blob: 28f2b39f6a25d338565fa00b526c667d3542e0a0 [file] [log] [blame]
Arne Jansenbed92ea2012-06-28 18:03:02 +02001/*
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
27#include "ctree.h"
28#include "transaction.h"
29#include "disk-io.h"
30#include "locking.h"
31#include "ulist.h"
32#include "ioctl.h"
33#include "backref.h"
34
35/* TODO XXX FIXME
36 * - subvol delete -> delete when ref goes to 0? delete limits also?
37 * - reorganize keys
38 * - compressed
39 * - sync
40 * - rescan
41 * - copy also limits on subvol creation
42 * - limit
43 * - caches fuer ulists
44 * - performance benchmarks
45 * - check all ioctl parameters
46 */
47
48/*
49 * one struct for each qgroup, organized in fs_info->qgroup_tree.
50 */
51struct btrfs_qgroup {
52 u64 qgroupid;
53
54 /*
55 * state
56 */
57 u64 rfer; /* referenced */
58 u64 rfer_cmpr; /* referenced compressed */
59 u64 excl; /* exclusive */
60 u64 excl_cmpr; /* exclusive compressed */
61
62 /*
63 * limits
64 */
65 u64 lim_flags; /* which limits are set */
66 u64 max_rfer;
67 u64 max_excl;
68 u64 rsv_rfer;
69 u64 rsv_excl;
70
71 /*
72 * reservation tracking
73 */
74 u64 reserved;
75
76 /*
77 * lists
78 */
79 struct list_head groups; /* groups this group is member of */
80 struct list_head members; /* groups that are members of this group */
81 struct list_head dirty; /* dirty groups */
82 struct rb_node node; /* tree of qgroups */
83
84 /*
85 * temp variables for accounting operations
86 */
87 u64 tag;
88 u64 refcnt;
89};
90
91/*
92 * glue structure to represent the relations between qgroups.
93 */
94struct btrfs_qgroup_list {
95 struct list_head next_group;
96 struct list_head next_member;
97 struct btrfs_qgroup *group;
98 struct btrfs_qgroup *member;
99};
100
101/* must be called with qgroup_lock held */
102static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
103 u64 qgroupid)
104{
105 struct rb_node *n = fs_info->qgroup_tree.rb_node;
106 struct btrfs_qgroup *qgroup;
107
108 while (n) {
109 qgroup = rb_entry(n, struct btrfs_qgroup, node);
110 if (qgroup->qgroupid < qgroupid)
111 n = n->rb_left;
112 else if (qgroup->qgroupid > qgroupid)
113 n = n->rb_right;
114 else
115 return qgroup;
116 }
117 return NULL;
118}
119
120/* must be called with qgroup_lock held */
121static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
122 u64 qgroupid)
123{
124 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
125 struct rb_node *parent = NULL;
126 struct btrfs_qgroup *qgroup;
127
128 while (*p) {
129 parent = *p;
130 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
131
132 if (qgroup->qgroupid < qgroupid)
133 p = &(*p)->rb_left;
134 else if (qgroup->qgroupid > qgroupid)
135 p = &(*p)->rb_right;
136 else
137 return qgroup;
138 }
139
140 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
141 if (!qgroup)
142 return ERR_PTR(-ENOMEM);
143
144 qgroup->qgroupid = qgroupid;
145 INIT_LIST_HEAD(&qgroup->groups);
146 INIT_LIST_HEAD(&qgroup->members);
147 INIT_LIST_HEAD(&qgroup->dirty);
148
149 rb_link_node(&qgroup->node, parent, p);
150 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
151
152 return qgroup;
153}
154
155/* must be called with qgroup_lock held */
156static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
157{
158 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
159 struct btrfs_qgroup_list *list;
160
161 if (!qgroup)
162 return -ENOENT;
163
164 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
165 list_del(&qgroup->dirty);
166
167 while (!list_empty(&qgroup->groups)) {
168 list = list_first_entry(&qgroup->groups,
169 struct btrfs_qgroup_list, next_group);
170 list_del(&list->next_group);
171 list_del(&list->next_member);
172 kfree(list);
173 }
174
175 while (!list_empty(&qgroup->members)) {
176 list = list_first_entry(&qgroup->members,
177 struct btrfs_qgroup_list, next_member);
178 list_del(&list->next_group);
179 list_del(&list->next_member);
180 kfree(list);
181 }
182 kfree(qgroup);
183
184 return 0;
185}
186
187/* must be called with qgroup_lock held */
188static int add_relation_rb(struct btrfs_fs_info *fs_info,
189 u64 memberid, u64 parentid)
190{
191 struct btrfs_qgroup *member;
192 struct btrfs_qgroup *parent;
193 struct btrfs_qgroup_list *list;
194
195 member = find_qgroup_rb(fs_info, memberid);
196 parent = find_qgroup_rb(fs_info, parentid);
197 if (!member || !parent)
198 return -ENOENT;
199
200 list = kzalloc(sizeof(*list), GFP_ATOMIC);
201 if (!list)
202 return -ENOMEM;
203
204 list->group = parent;
205 list->member = member;
206 list_add_tail(&list->next_group, &member->groups);
207 list_add_tail(&list->next_member, &parent->members);
208
209 return 0;
210}
211
212/* must be called with qgroup_lock held */
213static int del_relation_rb(struct btrfs_fs_info *fs_info,
214 u64 memberid, u64 parentid)
215{
216 struct btrfs_qgroup *member;
217 struct btrfs_qgroup *parent;
218 struct btrfs_qgroup_list *list;
219
220 member = find_qgroup_rb(fs_info, memberid);
221 parent = find_qgroup_rb(fs_info, parentid);
222 if (!member || !parent)
223 return -ENOENT;
224
225 list_for_each_entry(list, &member->groups, next_group) {
226 if (list->group == parent) {
227 list_del(&list->next_group);
