45868fd76209024dc2cfc790801ffac9133a3ef1
[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 fuer 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 umounting
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         spin_lock(&fs_info->qgroup_lock);
1017         fs_info->quota_root = quota_root;
1018         set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1019         spin_unlock(&fs_info->qgroup_lock);
1020
1021         ret = btrfs_commit_transaction(trans);
1022         trans = NULL;
1023         if (ret)
1024                 goto out_free_path;
1025
1026         ret = qgroup_rescan_init(fs_info, 0, 1);
1027         if (!ret) {
1028                 qgroup_rescan_zero_tracking(fs_info);
1029                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1030                                  &fs_info->qgroup_rescan_work);
1031         }
1032
1033 out_free_path:
1034         btrfs_free_path(path);
1035 out_free_root:
1036         if (ret) {
1037                 free_extent_buffer(quota_root->node);
1038                 free_extent_buffer(quota_root->commit_root);
1039                 kfree(quota_root);
1040         }
1041 out:
1042         if (ret) {
1043                 ulist_free(fs_info->qgroup_ulist);
1044                 fs_info->qgroup_ulist = NULL;
1045                 if (trans)
1046                         btrfs_end_transaction(trans);
1047         }
1048         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1049         return ret;
1050 }
1051
1052 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1053 {
1054         struct btrfs_root *quota_root;
1055         struct btrfs_trans_handle *trans = NULL;
1056         int ret = 0;
1057
1058         mutex_lock(&fs_info->qgroup_ioctl_lock);
1059         if (!fs_info->quota_root)
1060                 goto out;
1061
1062         /*
1063          * 1 For the root item
1064          *
1065          * We should also reserve enough items for the quota tree deletion in
1066          * btrfs_clean_quota_tree but this is not done.
1067          */
1068         trans = btrfs_start_transaction(fs_info->tree_root, 1);
1069         if (IS_ERR(trans)) {
1070                 ret = PTR_ERR(trans);
1071                 goto out;
1072         }
1073
1074         clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1075         btrfs_qgroup_wait_for_completion(fs_info, false);
1076         spin_lock(&fs_info->qgroup_lock);
1077         quota_root = fs_info->quota_root;
1078         fs_info->quota_root = NULL;
1079         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1080         spin_unlock(&fs_info->qgroup_lock);
1081
1082         btrfs_free_qgroup_config(fs_info);
1083
1084         ret = btrfs_clean_quota_tree(trans, quota_root);
1085         if (ret) {
1086                 btrfs_abort_transaction(trans, ret);
1087                 goto end_trans;
1088         }
1089
1090         ret = btrfs_del_root(trans, &quota_root->root_key);
1091         if (ret) {
1092                 btrfs_abort_transaction(trans, ret);
1093                 goto end_trans;
1094         }
1095
1096         list_del(&quota_root->dirty_list);
1097
1098         btrfs_tree_lock(quota_root->node);
1099         clean_tree_block(fs_info, quota_root->node);
1100         btrfs_tree_unlock(quota_root->node);
1101         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1102
1103         free_extent_buffer(quota_root->node);
1104         free_extent_buffer(quota_root->commit_root);
1105         kfree(quota_root);
1106
1107 end_trans:
1108         ret = btrfs_end_transaction(trans);
1109 out:
1110         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1111         return ret;
1112 }
1113
1114 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1115                          struct btrfs_qgroup *qgroup)
1116 {
1117         if (list_empty(&qgroup->dirty))
1118                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1119 }
1120
1121 /*
1122  * The easy accounting, we're updating qgroup relationship whose child qgroup
1123  * only has exclusive extents.
1124  *
1125  * In this case, all exclsuive extents will also be exlusive for parent, so
1126  * excl/rfer just get added/removed.
1127  *
1128  * So is qgroup reservation space, which should also be added/removed to
1129  * parent.
1130  * Or when child tries to release reservation space, parent will underflow its
1131  * reservation (for relationship adding case).
1132  *
1133  * Caller should hold fs_info->qgroup_lock.
1134  */
1135 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1136                                     struct ulist *tmp, u64 ref_root,
1137                                     struct btrfs_qgroup *src, int sign)
1138 {
1139         struct btrfs_qgroup *qgroup;
1140         struct btrfs_qgroup_list *glist;
1141         struct ulist_node *unode;
1142         struct ulist_iterator uiter;
1143         u64 num_bytes = src->excl;
1144         int ret = 0;
1145
1146         qgroup = find_qgroup_rb(fs_info, ref_root);
1147         if (!qgroup)
1148                 goto out;
1149
1150         qgroup->rfer += sign * num_bytes;
1151         qgroup->rfer_cmpr += sign * num_bytes;
1152
1153         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1154         qgroup->excl += sign * num_bytes;
1155         qgroup->excl_cmpr += sign * num_bytes;
1156
1157         if (sign > 0)
1158                 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1159         else
1160                 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1161
1162         qgroup_dirty(fs_info, qgroup);
1163
1164         /* Get all of the parent groups that contain this qgroup */
1165         list_for_each_entry(glist, &qgroup->groups, next_group) {
1166                 ret = ulist_add(tmp, glist->group->qgroupid,
1167                                 qgroup_to_aux(glist->group), GFP_ATOMIC);
1168                 if (ret < 0)
1169                         goto out;
1170         }
1171
1172         /* Iterate all of the parents and adjust their reference counts */
1173         ULIST_ITER_INIT(&uiter);
1174         while ((unode = ulist_next(tmp, &uiter))) {
1175                 qgroup = unode_aux_to_qgroup(unode);
1176                 qgroup->rfer += sign * num_bytes;
1177                 qgroup->rfer_cmpr += sign * num_bytes;
1178                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1179                 qgroup->excl += sign * num_bytes;
1180                 if (sign > 0)
1181                         qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1182                 else
1183                         qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1184                 qgroup->excl_cmpr += sign * num_bytes;
1185                 qgroup_dirty(fs_info, qgroup);
1186
1187                 /* Add any parents of the parents */
1188                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1189                         ret = ulist_add(tmp, glist->group->qgroupid,
1190                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
1191                         if (ret < 0)
1192                                 goto out;
1193                 }
1194         }
1195         ret = 0;
1196 out:
1197         return ret;
1198 }
1199
1200
1201 /*
1202  * Quick path for updating qgroup with only excl refs.
1203  *
1204  * In that case, just update all parent will be enough.
1205  * Or we needs to do a full rescan.
1206  * Caller should also hold fs_info->qgroup_lock.
1207  *
1208  * Return 0 for quick update, return >0 for need to full rescan
1209  * and mark INCONSISTENT flag.
1210  * Return < 0 for other error.
1211  */
1212 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1213                                    struct ulist *tmp, u64 src, u64 dst,
1214                                    int sign)
1215 {
1216         struct btrfs_qgroup *qgroup;
1217         int ret = 1;
1218         int err = 0;
1219
1220         qgroup = find_qgroup_rb(fs_info, src);
1221         if (!qgroup)
1222                 goto out;
1223         if (qgroup->excl == qgroup->rfer) {
1224                 ret = 0;
1225                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1226                                                qgroup, sign);
1227                 if (err < 0) {
1228                         ret = err;
1229                         goto out;
1230                 }
1231         }
1232 out:
1233         if (ret)
1234                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1235         return ret;
1236 }
1237
1238 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1239                               u64 dst)
1240 {
1241         struct btrfs_fs_info *fs_info = trans->fs_info;
1242         struct btrfs_root *quota_root;
1243         struct btrfs_qgroup *parent;
1244         struct btrfs_qgroup *member;
1245         struct btrfs_qgroup_list *list;
1246         struct ulist *tmp;
1247         int ret = 0;
1248
1249         /* Check the level of src and dst first */
1250         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1251                 return -EINVAL;
1252
1253         tmp = ulist_alloc(GFP_KERNEL);
1254         if (!tmp)
1255                 return -ENOMEM;
1256
1257         mutex_lock(&fs_info->qgroup_ioctl_lock);
1258         quota_root = fs_info->quota_root;
1259         if (!quota_root) {
1260                 ret = -EINVAL;
1261                 goto out;
1262         }
1263         member = find_qgroup_rb(fs_info, src);
1264         parent = find_qgroup_rb(fs_info, dst);
1265         if (!member || !parent) {
1266                 ret = -EINVAL;
1267                 goto out;
1268         }
1269
1270         /* check if such qgroup relation exist firstly */
1271         list_for_each_entry(list, &member->groups, next_group) {
1272                 if (list->group == parent) {
1273                         ret = -EEXIST;
1274                         goto out;
1275                 }
1276         }
1277
1278         ret = add_qgroup_relation_item(trans, src, dst);
1279         if (ret)
1280                 goto out;
1281
1282         ret = add_qgroup_relation_item(trans, dst, src);
1283         if (ret) {
1284                 del_qgroup_relation_item(trans, src, dst);
1285                 goto out;
1286         }
1287
1288         spin_lock(&fs_info->qgroup_lock);
1289         ret = add_relation_rb(fs_info, src, dst);
1290         if (ret < 0) {
1291                 spin_unlock(&fs_info->qgroup_lock);
1292                 goto out;
1293         }
1294         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1295         spin_unlock(&fs_info->qgroup_lock);
1296 out:
1297         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1298         ulist_free(tmp);
1299         return ret;
1300 }
1301
1302 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1303                                  u64 dst)
1304 {
1305         struct btrfs_fs_info *fs_info = trans->fs_info;
1306         struct btrfs_root *quota_root;
1307         struct btrfs_qgroup *parent;
1308         struct btrfs_qgroup *member;
1309         struct btrfs_qgroup_list *list;
1310         struct ulist *tmp;
1311         int ret = 0;
1312         int err;
1313
1314         tmp = ulist_alloc(GFP_KERNEL);
1315         if (!tmp)
1316                 return -ENOMEM;
1317
1318         quota_root = fs_info->quota_root;
1319         if (!quota_root) {
1320                 ret = -EINVAL;
1321                 goto out;
1322         }
1323
1324         member = find_qgroup_rb(fs_info, src);
1325         parent = find_qgroup_rb(fs_info, dst);
1326         if (!member || !parent) {
1327                 ret = -EINVAL;
1328                 goto out;
1329         }
1330
1331         /* check if such qgroup relation exist firstly */
1332         list_for_each_entry(list, &member->groups, next_group) {
1333                 if (list->group == parent)
1334                         goto exist;
1335         }
1336         ret = -ENOENT;
1337         goto out;
1338 exist:
1339         ret = del_qgroup_relation_item(trans, src, dst);
1340         err = del_qgroup_relation_item(trans, dst, src);
1341         if (err && !ret)
1342                 ret = err;
1343
1344         spin_lock(&fs_info->qgroup_lock);
1345         del_relation_rb(fs_info, src, dst);
1346         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1347         spin_unlock(&fs_info->qgroup_lock);
1348 out:
1349         ulist_free(tmp);
1350         return ret;
1351 }
1352
1353 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1354                               u64 dst)
1355 {
1356         struct btrfs_fs_info *fs_info = trans->fs_info;
1357         int ret = 0;
1358
1359         mutex_lock(&fs_info->qgroup_ioctl_lock);
1360         ret = __del_qgroup_relation(trans, src, dst);
1361         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1362
1363         return ret;
1364 }
1365
1366 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1367 {
1368         struct btrfs_fs_info *fs_info = trans->fs_info;
1369         struct btrfs_root *quota_root;
1370         struct btrfs_qgroup *qgroup;
1371         int ret = 0;
1372
1373         mutex_lock(&fs_info->qgroup_ioctl_lock);
1374         quota_root = fs_info->quota_root;
1375         if (!quota_root) {
