478 
3479 #define rbtree_iterate_from_safe(node, next, start)				\
3480        for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3481 
qgroup_unreserve_range(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len)3482 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3483 				  struct extent_changeset *reserved, u64 start,
3484 				  u64 len)
3485 {
3486 	struct rb_node *node;
3487 	struct rb_node *next;
3488 	struct ulist_node *entry;
3489 	int ret = 0;
3490 
3491 	node = reserved->range_changed.root.rb_node;
3492 	if (!node)
3493 		return 0;
3494 	while (node) {
3495 		entry = rb_entry(node, struct ulist_node, rb_node);
3496 		if (entry->val < start)
3497 			node = node->rb_right;
3498 		else
3499 			node = node->rb_left;
3500 	}
3501 
3502 	if (entry->val > start && rb_prev(&entry->rb_node))
3503 		entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3504 				 rb_node);
3505 
3506 	rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3507 		u64 entry_start;
3508 		u64 entry_end;
3509 		u64 entry_len;
3510 		int clear_ret;
3511 
3512 		entry = rb_entry(node, struct ulist_node, rb_node);
3513 		entry_start = entry->val;
3514 		entry_end = entry->aux;
3515 		entry_len = entry_end - entry_start + 1;
3516 
3517 		if (entry_start >= start + len)
3518 			break;
3519 		if (entry_start + entry_len <= start)
3520 			continue;
3521 		/*
3522 		 * Now the entry is in [start, start + len), revert the
3523 		 * EXTENT_QGROUP_RESERVED bit.
3524 		 */
3525 		clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3526 					      entry_end, EXTENT_QGROUP_RESERVED);
3527 		if (!ret && clear_ret < 0)
3528 			ret = clear_ret;
3529 
3530 		ulist_del(&reserved->range_changed, entry->val, entry->aux);
3531 		if (likely(reserved->bytes_changed >= entry_len)) {
3532 			reserved->bytes_changed -= entry_len;
3533 		} else {
3534 			WARN_ON(1);
3535 			reserved->bytes_changed = 0;
3536 		}
3537 	}
3538 
3539 	return ret;
3540 }
3541 
3542 /*
3543  * Try to free some space for qgroup.
3544  *
3545  * For qgroup, there are only 3 ways to free qgroup space:
3546  * - Flush nodatacow write
3547  *   Any nodatacow write will free its reserved data space at run_delalloc_range().
3548  *   In theory, we should only flush nodatacow inodes, but it's not yet
3549  *   possible, so we need to flush the whole root.
3550  *
3551  * - Wait for ordered extents
3552  *   When ordered extents are finished, their reserved metadata is finally
3553  *   converted to per_trans status, which can be freed by later commit
3554  *   transaction.
3555  *
3556  * - Commit transaction
3557  *   This would free the meta_per_trans space.
3558  *   In theory this shouldn't provide much space, but any more qgroup space
3559  *   is needed.
3560  */
try_flush_qgroup(struct btrfs_root * root)3561 static int try_flush_qgroup(struct btrfs_root *root)
3562 {
3563 	struct btrfs_trans_handle *trans;
3564 	int ret;
3565 	bool can_commit = true;
3566 
3567 	/*
3568 	 * We don't want to run flush again and again, so if there is a running
3569 	 * one, we won't try to start a new flush, but exit directly.
3570 	 */
3571 	if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3572 		wait_event(root->qgroup_flush_wait,
3573 			!test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3574 		return 0;
3575 	}
3576 
3577 	/*
3578 	 * If current process holds a transaction, we shouldn't flush, as we
3579 	 * assume all space reservation happens before a transaction handle is
3580 	 * held.
3581 	 *
3582 	 * But there are cases like btrfs_delayed_item_reserve_metadata() where
3583 	 * we try to reserve space with one transction handle already held.
3584 	 * In that case we can't commit transaction, but at least try to end it
3585 	 * and hope the started data writes can free some space.
