ad041668ee3fd8e8ee595d06f17c0f2ba4938263
[sfrench/cifs-2.6.git] / fs / btrfs / qgroup.c
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/sizes.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
25
26 /* TODO XXX FIXME
27  *  - subvol delete -> delete when ref goes to 0? delete limits also?
28  *  - reorganize keys
29  *  - compressed
30  *  - sync
31  *  - copy also limits on subvol creation
32  *  - limit
33  *  - caches for ulists
34  *  - performance benchmarks
35  *  - check all ioctl parameters
36  */
37
38 /*
39  * Helpers to access qgroup reservation
40  *
41  * Callers should ensure the lock context and type are valid
42  */
43
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
45 {
46         u64 ret = 0;
47         int i;
48
49         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50                 ret += qgroup->rsv.values[i];
51
52         return ret;
53 }
54
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57 {
58         if (type == BTRFS_QGROUP_RSV_DATA)
59                 return "data";
60         if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61                 return "meta_pertrans";
62         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63                 return "meta_prealloc";
64         return NULL;
65 }
66 #endif
67
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69                            struct btrfs_qgroup *qgroup, u64 num_bytes,
70                            enum btrfs_qgroup_rsv_type type)
71 {
72         trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73         qgroup->rsv.values[type] += num_bytes;
74 }
75
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77                                struct btrfs_qgroup *qgroup, u64 num_bytes,
78                                enum btrfs_qgroup_rsv_type type)
79 {
80         trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81         if (qgroup->rsv.values[type] >= num_bytes) {
82                 qgroup->rsv.values[type] -= num_bytes;
83                 return;
84         }
85 #ifdef CONFIG_BTRFS_DEBUG
86         WARN_RATELIMIT(1,
87                 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
88                 qgroup->qgroupid, qgroup_rsv_type_str(type),
89                 qgroup->rsv.values[type], num_bytes);
90 #endif
91         qgroup->rsv.values[type] = 0;
92 }
93
94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95                                      struct btrfs_qgroup *dest,
96                                      struct btrfs_qgroup *src)
97 {
98         int i;
99
100         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101                 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
102 }
103
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105                                          struct btrfs_qgroup *dest,
106                                           struct btrfs_qgroup *src)
107 {
108         int i;
109
110         for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111                 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
112 }
113
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
115                                            int mod)
116 {
117         if (qg->old_refcnt < seq)
118                 qg->old_refcnt = seq;
119         qg->old_refcnt += mod;
120 }
121
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
123                                            int mod)
124 {
125         if (qg->new_refcnt < seq)
126                 qg->new_refcnt = seq;
127         qg->new_refcnt += mod;
128 }
129
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131 {
132         if (qg->old_refcnt < seq)
133                 return 0;
134         return qg->old_refcnt - seq;
135 }
136
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138 {
139         if (qg->new_refcnt < seq)
140                 return 0;
141         return qg->new_refcnt - seq;
142 }
143
144 /*
145  * glue structure to represent the relations between qgroups.
146  */
147 struct btrfs_qgroup_list {
148         struct list_head next_group;
149         struct list_head next_member;
150         struct btrfs_qgroup *group;
151         struct btrfs_qgroup *member;
152 };
153
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 {
156         return (u64)(uintptr_t)qg;
157 }
158
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 {
161         return (struct btrfs_qgroup *)(uintptr_t)n->aux;
162 }
163
164 static int
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166                    int init_flags);
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
171                                            u64 qgroupid)
172 {
173         struct rb_node *n = fs_info->qgroup_tree.rb_node;
174         struct btrfs_qgroup *qgroup;
175
176         while (n) {
177                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
178                 if (qgroup->qgroupid < qgroupid)
179                         n = n->rb_left;
180                 else if (qgroup->qgroupid > qgroupid)
181                         n = n->rb_right;
182                 else
183                         return qgroup;
184         }
185         return NULL;
186 }
187
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
190                                           u64 qgroupid)
191 {
192         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193         struct rb_node *parent = NULL;
194         struct btrfs_qgroup *qgroup;
195
196         while (*p) {
197                 parent = *p;
198                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199
200                 if (qgroup->qgroupid < qgroupid)
201                         p = &(*p)->rb_left;
202                 else if (qgroup->qgroupid > qgroupid)
203                         p = &(*p)->rb_right;
204                 else
205                         return qgroup;
206         }
207
208         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209         if (!qgroup)
210                 return ERR_PTR(-ENOMEM);
211
212         qgroup->qgroupid = qgroupid;
213         INIT_LIST_HEAD(&qgroup->groups);
214         INIT_LIST_HEAD(&qgroup->members);
215         INIT_LIST_HEAD(&qgroup->dirty);
216
217         rb_link_node(&qgroup->node, parent, p);
218         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
219
220         return qgroup;
221 }
222
223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
224 {
225         struct btrfs_qgroup_list *list;
226
227         list_del(&qgroup->dirty);
228         while (!list_empty(&qgroup->groups)) {
229                 list = list_first_entry(&qgroup->groups,
230                                         struct btrfs_qgroup_list, next_group);
231                 list_del(&list->next_group);
232                 list_del(&list->next_member);
233                 kfree(list);
234         }
235
236         while (!list_empty(&qgroup->members)) {
237                 list = list_first_entry(&qgroup->members,
238                                         struct btrfs_qgroup_list, next_member);
239                 list_del(&list->next_group);
240                 list_del(&list->next_member);
241                 kfree(list);
242         }
243         kfree(qgroup);
244 }
245
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248 {
249         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
250
251         if (!qgroup)
252                 return -ENOENT;
253
254         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255         __del_qgroup_rb(qgroup);
256         return 0;
257 }
258
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261                            u64 memberid, u64 parentid)
262 {
263         struct btrfs_qgroup *member;
264         struct btrfs_qgroup *parent;
265         struct btrfs_qgroup_list *list;
266
267         member = find_qgroup_rb(fs_info, memberid);
268         parent = find_qgroup_rb(fs_info, parentid);
269         if (!member || !parent)
270                 return -ENOENT;
271
272         list = kzalloc(sizeof(*list), GFP_ATOMIC);
273         if (!list)
274                 return -ENOMEM;
275
276         list->group = parent;
277         list->member = member;
278         list_add_tail(&list->next_group, &member->groups);
279         list_add_tail(&list->next_member, &parent->members);
280
281         return 0;
282 }
283
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286                            u64 memberid, u64 parentid)
287 {
288         struct btrfs_qgroup *member;
289         struct btrfs_qgroup *parent;
290         struct btrfs_qgroup_list *list;
291
292         member = find_qgroup_rb(fs_info, memberid);
293         parent = find_qgroup_rb(fs_info, parentid);
294         if (!member || !parent)
295                 return -ENOENT;
296
297         list_for_each_entry(list, &member->groups, next_group) {
298                 if (list->group == parent) {
299                         list_del(&list->next_group);
300                         list_del(&list->next_member);
301                         kfree(list);
302                         return 0;
303                 }
304         }
305         return -ENOENT;
306 }
307
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
310                                u64 rfer, u64 excl)
311 {
312         struct btrfs_qgroup *qgroup;
313
314         qgroup = find_qgroup_rb(fs_info, qgroupid);
315         if (!qgroup)
316                 return -EINVAL;
317         if (qgroup->rfer != rfer || qgroup->excl != excl)
318                 return -EINVAL;
319         return 0;
320 }
321 #endif
322
323 /*
324  * The full config is read in one go, only called from open_ctree()
325  * It doesn't use any locking, as at this point we're still single-threaded
326  */
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328 {
329         struct btrfs_key key;
330         struct btrfs_key found_key;
331         struct btrfs_root *quota_root = fs_info->quota_root;
332         struct btrfs_path *path = NULL;
333         struct extent_buffer *l;
334         int slot;
335         int ret = 0;
336         u64 flags = 0;
337         u64 rescan_progress = 0;
338
339         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
340                 return 0;
341
342         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343         if (!fs_info->qgroup_ulist) {
344                 ret = -ENOMEM;
345                 goto out;
346         }
347
348         path = btrfs_alloc_path();
349         if (!path) {
350                 ret = -ENOMEM;
351                 goto out;
352         }
353
354         /* default this to quota off, in case no status key is found */
355         fs_info->qgroup_flags = 0;
356
357         /*
358          * pass 1: read status, all qgroup infos and limits
359          */
360         key.objectid = 0;
361         key.type = 0;
362         key.offset = 0;
363         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
364         if (ret)
365                 goto out;
366
367         while (1) {
368                 struct btrfs_qgroup *qgroup;
369
370                 slot = path->slots[0];
371                 l = path->nodes[0];
372                 btrfs_item_key_to_cpu(l, &found_key, slot);
373
374                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375                         struct btrfs_qgroup_status_item *ptr;
376
377                         ptr = btrfs_item_ptr(l, slot,
378                                              struct btrfs_qgroup_status_item);
379
380                         if (btrfs_qgroup_status_version(l, ptr) !=
381                             BTRFS_QGROUP_STATUS_VERSION) {
382                                 btrfs_err(fs_info,
383                                  "old qgroup version, quota disabled");
384                                 goto out;
385                         }
386                         if (btrfs_qgroup_status_generation(l, ptr) !=
387                             fs_info->generation) {
388                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389                                 btrfs_err(fs_info,
390                                         "qgroup generation mismatch, marked as inconsistent");
391                         }
392                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393                                                                           ptr);
394                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
395                         goto next1;
396                 }
397
398                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
400                         goto next1;
401
402                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
403                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405                         btrfs_err(fs_info, "inconsistent qgroup config");
406                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
407                 }
408                 if (!qgroup) {
409                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
410                         if (IS_ERR(qgroup)) {
411                                 ret = PTR_ERR(qgroup);
412                                 goto out;
413                         }
414                 }
415                 switch (found_key.type) {
416                 case BTRFS_QGROUP_INFO_KEY: {
417                         struct btrfs_qgroup_info_item *ptr;
418
419                         ptr = btrfs_item_ptr(l, slot,
420                                              struct btrfs_qgroup_info_item);
421                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425                         /* generation currently unused */
426                         break;
427                 }
428                 case BTRFS_QGROUP_LIMIT_KEY: {
429                         struct btrfs_qgroup_limit_item *ptr;
430
431                         ptr = btrfs_item_ptr(l, slot,
432                                              struct btrfs_qgroup_limit_item);
433                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
438                         break;
439                 }
440                 }
441 next1:
442                 ret = btrfs_next_item(quota_root, path);
443                 if (ret < 0)
444                         goto out;
445                 if (ret)
446                         break;
447         }
448         btrfs_release_path(path);
449
450         /*
451          * pass 2: read all qgroup relations
452          */
453         key.objectid = 0;
454         key.type = BTRFS_QGROUP_RELATION_KEY;
455         key.offset = 0;
456         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
457         if (ret)
458                 goto out;
459         while (1) {
460                 slot = path->slots[0];
461                 l = path->nodes[0];
462                 btrfs_item_key_to_cpu(l, &found_key, slot);
463
464                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
465                         goto next2;
466
467                 if (found_key.objectid > found_key.offset) {
468                         /* parent <- member, not needed to build config */
469                         /* FIXME should we omit the key completely? */
470                         goto next2;
471                 }
472
473                 ret = add_relation_rb(fs_info, found_key.objectid,
474                                       found_key.offset);
475                 if (ret == -ENOENT) {
476                         btrfs_warn(fs_info,
477                                 "orphan qgroup relation 0x%llx->0x%llx",
478                                 found_key.objectid, found_key.offset);
479                         ret = 0;        /* ignore the error */
480                 }
481                 if (ret)
482                         goto out;
483 next2:
484                 ret = btrfs_next_item(quota_root, path);
485                 if (ret < 0)
486                         goto out;
487                 if (ret)
488                         break;
489         }
490 out:
491         fs_info->qgroup_flags |= flags;
492         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493                 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494         else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495                  ret >= 0)
496                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497         btrfs_free_path(path);
498
499         if (ret < 0) {
500                 ulist_free(fs_info->qgroup_ulist);
501                 fs_info->qgroup_ulist = NULL;
502                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
503         }
504
505         return ret < 0 ? ret : 0;
506 }
507
508 /*
509  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510  * first two are in single-threaded paths.And for the third one, we have set
511  * quota_root to be null with qgroup_lock held before, so it is safe to clean
512  * up the in-memory structures without qgroup_lock held.
513  */
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
515 {
516         struct rb_node *n;
517         struct btrfs_qgroup *qgroup;
518
519         while ((n = rb_first(&fs_info->qgroup_tree))) {
520                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
521                 rb_erase(n, &fs_info->qgroup_tree);
522                 __del_qgroup_rb(qgroup);
523         }
524         /*
525          * We call btrfs_free_qgroup_config() when unmounting
526          * filesystem and disabling quota, so we set qgroup_ulist
527          * to be null here to avoid double free.
