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 | ||
f59c0347 QW |
50 | /* |
51 | * Helpers to access qgroup reservation | |
52 | * | |
53 | * Callers should ensure the lock context and type are valid | |
54 | */ | |
55 | ||
56 | static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup) | |
57 | { | |
58 | u64 ret = 0; | |
59 | int i; | |
60 | ||
61 | for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) | |
62 | ret += qgroup->rsv.values[i]; | |
63 | ||
64 | return ret; | |
65 | } | |
66 | ||
67 | #ifdef CONFIG_BTRFS_DEBUG | |
68 | static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type) | |
69 | { | |
70 | if (type == BTRFS_QGROUP_RSV_DATA) | |
71 | return "data"; | |
733e03a0 QW |
72 | if (type == BTRFS_QGROUP_RSV_META_PERTRANS) |
73 | return "meta_pertrans"; | |
74 | if (type == BTRFS_QGROUP_RSV_META_PREALLOC) | |
75 | return "meta_prealloc"; | |
f59c0347 QW |
76 | return NULL; |
77 | } | |
78 | #endif | |
79 | ||
64ee4e75 QW |
80 | static void qgroup_rsv_add(struct btrfs_fs_info *fs_info, |
81 | struct btrfs_qgroup *qgroup, u64 num_bytes, | |
f59c0347 QW |
82 | enum btrfs_qgroup_rsv_type type) |
83 | { | |
64ee4e75 | 84 | trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type); |
f59c0347 QW |
85 | qgroup->rsv.values[type] += num_bytes; |
86 | } | |
87 | ||
64ee4e75 QW |
88 | static void qgroup_rsv_release(struct btrfs_fs_info *fs_info, |
89 | struct btrfs_qgroup *qgroup, u64 num_bytes, | |
f59c0347 QW |
90 | enum btrfs_qgroup_rsv_type type) |
91 | { | |
64ee4e75 | 92 | trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type); |
f59c0347 QW |
93 | if (qgroup->rsv.values[type] >= num_bytes) { |
94 | qgroup->rsv.values[type] -= num_bytes; | |
95 | return; | |
96 | } | |
97 | #ifdef CONFIG_BTRFS_DEBUG | |
98 | WARN_RATELIMIT(1, | |
99 | "qgroup %llu %s reserved space underflow, have %llu to free %llu", | |
100 | qgroup->qgroupid, qgroup_rsv_type_str(type), | |
101 | qgroup->rsv.values[type], num_bytes); | |
102 | #endif | |
103 | qgroup->rsv.values[type] = 0; | |
104 | } | |
105 | ||
64ee4e75 QW |
106 | static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info, |
107 | struct btrfs_qgroup *dest, | |
108 | struct btrfs_qgroup *src) | |
f59c0347 QW |
109 | { |
110 | int i; | |
111 | ||
112 | for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) | |
64ee4e75 | 113 | qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i); |
f59c0347 QW |
114 | } |
115 | ||
64ee4e75 QW |
116 | static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info, |
117 | struct btrfs_qgroup *dest, | |
f59c0347 QW |
118 | struct btrfs_qgroup *src) |
119 | { | |
120 | int i; | |
121 | ||
122 | for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) | |
64ee4e75 | 123 | qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i); |
f59c0347 QW |
124 | } |
125 | ||
9c542136 QW |
126 | static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq, |
127 | int mod) | |
128 | { | |
129 | if (qg->old_refcnt < seq) | |
130 | qg->old_refcnt = seq; | |
131 | qg->old_refcnt += mod; | |
132 | } | |
133 | ||
134 | static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq, | |
135 | int mod) | |
136 | { | |
137 | if (qg->new_refcnt < seq) | |
138 | qg->new_refcnt = seq; | |
139 | qg->new_refcnt += mod; | |
140 | } | |
141 | ||
142 | static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq) | |
143 | { | |
144 | if (qg->old_refcnt < seq) | |
145 | return 0; | |
146 | return qg->old_refcnt - seq; | |
147 | } | |
148 | ||
149 | static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq) | |
150 | { | |
151 | if (qg->new_refcnt < seq) | |
152 | return 0; | |
153 | return qg->new_refcnt - seq; | |
154 | } | |
155 | ||
bed92eae AJ |
156 | /* |
157 | * glue structure to represent the relations between qgroups. | |
158 | */ | |
159 | struct btrfs_qgroup_list { | |
160 | struct list_head next_group; | |
161 | struct list_head next_member; | |
162 | struct btrfs_qgroup *group; | |
163 | struct btrfs_qgroup *member; | |
164 | }; | |
165 | ||
ef2fff64 DS |
166 | static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg) |
167 | { | |
168 | return (u64)(uintptr_t)qg; | |
169 | } | |
170 | ||
171 | static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n) | |
172 | { | |
173 | return (struct btrfs_qgroup *)(uintptr_t)n->aux; | |
174 | } | |
fcebe456 | 175 | |
b382a324 JS |
176 | static int |
177 | qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid, | |
178 | int init_flags); | |
179 | static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info); | |
2f232036 | 180 | |
58400fce | 181 | /* must be called with qgroup_ioctl_lock held */ |
bed92eae AJ |
182 | static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info, |