228 list_del(&list->next_member);
229 kfree(list);
230 return 0;
231 }
232 }
233 return -ENOENT;
234}
235
236/*
237 * The full config is read in one go, only called from open_ctree()
238 * It doesn't use any locking, as at this point we're still single-threaded
239 */
240int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
241{
242 struct btrfs_key key;
243 struct btrfs_key found_key;
244 struct btrfs_root *quota_root = fs_info->quota_root;
245 struct btrfs_path *path = NULL;
246 struct extent_buffer *l;
247 int slot;
248 int ret = 0;
249 u64 flags = 0;
250
251 if (!fs_info->quota_enabled)
252 return 0;
253
254 path = btrfs_alloc_path();
255 if (!path) {
256 ret = -ENOMEM;
257 goto out;
258 }
259
260 /* default this to quota off, in case no status key is found */
261 fs_info->qgroup_flags = 0;
262
263 /*
264 * pass 1: read status, all qgroup infos and limits
265 */
266 key.objectid = 0;
267 key.type = 0;
268 key.offset = 0;
269 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
270 if (ret)
271 goto out;
272
273 while (1) {
274 struct btrfs_qgroup *qgroup;
275
276 slot = path->slots[0];
277 l = path->nodes[0];
278 btrfs_item_key_to_cpu(l, &found_key, slot);
279
280 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
281 struct btrfs_qgroup_status_item *ptr;
282
283 ptr = btrfs_item_ptr(l, slot,
284 struct btrfs_qgroup_status_item);
285
286 if (btrfs_qgroup_status_version(l, ptr) !=
287 BTRFS_QGROUP_STATUS_VERSION) {
288 printk(KERN_ERR
289 "btrfs: old qgroup version, quota disabled\n");
290 goto out;
291 }
292 if (btrfs_qgroup_status_generation(l, ptr) !=
293 fs_info->generation) {
294 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
295 printk(KERN_ERR
296 "btrfs: qgroup generation mismatch, "
297 "marked as inconsistent\n");
298 }
299 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
300 ptr);
301 /* FIXME read scan element */
302 goto next1;
303 }
304
305 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
306 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
307 goto next1;
308
309 qgroup = find_qgroup_rb(fs_info, found_key.offset);
310 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
311 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
312 printk(KERN_ERR "btrfs: inconsitent qgroup config\n");
313 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
314 }
315 if (!qgroup) {
316 qgroup = add_qgroup_rb(fs_info, found_key.offset);
317 if (IS_ERR(qgroup)) {
318 ret = PTR_ERR(qgroup);
319 goto out;
320 }
321 }
322 switch (found_key.type) {
323 case BTRFS_QGROUP_INFO_KEY: {
324 struct btrfs_qgroup_info_item *ptr;
325
326 ptr = btrfs_item_ptr(l, slot,
327 struct btrfs_qgroup_info_item);
328 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
329 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
330 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
331 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
332 /* generation currently unused */
333 break;
334 }
335 case BTRFS_QGROUP_LIMIT_KEY: {
336 struct btrfs_qgroup_limit_item *ptr;
337
338 ptr = btrfs_item_ptr(l, slot,
339 struct btrfs_qgroup_limit_item);
340 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
341 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
342 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
343 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
344 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
345 break;
346 }
347 }
348next1:
349 ret = btrfs_next_item(quota_root, path);
350 if (ret < 0)
351 goto out;
352 if (ret)
353 break;
354 }
355 btrfs_release_path(path);
356
357 /*
358 * pass 2: read all qgroup relations
359 */
360 key.objectid = 0;
361 key.type = BTRFS_QGROUP_RELATION_KEY;
362 key.offset = 0;
363 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
364 if (ret)
365 goto out;
366 while (1) {
367 slot = path->slots[0];
368 l = path->nodes[0];
369 btrfs_item_key_to_cpu(l, &found_key, slot);
370
371 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
372 goto next2;
373
374 if (found_key.objectid > found_key.offset) {
375 /* parent <- member, not needed to build config */
376 /* FIXME should we omit the key completely? */
377 goto next2;
378 }
379
380 ret = add_relation_rb(fs_info, found_key.objectid,
381 found_key.offset);
Arne Jansenff248582013-01-17 01:22:08 -0700382 if (ret == -ENOENT) {
383 printk(KERN_WARNING
384 "btrfs: orphan qgroup relation 0x%llx->0x%llx\n",
385 (unsigned long long)found_key.objectid,
386 (unsigned long long)found_key.offset);
387 ret = 0; /* ignore the error */
388 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200389 if (ret)
390 goto out;
391next2:
392 ret = btrfs_next_item(quota_root, path);
393 if (ret < 0)
394 goto out;
395 if (ret)
396 break;
397 }
398out:
399 fs_info->qgroup_flags |= flags;
400 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
401 fs_info->quota_enabled = 0;
402 fs_info->pending_quota_state = 0;
403 }
404 btrfs_free_path(path);
405
406 return ret < 0 ? ret : 0;
407}
408
409/*
410 * This is only called from close_ctree() or open_ctree(), both in single-
411 * treaded paths. Clean up the in-memory structures. No locking needed.