1376                 ret = -EINVAL;
1377                 goto out;
1378         }
1379         qgroup = find_qgroup_rb(fs_info, qgroupid);
1380         if (qgroup) {
1381                 ret = -EEXIST;
1382                 goto out;
1383         }
1384
1385         ret = add_qgroup_item(trans, quota_root, qgroupid);
1386         if (ret)
1387                 goto out;
1388
1389         spin_lock(&fs_info->qgroup_lock);
1390         qgroup = add_qgroup_rb(fs_info, qgroupid);
1391         spin_unlock(&fs_info->qgroup_lock);
1392
1393         if (IS_ERR(qgroup))
1394                 ret = PTR_ERR(qgroup);
1395 out:
1396         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1397         return ret;
1398 }
1399
1400 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1401 {
1402         struct btrfs_fs_info *fs_info = trans->fs_info;
1403         struct btrfs_root *quota_root;
1404         struct btrfs_qgroup *qgroup;
1405         struct btrfs_qgroup_list *list;
1406         int ret = 0;
1407
1408         mutex_lock(&fs_info->qgroup_ioctl_lock);
1409         quota_root = fs_info->quota_root;
1410         if (!quota_root) {
1411                 ret = -EINVAL;
1412                 goto out;
1413         }
1414
1415         qgroup = find_qgroup_rb(fs_info, qgroupid);
1416         if (!qgroup) {
1417                 ret = -ENOENT;
1418                 goto out;
1419         }
1420
1421         /* Check if there are no children of this qgroup */
1422         if (!list_empty(&qgroup->members)) {
1423                 ret = -EBUSY;
1424                 goto out;
1425         }
1426
1427         ret = del_qgroup_item(trans, qgroupid);
1428         if (ret && ret != -ENOENT)
1429                 goto out;
1430
1431         while (!list_empty(&qgroup->groups)) {
1432                 list = list_first_entry(&qgroup->groups,
1433                                         struct btrfs_qgroup_list, next_group);
1434                 ret = __del_qgroup_relation(trans, qgroupid,
1435                                             list->group->qgroupid);
1436                 if (ret)
1437                         goto out;
1438         }
1439
1440         spin_lock(&fs_info->qgroup_lock);
1441         del_qgroup_rb(fs_info, qgroupid);
1442         spin_unlock(&fs_info->qgroup_lock);
1443 out:
1444         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1445         return ret;
1446 }
1447
1448 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1449                        struct btrfs_qgroup_limit *limit)
1450 {
1451         struct btrfs_fs_info *fs_info = trans->fs_info;
1452         struct btrfs_root *quota_root;
1453         struct btrfs_qgroup *qgroup;
1454         int ret = 0;
1455         /* Sometimes we would want to clear the limit on this qgroup.
1456          * To meet this requirement, we treat the -1 as a special value
1457          * which tell kernel to clear the limit on this qgroup.
1458          */
1459         const u64 CLEAR_VALUE = -1;
1460
1461         mutex_lock(&fs_info->qgroup_ioctl_lock);
1462         quota_root = fs_info->quota_root;
1463         if (!quota_root) {
1464                 ret = -EINVAL;
1465                 goto out;
1466         }
1467
1468         qgroup = find_qgroup_rb(fs_info, qgroupid);
1469         if (!qgroup) {
1470                 ret = -ENOENT;
1471                 goto out;
1472         }
1473
1474         spin_lock(&fs_info->qgroup_lock);
1475         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1476                 if (limit->max_rfer == CLEAR_VALUE) {
1477                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1478                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1479                         qgroup->max_rfer = 0;
1480                 } else {
1481                         qgroup->max_rfer = limit->max_rfer;
1482                 }
1483         }
1484         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1485                 if (limit->max_excl == CLEAR_VALUE) {
1486                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1487                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1488                         qgroup->max_excl = 0;
1489                 } else {
1490                         qgroup->max_excl = limit->max_excl;
1491                 }
1492         }
1493         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1494                 if (limit->rsv_rfer == CLEAR_VALUE) {
1495                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1496                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1497                         qgroup->rsv_rfer = 0;
1498                 } else {
1499                         qgroup->rsv_rfer = limit->rsv_rfer;
1500                 }
1501         }
1502         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1503                 if (limit->rsv_excl == CLEAR_VALUE) {
1504                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1505                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1506                         qgroup->rsv_excl = 0;
1507                 } else {
1508                         qgroup->rsv_excl = limit->rsv_excl;
1509                 }
1510         }
1511         qgroup->lim_flags |= limit->flags;
1512
1513         spin_unlock(&fs_info->qgroup_lock);
1514
1515         ret = update_qgroup_limit_item(trans, qgroup);
1516         if (ret) {
1517                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1518                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1519                        qgroupid);
1520         }
1521
1522 out:
1523         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1524         return ret;
1525 }
1526
1527 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1528                                 struct btrfs_delayed_ref_root *delayed_refs,
1529                                 struct btrfs_qgroup_extent_record *record)
1530 {
1531         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1532         struct rb_node *parent_node = NULL;
1533         struct btrfs_qgroup_extent_record *entry;
1534         u64 bytenr = record->bytenr;
1535
1536         lockdep_assert_held(&delayed_refs->lock);
1537         trace_btrfs_qgroup_trace_extent(fs_info, record);
1538
1539         while (*p) {
1540                 parent_node = *p;
1541                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1542                                  node);
1543                 if (bytenr < entry->bytenr)
1544                         p = &(*p)->rb_left;
1545                 else if (bytenr > entry->bytenr)
1546                         p = &(*p)->rb_right;
1547                 else
1548                         return 1;
1549         }
1550
1551         rb_link_node(&record->node, parent_node, p);
1552         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1553         return 0;
1554 }
1555
1556 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1557                                    struct btrfs_qgroup_extent_record *qrecord)
1558 {
1559         struct ulist *old_root;
1560         u64 bytenr = qrecord->bytenr;
1561         int ret;
1562
1563         ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1564         if (ret < 0) {
1565                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1566                 btrfs_warn(fs_info,
1567 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1568                         ret);
1569                 return 0;
1570         }
1571
1572         /*
1573          * Here we don't need to get the lock of
1574          * trans->transaction->delayed_refs, since inserted qrecord won't
1575          * be deleted, only qrecord->node may be modified (new qrecord insert)
1576          *
1577          * So modifying qrecord->old_roots is safe here
1578          */
1579         qrecord->old_roots = old_root;
1580         return 0;
1581 }
1582
1583 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1584                               u64 num_bytes, gfp_t gfp_flag)
1585 {
1586         struct btrfs_fs_info *fs_info = trans->fs_info;
1587         struct btrfs_qgroup_extent_record *record;
1588         struct btrfs_delayed_ref_root *delayed_refs;
1589         int ret;
1590
1591         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1592             || bytenr == 0 || num_bytes == 0)
1593                 return 0;
1594         record = kmalloc(sizeof(*record), gfp_flag);
1595         if (!record)
1596                 return -ENOMEM;
1597
1598         delayed_refs = &trans->transaction->delayed_refs;
1599         record->bytenr = bytenr;
1600         record->num_bytes = num_bytes;
1601         record->old_roots = NULL;
1602
1603         spin_lock(&delayed_refs->lock);
1604         ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1605         spin_unlock(&delayed_refs->lock);
1606         if (ret > 0) {
1607                 kfree(record);
1608                 return 0;
1609         }
1610         return btrfs_qgroup_trace_extent_post(fs_info, record);
1611 }
1612
1613 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1614                                   struct extent_buffer *eb)
1615 {
1616         struct btrfs_fs_info *fs_info = trans->fs_info;
1617         int nr = btrfs_header_nritems(eb);
1618         int i, extent_type, ret;
1619         struct btrfs_key key;
1620         struct btrfs_file_extent_item *fi;
1621         u64 bytenr, num_bytes;
1622
1623         /* We can be called directly from walk_up_proc() */
1624         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1625                 return 0;
1626
1627         for (i = 0; i < nr; i++) {
1628                 btrfs_item_key_to_cpu(eb, &key, i);
1629
1630                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1631                         continue;
1632
1633                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1634                 /* filter out non qgroup-accountable extents  */
1635                 extent_type = btrfs_file_extent_type(eb, fi);
1636
1637                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1638                         continue;
1639
1640                 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1641                 if (!bytenr)
1642                         continue;
1643
1644                 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1645
1646                 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1647                                                 GFP_NOFS);
1648                 if (ret)
1649                         return ret;
1650         }
1651         cond_resched();
1652         return 0;
1653 }
1654
1655 /*
1656  * Walk up the tree from the bottom, freeing leaves and any interior
1657  * nodes which have had all slots visited. If a node (leaf or
1658  * interior) is freed, the node above it will have it's slot
1659  * incremented. The root node will never be freed.
1660  *
1661  * At the end of this function, we should have a path which has all
1662  * slots incremented to the next position for a search. If we need to
1663  * read a new node it will be NULL and the node above it will have the
1664  * correct slot selected for a later read.
1665  *
1666  * If we increment the root nodes slot counter past the number of
1667  * elements, 1 is returned to signal completion of the search.
1668  */
1669 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1670 {
1671         int level = 0;
1672         int nr, slot;
1673         struct extent_buffer *eb;
1674
1675         if (root_level == 0)
1676                 return 1;
1677
1678         while (level <= root_level) {
1679                 eb = path->nodes[level];
1680                 nr = btrfs_header_nritems(eb);
1681                 path->slots[level]++;
1682                 slot = path->slots[level];
1683                 if (slot >= nr || level == 0) {
1684                         /*
1685                          * Don't free the root -  we will detect this
1686                          * condition after our loop and return a
1687                          * positive value for caller to stop walking the tree.
1688                          */
1689                         if (level != root_level) {
1690                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
1691                                 path->locks[level] = 0;
1692
1693                                 free_extent_buffer(eb);
1694                                 path->nodes[level] = NULL;
1695                                 path->slots[level] = 0;
1696                         }
1697                 } else {
1698                         /*
1699                          * We have a valid slot to walk back down
1700                          * from. Stop here so caller can process these
1701                          * new nodes.
1702                          */
1703                         break;
1704                 }
1705
1706                 level++;
1707         }
1708
1709         eb = path->nodes[root_level];
1710         if (path->slots[root_level] >= btrfs_header_nritems(eb))
1711                 return 1;
1712
1713         return 0;
1714 }
1715
1716 /*
1717  * Helper function to trace a subtree tree block swap.
