~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/qgroup.c

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2011 STRATO.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 
 19 #include <linux/sched.h>
 20 #include <linux/pagemap.h>
 21 #include <linux/writeback.h>
 22 #include <linux/blkdev.h>
 23 #include <linux/rbtree.h>
 24 #include <linux/slab.h>
 25 #include <linux/workqueue.h>
 26 #include <linux/btrfs.h>
 27 
 28 #include "ctree.h"
 29 #include "transaction.h"
 30 #include "disk-io.h"
 31 #include "locking.h"
 32 #include "ulist.h"
 33 #include "backref.h"
 34 #include "extent_io.h"
 35 #include "qgroup.h"
 36 
 37 
 38 /* TODO XXX FIXME
 39  *  - subvol delete -> delete when ref goes to 0? delete limits also?
 40  *  - reorganize keys
 41  *  - compressed
 42  *  - sync
 43  *  - copy also limits on subvol creation
 44  *  - limit
 45  *  - caches fuer ulists
 46  *  - performance benchmarks
 47  *  - check all ioctl parameters
 48  */
 49 
 50 /*
 51  * one struct for each qgroup, organized in fs_info->qgroup_tree.
 52  */
 53 struct btrfs_qgroup {
 54         u64 qgroupid;
 55 
 56         /*
 57          * state
 58          */
 59         u64 rfer;       /* referenced */
 60         u64 rfer_cmpr;  /* referenced compressed */
 61         u64 excl;       /* exclusive */
 62         u64 excl_cmpr;  /* exclusive compressed */
 63 
 64         /*
 65          * limits
 66          */
 67         u64 lim_flags;  /* which limits are set */
 68         u64 max_rfer;
 69         u64 max_excl;
 70         u64 rsv_rfer;
 71         u64 rsv_excl;
 72 
 73         /*
 74          * reservation tracking
 75          */
 76         u64 reserved;
 77 
 78         /*
 79          * lists
 80          */
 81         struct list_head groups;  /* groups this group is member of */
 82         struct list_head members; /* groups that are members of this group */
 83         struct list_head dirty;   /* dirty groups */
 84         struct rb_node node;      /* tree of qgroups */
 85 
 86         /*
 87          * temp variables for accounting operations
 88          * Refer to qgroup_shared_accounting() for details.
 89          */
 90         u64 old_refcnt;
 91         u64 new_refcnt;
 92 };
 93 
 94 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
 95                                            int mod)
 96 {
 97         if (qg->old_refcnt < seq)
 98                 qg->old_refcnt = seq;
 99         qg->old_refcnt += mod;
100 }
101 
102 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
103                                            int mod)
104 {
105         if (qg->new_refcnt < seq)
106                 qg->new_refcnt = seq;
107         qg->new_refcnt += mod;
108 }
109 
110 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
111 {
112         if (qg->old_refcnt < seq)
113                 return 0;
114         return qg->old_refcnt - seq;
115 }
116 
117 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
118 {
119         if (qg->new_refcnt < seq)
120                 return 0;
121         return qg->new_refcnt - seq;
122 }
123 
124 /*
125  * glue structure to represent the relations between qgroups.
126  */
127 struct btrfs_qgroup_list {
128         struct list_head next_group;
129         struct list_head next_member;
130         struct btrfs_qgroup *group;
131         struct btrfs_qgroup *member;
132 };
133 
134 #define ptr_to_u64(x) ((u64)(uintptr_t)x)
135 #define u64_to_ptr(x) ((struct btrfs_qgroup *)(uintptr_t)x)
136 
137 static int
138 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
139                    int init_flags);
140 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
141 
142 /* must be called with qgroup_ioctl_lock held */
143 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
144                                            u64 qgroupid)
145 {
146         struct rb_node *n = fs_info->qgroup_tree.rb_node;
147         struct btrfs_qgroup *qgroup;
148 
149         while (n) {
150                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
151                 if (qgroup->qgroupid < qgroupid)
152                         n = n->rb_left;
153                 else if (qgroup->qgroupid > qgroupid)
154                         n = n->rb_right;
155                 else
156                         return qgroup;
157         }
158         return NULL;
159 }
160 
161 /* must be called with qgroup_lock held */
162 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
163                                           u64 qgroupid)
164 {
165         struct rb_node **p = &fs_info->qgroup_tree.rb_node;
166         struct rb_node *parent = NULL;
167         struct btrfs_qgroup *qgroup;
168 
169         while (*p) {
170                 parent = *p;
171                 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
172 
173                 if (qgroup->qgroupid < qgroupid)
174                         p = &(*p)->rb_left;
175                 else if (qgroup->qgroupid > qgroupid)
176                         p = &(*p)->rb_right;
177                 else
178                         return qgroup;
179         }
180 
181         qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
182         if (!qgroup)
183                 return ERR_PTR(-ENOMEM);
184 
185         qgroup->qgroupid = qgroupid;
186         INIT_LIST_HEAD(&qgroup->groups);
187         INIT_LIST_HEAD(&qgroup->members);
188         INIT_LIST_HEAD(&qgroup->dirty);
189 
190         rb_link_node(&qgroup->node, parent, p);
191         rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
192 
193         return qgroup;
194 }
195 
196 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
197 {
198         struct btrfs_qgroup_list *list;
199 
200         list_del(&qgroup->dirty);
201         while (!list_empty(&qgroup->groups)) {
202                 list = list_first_entry(&qgroup->groups,
203                                         struct btrfs_qgroup_list, next_group);
204                 list_del(&list->next_group);
205                 list_del(&list->next_member);
206                 kfree(list);
207         }
208 
209         while (!list_empty(&qgroup->members)) {
210                 list = list_first_entry(&qgroup->members,
211                                         struct btrfs_qgroup_list, next_member);
212                 list_del(&list->next_group);
213                 list_del(&list->next_member);
214                 kfree(list);
215         }
216         kfree(qgroup);
217 }
218 
219 /* must be called with qgroup_lock held */
220 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
221 {
222         struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
223 
224         if (!qgroup)
225                 return -ENOENT;
226 
227         rb_erase(&qgroup->node, &fs_info->qgroup_tree);
228         __del_qgroup_rb(qgroup);
229         return 0;
230 }
231 
232 /* must be called with qgroup_lock held */
233 static int add_relation_rb(struct btrfs_fs_info *fs_info,
234                            u64 memberid, u64 parentid)
235 {
236         struct btrfs_qgroup *member;
237         struct btrfs_qgroup *parent;
238         struct btrfs_qgroup_list *list;
239 
240         member = find_qgroup_rb(fs_info, memberid);
241         parent = find_qgroup_rb(fs_info, parentid);
242         if (!member || !parent)
243                 return -ENOENT;
244 
245         list = kzalloc(sizeof(*list), GFP_ATOMIC);
246         if (!list)
247                 return -ENOMEM;
248 
249         list->group = parent;
250         list->member = member;
251         list_add_tail(&list->next_group, &member->groups);
252         list_add_tail(&list->next_member, &parent->members);
253 
254         return 0;
255 }
256 
257 /* must be called with qgroup_lock held */
258 static int del_relation_rb(struct btrfs_fs_info *fs_info,
259                            u64 memberid, u64 parentid)
260 {
261         struct btrfs_qgroup *member;
262         struct btrfs_qgroup *parent;
263         struct btrfs_qgroup_list *list;
264 
265         member = find_qgroup_rb(fs_info, memberid);
266         parent = find_qgroup_rb(fs_info, parentid);
267         if (!member || !parent)
268                 return -ENOENT;
269 
270         list_for_each_entry(list, &member->groups, next_group) {
271                 if (list->group == parent) {
272                         list_del(&list->next_group);
273                         list_del(&list->next_member);
274                         kfree(list);
275                         return 0;
276                 }
277         }
278         return -ENOENT;
279 }
280 
281 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
282 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
283                                u64 rfer, u64 excl)
284 {
285         struct btrfs_qgroup *qgroup;
286 
287         qgroup = find_qgroup_rb(fs_info, qgroupid);
288         if (!qgroup)
289                 return -EINVAL;
290         if (qgroup->rfer != rfer || qgroup->excl != excl)
291                 return -EINVAL;
292         return 0;
293 }
294 #endif
295 
296 /*
297  * The full config is read in one go, only called from open_ctree()
298  * It doesn't use any locking, as at this point we're still single-threaded
299  */
300 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
301 {
302         struct btrfs_key key;
303         struct btrfs_key found_key;
304         struct btrfs_root *quota_root = fs_info->quota_root;
305         struct btrfs_path *path = NULL;
306         struct extent_buffer *l;
307         int slot;
308         int ret = 0;
309         u64 flags = 0;
310         u64 rescan_progress = 0;
311 
312         if (!fs_info->quota_enabled)
313                 return 0;
314 
315         fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
316         if (!fs_info->qgroup_ulist) {
317                 ret = -ENOMEM;
318                 goto out;
319         }
320 
321         path = btrfs_alloc_path();
322         if (!path) {
323                 ret = -ENOMEM;
324                 goto out;
325         }
326 
327         /* default this to quota off, in case no status key is found */
328         fs_info->qgroup_flags = 0;
329 
330         /*
331          * pass 1: read status, all qgroup infos and limits
332          */
333         key.objectid = 0;
334         key.type = 0;
335         key.offset = 0;
336         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
337         if (ret)
338                 goto out;
339 
340         while (1) {
341                 struct btrfs_qgroup *qgroup;
342 
343                 slot = path->slots[0];
344                 l = path->nodes[0];
345                 btrfs_item_key_to_cpu(l, &found_key, slot);
346 
347                 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
348                         struct btrfs_qgroup_status_item *ptr;
349 
350                         ptr = btrfs_item_ptr(l, slot,
351                                              struct btrfs_qgroup_status_item);
352 
353                         if (btrfs_qgroup_status_version(l, ptr) !=
354                             BTRFS_QGROUP_STATUS_VERSION) {
355                                 btrfs_err(fs_info,
356                                  "old qgroup version, quota disabled");
357                                 goto out;
358                         }
359                         if (btrfs_qgroup_status_generation(l, ptr) !=
360                             fs_info->generation) {
361                                 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
362                                 btrfs_err(fs_info,
363                                         "qgroup generation mismatch, "
364                                         "marked as inconsistent");
365                         }
366                         fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
367                                                                           ptr);
368                         rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
369                         goto next1;
370                 }
371 
372                 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
373                     found_key.type != BTRFS_QGROUP_LIMIT_KEY)
374                         goto next1;
375 
376                 qgroup = find_qgroup_rb(fs_info, found_key.offset);
377                 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
378                     (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
379                         btrfs_err(fs_info, "inconsistent qgroup config");
380                         flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
381                 }
382                 if (!qgroup) {
383                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
384                         if (IS_ERR(qgroup)) {
385                                 ret = PTR_ERR(qgroup);
386                                 goto out;
387                         }
388                 }
389                 switch (found_key.type) {
390                 case BTRFS_QGROUP_INFO_KEY: {
391                         struct btrfs_qgroup_info_item *ptr;
392 
393                         ptr = btrfs_item_ptr(l, slot,
394                                              struct btrfs_qgroup_info_item);
395                         qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
396                         qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
397                         qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
398                         qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
399                         /* generation currently unused */
400                         break;
401                 }
402                 case BTRFS_QGROUP_LIMIT_KEY: {
403                         struct btrfs_qgroup_limit_item *ptr;
404 
405                         ptr = btrfs_item_ptr(l, slot,
406                                              struct btrfs_qgroup_limit_item);
407                         qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
408                         qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
409                         qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
410                         qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