3586 	 */
3587 	if (current->journal_info &&
3588 	    current->journal_info != BTRFS_SEND_TRANS_STUB)
3589 		can_commit = false;
3590 
3591 	ret = btrfs_start_delalloc_snapshot(root);
3592 	if (ret < 0)
3593 		goto out;
3594 	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3595 
3596 	trans = btrfs_join_transaction(root);
3597 	if (IS_ERR(trans)) {
3598 		ret = PTR_ERR(trans);
3599 		goto out;
3600 	}
3601 
3602 	if (can_commit)
3603 		ret = btrfs_commit_transaction(trans);
3604 	else
3605 		ret = btrfs_end_transaction(trans);
3606 out:
3607 	clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3608 	wake_up(&root->qgroup_flush_wait);
3609 	return ret;
3610 }
3611 
qgroup_reserve_data(struct btrfs_inode * inode,struct extent_changeset ** reserved_ret,u64 start,u64 len)3612 static int qgroup_reserve_data(struct btrfs_inode *inode,
3613 			struct extent_changeset **reserved_ret, u64 start,
3614 			u64 len)
3615 {
3616 	struct btrfs_root *root = inode->root;
3617 	struct extent_changeset *reserved;
3618 	bool new_reserved = false;
3619 	u64 orig_reserved;
3620 	u64 to_reserve;
3621 	int ret;
3622 
3623 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3624 	    !is_fstree(root->root_key.objectid) || len == 0)
3625 		return 0;
3626 
3627 	/* @reserved parameter is mandatory for qgroup */
3628 	if (WARN_ON(!reserved_ret))
3629 		return -EINVAL;
3630 	if (!*reserved_ret) {
3631 		new_reserved = true;
3632 		*reserved_ret = extent_changeset_alloc();
3633 		if (!*reserved_ret)
3634 			return -ENOMEM;
3635 	}
3636 	reserved = *reserved_ret;
3637 	/* Record already reserved space */
3638 	orig_reserved = reserved->bytes_changed;
3639 	ret = set_record_extent_bits(&inode->io_tree, start,
3640 			start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3641 
3642 	/* Newly reserved space */
3643 	to_reserve = reserved->bytes_changed - orig_reserved;
3644 	trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3645 					to_reserve, QGROUP_RESERVE);
3646 	if (ret < 0)
3647 		goto out;
3648 	ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3649 	if (ret < 0)
3650 		goto cleanup;
3651 
3652 	return ret;
3653 
3654 cleanup:
3655 	qgroup_unreserve_range(inode, reserved, start, len);
3656 out:
3657 	if (new_reserved) {
3658 		extent_changeset_release(reserved);
3659 		kfree(reserved);
3660 		*reserved_ret = NULL;
3661 	}
3662 	return ret;
3663 }
3664 
3665 /*
3666  * Reserve qgroup space for range [start, start + len).
3667  *
3668  * This function will either reserve space from related qgroups or do nothing
3669  * if the range is already reserved.
3670  *
3671  * Return 0 for successful reservation
3672  * Return <0 for error (including -EQUOT)
3673  *
3674  * NOTE: This function may sleep for memory allocation, dirty page flushing and
3675  *	 commit transaction. So caller should not hold any dirty page locked.
3676  */
btrfs_qgroup_reserve_data(struct btrfs_inode * inode,struct extent_changeset ** reserved_ret,u64 start,u64 len)3677 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3678 			struct extent_changeset **reserved_ret, u64 start,
3679 			u64 len)
3680 {
3681 	int ret;
3682 
3683 	ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3684 	if (ret <= 0 && ret != -EDQUOT)
3685 		return ret;
3686 
3687 	ret = try_flush_qgroup(inode->root);
3688 	if (ret < 0)
3689 		return ret;
3690 	return qgroup_reserve_data(inode, reserved_ret, start, len);
3691 }
3692 
3693 /* 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)3694 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3695 			struct extent_changeset *reserved, u64 start, u64 len)
3696 {
3697 	struct btrfs_root *root = inode->root;
3698 	struct ulist_node *unode;
3699 	struct ulist_iterator uiter;
3700 	struct extent_changeset changeset;
3701 	int freed = 0;
3702 	int ret;
3703 
3704 	extent_changeset_init(&changeset);
3705 	len = round_up(start + len, root->fs_info->sectorsize);
3706 	start = round_down(start, root->fs_info->sectorsize);
3707 
3708 	ULIST_ITER_INIT(&uiter);
3709 	while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3710 		u64 range_start = unode->val;
3711 		/* unode->aux is the inclusive end */
3712 		u64 range_len = unode->aux - range_start + 1;
3713 		u64 free_start;
3714 		u64 free_len;
3715 
3716 		extent_changeset_release(&changeset);
3717 
3718 		/* Only free range in range [start, start + len) */
3719 		if (range_start >= start + len ||
3720 		    range_start + range_len <= start)
3721 			continue;
3722 		free_start = max(range_start, start);
3723 		free_len = min(start + len, range_start + range_len) -
3724 			   free_start;
3725 		/*
3726 		 * TODO: To also modify reserved->ranges_reserved to reflect
3727 		 * the modification.
