projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: log ram bytes properly
[deliverable/linux.git] / fs / btrfs / qgroup.c
CommitLineData
bed92eae
AJ		 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>
55e301fd 26#include <linux/btrfs.h>
bed92eae
AJ		 27
28#include "ctree.h"
29#include "transaction.h"
30#include "disk-io.h"
31#include "locking.h"
32#include "ulist.h"
bed92eae
AJ		 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);
ff24858c
AJ		 382 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 }
bed92eae
AJ		 389 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
bed92eae
AJ		 423 while (!list_empty(&qgroup->groups)) {
424 list = list_first_entry(&qgroup->groups,
425 struct btrfs_qgroup_list,
426 next_group);
427 list_del(&list->next_group);
428 list_del(&list->next_member);
429 kfree(list);
430 }
431
432 while (!list_empty(&qgroup->members)) {
433 list = list_first_entry(&qgroup->members,
434 struct btrfs_qgroup_list,
435 next_member);
436 list_del(&list->next_group);
437 list_del(&list->next_member);
438 kfree(list);
439 }
440 kfree(qgroup);
441 }
442}
443
444static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
445 struct btrfs_root *quota_root,
446 u64 src, u64 dst)
447{
448 int ret;
449 struct btrfs_path *path;
450 struct btrfs_key key;
451
452 path = btrfs_alloc_path();
453 if (!path)
454 return -ENOMEM;
455
456 key.objectid = src;
457 key.type = BTRFS_QGROUP_RELATION_KEY;
458 key.offset = dst;
459
460 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
461
462 btrfs_mark_buffer_dirty(path->nodes[0]);
463
464 btrfs_free_path(path);
465 return ret;
466}
467
468static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
469 struct btrfs_root *quota_root,
470 u64 src, u64 dst)
471{
472 int ret;
473 struct btrfs_path *path;
474 struct btrfs_key key;
475
476 path = btrfs_alloc_path();
477 if (!path)
478 return -ENOMEM;
479
480 key.objectid = src;
481 key.type = BTRFS_QGROUP_RELATION_KEY;
482 key.offset = dst;
483
484 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
485 if (ret < 0)
486 goto out;
487
488 if (ret > 0) {
489 ret = -ENOENT;
490 goto out;
491 }
492
493 ret = btrfs_del_item(trans, quota_root, path);
494out:
495 btrfs_free_path(path);
496 return ret;
497}
498
499static int add_qgroup_item(struct btrfs_trans_handle *trans,
500 struct btrfs_root *quota_root, u64 qgroupid)
501{
502 int ret;
503 struct btrfs_path *path;
504 struct btrfs_qgroup_info_item *qgroup_info;
505 struct btrfs_qgroup_limit_item *qgroup_limit;
506 struct extent_buffer *leaf;
507 struct btrfs_key key;
508
509 path = btrfs_alloc_path();
510 if (!path)
511 return -ENOMEM;
512
513 key.objectid = 0;
514 key.type = BTRFS_QGROUP_INFO_KEY;
515 key.offset = qgroupid;
516
517 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
518 sizeof(*qgroup_info));
519 if (ret)
520 goto out;
521
522 leaf = path->nodes[0];
523 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
524 struct btrfs_qgroup_info_item);
525 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
526 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
527 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
528 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
529 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
530
531 btrfs_mark_buffer_dirty(leaf);
532
533 btrfs_release_path(path);
534
535 key.type = BTRFS_QGROUP_LIMIT_KEY;
536 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
537 sizeof(*qgroup_limit));
538 if (ret)
539 goto out;
540
541 leaf = path->nodes[0];
542 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
543 struct btrfs_qgroup_limit_item);
544 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
545 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
546 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
547 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
548 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
549
550 btrfs_mark_buffer_dirty(leaf);
551
552 ret = 0;
553out:
554 btrfs_free_path(path);
555 return ret;
556}
557