411                         qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
412                         break;
413                 }
414                 }
415 next1:
416                 ret = btrfs_next_item(quota_root, path);
417                 if (ret < 0)
418                         goto out;
419                 if (ret)
420                         break;
421         }
422         btrfs_release_path(path);
423 
424         /*
425          * pass 2: read all qgroup relations
426          */
427         key.objectid = 0;
428         key.type = BTRFS_QGROUP_RELATION_KEY;
429         key.offset = 0;
430         ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
431         if (ret)
432                 goto out;
433         while (1) {
434                 slot = path->slots[0];
435                 l = path->nodes[0];
436                 btrfs_item_key_to_cpu(l, &found_key, slot);
437 
438                 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
439                         goto next2;
440 
441                 if (found_key.objectid > found_key.offset) {
442                         /* parent <- member, not needed to build config */
443                         /* FIXME should we omit the key completely? */
444                         goto next2;
445                 }
446 
447                 ret = add_relation_rb(fs_info, found_key.objectid,
448                                       found_key.offset);
449                 if (ret == -ENOENT) {
450                         btrfs_warn(fs_info,
451                                 "orphan qgroup relation 0x%llx->0x%llx",
452                                 found_key.objectid, found_key.offset);
453                         ret = 0;        /* ignore the error */
454                 }
455                 if (ret)
456                         goto out;
457 next2:
458                 ret = btrfs_next_item(quota_root, path);
459                 if (ret < 0)
460                         goto out;
461                 if (ret)
462                         break;
463         }
464 out:
465         fs_info->qgroup_flags |= flags;
466         if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
467                 fs_info->quota_enabled = 0;
468                 fs_info->pending_quota_state = 0;
469         } else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
470                    ret >= 0) {
471                 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
472         }
473         btrfs_free_path(path);
474 
475         if (ret < 0) {
476                 ulist_free(fs_info->qgroup_ulist);
477                 fs_info->qgroup_ulist = NULL;
478                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
479         }
480 
481         return ret < 0 ? ret : 0;
482 }
483 
484 /*
485  * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
486  * first two are in single-threaded paths.And for the third one, we have set
487  * quota_root to be null with qgroup_lock held before, so it is safe to clean
488  * up the in-memory structures without qgroup_lock held.
489  */
490 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
491 {
492         struct rb_node *n;
493         struct btrfs_qgroup *qgroup;
494 
495         while ((n = rb_first(&fs_info->qgroup_tree))) {
496                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
497                 rb_erase(n, &fs_info->qgroup_tree);
498                 __del_qgroup_rb(qgroup);
499         }
500         /*
501          * we call btrfs_free_qgroup_config() when umounting
502          * filesystem and disabling quota, so we set qgroup_ulist
503          * to be null here to avoid double free.
504          */
505         ulist_free(fs_info->qgroup_ulist);
506         fs_info->qgroup_ulist = NULL;
507 }
508 
509 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
510                                     struct btrfs_root *quota_root,
511                                     u64 src, u64 dst)
512 {
513         int ret;
514         struct btrfs_path *path;
515         struct btrfs_key key;
516 
517         path = btrfs_alloc_path();
518         if (!path)
519                 return -ENOMEM;
520 
521         key.objectid = src;
522         key.type = BTRFS_QGROUP_RELATION_KEY;
523         key.offset = dst;
524 
525         ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
526 
527         btrfs_mark_buffer_dirty(path->nodes[0]);
528 
529         btrfs_free_path(path);
530         return ret;
531 }
532 
533 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
534                                     struct btrfs_root *quota_root,
535                                     u64 src, u64 dst)
536 {
537         int ret;
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_search_slot(trans, quota_root, &key, path, -1, 1);
550         if (ret < 0)
551                 goto out;
552 
553         if (ret > 0) {
554                 ret = -ENOENT;
555                 goto out;
556         }
557 
558         ret = btrfs_del_item(trans, quota_root, path);
559 out:
560         btrfs_free_path(path);
561         return ret;
562 }
563 
564 static int add_qgroup_item(struct btrfs_trans_handle *trans,
565                            struct btrfs_root *quota_root, u64 qgroupid)
566 {
567         int ret;
568         struct btrfs_path *path;
569         struct btrfs_qgroup_info_item *qgroup_info;
570         struct btrfs_qgroup_limit_item *qgroup_limit;
571         struct extent_buffer *leaf;
572         struct btrfs_key key;
573 
574         if (btrfs_is_testing(quota_root->fs_info))
575                 return 0;
576 
577         path = btrfs_alloc_path();
578         if (!path)
579                 return -ENOMEM;
580 
581         key.objectid = 0;
582         key.type = BTRFS_QGROUP_INFO_KEY;
583         key.offset = qgroupid;
584 
585         /*
586          * Avoid a transaction abort by catching -EEXIST here. In that
587          * case, we proceed by re-initializing the existing structure
588          * on disk.
589          */
590 
591         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
592                                       sizeof(*qgroup_info));
593         if (ret && ret != -EEXIST)
594                 goto out;
595 
596         leaf = path->nodes[0];
597         qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
598                                  struct btrfs_qgroup_info_item);
599         btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
600         btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
601         btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
602         btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
603         btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
604 
605         btrfs_mark_buffer_dirty(leaf);
606 
607         btrfs_release_path(path);
608 
609         key.type = BTRFS_QGROUP_LIMIT_KEY;
610         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
611                                       sizeof(*qgroup_limit));
612         if (ret && ret != -EEXIST)
613                 goto out;
614 
615         leaf = path->nodes[0];
616         qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
617                                   struct btrfs_qgroup_limit_item);
618         btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
619         btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
620         btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
621         btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
622         btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
623 
624         btrfs_mark_buffer_dirty(leaf);
625 
626         ret = 0;
627 out:
628         btrfs_free_path(path);
629         return ret;
630 }
631 
632 static int del_qgroup_item(struct btrfs_trans_handle *trans,
633                            struct btrfs_root *quota_root, u64 qgroupid)
634 {
635         int ret;
636         struct btrfs_path *path;
637         struct btrfs_key key;
638 
639         path = btrfs_alloc_path();
640         if (!path)
641                 return -ENOMEM;
642 
643         key.objectid = 0;
644         key.type = BTRFS_QGROUP_INFO_KEY;
645         key.offset = qgroupid;
646         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
647         if (ret < 0)
648                 goto out;
649 
650         if (ret > 0) {
651                 ret = -ENOENT;
652                 goto out;
653         }
654 
655         ret = btrfs_del_item(trans, quota_root, path);
656         if (ret)
657                 goto out;
658 
659         btrfs_release_path(path);
660 
661         key.type = BTRFS_QGROUP_LIMIT_KEY;
662         ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
663         if (ret < 0)
664                 goto out;
665 
666         if (ret > 0) {
667                 ret = -ENOENT;
668                 goto out;
669         }
670 
671         ret = btrfs_del_item(trans, quota_root, path);
672 
673 out:
674         btrfs_free_path(path);
675         return ret;
676 }
677 
678 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
679                                     struct btrfs_root *root,
680                                     struct btrfs_qgroup *qgroup)
681 {
682         struct btrfs_path *path;
683         struct btrfs_key key;
684         struct extent_buffer *l;
685         struct btrfs_qgroup_limit_item *qgroup_limit;
686         int ret;
687         int slot;
688 
689         key.objectid = 0;
690         key.type = BTRFS_QGROUP_LIMIT_KEY;
691         key.offset = qgroup->qgroupid;
692 
693         path = btrfs_alloc_path();
694         if (!path)
695                 return -ENOMEM;
696 
697         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
698         if (ret > 0)
699                 ret = -ENOENT;
700 
701         if (ret)
702                 goto out;
703 
704         l = path->nodes[0];
705         slot = path->slots[0];
706         qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
707         btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
708         btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
709         btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
710         btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
711         btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
712 
713         btrfs_mark_buffer_dirty(l);
714 
715 out:
716         btrfs_free_path(path);
717         return ret;
718 }
719 
720 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
721                                    struct btrfs_root *root,
722                                    struct btrfs_qgroup *qgroup)
723 {
724         struct btrfs_path *path;
725         struct btrfs_key key;
726         struct extent_buffer *l;
727         struct btrfs_qgroup_info_item *qgroup_info;
728         int ret;
729         int slot;
730 
731         if (btrfs_is_testing(root->fs_info))
732                 return 0;
733 
734         key.objectid = 0;
735         key.type = BTRFS_QGROUP_INFO_KEY;
736         key.offset = qgroup->qgroupid;
737 
738         path = btrfs_alloc_path();
739         if (!path)
740                 return -ENOMEM;
741 
742         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
743         if (ret > 0)
744                 ret = -ENOENT;
745 
746         if (ret)
747                 goto out;
748 
749         l = path->nodes[0];
750         slot = path->slots[0];
751         qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
752         btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
753         btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
754         btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
755         btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
756         btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
757 
758         btrfs_mark_buffer_dirty(l);
759 
760 out:
761         btrfs_free_path(path);
762         return ret;
763 }
764 
765 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
766                                      struct btrfs_fs_info *fs_info,
767                                     struct btrfs_root *root)
768 {
769         struct btrfs_path *path;
770         struct btrfs_key key;
771         struct extent_buffer *l;
772         struct btrfs_qgroup_status_item *ptr;
773         int ret;
774         int slot;
775 
776         key.objectid = 0;
777         key.type = BTRFS_QGROUP_STATUS_KEY;
778         key.offset = 0;
779 
780         path = btrfs_alloc_path();
781         if (!path)
782                 return -ENOMEM;
783 
784         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
785         if (ret > 0)
786                 ret = -ENOENT;
787 
788         if (ret)
789                 goto out;
790 
791         l = path->nodes[0];
792         slot = path->slots[0];
793         ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
794         btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
795         btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
796         btrfs_set_qgroup_status_rescan(l, ptr,
797                                 fs_info->qgroup_rescan_progress.objectid);
798 
799         btrfs_mark_buffer_dirty(l);
800 
801 out:
802         btrfs_free_path(path);
803         return ret;
804 }
805 
806 /*
807  * called with qgroup_lock held
808  */
809 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
810                                   struct btrfs_root *root)
811 {
812         struct btrfs_path *path;
813         struct btrfs_key key;
814         struct extent_buffer *leaf = NULL;
815         int ret;
816         int nr = 0;
817 
818         path = btrfs_alloc_path();
819         if (!path)
820                 return -ENOMEM;
821 
822         path->leave_spinning = 1;
823 
824         key.objectid = 0;
825         key.offset = 0;
826         key.type = 0;
827 
828         while (1) {
829                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
830                 if (ret < 0)
831                         goto out;
832                 leaf = path->nodes[0];
833                 nr = btrfs_header_nritems(leaf);
834                 if (!nr)
835                         break;
836                 /*
837                  * delete the leaf one by one
838                  * since the whole tree is going
839                  * to be deleted.