528          */
529         ulist_free(fs_info->qgroup_ulist);
530         fs_info->qgroup_ulist = NULL;
531 }
532
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
534                                     u64 dst)
535 {
536         int ret;
537         struct btrfs_root *quota_root = trans->fs_info->quota_root;
538         struct btrfs_path *path;
539         struct btrfs_key key;
540
541         path = btrfs_alloc_path();
542         if (!path)
543                 return -ENOMEM;
544
545         key.objectid = src;
546         key.type = BTRFS_QGROUP_RELATION_KEY;
547         key.offset = dst;
548
549         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550
551         btrfs_mark_buffer_dirty(path->nodes[0]);
552
553         btrfs_free_path(path);
554         return ret;
555 }
556
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
558                                     u64 dst)
559 {
560         int ret;
561         struct btrfs_root *quota_root = trans->fs_info->quota_root;
562         struct btrfs_path *path;
563         struct btrfs_key key;
564
565         path = btrfs_alloc_path();
566         if (!path)
567                 return -ENOMEM;
568
569         key.objectid = src;
570         key.type = BTRFS_QGROUP_RELATION_KEY;
571         key.offset = dst;
572
573         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
574         if (ret < 0)
575                 goto out;
576
577         if (ret > 0) {
578                 ret = -ENOENT;
579                 goto out;
580         }
581
582         ret = btrfs_del_item(trans, quota_root, path);
583 out:
584         btrfs_free_path(path);
585         return ret;
586 }
587
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589                            struct btrfs_root *quota_root, u64 qgroupid)
590 {
591         int ret;
592         struct btrfs_path *path;
593         struct btrfs_qgroup_info_item *qgroup_info;
594         struct btrfs_qgroup_limit_item *qgroup_limit;
595         struct extent_buffer *leaf;
596         struct btrfs_key key;
597
598         if (btrfs_is_testing(quota_root->fs_info))
599                 return 0;
600
601         path = btrfs_alloc_path();
602         if (!path)
603                 return -ENOMEM;
604
605         key.objectid = 0;
606         key.type = BTRFS_QGROUP_INFO_KEY;
607         key.offset = qgroupid;
608
609         /*
610          * Avoid a transaction abort by catching -EEXIST here. In that
611          * case, we proceed by re-initializing the existing structure
612          * on disk.
613          */
614
615         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616                                       sizeof(*qgroup_info));
617         if (ret && ret != -EEXIST)
618                 goto out;
619
620         leaf = path->nodes[0];
621         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622                                  struct btrfs_qgroup_info_item);
623         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628
629         btrfs_mark_buffer_dirty(leaf);
630
631         btrfs_release_path(path);
632
633         key.type = BTRFS_QGROUP_LIMIT_KEY;
634         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635                                       sizeof(*qgroup_limit));
636         if (ret && ret != -EEXIST)
637                 goto out;
638
639         leaf = path->nodes[0];
640         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641                                   struct btrfs_qgroup_limit_item);
642         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647
648         btrfs_mark_buffer_dirty(leaf);
649
650         ret = 0;
651 out:
652         btrfs_free_path(path);
653         return ret;
654 }
655
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
657 {
658         int ret;
659         struct btrfs_root *quota_root = trans->fs_info->quota_root;
660         struct btrfs_path *path;
661         struct btrfs_key key;
662
663         path = btrfs_alloc_path();
664         if (!path)
665                 return -ENOMEM;
666
667         key.objectid = 0;
668         key.type = BTRFS_QGROUP_INFO_KEY;
669         key.offset = qgroupid;
670         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671         if (ret < 0)
672                 goto out;
673
674         if (ret > 0) {
675                 ret = -ENOENT;
676                 goto out;
677         }
678
679         ret = btrfs_del_item(trans, quota_root, path);
680         if (ret)
681                 goto out;
682
683         btrfs_release_path(path);
684
685         key.type = BTRFS_QGROUP_LIMIT_KEY;
686         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
687         if (ret < 0)
688                 goto out;
689
690         if (ret > 0) {
691                 ret = -ENOENT;
692                 goto out;
693         }
694
695         ret = btrfs_del_item(trans, quota_root, path);
696
697 out:
698         btrfs_free_path(path);
699         return ret;
700 }
701
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703                                     struct btrfs_qgroup *qgroup)
704 {
705         struct btrfs_root *quota_root = trans->fs_info->quota_root;
706         struct btrfs_path *path;
707         struct btrfs_key key;
708         struct extent_buffer *l;
709         struct btrfs_qgroup_limit_item *qgroup_limit;
710         int ret;
711         int slot;
712
713         key.objectid = 0;
714         key.type = BTRFS_QGROUP_LIMIT_KEY;
715         key.offset = qgroup->qgroupid;
716
717         path = btrfs_alloc_path();
718         if (!path)
719                 return -ENOMEM;
720
721         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
722         if (ret > 0)
723                 ret = -ENOENT;
724
725         if (ret)
726                 goto out;
727
728         l = path->nodes[0];
729         slot = path->slots[0];
730         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736
737         btrfs_mark_buffer_dirty(l);
738
739 out:
740         btrfs_free_path(path);
741         return ret;
742 }
743
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745                                    struct btrfs_qgroup *qgroup)
746 {
747         struct btrfs_fs_info *fs_info = trans->fs_info;
748         struct btrfs_root *quota_root = fs_info->quota_root;
749         struct btrfs_path *path;
750         struct btrfs_key key;
751         struct extent_buffer *l;
752         struct btrfs_qgroup_info_item *qgroup_info;
753         int ret;
754         int slot;
755
756         if (btrfs_is_testing(fs_info))
757                 return 0;
758
759         key.objectid = 0;
760         key.type = BTRFS_QGROUP_INFO_KEY;
761         key.offset = qgroup->qgroupid;
762
763         path = btrfs_alloc_path();
764         if (!path)
765                 return -ENOMEM;
766
767         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
768         if (ret > 0)
769                 ret = -ENOENT;
770
771         if (ret)
772                 goto out;
773
774         l = path->nodes[0];
775         slot = path->slots[0];
776         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
782
783         btrfs_mark_buffer_dirty(l);
784
785 out:
786         btrfs_free_path(path);
787         return ret;
788 }
789
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
791 {
792         struct btrfs_fs_info *fs_info = trans->fs_info;
793         struct btrfs_root *quota_root = fs_info->quota_root;
794         struct btrfs_path *path;
795         struct btrfs_key key;
796         struct extent_buffer *l;
797         struct btrfs_qgroup_status_item *ptr;
798         int ret;
799         int slot;
800
801         key.objectid = 0;
802         key.type = BTRFS_QGROUP_STATUS_KEY;
803         key.offset = 0;
804
805         path = btrfs_alloc_path();
806         if (!path)
807                 return -ENOMEM;
808
809         ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
810         if (ret > 0)
811                 ret = -ENOENT;
812
813         if (ret)
814                 goto out;
815
816         l = path->nodes[0];
817         slot = path->slots[0];
818         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821         btrfs_set_qgroup_status_rescan(l, ptr,
822                                 fs_info->qgroup_rescan_progress.objectid);
823
824         btrfs_mark_buffer_dirty(l);
825
826 out:
827         btrfs_free_path(path);
828         return ret;
829 }
830
831 /*
832  * called with qgroup_lock held
833  */
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835                                   struct btrfs_root *root)
836 {
837         struct btrfs_path *path;
838         struct btrfs_key key;
839         struct extent_buffer *leaf = NULL;
840         int ret;
841         int nr = 0;
842
843         path = btrfs_alloc_path();
844         if (!path)
845                 return -ENOMEM;
846
847         path->leave_spinning = 1;
848
849         key.objectid = 0;
850         key.offset = 0;
851         key.type = 0;
852
853         while (1) {
854                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855                 if (ret < 0)
856                         goto out;
857                 leaf = path->nodes[0];
858                 nr = btrfs_header_nritems(leaf);
859                 if (!nr)
860                         break;
861                 /*
862                  * delete the leaf one by one
863                  * since the whole tree is going
864                  * to be deleted.
865                  */
866                 path->slots[0] = 0;
867                 ret = btrfs_del_items(trans, root, path, 0, nr);
868                 if (ret)
869                         goto out;
870
871                 btrfs_release_path(path);
872         }
873         ret = 0;
874 out:
875         btrfs_free_path(path);
876         return ret;
877 }
878
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
880 {
881         struct btrfs_root *quota_root;
882         struct btrfs_root *tree_root = fs_info->tree_root;
883         struct btrfs_path *path = NULL;
884         struct btrfs_qgroup_status_item *ptr;
885         struct extent_buffer *leaf;
886         struct btrfs_key key;
887         struct btrfs_key found_key;
888         struct btrfs_qgroup *qgroup = NULL;
889         struct btrfs_trans_handle *trans = NULL;
890         int ret = 0;
891         int slot;
892
893         mutex_lock(&fs_info->qgroup_ioctl_lock);
894         if (fs_info->quota_root)
895                 goto out;
896
897         /*
898          * 1 for quota root item
899          * 1 for BTRFS_QGROUP_STATUS item
900          *
901          * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
902          * per subvolume. However those are not currently reserved since it
903          * would be a lot of overkill.
904          */
905         trans = btrfs_start_transaction(tree_root, 2);
906         if (IS_ERR(trans)) {
907                 ret = PTR_ERR(trans);
908                 trans = NULL;
909                 goto out;
910         }
911
912         fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
913         if (!fs_info->qgroup_ulist) {
914                 ret = -ENOMEM;
915                 btrfs_abort_transaction(trans, ret);
916                 goto out;
917         }
918
919         /*
920          * initially create the quota tree
921          */
922         quota_root = btrfs_create_tree(trans, fs_info,
923                                        BTRFS_QUOTA_TREE_OBJECTID);
924         if (IS_ERR(quota_root)) {
925                 ret =  PTR_ERR(quota_root);
926                 btrfs_abort_transaction(trans, ret);
927                 goto out;
928         }
929
930         path = btrfs_alloc_path();
931         if (!path) {
932                 ret = -ENOMEM;
933                 btrfs_abort_transaction(trans, ret);
934                 goto out_free_root;
935         }
936
937         key.objectid = 0;
938         key.type = BTRFS_QGROUP_STATUS_KEY;
939         key.offset = 0;
940
941         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
942                                       sizeof(*ptr));
943         if (ret) {
944                 btrfs_abort_transaction(trans, ret);
945                 goto out_free_path;
946         }
947
948         leaf = path->nodes[0];
949         ptr = btrfs_item_ptr(leaf, path->slots[0],
950                                  struct btrfs_qgroup_status_item);
951         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
952         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
953         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
954                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
955         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
956         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
957
958         btrfs_mark_buffer_dirty(leaf);
959
960         key.objectid = 0;
961         key.type = BTRFS_ROOT_REF_KEY;
962         key.offset = 0;
963
964         btrfs_release_path(path);
965         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
966         if (ret > 0)
967                 goto out_add_root;
968         if (ret < 0) {
969                 btrfs_abort_transaction(trans, ret);
970                 goto out_free_path;
971         }
972
973         while (1) {
974                 slot = path->slots[0];
975                 leaf = path->nodes[0];
976                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
977
978                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
979                         ret = add_qgroup_item(trans, quota_root,
980                                               found_key.offset);
981                         if (ret) {
982                                 btrfs_abort_transaction(trans, ret);
983                                 goto out_free_path;
984                         }
985
986                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
987                         if (IS_ERR(qgroup)) {
988                                 ret = PTR_ERR(qgroup);
989                                 btrfs_abort_transaction(trans, ret);
990                                 goto out_free_path;
991                         }
992                 }
993                 ret = btrfs_next_item(tree_root, path);
994                 if (ret < 0) {
995                         btrfs_abort_transaction(trans, ret);
996                         goto out_free_path;
997                 }
998                 if (ret)
999                         break;
1000         }
1001
1002 out_add_root:
1003         btrfs_release_path(path);
1004         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1005         if (ret) {
1006                 btrfs_abort_transaction(trans, ret);
1007                 goto out_free_path;
1008         }
1009
1010         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011         if (IS_ERR(qgroup)) {
1012                 ret = PTR_ERR(qgroup);
1013                 btrfs_abort_transaction(trans, ret);
1014                 goto out_free_path;
1015         }
1016
1017         ret = btrfs_commit_transaction(trans);
1018         trans = NULL;
1019         if (ret)
1020                 goto out_free_path;
1021
1022         /*
1023          * Set quota enabled flag after committing the transaction, to avoid
1024          * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1025          * creation.
1026          */
1027         spin_lock(&fs_info->qgroup_lock);
1028         fs_info->quota_root = quota_root;
1029         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1030         spin_unlock(&fs_info->qgroup_lock);
1031
1032         ret = qgroup_rescan_init(fs_info, 0, 1);
1033         if (!ret) {
1034                 qgroup_rescan_zero_tracking(fs_info);
1035                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1036                                  &fs_info->qgroup_rescan_work);
1037         }
1038
1039 out_free_path:
1040         btrfs_free_path(path);
1041 out_free_root:
1042         if (ret) {
1043                 free_extent_buffer(quota_root->node);
1044                 free_extent_buffer(quota_root->commit_root);
1045                 kfree(quota_root);
1046         }
1047 out:
1048         if (ret) {
1049                 ulist_free(fs_info->qgroup_ulist);
1050                 fs_info->qgroup_ulist = NULL;
1051                 if (trans)
1052                         btrfs_end_transaction(trans);
1053         }
1054         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1055         return ret;
1056 }
1057
1058 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1059 {
1060         struct btrfs_root *quota_root;
1061         struct btrfs_trans_handle *trans = NULL;
1062         int ret = 0;
1063
1064         mutex_lock(&fs_info->qgroup_ioctl_lock);
1065         if (!fs_info->quota_root)
1066                 goto out;
1067
1068         /*
1069          * 1 For the root item
1070          *
1071          * We should also reserve enough items for the quota tree deletion in
1072          * btrfs_clean_quota_tree but this is not done.