558static int del_qgroup_item(struct btrfs_trans_handle *trans,
559 struct btrfs_root *quota_root, u64 qgroupid)
560{
561 int ret;
562 struct btrfs_path *path;
563 struct btrfs_key key;
564
565 path = btrfs_alloc_path();
566 if (!path)
567 return -ENOMEM;
568
569 key.objectid = 0;
570 key.type = BTRFS_QGROUP_INFO_KEY;
571 key.offset = qgroupid;
572 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
573 if (ret < 0)
574 goto out;
575
576 if (ret > 0) {
577 ret = -ENOENT;
578 goto out;
579 }
580
581 ret = btrfs_del_item(trans, quota_root, path);
582 if (ret)
583 goto out;
584
585 btrfs_release_path(path);
586
587 key.type = BTRFS_QGROUP_LIMIT_KEY;
588 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
589 if (ret < 0)
590 goto out;
591
592 if (ret > 0) {
593 ret = -ENOENT;
594 goto out;
595 }
596
597 ret = btrfs_del_item(trans, quota_root, path);
598
599out:
600 btrfs_free_path(path);
601 return ret;
602}
603
604static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
605 struct btrfs_root *root, u64 qgroupid,
606 u64 flags, u64 max_rfer, u64 max_excl,
607 u64 rsv_rfer, u64 rsv_excl)
608{
609 struct btrfs_path *path;
610 struct btrfs_key key;
611 struct extent_buffer *l;
612 struct btrfs_qgroup_limit_item *qgroup_limit;
613 int ret;
614 int slot;
615
616 key.objectid = 0;
617 key.type = BTRFS_QGROUP_LIMIT_KEY;
618 key.offset = qgroupid;
619
620 path = btrfs_alloc_path();
84cbe2f7
WS		 621 if (!path)
622 return -ENOMEM;
623
bed92eae
AJ		 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();
84cbe2f7
WS		 664 if (!path)
665 return -ENOMEM;
666
bed92eae
AJ		 667 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
668 if (ret > 0)
669 ret = -ENOENT;
670
671 if (ret)
672 goto out;
673
674 l = path->nodes[0];
675 slot = path->slots[0];
676 qgroup_info = btrfs_item_ptr(l, path->slots[0],
677 struct btrfs_qgroup_info_item);
678 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
679 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
680 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
681 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
682 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
683
684 btrfs_mark_buffer_dirty(l);
685
686out:
687 btrfs_free_path(path);
688 return ret;
689}
690
691static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
692 struct btrfs_fs_info *fs_info,
693 struct btrfs_root *root)
694{
695 struct btrfs_path *path;
696 struct btrfs_key key;
697 struct extent_buffer *l;
698 struct btrfs_qgroup_status_item *ptr;
699 int ret;
700 int slot;
701
702 key.objectid = 0;
703 key.type = BTRFS_QGROUP_STATUS_KEY;
704 key.offset = 0;
705
706 path = btrfs_alloc_path();
84cbe2f7
WS		 707 if (!path)
708 return -ENOMEM;
709
bed92eae
AJ		 710 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
711 if (ret > 0)
712 ret = -ENOENT;
713
714 if (ret)
715 goto out;
716
717 l = path->nodes[0];
718 slot = path->slots[0];
719 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
720 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
721 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
722 /* XXX scan */
723
724 btrfs_mark_buffer_dirty(l);
725
726out:
727 btrfs_free_path(path);
728 return ret;
729}
730
731/*
732 * called with qgroup_lock held
733 */
734static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root)
736{
737 struct btrfs_path *path;
738 struct btrfs_key key;
06b3a860 739 struct extent_buffer *leaf = NULL;
bed92eae 740 int ret;
06b3a860 741 int nr = 0;
bed92eae 742
bed92eae
AJ		 743 path = btrfs_alloc_path();
744 if (!path)
745 return -ENOMEM;
746
06b3a860
WS		 747 path->leave_spinning = 1;
748
749 key.objectid = 0;
750 key.offset = 0;
751 key.type = 0;
bed92eae 752
06b3a860 753 while (1) {
bed92eae 754 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
06b3a860
WS		 755 if (ret < 0)
756 goto out;
757 leaf = path->nodes[0];
758 nr = btrfs_header_nritems(leaf);
759 if (!nr)
bed92eae 760 break;
06b3a860
WS		 761 /*
762 * delete the leaf one by one
763 * since the whole tree is going
764 * to be deleted.