840                  */
841                 path->slots[0] = 0;
842                 ret = btrfs_del_items(trans, root, path, 0, nr);
843                 if (ret)
844                         goto out;
845 
846                 btrfs_release_path(path);
847         }
848         ret = 0;
849 out:
850         root->fs_info->pending_quota_state = 0;
851         btrfs_free_path(path);
852         return ret;
853 }
854 
855 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
856                        struct btrfs_fs_info *fs_info)
857 {
858         struct btrfs_root *quota_root;
859         struct btrfs_root *tree_root = fs_info->tree_root;
860         struct btrfs_path *path = NULL;
861         struct btrfs_qgroup_status_item *ptr;
862         struct extent_buffer *leaf;
863         struct btrfs_key key;
864         struct btrfs_key found_key;
865         struct btrfs_qgroup *qgroup = NULL;
866         int ret = 0;
867         int slot;
868 
869         mutex_lock(&fs_info->qgroup_ioctl_lock);
870         if (fs_info->quota_root) {
871                 fs_info->pending_quota_state = 1;
872                 goto out;
873         }
874 
875         fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
876         if (!fs_info->qgroup_ulist) {
877                 ret = -ENOMEM;
878                 goto out;
879         }
880 
881         /*
882          * initially create the quota tree
883          */
884         quota_root = btrfs_create_tree(trans, fs_info,
885                                        BTRFS_QUOTA_TREE_OBJECTID);
886         if (IS_ERR(quota_root)) {
887                 ret =  PTR_ERR(quota_root);
888                 goto out;
889         }
890 
891         path = btrfs_alloc_path();
892         if (!path) {
893                 ret = -ENOMEM;
894                 goto out_free_root;
895         }
896 
897         key.objectid = 0;
898         key.type = BTRFS_QGROUP_STATUS_KEY;
899         key.offset = 0;
900 
901         ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
902                                       sizeof(*ptr));
903         if (ret)
904                 goto out_free_path;
905 
906         leaf = path->nodes[0];
907         ptr = btrfs_item_ptr(leaf, path->slots[0],
908                                  struct btrfs_qgroup_status_item);
909         btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
910         btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
911         fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
912                                 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
913         btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
914         btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
915 
916         btrfs_mark_buffer_dirty(leaf);
917 
918         key.objectid = 0;
919         key.type = BTRFS_ROOT_REF_KEY;
920         key.offset = 0;
921 
922         btrfs_release_path(path);
923         ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
924         if (ret > 0)
925                 goto out_add_root;
926         if (ret < 0)
927                 goto out_free_path;
928 
929 
930         while (1) {
931                 slot = path->slots[0];
932                 leaf = path->nodes[0];
933                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
934 
935                 if (found_key.type == BTRFS_ROOT_REF_KEY) {
936                         ret = add_qgroup_item(trans, quota_root,
937                                               found_key.offset);
938                         if (ret)
939                                 goto out_free_path;
940 
941                         qgroup = add_qgroup_rb(fs_info, found_key.offset);
942                         if (IS_ERR(qgroup)) {
943                                 ret = PTR_ERR(qgroup);
944                                 goto out_free_path;
945                         }
946                 }
947                 ret = btrfs_next_item(tree_root, path);
948                 if (ret < 0)
949                         goto out_free_path;
950                 if (ret)
951                         break;
952         }
953 
954 out_add_root:
955         btrfs_release_path(path);
956         ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
957         if (ret)
958                 goto out_free_path;
959 
960         qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
961         if (IS_ERR(qgroup)) {
962                 ret = PTR_ERR(qgroup);
963                 goto out_free_path;
964         }
965         spin_lock(&fs_info->qgroup_lock);
966         fs_info->quota_root = quota_root;
967         fs_info->pending_quota_state = 1;
968         spin_unlock(&fs_info->qgroup_lock);
969 out_free_path:
970         btrfs_free_path(path);
971 out_free_root:
972         if (ret) {
973                 free_extent_buffer(quota_root->node);
974                 free_extent_buffer(quota_root->commit_root);
975                 kfree(quota_root);
976         }
977 out:
978         if (ret) {
979                 ulist_free(fs_info->qgroup_ulist);
980                 fs_info->qgroup_ulist = NULL;
981         }
982         mutex_unlock(&fs_info->qgroup_ioctl_lock);
983         return ret;
984 }
985 
986 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
987                         struct btrfs_fs_info *fs_info)
988 {
989         struct btrfs_root *tree_root = fs_info->tree_root;
990         struct btrfs_root *quota_root;
991         int ret = 0;
992 
993         mutex_lock(&fs_info->qgroup_ioctl_lock);
994         if (!fs_info->quota_root)
995                 goto out;
996         fs_info->quota_enabled = 0;
997         fs_info->pending_quota_state = 0;
998         btrfs_qgroup_wait_for_completion(fs_info, false);
999         spin_lock(&fs_info->qgroup_lock);
1000         quota_root = fs_info->quota_root;
1001         fs_info->quota_root = NULL;
1002         fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1003         spin_unlock(&fs_info->qgroup_lock);
1004 
1005         btrfs_free_qgroup_config(fs_info);
1006 
1007         ret = btrfs_clean_quota_tree(trans, quota_root);
1008         if (ret)
1009                 goto out;
1010 
1011         ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
1012         if (ret)
1013                 goto out;
1014 
1015         list_del(&quota_root->dirty_list);
1016 
1017         btrfs_tree_lock(quota_root->node);
1018         clean_tree_block(trans, tree_root->fs_info, quota_root->node);
1019         btrfs_tree_unlock(quota_root->node);
1020         btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1021 
1022         free_extent_buffer(quota_root->node);
1023         free_extent_buffer(quota_root->commit_root);
1024         kfree(quota_root);
1025 out:
1026         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1027         return ret;
1028 }
1029 
1030 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1031                          struct btrfs_qgroup *qgroup)
1032 {
1033         if (list_empty(&qgroup->dirty))
1034                 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1035 }
1036 
1037 /*
1038  * The easy accounting, if we are adding/removing the only ref for an extent
1039  * then this qgroup and all of the parent qgroups get their reference and
1040  * exclusive counts adjusted.
1041  *
1042  * Caller should hold fs_info->qgroup_lock.
1043  */
1044 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1045                                     struct ulist *tmp, u64 ref_root,
1046                                     u64 num_bytes, int sign)
1047 {
1048         struct btrfs_qgroup *qgroup;
1049         struct btrfs_qgroup_list *glist;
1050         struct ulist_node *unode;
1051         struct ulist_iterator uiter;
1052         int ret = 0;
1053 
1054         qgroup = find_qgroup_rb(fs_info, ref_root);
1055         if (!qgroup)
1056                 goto out;
1057 
1058         qgroup->rfer += sign * num_bytes;
1059         qgroup->rfer_cmpr += sign * num_bytes;
1060 
1061         WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1062         qgroup->excl += sign * num_bytes;
1063         qgroup->excl_cmpr += sign * num_bytes;
1064         if (sign > 0)
1065                 qgroup->reserved -= num_bytes;
1066 
1067         qgroup_dirty(fs_info, qgroup);
1068 
1069         /* Get all of the parent groups that contain this qgroup */
1070         list_for_each_entry(glist, &qgroup->groups, next_group) {
1071                 ret = ulist_add(tmp, glist->group->qgroupid,
1072                                 ptr_to_u64(glist->group), GFP_ATOMIC);
1073                 if (ret < 0)
1074                         goto out;
1075         }
1076 
1077         /* Iterate all of the parents and adjust their reference counts */
1078         ULIST_ITER_INIT(&uiter);
1079         while ((unode = ulist_next(tmp, &uiter))) {
1080                 qgroup = u64_to_ptr(unode->aux);
1081                 qgroup->rfer += sign * num_bytes;
1082                 qgroup->rfer_cmpr += sign * num_bytes;
1083                 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1084                 qgroup->excl += sign * num_bytes;
1085                 if (sign > 0)
1086                         qgroup->reserved -= num_bytes;
1087                 qgroup->excl_cmpr += sign * num_bytes;
1088                 qgroup_dirty(fs_info, qgroup);
1089 
1090                 /* Add any parents of the parents */
1091                 list_for_each_entry(glist, &qgroup->groups, next_group) {
1092                         ret = ulist_add(tmp, glist->group->qgroupid,
1093                                         ptr_to_u64(glist->group), GFP_ATOMIC);
1094                         if (ret < 0)
1095                                 goto out;
1096                 }
1097         }
1098         ret = 0;
1099 out:
1100         return ret;
1101 }
1102 
1103 
1104 /*
1105  * Quick path for updating qgroup with only excl refs.
1106  *
1107  * In that case, just update all parent will be enough.
1108  * Or we needs to do a full rescan.
1109  * Caller should also hold fs_info->qgroup_lock.
1110  *
1111  * Return 0 for quick update, return >0 for need to full rescan
1112  * and mark INCONSISTENT flag.
1113  * Return < 0 for other error.