1718  *
1719  * The swap will happen in highest tree block, but there may be a lot of
1720  * tree blocks involved.
1721  *
1722  * For example:
1723  *  OO = Old tree blocks
1724  *  NN = New tree blocks allocated during balance
1725  *
1726  *           File tree (257)                  Reloc tree for 257
1727  * L2              OO                                NN
1728  *               /    \                            /    \
1729  * L1          OO      OO (a)                    OO      NN (a)
1730  *            / \     / \                       / \     / \
1731  * L0       OO   OO OO   OO                   OO   OO NN   NN
1732  *                  (b)  (c)                          (b)  (c)
1733  *
1734  * When calling qgroup_trace_extent_swap(), we will pass:
1735  * @src_eb = OO(a)
1736  * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1737  * @dst_level = 0
1738  * @root_level = 1
1739  *
1740  * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1741  * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1742  *
1743  * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1744  *
1745  * 1) Tree search from @src_eb
1746  *    It should acts as a simplified btrfs_search_slot().
1747  *    The key for search can be extracted from @dst_path->nodes[dst_level]
1748  *    (first key).
1749  *
1750  * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1751  *    NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1752  *    They should be marked during preivous (@dst_level = 1) iteration.
1753  *
1754  * 3) Mark file extents in leaves dirty
1755  *    We don't have good way to pick out new file extents only.
1756  *    So we still follow the old method by scanning all file extents in
1757  *    the leave.
1758  *
1759  * This function can free us from keeping two pathes, thus later we only need
1760  * to care about how to iterate all new tree blocks in reloc tree.
1761  */
1762 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1763                                     struct extent_buffer *src_eb,
1764                                     struct btrfs_path *dst_path,
1765                                     int dst_level, int root_level,
1766                                     bool trace_leaf)
1767 {
1768         struct btrfs_key key;
1769         struct btrfs_path *src_path;
1770         struct btrfs_fs_info *fs_info = trans->fs_info;
1771         u32 nodesize = fs_info->nodesize;
1772         int cur_level = root_level;
1773         int ret;
1774
1775         BUG_ON(dst_level > root_level);
1776         /* Level mismatch */
1777         if (btrfs_header_level(src_eb) != root_level)
1778                 return -EINVAL;
1779
1780         src_path = btrfs_alloc_path();
1781         if (!src_path) {
1782                 ret = -ENOMEM;
1783                 goto out;
1784         }
1785
1786         if (dst_level)
1787                 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1788         else
1789                 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1790
1791         /* For src_path */
1792         extent_buffer_get(src_eb);
1793         src_path->nodes[root_level] = src_eb;
1794         src_path->slots[root_level] = dst_path->slots[root_level];
1795         src_path->locks[root_level] = 0;
1796
1797         /* A simplified version of btrfs_search_slot() */
1798         while (cur_level >= dst_level) {
1799                 struct btrfs_key src_key;
1800                 struct btrfs_key dst_key;
1801
1802                 if (src_path->nodes[cur_level] == NULL) {
1803                         struct btrfs_key first_key;
1804                         struct extent_buffer *eb;
1805                         int parent_slot;
1806                         u64 child_gen;
1807                         u64 child_bytenr;
1808
1809                         eb = src_path->nodes[cur_level + 1];
1810                         parent_slot = src_path->slots[cur_level + 1];
1811                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1812                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1813                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1814
1815                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
1816                                              cur_level, &first_key);
1817                         if (IS_ERR(eb)) {
1818                                 ret = PTR_ERR(eb);
1819                                 goto out;
1820                         } else if (!extent_buffer_uptodate(eb)) {
1821                                 free_extent_buffer(eb);
1822                                 ret = -EIO;
1823                                 goto out;
1824                         }
1825
1826                         src_path->nodes[cur_level] = eb;
1827
1828                         btrfs_tree_read_lock(eb);
1829                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1830                         src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1831                 }
1832
1833                 src_path->slots[cur_level] = dst_path->slots[cur_level];
1834                 if (cur_level) {
1835                         btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1836                                         &dst_key, dst_path->slots[cur_level]);
1837                         btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1838                                         &src_key, src_path->slots[cur_level]);
1839                 } else {
1840                         btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1841                                         &dst_key, dst_path->slots[cur_level]);
1842                         btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1843                                         &src_key, src_path->slots[cur_level]);
1844                 }
1845                 /* Content mismatch, something went wrong */
1846                 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1847                         ret = -ENOENT;
1848                         goto out;
1849                 }
1850                 cur_level--;
1851         }
1852
1853         /*
1854          * Now both @dst_path and @src_path have been populated, record the tree
1855          * blocks for qgroup accounting.
1856          */
1857         ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1858                         nodesize, GFP_NOFS);
1859         if (ret < 0)
1860                 goto out;
1861         ret = btrfs_qgroup_trace_extent(trans,
1862                         dst_path->nodes[dst_level]->start,
1863                         nodesize, GFP_NOFS);
1864         if (ret < 0)
1865                 goto out;
1866
1867         /* Record leaf file extents */
1868         if (dst_level == 0 && trace_leaf) {
1869                 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1870                 if (ret < 0)
1871                         goto out;
1872                 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1873         }
1874 out:
1875         btrfs_free_path(src_path);
1876         return ret;
1877 }
1878
1879 /*
1880  * Helper function to do recursive generation-aware depth-first search, to
1881  * locate all new tree blocks in a subtree of reloc tree.
1882  *
1883  * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1884  *         reloc tree
1885  * L2         NN (a)
1886  *          /    \
1887  * L1    OO        NN (b)
1888  *      /  \      /  \
1889  * L0  OO  OO    OO  NN
1890  *               (c) (d)
1891  * If we pass:
1892  * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1893  * @cur_level = 1
1894  * @root_level = 1
1895  *
1896  * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1897  * above tree blocks along with their counter parts in file tree.
1898  * While during search, old tree blocsk OO(c) will be skiped as tree block swap
1899  * won't affect OO(c).
1900  */
1901 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1902                                            struct extent_buffer *src_eb,
1903                                            struct btrfs_path *dst_path,
1904                                            int cur_level, int root_level,
1905                                            u64 last_snapshot, bool trace_leaf)
1906 {
1907         struct btrfs_fs_info *fs_info = trans->fs_info;
1908         struct extent_buffer *eb;
1909         bool need_cleanup = false;
1910         int ret = 0;
1911         int i;
1912
1913         /* Level sanity check */
1914         if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL ||
1915             root_level < 0 || root_level >= BTRFS_MAX_LEVEL ||
1916             root_level < cur_level) {
1917                 btrfs_err_rl(fs_info,
1918                         "%s: bad levels, cur_level=%d root_level=%d",
1919                         __func__, cur_level, root_level);
1920                 return -EUCLEAN;
1921         }
1922
1923         /* Read the tree block if needed */
1924         if (dst_path->nodes[cur_level] == NULL) {
1925                 struct btrfs_key first_key;
1926                 int parent_slot;
1927                 u64 child_gen;
1928                 u64 child_bytenr;
1929
1930                 /*
1931                  * dst_path->nodes[root_level] must be initialized before
1932                  * calling this function.
1933                  */
1934                 if (cur_level == root_level) {
1935                         btrfs_err_rl(fs_info,
1936         "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1937                                 __func__, root_level, root_level, cur_level);
1938                         return -EUCLEAN;
1939                 }
1940
1941                 /*
1942                  * We need to get child blockptr/gen from parent before we can
1943                  * read it.
1944                   */
1945                 eb = dst_path->nodes[cur_level + 1];
1946                 parent_slot = dst_path->slots[cur_level + 1];
1947                 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1948                 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1949                 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1950
1951                 /* This node is old, no need to trace */
1952                 if (child_gen < last_snapshot)
1953                         goto out;
1954
1955                 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1956                                      cur_level, &first_key);
1957                 if (IS_ERR(eb)) {
1958                         ret = PTR_ERR(eb);
1959                         goto out;
1960                 } else if (!extent_buffer_uptodate(eb)) {
1961                         free_extent_buffer(eb);
1962                         ret = -EIO;
1963                         goto out;
1964                 }
1965
1966                 dst_path->nodes[cur_level] = eb;
1967                 dst_path->slots[cur_level] = 0;
1968
1969                 btrfs_tree_read_lock(eb);
1970                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1971                 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1972                 need_cleanup = true;
1973         }
1974
1975         /* Now record this tree block and its counter part for qgroups */
1976         ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1977                                        root_level, trace_leaf);
1978         if (ret < 0)
1979                 goto cleanup;
1980
1981         eb = dst_path->nodes[cur_level];
1982
1983         if (cur_level > 0) {
1984                 /* Iterate all child tree blocks */
1985                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1986                         /* Skip old tree blocks as they won't be swapped */
1987                         if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1988                                 continue;
1989                         dst_path->slots[cur_level] = i;
1990
1991                         /* Recursive call (at most 7 times) */
1992                         ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
1993                                         dst_path, cur_level - 1, root_level,
1994                                         last_snapshot, trace_leaf);
1995                         if (ret < 0)
1996                                 goto cleanup;
1997                 }
1998         }
1999
2000 cleanup:
2001         if (need_cleanup) {
2002                 /* Clean up */
2003                 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2004                                      dst_path->locks[cur_level]);
2005                 free_extent_buffer(dst_path->nodes[cur_level]);
2006                 dst_path->nodes[cur_level] = NULL;
2007                 dst_path->slots[cur_level] = 0;
2008                 dst_path->locks[cur_level] = 0;
2009         }
2010 out:
2011         return ret;
2012 }
2013
2014 /*
2015  * Inform qgroup to trace subtree swap used in balance.
2016  *
2017  * Unlike btrfs_qgroup_trace_subtree(), this function will only trace
2018  * new tree blocks whose generation is equal to (or larger than) @last_snapshot.
2019  *
2020  * Will go down the tree block pointed by @dst_eb (pointed by @dst_parent and
2021  * @dst_slot), and find any tree blocks whose generation is at @last_snapshot,
2022  * and then go down @src_eb (pointed by @src_parent and @src_slot) to find
2023  * the conterpart of the tree block, then mark both tree blocks as qgroup dirty,
2024  * and skip all tree blocks whose generation is smaller than last_snapshot.
2025  *
2026  * This would skip tons of tree blocks of original btrfs_qgroup_trace_subtree(),
2027  * which could be the cause of very slow balance if the file tree is large.
2028  *
2029  * @src_parent, @src_slot: pointer to src (file tree) eb.
2030  * @dst_parent, @dst_slot: pointer to dst (reloc tree) eb.