765 */
766 path->slots[0] = 0;
767 ret = btrfs_del_items(trans, root, path, 0, nr);
bed92eae
AJ		 768 if (ret)
769 goto out;
06b3a860 770
bed92eae
AJ		 771 btrfs_release_path(path);
772 }
773 ret = 0;
774out:
775 root->fs_info->pending_quota_state = 0;
776 btrfs_free_path(path);
777 return ret;
778}
779
780int btrfs_quota_enable(struct btrfs_trans_handle *trans,
781 struct btrfs_fs_info *fs_info)
782{
783 struct btrfs_root *quota_root;
784 struct btrfs_path *path = NULL;
785 struct btrfs_qgroup_status_item *ptr;
786 struct extent_buffer *leaf;
787 struct btrfs_key key;
788 int ret = 0;
789
790 spin_lock(&fs_info->qgroup_lock);
791 if (fs_info->quota_root) {
792 fs_info->pending_quota_state = 1;
793 spin_unlock(&fs_info->qgroup_lock);
794 goto out;
795 }
796 spin_unlock(&fs_info->qgroup_lock);
797
798 /*
799 * initially create the quota tree
800 */
801 quota_root = btrfs_create_tree(trans, fs_info,
802 BTRFS_QUOTA_TREE_OBJECTID);
803 if (IS_ERR(quota_root)) {
804 ret = PTR_ERR(quota_root);
805 goto out;
806 }
807
808 path = btrfs_alloc_path();
5b7ff5b3
TI		 809 if (!path) {
810 ret = -ENOMEM;
811 goto out_free_root;
812 }
bed92eae
AJ		 813
814 key.objectid = 0;
815 key.type = BTRFS_QGROUP_STATUS_KEY;
816 key.offset = 0;
817
818 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
819 sizeof(*ptr));
820 if (ret)
5b7ff5b3 821 goto out_free_path;
bed92eae
AJ		 822
823 leaf = path->nodes[0];
824 ptr = btrfs_item_ptr(leaf, path->slots[0],
825 struct btrfs_qgroup_status_item);
826 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
827 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
828 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
829 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
830 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
831 btrfs_set_qgroup_status_scan(leaf, ptr, 0);
832
833 btrfs_mark_buffer_dirty(leaf);
834
835 spin_lock(&fs_info->qgroup_lock);
836 fs_info->quota_root = quota_root;
837 fs_info->pending_quota_state = 1;
838 spin_unlock(&fs_info->qgroup_lock);
5b7ff5b3 839out_free_path:
bed92eae 840 btrfs_free_path(path);
5b7ff5b3
TI		 841out_free_root:
842 if (ret) {
843 free_extent_buffer(quota_root->node);
844 free_extent_buffer(quota_root->commit_root);
845 kfree(quota_root);
846 }
847out:
bed92eae
AJ		 848 return ret;
849}
850
851int btrfs_quota_disable(struct btrfs_trans_handle *trans,
852 struct btrfs_fs_info *fs_info)
853{
854 struct btrfs_root *tree_root = fs_info->tree_root;
855 struct btrfs_root *quota_root;
856 int ret = 0;
857
858 spin_lock(&fs_info->qgroup_lock);
683cebda
WS		 859 if (!fs_info->quota_root) {
860 spin_unlock(&fs_info->qgroup_lock);
861 return 0;
862 }
bed92eae
AJ		 863 fs_info->quota_enabled = 0;
864 fs_info->pending_quota_state = 0;
865 quota_root = fs_info->quota_root;
866 fs_info->quota_root = NULL;
867 btrfs_free_qgroup_config(fs_info);
868 spin_unlock(&fs_info->qgroup_lock);
869
870 if (!quota_root)
871 return -EINVAL;
872
873 ret = btrfs_clean_quota_tree(trans, quota_root);
874 if (ret)
875 goto out;
876
877 ret = btrfs_del_root(trans, tree_root, &quota_root->root_key);
878 if (ret)
879 goto out;
880
881 list_del(&quota_root->dirty_list);
882
883 btrfs_tree_lock(quota_root->node);
884 clean_tree_block(trans, tree_root, quota_root->node);
885 btrfs_tree_unlock(quota_root->node);
886 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
887
888 free_extent_buffer(quota_root->node);
889 free_extent_buffer(quota_root->commit_root);
890 kfree(quota_root);
891out:
892 return ret;
893}
894
895int btrfs_quota_rescan(struct btrfs_fs_info *fs_info)
896{
897 /* FIXME */
898 return 0;
899}
900
901int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
902 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
903{
904 struct btrfs_root *quota_root;
905 int ret = 0;
906
907 quota_root = fs_info->quota_root;
908 if (!quota_root)
909 return -EINVAL;
910
911 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
912 if (ret)
913 return ret;
914
915 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
916 if (ret) {
917 del_qgroup_relation_item(trans, quota_root, src, dst);
918 return ret;
919 }
920
921 spin_lock(&fs_info->qgroup_lock);
922 ret = add_relation_rb(quota_root->fs_info, src, dst);
923 spin_unlock(&fs_info->qgroup_lock);
924
925 return ret;
926}
927