1114  */
1115 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1116                                    struct ulist *tmp, u64 src, u64 dst,
1117                                    int sign)
1118 {
1119         struct btrfs_qgroup *qgroup;
1120         int ret = 1;
1121         int err = 0;
1122 
1123         qgroup = find_qgroup_rb(fs_info, src);
1124         if (!qgroup)
1125                 goto out;
1126         if (qgroup->excl == qgroup->rfer) {
1127                 ret = 0;
1128                 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1129                                                qgroup->excl, sign);
1130                 if (err < 0) {
1131                         ret = err;
1132                         goto out;
1133                 }
1134         }
1135 out:
1136         if (ret)
1137                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1138         return ret;
1139 }
1140 
1141 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1142                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1143 {
1144         struct btrfs_root *quota_root;
1145         struct btrfs_qgroup *parent;
1146         struct btrfs_qgroup *member;
1147         struct btrfs_qgroup_list *list;
1148         struct ulist *tmp;
1149         int ret = 0;
1150 
1151         /* Check the level of src and dst first */
1152         if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1153                 return -EINVAL;
1154 
1155         tmp = ulist_alloc(GFP_NOFS);
1156         if (!tmp)
1157                 return -ENOMEM;
1158 
1159         mutex_lock(&fs_info->qgroup_ioctl_lock);
1160         quota_root = fs_info->quota_root;
1161         if (!quota_root) {
1162                 ret = -EINVAL;
1163                 goto out;
1164         }
1165         member = find_qgroup_rb(fs_info, src);
1166         parent = find_qgroup_rb(fs_info, dst);
1167         if (!member || !parent) {
1168                 ret = -EINVAL;
1169                 goto out;
1170         }
1171 
1172         /* check if such qgroup relation exist firstly */
1173         list_for_each_entry(list, &member->groups, next_group) {
1174                 if (list->group == parent) {
1175                         ret = -EEXIST;
1176                         goto out;
1177                 }
1178         }
1179 
1180         ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1181         if (ret)
1182                 goto out;
1183 
1184         ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1185         if (ret) {
1186                 del_qgroup_relation_item(trans, quota_root, src, dst);
1187                 goto out;
1188         }
1189 
1190         spin_lock(&fs_info->qgroup_lock);
1191         ret = add_relation_rb(quota_root->fs_info, src, dst);
1192         if (ret < 0) {
1193                 spin_unlock(&fs_info->qgroup_lock);
1194                 goto out;
1195         }
1196         ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1197         spin_unlock(&fs_info->qgroup_lock);
1198 out:
1199         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1200         ulist_free(tmp);
1201         return ret;
1202 }
1203 
1204 int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1205                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1206 {
1207         struct btrfs_root *quota_root;
1208         struct btrfs_qgroup *parent;
1209         struct btrfs_qgroup *member;
1210         struct btrfs_qgroup_list *list;
1211         struct ulist *tmp;
1212         int ret = 0;
1213         int err;
1214 
1215         tmp = ulist_alloc(GFP_NOFS);
1216         if (!tmp)
1217                 return -ENOMEM;
1218 
1219         quota_root = fs_info->quota_root;
1220         if (!quota_root) {
1221                 ret = -EINVAL;
1222                 goto out;
1223         }
1224 
1225         member = find_qgroup_rb(fs_info, src);
1226         parent = find_qgroup_rb(fs_info, dst);
1227         if (!member || !parent) {
1228                 ret = -EINVAL;
1229                 goto out;
1230         }
1231 
1232         /* check if such qgroup relation exist firstly */
1233         list_for_each_entry(list, &member->groups, next_group) {
1234                 if (list->group == parent)
1235                         goto exist;
1236         }
1237         ret = -ENOENT;
1238         goto out;
1239 exist:
1240         ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1241         err = del_qgroup_relation_item(trans, quota_root, dst, src);
1242         if (err && !ret)
1243                 ret = err;
1244 
1245         spin_lock(&fs_info->qgroup_lock);
1246         del_relation_rb(fs_info, src, dst);
1247         ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1248         spin_unlock(&fs_info->qgroup_lock);
1249 out:
1250         ulist_free(tmp);
1251         return ret;
1252 }
1253 
1254 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1255                               struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1256 {
1257         int ret = 0;
1258 
1259         mutex_lock(&fs_info->qgroup_ioctl_lock);
1260         ret = __del_qgroup_relation(trans, fs_info, src, dst);
1261         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1262 
1263         return ret;
1264 }
1265 
1266 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1267                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1268 {
1269         struct btrfs_root *quota_root;
1270         struct btrfs_qgroup *qgroup;
1271         int ret = 0;
1272 
1273         mutex_lock(&fs_info->qgroup_ioctl_lock);
1274         quota_root = fs_info->quota_root;
1275         if (!quota_root) {
1276                 ret = -EINVAL;
1277                 goto out;
1278         }
1279         qgroup = find_qgroup_rb(fs_info, qgroupid);
1280         if (qgroup) {
1281                 ret = -EEXIST;
1282                 goto out;
1283         }
1284 
1285         ret = add_qgroup_item(trans, quota_root, qgroupid);
1286         if (ret)
1287                 goto out;
1288 
1289         spin_lock(&fs_info->qgroup_lock);
1290         qgroup = add_qgroup_rb(fs_info, qgroupid);
1291         spin_unlock(&fs_info->qgroup_lock);
1292 
1293         if (IS_ERR(qgroup))
1294                 ret = PTR_ERR(qgroup);
1295 out:
1296         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1297         return ret;
1298 }
1299 
1300 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1301                         struct btrfs_fs_info *fs_info, u64 qgroupid)
1302 {
1303         struct btrfs_root *quota_root;
1304         struct btrfs_qgroup *qgroup;
1305         struct btrfs_qgroup_list *list;
1306         int ret = 0;
1307 
1308         mutex_lock(&fs_info->qgroup_ioctl_lock);
1309         quota_root = fs_info->quota_root;
1310         if (!quota_root) {
1311                 ret = -EINVAL;
1312                 goto out;
1313         }
1314 
1315         qgroup = find_qgroup_rb(fs_info, qgroupid);
1316         if (!qgroup) {
1317                 ret = -ENOENT;
1318                 goto out;
1319         } else {
1320                 /* check if there are no children of this qgroup */
1321                 if (!list_empty(&qgroup->members)) {
1322                         ret = -EBUSY;
1323                         goto out;
1324                 }
1325         }
1326         ret = del_qgroup_item(trans, quota_root, qgroupid);
1327 
1328         while (!list_empty(&qgroup->groups)) {
1329                 list = list_first_entry(&qgroup->groups,
1330                                         struct btrfs_qgroup_list, next_group);
1331                 ret = __del_qgroup_relation(trans, fs_info,
1332                                            qgroupid,
1333                                            list->group->qgroupid);
1334                 if (ret)
1335                         goto out;
1336         }
1337 
1338         spin_lock(&fs_info->qgroup_lock);
1339         del_qgroup_rb(quota_root->fs_info, qgroupid);
1340         spin_unlock(&fs_info->qgroup_lock);
1341 out:
1342         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1343         return ret;
1344 }
1345 
1346 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1347                        struct btrfs_fs_info *fs_info, u64 qgroupid,
1348                        struct btrfs_qgroup_limit *limit)
1349 {
1350         struct btrfs_root *quota_root;
1351         struct btrfs_qgroup *qgroup;
1352         int ret = 0;
1353         /* Sometimes we would want to clear the limit on this qgroup.
1354          * To meet this requirement, we treat the -1 as a special value
1355          * which tell kernel to clear the limit on this qgroup.
1356          */
1357         const u64 CLEAR_VALUE = -1;
1358 
1359         mutex_lock(&fs_info->qgroup_ioctl_lock);
1360         quota_root = fs_info->quota_root;
1361         if (!quota_root) {
1362                 ret = -EINVAL;
1363                 goto out;
1364         }
1365 
1366         qgroup = find_qgroup_rb(fs_info, qgroupid);
1367         if (!qgroup) {
1368                 ret = -ENOENT;
1369                 goto out;
1370         }
1371 
1372         spin_lock(&fs_info->qgroup_lock);
1373         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1374                 if (limit->max_rfer == CLEAR_VALUE) {
1375                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1376                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1377                         qgroup->max_rfer = 0;
1378                 } else {
1379                         qgroup->max_rfer = limit->max_rfer;
1380                 }
1381         }
1382         if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1383                 if (limit->max_excl == CLEAR_VALUE) {
1384                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1385                         limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1386                         qgroup->max_excl = 0;
1387                 } else {
1388                         qgroup->max_excl = limit->max_excl;
1389                 }
1390         }
1391         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1392                 if (limit->rsv_rfer == CLEAR_VALUE) {
1393                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1394                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1395                         qgroup->rsv_rfer = 0;
1396                 } else {
1397                         qgroup->rsv_rfer = limit->rsv_rfer;
1398                 }
1399         }
1400         if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1401                 if (limit->rsv_excl == CLEAR_VALUE) {
1402                         qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1403                         limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1404                         qgroup->rsv_excl = 0;
1405                 } else {
1406                         qgroup->rsv_excl = limit->rsv_excl;
1407                 }
1408         }
1409         qgroup->lim_flags |= limit->flags;
1410 
1411         spin_unlock(&fs_info->qgroup_lock);
1412 
1413         ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1414         if (ret) {
1415                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1416                 btrfs_info(fs_info, "unable to update quota limit for %llu",
1417                        qgroupid);
1418         }
1419 
1420 out:
1421         mutex_unlock(&fs_info->qgroup_ioctl_lock);
1422         return ret;
1423 }
1424 
1425 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1426                                          struct btrfs_fs_info *fs_info)
1427 {
1428         struct btrfs_qgroup_extent_record *record;
1429         struct btrfs_delayed_ref_root *delayed_refs;
1430         struct rb_node *node;
1431         u64 qgroup_to_skip;
1432         int ret = 0;
1433 
1434         delayed_refs = &trans->transaction->delayed_refs;
1435         qgroup_to_skip = delayed_refs->qgroup_to_skip;
1436 
1437         /*
1438          * No need to do lock, since this function will only be called in
1439          * btrfs_commit_transaction().