1073          */
1074         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1075         if (IS_ERR(trans)) {
1076                 ret = PTR_ERR(trans);
1077                 goto out;
1078         }
1079
1080         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1081         btrfs_qgroup_wait_for_completion(fs_info, false);
1082         spin_lock(&fs_info->qgroup_lock);
1083         quota_root = fs_info->quota_root;
1084         fs_info->quota_root = NULL;
1085         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1086         spin_unlock(&fs_info->qgroup_lock);
1087
1088         btrfs_free_qgroup_config(fs_info);
1089
1090         ret = btrfs_clean_quota_tree(trans, quota_root);
1091         if (ret) {
1092                 btrfs_abort_transaction(trans, ret);
1093                 goto end_trans;
1094         }
1095
1096         ret = btrfs_del_root(trans, &quota_root->root_key);
1097         if (ret) {
1098                 btrfs_abort_transaction(trans, ret);
1099                 goto end_trans;
1100         }
1101
1102         list_del(&quota_root->dirty_list);
1103
1104         btrfs_tree_lock(quota_root->node);
1105         clean_tree_block(fs_info, quota_root->node);
1106         btrfs_tree_unlock(quota_root->node);
1107         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1108
1109         free_extent_buffer(quota_root->node);
1110         free_extent_buffer(quota_root->commit_root);
1111         kfree(quota_root);
1112
1113 end_trans:
1114         ret = btrfs_end_transaction(trans);
1115 out:
1116         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1117         return ret;
1118 }
1119
1120 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1121                          struct btrfs_qgroup *qgroup)
1122 {
1123         if (list_empty(&qgroup->dirty))
1124                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1125 }
1126
1127 /*
1128  * The easy accounting, we're updating qgroup relationship whose child qgroup
1129  * only has exclusive extents.
1130  *
1131  * In this case, all exclusive extents will also be exclusive for parent, so
1132  * excl/rfer just get added/removed.
1133  *
1134  * So is qgroup reservation space, which should also be added/removed to
1135  * parent.
1136  * Or when child tries to release reservation space, parent will underflow its
1137  * reservation (for relationship adding case).
1138  *
1139  * Caller should hold fs_info->qgroup_lock.
1140  */
1141 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1142                                     struct ulist *tmp, u64 ref_root,
1143                                     struct btrfs_qgroup *src, int sign)
1144 {
1145         struct btrfs_qgroup *qgroup;
1146         struct btrfs_qgroup_list *glist;
1147         struct ulist_node *unode;
1148         struct ulist_iterator uiter;
1149         u64 num_bytes = src->excl;
1150         int ret = 0;
1151
1152         qgroup = find_qgroup_rb(fs_info, ref_root);
1153         if (!qgroup)
1154                 goto out;
1155
1156         qgroup->rfer += sign * num_bytes;
1157         qgroup->rfer_cmpr += sign * num_bytes;
1158
1159         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1160         qgroup->excl += sign * num_bytes;
1161         qgroup->excl_cmpr += sign * num_bytes;
1162
1163         if (sign > 0)
1164                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1165         else
1166                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1167
1168         qgroup_dirty(fs_info, qgroup);
1169
1170         /* Get all of the parent groups that contain this qgroup */
1171         list_for_each_entry(glist, &qgroup->groups, next_group) {
1172                 ret = ulist_add(tmp, glist->group->qgroupid,
1173                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1174                 if (ret < 0)
1175                         goto out;
1176         }
1177
1178         /* Iterate all of the parents and adjust their reference counts */
1179         ULIST_ITER_INIT(&uiter);
1180         while ((unode = ulist_next(tmp, &uiter))) {
1181                 qgroup = unode_aux_to_qgroup(unode);
1182                 qgroup->rfer += sign * num_bytes;
1183                 qgroup->rfer_cmpr += sign * num_bytes;
1184                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1185                 qgroup->excl += sign * num_bytes;
1186                 if (sign > 0)
1187                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1188                 else
1189                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1190                 qgroup->excl_cmpr += sign * num_bytes;
1191                 qgroup_dirty(fs_info, qgroup);
1192
1193                 /* Add any parents of the parents */
1194                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1195                         ret = ulist_add(tmp, glist->group->qgroupid,
1196                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1197                         if (ret < 0)
1198                                 goto out;
1199                 }
1200         }
1201         ret = 0;
1202 out:
1203         return ret;
1204 }
1205
1206
1207 /*
1208  * Quick path for updating qgroup with only excl refs.
1209  *
1210  * In that case, just update all parent will be enough.
1211  * Or we needs to do a full rescan.
1212  * Caller should also hold fs_info->qgroup_lock.
1213  *
1214  * Return 0 for quick update, return >0 for need to full rescan
1215  * and mark INCONSISTENT flag.
1216  * Return < 0 for other error.
1217  */
1218 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1219                                    struct ulist *tmp, u64 src, u64 dst,
1220                                    int sign)
1221 {
1222         struct btrfs_qgroup *qgroup;
1223         int ret = 1;
1224         int err = 0;
1225
1226         qgroup = find_qgroup_rb(fs_info, src);
1227         if (!qgroup)
1228                 goto out;
1229         if (qgroup->excl == qgroup->rfer) {
1230                 ret = 0;
1231                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1232                                                qgroup, sign);
1233                 if (err < 0) {
1234                         ret = err;
1235                         goto out;
1236                 }
1237         }
1238 out:
1239         if (ret)
1240                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1241         return ret;
1242 }
1243
1244 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1245                               u64 dst)
1246 {
1247         struct btrfs_fs_info *fs_info = trans->fs_info;
1248         struct btrfs_root *quota_root;
1249         struct btrfs_qgroup *parent;
1250         struct btrfs_qgroup *member;
1251         struct btrfs_qgroup_list *list;
1252         struct ulist *tmp;
1253         int ret = 0;
1254
1255         /* Check the level of src and dst first */
1256         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1257                 return -EINVAL;
1258
1259         tmp = ulist_alloc(GFP_KERNEL);
1260         if (!tmp)
1261                 return -ENOMEM;
1262
1263         mutex_lock(&fs_info->qgroup_ioctl_lock);
1264         quota_root = fs_info->quota_root;
1265         if (!quota_root) {
1266                 ret = -EINVAL;
1267                 goto out;
1268         }
1269         member = find_qgroup_rb(fs_info, src);
1270         parent = find_qgroup_rb(fs_info, dst);
1271         if (!member || !parent) {
1272                 ret = -EINVAL;
1273                 goto out;
1274         }
1275
1276         /* check if such qgroup relation exist firstly */
1277         list_for_each_entry(list, &member->groups, next_group) {
1278                 if (list->group == parent) {
1279                         ret = -EEXIST;
1280                         goto out;
1281                 }
1282         }
1283
1284         ret = add_qgroup_relation_item(trans, src, dst);
1285         if (ret)
1286                 goto out;
1287
1288         ret = add_qgroup_relation_item(trans, dst, src);
1289         if (ret) {
1290                 del_qgroup_relation_item(trans, src, dst);
1291                 goto out;
1292         }
1293
1294         spin_lock(&fs_info->qgroup_lock);
1295         ret = add_relation_rb(fs_info, src, dst);
1296         if (ret < 0) {
1297                 spin_unlock(&fs_info->qgroup_lock);
1298                 goto out;
1299         }
1300         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1301         spin_unlock(&fs_info->qgroup_lock);
1302 out:
1303         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1304         ulist_free(tmp);
1305         return ret;
1306 }
1307
1308 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1309                                  u64 dst)
1310 {
1311         struct btrfs_fs_info *fs_info = trans->fs_info;
1312         struct btrfs_root *quota_root;
1313         struct btrfs_qgroup *parent;
1314         struct btrfs_qgroup *member;
1315         struct btrfs_qgroup_list *list;
1316         struct ulist *tmp;
1317         int ret = 0;
1318         int err;
1319
1320         tmp = ulist_alloc(GFP_KERNEL);
1321         if (!tmp)
1322                 return -ENOMEM;
1323
1324         quota_root = fs_info->quota_root;
1325         if (!quota_root) {
1326                 ret = -EINVAL;
1327                 goto out;
1328         }
1329
1330         member = find_qgroup_rb(fs_info, src);
1331         parent = find_qgroup_rb(fs_info, dst);
1332         if (!member || !parent) {
1333                 ret = -EINVAL;
1334                 goto out;
1335         }
1336
1337         /* check if such qgroup relation exist firstly */
1338         list_for_each_entry(list, &member->groups, next_group) {
1339                 if (list->group == parent)
1340                         goto exist;
1341         }
1342         ret = -ENOENT;
1343         goto out;
1344 exist:
1345         ret = del_qgroup_relation_item(trans, src, dst);
1346         err = del_qgroup_relation_item(trans, dst, src);
1347         if (err && !ret)
1348                 ret = err;
1349
1350         spin_lock(&fs_info->qgroup_lock);
1351         del_relation_rb(fs_info, src, dst);
1352         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1353         spin_unlock(&fs_info->qgroup_lock);
1354 out:
1355         ulist_free(tmp);
1356         return ret;
1357 }
1358
1359 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1360                               u64 dst)
1361 {
1362         struct btrfs_fs_info *fs_info = trans->fs_info;
1363         int ret = 0;
1364
1365         mutex_lock(&fs_info->qgroup_ioctl_lock);
1366         ret = __del_qgroup_relation(trans, src, dst);
1367         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1368
1369         return ret;
1370 }
1371
1372 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1373 {
1374         struct btrfs_fs_info *fs_info = trans->fs_info;
1375         struct btrfs_root *quota_root;
1376         struct btrfs_qgroup *qgroup;
1377         int ret = 0;
1378
1379         mutex_lock(&fs_info->qgroup_ioctl_lock);
1380         quota_root = fs_info->quota_root;
1381         if (!quota_root) {
1382                 ret = -EINVAL;
1383                 goto out;
1384         }
1385         qgroup = find_qgroup_rb(fs_info, qgroupid);
1386         if (qgroup) {
1387                 ret = -EEXIST;
1388                 goto out;
1389         }
1390
1391         ret = add_qgroup_item(trans, quota_root, qgroupid);
1392         if (ret)
1393                 goto out;
1394
1395         spin_lock(&fs_info->qgroup_lock);
1396         qgroup = add_qgroup_rb(fs_info, qgroupid);
1397         spin_unlock(&fs_info->qgroup_lock);
1398
1399         if (IS_ERR(qgroup))
1400                 ret = PTR_ERR(qgroup);
1401 out:
1402         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1403         return ret;
1404 }
1405
1406 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1407 {
1408         struct btrfs_fs_info *fs_info = trans->fs_info;
1409         struct btrfs_root *quota_root;
1410         struct btrfs_qgroup *qgroup;
1411         struct btrfs_qgroup_list *list;
1412         int ret = 0;
1413
1414         mutex_lock(&fs_info->qgroup_ioctl_lock);
1415         quota_root = fs_info->quota_root;
1416         if (!quota_root) {
1417                 ret = -EINVAL;
1418                 goto out;
1419         }
1420
1421         qgroup = find_qgroup_rb(fs_info, qgroupid);
1422         if (!qgroup) {
1423                 ret = -ENOENT;
1424                 goto out;
1425         }
1426
1427         /* Check if there are no children of this qgroup */
1428         if (!list_empty(&qgroup->members)) {
1429                 ret = -EBUSY;
1430                 goto out;
1431         }
1432
1433         ret = del_qgroup_item(trans, qgroupid);
1434         if (ret && ret != -ENOENT)
1435                 goto out;
1436
1437         while (!list_empty(&qgroup->groups)) {
1438                 list = list_first_entry(&qgroup->groups,
1439                                         struct btrfs_qgroup_list, next_group);
1440                 ret = __del_qgroup_relation(trans, qgroupid,
1441                                             list->group->qgroupid);
1442                 if (ret)
1443                         goto out;
1444         }
1445
1446         spin_lock(&fs_info->qgroup_lock);
1447         del_qgroup_rb(fs_info, qgroupid);
1448         spin_unlock(&fs_info->qgroup_lock);
1449 out:
1450         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1451         return ret;
1452 }
1453
1454 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1455                        struct btrfs_qgroup_limit *limit)
1456 {
1457         struct btrfs_fs_info *fs_info = trans->fs_info;
1458         struct btrfs_root *quota_root;
1459         struct btrfs_qgroup *qgroup;
1460         int ret = 0;
1461         /* Sometimes we would want to clear the limit on this qgroup.
1462          * To meet this requirement, we treat the -1 as a special value
1463          * which tell kernel to clear the limit on this qgroup.