2031  */
2032 int btrfs_qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2033                                 struct btrfs_block_group_cache *bg_cache,
2034                                 struct extent_buffer *src_parent, int src_slot,
2035                                 struct extent_buffer *dst_parent, int dst_slot,
2036                                 u64 last_snapshot)
2037 {
2038         struct btrfs_fs_info *fs_info = trans->fs_info;
2039         struct btrfs_path *dst_path = NULL;
2040         struct btrfs_key first_key;
2041         struct extent_buffer *src_eb = NULL;
2042         struct extent_buffer *dst_eb = NULL;
2043         bool trace_leaf = false;
2044         u64 child_gen;
2045         u64 child_bytenr;
2046         int level;
2047         int ret;
2048
2049         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2050                 return 0;
2051
2052         /* Check parameter order */
2053         if (btrfs_node_ptr_generation(src_parent, src_slot) >
2054             btrfs_node_ptr_generation(dst_parent, dst_slot)) {
2055                 btrfs_err_rl(fs_info,
2056                 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2057                         btrfs_node_ptr_generation(src_parent, src_slot),
2058                         btrfs_node_ptr_generation(dst_parent, dst_slot));
2059                 return -EUCLEAN;
2060         }
2061
2062         /*
2063          * Only trace leaf if we're relocating data block groups, this could
2064          * reduce tons of data extents tracing for meta/sys bg relocation.
2065          */
2066         if (bg_cache->flags & BTRFS_BLOCK_GROUP_DATA)
2067                 trace_leaf = true;
2068         /* Read out real @src_eb, pointed by @src_parent and @src_slot */
2069         child_bytenr = btrfs_node_blockptr(src_parent, src_slot);
2070         child_gen = btrfs_node_ptr_generation(src_parent, src_slot);
2071         btrfs_node_key_to_cpu(src_parent, &first_key, src_slot);
2072
2073         src_eb = read_tree_block(fs_info, child_bytenr, child_gen,
2074                         btrfs_header_level(src_parent) - 1, &first_key);
2075         if (IS_ERR(src_eb)) {
2076                 ret = PTR_ERR(src_eb);
2077                 goto out;
2078         }
2079
2080         /* Read out real @dst_eb, pointed by @src_parent and @src_slot */
2081         child_bytenr = btrfs_node_blockptr(dst_parent, dst_slot);
2082         child_gen = btrfs_node_ptr_generation(dst_parent, dst_slot);
2083         btrfs_node_key_to_cpu(dst_parent, &first_key, dst_slot);
2084
2085         dst_eb = read_tree_block(fs_info, child_bytenr, child_gen,
2086                         btrfs_header_level(dst_parent) - 1, &first_key);
2087         if (IS_ERR(dst_eb)) {
2088                 ret = PTR_ERR(dst_eb);
2089                 goto out;
2090         }
2091
2092         if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2093                 ret = -EINVAL;
2094                 goto out;
2095         }
2096
2097         level = btrfs_header_level(dst_eb);
2098         dst_path = btrfs_alloc_path();
2099         if (!dst_path) {
2100                 ret = -ENOMEM;
2101                 goto out;
2102         }
2103
2104         /* For dst_path */
2105         extent_buffer_get(dst_eb);
2106         dst_path->nodes[level] = dst_eb;
2107         dst_path->slots[level] = 0;
2108         dst_path->locks[level] = 0;
2109
2110         /* Do the generation-aware breadth-first search */
2111         ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2112                                               level, last_snapshot, trace_leaf);
2113         if (ret < 0)
2114                 goto out;
2115         ret = 0;
2116
2117 out:
2118         free_extent_buffer(src_eb);
2119         free_extent_buffer(dst_eb);
2120         btrfs_free_path(dst_path);
2121         if (ret < 0)
2122                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2123         return ret;
2124 }
2125
2126 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2127                                struct extent_buffer *root_eb,
2128                                u64 root_gen, int root_level)
2129 {
2130         struct btrfs_fs_info *fs_info = trans->fs_info;
2131         int ret = 0;
2132         int level;
2133         struct extent_buffer *eb = root_eb;
2134         struct btrfs_path *path = NULL;
2135
2136         BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2137         BUG_ON(root_eb == NULL);
2138
2139         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2140                 return 0;
2141
2142         if (!extent_buffer_uptodate(root_eb)) {
2143                 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2144                 if (ret)
2145                         goto out;
2146         }
2147
2148         if (root_level == 0) {
2149                 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2150                 goto out;
2151         }
2152
2153         path = btrfs_alloc_path();
2154         if (!path)
2155                 return -ENOMEM;
2156
2157         /*
2158          * Walk down the tree.  Missing extent blocks are filled in as
2159          * we go. Metadata is accounted every time we read a new
2160          * extent block.
2161          *
2162          * When we reach a leaf, we account for file extent items in it,
2163          * walk back up the tree (adjusting slot pointers as we go)
2164          * and restart the search process.
2165          */
2166         extent_buffer_get(root_eb); /* For path */
2167         path->nodes[root_level] = root_eb;
2168         path->slots[root_level] = 0;
2169         path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2170 walk_down:
2171         level = root_level;
2172         while (level >= 0) {
2173                 if (path->nodes[level] == NULL) {
2174                         struct btrfs_key first_key;
2175                         int parent_slot;
2176                         u64 child_gen;
2177                         u64 child_bytenr;
2178
2179                         /*
2180                          * We need to get child blockptr/gen from parent before
2181                          * we can read it.
2182                           */
2183                         eb = path->nodes[level + 1];
2184                         parent_slot = path->slots[level + 1];
2185                         child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2186                         child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2187                         btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2188
2189                         eb = read_tree_block(fs_info, child_bytenr, child_gen,
2190                                              level, &first_key);
2191                         if (IS_ERR(eb)) {
2192                                 ret = PTR_ERR(eb);
2193                                 goto out;
2194                         } else if (!extent_buffer_uptodate(eb)) {
2195                                 free_extent_buffer(eb);
2196                                 ret = -EIO;
2197                                 goto out;
2198                         }
2199
2200                         path->nodes[level] = eb;
2201                         path->slots[level] = 0;
2202
2203                         btrfs_tree_read_lock(eb);
2204                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
2205                         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2206
2207                         ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2208                                                         fs_info->nodesize,
2209                                                         GFP_NOFS);
2210                         if (ret)
2211                                 goto out;
2212                 }
2213
2214                 if (level == 0) {
2215                         ret = btrfs_qgroup_trace_leaf_items(trans,
2216                                                             path->nodes[level]);
2217                         if (ret)
2218                                 goto out;
2219
2220                         /* Nonzero return here means we completed our search */
2221                         ret = adjust_slots_upwards(path, root_level);
2222                         if (ret)
2223                                 break;
2224
2225                         /* Restart search with new slots */
2226                         goto walk_down;
2227                 }
2228
2229                 level--;
2230         }
2231
2232         ret = 0;
2233 out:
2234         btrfs_free_path(path);
2235
2236         return ret;
2237 }
2238
2239 #define UPDATE_NEW      0
2240 #define UPDATE_OLD      1
2241 /*
2242  * Walk all of the roots that points to the bytenr and adjust their refcnts.
2243  */
2244 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2245                                 struct ulist *roots, struct ulist *tmp,
2246                                 struct ulist *qgroups, u64 seq, int update_old)
2247 {
2248         struct ulist_node *unode;
2249         struct ulist_iterator uiter;
2250         struct ulist_node *tmp_unode;
2251         struct ulist_iterator tmp_uiter;
2252         struct btrfs_qgroup *qg;
2253         int ret = 0;
2254
2255         if (!roots)
2256                 return 0;
2257         ULIST_ITER_INIT(&uiter);
2258         while ((unode = ulist_next(roots, &uiter))) {
2259                 qg = find_qgroup_rb(fs_info, unode->val);
2260                 if (!qg)
2261                         continue;
2262
2263                 ulist_reinit(tmp);
2264                 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2265                                 GFP_ATOMIC);
2266                 if (ret < 0)
2267                         return ret;
2268                 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2269                 if (ret < 0)
2270                         return ret;
2271                 ULIST_ITER_INIT(&tmp_uiter);
2272                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2273                         struct btrfs_qgroup_list *glist;
2274
2275                         qg = unode_aux_to_qgroup(tmp_unode);
2276                         if (update_old)
2277                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2278                         else
2279                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2280                         list_for_each_entry(glist, &qg->groups, next_group) {
2281                                 ret = ulist_add(qgroups, glist->group->qgroupid,
2282                                                 qgroup_to_aux(glist->group),
2283                                                 GFP_ATOMIC);
2284                                 if (ret < 0)
2285                                         return ret;
2286                                 ret = ulist_add(tmp, glist->group->qgroupid,
2287                                                 qgroup_to_aux(glist->group),
2288                                                 GFP_ATOMIC);
2289                                 if (ret < 0)
2290                                         return ret;
2291                         }
2292                 }
2293         }
2294         return 0;
2295 }
2296
2297 /*
2298  * Update qgroup rfer/excl counters.
2299  * Rfer update is easy, codes can explain themselves.
2300  *
2301  * Excl update is tricky, the update is split into 2 part.
2302  * Part 1: Possible exclusive <-> sharing detect:
2303  *      |       A       |       !A      |
2304  *  -------------------------------------
2305  *  B   |       *       |       -       |
2306  *  -------------------------------------
2307  *  !B  |       +       |       **      |
2308  *  -------------------------------------
2309  *
2310  * Conditions:
2311  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
2312  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
2313  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
2314  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
2315  *
2316  * Results:
2317  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
2318  * *: Definitely not changed.           **: Possible unchanged.
2319  *
2320  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2321  *
2322  * To make the logic clear, we first use condition A and B to split
2323  * combination into 4 results.
2324  *
2325  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2326  * only on variant maybe 0.
2327  *
2328  * Lastly, check result **, since there are 2 variants maybe 0, split them
2329  * again(2x2).
2330  * But this time we don't need to consider other things, the codes and logic
2331  * is easy to understand now.