3728 		 *
3729 		 * However as long as we free qgroup reserved according to
3730 		 * EXTENT_QGROUP_RESERVED, we won't double free.
3731 		 * So not need to rush.
3732 		 */
3733 		ret = clear_record_extent_bits(&inode->io_tree, free_start,
3734 				free_start + free_len - 1,
3735 				EXTENT_QGROUP_RESERVED, &changeset);
3736 		if (ret < 0)
3737 			goto out;
3738 		freed += changeset.bytes_changed;
3739 	}
3740 	btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3741 				  BTRFS_QGROUP_RSV_DATA);
3742 	ret = freed;
3743 out:
3744 	extent_changeset_release(&changeset);
3745 	return ret;
3746 }
3747 
__btrfs_qgroup_release_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len,int free)3748 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3749 			struct extent_changeset *reserved, u64 start, u64 len,
3750 			int free)
3751 {
3752 	struct extent_changeset changeset;
3753 	int trace_op = QGROUP_RELEASE;
3754 	int ret;
3755 
3756 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3757 		return 0;
3758 
3759 	/* In release case, we shouldn't have @reserved */
3760 	WARN_ON(!free && reserved);
3761 	if (free && reserved)
3762 		return qgroup_free_reserved_data(inode, reserved, start, len);
3763 	extent_changeset_init(&changeset);
3764 	ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3765 				       EXTENT_QGROUP_RESERVED, &changeset);
3766 	if (ret < 0)
3767 		goto out;
3768 
3769 	if (free)
3770 		trace_op = QGROUP_FREE;
3771 	trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3772 					changeset.bytes_changed, trace_op);
3773 	if (free)
3774 		btrfs_qgroup_free_refroot(inode->root->fs_info,
3775 				inode->root->root_key.objectid,
3776 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3777 	ret = changeset.bytes_changed;
3778 out:
3779 	extent_changeset_release(&changeset);
3780 	return ret;
3781 }
3782 
3783 /*
3784  * Free a reserved space range from io_tree and related qgroups
3785  *
3786  * Should be called when a range of pages get invalidated before reaching disk.
3787  * Or for error cleanup case.
3788  * if @reserved is given, only reserved range in [@start, @start + @len) will
3789  * be freed.
3790  *
3791  * For data written to disk, use btrfs_qgroup_release_data().
3792  *
3793  * NOTE: This function may sleep for memory allocation.
3794  */
btrfs_qgroup_free_data(struct btrfs_inode * inode,struct extent_changeset * reserved,u64 start,u64 len)3795 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3796 			struct extent_changeset *reserved, u64 start, u64 len)
3797 {
3798 	return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3799 }
3800 
3801 /*
3802  * Release a reserved space range from io_tree only.
3803  *
3804  * Should be called when a range of pages get written to disk and corresponding
3805  * FILE_EXTENT is inserted into corresponding root.
3806  *
3807  * Since new qgroup accounting framework will only update qgroup numbers at
3808  * commit_transaction() time, its reserved space shouldn't be freed from
3809  * related qgroups.
3810  *
3811  * But we should release the range from io_tree, to allow further write to be
3812  * COWed.
3813  *
3814  * NOTE: This function may sleep for memory allocation.