1440          */
1441         node = rb_first(&delayed_refs->dirty_extent_root);
1442         while (node) {
1443                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1444                                   node);
1445                 ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0,
1446                                            &record->old_roots);
1447                 if (ret < 0)
1448                         break;
1449                 if (qgroup_to_skip)
1450                         ulist_del(record->old_roots, qgroup_to_skip, 0);
1451                 node = rb_next(node);
1452         }
1453         return ret;
1454 }
1455 
1456 int btrfs_qgroup_insert_dirty_extent_nolock(struct btrfs_fs_info *fs_info,
1457                                 struct btrfs_delayed_ref_root *delayed_refs,
1458                                 struct btrfs_qgroup_extent_record *record)
1459 {
1460         struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1461         struct rb_node *parent_node = NULL;
1462         struct btrfs_qgroup_extent_record *entry;
1463         u64 bytenr = record->bytenr;
1464 
1465         assert_spin_locked(&delayed_refs->lock);
1466         trace_btrfs_qgroup_insert_dirty_extent(fs_info, record);
1467 
1468         while (*p) {
1469                 parent_node = *p;
1470                 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1471                                  node);
1472                 if (bytenr < entry->bytenr)
1473                         p = &(*p)->rb_left;
1474                 else if (bytenr > entry->bytenr)
1475                         p = &(*p)->rb_right;
1476                 else
1477                         return 1;
1478         }
1479 
1480         rb_link_node(&record->node, parent_node, p);
1481         rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1482         return 0;
1483 }
1484 
1485 int btrfs_qgroup_insert_dirty_extent(struct btrfs_trans_handle *trans,
1486                 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1487                 gfp_t gfp_flag)
1488 {
1489         struct btrfs_qgroup_extent_record *record;
1490         struct btrfs_delayed_ref_root *delayed_refs;
1491         int ret;
1492 
1493         if (!fs_info->quota_enabled || bytenr == 0 || num_bytes == 0)
1494                 return 0;
1495         if (WARN_ON(trans == NULL))
1496                 return -EINVAL;
1497         record = kmalloc(sizeof(*record), gfp_flag);
1498         if (!record)
1499                 return -ENOMEM;
1500 
1501         delayed_refs = &trans->transaction->delayed_refs;
1502         record->bytenr = bytenr;
1503         record->num_bytes = num_bytes;
1504         record->old_roots = NULL;
1505 
1506         spin_lock(&delayed_refs->lock);
1507         ret = btrfs_qgroup_insert_dirty_extent_nolock(fs_info, delayed_refs,
1508                                                       record);
1509         spin_unlock(&delayed_refs->lock);
1510         if (ret > 0)
1511                 kfree(record);
1512         return 0;
1513 }
1514 
1515 #define UPDATE_NEW      0
1516 #define UPDATE_OLD      1
1517 /*
1518  * Walk all of the roots that points to the bytenr and adjust their refcnts.
1519  */
1520 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1521                                 struct ulist *roots, struct ulist *tmp,
1522                                 struct ulist *qgroups, u64 seq, int update_old)
1523 {
1524         struct ulist_node *unode;
1525         struct ulist_iterator uiter;
1526         struct ulist_node *tmp_unode;
1527         struct ulist_iterator tmp_uiter;
1528         struct btrfs_qgroup *qg;
1529         int ret = 0;
1530 
1531         if (!roots)
1532                 return 0;
1533         ULIST_ITER_INIT(&uiter);
1534         while ((unode = ulist_next(roots, &uiter))) {
1535                 qg = find_qgroup_rb(fs_info, unode->val);
1536                 if (!qg)
1537                         continue;
1538 
1539                 ulist_reinit(tmp);
1540                 ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg),
1541                                 GFP_ATOMIC);
1542                 if (ret < 0)
1543                         return ret;
1544                 ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg), GFP_ATOMIC);
1545                 if (ret < 0)
1546                         return ret;
1547                 ULIST_ITER_INIT(&tmp_uiter);
1548                 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1549                         struct btrfs_qgroup_list *glist;
1550 
1551                         qg = u64_to_ptr(tmp_unode->aux);
1552                         if (update_old)
1553                                 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1554                         else
1555                                 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1556                         list_for_each_entry(glist, &qg->groups, next_group) {
1557                                 ret = ulist_add(qgroups, glist->group->qgroupid,
1558                                                 ptr_to_u64(glist->group),
1559                                                 GFP_ATOMIC);
1560                                 if (ret < 0)
1561                                         return ret;
1562                                 ret = ulist_add(tmp, glist->group->qgroupid,
1563                                                 ptr_to_u64(glist->group),
1564                                                 GFP_ATOMIC);
1565                                 if (ret < 0)
1566                                         return ret;
1567                         }
1568                 }
1569         }
1570         return 0;
1571 }
1572 
1573 /*
1574  * Update qgroup rfer/excl counters.
1575  * Rfer update is easy, codes can explain themselves.
1576  *
1577  * Excl update is tricky, the update is split into 2 part.
1578  * Part 1: Possible exclusive <-> sharing detect:
1579  *      |       A       |       !A      |
1580  *  -------------------------------------
1581  *  B   |       *       |       -       |
1582  *  -------------------------------------
1583  *  !B  |       +       |       **      |
1584  *  -------------------------------------
1585  *
1586  * Conditions:
1587  * A:   cur_old_roots < nr_old_roots    (not exclusive before)
1588  * !A:  cur_old_roots == nr_old_roots   (possible exclusive before)
1589  * B:   cur_new_roots < nr_new_roots    (not exclusive now)
1590  * !B:  cur_new_roots == nr_new_roots   (possible exclusive now)
1591  *
1592  * Results:
1593  * +: Possible sharing -> exclusive     -: Possible exclusive -> sharing
1594  * *: Definitely not changed.           **: Possible unchanged.
1595  *
1596  * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1597  *
1598  * To make the logic clear, we first use condition A and B to split
1599  * combination into 4 results.
1600  *
1601  * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1602  * only on variant maybe 0.
1603  *
1604  * Lastly, check result **, since there are 2 variants maybe 0, split them
1605  * again(2x2).
1606  * But this time we don't need to consider other things, the codes and logic
1607  * is easy to understand now.
1608  */
1609 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1610                                   struct ulist *qgroups,
1611                                   u64 nr_old_roots,
1612                                   u64 nr_new_roots,
1613                                   u64 num_bytes, u64 seq)
1614 {
1615         struct ulist_node *unode;
1616         struct ulist_iterator uiter;
1617         struct btrfs_qgroup *qg;
1618         u64 cur_new_count, cur_old_count;
1619 
1620         ULIST_ITER_INIT(&uiter);
1621         while ((unode = ulist_next(qgroups, &uiter))) {
1622                 bool dirty = false;
1623 
1624                 qg = u64_to_ptr(unode->aux);
1625                 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1626                 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1627 
1628                 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1629                                              cur_old_count, cur_new_count);
1630 
1631                 /* Rfer update part */
1632                 if (cur_old_count == 0 && cur_new_count > 0) {
1633                         qg->rfer += num_bytes;
1634                         qg->rfer_cmpr += num_bytes;
1635                         dirty = true;
1636                 }
1637                 if (cur_old_count > 0 && cur_new_count == 0) {
1638                         qg->rfer -= num_bytes;
1639                         qg->rfer_cmpr -= num_bytes;
1640                         dirty = true;
1641                 }
1642 
1643                 /* Excl update part */
1644                 /* Exclusive/none -> shared case */
1645                 if (cur_old_count == nr_old_roots &&
1646                     cur_new_count < nr_new_roots) {
1647                         /* Exclusive -> shared */
1648                         if (cur_old_count != 0) {
1649                                 qg->excl -= num_bytes;
1650                                 qg->excl_cmpr -= num_bytes;
1651                                 dirty = true;
1652                         }
1653                 }
1654 
1655                 /* Shared -> exclusive/none case */
1656                 if (cur_old_count < nr_old_roots &&
1657                     cur_new_count == nr_new_roots) {
1658                         /* Shared->exclusive */
1659                         if (cur_new_count != 0) {
1660                                 qg->excl += num_bytes;
1661                                 qg->excl_cmpr += num_bytes;
1662                                 dirty = true;
1663                         }
1664                 }
1665 
1666                 /* Exclusive/none -> exclusive/none case */
1667                 if (cur_old_count == nr_old_roots &&
1668                     cur_new_count == nr_new_roots) {
1669                         if (cur_old_count == 0) {
1670                                 /* None -> exclusive/none */
1671 
1672                                 if (cur_new_count != 0) {
1673                                         /* None -> exclusive */
1674                                         qg->excl += num_bytes;
1675                                         qg->excl_cmpr += num_bytes;
1676                                         dirty = true;
1677                                 }
1678                                 /* None -> none, nothing changed */
1679                         } else {
1680                                 /* Exclusive -> exclusive/none */
1681 
1682                                 if (cur_new_count == 0) {
1683                                         /* Exclusive -> none */
1684                                         qg->excl -= num_bytes;
1685                                         qg->excl_cmpr -= num_bytes;
1686                                         dirty = true;
1687                                 }
1688                                 /* Exclusive -> exclusive, nothing changed */
1689                         }
1690                 }
1691 
1692                 if (dirty)
1693                         qgroup_dirty(fs_info, qg);
1694         }
1695         return 0;
1696 }
1697 
1698 int
1699 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1700                             struct btrfs_fs_info *fs_info,
1701                             u64 bytenr, u64 num_bytes,
1702                             struct ulist *old_roots, struct ulist *new_roots)
1703 {
1704         struct ulist *qgroups = NULL;
1705         struct ulist *tmp = NULL;
1706         u64 seq;
1707         u64 nr_new_roots = 0;
1708         u64 nr_old_roots = 0;
1709         int ret = 0;
1710 
1711         if (new_roots)
1712                 nr_new_roots = new_roots->nnodes;
1713         if (old_roots)
1714                 nr_old_roots = old_roots->nnodes;
1715 
1716         if (!fs_info->quota_enabled)
1717                 goto out_free;
1718         BUG_ON(!fs_info->quota_root);
1719 
1720         trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1721                                           nr_old_roots, nr_new_roots);
1722 
1723         qgroups = ulist_alloc(GFP_NOFS);
1724         if (!qgroups) {
1725                 ret = -ENOMEM;
1726                 goto out_free;
1727         }
1728         tmp = ulist_alloc(GFP_NOFS);
1729         if (!tmp) {
1730                 ret = -ENOMEM;
1731                 goto out_free;
1732         }
1733 
1734         mutex_lock(&fs_info->qgroup_rescan_lock);
1735         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1736                 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1737                         mutex_unlock(&fs_info->qgroup_rescan_lock);
1738                         ret = 0;
1739                         goto out_free;
1740                 }
1741         }
1742         mutex_unlock(&fs_info->qgroup_rescan_lock);
1743 
1744         spin_lock(&fs_info->qgroup_lock);
1745         seq = fs_info->qgroup_seq;
1746 
1747         /* Update old refcnts using old_roots */
1748         ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1749                                    UPDATE_OLD);
1750         if (ret < 0)
1751                 goto out;
1752 
1753         /* Update new refcnts using new_roots */
1754         ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1755                                    UPDATE_NEW);
1756         if (ret < 0)
1757                 goto out;
1758 
1759         qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1760                                num_bytes, seq);
1761 
1762         /*
1763          * Bump qgroup_seq to avoid seq overlap
1764          */
1765         fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1766 out:
1767         spin_unlock(&fs_info->qgroup_lock);
1768 out_free:
1769         ulist_free(tmp);
1770         ulist_free(qgroups);
1771         ulist_free(old_roots);
1772         ulist_free(new_roots);
1773         return ret;
1774 }
1775 
1776 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1777                                  struct btrfs_fs_info *fs_info)
1778 {
1779         struct btrfs_qgroup_extent_record *record;
1780         struct btrfs_delayed_ref_root *delayed_refs;
1781         struct ulist *new_roots = NULL;
1782         struct rb_node *node;
1783         u64 qgroup_to_skip;
1784         int ret = 0;
1785 
1786         delayed_refs = &trans->transaction->delayed_refs;
1787         qgroup_to_skip = delayed_refs->qgroup_to_skip;
1788         while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
1789                 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1790                                   node);
1791 
1792                 trace_btrfs_qgroup_account_extents(fs_info, record);
1793 
1794                 if (!ret) {
1795                         /*
1796                          * Use (u64)-1 as time_seq to do special search, which
1797                          * doesn't lock tree or delayed_refs and search current
1798                          * root. It's safe inside commit_transaction().