412 */
413void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
414{
415 struct rb_node *n;
416 struct btrfs_qgroup *qgroup;
417 struct btrfs_qgroup_list *list;
418
419 while ((n = rb_first(&fs_info->qgroup_tree))) {
420 qgroup = rb_entry(n, struct btrfs_qgroup, node);
421 rb_erase(n, &fs_info->qgroup_tree);
422
423 WARN_ON(!list_empty(&qgroup->dirty));
424
425 while (!list_empty(&qgroup->groups)) {
426 list = list_first_entry(&qgroup->groups,
427 struct btrfs_qgroup_list,
428 next_group);
429 list_del(&list->next_group);
430 list_del(&list->next_member);
431 kfree(list);
432 }
433
434 while (!list_empty(&qgroup->members)) {
435 list = list_first_entry(&qgroup->members,
436 struct btrfs_qgroup_list,
437 next_member);
438 list_del(&list->next_group);
439 list_del(&list->next_member);
440 kfree(list);
441 }
442 kfree(qgroup);
443 }
444}
445
446static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
447 struct btrfs_root *quota_root,
448 u64 src, u64 dst)
449{
450 int ret;
451 struct btrfs_path *path;
452 struct btrfs_key key;
453
454 path = btrfs_alloc_path();
455 if (!path)
456 return -ENOMEM;
457
458 key.objectid = src;
459 key.type = BTRFS_QGROUP_RELATION_KEY;
460 key.offset = dst;
461
462 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
463
464 btrfs_mark_buffer_dirty(path->nodes[0]);
465
466 btrfs_free_path(path);
467 return ret;
468}
469
470static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
471 struct btrfs_root *quota_root,
472 u64 src, u64 dst)
473{
474 int ret;
475 struct btrfs_path *path;
476 struct btrfs_key key;
477
478 path = btrfs_alloc_path();
479 if (!path)
480 return -ENOMEM;
481
482 key.objectid = src;
483 key.type = BTRFS_QGROUP_RELATION_KEY;
484 key.offset = dst;
485
486 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
487 if (ret < 0)
488 goto out;
489
490 if (ret > 0) {
491 ret = -ENOENT;
492 goto out;
493 }
494
495 ret = btrfs_del_item(trans, quota_root, path);
496out:
497 btrfs_free_path(path);
498 return ret;
499}
500
501static int add_qgroup_item(struct btrfs_trans_handle *trans,
502 struct btrfs_root *quota_root, u64 qgroupid)
503{
504 int ret;
505 struct btrfs_path *path;
506 struct btrfs_qgroup_info_item *qgroup_info;
507 struct btrfs_qgroup_limit_item *qgroup_limit;
508 struct extent_buffer *leaf;
509 struct btrfs_key key;
510
511 path = btrfs_alloc_path();
512 if (!path)
513 return -ENOMEM;
514
515 key.objectid = 0;
516 key.type = BTRFS_QGROUP_INFO_KEY;
517 key.offset = qgroupid;
518
519 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
520 sizeof(*qgroup_info));
521 if (ret)
522 goto out;
523
524 leaf = path->nodes[0];
525 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
526 struct btrfs_qgroup_info_item);
527 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
528 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
529 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
530 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
531 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
532
533 btrfs_mark_buffer_dirty(leaf);
534
535 btrfs_release_path(path);
536
537 key.type = BTRFS_QGROUP_LIMIT_KEY;
538 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
539 sizeof(*qgroup_limit));
540 if (ret)
541 goto out;
542
543 leaf = path->nodes[0];
544 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
545 struct btrfs_qgroup_limit_item);
546 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
547 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
548 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
549 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
550 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
551
552 btrfs_mark_buffer_dirty(leaf);
553
554 ret = 0;
555out:
556 btrfs_free_path(path);
557 return ret;
558}
559
560static int del_qgroup_item(struct btrfs_trans_handle *trans,
561 struct btrfs_root *quota_root, u64 qgroupid)
562{
563 int ret;
564 struct btrfs_path *path;
565 struct btrfs_key key;
566
567 path = btrfs_alloc_path();
568 if (!path)
569 return -ENOMEM;
570
571 key.objectid = 0;
572 key.type = BTRFS_QGROUP_INFO_KEY;
573 key.offset = qgroupid;
574 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
575 if (ret < 0)
576 goto out;
577
578 if (ret > 0) {
579 ret = -ENOENT;
580 goto out;
581 }
582
583 ret = btrfs_del_item(trans, quota_root, path);
584 if (ret)
585 goto out;
586
587 btrfs_release_path(path);
588
589 key.type = BTRFS_QGROUP_LIMIT_KEY;
590 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
591 if (ret < 0)
592 goto out;
593
594 if (ret > 0) {
595 ret = -ENOENT;
596 goto out;
597 }
598
599 ret = btrfs_del_item(trans, quota_root, path);
600
601out:
602 btrfs_free_path(path);
603 return ret;
604}
605
606static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
607 struct btrfs_root *root, u64 qgroupid,
608 u64 flags, u64 max_rfer, u64 max_excl,
609 u64 rsv_rfer, u64 rsv_excl)
610{
611 struct btrfs_path *path;
612 struct btrfs_key key;
613 struct extent_buffer *l;
614 struct btrfs_qgroup_limit_item *qgroup_limit;
615 int ret;
616 int slot;
617
618 key.objectid = 0;
619 key.type = BTRFS_QGROUP_LIMIT_KEY;
620 key.offset = qgroupid;
621
622 path = btrfs_alloc_path();
623 BUG_ON(!path);
624 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
625 if (ret > 0)
626 ret = -ENOENT;
627
628 if (ret)
629 goto out;
630
631 l = path->nodes[0];
632 slot = path->slots[0];
633 qgroup_limit = btrfs_item_ptr(l, path->slots[0],
634 struct btrfs_qgroup_limit_item);
635 btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
636 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
637 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
638 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
639 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
640
641 btrfs_mark_buffer_dirty(l);
642
643out:
644 btrfs_free_path(path);
645 return ret;
646}
647
648static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
649 struct btrfs_root *root,
650 struct btrfs_qgroup *qgroup)
651{
652 struct btrfs_path *path;
653 struct btrfs_key key;
654 struct extent_buffer *l;
655 struct btrfs_qgroup_info_item *qgroup_info;
656 int ret;
657 int slot;
658
659 key.objectid = 0;
660 key.type = BTRFS_QGROUP_INFO_KEY;
661 key.offset = qgroup->qgroupid;
662
663 path = btrfs_alloc_path();