1464          */
1465         const u64 CLEAR_VALUE = -1;
1466
1467         mutex_lock(&fs_info->qgroup_ioctl_lock);
1468         quota_root = fs_info->quota_root;
1469         if (!quota_root) {
1470                 ret = -EINVAL;
1471                 goto out;
1472         }
1473
1474         qgroup = find_qgroup_rb(fs_info, qgroupid);
1475         if (!qgroup) {
1476                 ret = -ENOENT;
1477                 goto out;
1478         }
1479
1480         spin_lock(&fs_info->qgroup_lock);
1481         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1482                 if (limit->max_rfer == CLEAR_VALUE) {
1483                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1484                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1485                         qgroup->max_rfer = 0;
1486                 } else {
1487                         qgroup->max_rfer = limit->max_rfer;
1488                 }
1489         }
1490         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1491                 if (limit->max_excl == CLEAR_VALUE) {
1492                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1493                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1494                         qgroup->max_excl = 0;
1495                 } else {
1496                         qgroup->max_excl = limit->max_excl;
1497                 }
1498         }
1499         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1500                 if (limit->rsv_rfer == CLEAR_VALUE) {
1501                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1502                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1503                         qgroup->rsv_rfer = 0;
1504                 } else {
1505                         qgroup->rsv_rfer = limit->rsv_rfer;
1506                 }
1507         }
1508         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1509                 if (limit->rsv_excl == CLEAR_VALUE) {
1510                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1511                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1512                         qgroup->rsv_excl = 0;
1513                 } else {
1514                         qgroup->rsv_excl = limit->rsv_excl;
1515                 }
1516         }
1517         qgroup->lim_flags |= limit->flags;
1518
1519         spin_unlock(&fs_info->qgroup_lock);
1520
1521         ret = update_qgroup_limit_item(trans, qgroup);
1522         if (ret) {
1523                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1524                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1525                        qgroupid);
1526         }
1527
1528 out:
1529         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1530         return ret;
1531 }
1532
1533 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1534                                 struct btrfs_delayed_ref_root *delayed_refs,
1535                                 struct btrfs_qgroup_extent_record *record)
1536 {
1537         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1538         struct rb_node *parent_node = NULL;
1539         struct btrfs_qgroup_extent_record *entry;
1540         u64 bytenr = record->bytenr;
1541
1542         lockdep_assert_held(&delayed_refs->lock);
1543         trace_btrfs_qgroup_trace_extent(fs_info, record);
1544
1545         while (*p) {
1546                 parent_node = *p;
1547                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1548                                  node);
1549                 if (bytenr < entry->bytenr)
1550                         p = &(*p)->rb_left;
1551                 else if (bytenr > entry->bytenr)
1552                         p = &(*p)->rb_right;
1553                 else
1554                         return 1;
1555         }
1556
1557         rb_link_node(&record->node, parent_node, p);
1558         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1559         return 0;
1560 }
1561
1562 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1563                                    struct btrfs_qgroup_extent_record *qrecord)
1564 {
1565         struct ulist *old_root;
1566         u64 bytenr = qrecord->bytenr;
1567         int ret;
1568
1569         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1570         if (ret < 0) {
1571                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1572                 btrfs_warn(fs_info,
1573 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1574                         ret);
1575                 return 0;
1576         }
1577
1578         /*
1579          * Here we don't need to get the lock of
1580          * trans->transaction->delayed_refs, since inserted qrecord won't
1581          * be deleted, only qrecord->node may be modified (new qrecord insert)
1582          *
1583          * So modifying qrecord->old_roots is safe here
1584          */
1585         qrecord->old_roots = old_root;
1586         return 0;
1587 }
1588
1589 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1590                               u64 num_bytes, gfp_t gfp_flag)
1591 {
1592         struct btrfs_fs_info *fs_info = trans->fs_info;
1593         struct btrfs_qgroup_extent_record *record;
1594         struct btrfs_delayed_ref_root *delayed_refs;
1595         int ret;
1596
1597         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1598             || bytenr == 0 || num_bytes == 0)
1599                 return 0;
1600         record = kmalloc(sizeof(*record), gfp_flag);
1601         if (!record)
1602                 return -ENOMEM;
1603
1604         delayed_refs = &trans->transaction->delayed_refs;
1605         record->bytenr = bytenr;
1606         record->num_bytes = num_bytes;
1607         record->old_roots = NULL;
1608
1609         spin_lock(&delayed_refs->lock);
1610         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1611         spin_unlock(&delayed_refs->lock);
1612         if (ret > 0) {
1613                 kfree(record);
1614                 return 0;
1615         }
1616         return btrfs_qgroup_trace_extent_post(fs_info, record);
1617 }
1618
1619 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1620                                   struct extent_buffer *eb)
1621 {
1622         struct btrfs_fs_info *fs_info = trans->fs_info;
1623         int nr = btrfs_header_nritems(eb);
1624         int i, extent_type, ret;
1625         struct btrfs_key key;
1626         struct btrfs_file_extent_item *fi;
1627         u64 bytenr, num_bytes;
1628
1629         /* We can be called directly from walk_up_proc() */
1630         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1631                 return 0;
1632
1633         for (i = 0; i < nr; i++) {
1634                 btrfs_item_key_to_cpu(eb, &key, i);
1635
1636                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1637                         continue;
1638
1639                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1640                 /* filter out non qgroup-accountable extents  */
1641                 extent_type = btrfs_file_extent_type(eb, fi);
1642
1643                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1644                         continue;
1645
1646                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1647                 if (!bytenr)
1648                         continue;
1649
1650                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1651
1652                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1653                                                 GFP_NOFS);
1654                 if (ret)
1655                         return ret;
1656         }
1657         cond_resched();
1658         return 0;
1659 }
1660
1661 /*
1662  * Walk up the tree from the bottom, freeing leaves and any interior
1663  * nodes which have had all slots visited. If a node (leaf or
1664  * interior) is freed, the node above it will have it's slot
1665  * incremented. The root node will never be freed.
1666  *
1667  * At the end of this function, we should have a path which has all
1668  * slots incremented to the next position for a search. If we need to
1669  * read a new node it will be NULL and the node above it will have the
1670  * correct slot selected for a later read.
1671  *
1672  * If we increment the root nodes slot counter past the number of
1673  * elements, 1 is returned to signal completion of the search.
1674  */
1675 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1676 {
1677         int level = 0;
1678         int nr, slot;
1679         struct extent_buffer *eb;
1680
1681         if (root_level == 0)
1682                 return 1;
1683
1684         while (level <= root_level) {
1685                 eb = path->nodes[level];
1686                 nr = btrfs_header_nritems(eb);
1687                 path->slots[level]++;
1688                 slot = path->slots[level];
1689                 if (slot >= nr || level == 0) {
1690                         /*
1691                          * Don't free the root -  we will detect this
1692                          * condition after our loop and return a
1693                          * positive value for caller to stop walking the tree.
1694                          */
1695                         if (level != root_level) {
1696                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1697                                 path->locks[level] = 0;
1698
1699                                 free_extent_buffer(eb);
1700                                 path->nodes[level] = NULL;
1701                                 path->slots[level] = 0;
1702                         }
1703                 } else {
1704                         /*
1705                          * We have a valid slot to walk back down
1706                          * from. Stop here so caller can process these
1707                          * new nodes.
1708                          */
1709                         break;
1710                 }
1711
1712                 level++;
1713         }
1714
1715         eb = path->nodes[root_level];
1716         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1717                 return 1;
1718
1719         return 0;
1720 }
1721
1722 /*
1723  * Helper function to trace a subtree tree block swap.
1724  *
1725  * The swap will happen in highest tree block, but there may be a lot of
1726  * tree blocks involved.
1727  *
1728  * For example:
1729  *  OO = Old tree blocks
1730  *  NN = New tree blocks allocated during balance
1731  *
1732  *           File tree (257)                  Reloc tree for 257
1733  * L2              OO                                NN
1734  *               /    \                            /    \
1735  * L1          OO      OO (a)                    OO      NN (a)
1736  *            / \     / \                       / \     / \
1737  * L0       OO   OO OO   OO                   OO   OO NN   NN
1738  *                  (b)  (c)                          (b)  (c)
1739  *
1740  * When calling qgroup_trace_extent_swap(), we will pass:
1741  * @src_eb = OO(a)
1742  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1743  * @dst_level = 0
1744  * @root_level = 1
1745  *
1746  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1747  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1748  *
1749  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1750  *
1751  * 1) Tree search from @src_eb
1752  *    It should acts as a simplified btrfs_search_slot().
1753  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1754  *    (first key).
1755  *
1756  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1757  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1758  *    They should be marked during previous (@dst_level = 1) iteration.
1759  *
1760  * 3) Mark file extents in leaves dirty
1761  *    We don't have good way to pick out new file extents only.
1762  *    So we still follow the old method by scanning all file extents in
1763  *    the leave.
1764  *
1765  * This function can free us from keeping two paths, thus later we only need
1766  * to care about how to iterate all new tree blocks in reloc tree.
1767  */
1768 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1769                                     struct extent_buffer *src_eb,
1770                                     struct btrfs_path *dst_path,
1771                                     int dst_level, int root_level,
1772                                     bool trace_leaf)
1773 {
1774         struct btrfs_key key;
1775         struct btrfs_path *src_path;
1776         struct btrfs_fs_info *fs_info = trans->fs_info;
1777         u32 nodesize = fs_info->nodesize;
1778         int cur_level = root_level;
1779         int ret;
1780
1781         BUG_ON(dst_level > root_level);
1782         /* Level mismatch */
1783         if (btrfs_header_level(src_eb) != root_level)
1784                 return -EINVAL;
1785
1786         src_path = btrfs_alloc_path();
1787         if (!src_path) {
1788                 ret = -ENOMEM;
1789                 goto out;
1790         }
1791
1792         if (dst_level)
1793                 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1794         else
1795                 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1796
1797         /* For src_path */
1798         extent_buffer_get(src_eb);
1799         src_path->nodes[root_level] = src_eb;
1800         src_path->slots[root_level] = dst_path->slots[root_level];
1801         src_path->locks[root_level] = 0;
1802
1803         /* A simplified version of btrfs_search_slot() */
1804         while (cur_level >= dst_level) {
1805                 struct btrfs_key src_key;
1806                 struct btrfs_key dst_key;
1807
1808                 if (src_path->nodes[cur_level] == NULL) {
1809                         struct btrfs_key first_key;
1810                         struct extent_buffer *eb;
1811                         int parent_slot;
1812                         u64 child_gen;
1813                         u64 child_bytenr;
1814
1815                         eb = src_path->nodes[cur_level + 1];
1816                         parent_slot = src_path->slots[cur_level + 1];
1817                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1818                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1819                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1820
1821                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1822                                              cur_level, &first_key);
1823                         if (IS_ERR(eb)) {
1824                                 ret = PTR_ERR(eb);
1825                                 goto out;
1826                         } else if (!extent_buffer_uptodate(eb)) {
1827                                 free_extent_buffer(eb);
1828                                 ret = -EIO;
1829                                 goto out;
1830                         }
1831
1832                         src_path->nodes[cur_level] = eb;
1833
1834                         btrfs_tree_read_lock(eb);
1835                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1836                         src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1837                 }
1838
1839                 src_path->slots[cur_level] = dst_path->slots[cur_level];
1840                 if (cur_level) {
1841                         btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1842                                         &dst_key, dst_path->slots[cur_level]);
1843                         btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1844                                         &src_key, src_path->slots[cur_level]);
1845                 } else {
1846                         btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1847                                         &dst_key, dst_path->slots[cur_level]);
1848                         btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1849                                         &src_key, src_path->slots[cur_level]);
1850                 }
1851                 /* Content mismatch, something went wrong */
1852                 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1853                         ret = -ENOENT;
1854                         goto out;
1855                 }
1856                 cur_level--;
1857         }
1858
1859         /*
1860          * Now both @dst_path and @src_path have been populated, record the tree
1861          * blocks for qgroup accounting.
1862          */
1863         ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1864                         nodesize, GFP_NOFS);
1865         if (ret < 0)
1866                 goto out;
1867         ret = btrfs_qgroup_trace_extent(trans,
1868                         dst_path->nodes[dst_level]->start,
1869                         nodesize, GFP_NOFS);
1870         if (ret < 0)
1871                 goto out;
1872
1873         /* Record leaf file extents */
1874         if (dst_level == 0 && trace_leaf) {
1875                 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1876                 if (ret < 0)
1877                         goto out;
1878                 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1879         }
1880 out:
1881         btrfs_free_path(src_path);
1882         return ret;
1883 }
1884
1885 /*
1886  * Helper function to do recursive generation-aware depth-first search, to
1887  * locate all new tree blocks in a subtree of reloc tree.
1888  *
1889  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1890  *         reloc tree
1891  * L2         NN (a)
1892  *          /    \
1893  * L1    OO        NN (b)
1894  *      /  \      /  \
1895  * L0  OO  OO    OO  NN
1896  *               (c) (d)
1897  * If we pass:
1898  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1899  * @cur_level = 1
1900  * @root_level = 1
1901  *
1902  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1903  * above tree blocks along with their counter parts in file tree.
1904  * While during search, old tree blocks OO(c) will be skipped as tree block swap
1905  * won't affect OO(c).
1906  */
1907 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1908                                            struct extent_buffer *src_eb,
1909                                            struct btrfs_path *dst_path,
1910                                            int cur_level, int root_level,
1911                                            u64 last_snapshot, bool trace_leaf)
1912 {
1913         struct btrfs_fs_info *fs_info = trans->fs_info;
1914         struct extent_buffer *eb;
1915         bool need_cleanup = false;
1916         int ret = 0;
1917         int i;
1918
1919         /* Level sanity check */
1920         if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL ||
1921             root_level < 0 || root_level >= BTRFS_MAX_LEVEL ||
1922             root_level < cur_level) {
1923                 btrfs_err_rl(fs_info,
1924                         "%s: bad levels, cur_level=%d root_level=%d",
1925                         __func__, cur_level, root_level);
1926                 return -EUCLEAN;
1927         }
1928
1929         /* Read the tree block if needed */
1930         if (dst_path->nodes[cur_level] == NULL) {
1931                 struct btrfs_key first_key;
1932                 int parent_slot;
1933                 u64 child_gen;
1934                 u64 child_bytenr;
1935
1936                 /*
1937                  * dst_path->nodes[root_level] must be initialized before
1938                  * calling this function.
1939                  */
1940                 if (cur_level == root_level) {
1941                         btrfs_err_rl(fs_info,
1942         "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1943                                 __func__, root_level, root_level, cur_level);
1944                         return -EUCLEAN;
1945                 }
1946
1947                 /*
1948                  * We need to get child blockptr/gen from parent before we can
1949                  * read it.