1799                          */
1800                         ret = btrfs_find_all_roots(trans, fs_info,
1801                                         record->bytenr, (u64)-1, &new_roots);
1802                         if (ret < 0)
1803                                 goto cleanup;
1804                         if (qgroup_to_skip)
1805                                 ulist_del(new_roots, qgroup_to_skip, 0);
1806                         ret = btrfs_qgroup_account_extent(trans, fs_info,
1807                                         record->bytenr, record->num_bytes,
1808                                         record->old_roots, new_roots);
1809                         record->old_roots = NULL;
1810                         new_roots = NULL;
1811                 }
1812 cleanup:
1813                 ulist_free(record->old_roots);
1814                 ulist_free(new_roots);
1815                 new_roots = NULL;
1816                 rb_erase(node, &delayed_refs->dirty_extent_root);
1817                 kfree(record);
1818 
1819         }
1820         return ret;
1821 }
1822 
1823 /*
1824  * called from commit_transaction. Writes all changed qgroups to disk.
1825  */
1826 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1827                       struct btrfs_fs_info *fs_info)
1828 {
1829         struct btrfs_root *quota_root = fs_info->quota_root;
1830         int ret = 0;
1831         int start_rescan_worker = 0;
1832 
1833         if (!quota_root)
1834                 goto out;
1835 
1836         if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1837                 start_rescan_worker = 1;
1838 
1839         fs_info->quota_enabled = fs_info->pending_quota_state;
1840 
1841         spin_lock(&fs_info->qgroup_lock);
1842         while (!list_empty(&fs_info->dirty_qgroups)) {
1843                 struct btrfs_qgroup *qgroup;
1844                 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1845                                           struct btrfs_qgroup, dirty);
1846                 list_del_init(&qgroup->dirty);
1847                 spin_unlock(&fs_info->qgroup_lock);
1848                 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1849                 if (ret)
1850                         fs_info->qgroup_flags |=
1851                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1852                 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1853                 if (ret)
1854                         fs_info->qgroup_flags |=
1855                                         BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1856                 spin_lock(&fs_info->qgroup_lock);
1857         }
1858         if (fs_info->quota_enabled)
1859                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1860         else
1861                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1862         spin_unlock(&fs_info->qgroup_lock);
1863 
1864         ret = update_qgroup_status_item(trans, fs_info, quota_root);
1865         if (ret)
1866                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1867 
1868         if (!ret && start_rescan_worker) {
1869                 ret = qgroup_rescan_init(fs_info, 0, 1);
1870                 if (!ret) {
1871                         qgroup_rescan_zero_tracking(fs_info);
1872                         btrfs_queue_work(fs_info->qgroup_rescan_workers,
1873                                          &fs_info->qgroup_rescan_work);
1874                 }
1875                 ret = 0;
1876         }
1877 
1878 out:
1879 
1880         return ret;
1881 }
1882 
1883 /*
1884  * Copy the accounting information between qgroups. This is necessary
1885  * when a snapshot or a subvolume is created. Throwing an error will
1886  * cause a transaction abort so we take extra care here to only error
1887  * when a readonly fs is a reasonable outcome.
1888  */
1889 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1890                          struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1891                          struct btrfs_qgroup_inherit *inherit)
1892 {
1893         int ret = 0;
1894         int i;
1895         u64 *i_qgroups;
1896         struct btrfs_root *quota_root = fs_info->quota_root;
1897         struct btrfs_qgroup *srcgroup;
1898         struct btrfs_qgroup *dstgroup;
1899         u32 level_size = 0;
1900         u64 nums;
1901 
1902         mutex_lock(&fs_info->qgroup_ioctl_lock);
1903         if (!fs_info->quota_enabled)
1904                 goto out;
1905 
1906         if (!quota_root) {
1907                 ret = -EINVAL;
1908                 goto out;
1909         }
1910 
1911         if (inherit) {
1912                 i_qgroups = (u64 *)(inherit + 1);
1913                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1914                        2 * inherit->num_excl_copies;
1915                 for (i = 0; i < nums; ++i) {
1916                         srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1917 
1918                         /*
1919                          * Zero out invalid groups so we can ignore
1920                          * them later.
1921                          */
1922                         if (!srcgroup ||
1923                             ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
1924                                 *i_qgroups = 0ULL;
1925 
1926                         ++i_qgroups;
1927                 }
1928         }
1929 
1930         /*
1931          * create a tracking group for the subvol itself
1932          */
1933         ret = add_qgroup_item(trans, quota_root, objectid);
1934         if (ret)
1935                 goto out;
1936 
1937         if (srcid) {
1938                 struct btrfs_root *srcroot;
1939                 struct btrfs_key srckey;
1940 
1941                 srckey.objectid = srcid;
1942                 srckey.type = BTRFS_ROOT_ITEM_KEY;
1943                 srckey.offset = (u64)-1;
1944                 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1945                 if (IS_ERR(srcroot)) {
1946                         ret = PTR_ERR(srcroot);
1947                         goto out;
1948                 }
1949 
1950                 rcu_read_lock();
1951                 level_size = srcroot->nodesize;
1952                 rcu_read_unlock();
1953         }
1954 
1955         /*
1956          * add qgroup to all inherited groups
1957          */
1958         if (inherit) {
1959                 i_qgroups = (u64 *)(inherit + 1);
1960                 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
1961                         if (*i_qgroups == 0)
1962                                 continue;
1963                         ret = add_qgroup_relation_item(trans, quota_root,
1964                                                        objectid, *i_qgroups);
1965                         if (ret && ret != -EEXIST)
1966                                 goto out;
1967                         ret = add_qgroup_relation_item(trans, quota_root,
1968                                                        *i_qgroups, objectid);
1969                         if (ret && ret != -EEXIST)
1970                                 goto out;
1971                 }
1972                 ret = 0;
1973         }
1974 
1975 
1976         spin_lock(&fs_info->qgroup_lock);
1977 
1978         dstgroup = add_qgroup_rb(fs_info, objectid);
1979         if (IS_ERR(dstgroup)) {
1980                 ret = PTR_ERR(dstgroup);
1981                 goto unlock;
1982         }
1983 
1984         if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1985                 dstgroup->lim_flags = inherit->lim.flags;
1986                 dstgroup->max_rfer = inherit->lim.max_rfer;
1987                 dstgroup->max_excl = inherit->lim.max_excl;
1988                 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
1989                 dstgroup->rsv_excl = inherit->lim.rsv_excl;
1990 
1991                 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
1992                 if (ret) {
1993                         fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1994                         btrfs_info(fs_info, "unable to update quota limit for %llu",
1995                                dstgroup->qgroupid);
1996                         goto unlock;
1997                 }
1998         }
1999 
2000         if (srcid) {
2001                 srcgroup = find_qgroup_rb(fs_info, srcid);
2002                 if (!srcgroup)
2003                         goto unlock;
2004 
2005                 /*
2006                  * We call inherit after we clone the root in order to make sure
2007                  * our counts don't go crazy, so at this point the only
2008                  * difference between the two roots should be the root node.