2332  */
2333 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2334                                   struct ulist *qgroups,
2335                                   u64 nr_old_roots,
2336                                   u64 nr_new_roots,
2337                                   u64 num_bytes, u64 seq)
2338 {
2339         struct ulist_node *unode;
2340         struct ulist_iterator uiter;
2341         struct btrfs_qgroup *qg;
2342         u64 cur_new_count, cur_old_count;
2343
2344         ULIST_ITER_INIT(&uiter);
2345         while ((unode = ulist_next(qgroups, &uiter))) {
2346                 bool dirty = false;
2347
2348                 qg = unode_aux_to_qgroup(unode);
2349                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2350                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2351
2352                 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2353                                              cur_new_count);
2354
2355                 /* Rfer update part */
2356                 if (cur_old_count == 0 && cur_new_count > 0) {
2357                         qg->rfer += num_bytes;
2358                         qg->rfer_cmpr += num_bytes;
2359                         dirty = true;
2360                 }
2361                 if (cur_old_count > 0 && cur_new_count == 0) {
2362                         qg->rfer -= num_bytes;
2363                         qg->rfer_cmpr -= num_bytes;
2364                         dirty = true;
2365                 }
2366
2367                 /* Excl update part */
2368                 /* Exclusive/none -> shared case */
2369                 if (cur_old_count == nr_old_roots &&
2370                     cur_new_count < nr_new_roots) {
2371                         /* Exclusive -> shared */
2372                         if (cur_old_count != 0) {
2373                                 qg->excl -= num_bytes;
2374                                 qg->excl_cmpr -= num_bytes;
2375                                 dirty = true;
2376                         }
2377                 }
2378
2379                 /* Shared -> exclusive/none case */
2380                 if (cur_old_count < nr_old_roots &&
2381                     cur_new_count == nr_new_roots) {
2382                         /* Shared->exclusive */
2383                         if (cur_new_count != 0) {
2384                                 qg->excl += num_bytes;
2385                                 qg->excl_cmpr += num_bytes;
2386                                 dirty = true;
2387                         }
2388                 }
2389
2390                 /* Exclusive/none -> exclusive/none case */
2391                 if (cur_old_count == nr_old_roots &&
2392                     cur_new_count == nr_new_roots) {
2393                         if (cur_old_count == 0) {
2394                                 /* None -> exclusive/none */
2395
2396                                 if (cur_new_count != 0) {
2397                                         /* None -> exclusive */
2398                                         qg->excl += num_bytes;
2399                                         qg->excl_cmpr += num_bytes;
2400                                         dirty = true;
2401                                 }
2402                                 /* None -> none, nothing changed */
2403                         } else {
2404                                 /* Exclusive -> exclusive/none */
2405
2406                                 if (cur_new_count == 0) {
2407                                         /* Exclusive -> none */
2408                                         qg->excl -= num_bytes;
2409                                         qg->excl_cmpr -= num_bytes;
2410                                         dirty = true;
2411                                 }
2412                                 /* Exclusive -> exclusive, nothing changed */
2413                         }
2414                 }
2415
2416                 if (dirty)
2417                         qgroup_dirty(fs_info, qg);
2418         }
2419         return 0;
2420 }
2421
2422 /*
2423  * Check if the @roots potentially is a list of fs tree roots
2424  *
2425  * Return 0 for definitely not a fs/subvol tree roots ulist
2426  * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2427  *          one as well)
2428  */
2429 static int maybe_fs_roots(struct ulist *roots)
2430 {
2431         struct ulist_node *unode;
2432         struct ulist_iterator uiter;
2433
2434         /* Empty one, still possible for fs roots */
2435         if (!roots || roots->nnodes == 0)
2436                 return 1;
2437
2438         ULIST_ITER_INIT(&uiter);
2439         unode = ulist_next(roots, &uiter);
2440         if (!unode)
2441                 return 1;
2442
2443         /*
2444          * If it contains fs tree roots, then it must belong to fs/subvol
2445          * trees.
2446          * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2447          */
2448         return is_fstree(unode->val);
2449 }
2450
2451 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2452                                 u64 num_bytes, struct ulist *old_roots,
2453                                 struct ulist *new_roots)
2454 {
2455         struct btrfs_fs_info *fs_info = trans->fs_info;
2456         struct ulist *qgroups = NULL;
2457         struct ulist *tmp = NULL;
2458         u64 seq;
2459         u64 nr_new_roots = 0;
2460         u64 nr_old_roots = 0;
2461         int ret = 0;
2462
2463         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2464                 return 0;
2465
2466         if (new_roots) {
2467                 if (!maybe_fs_roots(new_roots))
2468                         goto out_free;
2469                 nr_new_roots = new_roots->nnodes;
2470         }
2471         if (old_roots) {
2472                 if (!maybe_fs_roots(old_roots))
2473                         goto out_free;
2474                 nr_old_roots = old_roots->nnodes;
2475         }
2476
2477         /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2478         if (nr_old_roots == 0 && nr_new_roots == 0)
2479                 goto out_free;
2480
2481         BUG_ON(!fs_info->quota_root);
2482
2483         trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2484                                         num_bytes, nr_old_roots, nr_new_roots);
2485
2486         qgroups = ulist_alloc(GFP_NOFS);
2487         if (!qgroups) {
2488                 ret = -ENOMEM;
2489                 goto out_free;
2490         }
2491         tmp = ulist_alloc(GFP_NOFS);
2492         if (!tmp) {
2493                 ret = -ENOMEM;
2494                 goto out_free;
2495         }
2496
2497         mutex_lock(&fs_info->qgroup_rescan_lock);
2498         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2499                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2500                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2501                         ret = 0;
2502                         goto out_free;
2503                 }
2504         }
2505         mutex_unlock(&fs_info->qgroup_rescan_lock);
2506
2507         spin_lock(&fs_info->qgroup_lock);
2508         seq = fs_info->qgroup_seq;
2509
2510         /* Update old refcnts using old_roots */
2511         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2512                                    UPDATE_OLD);
2513         if (ret < 0)
2514                 goto out;
2515
2516         /* Update new refcnts using new_roots */
2517         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2518                                    UPDATE_NEW);
2519         if (ret < 0)
2520                 goto out;
2521
2522         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2523                                num_bytes, seq);
2524
2525         /*
2526          * Bump qgroup_seq to avoid seq overlap
2527          */
2528         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2529 out:
2530         spin_unlock(&fs_info->qgroup_lock);
2531 out_free:
2532         ulist_free(tmp);
2533         ulist_free(qgroups);
2534         ulist_free(old_roots);
2535         ulist_free(new_roots);
2536         return ret;
2537 }
2538
2539 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2540 {
2541         struct btrfs_fs_info *fs_info = trans->fs_info;
2542         struct btrfs_qgroup_extent_record *record;
2543         struct btrfs_delayed_ref_root *delayed_refs;
2544         struct ulist *new_roots = NULL;
2545         struct rb_node *node;
2546         u64 num_dirty_extents = 0;
2547         u64 qgroup_to_skip;
2548         int ret = 0;
2549
2550         delayed_refs = &trans->transaction->delayed_refs;
2551         qgroup_to_skip = delayed_refs->qgroup_to_skip;
2552         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2553                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2554                                   node);
2555
2556                 num_dirty_extents++;
2557                 trace_btrfs_qgroup_account_extents(fs_info, record);
2558
2559                 if (!ret) {
2560                         /*
2561                          * Old roots should be searched when inserting qgroup
2562                          * extent record
2563                          */
2564                         if (WARN_ON(!record->old_roots)) {
2565                                 /* Search commit root to find old_roots */
2566                                 ret = btrfs_find_all_roots(NULL, fs_info,
2567                                                 record->bytenr, 0,
2568                                                 &record->old_roots, false);
2569                                 if (ret < 0)
2570                                         goto cleanup;
2571                         }
2572
2573                         /*
2574                          * Use SEQ_LAST as time_seq to do special search, which
2575                          * doesn't lock tree or delayed_refs and search current
2576                          * root. It's safe inside commit_transaction().
2577                          */
2578                         ret = btrfs_find_all_roots(trans, fs_info,
2579                                 record->bytenr, SEQ_LAST, &new_roots, false);
2580                         if (ret < 0)
2581                                 goto cleanup;
2582                         if (qgroup_to_skip) {
2583                                 ulist_del(new_roots, qgroup_to_skip, 0);
2584                                 ulist_del(record->old_roots, qgroup_to_skip,
2585                                           0);
2586                         }
2587                         ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2588                                                           record->num_bytes,
2589                                                           record->old_roots,
2590                                                           new_roots);
2591                         record->old_roots = NULL;
2592                         new_roots = NULL;
2593                 }
2594 cleanup:
2595                 ulist_free(record->old_roots);
2596                 ulist_free(new_roots);
2597                 new_roots = NULL;
2598                 rb_erase(node, &delayed_refs->dirty_extent_root);
2599                 kfree(record);
2600
2601         }
2602         trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2603                                        num_dirty_extents);
2604         return ret;
2605 }
2606
2607 /*
2608  * called from commit_transaction. Writes all changed qgroups to disk.
2609  */
2610 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2611 {
2612         struct btrfs_fs_info *fs_info = trans->fs_info;
2613         struct btrfs_root *quota_root = fs_info->quota_root;
2614         int ret = 0;
2615
2616         if (!quota_root)
2617                 return ret;
2618
2619         spin_lock(&fs_info->qgroup_lock);
2620         while (!list_empty(&fs_info->dirty_qgroups)) {
2621                 struct btrfs_qgroup *qgroup;
2622                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2623                                           struct btrfs_qgroup, dirty);
2624                 list_del_init(&qgroup->dirty);
2625                 spin_unlock(&fs_info->qgroup_lock);
2626                 ret = update_qgroup_info_item(trans, qgroup);
2627                 if (ret)
2628                         fs_info->qgroup_flags |=
2629                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2630                 ret = update_qgroup_limit_item(trans, qgroup);
2631                 if (ret)
2632                         fs_info->qgroup_flags |=
2633                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2634                 spin_lock(&fs_info->qgroup_lock);
2635         }
2636         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2637                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2638         else
2639                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2640         spin_unlock(&fs_info->qgroup_lock);
2641
2642         ret = update_qgroup_status_item(trans);
2643         if (ret)
2644                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2645
2646         return ret;
2647 }
2648
2649 /*
2650  * Copy the accounting information between qgroups. This is necessary
2651  * when a snapshot or a subvolume is created. Throwing an error will
2652  * cause a transaction abort so we take extra care here to only error
2653  * when a readonly fs is a reasonable outcome.
2654  */
2655 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2656                          u64 objectid, struct btrfs_qgroup_inherit *inherit)
2657 {
2658         int ret = 0;
2659         int i;
2660         u64 *i_qgroups;
2661         struct btrfs_fs_info *fs_info = trans->fs_info;
2662         struct btrfs_root *quota_root = fs_info->quota_root;
2663         struct btrfs_qgroup *srcgroup;
2664         struct btrfs_qgroup *dstgroup;
2665         u32 level_size = 0;
2666         u64 nums;
2667
2668         mutex_lock(&fs_info->qgroup_ioctl_lock);
2669         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2670                 goto out;
2671
2672         if (!quota_root) {
2673                 ret = -EINVAL;
2674                 goto out;
2675         }
2676
2677         if (inherit) {
2678                 i_qgroups = (u64 *)(inherit + 1);
2679                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2680                        2 * inherit->num_excl_copies;
2681                 for (i = 0; i < nums; ++i) {
2682                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2683
2684                         /*
2685                          * Zero out invalid groups so we can ignore
2686                          * them later.