928int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
929 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
930{
931 struct btrfs_root *quota_root;
932 int ret = 0;
933 int err;
934
935 quota_root = fs_info->quota_root;
936 if (!quota_root)
937 return -EINVAL;
938
939 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
940 err = del_qgroup_relation_item(trans, quota_root, dst, src);
941 if (err && !ret)
942 ret = err;
943
944 spin_lock(&fs_info->qgroup_lock);
945 del_relation_rb(fs_info, src, dst);
946
947 spin_unlock(&fs_info->qgroup_lock);
948
949 return ret;
950}
951
952int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
953 struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
954{
955 struct btrfs_root *quota_root;
956 struct btrfs_qgroup *qgroup;
957 int ret = 0;
958
959 quota_root = fs_info->quota_root;
960 if (!quota_root)
961 return -EINVAL;
962
963 ret = add_qgroup_item(trans, quota_root, qgroupid);
964
965 spin_lock(&fs_info->qgroup_lock);
966 qgroup = add_qgroup_rb(fs_info, qgroupid);
967 spin_unlock(&fs_info->qgroup_lock);
968
969 if (IS_ERR(qgroup))
970 ret = PTR_ERR(qgroup);
971
972 return ret;
973}
974
975int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
976 struct btrfs_fs_info *fs_info, u64 qgroupid)
977{
978 struct btrfs_root *quota_root;
2cf68703 979 struct btrfs_qgroup *qgroup;
bed92eae
AJ		 980 int ret = 0;
981
982 quota_root = fs_info->quota_root;
983 if (!quota_root)
984 return -EINVAL;
985
2cf68703
AJ		 986 /* check if there are no relations to this qgroup */
987 spin_lock(&fs_info->qgroup_lock);
988 qgroup = find_qgroup_rb(fs_info, qgroupid);
989 if (qgroup) {
990 if (!list_empty(&qgroup->groups) || !list_empty(&qgroup->members)) {
991 spin_unlock(&fs_info->qgroup_lock);
992 return -EBUSY;
993 }
994 }
995 spin_unlock(&fs_info->qgroup_lock);
996
bed92eae
AJ		 997 ret = del_qgroup_item(trans, quota_root, qgroupid);
998
999 spin_lock(&fs_info->qgroup_lock);
1000 del_qgroup_rb(quota_root->fs_info, qgroupid);
bed92eae
AJ		 1001 spin_unlock(&fs_info->qgroup_lock);
1002
1003 return ret;
1004}
1005
1006int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1007 struct btrfs_fs_info *fs_info, u64 qgroupid,
1008 struct btrfs_qgroup_limit *limit)
1009{
1010 struct btrfs_root *quota_root = fs_info->quota_root;
1011 struct btrfs_qgroup *qgroup;
1012 int ret = 0;
1013
1014 if (!quota_root)
1015 return -EINVAL;
1016
1017 ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1018 limit->flags, limit->max_rfer,
1019 limit->max_excl, limit->rsv_rfer,
1020 limit->rsv_excl);
1021 if (ret) {
1022 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1023 printk(KERN_INFO "unable to update quota limit for %llu\n",
1024 (unsigned long long)qgroupid);
1025 }
1026
1027 spin_lock(&fs_info->qgroup_lock);
1028
1029 qgroup = find_qgroup_rb(fs_info, qgroupid);
1030 if (!qgroup) {
1031 ret = -ENOENT;
1032 goto unlock;
1033 }
1034 qgroup->lim_flags = limit->flags;
1035 qgroup->max_rfer = limit->max_rfer;
1036 qgroup->max_excl = limit->max_excl;
1037 qgroup->rsv_rfer = limit->rsv_rfer;
1038 qgroup->rsv_excl = limit->rsv_excl;
1039
1040unlock:
1041 spin_unlock(&fs_info->qgroup_lock);
1042
1043 return ret;
1044}
1045
1046static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1047 struct btrfs_qgroup *qgroup)
1048{
1049 if (list_empty(&qgroup->dirty))
1050 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1051}
1052
1053/*
1054 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1055 * the modification into a list that's later used by btrfs_end_transaction to
1056 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1057 */
1058int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1059 struct btrfs_delayed_ref_node *node,
1060 struct btrfs_delayed_extent_op *extent_op)
1061{
1062 struct qgroup_update *u;
1063
1064 BUG_ON(!trans->delayed_ref_elem.seq);
1065 u = kmalloc(sizeof(*u), GFP_NOFS);
1066 if (!u)
1067 return -ENOMEM;
1068
1069 u->node = node;
1070 u->extent_op = extent_op;
1071 list_add_tail(&u->list, &trans->qgroup_ref_list);
1072
1073 return 0;
1074}
1075
1076/*
1077 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1078 * from the fs. First, all roots referencing the extent are searched, and
1079 * then the space is accounted accordingly to the different roots. The
1080 * accounting algorithm works in 3 steps documented inline.