1950                   */
1951                 eb = dst_path->nodes[cur_level + 1];
1952                 parent_slot = dst_path->slots[cur_level + 1];
1953                 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1954                 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1955                 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1956
1957                 /* This node is old, no need to trace */
1958                 if (child_gen < last_snapshot)
1959                         goto out;
1960
1961                 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1962                                      cur_level, &first_key);
1963                 if (IS_ERR(eb)) {
1964                         ret = PTR_ERR(eb);
1965                         goto out;
1966                 } else if (!extent_buffer_uptodate(eb)) {
1967                         free_extent_buffer(eb);
1968                         ret = -EIO;
1969                         goto out;
1970                 }
1971
1972                 dst_path->nodes[cur_level] = eb;
1973                 dst_path->slots[cur_level] = 0;
1974
1975                 btrfs_tree_read_lock(eb);
1976                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1977                 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1978                 need_cleanup = true;
1979         }
1980
1981         /* Now record this tree block and its counter part for qgroups */
1982         ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1983                                        root_level, trace_leaf);
1984         if (ret < 0)
1985                 goto cleanup;
1986
1987         eb = dst_path->nodes[cur_level];
1988
1989         if (cur_level > 0) {
1990                 /* Iterate all child tree blocks */
1991                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1992                         /* Skip old tree blocks as they won't be swapped */
1993                         if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1994                                 continue;
1995                         dst_path->slots[cur_level] = i;
1996
1997                         /* Recursive call (at most 7 times) */
1998                         ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
1999                                         dst_path, cur_level - 1, root_level,
2000                                         last_snapshot, trace_leaf);
2001                         if (ret < 0)
2002                                 goto cleanup;
2003                 }
2004         }
2005
2006 cleanup:
2007         if (need_cleanup) {
2008                 /* Clean up */
2009                 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2010                                      dst_path->locks[cur_level]);
2011                 free_extent_buffer(dst_path->nodes[cur_level]);
2012                 dst_path->nodes[cur_level] = NULL;
2013                 dst_path->slots[cur_level] = 0;
2014                 dst_path->locks[cur_level] = 0;
2015         }
2016 out:
2017         return ret;
2018 }
2019
2020 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2021                                 struct extent_buffer *src_eb,
2022                                 struct extent_buffer *dst_eb,
2023                                 u64 last_snapshot, bool trace_leaf)
2024 {
2025         struct btrfs_fs_info *fs_info = trans->fs_info;
2026         struct btrfs_path *dst_path = NULL;
2027         int level;
2028         int ret;
2029
2030         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2031                 return 0;
2032
2033         /* Wrong parameter order */
2034         if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2035                 btrfs_err_rl(fs_info,
2036                 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2037                              btrfs_header_generation(src_eb),
2038                              btrfs_header_generation(dst_eb));
2039                 return -EUCLEAN;
2040         }
2041
2042         if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2043                 ret = -EIO;
2044                 goto out;
2045         }
2046
2047         level = btrfs_header_level(dst_eb);
2048         dst_path = btrfs_alloc_path();
2049         if (!dst_path) {
2050                 ret = -ENOMEM;
2051                 goto out;
2052         }
2053         /* For dst_path */
2054         extent_buffer_get(dst_eb);
2055         dst_path->nodes[level] = dst_eb;
2056         dst_path->slots[level] = 0;
2057         dst_path->locks[level] = 0;
2058
2059         /* Do the generation aware breadth-first search */
2060         ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2061                                               level, last_snapshot, trace_leaf);
2062         if (ret < 0)
2063                 goto out;
2064         ret = 0;
2065
2066 out:
2067         btrfs_free_path(dst_path);
2068         if (ret < 0)
2069                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2070         return ret;
2071 }
2072
2073 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2074                                struct extent_buffer *root_eb,
2075                                u64 root_gen, int root_level)
2076 {
2077         struct btrfs_fs_info *fs_info = trans->fs_info;
2078         int ret = 0;
2079         int level;
2080         struct extent_buffer *eb = root_eb;
2081         struct btrfs_path *path = NULL;
2082
2083         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2084         BUG_ON(root_eb == NULL);
2085
2086         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2087                 return 0;
2088
2089         if (!extent_buffer_uptodate(root_eb)) {
2090                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2091                 if (ret)
2092                         goto out;
2093         }
2094
2095         if (root_level == 0) {
2096                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2097                 goto out;
2098         }
2099
2100         path = btrfs_alloc_path();
2101         if (!path)
2102                 return -ENOMEM;
2103
2104         /*
2105          * Walk down the tree.  Missing extent blocks are filled in as
2106          * we go. Metadata is accounted every time we read a new
2107          * extent block.
2108          *
2109          * When we reach a leaf, we account for file extent items in it,
2110          * walk back up the tree (adjusting slot pointers as we go)
2111          * and restart the search process.
2112          */
2113         extent_buffer_get(root_eb); /* For path */
2114         path->nodes[root_level] = root_eb;
2115         path->slots[root_level] = 0;
2116         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2117 walk_down:
2118         level = root_level;
2119         while (level >= 0) {
2120                 if (path->nodes[level] == NULL) {
2121                         struct btrfs_key first_key;
2122                         int parent_slot;
2123                         u64 child_gen;
2124                         u64 child_bytenr;
2125
2126                         /*
2127                          * We need to get child blockptr/gen from parent before
2128                          * we can read it.
2129                           */
2130                         eb = path->nodes[level + 1];
2131                         parent_slot = path->slots[level + 1];
2132                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2133                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2134                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2135
2136                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
2137                                              level, &first_key);
2138                         if (IS_ERR(eb)) {
2139                                 ret = PTR_ERR(eb);
2140                                 goto out;
2141                         } else if (!extent_buffer_uptodate(eb)) {
2142                                 free_extent_buffer(eb);
2143                                 ret = -EIO;
2144                                 goto out;
2145                         }
2146
2147                         path->nodes[level] = eb;
2148                         path->slots[level] = 0;
2149
2150                         btrfs_tree_read_lock(eb);
2151                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
2152                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2153
2154                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2155                                                         fs_info->nodesize,
2156                                                         GFP_NOFS);
2157                         if (ret)
2158                                 goto out;
2159                 }
2160
2161                 if (level == 0) {
2162                         ret = btrfs_qgroup_trace_leaf_items(trans,
2163                                                             path->nodes[level]);
2164                         if (ret)
2165                                 goto out;
2166
2167                         /* Nonzero return here means we completed our search */
2168                         ret = adjust_slots_upwards(path, root_level);
2169                         if (ret)
2170                                 break;
2171
2172                         /* Restart search with new slots */
2173                         goto walk_down;
2174                 }
2175
2176                 level--;
2177         }
2178
2179         ret = 0;
2180 out:
2181         btrfs_free_path(path);
2182
2183         return ret;
2184 }
2185
2186 #define UPDATE_NEW      0
2187 #define UPDATE_OLD      1
2188 /*
2189  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2190  */
2191 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2192                                 struct ulist *roots, struct ulist *tmp,
2193                                 struct ulist *qgroups, u64 seq, int update_old)
2194 {
2195         struct ulist_node *unode;
2196         struct ulist_iterator uiter;
2197         struct ulist_node *tmp_unode;
2198         struct ulist_iterator tmp_uiter;
2199         struct btrfs_qgroup *qg;
2200         int ret = 0;
2201
2202         if (!roots)
2203                 return 0;
2204         ULIST_ITER_INIT(&uiter);
2205         while ((unode = ulist_next(roots, &uiter))) {
2206                 qg = find_qgroup_rb(fs_info, unode->val);
2207                 if (!qg)
2208                         continue;
2209
2210                 ulist_reinit(tmp);
2211                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2212                                 GFP_ATOMIC);
2213                 if (ret < 0)
2214                         return ret;
2215                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2216                 if (ret < 0)
2217                         return ret;
2218                 ULIST_ITER_INIT(&tmp_uiter);
2219                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2220                         struct btrfs_qgroup_list *glist;
2221
2222                         qg = unode_aux_to_qgroup(tmp_unode);
2223                         if (update_old)
2224                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2225                         else
2226                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2227                         list_for_each_entry(glist, &qg->groups, next_group) {
2228                                 ret = ulist_add(qgroups, glist->group->qgroupid,
2229                                                 qgroup_to_aux(glist->group),
2230                                                 GFP_ATOMIC);
2231                                 if (ret < 0)
2232                                         return ret;
2233                                 ret = ulist_add(tmp, glist->group->qgroupid,
2234                                                 qgroup_to_aux(glist->group),
2235                                                 GFP_ATOMIC);
2236                                 if (ret < 0)
2237                                         return ret;
2238                         }
2239                 }
2240         }
2241         return 0;
2242 }
2243
2244 /*
2245  * Update qgroup rfer/excl counters.
2246  * Rfer update is easy, codes can explain themselves.
2247  *
2248  * Excl update is tricky, the update is split into 2 part.
2249  * Part 1: Possible exclusive <-> sharing detect:
2250  *      |       A       |       !A      |
2251  *  -------------------------------------
2252  *  B   |       *       |       -       |
2253  *  -------------------------------------
2254  *  !B  |       +       |       **      |
2255  *  -------------------------------------
2256  *
2257  * Conditions:
2258  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
2259  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
2260  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
2261  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
2262  *
2263  * Results:
2264  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
2265  * *: Definitely not changed.           **: Possible unchanged.
2266  *
2267  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2268  *
2269  * To make the logic clear, we first use condition A and B to split
2270  * combination into 4 results.
2271  *
2272  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2273  * only on variant maybe 0.
2274  *
2275  * Lastly, check result **, since there are 2 variants maybe 0, split them
2276  * again(2x2).
2277  * But this time we don't need to consider other things, the codes and logic
2278  * is easy to understand now.
2279  */
2280 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2281                                   struct ulist *qgroups,
2282                                   u64 nr_old_roots,
2283                                   u64 nr_new_roots,
2284                                   u64 num_bytes, u64 seq)
2285 {
2286         struct ulist_node *unode;
2287         struct ulist_iterator uiter;
2288         struct btrfs_qgroup *qg;
2289         u64 cur_new_count, cur_old_count;
2290
2291         ULIST_ITER_INIT(&uiter);
2292         while ((unode = ulist_next(qgroups, &uiter))) {
2293                 bool dirty = false;
2294
2295                 qg = unode_aux_to_qgroup(unode);
2296                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2297                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2298
2299                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2300                                              cur_new_count);
2301
2302                 /* Rfer update part */
2303                 if (cur_old_count == 0 && cur_new_count > 0) {
2304                         qg->rfer += num_bytes;
2305                         qg->rfer_cmpr += num_bytes;
2306                         dirty = true;
2307                 }
2308                 if (cur_old_count > 0 && cur_new_count == 0) {
2309                         qg->rfer -= num_bytes;
2310                         qg->rfer_cmpr -= num_bytes;
2311                         dirty = true;
2312                 }
2313
2314                 /* Excl update part */
2315                 /* Exclusive/none -> shared case */
2316                 if (cur_old_count == nr_old_roots &&
2317                     cur_new_count < nr_new_roots) {
2318                         /* Exclusive -> shared */
2319                         if (cur_old_count != 0) {
2320                                 qg->excl -= num_bytes;
2321                                 qg->excl_cmpr -= num_bytes;
2322                                 dirty = true;
2323                         }
2324                 }
2325
2326                 /* Shared -> exclusive/none case */
2327                 if (cur_old_count < nr_old_roots &&
2328                     cur_new_count == nr_new_roots) {
2329                         /* Shared->exclusive */
2330                         if (cur_new_count != 0) {
2331                                 qg->excl += num_bytes;
2332                                 qg->excl_cmpr += num_bytes;
2333                                 dirty = true;
2334                         }
2335                 }
2336
2337                 /* Exclusive/none -> exclusive/none case */
2338                 if (cur_old_count == nr_old_roots &&
2339                     cur_new_count == nr_new_roots) {
2340                         if (cur_old_count == 0) {
2341                                 /* None -> exclusive/none */
2342
2343                                 if (cur_new_count != 0) {
2344                                         /* None -> exclusive */
2345                                         qg->excl += num_bytes;
2346                                         qg->excl_cmpr += num_bytes;
2347                                         dirty = true;
2348                                 }
2349                                 /* None -> none, nothing changed */
2350                         } else {
2351                                 /* Exclusive -> exclusive/none */
2352
2353                                 if (cur_new_count == 0) {
2354                                         /* Exclusive -> none */
2355                                         qg->excl -= num_bytes;
2356                                         qg->excl_cmpr -= num_bytes;
2357                                         dirty = true;
2358                                 }
2359                                 /* Exclusive -> exclusive, nothing changed */
2360                         }
2361                 }
2362
2363                 if (dirty)
2364                         qgroup_dirty(fs_info, qg);
2365         }
2366         return 0;
2367 }
2368
2369 /*
2370  * Check if the @roots potentially is a list of fs tree roots
2371  *
2372  * Return 0 for definitely not a fs/subvol tree roots ulist
2373  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2374  *          one as well)
2375  */
2376 static int maybe_fs_roots(struct ulist *roots)
2377 {
2378         struct ulist_node *unode;
2379         struct ulist_iterator uiter;
2380
2381         /* Empty one, still possible for fs roots */
2382         if (!roots || roots->nnodes == 0)
2383                 return 1;
2384
2385         ULIST_ITER_INIT(&uiter);
2386         unode = ulist_next(roots, &uiter);
2387         if (!unode)
2388                 return 1;
2389
2390         /*
2391          * If it contains fs tree roots, then it must belong to fs/subvol
2392          * trees.