2687                          */
2688                         if (!srcgroup ||
2689                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2690                                 *i_qgroups = 0ULL;
2691
2692                         ++i_qgroups;
2693                 }
2694         }
2695
2696         /*
2697          * create a tracking group for the subvol itself
2698          */
2699         ret = add_qgroup_item(trans, quota_root, objectid);
2700         if (ret)
2701                 goto out;
2702
2703         /*
2704          * add qgroup to all inherited groups
2705          */
2706         if (inherit) {
2707                 i_qgroups = (u64 *)(inherit + 1);
2708                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2709                         if (*i_qgroups == 0)
2710                                 continue;
2711                         ret = add_qgroup_relation_item(trans, objectid,
2712                                                        *i_qgroups);
2713                         if (ret && ret != -EEXIST)
2714                                 goto out;
2715                         ret = add_qgroup_relation_item(trans, *i_qgroups,
2716                                                        objectid);
2717                         if (ret && ret != -EEXIST)
2718                                 goto out;
2719                 }
2720                 ret = 0;
2721         }
2722
2723
2724         spin_lock(&fs_info->qgroup_lock);
2725
2726         dstgroup = add_qgroup_rb(fs_info, objectid);
2727         if (IS_ERR(dstgroup)) {
2728                 ret = PTR_ERR(dstgroup);
2729                 goto unlock;
2730         }
2731
2732         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2733                 dstgroup->lim_flags = inherit->lim.flags;
2734                 dstgroup->max_rfer = inherit->lim.max_rfer;
2735                 dstgroup->max_excl = inherit->lim.max_excl;
2736                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2737                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2738
2739                 ret = update_qgroup_limit_item(trans, dstgroup);
2740                 if (ret) {
2741                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2742                         btrfs_info(fs_info,
2743                                    "unable to update quota limit for %llu",
2744                                    dstgroup->qgroupid);
2745                         goto unlock;
2746                 }
2747         }
2748
2749         if (srcid) {
2750                 srcgroup = find_qgroup_rb(fs_info, srcid);
2751                 if (!srcgroup)
2752                         goto unlock;
2753
2754                 /*
2755                  * We call inherit after we clone the root in order to make sure
2756                  * our counts don't go crazy, so at this point the only
2757                  * difference between the two roots should be the root node.
2758                  */
2759                 level_size = fs_info->nodesize;
2760                 dstgroup->rfer = srcgroup->rfer;
2761                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2762                 dstgroup->excl = level_size;
2763                 dstgroup->excl_cmpr = level_size;
2764                 srcgroup->excl = level_size;
2765                 srcgroup->excl_cmpr = level_size;
2766
2767                 /* inherit the limit info */
2768                 dstgroup->lim_flags = srcgroup->lim_flags;
2769                 dstgroup->max_rfer = srcgroup->max_rfer;
2770                 dstgroup->max_excl = srcgroup->max_excl;
2771                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2772                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2773
2774                 qgroup_dirty(fs_info, dstgroup);
2775                 qgroup_dirty(fs_info, srcgroup);
2776         }
2777
2778         if (!inherit)
2779                 goto unlock;
2780
2781         i_qgroups = (u64 *)(inherit + 1);
2782         for (i = 0; i < inherit->num_qgroups; ++i) {
2783                 if (*i_qgroups) {
2784                         ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2785                         if (ret)
2786                                 goto unlock;
2787                 }
2788                 ++i_qgroups;
2789         }
2790
2791         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2792                 struct btrfs_qgroup *src;
2793                 struct btrfs_qgroup *dst;
2794
2795                 if (!i_qgroups[0] || !i_qgroups[1])
2796                         continue;
2797
2798                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2799                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2800
2801                 if (!src || !dst) {
2802                         ret = -EINVAL;
2803                         goto unlock;
2804                 }
2805
2806                 dst->rfer = src->rfer - level_size;
2807                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2808         }
2809         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2810                 struct btrfs_qgroup *src;
2811                 struct btrfs_qgroup *dst;
2812
2813                 if (!i_qgroups[0] || !i_qgroups[1])
2814                         continue;
2815
2816                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2817                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2818
2819                 if (!src || !dst) {
2820                         ret = -EINVAL;
2821                         goto unlock;
2822                 }
2823
2824                 dst->excl = src->excl + level_size;
2825                 dst->excl_cmpr = src->excl_cmpr + level_size;
2826         }
2827
2828 unlock:
2829         spin_unlock(&fs_info->qgroup_lock);
2830 out:
2831         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2832         return ret;
2833 }
2834
2835 /*
2836  * Two limits to commit transaction in advance.
2837  *
2838  * For RATIO, it will be 1/RATIO of the remaining limit
2839  * (excluding data and prealloc meta) as threshold.
2840  * For SIZE, it will be in byte unit as threshold.
2841  */
2842 #define QGROUP_PERTRANS_RATIO           32
2843 #define QGROUP_PERTRANS_SIZE            SZ_32M
2844 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2845                                 const struct btrfs_qgroup *qg, u64 num_bytes)
2846 {
2847         u64 limit;
2848         u64 threshold;
2849
2850         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2851             qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2852                 return false;
2853
2854         if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2855             qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2856                 return false;
2857
2858         /*
2859          * Even if we passed the check, it's better to check if reservation
2860          * for meta_pertrans is pushing us near limit.
2861          * If there is too much pertrans reservation or it's near the limit,
2862          * let's try commit transaction to free some, using transaction_kthread
2863          */
2864         if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2865                               BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2866                 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2867                         limit = qg->max_excl;
2868                 else
2869                         limit = qg->max_rfer;
2870                 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2871                             qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2872                             QGROUP_PERTRANS_RATIO;
2873                 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2874
2875                 /*
2876                  * Use transaction_kthread to commit transaction, so we no
2877                  * longer need to bother nested transaction nor lock context.
2878                  */
2879                 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2880                         btrfs_commit_transaction_locksafe(fs_info);
2881         }
2882
2883         return true;
2884 }
2885
2886 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2887                           enum btrfs_qgroup_rsv_type type)
2888 {
2889         struct btrfs_root *quota_root;
2890         struct btrfs_qgroup *qgroup;
2891         struct btrfs_fs_info *fs_info = root->fs_info;
2892         u64 ref_root = root->root_key.objectid;
2893         int ret = 0;
2894         struct ulist_node *unode;
2895         struct ulist_iterator uiter;
2896
2897         if (!is_fstree(ref_root))
2898                 return 0;
2899
2900         if (num_bytes == 0)
2901                 return 0;
2902
2903         if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2904             capable(CAP_SYS_RESOURCE))
2905                 enforce = false;
2906
2907         spin_lock(&fs_info->qgroup_lock);
2908         quota_root = fs_info->quota_root;
2909         if (!quota_root)
2910                 goto out;
2911
2912         qgroup = find_qgroup_rb(fs_info, ref_root);
2913         if (!qgroup)
2914                 goto out;
2915
2916         /*
2917          * in a first step, we check all affected qgroups if any limits would
2918          * be exceeded
2919          */
2920         ulist_reinit(fs_info->qgroup_ulist);
2921         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2922                         qgroup_to_aux(qgroup), GFP_ATOMIC);
2923         if (ret < 0)
2924                 goto out;
2925         ULIST_ITER_INIT(&uiter);
2926         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2927                 struct btrfs_qgroup *qg;
2928                 struct btrfs_qgroup_list *glist;
2929
2930                 qg = unode_aux_to_qgroup(unode);
2931
2932                 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2933                         ret = -EDQUOT;
2934                         goto out;
2935                 }
2936
2937                 list_for_each_entry(glist, &qg->groups, next_group) {
2938                         ret = ulist_add(fs_info->qgroup_ulist,
2939                                         glist->group->qgroupid,
2940                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
2941                         if (ret < 0)
2942                                 goto out;
2943                 }
2944         }
2945         ret = 0;
2946         /*
2947          * no limits exceeded, now record the reservation into all qgroups
2948          */
2949         ULIST_ITER_INIT(&uiter);
2950         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2951                 struct btrfs_qgroup *qg;
2952
2953                 qg = unode_aux_to_qgroup(unode);
2954
2955                 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2956                 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2957         }
2958
2959 out:
2960         spin_unlock(&fs_info->qgroup_lock);
2961         return ret;
2962 }
2963
2964 /*
2965  * Free @num_bytes of reserved space with @type for qgroup.  (Normally level 0
2966  * qgroup).
2967  *
2968  * Will handle all higher level qgroup too.
2969  *
2970  * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2971  * This special case is only used for META_PERTRANS type.
2972  */
2973 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2974                                u64 ref_root, u64 num_bytes,
2975                                enum btrfs_qgroup_rsv_type type)
2976 {
2977         struct btrfs_root *quota_root;
2978         struct btrfs_qgroup *qgroup;
2979         struct ulist_node *unode;
2980         struct ulist_iterator uiter;
2981         int ret = 0;
2982
2983         if (!is_fstree(ref_root))
2984                 return;
2985
2986         if (num_bytes == 0)
2987                 return;
2988
2989         if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2990                 WARN(1, "%s: Invalid type to free", __func__);
2991                 return;
2992         }
2993         spin_lock(&fs_info->qgroup_lock);
2994
2995         quota_root = fs_info->quota_root;
2996         if (!quota_root)
2997                 goto out;
2998
2999         qgroup = find_qgroup_rb(fs_info, ref_root);
3000         if (!qgroup)
3001                 goto out;
3002
3003         if (num_bytes == (u64)-1)
3004                 /*
3005                  * We're freeing all pertrans rsv, get reserved value from
3006                  * level 0 qgroup as real num_bytes to free.
3007                  */
3008                 num_bytes = qgroup->rsv.values[type];
3009
3010         ulist_reinit(fs_info->qgroup_ulist);
3011         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3012                         qgroup_to_aux(qgroup), GFP_ATOMIC);
3013         if (ret < 0)
3014                 goto out;
3015         ULIST_ITER_INIT(&uiter);
3016         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3017                 struct btrfs_qgroup *qg;
3018                 struct btrfs_qgroup_list *glist;
3019
3020                 qg = unode_aux_to_qgroup(unode);
3021
3022                 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
3023                 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3024
3025                 list_for_each_entry(glist, &qg->groups, next_group) {
3026                         ret = ulist_add(fs_info->qgroup_ulist,
3027                                         glist->group->qgroupid,
3028                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3029                         if (ret < 0)
3030                                 goto out;
3031                 }
3032         }
3033
3034 out:
3035         spin_unlock(&fs_info->qgroup_lock);
3036 }
3037
3038 /*
3039  * Check if the leaf is the last leaf. Which means all node pointers
3040  * are at their last position.
3041  */
3042 static bool is_last_leaf(struct btrfs_path *path)
3043 {
3044         int i;
3045
3046         for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3047                 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3048                         return false;
3049         }
3050         return true;
3051 }
3052
3053 /*
3054  * returns < 0 on error, 0 when more leafs are to be scanned.
3055  * returns 1 when done.
3056  */
3057 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3058                               struct btrfs_path *path)
3059 {
3060         struct btrfs_fs_info *fs_info = trans->fs_info;
3061         struct btrfs_key found;
3062         struct extent_buffer *scratch_leaf = NULL;
3063         struct ulist *roots = NULL;
3064         u64 num_bytes;
3065         bool done;
3066         int slot;
3067         int ret;
3068
3069         mutex_lock(&fs_info->qgroup_rescan_lock);
3070         ret = btrfs_search_slot_for_read(fs_info->extent_root,
3071                                          &fs_info->qgroup_rescan_progress,
3072                                          path, 1, 0);
3073
3074         btrfs_debug(fs_info,
3075                 "current progress key (%llu %u %llu), search_slot ret %d",
3076                 fs_info->qgroup_rescan_progress.objectid,
3077                 fs_info->qgroup_rescan_progress.type,
3078                 fs_info->qgroup_rescan_progress.offset, ret);
3079
3080         if (ret) {
3081                 /*
3082                  * The rescan is about to end, we will not be scanning any
3083                  * further blocks. We cannot unset the RESCAN flag here, because
3084                  * we want to commit the transaction if everything went well.