1081 */
1082int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1083 struct btrfs_fs_info *fs_info,
1084 struct btrfs_delayed_ref_node *node,
1085 struct btrfs_delayed_extent_op *extent_op)
1086{
1087 struct btrfs_key ins;
1088 struct btrfs_root *quota_root;
1089 u64 ref_root;
1090 struct btrfs_qgroup *qgroup;
1091 struct ulist_node *unode;
1092 struct ulist *roots = NULL;
1093 struct ulist *tmp = NULL;
1094 struct ulist_iterator uiter;
1095 u64 seq;
1096 int ret = 0;
1097 int sgn;
1098
1099 if (!fs_info->quota_enabled)
1100 return 0;
1101
1102 BUG_ON(!fs_info->quota_root);
1103
1104 ins.objectid = node->bytenr;
1105 ins.offset = node->num_bytes;
1106 ins.type = BTRFS_EXTENT_ITEM_KEY;
1107
1108 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1109 node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1110 struct btrfs_delayed_tree_ref *ref;
1111 ref = btrfs_delayed_node_to_tree_ref(node);
1112 ref_root = ref->root;
1113 } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1114 node->type == BTRFS_SHARED_DATA_REF_KEY) {
1115 struct btrfs_delayed_data_ref *ref;
1116 ref = btrfs_delayed_node_to_data_ref(node);
1117 ref_root = ref->root;
1118 } else {
1119 BUG();
1120 }
1121
1122 if (!is_fstree(ref_root)) {
1123 /*
1124 * non-fs-trees are not being accounted
1125 */
1126 return 0;
1127 }
1128
1129 switch (node->action) {
1130 case BTRFS_ADD_DELAYED_REF:
1131 case BTRFS_ADD_DELAYED_EXTENT:
1132 sgn = 1;
1133 break;
1134 case BTRFS_DROP_DELAYED_REF:
1135 sgn = -1;
1136 break;
1137 case BTRFS_UPDATE_DELAYED_HEAD:
1138 return 0;
1139 default:
1140 BUG();
1141 }
1142
1143 /*
1144 * the delayed ref sequence number we pass depends on the direction of
1145 * the operation. for add operations, we pass (node->seq - 1) to skip
1146 * the delayed ref's current sequence number, because we need the state
1147 * of the tree before the add operation. for delete operations, we pass
1148 * (node->seq) to include the delayed ref's current sequence number,
1149 * because we need the state of the tree after the delete operation.
1150 */
1151 ret = btrfs_find_all_roots(trans, fs_info, node->bytenr,
1152 sgn > 0 ? node->seq - 1 : node->seq, &roots);
1153 if (ret < 0)
a7975026 1154 return ret;
bed92eae
AJ		 1155
1156 spin_lock(&fs_info->qgroup_lock);
1157 quota_root = fs_info->quota_root;
1158 if (!quota_root)
1159 goto unlock;
1160
1161 qgroup = find_qgroup_rb(fs_info, ref_root);
1162 if (!qgroup)
1163 goto unlock;
1164
1165 /*
1166 * step 1: for each old ref, visit all nodes once and inc refcnt
1167 */
1168 tmp = ulist_alloc(GFP_ATOMIC);
1169 if (!tmp) {
1170 ret = -ENOMEM;
1171 goto unlock;
1172 }
1173 seq = fs_info->qgroup_seq;
1174 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1175
1176 ULIST_ITER_INIT(&uiter);
1177 while ((unode = ulist_next(roots, &uiter))) {
1178 struct ulist_node *tmp_unode;
1179 struct ulist_iterator tmp_uiter;
1180 struct btrfs_qgroup *qg;
1181
1182 qg = find_qgroup_rb(fs_info, unode->val);
1183 if (!qg)
1184 continue;
1185
1186 ulist_reinit(tmp);
1187 /* XXX id not needed */
995e01b7 1188 ulist_add(tmp, qg->qgroupid, (u64)(uintptr_t)qg, GFP_ATOMIC);
bed92eae
AJ		 1189 ULIST_ITER_INIT(&tmp_uiter);
1190 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1191 struct btrfs_qgroup_list *glist;
1192
995e01b7 1193 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
bed92eae
AJ		 1194 if (qg->refcnt < seq)
1195 qg->refcnt = seq + 1;
1196 else
1197 ++qg->refcnt;
1198
1199 list_for_each_entry(glist, &qg->groups, next_group) {
1200 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1201 (u64)(uintptr_t)glist->group,
bed92eae
AJ		 1202 GFP_ATOMIC);
1203 }
1204 }
1205 }
1206
1207 /*
1208 * step 2: walk from the new root
1209 */
1210 ulist_reinit(tmp);