2009                  */
2010                 dstgroup->rfer = srcgroup->rfer;
2011                 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2012                 dstgroup->excl = level_size;
2013                 dstgroup->excl_cmpr = level_size;
2014                 srcgroup->excl = level_size;
2015                 srcgroup->excl_cmpr = level_size;
2016 
2017                 /* inherit the limit info */
2018                 dstgroup->lim_flags = srcgroup->lim_flags;
2019                 dstgroup->max_rfer = srcgroup->max_rfer;
2020                 dstgroup->max_excl = srcgroup->max_excl;
2021                 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2022                 dstgroup->rsv_excl = srcgroup->rsv_excl;
2023 
2024                 qgroup_dirty(fs_info, dstgroup);
2025                 qgroup_dirty(fs_info, srcgroup);
2026         }
2027 
2028         if (!inherit)
2029                 goto unlock;
2030 
2031         i_qgroups = (u64 *)(inherit + 1);
2032         for (i = 0; i < inherit->num_qgroups; ++i) {
2033                 if (*i_qgroups) {
2034                         ret = add_relation_rb(quota_root->fs_info, objectid,
2035                                               *i_qgroups);
2036                         if (ret)
2037                                 goto unlock;
2038                 }
2039                 ++i_qgroups;
2040         }
2041 
2042         for (i = 0; i <  inherit->num_ref_copies; ++i, i_qgroups += 2) {
2043                 struct btrfs_qgroup *src;
2044                 struct btrfs_qgroup *dst;
2045 
2046                 if (!i_qgroups[0] || !i_qgroups[1])
2047                         continue;
2048 
2049                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2050                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2051 
2052                 if (!src || !dst) {
2053                         ret = -EINVAL;
2054                         goto unlock;
2055                 }
2056 
2057                 dst->rfer = src->rfer - level_size;
2058                 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2059         }
2060         for (i = 0; i <  inherit->num_excl_copies; ++i, i_qgroups += 2) {
2061                 struct btrfs_qgroup *src;
2062                 struct btrfs_qgroup *dst;
2063 
2064                 if (!i_qgroups[0] || !i_qgroups[1])
2065                         continue;
2066 
2067                 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2068                 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2069 
2070                 if (!src || !dst) {
2071                         ret = -EINVAL;
2072                         goto unlock;
2073                 }
2074 
2075                 dst->excl = src->excl + level_size;
2076                 dst->excl_cmpr = src->excl_cmpr + level_size;
2077         }
2078 
2079 unlock:
2080         spin_unlock(&fs_info->qgroup_lock);
2081 out:
2082         mutex_unlock(&fs_info->qgroup_ioctl_lock);
2083         return ret;
2084 }
2085 
2086 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2087 {
2088         struct btrfs_root *quota_root;
2089         struct btrfs_qgroup *qgroup;
2090         struct btrfs_fs_info *fs_info = root->fs_info;
2091         u64 ref_root = root->root_key.objectid;
2092         int ret = 0;
2093         struct ulist_node *unode;
2094         struct ulist_iterator uiter;
2095 
2096         if (!is_fstree(ref_root))
2097                 return 0;
2098 
2099         if (num_bytes == 0)
2100                 return 0;
2101 
2102         spin_lock(&fs_info->qgroup_lock);
2103         quota_root = fs_info->quota_root;
2104         if (!quota_root)
2105                 goto out;
2106 
2107         qgroup = find_qgroup_rb(fs_info, ref_root);
2108         if (!qgroup)
2109                 goto out;
2110 
2111         /*
2112          * in a first step, we check all affected qgroups if any limits would
2113          * be exceeded
2114          */
2115         ulist_reinit(fs_info->qgroup_ulist);
2116         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2117                         (uintptr_t)qgroup, GFP_ATOMIC);
2118         if (ret < 0)
2119                 goto out;
2120         ULIST_ITER_INIT(&uiter);
2121         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2122                 struct btrfs_qgroup *qg;
2123                 struct btrfs_qgroup_list *glist;
2124 
2125                 qg = u64_to_ptr(unode->aux);
2126 
2127                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2128                     qg->reserved + (s64)qg->rfer + num_bytes >
2129                     qg->max_rfer) {
2130                         ret = -EDQUOT;
2131                         goto out;
2132                 }
2133 
2134                 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2135                     qg->reserved + (s64)qg->excl + num_bytes >
2136                     qg->max_excl) {
2137                         ret = -EDQUOT;
2138                         goto out;
2139                 }
2140 
2141                 list_for_each_entry(glist, &qg->groups, next_group) {
2142                         ret = ulist_add(fs_info->qgroup_ulist,
2143                                         glist->group->qgroupid,
2144                                         (uintptr_t)glist->group, GFP_ATOMIC);
2145                         if (ret < 0)
2146                                 goto out;
2147                 }
2148         }
2149         ret = 0;
2150         /*
2151          * no limits exceeded, now record the reservation into all qgroups
2152          */
2153         ULIST_ITER_INIT(&uiter);
2154         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2155                 struct btrfs_qgroup *qg;
2156 
2157                 qg = u64_to_ptr(unode->aux);
2158 
2159                 qg->reserved += num_bytes;
2160         }
2161 
2162 out:
2163         spin_unlock(&fs_info->qgroup_lock);
2164         return ret;
2165 }
2166 
2167 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2168                                u64 ref_root, u64 num_bytes)
2169 {
2170         struct btrfs_root *quota_root;
2171         struct btrfs_qgroup *qgroup;
2172         struct ulist_node *unode;
2173         struct ulist_iterator uiter;
2174         int ret = 0;
2175 
2176         if (!is_fstree(ref_root))
2177                 return;
2178 
2179         if (num_bytes == 0)
2180                 return;
2181 
2182         spin_lock(&fs_info->qgroup_lock);
2183 
2184         quota_root = fs_info->quota_root;
2185         if (!quota_root)
2186                 goto out;
2187 
2188         qgroup = find_qgroup_rb(fs_info, ref_root);
2189         if (!qgroup)
2190                 goto out;
2191 
2192         ulist_reinit(fs_info->qgroup_ulist);
2193         ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2194                         (uintptr_t)qgroup, GFP_ATOMIC);
2195         if (ret < 0)
2196                 goto out;
2197         ULIST_ITER_INIT(&uiter);
2198         while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2199                 struct btrfs_qgroup *qg;
2200                 struct btrfs_qgroup_list *glist;
2201 
2202                 qg = u64_to_ptr(unode->aux);
2203 
2204                 qg->reserved -= num_bytes;
2205 
2206                 list_for_each_entry(glist, &qg->groups, next_group) {
2207                         ret = ulist_add(fs_info->qgroup_ulist,
2208                                         glist->group->qgroupid,
2209                                         (uintptr_t)glist->group, GFP_ATOMIC);
2210                         if (ret < 0)
2211                                 goto out;
2212                 }
2213         }
2214 
2215 out:
2216         spin_unlock(&fs_info->qgroup_lock);
2217 }
2218 
2219 static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2220 {
2221         return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2222                                          num_bytes);
2223 }
2224 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2225 {
2226         if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2227                 return;
2228         btrfs_err(trans->fs_info,
2229                 "qgroups not uptodate in trans handle %p:  list is%s empty, "
2230                 "seq is %#x.%x",
2231                 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2232                 (u32)(trans->delayed_ref_elem.seq >> 32),
2233                 (u32)trans->delayed_ref_elem.seq);
2234         BUG();
2235 }
2236 
2237 /*
2238  * returns < 0 on error, 0 when more leafs are to be scanned.
2239  * returns 1 when done.
2240  */
2241 static int
2242 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2243                    struct btrfs_trans_handle *trans)
2244 {
2245         struct btrfs_key found;
2246         struct extent_buffer *scratch_leaf = NULL;
2247         struct ulist *roots = NULL;
2248         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2249         u64 num_bytes;
2250         int slot;
2251         int ret;
2252 
2253         mutex_lock(&fs_info->qgroup_rescan_lock);
2254         ret = btrfs_search_slot_for_read(fs_info->extent_root,
2255                                          &fs_info->qgroup_rescan_progress,
2256                                          path, 1, 0);
2257 
2258         pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2259                  fs_info->qgroup_rescan_progress.objectid,
2260                  fs_info->qgroup_rescan_progress.type,
2261                  fs_info->qgroup_rescan_progress.offset, ret);
2262 
2263         if (ret) {
2264                 /*
2265                  * The rescan is about to end, we will not be scanning any
2266                  * further blocks. We cannot unset the RESCAN flag here, because
2267                  * we want to commit the transaction if everything went well.
2268                  * To make the live accounting work in this phase, we set our
2269                  * scan progress pointer such that every real extent objectid
2270                  * will be smaller.
2271                  */
2272                 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2273                 btrfs_release_path(path);
2274                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2275                 return ret;
2276         }
2277 
2278         btrfs_item_key_to_cpu(path->nodes[0], &found,
2279                               btrfs_header_nritems(path->nodes[0]) - 1);
2280         fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2281 
2282         btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2283         scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2284         if (!scratch_leaf) {
2285                 ret = -ENOMEM;
2286                 mutex_unlock(&fs_info->qgroup_rescan_lock);
2287                 goto out;
2288         }
2289         extent_buffer_get(scratch_leaf);
2290         btrfs_tree_read_lock(scratch_leaf);
2291         btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2292         slot = path->slots[0];
2293         btrfs_release_path(path);
2294         mutex_unlock(&fs_info->qgroup_rescan_lock);
2295 
2296         for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2297                 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2298                 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2299                     found.type != BTRFS_METADATA_ITEM_KEY)
2300                         continue;
2301                 if (found.type == BTRFS_METADATA_ITEM_KEY)
2302                         num_bytes = fs_info->extent_root->nodesize;
2303                 else
2304                         num_bytes = found.offset;
2305 
2306                 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2307                                            &roots);
2308                 if (ret < 0)
2309                         goto out;
2310                 /* For rescan, just pass old_roots as NULL */
2311                 ret = btrfs_qgroup_account_extent(trans, fs_info,
2312                                 found.objectid, num_bytes, NULL, roots);
2313                 if (ret < 0)
2314                         goto out;
2315         }
2316 out:
2317         if (scratch_leaf) {
2318                 btrfs_tree_read_unlock_blocking(scratch_leaf);
2319                 free_extent_buffer(scratch_leaf);
2320         }
2321         btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2322 
2323         return ret;
2324 }
2325 
2326 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2327 {
2328         struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2329                                                      qgroup_rescan_work);
2330         struct btrfs_path *path;
2331         struct btrfs_trans_handle *trans = NULL;
2332         int err = -ENOMEM;
2333         int ret = 0;
2334 
2335         path = btrfs_alloc_path();
2336         if (!path)
2337                 goto out;
2338 
2339         err = 0;
2340         while (!err && !btrfs_fs_closing(fs_info)) {
2341                 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2342                 if (IS_ERR(trans)) {
2343                         err = PTR_ERR(trans);
2344                         break;
2345                 }
2346                 if (!fs_info->quota_enabled) {
2347                         err = -EINTR;
2348                 } else {
2349                         err = qgroup_rescan_leaf(fs_info, path, trans);
2350                 }
2351                 if (err > 0)
2352                         btrfs_commit_transaction(trans, fs_info->fs_root);
2353                 else
2354                         btrfs_end_transaction(trans, fs_info->fs_root);
2355         }
2356 
2357 out:
2358         btrfs_free_path(path);
2359 
2360         mutex_lock(&fs_info->qgroup_rescan_lock);
2361         if (!btrfs_fs_closing(fs_info))
2362                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2363 
2364         if (err > 0 &&
2365             fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2366                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2367         } else if (err < 0) {
2368                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2369         }
2370         mutex_unlock(&fs_info->qgroup_rescan_lock);
2371 
2372         /*
2373          * only update status, since the previous part has already updated the
2374          * qgroup info.