664 BUG_ON(!path);
665 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
666 if (ret > 0)
667 ret = -ENOENT;
668
669 if (ret)
670 goto out;
671
672 l = path->nodes[0];
673 slot = path->slots[0];
674 qgroup_info = btrfs_item_ptr(l, path->slots[0],
675 struct btrfs_qgroup_info_item);
676 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
677 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
678 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
679 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
680 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
681
682 btrfs_mark_buffer_dirty(l);
683
684out:
685 btrfs_free_path(path);
686 return ret;
687}
688
689static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
690 struct btrfs_fs_info *fs_info,
691 struct btrfs_root *root)
692{
693 struct btrfs_path *path;
694 struct btrfs_key key;
695 struct extent_buffer *l;
696 struct btrfs_qgroup_status_item *ptr;
697 int ret;
698 int slot;
699
700 key.objectid = 0;
701 key.type = BTRFS_QGROUP_STATUS_KEY;
702 key.offset = 0;
703
704 path = btrfs_alloc_path();
705 BUG_ON(!path);
706 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
707 if (ret > 0)
708 ret = -ENOENT;
709
710 if (ret)
711 goto out;
712
713 l = path->nodes[0];
714 slot = path->slots[0];
715 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
716 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
717 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
718 /* XXX scan */
719
720 btrfs_mark_buffer_dirty(l);
721
722out:
723 btrfs_free_path(path);
724 return ret;
725}
726
727/*
728 * called with qgroup_lock held
729 */
730static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
731 struct btrfs_root *root)
732{
733 struct btrfs_path *path;
734 struct btrfs_key key;
735 int ret;
736
737 if (!root)
738 return -EINVAL;
739
740 path = btrfs_alloc_path();
741 if (!path)
742 return -ENOMEM;
743
744 while (1) {
745 key.objectid = 0;
746 key.offset = 0;
747 key.type = 0;
748
749 path->leave_spinning = 1;
750 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
751 if (ret > 0) {
752 if (path->slots[0] == 0)
753 break;
754 path->slots[0]--;
755 } else if (ret < 0) {
756 break;
757 }
758
759 ret = btrfs_del_item(trans, root, path);
760 if (ret)
761 goto out;
762 btrfs_release_path(path);
763 }
764 ret = 0;
765out:
766 root->fs_info->pending_quota_state = 0;
767 btrfs_free_path(path);
768 return ret;
769}
770
771int btrfs_quota_enable(struct btrfs_trans_handle *trans,
772 struct btrfs_fs_info *fs_info)
773{
774 struct btrfs_root *quota_root;
775 struct btrfs_path *path = NULL;
776 struct btrfs_qgroup_status_item *ptr;
777 struct extent_buffer *leaf;
778 struct btrfs_key key;
779 int ret = 0;
780
781 spin_lock(&fs_info->qgroup_lock);
782 if (fs_info->quota_root) {
783 fs_info->pending_quota_state = 1;
784 spin_unlock(&fs_info->qgroup_lock);
785 goto out;
786 }
787 spin_unlock(&fs_info->qgroup_lock);
788
789 /*
790 * initially create the quota tree
791 */
792 quota_root = btrfs_create_tree(trans, fs_info,
793 BTRFS_QUOTA_TREE_OBJECTID);
794 if (IS_ERR(quota_root)) {
795 ret = PTR_ERR(quota_root);
796 goto out;
797 }
798
799 path = btrfs_alloc_path();
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000800 if (!path) {
801 ret = -ENOMEM;
802 goto out_free_root;
803 }
Arne Jansenbed92ea2012-06-28 18:03:02 +0200804
805 key.objectid = 0;
806 key.type = BTRFS_QGROUP_STATUS_KEY;
807 key.offset = 0;
808
809 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
810 sizeof(*ptr));
811 if (ret)
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000812 goto out_free_path;
Arne Jansenbed92ea2012-06-28 18:03:02 +0200813
814 leaf = path->nodes[0];
815 ptr = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_qgroup_status_item);
817 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
818 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
819 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
820 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
821 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
822 btrfs_set_qgroup_status_scan(leaf, ptr, 0);
823
824 btrfs_mark_buffer_dirty(leaf);
825
826 spin_lock(&fs_info->qgroup_lock);
827 fs_info->quota_root = quota_root;
828 fs_info->pending_quota_state = 1;
829 spin_unlock(&fs_info->qgroup_lock);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000830out_free_path:
Arne Jansenbed92ea2012-06-28 18:03:02 +0200831 btrfs_free_path(path);
Tsutomu Itoh5b7ff5b2012-10-16 05:44:21 +0000832out_free_root:
833 if (ret) {
834 free_extent_buffer(quota_root->node);
835 free_extent_buffer(quota_root->commit_root);
836 kfree(quota_root);
837 }
838out:
Arne Jansenbed92ea2012-06-28 18:03:02 +0200839 return ret;
840}
841
842int btrfs_quota_disable(struct btrfs_trans_handle *trans,
843 struct btrfs_fs_info *fs_info)
844{
845 struct btrfs_root *tree_root = fs_info->tree_root;
846 struct btrfs_root *quota_root;
847 int ret = 0;
848
849 spin_lock(&fs_info->qgroup_lock);
850 fs_info->quota_enabled = 0;
851 fs_info->pending_quota_state = 0;
852 quota_root = fs_info->quota_root;
853 fs_info->quota_root = NULL;
854 btrfs_free_qgroup_config(fs_info);
855 spin_unlock(&fs_info->qgroup_lock);
856
857 if (!quota_root)
858 return -EINVAL;
859
860 ret = btrfs_clean_quota_tree(trans, quota_root);
861 if (ret)
862 goto out;
863
864 ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
865 if (ret)
866 goto out;
867
868 list_del(&quota_root->dirty_list);
869
870 btrfs_tree_lock(quota_root->node);
871 clean_tree_block(trans, tree_root, quota_root->node);
872 btrfs_tree_unlock(quota_root->node);
873 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
874
875 free_extent_buffer(quota_root->node);
876 free_extent_buffer(quota_root->commit_root);
877 kfree(quota_root);
878out:
879 return ret;
880}
881
882int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
883{
884 /* FIXME */
885 return 0;
886}
887
888int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
889 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
890{
891 struct btrfs_root *quota_root;
892 int ret = 0;
893
894 quota_root = fs_info->quota_root;
895 if (!quota_root)
896 return -EINVAL;
897
898 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
899 if (ret)
900 return ret;