995e01b7 1211 ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1212 ULIST_ITER_INIT(&uiter);
1213 while ((unode = ulist_next(tmp, &uiter))) {
1214 struct btrfs_qgroup *qg;
1215 struct btrfs_qgroup_list *glist;
1216
995e01b7 1217 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1218 if (qg->refcnt < seq) {
1219 /* not visited by step 1 */
1220 qg->rfer += sgn * node->num_bytes;
1221 qg->rfer_cmpr += sgn * node->num_bytes;
1222 if (roots->nnodes == 0) {
1223 qg->excl += sgn * node->num_bytes;
1224 qg->excl_cmpr += sgn * node->num_bytes;
1225 }
1226 qgroup_dirty(fs_info, qg);
1227 }
1228 WARN_ON(qg->tag >= seq);
1229 qg->tag = seq;
1230
1231 list_for_each_entry(glist, &qg->groups, next_group) {
1232 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1233 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1234 }
1235 }
1236
1237 /*
1238 * step 3: walk again from old refs
1239 */
1240 ULIST_ITER_INIT(&uiter);
1241 while ((unode = ulist_next(roots, &uiter))) {
1242 struct btrfs_qgroup *qg;
1243 struct ulist_node *tmp_unode;
1244 struct ulist_iterator tmp_uiter;
1245
1246 qg = find_qgroup_rb(fs_info, unode->val);
1247 if (!qg)
1248 continue;
1249
1250 ulist_reinit(tmp);
995e01b7 1251 ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
bed92eae
AJ		 1252 ULIST_ITER_INIT(&tmp_uiter);
1253 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1254 struct btrfs_qgroup_list *glist;
1255
995e01b7 1256 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
bed92eae
AJ		 1257 if (qg->tag == seq)
1258 continue;
1259
1260 if (qg->refcnt - seq == roots->nnodes) {
1261 qg->excl -= sgn * node->num_bytes;
1262 qg->excl_cmpr -= sgn * node->num_bytes;
1263 qgroup_dirty(fs_info, qg);
1264 }
1265
1266 list_for_each_entry(glist, &qg->groups, next_group) {
1267 ulist_add(tmp, glist->group->qgroupid,
995e01b7 1268 (uintptr_t)glist->group,
bed92eae
AJ		 1269 GFP_ATOMIC);
1270 }
1271 }
1272 }
1273 ret = 0;
1274unlock:
1275 spin_unlock(&fs_info->qgroup_lock);
bed92eae
AJ		 1276 ulist_free(roots);
1277 ulist_free(tmp);
1278
1279 return ret;
1280}
1281
1282/*
1283 * called from commit_transaction. Writes all changed qgroups to disk.
1284 */
1285int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1286 struct btrfs_fs_info *fs_info)
1287{
1288 struct btrfs_root *quota_root = fs_info->quota_root;
1289 int ret = 0;
1290
1291 if (!quota_root)
1292 goto out;
1293
1294 fs_info->quota_enabled = fs_info->pending_quota_state;
1295
1296 spin_lock(&fs_info->qgroup_lock);
1297 while (!list_empty(&fs_info->dirty_qgroups)) {
1298 struct btrfs_qgroup *qgroup;
1299 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1300 struct btrfs_qgroup, dirty);
1301 list_del_init(&qgroup->dirty);
1302 spin_unlock(&fs_info->qgroup_lock);
1303 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1304 if (ret)
1305 fs_info->qgroup_flags |=
1306 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1307 spin_lock(&fs_info->qgroup_lock);
1308 }
1309 if (fs_info->quota_enabled)
1310 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1311 else
1312 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1313 spin_unlock(&fs_info->qgroup_lock);
1314
1315 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1316 if (ret)
1317 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1318
1319out:
1320
1321 return ret;
1322}
1323
1324/*
1325 * copy the acounting information between qgroups. This is necessary when a
1326 * snapshot or a subvolume is created
1327 */
1328int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1329 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1330 struct btrfs_qgroup_inherit *inherit)
1331{
1332 int ret = 0;
1333 int i;
1334 u64 *i_qgroups;
1335 struct btrfs_root *quota_root = fs_info->quota_root;
1336 struct btrfs_qgroup *srcgroup;