2393          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2394          */
2395         return is_fstree(unode->val);
2396 }
2397
2398 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2399                                 u64 num_bytes, struct ulist *old_roots,
2400                                 struct ulist *new_roots)
2401 {
2402         struct btrfs_fs_info *fs_info = trans->fs_info;
2403         struct ulist *qgroups = NULL;
2404         struct ulist *tmp = NULL;
2405         u64 seq;
2406         u64 nr_new_roots = 0;
2407         u64 nr_old_roots = 0;
2408         int ret = 0;
2409
2410         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2411                 return 0;
2412
2413         if (new_roots) {
2414                 if (!maybe_fs_roots(new_roots))
2415                         goto out_free;
2416                 nr_new_roots = new_roots->nnodes;
2417         }
2418         if (old_roots) {
2419                 if (!maybe_fs_roots(old_roots))
2420                         goto out_free;
2421                 nr_old_roots = old_roots->nnodes;
2422         }
2423
2424         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2425         if (nr_old_roots == 0 && nr_new_roots == 0)
2426                 goto out_free;
2427
2428         BUG_ON(!fs_info->quota_root);
2429
2430         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2431                                         num_bytes, nr_old_roots, nr_new_roots);
2432
2433         qgroups = ulist_alloc(GFP_NOFS);
2434         if (!qgroups) {
2435                 ret = -ENOMEM;
2436                 goto out_free;
2437         }
2438         tmp = ulist_alloc(GFP_NOFS);
2439         if (!tmp) {
2440                 ret = -ENOMEM;
2441                 goto out_free;
2442         }
2443
2444         mutex_lock(&fs_info->qgroup_rescan_lock);
2445         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2446                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2447                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2448                         ret = 0;
2449                         goto out_free;
2450                 }
2451         }
2452         mutex_unlock(&fs_info->qgroup_rescan_lock);
2453
2454         spin_lock(&fs_info->qgroup_lock);
2455         seq = fs_info->qgroup_seq;
2456
2457         /* Update old refcnts using old_roots */
2458         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2459                                    UPDATE_OLD);
2460         if (ret < 0)
2461                 goto out;
2462
2463         /* Update new refcnts using new_roots */
2464         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2465                                    UPDATE_NEW);
2466         if (ret < 0)
2467                 goto out;
2468
2469         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2470                                num_bytes, seq);
2471
2472         /*
2473          * Bump qgroup_seq to avoid seq overlap
2474          */
2475         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2476 out:
2477         spin_unlock(&fs_info->qgroup_lock);
2478 out_free:
2479         ulist_free(tmp);
2480         ulist_free(qgroups);
2481         ulist_free(old_roots);
2482         ulist_free(new_roots);
2483         return ret;
2484 }
2485
2486 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2487 {
2488         struct btrfs_fs_info *fs_info = trans->fs_info;
2489         struct btrfs_qgroup_extent_record *record;
2490         struct btrfs_delayed_ref_root *delayed_refs;
2491         struct ulist *new_roots = NULL;
2492         struct rb_node *node;
2493         u64 num_dirty_extents = 0;
2494         u64 qgroup_to_skip;
2495         int ret = 0;
2496
2497         delayed_refs = &trans->transaction->delayed_refs;
2498         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2499         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2500                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2501                                   node);
2502
2503                 num_dirty_extents++;
2504                 trace_btrfs_qgroup_account_extents(fs_info, record);
2505
2506                 if (!ret) {
2507                         /*
2508                          * Old roots should be searched when inserting qgroup
2509                          * extent record
2510                          */
2511                         if (WARN_ON(!record->old_roots)) {
2512                                 /* Search commit root to find old_roots */
2513                                 ret = btrfs_find_all_roots(NULL, fs_info,
2514                                                 record->bytenr, 0,
2515                                                 &record->old_roots, false);
2516                                 if (ret < 0)
2517                                         goto cleanup;
2518                         }
2519
2520                         /*
2521                          * Use SEQ_LAST as time_seq to do special search, which
2522                          * doesn't lock tree or delayed_refs and search current
2523                          * root. It's safe inside commit_transaction().
2524                          */
2525                         ret = btrfs_find_all_roots(trans, fs_info,
2526                                 record->bytenr, SEQ_LAST, &new_roots, false);
2527                         if (ret < 0)
2528                                 goto cleanup;
2529                         if (qgroup_to_skip) {
2530                                 ulist_del(new_roots, qgroup_to_skip, 0);
2531                                 ulist_del(record->old_roots, qgroup_to_skip,
2532                                           0);
2533                         }
2534                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2535                                                           record->num_bytes,
2536                                                           record->old_roots,
2537                                                           new_roots);
2538                         record->old_roots = NULL;
2539                         new_roots = NULL;
2540                 }
2541 cleanup:
2542                 ulist_free(record->old_roots);
2543                 ulist_free(new_roots);
2544                 new_roots = NULL;
2545                 rb_erase(node, &delayed_refs->dirty_extent_root);
2546                 kfree(record);
2547
2548         }
2549         trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2550                                        num_dirty_extents);
2551         return ret;
2552 }
2553
2554 /*
2555  * called from commit_transaction. Writes all changed qgroups to disk.
2556  */
2557 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2558 {
2559         struct btrfs_fs_info *fs_info = trans->fs_info;
2560         struct btrfs_root *quota_root = fs_info->quota_root;
2561         int ret = 0;
2562
2563         if (!quota_root)
2564                 return ret;
2565
2566         spin_lock(&fs_info->qgroup_lock);
2567         while (!list_empty(&fs_info->dirty_qgroups)) {
2568                 struct btrfs_qgroup *qgroup;
2569                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2570                                           struct btrfs_qgroup, dirty);
2571                 list_del_init(&qgroup->dirty);
2572                 spin_unlock(&fs_info->qgroup_lock);
2573                 ret = update_qgroup_info_item(trans, qgroup);
2574                 if (ret)
2575                         fs_info->qgroup_flags |=
2576                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2577                 ret = update_qgroup_limit_item(trans, qgroup);
2578                 if (ret)
2579                         fs_info->qgroup_flags |=
2580                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2581                 spin_lock(&fs_info->qgroup_lock);
2582         }
2583         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2584                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2585         else
2586                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2587         spin_unlock(&fs_info->qgroup_lock);
2588
2589         ret = update_qgroup_status_item(trans);
2590         if (ret)
2591                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2592
2593         return ret;
2594 }
2595
2596 /*
2597  * Copy the accounting information between qgroups. This is necessary
2598  * when a snapshot or a subvolume is created. Throwing an error will
2599  * cause a transaction abort so we take extra care here to only error
2600  * when a readonly fs is a reasonable outcome.
2601  */
2602 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2603                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2604 {
2605         int ret = 0;
2606         int i;
2607         u64 *i_qgroups;
2608         struct btrfs_fs_info *fs_info = trans->fs_info;
2609         struct btrfs_root *quota_root;
2610         struct btrfs_qgroup *srcgroup;
2611         struct btrfs_qgroup *dstgroup;
2612         u32 level_size = 0;
2613         u64 nums;
2614
2615         mutex_lock(&fs_info->qgroup_ioctl_lock);
2616         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2617                 goto out;
2618
2619         quota_root = fs_info->quota_root;
2620         if (!quota_root) {
2621                 ret = -EINVAL;
2622                 goto out;
2623         }
2624
2625         if (inherit) {
2626                 i_qgroups = (u64 *)(inherit + 1);
2627                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2628                        2 * inherit->num_excl_copies;
2629                 for (i = 0; i < nums; ++i) {
2630                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2631
2632                         /*
2633                          * Zero out invalid groups so we can ignore
2634                          * them later.
2635                          */
2636                         if (!srcgroup ||
2637                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2638                                 *i_qgroups = 0ULL;
2639
2640                         ++i_qgroups;
2641                 }
2642         }
2643
2644         /*
2645          * create a tracking group for the subvol itself
2646          */
2647         ret = add_qgroup_item(trans, quota_root, objectid);
2648         if (ret)
2649                 goto out;
2650
2651         /*
2652          * add qgroup to all inherited groups
2653          */
2654         if (inherit) {
2655                 i_qgroups = (u64 *)(inherit + 1);
2656                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2657                         if (*i_qgroups == 0)
2658                                 continue;
2659                         ret = add_qgroup_relation_item(trans, objectid,
2660                                                        *i_qgroups);
2661                         if (ret && ret != -EEXIST)
2662                                 goto out;
2663                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2664                                                        objectid);
2665                         if (ret && ret != -EEXIST)
2666                                 goto out;
2667                 }
2668                 ret = 0;
2669         }
2670
2671
2672         spin_lock(&fs_info->qgroup_lock);
2673
2674         dstgroup = add_qgroup_rb(fs_info, objectid);
2675         if (IS_ERR(dstgroup)) {
2676                 ret = PTR_ERR(dstgroup);
2677                 goto unlock;
2678         }
2679
2680         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2681                 dstgroup->lim_flags = inherit->lim.flags;
2682                 dstgroup->max_rfer = inherit->lim.max_rfer;
2683                 dstgroup->max_excl = inherit->lim.max_excl;
2684                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2685                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2686
2687                 ret = update_qgroup_limit_item(trans, dstgroup);
2688                 if (ret) {
2689                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2690                         btrfs_info(fs_info,
2691                                    "unable to update quota limit for %llu",
2692                                    dstgroup->qgroupid);
2693                         goto unlock;
2694                 }
2695         }
2696
2697         if (srcid) {
2698                 srcgroup = find_qgroup_rb(fs_info, srcid);
2699                 if (!srcgroup)
2700                         goto unlock;
2701
2702                 /*
2703                  * We call inherit after we clone the root in order to make sure
2704                  * our counts don't go crazy, so at this point the only
2705                  * difference between the two roots should be the root node.
2706                  */
2707                 level_size = fs_info->nodesize;
2708                 dstgroup->rfer = srcgroup->rfer;
2709                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2710                 dstgroup->excl = level_size;
2711                 dstgroup->excl_cmpr = level_size;
2712                 srcgroup->excl = level_size;
2713                 srcgroup->excl_cmpr = level_size;
2714
2715                 /* inherit the limit info */
2716                 dstgroup->lim_flags = srcgroup->lim_flags;
2717                 dstgroup->max_rfer = srcgroup->max_rfer;
2718                 dstgroup->max_excl = srcgroup->max_excl;
2719                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2720                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2721
2722                 qgroup_dirty(fs_info, dstgroup);
2723                 qgroup_dirty(fs_info, srcgroup);
2724         }
2725
2726         if (!inherit)
2727                 goto unlock;
2728
2729         i_qgroups = (u64 *)(inherit + 1);
2730         for (i = 0; i < inherit->num_qgroups; ++i) {
2731                 if (*i_qgroups) {
2732                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2733                         if (ret)
2734                                 goto unlock;
2735                 }
2736                 ++i_qgroups;
2737         }
2738
2739         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2740                 struct btrfs_qgroup *src;
2741                 struct btrfs_qgroup *dst;
2742
2743                 if (!i_qgroups[0] || !i_qgroups[1])
2744                         continue;
2745
2746                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2747                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2748
2749                 if (!src || !dst) {
2750                         ret = -EINVAL;
2751                         goto unlock;
2752                 }
2753
2754                 dst->rfer = src->rfer - level_size;
2755                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2756         }
2757         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2758                 struct btrfs_qgroup *src;
2759                 struct btrfs_qgroup *dst;
2760
2761                 if (!i_qgroups[0] || !i_qgroups[1])
2762                         continue;
2763
2764                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2765                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2766
2767                 if (!src || !dst) {
2768                         ret = -EINVAL;
2769                         goto unlock;
2770                 }
2771
2772                 dst->excl = src->excl + level_size;
2773                 dst->excl_cmpr = src->excl_cmpr + level_size;
2774         }
2775
2776 unlock:
2777         spin_unlock(&fs_info->qgroup_lock);
2778 out:
2779         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2780         return ret;
2781 }
2782
2783 /*
2784  * Two limits to commit transaction in advance.
2785  *
2786  * For RATIO, it will be 1/RATIO of the remaining limit
2787  * (excluding data and prealloc meta) as threshold.
2788  * For SIZE, it will be in byte unit as threshold.
2789  */
2790 #define QGROUP_PERTRANS_RATIO           32
2791 #define QGROUP_PERTRANS_SIZE            SZ_32M
2792 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2793                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2794 {
2795         u64 limit;
2796         u64 threshold;
2797
2798         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2799             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2800                 return false;
2801
2802         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2803             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2804                 return false;
2805
2806         /*
2807          * Even if we passed the check, it's better to check if reservation
2808          * for meta_pertrans is pushing us near limit.
2809          * If there is too much pertrans reservation or it's near the limit,
2810          * let's try commit transaction to free some, using transaction_kthread
2811          */
2812         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2813                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2814                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2815                         limit = qg->max_excl;
2816                 else
2817                         limit = qg->max_rfer;
2818                 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2819                             qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2820                             QGROUP_PERTRANS_RATIO;
2821                 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2822
2823                 /*
2824                  * Use transaction_kthread to commit transaction, so we no
2825                  * longer need to bother nested transaction nor lock context.