901
902 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
903 if (ret) {
904 del_qgroup_relation_item(trans, quota_root, src, dst);
905 return ret;
906 }
907
908 spin_lock(&fs_info->qgroup_lock);
909 ret = add_relation_rb(quota_root->fs_info, src, dst);
910 spin_unlock(&fs_info->qgroup_lock);
911
912 return ret;
913}
914
915int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
916 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
917{
918 struct btrfs_root *quota_root;
919 int ret = 0;
920 int err;
921
922 quota_root = fs_info->quota_root;
923 if (!quota_root)
924 return -EINVAL;
925
926 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
927 err = del_qgroup_relation_item(trans, quota_root, dst, src);
928 if (err && !ret)
929 ret = err;
930
931 spin_lock(&fs_info->qgroup_lock);
932 del_relation_rb(fs_info, src, dst);
933
934 spin_unlock(&fs_info->qgroup_lock);
935
936 return ret;
937}
938
939int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
940 struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
941{
942 struct btrfs_root *quota_root;
943 struct btrfs_qgroup *qgroup;
944 int ret = 0;
945
946 quota_root = fs_info->quota_root;
947 if (!quota_root)
948 return -EINVAL;
949
950 ret = add_qgroup_item(trans, quota_root, qgroupid);
951
952 spin_lock(&fs_info->qgroup_lock);
953 qgroup = add_qgroup_rb(fs_info, qgroupid);
954 spin_unlock(&fs_info->qgroup_lock);
955
956 if (IS_ERR(qgroup))
957 ret = PTR_ERR(qgroup);
958
959 return ret;
960}
961
962int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
963 struct btrfs_fs_info *fs_info, u64 qgroupid)
964{
965 struct btrfs_root *quota_root;
966 int ret = 0;
967
968 quota_root = fs_info->quota_root;
969 if (!quota_root)
970 return -EINVAL;
971
972 ret = del_qgroup_item(trans, quota_root, qgroupid);
973
974 spin_lock(&fs_info->qgroup_lock);
975 del_qgroup_rb(quota_root->fs_info, qgroupid);
976
977 spin_unlock(&fs_info->qgroup_lock);
978
979 return ret;
980}
981
982int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
983 struct btrfs_fs_info *fs_info, u64 qgroupid,
984 struct btrfs_qgroup_limit *limit)
985{
986 struct btrfs_root *quota_root = fs_info->quota_root;
987 struct btrfs_qgroup *qgroup;
988 int ret = 0;
989
990 if (!quota_root)
991 return -EINVAL;
992
993 ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
994 limit->flags, limit->max_rfer,
995 limit->max_excl, limit->rsv_rfer,
996 limit->rsv_excl);
997 if (ret) {
998 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
999 printk(KERN_INFO "unable to update quota limit for %llu\n",
1000 (unsigned long long)qgroupid);
1001 }
1002
1003 spin_lock(&fs_info->qgroup_lock);
1004
1005 qgroup = find_qgroup_rb(fs_info, qgroupid);
1006 if (!qgroup) {
1007 ret = -ENOENT;
1008 goto unlock;
1009 }
1010 qgroup->lim_flags = limit->flags;
1011 qgroup->max_rfer = limit->max_rfer;
1012 qgroup->max_excl = limit->max_excl;
1013 qgroup->rsv_rfer = limit->rsv_rfer;
1014 qgroup->rsv_excl = limit->rsv_excl;
1015
1016unlock:
1017 spin_unlock(&fs_info->qgroup_lock);
1018
1019 return ret;
1020}
1021
1022static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1023 struct btrfs_qgroup *qgroup)
1024{
1025 if (list_empty(&qgroup->dirty))
1026 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1027}
1028
1029/*
1030 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1031 * the modification into a list that's later used by btrfs_end_transaction to
1032 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1033 */
1034int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1035 struct btrfs_delayed_ref_node *node,
1036 struct btrfs_delayed_extent_op *extent_op)
1037{
1038 struct qgroup_update *u;
1039
1040 BUG_ON(!trans->delayed_ref_elem.seq);
1041 u = kmalloc(sizeof(*u), GFP_NOFS);
1042 if (!u)
1043 return -ENOMEM;
1044
1045 u->node = node;
1046 u->extent_op = extent_op;
1047 list_add_tail(&u->list, &trans->qgroup_ref_list);
1048
1049 return 0;
1050}
1051
1052/*
1053 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1054 * from the fs. First, all roots referencing the extent are searched, and
1055 * then the space is accounted accordingly to the different roots. The
1056 * accounting algorithm works in 3 steps documented inline.
1057 */
1058int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1059 struct btrfs_fs_info *fs_info,
1060 struct btrfs_delayed_ref_node *node,
1061 struct btrfs_delayed_extent_op *extent_op)
1062{
1063 struct btrfs_key ins;
1064 struct btrfs_root *quota_root;
1065 u64 ref_root;
1066 struct btrfs_qgroup *qgroup;
1067 struct ulist_node *unode;
1068 struct ulist *roots = NULL;
1069 struct ulist *tmp = NULL;
1070 struct ulist_iterator uiter;
1071 u64 seq;
1072 int ret = 0;
1073 int sgn;
1074
1075 if (!fs_info->quota_enabled)
1076 return 0;
1077
1078 BUG_ON(!fs_info->quota_root);
1079
1080 ins.objectid = node->bytenr;
1081 ins.offset = node->num_bytes;
1082 ins.type = BTRFS_EXTENT_ITEM_KEY;
1083
1084 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1085 node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1086 struct btrfs_delayed_tree_ref *ref;
1087 ref = btrfs_delayed_node_to_tree_ref(node);
1088 ref_root = ref->root;
1089 } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1090 node->type == BTRFS_SHARED_DATA_REF_KEY) {
1091 struct btrfs_delayed_data_ref *ref;
1092 ref = btrfs_delayed_node_to_data_ref(node);
1093 ref_root = ref->root;
1094 } else {
1095 BUG();
1096 }
1097
1098 if (!is_fstree(ref_root)) {
1099 /*
1100 * non-fs-trees are not being accounted
1101 */
1102 return 0;
1103 }
1104
1105 switch (node->action) {
1106 case BTRFS_ADD_DELAYED_REF:
1107 case BTRFS_ADD_DELAYED_EXTENT:
1108 sgn = 1;
1109 break;
1110 case BTRFS_DROP_DELAYED_REF:
1111 sgn = -1;
1112 break;
1113 case BTRFS_UPDATE_DELAYED_HEAD:
1114 return 0;
1115 default:
1116 BUG();
1117 }
1118
1119 /*
1120 * the delayed ref sequence number we pass depends on the direction of
1121 * the operation. for add operations, we pass (node->seq - 1) to skip
1122 * the delayed ref's current sequence number, because we need the state
1123 * of the tree before the add operation. for delete operations, we pass
1124 * (node->seq) to include the delayed ref's current sequence number,
1125 * because we need the state of the tree after the delete operation.