1337 struct btrfs_qgroup *dstgroup;
1338 u32 level_size = 0;
1339
1340 if (!fs_info->quota_enabled)
1341 return 0;
1342
1343 if (!quota_root)
1344 return -EINVAL;
1345
1346 /*
1347 * create a tracking group for the subvol itself
1348 */
1349 ret = add_qgroup_item(trans, quota_root, objectid);
1350 if (ret)
1351 goto out;
1352
1353 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1354 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1355 inherit->lim.flags,
1356 inherit->lim.max_rfer,
1357 inherit->lim.max_excl,
1358 inherit->lim.rsv_rfer,
1359 inherit->lim.rsv_excl);
1360 if (ret)
1361 goto out;
1362 }
1363
1364 if (srcid) {
1365 struct btrfs_root *srcroot;
1366 struct btrfs_key srckey;
1367 int srcroot_level;
1368
1369 srckey.objectid = srcid;
1370 srckey.type = BTRFS_ROOT_ITEM_KEY;
1371 srckey.offset = (u64)-1;
1372 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1373 if (IS_ERR(srcroot)) {
1374 ret = PTR_ERR(srcroot);
1375 goto out;
1376 }
1377
1378 rcu_read_lock();
1379 srcroot_level = btrfs_header_level(srcroot->node);
1380 level_size = btrfs_level_size(srcroot, srcroot_level);
1381 rcu_read_unlock();
1382 }
1383
1384 /*
1385 * add qgroup to all inherited groups
1386 */
1387 if (inherit) {
1388 i_qgroups = (u64 *)(inherit + 1);
1389 for (i = 0; i < inherit->num_qgroups; ++i) {
1390 ret = add_qgroup_relation_item(trans, quota_root,
1391 objectid, *i_qgroups);
1392 if (ret)
1393 goto out;
1394 ret = add_qgroup_relation_item(trans, quota_root,
1395 *i_qgroups, objectid);
1396 if (ret)
1397 goto out;
1398 ++i_qgroups;
1399 }
1400 }
1401
1402
1403 spin_lock(&fs_info->qgroup_lock);
1404
1405 dstgroup = add_qgroup_rb(fs_info, objectid);
57a5a882
DC		 1406 if (IS_ERR(dstgroup)) {
1407 ret = PTR_ERR(dstgroup);
bed92eae 1408 goto unlock;
57a5a882 1409 }
bed92eae
AJ		 1410
1411 if (srcid) {
1412 srcgroup = find_qgroup_rb(fs_info, srcid);
f3a87f1b 1413 if (!srcgroup)
bed92eae
AJ		 1414 goto unlock;
1415 dstgroup->rfer = srcgroup->rfer - level_size;
1416 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1417 srcgroup->excl = level_size;
1418 srcgroup->excl_cmpr = level_size;
1419 qgroup_dirty(fs_info, dstgroup);
1420 qgroup_dirty(fs_info, srcgroup);
1421 }
1422
f3a87f1b 1423 if (!inherit)
bed92eae
AJ		 1424 goto unlock;
1425
1426 i_qgroups = (u64 *)(inherit + 1);
1427 for (i = 0; i < inherit->num_qgroups; ++i) {
1428 ret = add_relation_rb(quota_root->fs_info, objectid,
1429 *i_qgroups);
1430 if (ret)
1431 goto unlock;
1432 ++i_qgroups;
1433 }
1434
1435 for (i = 0; i < inherit->num_ref_copies; ++i) {
1436 struct btrfs_qgroup *src;
1437 struct btrfs_qgroup *dst;
1438
1439 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1440 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1441
1442 if (!src || !dst) {
1443 ret = -EINVAL;
1444 goto unlock;
1445 }
1446
1447 dst->rfer = src->rfer - level_size;
1448 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1449 i_qgroups += 2;
1450 }
1451 for (i = 0; i < inherit->num_excl_copies; ++i) {
1452 struct btrfs_qgroup *src;
1453 struct btrfs_qgroup *dst;
1454
1455 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1456 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1457
1458 if (!src || !dst) {
1459 ret = -EINVAL;
1460 goto unlock;
1461 }
1462
1463 dst->excl = src->excl + level_size;
1464 dst->excl_cmpr = src->excl_cmpr + level_size;
1465 i_qgroups += 2;
1466 }
1467
1468unlock:
1469 spin_unlock(&fs_info->qgroup_lock);
1470out:
1471 return ret;
1472}
1473
1474/*
1475 * reserve some space for a qgroup and all its parents. The reservation takes
1476 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1477 * accounting. If not enough space is available, EDQUOT is returned.
1478 * We assume that the requested space is new for all qgroups.
1479 */
1480int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1481{
1482 struct btrfs_root *quota_root;
1483 struct btrfs_qgroup *qgroup;
1484 struct btrfs_fs_info *fs_info = root->fs_info;
1485 u64 ref_root = root->root_key.objectid;
1486 int ret = 0;
1487 struct ulist *ulist = NULL;
1488 struct ulist_node *unode;
1489 struct ulist_iterator uiter;
1490
1491 if (!is_fstree(ref_root))
1492 return 0;
1493
1494 if (num_bytes == 0)
1495 return 0;
1496
1497 spin_lock(&fs_info->qgroup_lock);
1498 quota_root = fs_info->quota_root;
1499 if (!quota_root)
1500 goto out;
1501
1502 qgroup = find_qgroup_rb(fs_info, ref_root);
1503 if (!qgroup)
1504 goto out;
1505
1506 /*
1507 * in a first step, we check all affected qgroups if any limits would
1508 * be exceeded
1509 */
1510 ulist = ulist_alloc(GFP_ATOMIC);
3d6b5c3b
TI		 1511 if (!ulist) {
1512 ret = -ENOMEM;
1513 goto out;
1514 }
995e01b7 1515 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1516 ULIST_ITER_INIT(&uiter);
1517 while ((unode = ulist_next(ulist, &uiter))) {
1518 struct btrfs_qgroup *qg;
1519 struct btrfs_qgroup_list *glist;
1520
995e01b7 1521 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1522
1523 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1524 qg->reserved + qg->rfer + num_bytes >
720f1e20 1525 qg->max_rfer) {
bed92eae 1526 ret = -EDQUOT;
720f1e20
WS		 1527 goto out;
1528 }
bed92eae
AJ		 1529
1530 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1531 qg->reserved + qg->excl + num_bytes >
720f1e20 1532 qg->max_excl) {
bed92eae 1533 ret = -EDQUOT;
720f1e20
WS		 1534 goto out;
1535 }
bed92eae
AJ		 1536
1537 list_for_each_entry(glist, &qg->groups, next_group) {
1538 ulist_add(ulist, glist->group->qgroupid,
995e01b7 1539 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1540 }
1541 }
bed92eae
AJ		 1542
1543 /*
1544 * no limits exceeded, now record the reservation into all qgroups
1545 */
1546 ULIST_ITER_INIT(&uiter);
1547 while ((unode = ulist_next(ulist, &uiter))) {
1548 struct btrfs_qgroup *qg;
1549
995e01b7 1550 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1551
1552 qg->reserved += num_bytes;
1553 }
1554
1555out:
1556 spin_unlock(&fs_info->qgroup_lock);
1557 ulist_free(ulist);
1558
1559 return ret;
1560}
1561
1562void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1563{
1564 struct btrfs_root *quota_root;
1565 struct btrfs_qgroup *qgroup;
1566 struct btrfs_fs_info *fs_info = root->fs_info;
1567 struct ulist *ulist = NULL;
1568 struct ulist_node *unode;
1569 struct ulist_iterator uiter;
1570 u64 ref_root = root->root_key.objectid;
1571
1572 if (!is_fstree(ref_root))
1573 return;
1574
1575 if (num_bytes == 0)
1576 return;
1577
1578 spin_lock(&fs_info->qgroup_lock);
1579
1580 quota_root = fs_info->quota_root;
1581 if (!quota_root)
1582 goto out;
1583
1584 qgroup = find_qgroup_rb(fs_info, ref_root);
1585 if (!qgroup)
1586 goto out;
1587
1588 ulist = ulist_alloc(GFP_ATOMIC);
3d6b5c3b
TI		 1589 if (!ulist) {
1590 btrfs_std_error(fs_info, -ENOMEM);
1591 goto out;
1592 }
995e01b7 1593 ulist_add(ulist, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
bed92eae
AJ		 1594 ULIST_ITER_INIT(&uiter);
1595 while ((unode = ulist_next(ulist, &uiter))) {
1596 struct btrfs_qgroup *qg;
1597 struct btrfs_qgroup_list *glist;
1598
995e01b7 1599 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
bed92eae
AJ		 1600
1601 qg->reserved -= num_bytes;
1602
1603 list_for_each_entry(glist, &qg->groups, next_group) {
1604 ulist_add(ulist, glist->group->qgroupid,
995e01b7 1605 (uintptr_t)glist->group, GFP_ATOMIC);
bed92eae
AJ		 1606 }
1607 }
1608
1609out:
1610 spin_unlock(&fs_info->qgroup_lock);
1611 ulist_free(ulist);
1612}
1613
1614void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1615{
1616 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1617 return;
1618 printk(KERN_ERR "btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %llu\n",
1619 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1620 trans->delayed_ref_elem.seq);
1621 BUG();
1622}
Linux kernel - Deliverables
This page took 0.157357 seconds and 5 git commands to generate.