3085                  * To make the live accounting work in this phase, we set our
3086                  * scan progress pointer such that every real extent objectid
3087                  * will be smaller.
3088                  */
3089                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3090                 btrfs_release_path(path);
3091                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3092                 return ret;
3093         }
3094         done = is_last_leaf(path);
3095
3096         btrfs_item_key_to_cpu(path->nodes[0], &found,
3097                               btrfs_header_nritems(path->nodes[0]) - 1);
3098         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3099
3100         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3101         if (!scratch_leaf) {
3102                 ret = -ENOMEM;
3103                 mutex_unlock(&fs_info->qgroup_rescan_lock);
3104                 goto out;
3105         }
3106         extent_buffer_get(scratch_leaf);
3107         btrfs_tree_read_lock(scratch_leaf);
3108         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
3109         slot = path->slots[0];
3110         btrfs_release_path(path);
3111         mutex_unlock(&fs_info->qgroup_rescan_lock);
3112
3113         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3114                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3115                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3116                     found.type != BTRFS_METADATA_ITEM_KEY)
3117                         continue;
3118                 if (found.type == BTRFS_METADATA_ITEM_KEY)
3119                         num_bytes = fs_info->nodesize;
3120                 else
3121                         num_bytes = found.offset;
3122
3123                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3124                                            &roots, false);
3125                 if (ret < 0)
3126                         goto out;
3127                 /* For rescan, just pass old_roots as NULL */
3128                 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3129                                                   num_bytes, NULL, roots);
3130                 if (ret < 0)
3131                         goto out;
3132         }
3133 out:
3134         if (scratch_leaf) {
3135                 btrfs_tree_read_unlock_blocking(scratch_leaf);
3136                 free_extent_buffer(scratch_leaf);
3137         }
3138
3139         if (done && !ret) {
3140                 ret = 1;
3141                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3142         }
3143         return ret;
3144 }
3145
3146 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3147 {
3148         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3149                                                      qgroup_rescan_work);
3150         struct btrfs_path *path;
3151         struct btrfs_trans_handle *trans = NULL;
3152         int err = -ENOMEM;
3153         int ret = 0;
3154
3155         path = btrfs_alloc_path();
3156         if (!path)
3157                 goto out;
3158         /*
3159          * Rescan should only search for commit root, and any later difference
3160          * should be recorded by qgroup
3161          */
3162         path->search_commit_root = 1;
3163         path->skip_locking = 1;
3164
3165         err = 0;
3166         while (!err && !btrfs_fs_closing(fs_info)) {
3167                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3168                 if (IS_ERR(trans)) {
3169                         err = PTR_ERR(trans);
3170                         break;
3171                 }
3172                 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3173                         err = -EINTR;
3174                 } else {
3175                         err = qgroup_rescan_leaf(trans, path);
3176                 }
3177                 if (err > 0)
3178                         btrfs_commit_transaction(trans);
3179                 else
3180                         btrfs_end_transaction(trans);
3181         }
3182
3183 out:
3184         btrfs_free_path(path);
3185
3186         mutex_lock(&fs_info->qgroup_rescan_lock);
3187         if (!btrfs_fs_closing(fs_info))
3188                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3189
3190         if (err > 0 &&
3191             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3192                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3193         } else if (err < 0) {
3194                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3195         }
3196         mutex_unlock(&fs_info->qgroup_rescan_lock);
3197
3198         /*
3199          * only update status, since the previous part has already updated the
3200          * qgroup info.
3201          */
3202         trans = btrfs_start_transaction(fs_info->quota_root, 1);
3203         if (IS_ERR(trans)) {
3204                 err = PTR_ERR(trans);
3205                 btrfs_err(fs_info,
3206                           "fail to start transaction for status update: %d",
3207                           err);
3208                 goto done;
3209         }
3210         ret = update_qgroup_status_item(trans);
3211         if (ret < 0) {
3212                 err = ret;
3213                 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3214         }
3215         btrfs_end_transaction(trans);
3216
3217         if (btrfs_fs_closing(fs_info)) {
3218                 btrfs_info(fs_info, "qgroup scan paused");
3219         } else if (err >= 0) {
3220                 btrfs_info(fs_info, "qgroup scan completed%s",
3221                         err > 0 ? " (inconsistency flag cleared)" : "");
3222         } else {
3223                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3224         }
3225
3226 done:
3227         mutex_lock(&fs_info->qgroup_rescan_lock);
3228         fs_info->qgroup_rescan_running = false;
3229         mutex_unlock(&fs_info->qgroup_rescan_lock);
3230         complete_all(&fs_info->qgroup_rescan_completion);
3231 }
3232
3233 /*
3234  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3235  * memory required for the rescan context.
3236  */
3237 static int
3238 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3239                    int init_flags)
3240 {
3241         int ret = 0;
3242
3243         if (!init_flags) {
3244                 /* we're resuming qgroup rescan at mount time */
3245                 if (!(fs_info->qgroup_flags &
3246                       BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3247                         btrfs_warn(fs_info,
3248                         "qgroup rescan init failed, qgroup is not enabled");
3249                         ret = -EINVAL;
3250                 } else if (!(fs_info->qgroup_flags &
3251                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3252                         btrfs_warn(fs_info,
3253                         "qgroup rescan init failed, qgroup rescan is not queued");
3254                         ret = -EINVAL;
3255                 }
3256
3257                 if (ret)
3258                         return ret;
3259         }
3260
3261         mutex_lock(&fs_info->qgroup_rescan_lock);
3262         spin_lock(&fs_info->qgroup_lock);
3263
3264         if (init_flags) {
3265                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3266                         btrfs_warn(fs_info,
3267                                    "qgroup rescan is already in progress");
3268                         ret = -EINPROGRESS;
3269                 } else if (!(fs_info->qgroup_flags &
3270                              BTRFS_QGROUP_STATUS_FLAG_ON)) {
3271                         btrfs_warn(fs_info,
3272                         "qgroup rescan init failed, qgroup is not enabled");
3273                         ret = -EINVAL;
3274                 }
3275
3276                 if (ret) {
3277                         spin_unlock(&fs_info->qgroup_lock);
3278                         mutex_unlock(&fs_info->qgroup_rescan_lock);
3279                         return ret;
3280                 }
3281                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3282         }
3283
3284         memset(&fs_info->qgroup_rescan_progress, 0,
3285                 sizeof(fs_info->qgroup_rescan_progress));
3286         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3287         init_completion(&fs_info->qgroup_rescan_completion);
3288         fs_info->qgroup_rescan_running = true;
3289
3290         spin_unlock(&fs_info->qgroup_lock);
3291         mutex_unlock(&fs_info->qgroup_rescan_lock);
3292
3293         memset(&fs_info->qgroup_rescan_work, 0,
3294                sizeof(fs_info->qgroup_rescan_work));
3295         btrfs_init_work(&fs_info->qgroup_rescan_work,
3296                         btrfs_qgroup_rescan_helper,
3297                         btrfs_qgroup_rescan_worker, NULL, NULL);
3298         return 0;
3299 }
3300
3301 static void
3302 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3303 {
3304         struct rb_node *n;
3305         struct btrfs_qgroup *qgroup;
3306
3307         spin_lock(&fs_info->qgroup_lock);
3308         /* clear all current qgroup tracking information */
3309         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3310                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3311                 qgroup->rfer = 0;
3312                 qgroup->rfer_cmpr = 0;
3313                 qgroup->excl = 0;
3314                 qgroup->excl_cmpr = 0;
3315                 qgroup_dirty(fs_info, qgroup);
3316         }
3317         spin_unlock(&fs_info->qgroup_lock);
3318 }
3319
3320 int
3321 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3322 {
3323         int ret = 0;
3324         struct btrfs_trans_handle *trans;
3325
3326         ret = qgroup_rescan_init(fs_info, 0, 1);
3327         if (ret)
3328                 return ret;
3329
3330         /*
3331          * We have set the rescan_progress to 0, which means no more
3332          * delayed refs will be accounted by btrfs_qgroup_account_ref.
3333          * However, btrfs_qgroup_account_ref may be right after its call
3334          * to btrfs_find_all_roots, in which case it would still do the
3335          * accounting.
3336          * To solve this, we're committing the transaction, which will
3337          * ensure we run all delayed refs and only after that, we are
3338          * going to clear all tracking information for a clean start.
3339          */
3340
3341         trans = btrfs_join_transaction(fs_info->fs_root);
3342         if (IS_ERR(trans)) {
3343                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3344                 return PTR_ERR(trans);
3345         }
3346         ret = btrfs_commit_transaction(trans);
3347         if (ret) {
3348                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3349                 return ret;
3350         }
3351
3352         qgroup_rescan_zero_tracking(fs_info);
3353
3354         btrfs_queue_work(fs_info->qgroup_rescan_workers,
3355                          &fs_info->qgroup_rescan_work);
3356
3357         return 0;
3358 }
3359
3360 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3361                                      bool interruptible)
3362 {
3363         int running;
3364         int ret = 0;
3365
3366         mutex_lock(&fs_info->qgroup_rescan_lock);
3367         spin_lock(&fs_info->qgroup_lock);
3368         running = fs_info->qgroup_rescan_running;
3369         spin_unlock(&fs_info->qgroup_lock);
3370         mutex_unlock(&fs_info->qgroup_rescan_lock);
3371
3372         if (!running)
3373                 return 0;
3374
3375         if (interruptible)
3376                 ret = wait_for_completion_interruptible(
3377                                         &fs_info->qgroup_rescan_completion);
3378         else
3379                 wait_for_completion(&fs_info->qgroup_rescan_completion);
3380
3381         return ret;
3382 }
3383
3384 /*
3385  * this is only called from open_ctree where we're still single threaded, thus
3386  * locking is omitted here.
3387  */
3388 void
3389 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3390 {
3391         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3392                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3393                                  &fs_info->qgroup_rescan_work);
3394 }
3395
3396 /*
3397  * Reserve qgroup space for range [start, start + len).
3398  *
3399  * This function will either reserve space from related qgroups or doing
3400  * nothing if the range is already reserved.
3401  *
3402  * Return 0 for successful reserve
3403  * Return <0 for error (including -EQUOT)
3404  *
3405  * NOTE: this function may sleep for memory allocation.
3406  *       if btrfs_qgroup_reserve_data() is called multiple times with
3407  *       same @reserved, caller must ensure when error happens it's OK
3408  *       to free *ALL* reserved space.