1126 */
1127 ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1128 sgn > 0 ? node->seq - 1 : node->seq, &roots);
1129 if (ret < 0)
1130 goto out;
1131
1132 spin_lock(&fs_info->qgroup_lock);
1133 quota_root = fs_info->quota_root;
1134 if (!quota_root)
1135 goto unlock;
1136
1137 qgroup = find_qgroup_rb(fs_info, ref_root);
1138 if (!qgroup)
1139 goto unlock;
1140
1141 /*
1142 * step 1: for each old ref, visit all nodes once and inc refcnt
1143 */
1144 tmp = ulist_alloc(GFP_ATOMIC);
1145 if (!tmp) {
1146 ret = -ENOMEM;
1147 goto unlock;
1148 }
1149 seq = fs_info->qgroup_seq;
1150 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1151
1152 ULIST_ITER_INIT(&uiter);
1153 while ((unode = ulist_next(roots, &uiter))) {
1154 struct ulist_node *tmp_unode;
1155 struct ulist_iterator tmp_uiter;
1156 struct btrfs_qgroup *qg;
1157
1158 qg = find_qgroup_rb(fs_info, unode->val);
1159 if (!qg)
1160 continue;
1161
1162 ulist_reinit(tmp);
1163 /* XXX id not needed */
Jan Schmidt995e01b2012-08-13 02:52:38 -06001164 ulist_add(tmp, qg->qgroupid, (u64)(uintptr_t)qg, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001165 ULIST_ITER_INIT(&tmp_uiter);
1166 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1167 struct btrfs_qgroup_list *glist;
1168
Jan Schmidt995e01b2012-08-13 02:52:38 -06001169 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001170 if (qg->refcnt < seq)
1171 qg->refcnt = seq + 1;
1172 else
1173 ++qg->refcnt;
1174
1175 list_for_each_entry(glist, &qg->groups, next_group) {
1176 ulist_add(tmp, glist->group->qgroupid,
Jan Schmidt995e01b2012-08-13 02:52:38 -06001177 (u64)(uintptr_t)glist->group,
Arne Jansenbed92ea2012-06-28 18:03:02 +02001178 GFP_ATOMIC);
1179 }
1180 }
1181 }
1182
1183 /*
1184 * step 2: walk from the new root
1185 */
1186 ulist_reinit(tmp);
Jan Schmidt995e01b2012-08-13 02:52:38 -06001187 ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001188 ULIST_ITER_INIT(&uiter);
1189 while ((unode = ulist_next(tmp, &uiter))) {
1190 struct btrfs_qgroup *qg;
1191 struct btrfs_qgroup_list *glist;
1192
Jan Schmidt995e01b2012-08-13 02:52:38 -06001193 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001194 if (qg->refcnt < seq) {
1195 /* not visited by step 1 */
1196 qg->rfer += sgn * node->num_bytes;
1197 qg->rfer_cmpr += sgn * node->num_bytes;
1198 if (roots->nnodes == 0) {
1199 qg->excl += sgn * node->num_bytes;
1200 qg->excl_cmpr += sgn * node->num_bytes;
1201 }
1202 qgroup_dirty(fs_info, qg);
1203 }
1204 WARN_ON(qg->tag >= seq);
1205 qg->tag = seq;
1206
1207 list_for_each_entry(glist, &qg->groups, next_group) {
1208 ulist_add(tmp, glist->group->qgroupid,
Jan Schmidt995e01b2012-08-13 02:52:38 -06001209 (uintptr_t)glist->group, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001210 }
1211 }
1212
1213 /*
1214 * step 3: walk again from old refs
1215 */
1216 ULIST_ITER_INIT(&uiter);
1217 while ((unode = ulist_next(roots, &uiter))) {
1218 struct btrfs_qgroup *qg;
1219 struct ulist_node *tmp_unode;
1220 struct ulist_iterator tmp_uiter;
1221
1222 qg = find_qgroup_rb(fs_info, unode->val);
1223 if (!qg)
1224 continue;
1225
1226 ulist_reinit(tmp);
Jan Schmidt995e01b2012-08-13 02:52:38 -06001227 ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001228 ULIST_ITER_INIT(&tmp_uiter);
1229 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1230 struct btrfs_qgroup_list *glist;
1231
Jan Schmidt995e01b2012-08-13 02:52:38 -06001232 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001233 if (qg->tag == seq)
1234 continue;
1235
1236 if (qg->refcnt - seq == roots->nnodes) {
1237 qg->excl -= sgn * node->num_bytes;
1238 qg->excl_cmpr -= sgn * node->num_bytes;
1239 qgroup_dirty(fs_info, qg);
1240 }
1241
1242 list_for_each_entry(glist, &qg->groups, next_group) {
1243 ulist_add(tmp, glist->group->qgroupid,
Jan Schmidt995e01b2012-08-13 02:52:38 -06001244 (uintptr_t)glist->group,
Arne Jansenbed92ea2012-06-28 18:03:02 +02001245 GFP_ATOMIC);
1246 }
1247 }
1248 }
1249 ret = 0;
1250unlock:
1251 spin_unlock(&fs_info->qgroup_lock);
1252out:
1253 ulist_free(roots);
1254 ulist_free(tmp);
1255
1256 return ret;
1257}
1258
1259/*
1260 * called from commit_transaction. Writes all changed qgroups to disk.