2826                  */
2827                 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2828                         btrfs_commit_transaction_locksafe(fs_info);
2829         }
2830
2831         return true;
2832 }
2833
2834 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2835                           enum btrfs_qgroup_rsv_type type)
2836 {
2837         struct btrfs_root *quota_root;
2838         struct btrfs_qgroup *qgroup;
2839         struct btrfs_fs_info *fs_info = root->fs_info;
2840         u64 ref_root = root->root_key.objectid;
2841         int ret = 0;
2842         struct ulist_node *unode;
2843         struct ulist_iterator uiter;
2844
2845         if (!is_fstree(ref_root))
2846                 return 0;
2847
2848         if (num_bytes == 0)
2849                 return 0;
2850
2851         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2852             capable(CAP_SYS_RESOURCE))
2853                 enforce = false;
2854
2855         spin_lock(&fs_info->qgroup_lock);
2856         quota_root = fs_info->quota_root;
2857         if (!quota_root)
2858                 goto out;
2859
2860         qgroup = find_qgroup_rb(fs_info, ref_root);
2861         if (!qgroup)
2862                 goto out;
2863
2864         /*
2865          * in a first step, we check all affected qgroups if any limits would
2866          * be exceeded
2867          */
2868         ulist_reinit(fs_info->qgroup_ulist);
2869         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2870                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2871         if (ret < 0)
2872                 goto out;
2873         ULIST_ITER_INIT(&uiter);
2874         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2875                 struct btrfs_qgroup *qg;
2876                 struct btrfs_qgroup_list *glist;
2877
2878                 qg = unode_aux_to_qgroup(unode);
2879
2880                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2881                         ret = -EDQUOT;
2882                         goto out;
2883                 }
2884
2885                 list_for_each_entry(glist, &qg->groups, next_group) {
2886                         ret = ulist_add(fs_info->qgroup_ulist,
2887                                         glist->group->qgroupid,
2888                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2889                         if (ret < 0)
2890                                 goto out;
2891                 }
2892         }
2893         ret = 0;
2894         /*
2895          * no limits exceeded, now record the reservation into all qgroups
2896          */
2897         ULIST_ITER_INIT(&uiter);
2898         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2899                 struct btrfs_qgroup *qg;
2900
2901                 qg = unode_aux_to_qgroup(unode);
2902
2903                 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2904                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2905         }
2906
2907 out:
2908         spin_unlock(&fs_info->qgroup_lock);
2909         return ret;
2910 }
2911
2912 /*
2913  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2914  * qgroup).
2915  *
2916  * Will handle all higher level qgroup too.
2917  *
2918  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2919  * This special case is only used for META_PERTRANS type.
2920  */
2921 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2922                                u64 ref_root, u64 num_bytes,
2923                                enum btrfs_qgroup_rsv_type type)
2924 {
2925         struct btrfs_root *quota_root;
2926         struct btrfs_qgroup *qgroup;
2927         struct ulist_node *unode;
2928         struct ulist_iterator uiter;
2929         int ret = 0;
2930
2931         if (!is_fstree(ref_root))
2932                 return;
2933
2934         if (num_bytes == 0)
2935                 return;
2936
2937         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2938                 WARN(1, "%s: Invalid type to free", __func__);
2939                 return;
2940         }
2941         spin_lock(&fs_info->qgroup_lock);
2942
2943         quota_root = fs_info->quota_root;
2944         if (!quota_root)
2945                 goto out;
2946
2947         qgroup = find_qgroup_rb(fs_info, ref_root);
2948         if (!qgroup)
2949                 goto out;
2950
2951         if (num_bytes == (u64)-1)
2952                 /*
2953                  * We're freeing all pertrans rsv, get reserved value from
2954                  * level 0 qgroup as real num_bytes to free.
2955                  */
2956                 num_bytes = qgroup->rsv.values[type];
2957
2958         ulist_reinit(fs_info->qgroup_ulist);
2959         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2960                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2961         if (ret < 0)
2962                 goto out;
2963         ULIST_ITER_INIT(&uiter);
2964         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2965                 struct btrfs_qgroup *qg;
2966                 struct btrfs_qgroup_list *glist;
2967
2968                 qg = unode_aux_to_qgroup(unode);
2969
2970                 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2971                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2972
2973                 list_for_each_entry(glist, &qg->groups, next_group) {
2974                         ret = ulist_add(fs_info->qgroup_ulist,
2975                                         glist->group->qgroupid,
2976                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2977                         if (ret < 0)
2978                                 goto out;
2979                 }
2980         }
2981
2982 out:
2983         spin_unlock(&fs_info->qgroup_lock);
2984 }
2985
2986 /*
2987  * Check if the leaf is the last leaf. Which means all node pointers
2988  * are at their last position.
2989  */
2990 static bool is_last_leaf(struct btrfs_path *path)
2991 {
2992         int i;
2993
2994         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2995                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2996                         return false;
2997         }
2998         return true;
2999 }
3000
3001 /*
3002  * returns < 0 on error, 0 when more leafs are to be scanned.
3003  * returns 1 when done.
3004  */
3005 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3006                               struct btrfs_path *path)
3007 {
3008         struct btrfs_fs_info *fs_info = trans->fs_info;
3009         struct btrfs_key found;
3010         struct extent_buffer *scratch_leaf = NULL;
3011         struct ulist *roots = NULL;
3012         u64 num_bytes;
3013         bool done;
3014         int slot;
3015         int ret;
3016
3017         mutex_lock(&fs_info->qgroup_rescan_lock);
3018         ret = btrfs_search_slot_for_read(fs_info->extent_root,
3019                                          &fs_info->qgroup_rescan_progress,
3020                                          path, 1, 0);
3021
3022         btrfs_debug(fs_info,
3023                 "current progress key (%llu %u %llu), search_slot ret %d",
3024                 fs_info->qgroup_rescan_progress.objectid,
3025                 fs_info->qgroup_rescan_progress.type,
3026                 fs_info->qgroup_rescan_progress.offset, ret);
3027
3028         if (ret) {
3029                 /*
3030                  * The rescan is about to end, we will not be scanning any
3031                  * further blocks. We cannot unset the RESCAN flag here, because
3032                  * we want to commit the transaction if everything went well.
3033                  * To make the live accounting work in this phase, we set our
3034                  * scan progress pointer such that every real extent objectid
3035                  * will be smaller.
3036                  */
3037                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3038                 btrfs_release_path(path);
3039                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3040                 return ret;
3041         }
3042         done = is_last_leaf(path);
3043
3044         btrfs_item_key_to_cpu(path->nodes[0], &found,
3045                               btrfs_header_nritems(path->nodes[0]) - 1);
3046         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3047
3048         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3049         if (!scratch_leaf) {
3050                 ret = -ENOMEM;
3051                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3052                 goto out;
3053         }
3054         slot = path->slots[0];
3055         btrfs_release_path(path);
3056         mutex_unlock(&fs_info->qgroup_rescan_lock);
3057
3058         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3059                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3060                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3061                     found.type != BTRFS_METADATA_ITEM_KEY)
3062                         continue;
3063                 if (found.type == BTRFS_METADATA_ITEM_KEY)
3064                         num_bytes = fs_info->nodesize;
3065                 else
3066                         num_bytes = found.offset;
3067
3068                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3069                                            &roots, false);
3070                 if (ret < 0)
3071                         goto out;
3072                 /* For rescan, just pass old_roots as NULL */
3073                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3074                                                   num_bytes, NULL, roots);
3075                 if (ret < 0)
3076                         goto out;
3077         }
3078 out:
3079         if (scratch_leaf)
3080                 free_extent_buffer(scratch_leaf);
3081
3082         if (done && !ret) {
3083                 ret = 1;
3084                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3085         }
3086         return ret;
3087 }
3088
3089 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3090 {
3091         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3092                                                      qgroup_rescan_work);
3093         struct btrfs_path *path;
3094         struct btrfs_trans_handle *trans = NULL;
3095         int err = -ENOMEM;
3096         int ret = 0;
3097
3098         path = btrfs_alloc_path();
3099         if (!path)
3100                 goto out;
3101         /*
3102          * Rescan should only search for commit root, and any later difference
3103          * should be recorded by qgroup
3104          */
3105         path->search_commit_root = 1;
3106         path->skip_locking = 1;
3107
3108         err = 0;
3109         while (!err && !btrfs_fs_closing(fs_info)) {
3110                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3111                 if (IS_ERR(trans)) {
3112                         err = PTR_ERR(trans);
3113                         break;
3114                 }
3115                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3116                         err = -EINTR;
3117                 } else {
3118                         err = qgroup_rescan_leaf(trans, path);
3119                 }
3120                 if (err > 0)
3121                         btrfs_commit_transaction(trans);
3122                 else
3123                         btrfs_end_transaction(trans);
3124         }
3125
3126 out:
3127         btrfs_free_path(path);
3128
3129         mutex_lock(&fs_info->qgroup_rescan_lock);
3130         if (!btrfs_fs_closing(fs_info))
3131                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3132
3133         if (err > 0 &&
3134             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3135                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3136         } else if (err < 0) {
3137                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3138         }
3139         mutex_unlock(&fs_info->qgroup_rescan_lock);
3140
3141         /*
3142          * only update status, since the previous part has already updated the
3143          * qgroup info.
3144          */
3145         trans = btrfs_start_transaction(fs_info->quota_root, 1);
3146         if (IS_ERR(trans)) {
3147                 err = PTR_ERR(trans);
3148                 btrfs_err(fs_info,
3149                           "fail to start transaction for status update: %d",
3150                           err);
3151                 goto done;
3152         }
3153         ret = update_qgroup_status_item(trans);
3154         if (ret < 0) {
3155                 err = ret;
3156                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3157         }
3158         btrfs_end_transaction(trans);
3159
3160         if (btrfs_fs_closing(fs_info)) {
3161                 btrfs_info(fs_info, "qgroup scan paused");
3162         } else if (err >= 0) {
3163                 btrfs_info(fs_info, "qgroup scan completed%s",
3164                         err > 0 ? " (inconsistency flag cleared)" : "");
3165         } else {
3166                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3167         }
3168
3169 done:
3170         mutex_lock(&fs_info->qgroup_rescan_lock);
3171         fs_info->qgroup_rescan_running = false;
3172         mutex_unlock(&fs_info->qgroup_rescan_lock);
3173         complete_all(&fs_info->qgroup_rescan_completion);
3174 }
3175
3176 /*
3177  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3178  * memory required for the rescan context.
3179  */
3180 static int
3181 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3182                    int init_flags)
3183 {
3184         int ret = 0;
3185
3186         if (!init_flags) {
3187                 /* we're resuming qgroup rescan at mount time */
3188                 if (!(fs_info->qgroup_flags &
3189                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3190                         btrfs_warn(fs_info,
3191                         "qgroup rescan init failed, qgroup is not enabled");
3192                         ret = -EINVAL;
3193                 } else if (!(fs_info->qgroup_flags &
3194                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3195                         btrfs_warn(fs_info,
3196                         "qgroup rescan init failed, qgroup rescan is not queued");
3197                         ret = -EINVAL;
3198                 }
3199
3200                 if (ret)
3201                         return ret;
3202         }
3203
3204         mutex_lock(&fs_info->qgroup_rescan_lock);
3205         spin_lock(&fs_info->qgroup_lock);
3206
3207         if (init_flags) {
3208                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3209                         btrfs_warn(fs_info,
3210                                    "qgroup rescan is already in progress");
3211                         ret = -EINPROGRESS;
3212                 } else if (!(fs_info->qgroup_flags &
3213                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3214                         btrfs_warn(fs_info,
3215                         "qgroup rescan init failed, qgroup is not enabled");
3216                         ret = -EINVAL;
3217                 }
3218
3219                 if (ret) {
3220                         spin_unlock(&fs_info->qgroup_lock);
3221                         mutex_unlock(&fs_info->qgroup_rescan_lock);
3222                         return ret;
3223                 }
3224                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3225         }
3226
3227         memset(&fs_info->qgroup_rescan_progress, 0,
3228                 sizeof(fs_info->qgroup_rescan_progress));
3229         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3230         init_completion(&fs_info->qgroup_rescan_completion);
3231         fs_info->qgroup_rescan_running = true;
3232
3233         spin_unlock(&fs_info->qgroup_lock);
3234         mutex_unlock(&fs_info->qgroup_rescan_lock);
3235
3236         memset(&fs_info->qgroup_rescan_work, 0,
3237                sizeof(fs_info->qgroup_rescan_work));
3238         btrfs_init_work(&fs_info->qgroup_rescan_work,
3239                         btrfs_qgroup_rescan_helper,
3240                         btrfs_qgroup_rescan_worker, NULL, NULL);
3241         return 0;
3242 }
3243
3244 static void
3245 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3246 {
3247         struct rb_node *n;
3248         struct btrfs_qgroup *qgroup;
3249
3250         spin_lock(&fs_info->qgroup_lock);
3251         /* clear all current qgroup tracking information */
3252         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3253                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3254                 qgroup->rfer = 0;
3255                 qgroup->rfer_cmpr = 0;
3256                 qgroup->excl = 0;
3257                 qgroup->excl_cmpr = 0;
3258                 qgroup_dirty(fs_info, qgroup);
3259         }
3260         spin_unlock(&fs_info->qgroup_lock);
3261 }
3262
3263 int
3264 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3265 {
3266         int ret = 0;
3267         struct btrfs_trans_handle *trans;
3268
3269         ret = qgroup_rescan_init(fs_info, 0, 1);
3270         if (ret)
3271                 return ret;
3272
3273         /*
3274          * We have set the rescan_progress to 0, which means no more
3275          * delayed refs will be accounted by btrfs_qgroup_account_ref.
3276          * However, btrfs_qgroup_account_ref may be right after its call
3277          * to btrfs_find_all_roots, in which case it would still do the
3278          * accounting.
3279          * To solve this, we're committing the transaction, which will
3280          * ensure we run all delayed refs and only after that, we are
3281          * going to clear all tracking information for a clean start.