3815  */
btrfs_qgroup_release_data(struct btrfs_inode * inode,u64 start,u64 len)3816 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3817 {
3818 	return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3819 }
3820 
add_root_meta_rsv(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)3821 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3822 			      enum btrfs_qgroup_rsv_type type)
3823 {
3824 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3825 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3826 		return;
3827 	if (num_bytes == 0)
3828 		return;
3829 
3830 	spin_lock(&root->qgroup_meta_rsv_lock);
3831 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3832 		root->qgroup_meta_rsv_prealloc += num_bytes;
3833 	else
3834 		root->qgroup_meta_rsv_pertrans += num_bytes;
3835 	spin_unlock(&root->qgroup_meta_rsv_lock);
3836 }
3837 
sub_root_meta_rsv(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)3838 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3839 			     enum btrfs_qgroup_rsv_type type)
3840 {
3841 	if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3842 	    type != BTRFS_QGROUP_RSV_META_PERTRANS)
3843 		return 0;
3844 	if (num_bytes == 0)
3845 		return 0;
3846 
3847 	spin_lock(&root->qgroup_meta_rsv_lock);
3848 	if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3849 		num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3850 				  num_bytes);
3851 		root->qgroup_meta_rsv_prealloc -= num_bytes;
3852 	} else {
3853 		num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3854 				  num_bytes);
3855 		root->qgroup_meta_rsv_pertrans -= num_bytes;
3856 	}
3857 	spin_unlock(&root->qgroup_meta_rsv_lock);
3858 	return num_bytes;
3859 }
3860 
qgroup_reserve_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type,bool enforce)3861 static int qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3862 				enum btrfs_qgroup_rsv_type type, bool enforce)
3863 {
3864 	struct btrfs_fs_info *fs_info = root->fs_info;
3865 	int ret;
3866 
3867 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3868 	    !is_fstree(root->root_key.objectid) || num_bytes == 0)
3869 		return 0;
3870 
3871 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3872 	trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3873 	ret = qgroup_reserve(root, num_bytes, enforce, type);
3874 	if (ret < 0)
3875 		return ret;
3876 	/*
3877 	 * Record what we have reserved into root.
3878 	 *
3879 	 * To avoid quota disabled->enabled underflow.
3880 	 * In that case, we may try to free space we haven't reserved
3881 	 * (since quota was disabled), so record what we reserved into root.
3882 	 * And ensure later release won't underflow this number.
3883 	 */
3884 	add_root_meta_rsv(root, num_bytes, type);
3885 	return ret;
3886 }
3887 
__btrfs_qgroup_reserve_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type,bool enforce)3888 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3889 				enum btrfs_qgroup_rsv_type type, bool enforce)
3890 {
3891 	int ret;
3892 
3893 	ret = qgroup_reserve_meta(root, num_bytes, type, enforce);
3894 	if (ret <= 0 && ret != -EDQUOT)
3895 		return ret;
3896 
3897 	ret = try_flush_qgroup(root);
3898 	if (ret < 0)
3899 		return ret;
3900 	return qgroup_reserve_meta(root, num_bytes, type, enforce);
3901 }
3902 
btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root * root)3903 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3904 {
3905 	struct btrfs_fs_info *fs_info = root->fs_info;
3906 
3907 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3908 	    !is_fstree(root->root_key.objectid))
3909 		return;
3910 
3911 	/* TODO: Update trace point to handle such free */
3912 	trace_qgroup_meta_free_all_pertrans(root);
3913 	/* Special value -1 means to free all reserved space */
3914 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3915 				  BTRFS_QGROUP_RSV_META_PERTRANS);
3916 }
3917 
__btrfs_qgroup_free_meta(struct btrfs_root * root,int num_bytes,enum btrfs_qgroup_rsv_type type)3918 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3919 			      enum btrfs_qgroup_rsv_type type)
3920 {
3921 	struct btrfs_fs_info *fs_info = root->fs_info;
3922 
3923 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3924 	    !is_fstree(root->root_key.objectid))
3925 		return;
3926 
3927 	/*
3928 	 * reservation for META_PREALLOC can happen before quota is enabled,
3929 	 * which can lead to underflow.
3930 	 * Here ensure we will only free what we really have reserved.