1261 */
1262int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1263 struct btrfs_fs_info *fs_info)
1264{
1265 struct btrfs_root *quota_root = fs_info->quota_root;
1266 int ret = 0;
1267
1268 if (!quota_root)
1269 goto out;
1270
1271 fs_info->quota_enabled = fs_info->pending_quota_state;
1272
1273 spin_lock(&fs_info->qgroup_lock);
1274 while (!list_empty(&fs_info->dirty_qgroups)) {
1275 struct btrfs_qgroup *qgroup;
1276 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1277 struct btrfs_qgroup, dirty);
1278 list_del_init(&qgroup->dirty);
1279 spin_unlock(&fs_info->qgroup_lock);
1280 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1281 if (ret)
1282 fs_info->qgroup_flags |=
1283 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1284 spin_lock(&fs_info->qgroup_lock);
1285 }
1286 if (fs_info->quota_enabled)
1287 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1288 else
1289 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1290 spin_unlock(&fs_info->qgroup_lock);
1291
1292 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1293 if (ret)
1294 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1295
1296out:
1297
1298 return ret;
1299}
1300
1301/*
1302 * copy the acounting information between qgroups. This is necessary when a
1303 * snapshot or a subvolume is created
1304 */
1305int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1306 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1307 struct btrfs_qgroup_inherit *inherit)
1308{
1309 int ret = 0;
1310 int i;
1311 u64 *i_qgroups;
1312 struct btrfs_root *quota_root = fs_info->quota_root;
1313 struct btrfs_qgroup *srcgroup;
1314 struct btrfs_qgroup *dstgroup;
1315 u32 level_size = 0;
1316
1317 if (!fs_info->quota_enabled)
1318 return 0;
1319
1320 if (!quota_root)
1321 return -EINVAL;
1322
1323 /*
1324 * create a tracking group for the subvol itself
1325 */
1326 ret = add_qgroup_item(trans, quota_root, objectid);
1327 if (ret)
1328 goto out;
1329
1330 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1331 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1332 inherit->lim.flags,
1333 inherit->lim.max_rfer,
1334 inherit->lim.max_excl,
1335 inherit->lim.rsv_rfer,
1336 inherit->lim.rsv_excl);
1337 if (ret)
1338 goto out;
1339 }
1340
1341 if (srcid) {
1342 struct btrfs_root *srcroot;
1343 struct btrfs_key srckey;
1344 int srcroot_level;
1345
1346 srckey.objectid = srcid;
1347 srckey.type = BTRFS_ROOT_ITEM_KEY;
1348 srckey.offset = (u64)-1;
1349 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1350 if (IS_ERR(srcroot)) {
1351 ret = PTR_ERR(srcroot);
1352 goto out;
1353 }
1354
1355 rcu_read_lock();
1356 srcroot_level = btrfs_header_level(srcroot->node);
1357 level_size = btrfs_level_size(srcroot, srcroot_level);
1358 rcu_read_unlock();
1359 }
1360
1361 /*
1362 * add qgroup to all inherited groups
1363 */
1364 if (inherit) {
1365 i_qgroups = (u64 *)(inherit + 1);
1366 for (i = 0; i < inherit->num_qgroups; ++i) {
1367 ret = add_qgroup_relation_item(trans, quota_root,
1368 objectid, *i_qgroups);
1369 if (ret)
1370 goto out;
1371 ret = add_qgroup_relation_item(trans, quota_root,
1372 *i_qgroups, objectid);
1373 if (ret)
1374 goto out;
1375 ++i_qgroups;
1376 }
1377 }
1378
1379
1380 spin_lock(&fs_info->qgroup_lock);
1381
1382 dstgroup = add_qgroup_rb(fs_info, objectid);
Dan Carpenter57a5a882012-07-30 02:15:43 -06001383 if (IS_ERR(dstgroup)) {
1384 ret = PTR_ERR(dstgroup);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001385 goto unlock;
Dan Carpenter57a5a882012-07-30 02:15:43 -06001386 }
Arne Jansenbed92ea2012-06-28 18:03:02 +02001387
1388 if (srcid) {
1389 srcgroup = find_qgroup_rb(fs_info, srcid);
Chris Masonf3a87f12012-09-14 20:06:30 -04001390 if (!srcgroup)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001391 goto unlock;
1392 dstgroup->rfer = srcgroup->rfer - level_size;
1393 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1394 srcgroup->excl = level_size;
1395 srcgroup->excl_cmpr = level_size;
1396 qgroup_dirty(fs_info, dstgroup);
1397 qgroup_dirty(fs_info, srcgroup);
1398 }
1399
Chris Masonf3a87f12012-09-14 20:06:30 -04001400 if (!inherit)
Arne Jansenbed92ea2012-06-28 18:03:02 +02001401 goto unlock;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001402
1403 i_qgroups = (u64 *)(inherit + 1);
1404 for (i = 0; i < inherit->num_qgroups; ++i) {
1405 ret = add_relation_rb(quota_root->fs_info, objectid,
1406 *i_qgroups);
1407 if (ret)
1408 goto unlock;
1409 ++i_qgroups;
1410 }
1411
1412 for (i = 0; i < inherit->num_ref_copies; ++i) {
1413 struct btrfs_qgroup *src;
1414 struct btrfs_qgroup *dst;
1415
1416 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1417 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1418
1419 if (!src || !dst) {
1420 ret = -EINVAL;
1421 goto unlock;
1422 }
1423
1424 dst->rfer = src->rfer - level_size;
1425 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1426 i_qgroups += 2;
1427 }
1428 for (i = 0; i < inherit->num_excl_copies; ++i) {
1429 struct btrfs_qgroup *src;
1430 struct btrfs_qgroup *dst;
1431
1432 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1433 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1434
1435 if (!src || !dst) {
1436 ret = -EINVAL;
1437 goto unlock;
1438 }
1439
1440 dst->excl = src->excl + level_size;
1441 dst->excl_cmpr = src->excl_cmpr + level_size;
1442 i_qgroups += 2;
1443 }
1444
1445unlock:
1446 spin_unlock(&fs_info->qgroup_lock);
1447out:
1448 return ret;
1449}
1450
1451/*
1452 * reserve some space for a qgroup and all its parents. The reservation takes
1453 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1454 * accounting. If not enough space is available, EDQUOT is returned.