3409  */
3410 int btrfs_qgroup_reserve_data(struct inode *inode,
3411                         struct extent_changeset **reserved_ret, u64 start,
3412                         u64 len)
3413 {
3414         struct btrfs_root *root = BTRFS_I(inode)->root;
3415         struct ulist_node *unode;
3416         struct ulist_iterator uiter;
3417         struct extent_changeset *reserved;
3418         u64 orig_reserved;
3419         u64 to_reserve;
3420         int ret;
3421
3422         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3423             !is_fstree(root->root_key.objectid) || len == 0)
3424                 return 0;
3425
3426         /* @reserved parameter is mandatory for qgroup */
3427         if (WARN_ON(!reserved_ret))
3428                 return -EINVAL;
3429         if (!*reserved_ret) {
3430                 *reserved_ret = extent_changeset_alloc();
3431                 if (!*reserved_ret)
3432                         return -ENOMEM;
3433         }
3434         reserved = *reserved_ret;
3435         /* Record already reserved space */
3436         orig_reserved = reserved->bytes_changed;
3437         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3438                         start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3439
3440         /* Newly reserved space */
3441         to_reserve = reserved->bytes_changed - orig_reserved;
3442         trace_btrfs_qgroup_reserve_data(inode, start, len,
3443                                         to_reserve, QGROUP_RESERVE);
3444         if (ret < 0)
3445                 goto cleanup;
3446         ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3447         if (ret < 0)
3448                 goto cleanup;
3449
3450         return ret;
3451
3452 cleanup:
3453         /* cleanup *ALL* already reserved ranges */
3454         ULIST_ITER_INIT(&uiter);
3455         while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3456                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3457                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3458         extent_changeset_release(reserved);
3459         return ret;
3460 }
3461
3462 /* Free ranges specified by @reserved, normally in error path */
3463 static int qgroup_free_reserved_data(struct inode *inode,
3464                         struct extent_changeset *reserved, u64 start, u64 len)
3465 {
3466         struct btrfs_root *root = BTRFS_I(inode)->root;
3467         struct ulist_node *unode;
3468         struct ulist_iterator uiter;
3469         struct extent_changeset changeset;
3470         int freed = 0;
3471         int ret;
3472
3473         extent_changeset_init(&changeset);
3474         len = round_up(start + len, root->fs_info->sectorsize);
3475         start = round_down(start, root->fs_info->sectorsize);
3476
3477         ULIST_ITER_INIT(&uiter);
3478         while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3479                 u64 range_start = unode->val;
3480                 /* unode->aux is the inclusive end */
3481                 u64 range_len = unode->aux - range_start + 1;
3482                 u64 free_start;
3483                 u64 free_len;
3484
3485                 extent_changeset_release(&changeset);
3486
3487                 /* Only free range in range [start, start + len) */
3488                 if (range_start >= start + len ||
3489                     range_start + range_len <= start)
3490                         continue;
3491                 free_start = max(range_start, start);
3492                 free_len = min(start + len, range_start + range_len) -
3493                            free_start;
3494                 /*
3495                  * TODO: To also modify reserved->ranges_reserved to reflect
3496                  * the modification.
3497                  *
3498                  * However as long as we free qgroup reserved according to
3499                  * EXTENT_QGROUP_RESERVED, we won't double free.
3500                  * So not need to rush.
3501                  */
3502                 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3503                                 free_start, free_start + free_len - 1,
3504                                 EXTENT_QGROUP_RESERVED, &changeset);
3505                 if (ret < 0)
3506                         goto out;
3507                 freed += changeset.bytes_changed;
3508         }
3509         btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3510                                   BTRFS_QGROUP_RSV_DATA);
3511         ret = freed;
3512 out:
3513         extent_changeset_release(&changeset);
3514         return ret;
3515 }
3516
3517 static int __btrfs_qgroup_release_data(struct inode *inode,
3518                         struct extent_changeset *reserved, u64 start, u64 len,
3519                         int free)
3520 {
3521         struct extent_changeset changeset;
3522         int trace_op = QGROUP_RELEASE;
3523         int ret;
3524
3525         if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3526                       &BTRFS_I(inode)->root->fs_info->flags))
3527                 return 0;
3528
3529         /* In release case, we shouldn't have @reserved */
3530         WARN_ON(!free && reserved);
3531         if (free && reserved)
3532                 return qgroup_free_reserved_data(inode, reserved, start, len);
3533         extent_changeset_init(&changeset);
3534         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start, 
3535                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3536         if (ret < 0)
3537                 goto out;
3538
3539         if (free)
3540                 trace_op = QGROUP_FREE;
3541         trace_btrfs_qgroup_release_data(inode, start, len,
3542                                         changeset.bytes_changed, trace_op);
3543         if (free)
3544                 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3545                                 BTRFS_I(inode)->root->root_key.objectid,
3546                                 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3547         ret = changeset.bytes_changed;
3548 out:
3549         extent_changeset_release(&changeset);
3550         return ret;
3551 }
3552
3553 /*
3554  * Free a reserved space range from io_tree and related qgroups
3555  *
3556  * Should be called when a range of pages get invalidated before reaching disk.
3557  * Or for error cleanup case.
3558  * if @reserved is given, only reserved range in [@start, @start + @len) will
3559  * be freed.
3560  *
3561  * For data written to disk, use btrfs_qgroup_release_data().
3562  *
3563  * NOTE: This function may sleep for memory allocation.
3564  */
3565 int btrfs_qgroup_free_data(struct inode *inode,
3566                         struct extent_changeset *reserved, u64 start, u64 len)
3567 {
3568         return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3569 }
3570
3571 /*
3572  * Release a reserved space range from io_tree only.
3573  *
3574  * Should be called when a range of pages get written to disk and corresponding
3575  * FILE_EXTENT is inserted into corresponding root.
3576  *
3577  * Since new qgroup accounting framework will only update qgroup numbers at
3578  * commit_transaction() time, its reserved space shouldn't be freed from
3579  * related qgroups.
3580  *
3581  * But we should release the range from io_tree, to allow further write to be
3582  * COWed.
3583  *
3584  * NOTE: This function may sleep for memory allocation.
3585  */
3586 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3587 {
3588         return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3589 }
3590
3591 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3592                               enum btrfs_qgroup_rsv_type type)
3593 {
3594         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3595             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3596                 return;
3597         if (num_bytes == 0)
3598                 return;
3599
3600         spin_lock(&root->qgroup_meta_rsv_lock);
3601         if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3602                 root->qgroup_meta_rsv_prealloc += num_bytes;
3603         else
3604                 root->qgroup_meta_rsv_pertrans += num_bytes;
3605         spin_unlock(&root->qgroup_meta_rsv_lock);
3606 }
3607
3608 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3609                              enum btrfs_qgroup_rsv_type type)
3610 {
3611         if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3612             type != BTRFS_QGROUP_RSV_META_PERTRANS)
3613                 return 0;
3614         if (num_bytes == 0)
3615                 return 0;
3616
3617         spin_lock(&root->qgroup_meta_rsv_lock);
3618         if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3619                 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3620                                   num_bytes);
3621                 root->qgroup_meta_rsv_prealloc -= num_bytes;
3622         } else {
3623                 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3624                                   num_bytes);
3625                 root->qgroup_meta_rsv_pertrans -= num_bytes;
3626         }
3627         spin_unlock(&root->qgroup_meta_rsv_lock);
3628         return num_bytes;
3629 }
3630
3631 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3632                                 enum btrfs_qgroup_rsv_type type, bool enforce)
3633 {
3634         struct btrfs_fs_info *fs_info = root->fs_info;
3635         int ret;
3636
3637         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3638             !is_fstree(root->root_key.objectid) || num_bytes == 0)
3639                 return 0;
3640
3641         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3642         trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3643         ret = qgroup_reserve(root, num_bytes, enforce, type);
3644         if (ret < 0)
3645                 return ret;
3646         /*
3647          * Record what we have reserved into root.
3648          *
3649          * To avoid quota disabled->enabled underflow.
3650          * In that case, we may try to free space we haven't reserved
3651          * (since quota was disabled), so record what we reserved into root.
3652          * And ensure later release won't underflow this number.
3653          */
3654         add_root_meta_rsv(root, num_bytes, type);
3655         return ret;
3656 }
3657
3658 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3659 {
3660         struct btrfs_fs_info *fs_info = root->fs_info;
3661
3662         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3663             !is_fstree(root->root_key.objectid))
3664                 return;
3665
3666         /* TODO: Update trace point to handle such free */
3667         trace_qgroup_meta_free_all_pertrans(root);
3668         /* Special value -1 means to free all reserved space */
3669         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3670                                   BTRFS_QGROUP_RSV_META_PERTRANS);
3671 }
3672
3673 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3674                               enum btrfs_qgroup_rsv_type type)
3675 {
3676         struct btrfs_fs_info *fs_info = root->fs_info;
3677
3678         if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3679             !is_fstree(root->root_key.objectid))
3680                 return;
3681
3682         /*
3683          * reservation for META_PREALLOC can happen before quota is enabled,
3684          * which can lead to underflow.
3685          * Here ensure we will only free what we really have reserved.
3686          */
3687         num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3688         BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3689         trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3690         btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3691                                   num_bytes, type);
3692 }
3693
3694 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3695                                 int num_bytes)
3696 {
3697         struct btrfs_root *quota_root = fs_info->quota_root;
3698         struct btrfs_qgroup *qgroup;
3699         struct ulist_node *unode;
3700         struct ulist_iterator uiter;
3701         int ret = 0;
3702
3703         if (num_bytes == 0)
3704                 return;
3705         if (!quota_root)
3706                 return;
3707
3708         spin_lock(&fs_info->qgroup_lock);
3709         qgroup = find_qgroup_rb(fs_info, ref_root);
3710         if (!qgroup)
3711                 goto out;
3712         ulist_reinit(fs_info->qgroup_ulist);
3713         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3714                        qgroup_to_aux(qgroup), GFP_ATOMIC);
3715         if (ret < 0)
3716                 goto out;
3717         ULIST_ITER_INIT(&uiter);
3718         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3719                 struct btrfs_qgroup *qg;
3720                 struct btrfs_qgroup_list *glist;
3721
3722                 qg = unode_aux_to_qgroup(unode);
3723
3724                 qgroup_rsv_release(fs_info, qg, num_bytes,
3725                                 BTRFS_QGROUP_RSV_META_PREALLOC);
3726                 qgroup_rsv_add(fs_info, qg, num_bytes,
3727                                 BTRFS_QGROUP_RSV_META_PERTRANS);
3728                 list_for_each_entry(glist, &qg->groups, next_group) {
3729                         ret = ulist_add(fs_info->qgroup_ulist,
3730                                         glist->group->qgroupid,
3731                                         qgroup_to_aux(glist->group), GFP_ATOMIC);
3732                         if (ret < 0)
3733                                 goto out;
3734                 }
3735         }
3736 out:
3737         spin_unlock(&fs_info->qgroup_lock);
3738 }
3739
3740 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3741 {
3742         struct btrfs_fs_info *fs_info = root->fs_info;
3743