3931 	 */
3932 	num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3933 	BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3934 	trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3935 	btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3936 				  num_bytes, type);
3937 }
3938 
qgroup_convert_meta(struct btrfs_fs_info * fs_info,u64 ref_root,int num_bytes)3939 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3940 				int num_bytes)
3941 {
3942 	struct btrfs_qgroup *qgroup;
3943 	struct ulist_node *unode;
3944 	struct ulist_iterator uiter;
3945 	int ret = 0;
3946 
3947 	if (num_bytes == 0)
3948 		return;
3949 	if (!fs_info->quota_root)
3950 		return;
3951 
3952 	spin_lock(&fs_info->qgroup_lock);
3953 	qgroup = find_qgroup_rb(fs_info, ref_root);
3954 	if (!qgroup)
3955 		goto out;
3956 	ulist_reinit(fs_info->qgroup_ulist);
3957 	ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3958 		       qgroup_to_aux(qgroup), GFP_ATOMIC);
3959 	if (ret < 0)
3960 		goto out;
3961 	ULIST_ITER_INIT(&uiter);
3962 	while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3963 		struct btrfs_qgroup *qg;
3964 		struct btrfs_qgroup_list *glist;
3965 
3966 		qg = unode_aux_to_qgroup(unode);
3967 
3968 		qgroup_rsv_release(fs_info, qg, num_bytes,
3969 				BTRFS_QGROUP_RSV_META_PREALLOC);
3970 		qgroup_rsv_add(fs_info, qg, num_bytes,
3971 				BTRFS_QGROUP_RSV_META_PERTRANS);
3972 		list_for_each_entry(glist, &qg->groups, next_group) {
3973 			ret = ulist_add(fs_info->qgroup_ulist,
3974 					glist->group->qgroupid,
3975 					qgroup_to_aux(glist->group), GFP_ATOMIC);
3976 			if (ret < 0)
3977 				goto out;
3978 		}
3979 	}
3980 out:
3981 	spin_unlock(&fs_info->qgroup_lock);
3982 }
3983 
btrfs_qgroup_convert_reserved_meta(struct btrfs_root * root,int num_bytes)3984 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3985 {
3986 	struct btrfs_fs_info *fs_info = root->fs_info;
3987 
3988 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3989 	    !is_fstree(root->root_key.objectid))
3990 		return;
3991 	/* Same as btrfs_qgroup_free_meta_prealloc() */
3992 	num_bytes = sub_root_meta_rsv(root, num_bytes,
3993 				      BTRFS_QGROUP_RSV_META_PREALLOC);
3994 	trace_qgroup_meta_convert(root, num_bytes);
3995 	qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3996 }
3997 
3998 /*
3999  * Check qgroup reserved space leaking, normally at destroy inode
4000  * time
4001  */
btrfs_qgroup_check_reserved_leak(struct btrfs_inode * inode)4002 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4003 {
4004 	struct extent_changeset changeset;
4005 	struct ulist_node *unode;
4006 	struct ulist_iterator iter;
4007 	int ret;
4008 
4009 	extent_changeset_init(&changeset);
4010 	ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4011 			EXTENT_QGROUP_RESERVED, &changeset);
4012 
4013 	WARN_ON(ret < 0);
4014 	if (WARN_ON(changeset.bytes_changed)) {
4015 		ULIST_ITER_INIT(&iter);
4016 		while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4017 			btrfs_warn(inode->root->fs_info,
4018 		"leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4019 				btrfs_ino(inode), unode->val, unode->aux);
4020 		}
4021 		btrfs_qgroup_free_refroot(inode->root->fs_info,
4022 				inode->root->root_key.objectid,
4023 				changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4024 
4025 	}
4026 	extent_changeset_release(&changeset);
4027 }
4028 
btrfs_qgroup_init_swapped_blocks(struct btrfs_qgroup_swapped_blocks * swapped_blocks)4029 void btrfs_qgroup_init_swapped_blocks(
4030 	struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4031 {
4032 	int i;
4033 
4034 	spin_lock_init(&swapped_blocks->lock);
4035 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4036 		swapped_blocks->blocks[i] = RB_ROOT;
4037 	swapped_blocks->swapped = false;
4038 }
4039 
4040 /*
4041  * Delete all swapped blocks record of @root.