1455 * We assume that the requested space is new for all qgroups.
1456 */
1457int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1458{
1459 struct btrfs_root *quota_root;
1460 struct btrfs_qgroup *qgroup;
1461 struct btrfs_fs_info *fs_info = root->fs_info;
1462 u64 ref_root = root->root_key.objectid;
1463 int ret = 0;
1464 struct ulist *ulist = NULL;
1465 struct ulist_node *unode;
1466 struct ulist_iterator uiter;
1467
1468 if (!is_fstree(ref_root))
1469 return 0;
1470
1471 if (num_bytes == 0)
1472 return 0;
1473
1474 spin_lock(&fs_info->qgroup_lock);
1475 quota_root = fs_info->quota_root;
1476 if (!quota_root)
1477 goto out;
1478
1479 qgroup = find_qgroup_rb(fs_info, ref_root);
1480 if (!qgroup)
1481 goto out;
1482
1483 /*
1484 * in a first step, we check all affected qgroups if any limits would
1485 * be exceeded
1486 */
1487 ulist = ulist_alloc(GFP_ATOMIC);
Tsutomu Itoh3d6b5c32012-09-06 00:18:10 -06001488 if (!ulist) {
1489 ret = -ENOMEM;
1490 goto out;
1491 }
Jan Schmidt995e01b2012-08-13 02:52:38 -06001492 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001493 ULIST_ITER_INIT(&uiter);
1494 while ((unode = ulist_next(ulist, &uiter))) {
1495 struct btrfs_qgroup *qg;
1496 struct btrfs_qgroup_list *glist;
1497
Jan Schmidt995e01b2012-08-13 02:52:38 -06001498 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001499
1500 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1501 qg->reserved + qg->rfer + num_bytes >
1502 qg->max_rfer)
1503 ret = -EDQUOT;
1504
1505 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1506 qg->reserved + qg->excl + num_bytes >
1507 qg->max_excl)
1508 ret = -EDQUOT;
1509
1510 list_for_each_entry(glist, &qg->groups, next_group) {
1511 ulist_add(ulist, glist->group->qgroupid,
Jan Schmidt995e01b2012-08-13 02:52:38 -06001512 (uintptr_t)glist->group, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001513 }
1514 }
1515 if (ret)
1516 goto out;
1517
1518 /*
1519 * no limits exceeded, now record the reservation into all qgroups
1520 */
1521 ULIST_ITER_INIT(&uiter);
1522 while ((unode = ulist_next(ulist, &uiter))) {
1523 struct btrfs_qgroup *qg;
1524
Jan Schmidt995e01b2012-08-13 02:52:38 -06001525 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001526
1527 qg->reserved += num_bytes;
1528 }
1529
1530out:
1531 spin_unlock(&fs_info->qgroup_lock);
1532 ulist_free(ulist);
1533
1534 return ret;
1535}
1536
1537void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1538{
1539 struct btrfs_root *quota_root;
1540 struct btrfs_qgroup *qgroup;
1541 struct btrfs_fs_info *fs_info = root->fs_info;
1542 struct ulist *ulist = NULL;
1543 struct ulist_node *unode;
1544 struct ulist_iterator uiter;
1545 u64 ref_root = root->root_key.objectid;
1546
1547 if (!is_fstree(ref_root))
1548 return;
1549
1550 if (num_bytes == 0)
1551 return;
1552
1553 spin_lock(&fs_info->qgroup_lock);
1554
1555 quota_root = fs_info->quota_root;
1556 if (!quota_root)
1557 goto out;
1558
1559 qgroup = find_qgroup_rb(fs_info, ref_root);
1560 if (!qgroup)
1561 goto out;
1562
1563 ulist = ulist_alloc(GFP_ATOMIC);
Tsutomu Itoh3d6b5c32012-09-06 00:18:10 -06001564 if (!ulist) {
1565 btrfs_std_error(fs_info, -ENOMEM);
1566 goto out;
1567 }
Jan Schmidt995e01b2012-08-13 02:52:38 -06001568 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001569 ULIST_ITER_INIT(&uiter);
1570 while ((unode = ulist_next(ulist, &uiter))) {
1571 struct btrfs_qgroup *qg;
1572 struct btrfs_qgroup_list *glist;
1573
Jan Schmidt995e01b2012-08-13 02:52:38 -06001574 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
Arne Jansenbed92ea2012-06-28 18:03:02 +02001575
1576 qg->reserved -= num_bytes;
1577
1578 list_for_each_entry(glist, &qg->groups, next_group) {
1579 ulist_add(ulist, glist->group->qgroupid,
Jan Schmidt995e01b2012-08-13 02:52:38 -06001580 (uintptr_t)glist->group, GFP_ATOMIC);
Arne Jansenbed92ea2012-06-28 18:03:02 +02001581 }
1582 }
1583
1584out:
1585 spin_unlock(&fs_info->qgroup_lock);
1586 ulist_free(ulist);
1587}
1588
1589void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1590{
1591 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1592 return;
1593 printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1594 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1595 trans->delayed_ref_elem.seq);
1596 BUG();
1597}