2375          */
2376         trans = btrfs_start_transaction(fs_info->quota_root, 1);
2377         if (IS_ERR(trans)) {
2378                 err = PTR_ERR(trans);
2379                 btrfs_err(fs_info,
2380                           "fail to start transaction for status update: %d\n",
2381                           err);
2382                 goto done;
2383         }
2384         ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2385         if (ret < 0) {
2386                 err = ret;
2387                 btrfs_err(fs_info, "fail to update qgroup status: %d\n", err);
2388         }
2389         btrfs_end_transaction(trans, fs_info->quota_root);
2390 
2391         if (btrfs_fs_closing(fs_info)) {
2392                 btrfs_info(fs_info, "qgroup scan paused");
2393         } else if (err >= 0) {
2394                 btrfs_info(fs_info, "qgroup scan completed%s",
2395                         err > 0 ? " (inconsistency flag cleared)" : "");
2396         } else {
2397                 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2398         }
2399 
2400 done:
2401         mutex_lock(&fs_info->qgroup_rescan_lock);
2402         fs_info->qgroup_rescan_running = false;
2403         mutex_unlock(&fs_info->qgroup_rescan_lock);
2404         complete_all(&fs_info->qgroup_rescan_completion);
2405 }
2406 
2407 /*
2408  * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2409  * memory required for the rescan context.
2410  */
2411 static int
2412 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2413                    int init_flags)
2414 {
2415         int ret = 0;
2416 
2417         if (!init_flags &&
2418             (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2419              !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2420                 ret = -EINVAL;
2421                 goto err;
2422         }
2423 
2424         mutex_lock(&fs_info->qgroup_rescan_lock);
2425         spin_lock(&fs_info->qgroup_lock);
2426 
2427         if (init_flags) {
2428                 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2429                         ret = -EINPROGRESS;
2430                 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2431                         ret = -EINVAL;
2432 
2433                 if (ret) {
2434                         spin_unlock(&fs_info->qgroup_lock);
2435                         mutex_unlock(&fs_info->qgroup_rescan_lock);
2436                         goto err;
2437                 }
2438                 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2439         }
2440 
2441         memset(&fs_info->qgroup_rescan_progress, 0,
2442                 sizeof(fs_info->qgroup_rescan_progress));
2443         fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2444         init_completion(&fs_info->qgroup_rescan_completion);
2445         fs_info->qgroup_rescan_running = true;
2446 
2447         spin_unlock(&fs_info->qgroup_lock);
2448         mutex_unlock(&fs_info->qgroup_rescan_lock);
2449 
2450         memset(&fs_info->qgroup_rescan_work, 0,
2451                sizeof(fs_info->qgroup_rescan_work));
2452         btrfs_init_work(&fs_info->qgroup_rescan_work,
2453                         btrfs_qgroup_rescan_helper,
2454                         btrfs_qgroup_rescan_worker, NULL, NULL);
2455 
2456         if (ret) {
2457 err:
2458                 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2459                 return ret;
2460         }
2461 
2462         return 0;
2463 }
2464 
2465 static void
2466 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2467 {
2468         struct rb_node *n;
2469         struct btrfs_qgroup *qgroup;
2470 
2471         spin_lock(&fs_info->qgroup_lock);
2472         /* clear all current qgroup tracking information */
2473         for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2474                 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2475                 qgroup->rfer = 0;
2476                 qgroup->rfer_cmpr = 0;
2477                 qgroup->excl = 0;
2478                 qgroup->excl_cmpr = 0;
2479         }
2480         spin_unlock(&fs_info->qgroup_lock);
2481 }
2482 
2483 int
2484 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2485 {
2486         int ret = 0;
2487         struct btrfs_trans_handle *trans;
2488 
2489         ret = qgroup_rescan_init(fs_info, 0, 1);
2490         if (ret)
2491                 return ret;
2492 
2493         /*
2494          * We have set the rescan_progress to 0, which means no more
2495          * delayed refs will be accounted by btrfs_qgroup_account_ref.
2496          * However, btrfs_qgroup_account_ref may be right after its call
2497          * to btrfs_find_all_roots, in which case it would still do the
2498          * accounting.
2499          * To solve this, we're committing the transaction, which will
2500          * ensure we run all delayed refs and only after that, we are
2501          * going to clear all tracking information for a clean start.
2502          */
2503 
2504         trans = btrfs_join_transaction(fs_info->fs_root);
2505         if (IS_ERR(trans)) {
2506                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2507                 return PTR_ERR(trans);
2508         }
2509         ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2510         if (ret) {
2511                 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2512                 return ret;
2513         }
2514 
2515         qgroup_rescan_zero_tracking(fs_info);
2516 
2517         btrfs_queue_work(fs_info->qgroup_rescan_workers,
2518                          &fs_info->qgroup_rescan_work);
2519 
2520         return 0;
2521 }
2522 
2523 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2524                                      bool interruptible)
2525 {
2526         int running;
2527         int ret = 0;
2528 
2529         mutex_lock(&fs_info->qgroup_rescan_lock);
2530         spin_lock(&fs_info->qgroup_lock);
2531         running = fs_info->qgroup_rescan_running;
2532         spin_unlock(&fs_info->qgroup_lock);
2533         mutex_unlock(&fs_info->qgroup_rescan_lock);
2534 
2535         if (!running)
2536                 return 0;
2537 
2538         if (interruptible)
2539                 ret = wait_for_completion_interruptible(
2540                                         &fs_info->qgroup_rescan_completion);
2541         else
2542                 wait_for_completion(&fs_info->qgroup_rescan_completion);
2543 
2544         return ret;
2545 }
2546 
2547 /*
2548  * this is only called from open_ctree where we're still single threaded, thus
2549  * locking is omitted here.
2550  */
2551 void
2552 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2553 {
2554         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2555                 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2556                                  &fs_info->qgroup_rescan_work);
2557 }
2558 
2559 /*
2560  * Reserve qgroup space for range [start, start + len).
2561  *
2562  * This function will either reserve space from related qgroups or doing
2563  * nothing if the range is already reserved.
2564  *
2565  * Return 0 for successful reserve
2566  * Return <0 for error (including -EQUOT)
2567  *
2568  * NOTE: this function may sleep for memory allocation.
2569  */
2570 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2571 {
2572         struct btrfs_root *root = BTRFS_I(inode)->root;
2573         struct extent_changeset changeset;
2574         struct ulist_node *unode;
2575         struct ulist_iterator uiter;
2576         int ret;
2577 
2578         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2579             len == 0)
2580                 return 0;
2581 
2582         changeset.bytes_changed = 0;
2583         changeset.range_changed = ulist_alloc(GFP_NOFS);
2584         ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2585                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2586         trace_btrfs_qgroup_reserve_data(inode, start, len,
2587                                         changeset.bytes_changed,
2588                                         QGROUP_RESERVE);
2589         if (ret < 0)
2590                 goto cleanup;
2591         ret = qgroup_reserve(root, changeset.bytes_changed);
2592         if (ret < 0)
2593                 goto cleanup;
2594 
2595         ulist_free(changeset.range_changed);
2596         return ret;
2597 
2598 cleanup:
2599         /* cleanup already reserved ranges */
2600         ULIST_ITER_INIT(&uiter);
2601         while ((unode = ulist_next(changeset.range_changed, &uiter)))
2602                 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2603                                  unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2604                                  GFP_NOFS);
2605         ulist_free(changeset.range_changed);
2606         return ret;
2607 }
2608 
2609 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2610                                        int free)
2611 {
2612         struct extent_changeset changeset;
2613         int trace_op = QGROUP_RELEASE;
2614         int ret;
2615 
2616         changeset.bytes_changed = 0;
2617         changeset.range_changed = ulist_alloc(GFP_NOFS);
2618         if (!changeset.range_changed)
2619                 return -ENOMEM;
2620 
2621         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start, 
2622                         start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2623         if (ret < 0)
2624                 goto out;
2625 
2626         if (free) {
2627                 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2628                 trace_op = QGROUP_FREE;
2629         }
2630         trace_btrfs_qgroup_release_data(inode, start, len,
2631                                         changeset.bytes_changed, trace_op);
2632 out:
2633         ulist_free(changeset.range_changed);
2634         return ret;
2635 }
2636 
2637 /*
2638  * Free a reserved space range from io_tree and related qgroups
2639  *
2640  * Should be called when a range of pages get invalidated before reaching disk.
2641  * Or for error cleanup case.
2642  *
2643  * For data written to disk, use btrfs_qgroup_release_data().
2644  *
2645  * NOTE: This function may sleep for memory allocation.
2646  */
2647 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2648 {
2649         return __btrfs_qgroup_release_data(inode, start, len, 1);
2650 }
2651 
2652 /*
2653  * Release a reserved space range from io_tree only.
2654  *
2655  * Should be called when a range of pages get written to disk and corresponding
2656  * FILE_EXTENT is inserted into corresponding root.
2657  *
2658  * Since new qgroup accounting framework will only update qgroup numbers at
2659  * commit_transaction() time, its reserved space shouldn't be freed from
2660  * related qgroups.
2661  *
2662  * But we should release the range from io_tree, to allow further write to be
2663  * COWed.
2664  *
2665  * NOTE: This function may sleep for memory allocation.
2666  */
2667 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2668 {
2669         return __btrfs_qgroup_release_data(inode, start, len, 0);
2670 }
2671 
2672 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2673 {
2674         int ret;
2675 
2676         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2677             num_bytes == 0)
2678                 return 0;
2679 
2680         BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2681         ret = qgroup_reserve(root, num_bytes);
2682         if (ret < 0)
2683                 return ret;
2684         atomic_add(num_bytes, &root->qgroup_meta_rsv);
2685         return ret;
2686 }
2687 
2688 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2689 {
2690         int reserved;
2691 
2692         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2693                 return;
2694 
2695         reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2696         if (reserved == 0)
2697                 return;
2698         qgroup_free(root, reserved);
2699 }
2700 
2701 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2702 {
2703         if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2704                 return;
2705 
2706         BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2707         WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2708         atomic_sub(num_bytes, &root->qgroup_meta_rsv);
2709         qgroup_free(root, num_bytes);
2710 }
2711 
2712 /*
2713  * Check qgroup reserved space leaking, normally at destroy inode
2714  * time
2715  */
2716 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2717 {
2718         struct extent_changeset changeset;
2719         struct ulist_node *unode;
2720         struct ulist_iterator iter;
2721         int ret;
2722 
2723         changeset.bytes_changed = 0;
2724         changeset.range_changed = ulist_alloc(GFP_NOFS);
2725         if (WARN_ON(!changeset.range_changed))
2726                 return;
2727 
2728         ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2729                         EXTENT_QGROUP_RESERVED, &changeset);
2730 
2731         WARN_ON(ret < 0);
2732         if (WARN_ON(changeset.bytes_changed)) {
2733                 ULIST_ITER_INIT(&iter);
2734                 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2735                         btrfs_warn(BTRFS_I(inode)->root->fs_info,
2736                                 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2737                                 inode->i_ino, unode->val, unode->aux);
2738                 }
2739                 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2740         }
2741         ulist_free(changeset.range_changed);
2742 }
2743 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp