]>
Commit | Line | Data |
---|---|---|
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 | 33 | #include "backref.h" |
2f232036 | 34 | #include "extent_io.h" |
fcebe456 | 35 | #include "qgroup.h" |
bed92eae | 36 | |
e69bcee3 | 37 | |
bed92eae AJ |
38 | /* TODO XXX FIXME |
39 | * - subvol delete -> delete when ref goes to 0? delete limits also? | |
40 | * - reorganize keys | |
41 | * - compressed | |
42 | * - sync | |
bed92eae AJ |
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 | |
9c542136 | 88 | * Refer to qgroup_shared_accouting() for details. |
bed92eae | 89 | */ |
fcebe456 JB |
90 | u64 old_refcnt; |
91 | u64 new_refcnt; | |
bed92eae AJ |
92 | }; |
93 | ||
9c542136 QW |
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 | ||
bed92eae AJ |
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 | ||
fcebe456 JB |
134 | #define ptr_to_u64(x) ((u64)(uintptr_t)x) |
135 | #define u64_to_ptr(x) ((struct btrfs_qgroup *)(uintptr_t)x) | |
136 | ||
b382a324 JS |
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); | |
2f232036 | 141 | |
58400fce | 142 | /* must be called with qgroup_ioctl_lock held */ |
bed92eae AJ |
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 | ||
4082bd3d | 196 | static void __del_qgroup_rb(struct btrfs_qgroup *qgroup) |
bed92eae | 197 | { |
bed92eae AJ |
198 | struct btrfs_qgroup_list *list; |
199 | ||
bed92eae | 200 | list_del(&qgroup->dirty); |
bed92eae AJ |
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); | |
4082bd3d | 217 | } |
bed92eae | 218 | |
4082bd3d WS |
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); | |
bed92eae AJ |
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 | ||
faa2dbf0 JB |
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 | ||
bed92eae AJ |
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; | |
b382a324 | 310 | u64 rescan_progress = 0; |
bed92eae AJ |
311 | |
312 | if (!fs_info->quota_enabled) | |
313 | return 0; | |
314 | ||
1e8f9158 WS |
315 | fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS); |
316 | if (!fs_info->qgroup_ulist) { | |
317 | ret = -ENOMEM; | |
318 | goto out; | |
319 | } | |
320 | ||
bed92eae AJ |
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) { | |
efe120a0 FH |
355 | btrfs_err(fs_info, |
356 | "old qgroup version, quota disabled"); | |
bed92eae AJ |
357 | goto out; |
358 | } | |
359 | if (btrfs_qgroup_status_generation(l, ptr) != | |
360 | fs_info->generation) { | |
361 | flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
efe120a0 FH |
362 | btrfs_err(fs_info, |
363 | "qgroup generation mismatch, " | |
364 | "marked as inconsistent"); | |
bed92eae AJ |
365 | } |
366 | fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, | |
367 | ptr); | |
b382a324 | 368 | rescan_progress = btrfs_qgroup_status_rescan(l, ptr); |
bed92eae AJ |
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)) { | |
d41e36a0 | 379 | btrfs_err(fs_info, "inconsistent qgroup config"); |
bed92eae AJ |
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); | |
ff24858c | 449 | if (ret == -ENOENT) { |
efe120a0 FH |
450 | btrfs_warn(fs_info, |
451 | "orphan qgroup relation 0x%llx->0x%llx", | |
c1c9ff7c | 452 | found_key.objectid, found_key.offset); |
ff24858c AJ |
453 | ret = 0; /* ignore the error */ |
454 | } | |
bed92eae AJ |
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; | |
b382a324 JS |
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); | |
bed92eae AJ |
472 | } |
473 | btrfs_free_path(path); | |
474 | ||
eb1716af | 475 | if (ret < 0) { |
1e8f9158 | 476 | ulist_free(fs_info->qgroup_ulist); |
eb1716af | 477 | fs_info->qgroup_ulist = NULL; |
b382a324 | 478 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; |
eb1716af | 479 | } |
1e8f9158 | 480 | |
bed92eae AJ |
481 | return ret < 0 ? ret : 0; |
482 | } | |
483 | ||
484 | /* | |
e685da14 WS |
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. | |
bed92eae AJ |
489 | */ |
490 | void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info) | |
491 | { | |
492 | struct rb_node *n; | |
493 | struct btrfs_qgroup *qgroup; | |
bed92eae AJ |
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); | |
4082bd3d | 498 | __del_qgroup_rb(qgroup); |
bed92eae | 499 | } |
1e7bac1e WS |
500 | /* |
501 | * we call btrfs_free_qgroup_config() when umounting | |
502 | * filesystem and disabling quota, so we set qgroup_ulit | |
503 | * to be null here to avoid double free. | |
504 | */ | |
1e8f9158 | 505 | ulist_free(fs_info->qgroup_ulist); |
1e7bac1e | 506 | fs_info->qgroup_ulist = NULL; |
bed92eae AJ |
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 | ||
fccb84c9 | 574 | if (btrfs_test_is_dummy_root(quota_root)) |
faa2dbf0 | 575 | return 0; |
fccb84c9 | 576 | |
bed92eae AJ |
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 | ||
0b4699dc MF |
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 | ||
bed92eae AJ |
591 | ret = btrfs_insert_empty_item(trans, quota_root, path, &key, |
592 | sizeof(*qgroup_info)); | |
0b4699dc | 593 | if (ret && ret != -EEXIST) |
bed92eae AJ |
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)); | |
0b4699dc | 612 | if (ret && ret != -EEXIST) |
bed92eae AJ |
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, | |
1510e71c DY |
679 | struct btrfs_root *root, |
680 | struct btrfs_qgroup *qgroup) | |
bed92eae AJ |
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; | |
1510e71c | 691 | key.offset = qgroup->qgroupid; |
bed92eae AJ |
692 | |
693 | path = btrfs_alloc_path(); | |
84cbe2f7 WS |
694 | if (!path) |
695 | return -ENOMEM; | |
696 | ||
bed92eae AJ |
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]; | |
a3df41ee | 706 | qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item); |
1510e71c DY |
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); | |
bed92eae AJ |
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 | ||
fccb84c9 | 731 | if (btrfs_test_is_dummy_root(root)) |
faa2dbf0 | 732 | return 0; |
fccb84c9 | 733 | |
bed92eae AJ |
734 | key.objectid = 0; |
735 | key.type = BTRFS_QGROUP_INFO_KEY; | |
736 | key.offset = qgroup->qgroupid; | |
737 | ||
738 | path = btrfs_alloc_path(); | |
84cbe2f7 WS |
739 | if (!path) |
740 | return -ENOMEM; | |
741 | ||
bed92eae AJ |
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]; | |
a3df41ee | 751 | qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item); |
bed92eae AJ |
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(); | |
84cbe2f7 WS |
781 | if (!path) |
782 | return -ENOMEM; | |
783 | ||
bed92eae AJ |
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); | |
2f232036 JS |
796 | btrfs_set_qgroup_status_rescan(l, ptr, |
797 | fs_info->qgroup_rescan_progress.objectid); | |
bed92eae AJ |
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; | |
06b3a860 | 814 | struct extent_buffer *leaf = NULL; |
bed92eae | 815 | int ret; |
06b3a860 | 816 | int nr = 0; |
bed92eae | 817 | |
bed92eae AJ |
818 | path = btrfs_alloc_path(); |
819 | if (!path) | |
820 | return -ENOMEM; | |
821 | ||
06b3a860 WS |
822 | path->leave_spinning = 1; |
823 | ||
824 | key.objectid = 0; | |
825 | key.offset = 0; | |
826 | key.type = 0; | |
bed92eae | 827 | |
06b3a860 | 828 | while (1) { |
bed92eae | 829 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
06b3a860 WS |
830 | if (ret < 0) |
831 | goto out; | |
832 | leaf = path->nodes[0]; | |
833 | nr = btrfs_header_nritems(leaf); | |
834 | if (!nr) | |
bed92eae | 835 | break; |
06b3a860 WS |
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); | |
bed92eae AJ |
843 | if (ret) |
844 | goto out; | |
06b3a860 | 845 | |
bed92eae AJ |
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; | |
7708f029 | 859 | struct btrfs_root *tree_root = fs_info->tree_root; |
bed92eae AJ |
860 | struct btrfs_path *path = NULL; |
861 | struct btrfs_qgroup_status_item *ptr; | |
862 | struct extent_buffer *leaf; | |
863 | struct btrfs_key key; | |
7708f029 WS |
864 | struct btrfs_key found_key; |
865 | struct btrfs_qgroup *qgroup = NULL; | |
bed92eae | 866 | int ret = 0; |
7708f029 | 867 | int slot; |
bed92eae | 868 | |
f2f6ed3d | 869 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
870 | if (fs_info->quota_root) { |
871 | fs_info->pending_quota_state = 1; | |
bed92eae AJ |
872 | goto out; |
873 | } | |
bed92eae | 874 | |
1e8f9158 WS |
875 | fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS); |
876 | if (!fs_info->qgroup_ulist) { | |
877 | ret = -ENOMEM; | |
878 | goto out; | |
879 | } | |
880 | ||
bed92eae AJ |
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(); | |
5b7ff5b3 TI |
892 | if (!path) { |
893 | ret = -ENOMEM; | |
894 | goto out_free_root; | |
895 | } | |
bed92eae AJ |
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) | |
5b7ff5b3 | 904 | goto out_free_path; |
bed92eae AJ |
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); | |
2f232036 | 914 | btrfs_set_qgroup_status_rescan(leaf, ptr, 0); |
bed92eae AJ |
915 | |
916 | btrfs_mark_buffer_dirty(leaf); | |
917 | ||
7708f029 WS |
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 | ||
7708f029 WS |
941 | qgroup = add_qgroup_rb(fs_info, found_key.offset); |
942 | if (IS_ERR(qgroup)) { | |
7708f029 WS |
943 | ret = PTR_ERR(qgroup); |
944 | goto out_free_path; | |
945 | } | |
7708f029 WS |
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 | ||
7708f029 WS |
960 | qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID); |
961 | if (IS_ERR(qgroup)) { | |
7708f029 WS |
962 | ret = PTR_ERR(qgroup); |
963 | goto out_free_path; | |
964 | } | |
58400fce | 965 | spin_lock(&fs_info->qgroup_lock); |
bed92eae AJ |
966 | fs_info->quota_root = quota_root; |
967 | fs_info->pending_quota_state = 1; | |
968 | spin_unlock(&fs_info->qgroup_lock); | |
5b7ff5b3 | 969 | out_free_path: |
bed92eae | 970 | btrfs_free_path(path); |
5b7ff5b3 TI |
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: | |
eb1716af | 978 | if (ret) { |
1e8f9158 | 979 | ulist_free(fs_info->qgroup_ulist); |
eb1716af JS |
980 | fs_info->qgroup_ulist = NULL; |
981 | } | |
f2f6ed3d | 982 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
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 | ||
f2f6ed3d | 993 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
58400fce | 994 | if (!fs_info->quota_root) |
f2f6ed3d | 995 | goto out; |
58400fce | 996 | spin_lock(&fs_info->qgroup_lock); |
bed92eae AJ |
997 | fs_info->quota_enabled = 0; |
998 | fs_info->pending_quota_state = 0; | |
999 | quota_root = fs_info->quota_root; | |
1000 | fs_info->quota_root = NULL; | |
8ea0ec9e | 1001 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON; |
bed92eae AJ |
1002 | spin_unlock(&fs_info->qgroup_lock); |
1003 | ||
e685da14 WS |
1004 | btrfs_free_qgroup_config(fs_info); |
1005 | ||
bed92eae AJ |
1006 | ret = btrfs_clean_quota_tree(trans, quota_root); |
1007 | if (ret) | |
1008 | goto out; | |
1009 | ||
1010 | ret = btrfs_del_root(trans, tree_root, "a_root->root_key); | |
1011 | if (ret) | |
1012 | goto out; | |
1013 | ||
1014 | list_del("a_root->dirty_list); | |
1015 | ||
1016 | btrfs_tree_lock(quota_root->node); | |
01d58472 | 1017 | clean_tree_block(trans, tree_root->fs_info, quota_root->node); |
bed92eae AJ |
1018 | btrfs_tree_unlock(quota_root->node); |
1019 | btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1); | |
1020 | ||
1021 | free_extent_buffer(quota_root->node); | |
1022 | free_extent_buffer(quota_root->commit_root); | |
1023 | kfree(quota_root); | |
1024 | out: | |
f2f6ed3d | 1025 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
1026 | return ret; |
1027 | } | |
1028 | ||
2f232036 JS |
1029 | static void qgroup_dirty(struct btrfs_fs_info *fs_info, |
1030 | struct btrfs_qgroup *qgroup) | |
bed92eae | 1031 | { |
2f232036 JS |
1032 | if (list_empty(&qgroup->dirty)) |
1033 | list_add(&qgroup->dirty, &fs_info->dirty_qgroups); | |
bed92eae AJ |
1034 | } |
1035 | ||
9c8b35b1 QW |
1036 | /* |
1037 | * The easy accounting, if we are adding/removing the only ref for an extent | |
1038 | * then this qgroup and all of the parent qgroups get their refrence and | |
1039 | * exclusive counts adjusted. | |
1040 | * | |
1041 | * Caller should hold fs_info->qgroup_lock. | |
1042 | */ | |
1043 | static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info, | |
1044 | struct ulist *tmp, u64 ref_root, | |
1045 | u64 num_bytes, int sign) | |
1046 | { | |
1047 | struct btrfs_qgroup *qgroup; | |
1048 | struct btrfs_qgroup_list *glist; | |
1049 | struct ulist_node *unode; | |
1050 | struct ulist_iterator uiter; | |
1051 | int ret = 0; | |
1052 | ||
1053 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
1054 | if (!qgroup) | |
1055 | goto out; | |
1056 | ||
1057 | qgroup->rfer += sign * num_bytes; | |
1058 | qgroup->rfer_cmpr += sign * num_bytes; | |
1059 | ||
1060 | WARN_ON(sign < 0 && qgroup->excl < num_bytes); | |
1061 | qgroup->excl += sign * num_bytes; | |
1062 | qgroup->excl_cmpr += sign * num_bytes; | |
1063 | if (sign > 0) | |
1064 | qgroup->reserved -= num_bytes; | |
1065 | ||
1066 | qgroup_dirty(fs_info, qgroup); | |
1067 | ||
1068 | /* Get all of the parent groups that contain this qgroup */ | |
1069 | list_for_each_entry(glist, &qgroup->groups, next_group) { | |
1070 | ret = ulist_add(tmp, glist->group->qgroupid, | |
1071 | ptr_to_u64(glist->group), GFP_ATOMIC); | |
1072 | if (ret < 0) | |
1073 | goto out; | |
1074 | } | |
1075 | ||
1076 | /* Iterate all of the parents and adjust their reference counts */ | |
1077 | ULIST_ITER_INIT(&uiter); | |
1078 | while ((unode = ulist_next(tmp, &uiter))) { | |
1079 | qgroup = u64_to_ptr(unode->aux); | |
1080 | qgroup->rfer += sign * num_bytes; | |
1081 | qgroup->rfer_cmpr += sign * num_bytes; | |
1082 | WARN_ON(sign < 0 && qgroup->excl < num_bytes); | |
1083 | qgroup->excl += sign * num_bytes; | |
1084 | if (sign > 0) | |
1085 | qgroup->reserved -= num_bytes; | |
1086 | qgroup->excl_cmpr += sign * num_bytes; | |
1087 | qgroup_dirty(fs_info, qgroup); | |
1088 | ||
1089 | /* Add any parents of the parents */ | |
1090 | list_for_each_entry(glist, &qgroup->groups, next_group) { | |
1091 | ret = ulist_add(tmp, glist->group->qgroupid, | |
1092 | ptr_to_u64(glist->group), GFP_ATOMIC); | |
1093 | if (ret < 0) | |
1094 | goto out; | |
1095 | } | |
1096 | } | |
1097 | ret = 0; | |
1098 | out: | |
1099 | return ret; | |
1100 | } | |
1101 | ||
1102 | ||
1103 | /* | |
1104 | * Quick path for updating qgroup with only excl refs. | |
1105 | * | |
1106 | * In that case, just update all parent will be enough. | |
1107 | * Or we needs to do a full rescan. | |
1108 | * Caller should also hold fs_info->qgroup_lock. | |
1109 | * | |
1110 | * Return 0 for quick update, return >0 for need to full rescan | |
1111 | * and mark INCONSISTENT flag. | |
1112 | * Return < 0 for other error. | |
1113 | */ | |
1114 | static int quick_update_accounting(struct btrfs_fs_info *fs_info, | |
1115 | struct ulist *tmp, u64 src, u64 dst, | |
1116 | int sign) | |
1117 | { | |
1118 | struct btrfs_qgroup *qgroup; | |
1119 | int ret = 1; | |
1120 | int err = 0; | |
1121 | ||
1122 | qgroup = find_qgroup_rb(fs_info, src); | |
1123 | if (!qgroup) | |
1124 | goto out; | |
1125 | if (qgroup->excl == qgroup->rfer) { | |
1126 | ret = 0; | |
1127 | err = __qgroup_excl_accounting(fs_info, tmp, dst, | |
1128 | qgroup->excl, sign); | |
1129 | if (err < 0) { | |
1130 | ret = err; | |
1131 | goto out; | |
1132 | } | |
1133 | } | |
1134 | out: | |
1135 | if (ret) | |
1136 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1137 | return ret; | |
1138 | } | |
1139 | ||
bed92eae AJ |
1140 | int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, |
1141 | struct btrfs_fs_info *fs_info, u64 src, u64 dst) | |
1142 | { | |
1143 | struct btrfs_root *quota_root; | |
b7fef4f5 WS |
1144 | struct btrfs_qgroup *parent; |
1145 | struct btrfs_qgroup *member; | |
534e6623 | 1146 | struct btrfs_qgroup_list *list; |
9c8b35b1 | 1147 | struct ulist *tmp; |
bed92eae AJ |
1148 | int ret = 0; |
1149 | ||
8465ecec QW |
1150 | /* Check the level of src and dst first */ |
1151 | if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst)) | |
1152 | return -EINVAL; | |
1153 | ||
ab3680dd CE |
1154 | tmp = ulist_alloc(GFP_NOFS); |
1155 | if (!tmp) | |
1156 | return -ENOMEM; | |
1157 | ||
f2f6ed3d | 1158 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1159 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1160 | if (!quota_root) { |
1161 | ret = -EINVAL; | |
1162 | goto out; | |
1163 | } | |
b7fef4f5 WS |
1164 | member = find_qgroup_rb(fs_info, src); |
1165 | parent = find_qgroup_rb(fs_info, dst); | |
1166 | if (!member || !parent) { | |
1167 | ret = -EINVAL; | |
1168 | goto out; | |
1169 | } | |
bed92eae | 1170 | |
534e6623 WS |
1171 | /* check if such qgroup relation exist firstly */ |
1172 | list_for_each_entry(list, &member->groups, next_group) { | |
1173 | if (list->group == parent) { | |
1174 | ret = -EEXIST; | |
1175 | goto out; | |
1176 | } | |
1177 | } | |
1178 | ||
bed92eae AJ |
1179 | ret = add_qgroup_relation_item(trans, quota_root, src, dst); |
1180 | if (ret) | |
f2f6ed3d | 1181 | goto out; |
bed92eae AJ |
1182 | |
1183 | ret = add_qgroup_relation_item(trans, quota_root, dst, src); | |
1184 | if (ret) { | |
1185 | del_qgroup_relation_item(trans, quota_root, src, dst); | |
f2f6ed3d | 1186 | goto out; |
bed92eae AJ |
1187 | } |
1188 | ||
1189 | spin_lock(&fs_info->qgroup_lock); | |
1190 | ret = add_relation_rb(quota_root->fs_info, src, dst); | |
9c8b35b1 QW |
1191 | if (ret < 0) { |
1192 | spin_unlock(&fs_info->qgroup_lock); | |
1193 | goto out; | |
1194 | } | |
1195 | ret = quick_update_accounting(fs_info, tmp, src, dst, 1); | |
bed92eae | 1196 | spin_unlock(&fs_info->qgroup_lock); |
f2f6ed3d WS |
1197 | out: |
1198 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
9c8b35b1 | 1199 | ulist_free(tmp); |
bed92eae AJ |
1200 | return ret; |
1201 | } | |
1202 | ||
f5a6b1c5 | 1203 | int __del_qgroup_relation(struct btrfs_trans_handle *trans, |
bed92eae AJ |
1204 | struct btrfs_fs_info *fs_info, u64 src, u64 dst) |
1205 | { | |
1206 | struct btrfs_root *quota_root; | |
534e6623 WS |
1207 | struct btrfs_qgroup *parent; |
1208 | struct btrfs_qgroup *member; | |
1209 | struct btrfs_qgroup_list *list; | |
9c8b35b1 | 1210 | struct ulist *tmp; |
bed92eae AJ |
1211 | int ret = 0; |
1212 | int err; | |
1213 | ||
9c8b35b1 QW |
1214 | tmp = ulist_alloc(GFP_NOFS); |
1215 | if (!tmp) | |
1216 | return -ENOMEM; | |
1217 | ||
bed92eae | 1218 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1219 | if (!quota_root) { |
1220 | ret = -EINVAL; | |
1221 | goto out; | |
1222 | } | |
bed92eae | 1223 | |
534e6623 WS |
1224 | member = find_qgroup_rb(fs_info, src); |
1225 | parent = find_qgroup_rb(fs_info, dst); | |
1226 | if (!member || !parent) { | |
1227 | ret = -EINVAL; | |
1228 | goto out; | |
1229 | } | |
1230 | ||
1231 | /* check if such qgroup relation exist firstly */ | |
1232 | list_for_each_entry(list, &member->groups, next_group) { | |
1233 | if (list->group == parent) | |
1234 | goto exist; | |
1235 | } | |
1236 | ret = -ENOENT; | |
1237 | goto out; | |
1238 | exist: | |
bed92eae AJ |
1239 | ret = del_qgroup_relation_item(trans, quota_root, src, dst); |
1240 | err = del_qgroup_relation_item(trans, quota_root, dst, src); | |
1241 | if (err && !ret) | |
1242 | ret = err; | |
1243 | ||
1244 | spin_lock(&fs_info->qgroup_lock); | |
1245 | del_relation_rb(fs_info, src, dst); | |
9c8b35b1 | 1246 | ret = quick_update_accounting(fs_info, tmp, src, dst, -1); |
bed92eae | 1247 | spin_unlock(&fs_info->qgroup_lock); |
f2f6ed3d | 1248 | out: |
9c8b35b1 | 1249 | ulist_free(tmp); |
f5a6b1c5 DY |
1250 | return ret; |
1251 | } | |
1252 | ||
1253 | int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, | |
1254 | struct btrfs_fs_info *fs_info, u64 src, u64 dst) | |
1255 | { | |
1256 | int ret = 0; | |
1257 | ||
1258 | mutex_lock(&fs_info->qgroup_ioctl_lock); | |
1259 | ret = __del_qgroup_relation(trans, fs_info, src, dst); | |
f2f6ed3d | 1260 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
f5a6b1c5 | 1261 | |
bed92eae AJ |
1262 | return ret; |
1263 | } | |
1264 | ||
1265 | int btrfs_create_qgroup(struct btrfs_trans_handle *trans, | |
4087cf24 | 1266 | struct btrfs_fs_info *fs_info, u64 qgroupid) |
bed92eae AJ |
1267 | { |
1268 | struct btrfs_root *quota_root; | |
1269 | struct btrfs_qgroup *qgroup; | |
1270 | int ret = 0; | |
1271 | ||
f2f6ed3d | 1272 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1273 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1274 | if (!quota_root) { |
1275 | ret = -EINVAL; | |
1276 | goto out; | |
1277 | } | |
534e6623 WS |
1278 | qgroup = find_qgroup_rb(fs_info, qgroupid); |
1279 | if (qgroup) { | |
1280 | ret = -EEXIST; | |
1281 | goto out; | |
1282 | } | |
bed92eae AJ |
1283 | |
1284 | ret = add_qgroup_item(trans, quota_root, qgroupid); | |
534e6623 WS |
1285 | if (ret) |
1286 | goto out; | |
bed92eae AJ |
1287 | |
1288 | spin_lock(&fs_info->qgroup_lock); | |
1289 | qgroup = add_qgroup_rb(fs_info, qgroupid); | |
1290 | spin_unlock(&fs_info->qgroup_lock); | |
1291 | ||
1292 | if (IS_ERR(qgroup)) | |
1293 | ret = PTR_ERR(qgroup); | |
f2f6ed3d WS |
1294 | out: |
1295 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
bed92eae AJ |
1296 | return ret; |
1297 | } | |
1298 | ||
1299 | int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, | |
1300 | struct btrfs_fs_info *fs_info, u64 qgroupid) | |
1301 | { | |
1302 | struct btrfs_root *quota_root; | |
2cf68703 | 1303 | struct btrfs_qgroup *qgroup; |
f5a6b1c5 | 1304 | struct btrfs_qgroup_list *list; |
bed92eae AJ |
1305 | int ret = 0; |
1306 | ||
f2f6ed3d | 1307 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1308 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1309 | if (!quota_root) { |
1310 | ret = -EINVAL; | |
1311 | goto out; | |
1312 | } | |
bed92eae | 1313 | |
2cf68703 | 1314 | qgroup = find_qgroup_rb(fs_info, qgroupid); |
534e6623 WS |
1315 | if (!qgroup) { |
1316 | ret = -ENOENT; | |
1317 | goto out; | |
1318 | } else { | |
f5a6b1c5 DY |
1319 | /* check if there are no children of this qgroup */ |
1320 | if (!list_empty(&qgroup->members)) { | |
f2f6ed3d WS |
1321 | ret = -EBUSY; |
1322 | goto out; | |
2cf68703 AJ |
1323 | } |
1324 | } | |
bed92eae AJ |
1325 | ret = del_qgroup_item(trans, quota_root, qgroupid); |
1326 | ||
f5a6b1c5 DY |
1327 | while (!list_empty(&qgroup->groups)) { |
1328 | list = list_first_entry(&qgroup->groups, | |
1329 | struct btrfs_qgroup_list, next_group); | |
1330 | ret = __del_qgroup_relation(trans, fs_info, | |
1331 | qgroupid, | |
1332 | list->group->qgroupid); | |
1333 | if (ret) | |
1334 | goto out; | |
1335 | } | |
1336 | ||
bed92eae AJ |
1337 | spin_lock(&fs_info->qgroup_lock); |
1338 | del_qgroup_rb(quota_root->fs_info, qgroupid); | |
bed92eae | 1339 | spin_unlock(&fs_info->qgroup_lock); |
f2f6ed3d WS |
1340 | out: |
1341 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
bed92eae AJ |
1342 | return ret; |
1343 | } | |
1344 | ||
1345 | int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, | |
1346 | struct btrfs_fs_info *fs_info, u64 qgroupid, | |
1347 | struct btrfs_qgroup_limit *limit) | |
1348 | { | |
f2f6ed3d | 1349 | struct btrfs_root *quota_root; |
bed92eae AJ |
1350 | struct btrfs_qgroup *qgroup; |
1351 | int ret = 0; | |
fe759907 YD |
1352 | /* Sometimes we would want to clear the limit on this qgroup. |
1353 | * To meet this requirement, we treat the -1 as a special value | |
1354 | * which tell kernel to clear the limit on this qgroup. | |
1355 | */ | |
1356 | const u64 CLEAR_VALUE = -1; | |
bed92eae | 1357 | |
f2f6ed3d WS |
1358 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
1359 | quota_root = fs_info->quota_root; | |
1360 | if (!quota_root) { | |
1361 | ret = -EINVAL; | |
1362 | goto out; | |
1363 | } | |
bed92eae | 1364 | |
ddb47afa WS |
1365 | qgroup = find_qgroup_rb(fs_info, qgroupid); |
1366 | if (!qgroup) { | |
1367 | ret = -ENOENT; | |
1368 | goto out; | |
1369 | } | |
bed92eae | 1370 | |
58400fce | 1371 | spin_lock(&fs_info->qgroup_lock); |
fe759907 YD |
1372 | if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) { |
1373 | if (limit->max_rfer == CLEAR_VALUE) { | |
1374 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER; | |
1375 | limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER; | |
1376 | qgroup->max_rfer = 0; | |
1377 | } else { | |
1378 | qgroup->max_rfer = limit->max_rfer; | |
1379 | } | |
1380 | } | |
1381 | if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) { | |
1382 | if (limit->max_excl == CLEAR_VALUE) { | |
1383 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL; | |
1384 | limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL; | |
1385 | qgroup->max_excl = 0; | |
1386 | } else { | |
1387 | qgroup->max_excl = limit->max_excl; | |
1388 | } | |
1389 | } | |
1390 | if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) { | |
1391 | if (limit->rsv_rfer == CLEAR_VALUE) { | |
1392 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER; | |
1393 | limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER; | |
1394 | qgroup->rsv_rfer = 0; | |
1395 | } else { | |
1396 | qgroup->rsv_rfer = limit->rsv_rfer; | |
1397 | } | |
1398 | } | |
1399 | if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) { | |
1400 | if (limit->rsv_excl == CLEAR_VALUE) { | |
1401 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL; | |
1402 | limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL; | |
1403 | qgroup->rsv_excl = 0; | |
1404 | } else { | |
1405 | qgroup->rsv_excl = limit->rsv_excl; | |
1406 | } | |
1407 | } | |
03477d94 DY |
1408 | qgroup->lim_flags |= limit->flags; |
1409 | ||
bed92eae | 1410 | spin_unlock(&fs_info->qgroup_lock); |
1510e71c DY |
1411 | |
1412 | ret = update_qgroup_limit_item(trans, quota_root, qgroup); | |
1413 | if (ret) { | |
1414 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1415 | btrfs_info(fs_info, "unable to update quota limit for %llu", | |
1416 | qgroupid); | |
1417 | } | |
1418 | ||
f2f6ed3d WS |
1419 | out: |
1420 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
bed92eae AJ |
1421 | return ret; |
1422 | } | |
1152651a | 1423 | |
3b7d00f9 QW |
1424 | int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans, |
1425 | struct btrfs_fs_info *fs_info) | |
1426 | { | |
1427 | struct btrfs_qgroup_extent_record *record; | |
1428 | struct btrfs_delayed_ref_root *delayed_refs; | |
1429 | struct rb_node *node; | |
9086db86 | 1430 | u64 qgroup_to_skip; |
3b7d00f9 QW |
1431 | int ret = 0; |
1432 | ||
1433 | delayed_refs = &trans->transaction->delayed_refs; | |
9086db86 | 1434 | qgroup_to_skip = delayed_refs->qgroup_to_skip; |
3b7d00f9 QW |
1435 | |
1436 | /* | |
1437 | * No need to do lock, since this function will only be called in | |
1438 | * btrfs_commmit_transaction(). | |
1439 | */ | |
1440 | node = rb_first(&delayed_refs->dirty_extent_root); | |
1441 | while (node) { | |
1442 | record = rb_entry(node, struct btrfs_qgroup_extent_record, | |
1443 | node); | |
1444 | ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0, | |
1445 | &record->old_roots); | |
1446 | if (ret < 0) | |
1447 | break; | |
9086db86 QW |
1448 | if (qgroup_to_skip) |
1449 | ulist_del(record->old_roots, qgroup_to_skip, 0); | |
3b7d00f9 QW |
1450 | node = rb_next(node); |
1451 | } | |
1452 | return ret; | |
1453 | } | |
1454 | ||
3368d001 QW |
1455 | struct btrfs_qgroup_extent_record |
1456 | *btrfs_qgroup_insert_dirty_extent(struct btrfs_delayed_ref_root *delayed_refs, | |
1457 | struct btrfs_qgroup_extent_record *record) | |
1458 | { | |
1459 | struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node; | |
1460 | struct rb_node *parent_node = NULL; | |
1461 | struct btrfs_qgroup_extent_record *entry; | |
1462 | u64 bytenr = record->bytenr; | |
1463 | ||
1464 | while (*p) { | |
1465 | parent_node = *p; | |
1466 | entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record, | |
1467 | node); | |
1468 | if (bytenr < entry->bytenr) | |
1469 | p = &(*p)->rb_left; | |
1470 | else if (bytenr > entry->bytenr) | |
1471 | p = &(*p)->rb_right; | |
1472 | else | |
1473 | return entry; | |
1474 | } | |
1475 | ||
1476 | rb_link_node(&record->node, parent_node, p); | |
1477 | rb_insert_color(&record->node, &delayed_refs->dirty_extent_root); | |
1478 | return NULL; | |
1479 | } | |
1480 | ||
d810ef2b QW |
1481 | #define UPDATE_NEW 0 |
1482 | #define UPDATE_OLD 1 | |
1483 | /* | |
1484 | * Walk all of the roots that points to the bytenr and adjust their refcnts. | |
1485 | */ | |
1486 | static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info, | |
1487 | struct ulist *roots, struct ulist *tmp, | |
1488 | struct ulist *qgroups, u64 seq, int update_old) | |
1489 | { | |
1490 | struct ulist_node *unode; | |
1491 | struct ulist_iterator uiter; | |
1492 | struct ulist_node *tmp_unode; | |
1493 | struct ulist_iterator tmp_uiter; | |
1494 | struct btrfs_qgroup *qg; | |
1495 | int ret = 0; | |
1496 | ||
1497 | if (!roots) | |
1498 | return 0; | |
1499 | ULIST_ITER_INIT(&uiter); | |
1500 | while ((unode = ulist_next(roots, &uiter))) { | |
1501 | qg = find_qgroup_rb(fs_info, unode->val); | |
1502 | if (!qg) | |
1503 | continue; | |
1504 | ||
1505 | ulist_reinit(tmp); | |
1506 | ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg), | |
1507 | GFP_ATOMIC); | |
1508 | if (ret < 0) | |
1509 | return ret; | |
1510 | ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg), GFP_ATOMIC); | |
1511 | if (ret < 0) | |
1512 | return ret; | |
1513 | ULIST_ITER_INIT(&tmp_uiter); | |
1514 | while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) { | |
1515 | struct btrfs_qgroup_list *glist; | |
1516 | ||
1517 | qg = u64_to_ptr(tmp_unode->aux); | |
1518 | if (update_old) | |
1519 | btrfs_qgroup_update_old_refcnt(qg, seq, 1); | |
1520 | else | |
1521 | btrfs_qgroup_update_new_refcnt(qg, seq, 1); | |
1522 | list_for_each_entry(glist, &qg->groups, next_group) { | |
1523 | ret = ulist_add(qgroups, glist->group->qgroupid, | |
1524 | ptr_to_u64(glist->group), | |
1525 | GFP_ATOMIC); | |
1526 | if (ret < 0) | |
1527 | return ret; | |
1528 | ret = ulist_add(tmp, glist->group->qgroupid, | |
1529 | ptr_to_u64(glist->group), | |
1530 | GFP_ATOMIC); | |
1531 | if (ret < 0) | |
1532 | return ret; | |
1533 | } | |
1534 | } | |
1535 | } | |
1536 | return 0; | |
1537 | } | |
1538 | ||
823ae5b8 QW |
1539 | /* |
1540 | * Update qgroup rfer/excl counters. | |
1541 | * Rfer update is easy, codes can explain themselves. | |
e69bcee3 | 1542 | * |
823ae5b8 QW |
1543 | * Excl update is tricky, the update is split into 2 part. |
1544 | * Part 1: Possible exclusive <-> sharing detect: | |
1545 | * | A | !A | | |
1546 | * ------------------------------------- | |
1547 | * B | * | - | | |
1548 | * ------------------------------------- | |
1549 | * !B | + | ** | | |
1550 | * ------------------------------------- | |
1551 | * | |
1552 | * Conditions: | |
1553 | * A: cur_old_roots < nr_old_roots (not exclusive before) | |
1554 | * !A: cur_old_roots == nr_old_roots (possible exclusive before) | |
1555 | * B: cur_new_roots < nr_new_roots (not exclusive now) | |
1556 | * !B: cur_new_roots == nr_new_roots (possible exclsuive now) | |
1557 | * | |
1558 | * Results: | |
1559 | * +: Possible sharing -> exclusive -: Possible exclusive -> sharing | |
1560 | * *: Definitely not changed. **: Possible unchanged. | |
1561 | * | |
1562 | * For !A and !B condition, the exception is cur_old/new_roots == 0 case. | |
1563 | * | |
1564 | * To make the logic clear, we first use condition A and B to split | |
1565 | * combination into 4 results. | |
1566 | * | |
1567 | * Then, for result "+" and "-", check old/new_roots == 0 case, as in them | |
1568 | * only on variant maybe 0. | |
1569 | * | |
1570 | * Lastly, check result **, since there are 2 variants maybe 0, split them | |
1571 | * again(2x2). | |
1572 | * But this time we don't need to consider other things, the codes and logic | |
1573 | * is easy to understand now. | |
1574 | */ | |
1575 | static int qgroup_update_counters(struct btrfs_fs_info *fs_info, | |
1576 | struct ulist *qgroups, | |
1577 | u64 nr_old_roots, | |
1578 | u64 nr_new_roots, | |
1579 | u64 num_bytes, u64 seq) | |
1580 | { | |
1581 | struct ulist_node *unode; | |
1582 | struct ulist_iterator uiter; | |
1583 | struct btrfs_qgroup *qg; | |
1584 | u64 cur_new_count, cur_old_count; | |
1585 | ||
1586 | ULIST_ITER_INIT(&uiter); | |
1587 | while ((unode = ulist_next(qgroups, &uiter))) { | |
1588 | bool dirty = false; | |
1589 | ||
1590 | qg = u64_to_ptr(unode->aux); | |
1591 | cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq); | |
1592 | cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq); | |
1593 | ||
1594 | /* Rfer update part */ | |
1595 | if (cur_old_count == 0 && cur_new_count > 0) { | |
1596 | qg->rfer += num_bytes; | |
1597 | qg->rfer_cmpr += num_bytes; | |
1598 | dirty = true; | |
1599 | } | |
1600 | if (cur_old_count > 0 && cur_new_count == 0) { | |
1601 | qg->rfer -= num_bytes; | |
1602 | qg->rfer_cmpr -= num_bytes; | |
1603 | dirty = true; | |
1604 | } | |
1605 | ||
1606 | /* Excl update part */ | |
1607 | /* Exclusive/none -> shared case */ | |
1608 | if (cur_old_count == nr_old_roots && | |
1609 | cur_new_count < nr_new_roots) { | |
1610 | /* Exclusive -> shared */ | |
1611 | if (cur_old_count != 0) { | |
1612 | qg->excl -= num_bytes; | |
1613 | qg->excl_cmpr -= num_bytes; | |
1614 | dirty = true; | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | /* Shared -> exclusive/none case */ | |
1619 | if (cur_old_count < nr_old_roots && | |
1620 | cur_new_count == nr_new_roots) { | |
1621 | /* Shared->exclusive */ | |
1622 | if (cur_new_count != 0) { | |
1623 | qg->excl += num_bytes; | |
1624 | qg->excl_cmpr += num_bytes; | |
1625 | dirty = true; | |
1626 | } | |
1627 | } | |
1628 | ||
1629 | /* Exclusive/none -> exclusive/none case */ | |
1630 | if (cur_old_count == nr_old_roots && | |
1631 | cur_new_count == nr_new_roots) { | |
1632 | if (cur_old_count == 0) { | |
1633 | /* None -> exclusive/none */ | |
1634 | ||
1635 | if (cur_new_count != 0) { | |
1636 | /* None -> exclusive */ | |
1637 | qg->excl += num_bytes; | |
1638 | qg->excl_cmpr += num_bytes; | |
1639 | dirty = true; | |
1640 | } | |
1641 | /* None -> none, nothing changed */ | |
1642 | } else { | |
1643 | /* Exclusive -> exclusive/none */ | |
1644 | ||
1645 | if (cur_new_count == 0) { | |
1646 | /* Exclusive -> none */ | |
1647 | qg->excl -= num_bytes; | |
1648 | qg->excl_cmpr -= num_bytes; | |
1649 | dirty = true; | |
1650 | } | |
1651 | /* Exclusive -> exclusive, nothing changed */ | |
1652 | } | |
1653 | } | |
c05f9429 | 1654 | |
823ae5b8 QW |
1655 | if (dirty) |
1656 | qgroup_dirty(fs_info, qg); | |
1657 | } | |
1658 | return 0; | |
1659 | } | |
1660 | ||
442244c9 | 1661 | int |
550d7a2e QW |
1662 | btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, |
1663 | struct btrfs_fs_info *fs_info, | |
1664 | u64 bytenr, u64 num_bytes, | |
1665 | struct ulist *old_roots, struct ulist *new_roots) | |
1666 | { | |
1667 | struct ulist *qgroups = NULL; | |
1668 | struct ulist *tmp = NULL; | |
1669 | u64 seq; | |
1670 | u64 nr_new_roots = 0; | |
1671 | u64 nr_old_roots = 0; | |
1672 | int ret = 0; | |
1673 | ||
1674 | if (new_roots) | |
1675 | nr_new_roots = new_roots->nnodes; | |
1676 | if (old_roots) | |
1677 | nr_old_roots = old_roots->nnodes; | |
1678 | ||
1679 | if (!fs_info->quota_enabled) | |
1680 | goto out_free; | |
1681 | BUG_ON(!fs_info->quota_root); | |
1682 | ||
1683 | qgroups = ulist_alloc(GFP_NOFS); | |
1684 | if (!qgroups) { | |
1685 | ret = -ENOMEM; | |
1686 | goto out_free; | |
1687 | } | |
1688 | tmp = ulist_alloc(GFP_NOFS); | |
1689 | if (!tmp) { | |
1690 | ret = -ENOMEM; | |
1691 | goto out_free; | |
1692 | } | |
1693 | ||
1694 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
1695 | if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { | |
1696 | if (fs_info->qgroup_rescan_progress.objectid <= bytenr) { | |
1697 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
1698 | ret = 0; | |
1699 | goto out_free; | |
1700 | } | |
1701 | } | |
1702 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
1703 | ||
1704 | spin_lock(&fs_info->qgroup_lock); | |
1705 | seq = fs_info->qgroup_seq; | |
1706 | ||
1707 | /* Update old refcnts using old_roots */ | |
1708 | ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq, | |
1709 | UPDATE_OLD); | |
1710 | if (ret < 0) | |
1711 | goto out; | |
1712 | ||
1713 | /* Update new refcnts using new_roots */ | |
1714 | ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq, | |
1715 | UPDATE_NEW); | |
1716 | if (ret < 0) | |
1717 | goto out; | |
1718 | ||
1719 | qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots, | |
1720 | num_bytes, seq); | |
1721 | ||
1722 | /* | |
1723 | * Bump qgroup_seq to avoid seq overlap | |
1724 | */ | |
1725 | fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1; | |
1726 | out: | |
1727 | spin_unlock(&fs_info->qgroup_lock); | |
1728 | out_free: | |
1729 | ulist_free(tmp); | |
1730 | ulist_free(qgroups); | |
1731 | ulist_free(old_roots); | |
1732 | ulist_free(new_roots); | |
1733 | return ret; | |
1734 | } | |
1735 | ||
1736 | int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans, | |
1737 | struct btrfs_fs_info *fs_info) | |
1738 | { | |
1739 | struct btrfs_qgroup_extent_record *record; | |
1740 | struct btrfs_delayed_ref_root *delayed_refs; | |
1741 | struct ulist *new_roots = NULL; | |
1742 | struct rb_node *node; | |
9086db86 | 1743 | u64 qgroup_to_skip; |
550d7a2e QW |
1744 | int ret = 0; |
1745 | ||
1746 | delayed_refs = &trans->transaction->delayed_refs; | |
9086db86 | 1747 | qgroup_to_skip = delayed_refs->qgroup_to_skip; |
550d7a2e QW |
1748 | while ((node = rb_first(&delayed_refs->dirty_extent_root))) { |
1749 | record = rb_entry(node, struct btrfs_qgroup_extent_record, | |
1750 | node); | |
1751 | ||
1752 | if (!ret) { | |
1753 | /* | |
1754 | * Use (u64)-1 as time_seq to do special search, which | |
1755 | * doesn't lock tree or delayed_refs and search current | |
1756 | * root. It's safe inside commit_transaction(). | |
1757 | */ | |
1758 | ret = btrfs_find_all_roots(trans, fs_info, | |
1759 | record->bytenr, (u64)-1, &new_roots); | |
1760 | if (ret < 0) | |
1761 | goto cleanup; | |
9086db86 QW |
1762 | if (qgroup_to_skip) |
1763 | ulist_del(new_roots, qgroup_to_skip, 0); | |
550d7a2e QW |
1764 | ret = btrfs_qgroup_account_extent(trans, fs_info, |
1765 | record->bytenr, record->num_bytes, | |
1766 | record->old_roots, new_roots); | |
1767 | record->old_roots = NULL; | |
1768 | new_roots = NULL; | |
1769 | } | |
1770 | cleanup: | |
1771 | ulist_free(record->old_roots); | |
1772 | ulist_free(new_roots); | |
1773 | new_roots = NULL; | |
1774 | rb_erase(node, &delayed_refs->dirty_extent_root); | |
1775 | kfree(record); | |
1776 | ||
1777 | } | |
1778 | return ret; | |
1779 | } | |
1780 | ||
bed92eae AJ |
1781 | /* |
1782 | * called from commit_transaction. Writes all changed qgroups to disk. | |
1783 | */ | |
1784 | int btrfs_run_qgroups(struct btrfs_trans_handle *trans, | |
1785 | struct btrfs_fs_info *fs_info) | |
1786 | { | |
1787 | struct btrfs_root *quota_root = fs_info->quota_root; | |
1788 | int ret = 0; | |
3d7b5a28 | 1789 | int start_rescan_worker = 0; |
bed92eae AJ |
1790 | |
1791 | if (!quota_root) | |
1792 | goto out; | |
1793 | ||
3d7b5a28 JS |
1794 | if (!fs_info->quota_enabled && fs_info->pending_quota_state) |
1795 | start_rescan_worker = 1; | |
1796 | ||
bed92eae AJ |
1797 | fs_info->quota_enabled = fs_info->pending_quota_state; |
1798 | ||
1799 | spin_lock(&fs_info->qgroup_lock); | |
1800 | while (!list_empty(&fs_info->dirty_qgroups)) { | |
1801 | struct btrfs_qgroup *qgroup; | |
1802 | qgroup = list_first_entry(&fs_info->dirty_qgroups, | |
1803 | struct btrfs_qgroup, dirty); | |
1804 | list_del_init(&qgroup->dirty); | |
1805 | spin_unlock(&fs_info->qgroup_lock); | |
1806 | ret = update_qgroup_info_item(trans, quota_root, qgroup); | |
d3001ed3 DY |
1807 | if (ret) |
1808 | fs_info->qgroup_flags |= | |
1809 | BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1810 | ret = update_qgroup_limit_item(trans, quota_root, qgroup); | |
bed92eae AJ |
1811 | if (ret) |
1812 | fs_info->qgroup_flags |= | |
1813 | BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1814 | spin_lock(&fs_info->qgroup_lock); | |
1815 | } | |
1816 | if (fs_info->quota_enabled) | |
1817 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON; | |
1818 | else | |
1819 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON; | |
1820 | spin_unlock(&fs_info->qgroup_lock); | |
1821 | ||
1822 | ret = update_qgroup_status_item(trans, fs_info, quota_root); | |
1823 | if (ret) | |
1824 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1825 | ||
3d7b5a28 | 1826 | if (!ret && start_rescan_worker) { |
b382a324 JS |
1827 | ret = qgroup_rescan_init(fs_info, 0, 1); |
1828 | if (!ret) { | |
1829 | qgroup_rescan_zero_tracking(fs_info); | |
fc97fab0 QW |
1830 | btrfs_queue_work(fs_info->qgroup_rescan_workers, |
1831 | &fs_info->qgroup_rescan_work); | |
b382a324 | 1832 | } |
3d7b5a28 JS |
1833 | ret = 0; |
1834 | } | |
1835 | ||
bed92eae AJ |
1836 | out: |
1837 | ||
1838 | return ret; | |
1839 | } | |
1840 | ||
1841 | /* | |
1842 | * copy the acounting information between qgroups. This is necessary when a | |
1843 | * snapshot or a subvolume is created | |
1844 | */ | |
1845 | int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, | |
1846 | struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid, | |
1847 | struct btrfs_qgroup_inherit *inherit) | |
1848 | { | |
1849 | int ret = 0; | |
1850 | int i; | |
1851 | u64 *i_qgroups; | |
1852 | struct btrfs_root *quota_root = fs_info->quota_root; | |
1853 | struct btrfs_qgroup *srcgroup; | |
1854 | struct btrfs_qgroup *dstgroup; | |
1855 | u32 level_size = 0; | |
3f5e2d3b | 1856 | u64 nums; |
bed92eae | 1857 | |
f2f6ed3d | 1858 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1859 | if (!fs_info->quota_enabled) |
f2f6ed3d | 1860 | goto out; |
bed92eae | 1861 | |
f2f6ed3d WS |
1862 | if (!quota_root) { |
1863 | ret = -EINVAL; | |
1864 | goto out; | |
1865 | } | |
bed92eae | 1866 | |
3f5e2d3b WS |
1867 | if (inherit) { |
1868 | i_qgroups = (u64 *)(inherit + 1); | |
1869 | nums = inherit->num_qgroups + 2 * inherit->num_ref_copies + | |
1870 | 2 * inherit->num_excl_copies; | |
1871 | for (i = 0; i < nums; ++i) { | |
1872 | srcgroup = find_qgroup_rb(fs_info, *i_qgroups); | |
1873 | if (!srcgroup) { | |
1874 | ret = -EINVAL; | |
1875 | goto out; | |
1876 | } | |
09870d27 DY |
1877 | |
1878 | if ((srcgroup->qgroupid >> 48) <= (objectid >> 48)) { | |
1879 | ret = -EINVAL; | |
1880 | goto out; | |
1881 | } | |
3f5e2d3b WS |
1882 | ++i_qgroups; |
1883 | } | |
1884 | } | |
1885 | ||
bed92eae AJ |
1886 | /* |
1887 | * create a tracking group for the subvol itself | |
1888 | */ | |
1889 | ret = add_qgroup_item(trans, quota_root, objectid); | |
1890 | if (ret) | |
1891 | goto out; | |
1892 | ||
bed92eae AJ |
1893 | if (srcid) { |
1894 | struct btrfs_root *srcroot; | |
1895 | struct btrfs_key srckey; | |
bed92eae AJ |
1896 | |
1897 | srckey.objectid = srcid; | |
1898 | srckey.type = BTRFS_ROOT_ITEM_KEY; | |
1899 | srckey.offset = (u64)-1; | |
1900 | srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey); | |
1901 | if (IS_ERR(srcroot)) { | |
1902 | ret = PTR_ERR(srcroot); | |
1903 | goto out; | |
1904 | } | |
1905 | ||
1906 | rcu_read_lock(); | |
707e8a07 | 1907 | level_size = srcroot->nodesize; |
bed92eae AJ |
1908 | rcu_read_unlock(); |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * add qgroup to all inherited groups | |
1913 | */ | |
1914 | if (inherit) { | |
1915 | i_qgroups = (u64 *)(inherit + 1); | |
1916 | for (i = 0; i < inherit->num_qgroups; ++i) { | |
1917 | ret = add_qgroup_relation_item(trans, quota_root, | |
1918 | objectid, *i_qgroups); | |
1919 | if (ret) | |
1920 | goto out; | |
1921 | ret = add_qgroup_relation_item(trans, quota_root, | |
1922 | *i_qgroups, objectid); | |
1923 | if (ret) | |
1924 | goto out; | |
1925 | ++i_qgroups; | |
1926 | } | |
1927 | } | |
1928 | ||
1929 | ||
1930 | spin_lock(&fs_info->qgroup_lock); | |
1931 | ||
1932 | dstgroup = add_qgroup_rb(fs_info, objectid); | |
57a5a882 DC |
1933 | if (IS_ERR(dstgroup)) { |
1934 | ret = PTR_ERR(dstgroup); | |
bed92eae | 1935 | goto unlock; |
57a5a882 | 1936 | } |
bed92eae | 1937 | |
e8c8541a | 1938 | if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) { |
e8c8541a DY |
1939 | dstgroup->lim_flags = inherit->lim.flags; |
1940 | dstgroup->max_rfer = inherit->lim.max_rfer; | |
1941 | dstgroup->max_excl = inherit->lim.max_excl; | |
1942 | dstgroup->rsv_rfer = inherit->lim.rsv_rfer; | |
1943 | dstgroup->rsv_excl = inherit->lim.rsv_excl; | |
1510e71c DY |
1944 | |
1945 | ret = update_qgroup_limit_item(trans, quota_root, dstgroup); | |
1946 | if (ret) { | |
1947 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1948 | btrfs_info(fs_info, "unable to update quota limit for %llu", | |
1949 | dstgroup->qgroupid); | |
1950 | goto unlock; | |
1951 | } | |
e8c8541a DY |
1952 | } |
1953 | ||
bed92eae AJ |
1954 | if (srcid) { |
1955 | srcgroup = find_qgroup_rb(fs_info, srcid); | |
f3a87f1b | 1956 | if (!srcgroup) |
bed92eae | 1957 | goto unlock; |
fcebe456 JB |
1958 | |
1959 | /* | |
1960 | * We call inherit after we clone the root in order to make sure | |
1961 | * our counts don't go crazy, so at this point the only | |
1962 | * difference between the two roots should be the root node. | |
1963 | */ | |
1964 | dstgroup->rfer = srcgroup->rfer; | |
1965 | dstgroup->rfer_cmpr = srcgroup->rfer_cmpr; | |
1966 | dstgroup->excl = level_size; | |
1967 | dstgroup->excl_cmpr = level_size; | |
bed92eae AJ |
1968 | srcgroup->excl = level_size; |
1969 | srcgroup->excl_cmpr = level_size; | |
3eeb4d59 DY |
1970 | |
1971 | /* inherit the limit info */ | |
1972 | dstgroup->lim_flags = srcgroup->lim_flags; | |
1973 | dstgroup->max_rfer = srcgroup->max_rfer; | |
1974 | dstgroup->max_excl = srcgroup->max_excl; | |
1975 | dstgroup->rsv_rfer = srcgroup->rsv_rfer; | |
1976 | dstgroup->rsv_excl = srcgroup->rsv_excl; | |
1977 | ||
bed92eae AJ |
1978 | qgroup_dirty(fs_info, dstgroup); |
1979 | qgroup_dirty(fs_info, srcgroup); | |
1980 | } | |
1981 | ||
f3a87f1b | 1982 | if (!inherit) |
bed92eae AJ |
1983 | goto unlock; |
1984 | ||
1985 | i_qgroups = (u64 *)(inherit + 1); | |
1986 | for (i = 0; i < inherit->num_qgroups; ++i) { | |
1987 | ret = add_relation_rb(quota_root->fs_info, objectid, | |
1988 | *i_qgroups); | |
1989 | if (ret) | |
1990 | goto unlock; | |
1991 | ++i_qgroups; | |
1992 | } | |
1993 | ||
1994 | for (i = 0; i < inherit->num_ref_copies; ++i) { | |
1995 | struct btrfs_qgroup *src; | |
1996 | struct btrfs_qgroup *dst; | |
1997 | ||
1998 | src = find_qgroup_rb(fs_info, i_qgroups[0]); | |
1999 | dst = find_qgroup_rb(fs_info, i_qgroups[1]); | |
2000 | ||
2001 | if (!src || !dst) { | |
2002 | ret = -EINVAL; | |
2003 | goto unlock; | |
2004 | } | |
2005 | ||
2006 | dst->rfer = src->rfer - level_size; | |
2007 | dst->rfer_cmpr = src->rfer_cmpr - level_size; | |
2008 | i_qgroups += 2; | |
2009 | } | |
2010 | for (i = 0; i < inherit->num_excl_copies; ++i) { | |
2011 | struct btrfs_qgroup *src; | |
2012 | struct btrfs_qgroup *dst; | |
2013 | ||
2014 | src = find_qgroup_rb(fs_info, i_qgroups[0]); | |
2015 | dst = find_qgroup_rb(fs_info, i_qgroups[1]); | |
2016 | ||
2017 | if (!src || !dst) { | |
2018 | ret = -EINVAL; | |
2019 | goto unlock; | |
2020 | } | |
2021 | ||
2022 | dst->excl = src->excl + level_size; | |
2023 | dst->excl_cmpr = src->excl_cmpr + level_size; | |
2024 | i_qgroups += 2; | |
2025 | } | |
2026 | ||
2027 | unlock: | |
2028 | spin_unlock(&fs_info->qgroup_lock); | |
2029 | out: | |
f2f6ed3d | 2030 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
2031 | return ret; |
2032 | } | |
2033 | ||
7cf5b976 | 2034 | static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes) |
bed92eae AJ |
2035 | { |
2036 | struct btrfs_root *quota_root; | |
2037 | struct btrfs_qgroup *qgroup; | |
2038 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2039 | u64 ref_root = root->root_key.objectid; | |
2040 | int ret = 0; | |
bed92eae AJ |
2041 | struct ulist_node *unode; |
2042 | struct ulist_iterator uiter; | |
2043 | ||
2044 | if (!is_fstree(ref_root)) | |
2045 | return 0; | |
2046 | ||
2047 | if (num_bytes == 0) | |
2048 | return 0; | |
2049 | ||
2050 | spin_lock(&fs_info->qgroup_lock); | |
2051 | quota_root = fs_info->quota_root; | |
2052 | if (!quota_root) | |
2053 | goto out; | |
2054 | ||
2055 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
2056 | if (!qgroup) | |
2057 | goto out; | |
2058 | ||
2059 | /* | |
2060 | * in a first step, we check all affected qgroups if any limits would | |
2061 | * be exceeded | |
2062 | */ | |
1e8f9158 WS |
2063 | ulist_reinit(fs_info->qgroup_ulist); |
2064 | ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid, | |
3c97185c WS |
2065 | (uintptr_t)qgroup, GFP_ATOMIC); |
2066 | if (ret < 0) | |
2067 | goto out; | |
bed92eae | 2068 | ULIST_ITER_INIT(&uiter); |
1e8f9158 | 2069 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { |
bed92eae AJ |
2070 | struct btrfs_qgroup *qg; |
2071 | struct btrfs_qgroup_list *glist; | |
2072 | ||
fcebe456 | 2073 | qg = u64_to_ptr(unode->aux); |
bed92eae AJ |
2074 | |
2075 | if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) && | |
e2d1f923 | 2076 | qg->reserved + (s64)qg->rfer + num_bytes > |
720f1e20 | 2077 | qg->max_rfer) { |
bed92eae | 2078 | ret = -EDQUOT; |
720f1e20 WS |
2079 | goto out; |
2080 | } | |
bed92eae AJ |
2081 | |
2082 | if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) && | |
e2d1f923 | 2083 | qg->reserved + (s64)qg->excl + num_bytes > |
720f1e20 | 2084 | qg->max_excl) { |
bed92eae | 2085 | ret = -EDQUOT; |
720f1e20 WS |
2086 | goto out; |
2087 | } | |
bed92eae AJ |
2088 | |
2089 | list_for_each_entry(glist, &qg->groups, next_group) { | |
1e8f9158 WS |
2090 | ret = ulist_add(fs_info->qgroup_ulist, |
2091 | glist->group->qgroupid, | |
3c97185c WS |
2092 | (uintptr_t)glist->group, GFP_ATOMIC); |
2093 | if (ret < 0) | |
2094 | goto out; | |
bed92eae AJ |
2095 | } |
2096 | } | |
3c97185c | 2097 | ret = 0; |
bed92eae AJ |
2098 | /* |
2099 | * no limits exceeded, now record the reservation into all qgroups | |
2100 | */ | |
2101 | ULIST_ITER_INIT(&uiter); | |
1e8f9158 | 2102 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { |
bed92eae AJ |
2103 | struct btrfs_qgroup *qg; |
2104 | ||
fcebe456 | 2105 | qg = u64_to_ptr(unode->aux); |
bed92eae | 2106 | |
e2d1f923 | 2107 | qg->reserved += num_bytes; |
bed92eae AJ |
2108 | } |
2109 | ||
2110 | out: | |
2111 | spin_unlock(&fs_info->qgroup_lock); | |
bed92eae AJ |
2112 | return ret; |
2113 | } | |
2114 | ||
297d750b QW |
2115 | void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info, |
2116 | u64 ref_root, u64 num_bytes) | |
bed92eae AJ |
2117 | { |
2118 | struct btrfs_root *quota_root; | |
2119 | struct btrfs_qgroup *qgroup; | |
bed92eae AJ |
2120 | struct ulist_node *unode; |
2121 | struct ulist_iterator uiter; | |
3c97185c | 2122 | int ret = 0; |
bed92eae AJ |
2123 | |
2124 | if (!is_fstree(ref_root)) | |
2125 | return; | |
2126 | ||
2127 | if (num_bytes == 0) | |
2128 | return; | |
2129 | ||
2130 | spin_lock(&fs_info->qgroup_lock); | |
2131 | ||
2132 | quota_root = fs_info->quota_root; | |
2133 | if (!quota_root) | |
2134 | goto out; | |
2135 | ||
2136 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
2137 | if (!qgroup) | |
2138 | goto out; | |
2139 | ||
1e8f9158 WS |
2140 | ulist_reinit(fs_info->qgroup_ulist); |
2141 | ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid, | |
3c97185c WS |
2142 | (uintptr_t)qgroup, GFP_ATOMIC); |
2143 | if (ret < 0) | |
2144 | goto out; | |
bed92eae | 2145 | ULIST_ITER_INIT(&uiter); |
1e8f9158 | 2146 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { |
bed92eae AJ |
2147 | struct btrfs_qgroup *qg; |
2148 | struct btrfs_qgroup_list *glist; | |
2149 | ||
fcebe456 | 2150 | qg = u64_to_ptr(unode->aux); |
bed92eae | 2151 | |
e2d1f923 | 2152 | qg->reserved -= num_bytes; |
bed92eae AJ |
2153 | |
2154 | list_for_each_entry(glist, &qg->groups, next_group) { | |
1e8f9158 WS |
2155 | ret = ulist_add(fs_info->qgroup_ulist, |
2156 | glist->group->qgroupid, | |
3c97185c WS |
2157 | (uintptr_t)glist->group, GFP_ATOMIC); |
2158 | if (ret < 0) | |
2159 | goto out; | |
bed92eae AJ |
2160 | } |
2161 | } | |
2162 | ||
2163 | out: | |
2164 | spin_unlock(&fs_info->qgroup_lock); | |
bed92eae AJ |
2165 | } |
2166 | ||
7cf5b976 QW |
2167 | static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes) |
2168 | { | |
2169 | return btrfs_qgroup_free_refroot(root->fs_info, root->objectid, | |
2170 | num_bytes); | |
2171 | } | |
bed92eae AJ |
2172 | void assert_qgroups_uptodate(struct btrfs_trans_handle *trans) |
2173 | { | |
2174 | if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq) | |
2175 | return; | |
efe120a0 FH |
2176 | btrfs_err(trans->root->fs_info, |
2177 | "qgroups not uptodate in trans handle %p: list is%s empty, " | |
2178 | "seq is %#x.%x", | |
bed92eae | 2179 | trans, list_empty(&trans->qgroup_ref_list) ? "" : " not", |
fc36ed7e JS |
2180 | (u32)(trans->delayed_ref_elem.seq >> 32), |
2181 | (u32)trans->delayed_ref_elem.seq); | |
bed92eae AJ |
2182 | BUG(); |
2183 | } | |
2f232036 JS |
2184 | |
2185 | /* | |
2186 | * returns < 0 on error, 0 when more leafs are to be scanned. | |
3393168d | 2187 | * returns 1 when done. |
2f232036 JS |
2188 | */ |
2189 | static int | |
b382a324 | 2190 | qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
0a0e8b89 | 2191 | struct btrfs_trans_handle *trans) |
2f232036 JS |
2192 | { |
2193 | struct btrfs_key found; | |
0a0e8b89 | 2194 | struct extent_buffer *scratch_leaf = NULL; |
2f232036 | 2195 | struct ulist *roots = NULL; |
3284da7b | 2196 | struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem); |
fcebe456 | 2197 | u64 num_bytes; |
2f232036 JS |
2198 | int slot; |
2199 | int ret; | |
2200 | ||
2f232036 JS |
2201 | mutex_lock(&fs_info->qgroup_rescan_lock); |
2202 | ret = btrfs_search_slot_for_read(fs_info->extent_root, | |
2203 | &fs_info->qgroup_rescan_progress, | |
2204 | path, 1, 0); | |
2205 | ||
2206 | pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n", | |
c1c9ff7c | 2207 | fs_info->qgroup_rescan_progress.objectid, |
2f232036 | 2208 | fs_info->qgroup_rescan_progress.type, |
c1c9ff7c | 2209 | fs_info->qgroup_rescan_progress.offset, ret); |
2f232036 JS |
2210 | |
2211 | if (ret) { | |
2212 | /* | |
2213 | * The rescan is about to end, we will not be scanning any | |
2214 | * further blocks. We cannot unset the RESCAN flag here, because | |
2215 | * we want to commit the transaction if everything went well. | |
2216 | * To make the live accounting work in this phase, we set our | |
2217 | * scan progress pointer such that every real extent objectid | |
2218 | * will be smaller. | |
2219 | */ | |
2220 | fs_info->qgroup_rescan_progress.objectid = (u64)-1; | |
2221 | btrfs_release_path(path); | |
2222 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2223 | return ret; | |
2224 | } | |
2225 | ||
2226 | btrfs_item_key_to_cpu(path->nodes[0], &found, | |
2227 | btrfs_header_nritems(path->nodes[0]) - 1); | |
2228 | fs_info->qgroup_rescan_progress.objectid = found.objectid + 1; | |
2229 | ||
2230 | btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); | |
0a0e8b89 QW |
2231 | scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]); |
2232 | if (!scratch_leaf) { | |
2233 | ret = -ENOMEM; | |
2234 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2235 | goto out; | |
2236 | } | |
2237 | extent_buffer_get(scratch_leaf); | |
2238 | btrfs_tree_read_lock(scratch_leaf); | |
2239 | btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK); | |
2f232036 JS |
2240 | slot = path->slots[0]; |
2241 | btrfs_release_path(path); | |
2242 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2243 | ||
2244 | for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) { | |
2245 | btrfs_item_key_to_cpu(scratch_leaf, &found, slot); | |
3a6d75e8 JB |
2246 | if (found.type != BTRFS_EXTENT_ITEM_KEY && |
2247 | found.type != BTRFS_METADATA_ITEM_KEY) | |
2f232036 | 2248 | continue; |
3a6d75e8 | 2249 | if (found.type == BTRFS_METADATA_ITEM_KEY) |
707e8a07 | 2250 | num_bytes = fs_info->extent_root->nodesize; |
3a6d75e8 JB |
2251 | else |
2252 | num_bytes = found.offset; | |
2253 | ||
fcebe456 JB |
2254 | ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0, |
2255 | &roots); | |
2f232036 JS |
2256 | if (ret < 0) |
2257 | goto out; | |
9d220c95 QW |
2258 | /* For rescan, just pass old_roots as NULL */ |
2259 | ret = btrfs_qgroup_account_extent(trans, fs_info, | |
2260 | found.objectid, num_bytes, NULL, roots); | |
2261 | if (ret < 0) | |
fcebe456 | 2262 | goto out; |
2f232036 | 2263 | } |
2f232036 | 2264 | out: |
0a0e8b89 QW |
2265 | if (scratch_leaf) { |
2266 | btrfs_tree_read_unlock_blocking(scratch_leaf); | |
2267 | free_extent_buffer(scratch_leaf); | |
2268 | } | |
2f232036 JS |
2269 | btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
2270 | ||
2271 | return ret; | |
2272 | } | |
2273 | ||
d458b054 | 2274 | static void btrfs_qgroup_rescan_worker(struct btrfs_work *work) |
2f232036 | 2275 | { |
b382a324 JS |
2276 | struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info, |
2277 | qgroup_rescan_work); | |
2f232036 JS |
2278 | struct btrfs_path *path; |
2279 | struct btrfs_trans_handle *trans = NULL; | |
2f232036 | 2280 | int err = -ENOMEM; |
53b7cde9 | 2281 | int ret = 0; |
2f232036 JS |
2282 | |
2283 | path = btrfs_alloc_path(); | |
2284 | if (!path) | |
2285 | goto out; | |
2f232036 JS |
2286 | |
2287 | err = 0; | |
7343dd61 | 2288 | while (!err && !btrfs_fs_closing(fs_info)) { |
2f232036 JS |
2289 | trans = btrfs_start_transaction(fs_info->fs_root, 0); |
2290 | if (IS_ERR(trans)) { | |
2291 | err = PTR_ERR(trans); | |
2292 | break; | |
2293 | } | |
2294 | if (!fs_info->quota_enabled) { | |
2295 | err = -EINTR; | |
2296 | } else { | |
0a0e8b89 | 2297 | err = qgroup_rescan_leaf(fs_info, path, trans); |
2f232036 JS |
2298 | } |
2299 | if (err > 0) | |
2300 | btrfs_commit_transaction(trans, fs_info->fs_root); | |
2301 | else | |
2302 | btrfs_end_transaction(trans, fs_info->fs_root); | |
2303 | } | |
2304 | ||
2305 | out: | |
2f232036 | 2306 | btrfs_free_path(path); |
2f232036 JS |
2307 | |
2308 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
7343dd61 JM |
2309 | if (!btrfs_fs_closing(fs_info)) |
2310 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2f232036 | 2311 | |
3393168d | 2312 | if (err > 0 && |
2f232036 JS |
2313 | fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) { |
2314 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2315 | } else if (err < 0) { | |
2316 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2317 | } | |
2318 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2319 | ||
53b7cde9 QW |
2320 | /* |
2321 | * only update status, since the previous part has alreay updated the | |
2322 | * qgroup info. | |
2323 | */ | |
2324 | trans = btrfs_start_transaction(fs_info->quota_root, 1); | |
2325 | if (IS_ERR(trans)) { | |
2326 | err = PTR_ERR(trans); | |
2327 | btrfs_err(fs_info, | |
2328 | "fail to start transaction for status update: %d\n", | |
2329 | err); | |
2330 | goto done; | |
2331 | } | |
2332 | ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root); | |
2333 | if (ret < 0) { | |
2334 | err = ret; | |
9c8b35b1 | 2335 | btrfs_err(fs_info, "fail to update qgroup status: %d\n", err); |
53b7cde9 QW |
2336 | } |
2337 | btrfs_end_transaction(trans, fs_info->quota_root); | |
2338 | ||
7343dd61 JM |
2339 | if (btrfs_fs_closing(fs_info)) { |
2340 | btrfs_info(fs_info, "qgroup scan paused"); | |
2341 | } else if (err >= 0) { | |
efe120a0 | 2342 | btrfs_info(fs_info, "qgroup scan completed%s", |
3393168d | 2343 | err > 0 ? " (inconsistency flag cleared)" : ""); |
2f232036 | 2344 | } else { |
efe120a0 | 2345 | btrfs_err(fs_info, "qgroup scan failed with %d", err); |
2f232036 | 2346 | } |
57254b6e | 2347 | |
53b7cde9 | 2348 | done: |
57254b6e | 2349 | complete_all(&fs_info->qgroup_rescan_completion); |
2f232036 JS |
2350 | } |
2351 | ||
b382a324 JS |
2352 | /* |
2353 | * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all | |
2354 | * memory required for the rescan context. | |
2355 | */ | |
2356 | static int | |
2357 | qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid, | |
2358 | int init_flags) | |
2f232036 JS |
2359 | { |
2360 | int ret = 0; | |
2f232036 | 2361 | |
b382a324 JS |
2362 | if (!init_flags && |
2363 | (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) || | |
2364 | !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) { | |
2365 | ret = -EINVAL; | |
2366 | goto err; | |
2367 | } | |
2f232036 JS |
2368 | |
2369 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
2370 | spin_lock(&fs_info->qgroup_lock); | |
b382a324 JS |
2371 | |
2372 | if (init_flags) { | |
2373 | if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) | |
2374 | ret = -EINPROGRESS; | |
2375 | else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) | |
2376 | ret = -EINVAL; | |
2377 | ||
2378 | if (ret) { | |
2379 | spin_unlock(&fs_info->qgroup_lock); | |
2380 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2381 | goto err; | |
2382 | } | |
b382a324 | 2383 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN; |
2f232036 JS |
2384 | } |
2385 | ||
2f232036 JS |
2386 | memset(&fs_info->qgroup_rescan_progress, 0, |
2387 | sizeof(fs_info->qgroup_rescan_progress)); | |
b382a324 | 2388 | fs_info->qgroup_rescan_progress.objectid = progress_objectid; |
190631f1 | 2389 | init_completion(&fs_info->qgroup_rescan_completion); |
b382a324 JS |
2390 | |
2391 | spin_unlock(&fs_info->qgroup_lock); | |
2392 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2393 | ||
b382a324 JS |
2394 | memset(&fs_info->qgroup_rescan_work, 0, |
2395 | sizeof(fs_info->qgroup_rescan_work)); | |
fc97fab0 | 2396 | btrfs_init_work(&fs_info->qgroup_rescan_work, |
9e0af237 | 2397 | btrfs_qgroup_rescan_helper, |
fc97fab0 | 2398 | btrfs_qgroup_rescan_worker, NULL, NULL); |
b382a324 JS |
2399 | |
2400 | if (ret) { | |
2401 | err: | |
efe120a0 | 2402 | btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret); |
b382a324 JS |
2403 | return ret; |
2404 | } | |
2405 | ||
2406 | return 0; | |
2407 | } | |
2408 | ||
2409 | static void | |
2410 | qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info) | |
2411 | { | |
2412 | struct rb_node *n; | |
2413 | struct btrfs_qgroup *qgroup; | |
2414 | ||
2415 | spin_lock(&fs_info->qgroup_lock); | |
2f232036 JS |
2416 | /* clear all current qgroup tracking information */ |
2417 | for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) { | |
2418 | qgroup = rb_entry(n, struct btrfs_qgroup, node); | |
2419 | qgroup->rfer = 0; | |
2420 | qgroup->rfer_cmpr = 0; | |
2421 | qgroup->excl = 0; | |
2422 | qgroup->excl_cmpr = 0; | |
2423 | } | |
2424 | spin_unlock(&fs_info->qgroup_lock); | |
b382a324 | 2425 | } |
2f232036 | 2426 | |
b382a324 JS |
2427 | int |
2428 | btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info) | |
2429 | { | |
2430 | int ret = 0; | |
2431 | struct btrfs_trans_handle *trans; | |
2432 | ||
2433 | ret = qgroup_rescan_init(fs_info, 0, 1); | |
2434 | if (ret) | |
2435 | return ret; | |
2436 | ||
2437 | /* | |
2438 | * We have set the rescan_progress to 0, which means no more | |
2439 | * delayed refs will be accounted by btrfs_qgroup_account_ref. | |
2440 | * However, btrfs_qgroup_account_ref may be right after its call | |
2441 | * to btrfs_find_all_roots, in which case it would still do the | |
2442 | * accounting. | |
2443 | * To solve this, we're committing the transaction, which will | |
2444 | * ensure we run all delayed refs and only after that, we are | |
2445 | * going to clear all tracking information for a clean start. | |
2446 | */ | |
2447 | ||
2448 | trans = btrfs_join_transaction(fs_info->fs_root); | |
2449 | if (IS_ERR(trans)) { | |
2450 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2451 | return PTR_ERR(trans); | |
2452 | } | |
2453 | ret = btrfs_commit_transaction(trans, fs_info->fs_root); | |
2454 | if (ret) { | |
2455 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2456 | return ret; | |
2457 | } | |
2458 | ||
2459 | qgroup_rescan_zero_tracking(fs_info); | |
2460 | ||
fc97fab0 QW |
2461 | btrfs_queue_work(fs_info->qgroup_rescan_workers, |
2462 | &fs_info->qgroup_rescan_work); | |
2f232036 JS |
2463 | |
2464 | return 0; | |
2465 | } | |
57254b6e JS |
2466 | |
2467 | int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info) | |
2468 | { | |
2469 | int running; | |
2470 | int ret = 0; | |
2471 | ||
2472 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
2473 | spin_lock(&fs_info->qgroup_lock); | |
2474 | running = fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2475 | spin_unlock(&fs_info->qgroup_lock); | |
2476 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2477 | ||
2478 | if (running) | |
2479 | ret = wait_for_completion_interruptible( | |
2480 | &fs_info->qgroup_rescan_completion); | |
2481 | ||
2482 | return ret; | |
2483 | } | |
b382a324 JS |
2484 | |
2485 | /* | |
2486 | * this is only called from open_ctree where we're still single threaded, thus | |
2487 | * locking is omitted here. | |
2488 | */ | |
2489 | void | |
2490 | btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info) | |
2491 | { | |
2492 | if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) | |
fc97fab0 QW |
2493 | btrfs_queue_work(fs_info->qgroup_rescan_workers, |
2494 | &fs_info->qgroup_rescan_work); | |
b382a324 | 2495 | } |
52472553 QW |
2496 | |
2497 | /* | |
2498 | * Reserve qgroup space for range [start, start + len). | |
2499 | * | |
2500 | * This function will either reserve space from related qgroups or doing | |
2501 | * nothing if the range is already reserved. | |
2502 | * | |
2503 | * Return 0 for successful reserve | |
2504 | * Return <0 for error (including -EQUOT) | |
2505 | * | |
2506 | * NOTE: this function may sleep for memory allocation. | |
2507 | */ | |
2508 | int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len) | |
2509 | { | |
2510 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2511 | struct extent_changeset changeset; | |
2512 | struct ulist_node *unode; | |
2513 | struct ulist_iterator uiter; | |
2514 | int ret; | |
2515 | ||
2516 | if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) || | |
2517 | len == 0) | |
2518 | return 0; | |
2519 | ||
2520 | changeset.bytes_changed = 0; | |
2521 | changeset.range_changed = ulist_alloc(GFP_NOFS); | |
52472553 QW |
2522 | ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start, |
2523 | start + len -1, EXTENT_QGROUP_RESERVED, GFP_NOFS, | |
2524 | &changeset); | |
81fb6f77 QW |
2525 | trace_btrfs_qgroup_reserve_data(inode, start, len, |
2526 | changeset.bytes_changed, | |
2527 | QGROUP_RESERVE); | |
52472553 QW |
2528 | if (ret < 0) |
2529 | goto cleanup; | |
7cf5b976 | 2530 | ret = qgroup_reserve(root, changeset.bytes_changed); |
52472553 QW |
2531 | if (ret < 0) |
2532 | goto cleanup; | |
2533 | ||
2534 | ulist_free(changeset.range_changed); | |
2535 | return ret; | |
2536 | ||
2537 | cleanup: | |
2538 | /* cleanup already reserved ranges */ | |
2539 | ULIST_ITER_INIT(&uiter); | |
2540 | while ((unode = ulist_next(changeset.range_changed, &uiter))) | |
2541 | clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val, | |
2542 | unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL, | |
2543 | GFP_NOFS); | |
2544 | ulist_free(changeset.range_changed); | |
2545 | return ret; | |
2546 | } | |
f695fdce QW |
2547 | |
2548 | static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len, | |
2549 | int free) | |
2550 | { | |
2551 | struct extent_changeset changeset; | |
81fb6f77 | 2552 | int trace_op = QGROUP_RELEASE; |
f695fdce QW |
2553 | int ret; |
2554 | ||
2555 | changeset.bytes_changed = 0; | |
2556 | changeset.range_changed = ulist_alloc(GFP_NOFS); | |
2557 | if (!changeset.range_changed) | |
2558 | return -ENOMEM; | |
2559 | ||
2560 | ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start, | |
2561 | start + len -1, EXTENT_QGROUP_RESERVED, GFP_NOFS, | |
2562 | &changeset); | |
2563 | if (ret < 0) | |
2564 | goto out; | |
2565 | ||
81fb6f77 | 2566 | if (free) { |
7cf5b976 | 2567 | qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed); |
81fb6f77 QW |
2568 | trace_op = QGROUP_FREE; |
2569 | } | |
2570 | trace_btrfs_qgroup_release_data(inode, start, len, | |
2571 | changeset.bytes_changed, trace_op); | |
f695fdce QW |
2572 | out: |
2573 | ulist_free(changeset.range_changed); | |
2574 | return ret; | |
2575 | } | |
2576 | ||
2577 | /* | |
2578 | * Free a reserved space range from io_tree and related qgroups | |
2579 | * | |
2580 | * Should be called when a range of pages get invalidated before reaching disk. | |
2581 | * Or for error cleanup case. | |
2582 | * | |
2583 | * For data written to disk, use btrfs_qgroup_release_data(). | |
2584 | * | |
2585 | * NOTE: This function may sleep for memory allocation. | |
2586 | */ | |
2587 | int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len) | |
2588 | { | |
2589 | return __btrfs_qgroup_release_data(inode, start, len, 1); | |
2590 | } | |
2591 | ||
2592 | /* | |
2593 | * Release a reserved space range from io_tree only. | |
2594 | * | |
2595 | * Should be called when a range of pages get written to disk and corresponding | |
2596 | * FILE_EXTENT is inserted into corresponding root. | |
2597 | * | |
2598 | * Since new qgroup accounting framework will only update qgroup numbers at | |
2599 | * commit_transaction() time, its reserved space shouldn't be freed from | |
2600 | * related qgroups. | |
2601 | * | |
2602 | * But we should release the range from io_tree, to allow further write to be | |
2603 | * COWed. | |
2604 | * | |
2605 | * NOTE: This function may sleep for memory allocation. | |
2606 | */ | |
2607 | int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len) | |
2608 | { | |
2609 | return __btrfs_qgroup_release_data(inode, start, len, 0); | |
2610 | } | |
55eeaf05 QW |
2611 | |
2612 | int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes) | |
2613 | { | |
2614 | int ret; | |
2615 | ||
2616 | if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) || | |
2617 | num_bytes == 0) | |
2618 | return 0; | |
2619 | ||
2620 | BUG_ON(num_bytes != round_down(num_bytes, root->nodesize)); | |
7cf5b976 | 2621 | ret = qgroup_reserve(root, num_bytes); |
55eeaf05 QW |
2622 | if (ret < 0) |
2623 | return ret; | |
2624 | atomic_add(num_bytes, &root->qgroup_meta_rsv); | |
2625 | return ret; | |
2626 | } | |
2627 | ||
2628 | void btrfs_qgroup_free_meta_all(struct btrfs_root *root) | |
2629 | { | |
2630 | int reserved; | |
2631 | ||
2632 | if (!root->fs_info->quota_enabled || !is_fstree(root->objectid)) | |
2633 | return; | |
2634 | ||
2635 | reserved = atomic_xchg(&root->qgroup_meta_rsv, 0); | |
2636 | if (reserved == 0) | |
2637 | return; | |
7cf5b976 | 2638 | qgroup_free(root, reserved); |
55eeaf05 QW |
2639 | } |
2640 | ||
2641 | void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes) | |
2642 | { | |
2643 | if (!root->fs_info->quota_enabled || !is_fstree(root->objectid)) | |
2644 | return; | |
2645 | ||
2646 | BUG_ON(num_bytes != round_down(num_bytes, root->nodesize)); | |
2647 | WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes); | |
2648 | atomic_sub(num_bytes, &root->qgroup_meta_rsv); | |
7cf5b976 | 2649 | qgroup_free(root, num_bytes); |
55eeaf05 | 2650 | } |
56fa9d07 QW |
2651 | |
2652 | /* | |
2653 | * Check qgroup reserved space leaking, normally at destory inode | |
2654 | * time | |
2655 | */ | |
2656 | void btrfs_qgroup_check_reserved_leak(struct inode *inode) | |
2657 | { | |
2658 | struct extent_changeset changeset; | |
2659 | struct ulist_node *unode; | |
2660 | struct ulist_iterator iter; | |
2661 | int ret; | |
2662 | ||
2663 | changeset.bytes_changed = 0; | |
2664 | changeset.range_changed = ulist_alloc(GFP_NOFS); | |
2665 | if (WARN_ON(!changeset.range_changed)) | |
2666 | return; | |
2667 | ||
2668 | ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1, | |
2669 | EXTENT_QGROUP_RESERVED, GFP_NOFS, &changeset); | |
2670 | ||
2671 | WARN_ON(ret < 0); | |
2672 | if (WARN_ON(changeset.bytes_changed)) { | |
2673 | ULIST_ITER_INIT(&iter); | |
2674 | while ((unode = ulist_next(changeset.range_changed, &iter))) { | |
2675 | btrfs_warn(BTRFS_I(inode)->root->fs_info, | |
2676 | "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu", | |
2677 | inode->i_ino, unode->val, unode->aux); | |
2678 | } | |
2679 | qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed); | |
2680 | } | |
2681 | ulist_free(changeset.range_changed); | |
2682 | } |