4042  * Every record here means we skipped a full subtree scan for qgroup.
4043  *
4044  * Gets called when committing one transaction.
4045  */
btrfs_qgroup_clean_swapped_blocks(struct btrfs_root * root)4046 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4047 {
4048 	struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4049 	int i;
4050 
4051 	swapped_blocks = &root->swapped_blocks;
4052 
4053 	spin_lock(&swapped_blocks->lock);
4054 	if (!swapped_blocks->swapped)
4055 		goto out;
4056 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4057 		struct rb_root *cur_root = &swapped_blocks->blocks[i];
4058 		struct btrfs_qgroup_swapped_block *entry;
4059 		struct btrfs_qgroup_swapped_block *next;
4060 
4061 		rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4062 						     node)
4063 			kfree(entry);
4064 		swapped_blocks->blocks[i] = RB_ROOT;
4065 	}
4066 	swapped_blocks->swapped = false;
4067 out:
4068 	spin_unlock(&swapped_blocks->lock);
4069 }
4070 
4071 /*
4072  * Add subtree roots record into @subvol_root.
4073  *
4074  * @subvol_root:	tree root of the subvolume tree get swapped
4075  * @bg:			block group under balance
4076  * @subvol_parent/slot:	pointer to the subtree root in subvolume tree
4077  * @reloc_parent/slot:	pointer to the subtree root in reloc tree
4078  *			BOTH POINTERS ARE BEFORE TREE SWAP
4079  * @last_snapshot:	last snapshot generation of the subvolume tree
4080  */
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)4081 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4082 		struct btrfs_root *subvol_root,
4083 		struct btrfs_block_group *bg,
4084 		struct extent_buffer *subvol_parent, int subvol_slot,
4085 		struct extent_buffer *reloc_parent, int reloc_slot,
4086 		u64 last_snapshot)
4087 {
4088 	struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4089 	struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4090 	struct btrfs_qgroup_swapped_block *block;
4091 	struct rb_node **cur;
4092 	struct rb_node *parent = NULL;
4093 	int level = btrfs_header_level(subvol_parent) - 1;
4094 	int ret = 0;
4095 
4096 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4097 		return 0;
4098 
4099 	if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4100 	    btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4101 		btrfs_err_rl(fs_info,
4102 		"%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4103 			__func__,
4104 			btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4105 			btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4106 		return -EUCLEAN;
4107 	}
4108 
4109 	block = kmalloc(sizeof(*block), GFP_NOFS);
4110 	if (!block) {
4111 		ret = -ENOMEM;
4112 		goto out;
4113 	}
4114 
4115 	/*
4116 	 * @reloc_parent/slot is still before swap, while @block is going to
4117 	 * record the bytenr after swap, so we do the swap here.
4118 	 */
4119 	block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4120 	block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4121 							     reloc_slot);
4122 	block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4123 	block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4124 							    subvol_slot);
4125 	block->last_snapshot = last_snapshot;
4126 	block->level = level;
4127 
4128 	/*
4129 	 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4130 	 * no one else can modify tree blocks thus we qgroup will not change
4131 	 * no matter the value of trace_leaf.
4132 	 */
4133 	if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4134 		block->trace_leaf = true;
4135 	else
4136 		block->trace_leaf = false;
4137 	btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4138 
4139 	/* Insert @block into @blocks */
4140 	spin_lock(&blocks->lock);
4141 	cur = &blocks->blocks[level].rb_node;
4142 	while (*cur) {
4143 		struct btrfs_qgroup_swapped_block *entry;
4144 
4145 		parent = *cur;
4146 		entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4147 				 node);
4148 
4149 		if (entry->subvol_bytenr < block->subvol_bytenr) {
4150 			cur = &(*cur)->rb_left;
4151 		} else if (entry->subvol_bytenr > block->subvol_bytenr) {
4152 			cur = &(*cur)->rb_right;
4153 		} else {
4154 			if (entry->subvol_generation !=
4155 					block->subvol_generation ||
4156 			    entry->reloc_bytenr != block->reloc_bytenr ||
4157 			    entry->reloc_generation !=
4158 					block->reloc_generation) {
4159 				/*
4160 				 * Duplicated but mismatch entry found.