3282          */
3283
3284         trans = btrfs_join_transaction(fs_info->fs_root);
3285         if (IS_ERR(trans)) {
3286                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3287                 return PTR_ERR(trans);
3288         }
3289         ret = btrfs_commit_transaction(trans);
3290         if (ret) {
3291                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3292                 return ret;
3293         }
3294
3295         qgroup_rescan_zero_tracking(fs_info);
3296
3297         btrfs_queue_work(fs_info->qgroup_rescan_workers,
3298                          &fs_info->qgroup_rescan_work);
3299
3300         return 0;
3301 }
3302
3303 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3304                                      bool interruptible)
3305 {
3306         int running;
3307         int ret = 0;
3308
3309         mutex_lock(&fs_info->qgroup_rescan_lock);
3310         spin_lock(&fs_info->qgroup_lock);
3311         running = fs_info->qgroup_rescan_running;
3312         spin_unlock(&fs_info->qgroup_lock);
3313         mutex_unlock(&fs_info->qgroup_rescan_lock);
3314
3315         if (!running)
3316                 return 0;
3317
3318         if (interruptible)
3319                 ret = wait_for_completion_interruptible(
3320                                         &fs_info->qgroup_rescan_completion);
3321         else
3322                 wait_for_completion(&fs_info->qgroup_rescan_completion);
3323
3324         return ret;
3325 }
3326
3327 /*
3328  * this is only called from open_ctree where we're still single threaded, thus
3329  * locking is omitted here.
3330  */
3331 void
3332 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3333 {
3334         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3335                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3336                                  &fs_info->qgroup_rescan_work);
3337 }
3338
3339 /*
3340  * Reserve qgroup space for range [start, start + len).
3341  *
3342  * This function will either reserve space from related qgroups or doing
3343  * nothing if the range is already reserved.
3344  *
3345  * Return 0 for successful reserve
3346  * Return <0 for error (including -EQUOT)
3347  *
3348  * NOTE: this function may sleep for memory allocation.
3349  *       if btrfs_qgroup_reserve_data() is called multiple times with
3350  *       same @reserved, caller must ensure when error happens it's OK
3351  *       to free *ALL* reserved space.
3352  */
3353 int btrfs_qgroup_reserve_data(struct inode *inode,
3354                         struct extent_changeset **reserved_ret, u64 start,
3355                         u64 len)
3356 {
3357         struct btrfs_root *root = BTRFS_I(inode)->root;
3358         struct ulist_node *unode;
3359         struct ulist_iterator uiter;
3360         struct extent_changeset *reserved;
3361         u64 orig_reserved;
3362         u64 to_reserve;
3363         int ret;
3364
3365         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3366             !is_fstree(root->root_key.objectid) || len == 0)
3367                 return 0;
3368
3369         /* @reserved parameter is mandatory for qgroup */
3370         if (WARN_ON(!reserved_ret))
3371                 return -EINVAL;
3372         if (!*reserved_ret) {
3373                 *reserved_ret = extent_changeset_alloc();
3374                 if (!*reserved_ret)
3375                         return -ENOMEM;
3376         }
3377         reserved = *reserved_ret;
3378         /* Record already reserved space */
3379         orig_reserved = reserved->bytes_changed;
3380         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3381                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3382
3383         /* Newly reserved space */
3384         to_reserve = reserved->bytes_changed - orig_reserved;
3385         trace_btrfs_qgroup_reserve_data(inode, start, len,
3386                                         to_reserve, QGROUP_RESERVE);
3387         if (ret < 0)
3388                 goto cleanup;
3389         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3390         if (ret < 0)
3391                 goto cleanup;
3392
3393         return ret;
3394
3395 cleanup:
3396         /* cleanup *ALL* already reserved ranges */
3397         ULIST_ITER_INIT(&uiter);
3398         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3399                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3400                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3401         extent_changeset_release(reserved);
3402         return ret;
3403 }
3404
3405 /* Free ranges specified by @reserved, normally in error path */
3406 static int qgroup_free_reserved_data(struct inode *inode,
3407                         struct extent_changeset *reserved, u64 start, u64 len)
3408 {
3409         struct btrfs_root *root = BTRFS_I(inode)->root;
3410         struct ulist_node *unode;
3411         struct ulist_iterator uiter;
3412         struct extent_changeset changeset;
3413         int freed = 0;
3414         int ret;
3415
3416         extent_changeset_init(&changeset);
3417         len = round_up(start + len, root->fs_info->sectorsize);
3418         start = round_down(start, root->fs_info->sectorsize);
3419
3420         ULIST_ITER_INIT(&uiter);
3421         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3422                 u64 range_start = unode->val;
3423                 /* unode->aux is the inclusive end */
3424                 u64 range_len = unode->aux - range_start + 1;
3425                 u64 free_start;
3426                 u64 free_len;
3427
3428                 extent_changeset_release(&changeset);
3429
3430                 /* Only free range in range [start, start + len) */
3431                 if (range_start >= start + len ||
3432                     range_start + range_len <= start)
3433                         continue;
3434                 free_start = max(range_start, start);
3435                 free_len = min(start + len, range_start + range_len) -
3436                            free_start;
3437                 /*
3438                  * TODO: To also modify reserved->ranges_reserved to reflect
3439                  * the modification.
3440                  *
3441                  * However as long as we free qgroup reserved according to
3442                  * EXTENT_QGROUP_RESERVED, we won't double free.
3443                  * So not need to rush.
3444                  */
3445                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3446                                 free_start, free_start + free_len - 1,
3447                                 EXTENT_QGROUP_RESERVED, &changeset);
3448                 if (ret < 0)
3449                         goto out;
3450                 freed += changeset.bytes_changed;
3451         }
3452         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3453                                   BTRFS_QGROUP_RSV_DATA);
3454         ret = freed;
3455 out:
3456         extent_changeset_release(&changeset);
3457         return ret;
3458 }
3459
3460 static int __btrfs_qgroup_release_data(struct inode *inode,
3461                         struct extent_changeset *reserved, u64 start, u64 len,
3462                         int free)
3463 {
3464         struct extent_changeset changeset;
3465         int trace_op = QGROUP_RELEASE;
3466         int ret;
3467
3468         if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3469                       &BTRFS_I(inode)->root->fs_info->flags))
3470                 return 0;
3471
3472         /* In release case, we shouldn't have @reserved */
3473         WARN_ON(!free && reserved);
3474         if (free && reserved)
3475                 return qgroup_free_reserved_data(inode, reserved, start, len);
3476         extent_changeset_init(&changeset);
3477         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start, 
3478                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3479         if (ret < 0)
3480                 goto out;
3481
3482         if (free)
3483                 trace_op = QGROUP_FREE;
3484         trace_btrfs_qgroup_release_data(inode, start, len,
3485                                         changeset.bytes_changed, trace_op);
3486         if (free)
3487                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3488                                 BTRFS_I(inode)->root->root_key.objectid,
3489                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3490         ret = changeset.bytes_changed;
3491 out:
3492         extent_changeset_release(&changeset);
3493         return ret;
3494 }
3495
3496 /*
3497  * Free a reserved space range from io_tree and related qgroups
3498  *
3499  * Should be called when a range of pages get invalidated before reaching disk.
3500  * Or for error cleanup case.
3501  * if @reserved is given, only reserved range in [@start, @start + @len) will
3502  * be freed.
3503  *
3504  * For data written to disk, use btrfs_qgroup_release_data().
3505  *
3506  * NOTE: This function may sleep for memory allocation.
3507  */
3508 int btrfs_qgroup_free_data(struct inode *inode,
3509                         struct extent_changeset *reserved, u64 start, u64 len)
3510 {
3511         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3512 }
3513
3514 /*
3515  * Release a reserved space range from io_tree only.
3516  *
3517  * Should be called when a range of pages get written to disk and corresponding
3518  * FILE_EXTENT is inserted into corresponding root.
3519  *
3520  * Since new qgroup accounting framework will only update qgroup numbers at
3521  * commit_transaction() time, its reserved space shouldn't be freed from
3522  * related qgroups.
3523  *
3524  * But we should release the range from io_tree, to allow further write to be
3525  * COWed.
3526  *
3527  * NOTE: This function may sleep for memory allocation.
3528  */
3529 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3530 {
3531         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3532 }
3533
3534 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3535                               enum btrfs_qgroup_rsv_type type)
3536 {
3537         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3538             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3539                 return;
3540         if (num_bytes == 0)
3541                 return;
3542
3543         spin_lock(&root->qgroup_meta_rsv_lock);
3544         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3545                 root->qgroup_meta_rsv_prealloc += num_bytes;
3546         else
3547                 root->qgroup_meta_rsv_pertrans += num_bytes;
3548         spin_unlock(&root->qgroup_meta_rsv_lock);
3549 }
3550
3551 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3552                              enum btrfs_qgroup_rsv_type type)
3553 {
3554         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3555             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3556                 return 0;
3557         if (num_bytes == 0)
3558                 return 0;
3559
3560         spin_lock(&root->qgroup_meta_rsv_lock);
3561         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3562                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3563                                   num_bytes);
3564                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3565         } else {
3566                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3567                                   num_bytes);
3568                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3569         }
3570         spin_unlock(&root->qgroup_meta_rsv_lock);
3571         return num_bytes;
3572 }
3573
3574 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3575                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3576 {
3577         struct btrfs_fs_info *fs_info = root->fs_info;
3578         int ret;
3579
3580         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3581             !is_fstree(root->root_key.objectid) || num_bytes == 0)
3582                 return 0;
3583
3584         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3585         trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3586         ret = qgroup_reserve(root, num_bytes, enforce, type);
3587         if (ret < 0)
3588                 return ret;
3589         /*
3590          * Record what we have reserved into root.
3591          *
3592          * To avoid quota disabled->enabled underflow.
3593          * In that case, we may try to free space we haven't reserved
3594          * (since quota was disabled), so record what we reserved into root.
3595          * And ensure later release won't underflow this number.
3596          */
3597         add_root_meta_rsv(root, num_bytes, type);
3598         return ret;
3599 }
3600
3601 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3602 {
3603         struct btrfs_fs_info *fs_info = root->fs_info;
3604
3605         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3606             !is_fstree(root->root_key.objectid))
3607                 return;
3608
3609         /* TODO: Update trace point to handle such free */
3610         trace_qgroup_meta_free_all_pertrans(root);
3611         /* Special value -1 means to free all reserved space */
3612         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3613                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3614 }
3615
3616 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3617                               enum btrfs_qgroup_rsv_type type)
3618 {
3619         struct btrfs_fs_info *fs_info = root->fs_info;
3620
3621         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3622             !is_fstree(root->root_key.objectid))
3623                 return;
3624
3625         /*
3626          * reservation for META_PREALLOC can happen before quota is enabled,
3627          * which can lead to underflow.
3628          * Here ensure we will only free what we really have reserved.
3629          */
3630         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3631         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3632         trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3633         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3634                                   num_bytes, type);
3635 }
3636
3637 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3638                                 int num_bytes)
3639 {
3640         struct btrfs_root *quota_root = fs_info->quota_root;
3641         struct btrfs_qgroup *qgroup;
3642         struct ulist_node *unode;
3643         struct ulist_iterator uiter;
3644         int ret = 0;
3645
3646         if (num_bytes == 0)
3647                 return;
3648         if (!quota_root)
3649                 return;
3650
3651         spin_lock(&fs_info->qgroup_lock);
3652         qgroup = find_qgroup_rb(fs_info, ref_root);
3653         if (!qgroup)
3654                 goto out;
3655         ulist_reinit(fs_info->qgroup_ulist);
3656         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3657                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3658         if (ret < 0)
3659                 goto out;
3660         ULIST_ITER_INIT(&uiter);
3661         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3662                 struct btrfs_qgroup *qg;
3663                 struct btrfs_qgroup_list *glist;
3664
3665                 qg = unode_aux_to_qgroup(unode);
3666
3667                 qgroup_rsv_release(fs_info, qg, num_bytes,
3668                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3669                 qgroup_rsv_add(fs_info, qg, num_bytes,
3670                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3671                 list_for_each_entry(glist, &qg->groups, next_group) {
3672                         ret = ulist_add(fs_info->qgroup_ulist,
3673                                         glist->group->qgroupid,
3674                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3675                         if (ret < 0)
3676                                 goto out;
3677                 }
3678         }
3679 out:
3680         spin_unlock(&fs_info->qgroup_lock);
3681 }
3682
3683 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3684 {
3685         struct btrfs_fs_info *fs_info = root->fs_info;
3686
3687         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3688             !is_fstree(root->root_key.objectid))
3689                 return;
3690         /* Same as btrfs_qgroup_free_meta_prealloc() */
3691         num_bytes = sub_root_meta_rsv(root, num_bytes,
3692                                       BTRFS_QGROUP_RSV_META_PREALLOC);
3693         trace_qgroup_meta_convert(root, num_bytes);
3694         qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3695 }
3696
3697 /*
3698  * Check qgroup reserved space leaking, normally at destroy inode
3699  * time
3700  */
3701 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3702 {
3703         struct extent_changeset changeset;
3704         struct ulist_node *unode;
3705         struct ulist_iterator iter;
3706         int ret;
3707
3708         extent_changeset_init(&changeset);
3709         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3710                         EXTENT_QGROUP_RESERVED, &changeset);
3711
3712         WARN_ON(ret < 0);
3713         if (WARN_ON(changeset.bytes_changed)) {
3714                 ULIST_ITER_INIT(&iter);
3715                 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3716                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
3717                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3718                                 inode->i_ino, unode->val, unode->aux);
3719                 }
3720                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3721                                 BTRFS_I(inode)->root->root_key.objectid,
3722                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3723
3724         }
3725         extent_changeset_release(&changeset);
3726 }
3727
3728 void btrfs_qgroup_init_swapped_blocks(
3729         struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3730 {
3731         int i;
3732
3733         spin_lock_init(&swapped_blocks->lock);
3734         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3735                 swapped_blocks->blocks[i] = RB_ROOT;
3736         swapped_blocks->swapped = false;
3737 }
3738
3739 /*
3740  * Delete all swapped blocks record of @root.
3741  * Every record here means we skipped a full subtree scan for qgroup.
3742  *
3743  * Gets called when committing one transaction.
3744  */
3745 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3746 {