4161 				 * Shouldn't happen.
4162 				 *
4163 				 * Marking qgroup inconsistent should be enough
4164 				 * for end users.
4165 				 */
4166 				WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4167 				ret = -EEXIST;
4168 			}
4169 			kfree(block);
4170 			goto out_unlock;
4171 		}
4172 	}
4173 	rb_link_node(&block->node, parent, cur);
4174 	rb_insert_color(&block->node, &blocks->blocks[level]);
4175 	blocks->swapped = true;
4176 out_unlock:
4177 	spin_unlock(&blocks->lock);
4178 out:
4179 	if (ret < 0)
4180 		fs_info->qgroup_flags |=
4181 			BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4182 	return ret;
4183 }
4184 
4185 /*
4186  * Check if the tree block is a subtree root, and if so do the needed
4187  * delayed subtree trace for qgroup.
4188  *
4189  * This is called during btrfs_cow_block().
4190  */
btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * subvol_eb)4191 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4192 					 struct btrfs_root *root,
4193 					 struct extent_buffer *subvol_eb)
4194 {
4195 	struct btrfs_fs_info *fs_info = root->fs_info;
4196 	struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4197 	struct btrfs_qgroup_swapped_block *block;
4198 	struct extent_buffer *reloc_eb = NULL;
4199 	struct rb_node *node;
4200 	bool found = false;
4201 	bool swapped = false;
4202 	int level = btrfs_header_level(subvol_eb);
4203 	int ret = 0;
4204 	int i;
4205 
4206 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4207 		return 0;
4208 	if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4209 		return 0;
4210 
4211 	spin_lock(&blocks->lock);
4212 	if (!blocks->swapped) {
4213 		spin_unlock(&blocks->lock);
4214 		return 0;
4215 	}
4216 	node = blocks->blocks[level].rb_node;
4217 
4218 	while (node) {
4219 		block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4220 		if (block->subvol_bytenr < subvol_eb->start) {
4221 			node = node->rb_left;
4222 		} else if (block->subvol_bytenr > subvol_eb->start) {
4223 			node = node->rb_right;
4224 		} else {
4225 			found = true;
4226 			break;
4227 		}
4228 	}
4229 	if (!found) {
4230 		spin_unlock(&blocks->lock);
4231 		goto out;
4232 	}
4233 	/* Found one, remove it from @blocks first and update blocks->swapped */
4234 	rb_erase(&block->node, &blocks->blocks[level]);
4235 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4236 		if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4237 			swapped = true;
4238 			break;
4239 		}
4240 	}
4241 	blocks->swapped = swapped;
4242 	spin_unlock(&blocks->lock);
4243 
4244 	/* Read out reloc subtree root */
4245 	reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
4246 				   block->reloc_generation, block->level,
4247 				   &block->first_key);
4248 	if (IS_ERR(reloc_eb)) {
4249 		ret = PTR_ERR(reloc_eb);
4250 		reloc_eb = NULL;
4251 		goto free_out;
4252 	}
4253 	if (!extent_buffer_uptodate(reloc_eb)) {
4254 		ret = -EIO;
4255 		goto free_out;
4256 	}
4257 
4258 	ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4259 			block->last_snapshot, block->trace_leaf);
4260 free_out:
4261 	kfree(block);
4262 	free_extent_buffer(reloc_eb);
4263 out:
4264 	if (ret < 0) {
4265 		btrfs_err_rl(fs_info,
4266 			     "failed to account subtree at bytenr %llu: %d",
4267 			     subvol_eb->start, ret);
4268 		fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4269 	}
4270 	return ret;
4271 }
4272 
btrfs_qgroup_destroy_extent_records(struct btrfs_transaction * trans)4273 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4274 {
4275 	struct btrfs_qgroup_extent_record *entry;
4276 	struct btrfs_qgroup_extent_record *next;
4277 	struct rb_root *root;
4278 
4279 	root = &trans->delayed_refs.dirty_extent_root;
4280 	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4281 		ulist_free(entry->old_roots);
4282 		kfree(entry);
4283 	}
4284 }
4285