183 | u64 qgroupid) | |
184 | { | |
185 | struct rb_node *n = fs_info->qgroup_tree.rb_node; | |
186 | struct btrfs_qgroup *qgroup; | |
187 | ||
188 | while (n) { | |
189 | qgroup = rb_entry(n, struct btrfs_qgroup, node); | |
190 | if (qgroup->qgroupid < qgroupid) | |
191 | n = n->rb_left; | |
192 | else if (qgroup->qgroupid > qgroupid) | |
193 | n = n->rb_right; | |
194 | else | |
195 | return qgroup; | |
196 | } | |
197 | return NULL; | |
198 | } | |
199 | ||
200 | /* must be called with qgroup_lock held */ | |
201 | static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info, | |
202 | u64 qgroupid) | |
203 | { | |
204 | struct rb_node **p = &fs_info->qgroup_tree.rb_node; | |
205 | struct rb_node *parent = NULL; | |
206 | struct btrfs_qgroup *qgroup; | |
207 | ||
208 | while (*p) { | |
209 | parent = *p; | |
210 | qgroup = rb_entry(parent, struct btrfs_qgroup, node); | |
211 | ||
212 | if (qgroup->qgroupid < qgroupid) | |
213 | p = &(*p)->rb_left; | |
214 | else if (qgroup->qgroupid > qgroupid) | |
215 | p = &(*p)->rb_right; | |
216 | else | |
217 | return qgroup; | |
218 | } | |
219 | ||
220 | qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC); | |
221 | if (!qgroup) | |
222 | return ERR_PTR(-ENOMEM); | |
223 | ||
224 | qgroup->qgroupid = qgroupid; | |
225 | INIT_LIST_HEAD(&qgroup->groups); | |
226 | INIT_LIST_HEAD(&qgroup->members); | |
227 | INIT_LIST_HEAD(&qgroup->dirty); | |
228 | ||
229 | rb_link_node(&qgroup->node, parent, p); | |
230 | rb_insert_color(&qgroup->node, &fs_info->qgroup_tree); | |
231 | ||
232 | return qgroup; | |
233 | } | |
234 | ||
4082bd3d | 235 | static void __del_qgroup_rb(struct btrfs_qgroup *qgroup) |
bed92eae | 236 | { |
bed92eae AJ |
237 | struct btrfs_qgroup_list *list; |
238 | ||
bed92eae | 239 | list_del(&qgroup->dirty); |
bed92eae AJ |
240 | while (!list_empty(&qgroup->groups)) { |
241 | list = list_first_entry(&qgroup->groups, | |
242 | struct btrfs_qgroup_list, next_group); | |
243 | list_del(&list->next_group); | |
244 | list_del(&list->next_member); | |
245 | kfree(list); | |
246 | } | |
247 | ||
248 | while (!list_empty(&qgroup->members)) { | |
249 | list = list_first_entry(&qgroup->members, | |
250 | struct btrfs_qgroup_list, next_member); | |
251 | list_del(&list->next_group); | |
252 | list_del(&list->next_member); | |
253 | kfree(list); | |
254 | } | |
255 | kfree(qgroup); | |
4082bd3d | 256 | } |
bed92eae | 257 | |
4082bd3d WS |
258 | /* must be called with qgroup_lock held */ |
259 | static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid) | |
260 | { | |
261 | struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid); | |
262 | ||
263 | if (!qgroup) | |
264 | return -ENOENT; | |
265 | ||
266 | rb_erase(&qgroup->node, &fs_info->qgroup_tree); | |
267 | __del_qgroup_rb(qgroup); | |
bed92eae AJ |
268 | return 0; |
269 | } | |
270 | ||
271 | /* must be called with qgroup_lock held */ | |
272 | static int add_relation_rb(struct btrfs_fs_info *fs_info, | |
273 | u64 memberid, u64 parentid) | |
274 | { | |
275 | struct btrfs_qgroup *member; | |
276 | struct btrfs_qgroup *parent; | |
277 | struct btrfs_qgroup_list *list; | |
278 | ||
279 | member = find_qgroup_rb(fs_info, memberid); | |
280 | parent = find_qgroup_rb(fs_info, parentid); | |
281 | if (!member || !parent) | |
282 | return -ENOENT; | |
283 | ||
284 | list = kzalloc(sizeof(*list), GFP_ATOMIC); | |
285 | if (!list) | |
286 | return -ENOMEM; | |
287 | ||
288 | list->group = parent; | |
289 | list->member = member; | |
290 | list_add_tail(&list->next_group, &member->groups); | |
291 | list_add_tail(&list->next_member, &parent->members); | |
292 | ||
293 | return 0; | |
294 | } | |
295 | ||
296 | /* must be called with qgroup_lock held */ | |
297 | static int del_relation_rb(struct btrfs_fs_info *fs_info, | |
298 | u64 memberid, u64 parentid) | |
299 | { | |
300 | struct btrfs_qgroup *member; | |
301 | struct btrfs_qgroup *parent; | |
302 | struct btrfs_qgroup_list *list; | |
303 | ||
304 | member = find_qgroup_rb(fs_info, memberid); | |
305 | parent = find_qgroup_rb(fs_info, parentid); | |
306 | if (!member || !parent) | |
307 | return -ENOENT; | |
308 | ||
309 | list_for_each_entry(list, &member->groups, next_group) { | |
310 | if (list->group == parent) { | |
311 | list_del(&list->next_group); | |
312 | list_del(&list->next_member); | |
313 | kfree(list); | |
314 | return 0; | |
315 | } | |
316 | } | |
317 | return -ENOENT; | |
318 | } | |
319 | ||
faa2dbf0 JB |
320 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
321 | int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid, | |
322 | u64 rfer, u64 excl) | |
323 | { | |
324 | struct btrfs_qgroup *qgroup; | |
325 | ||
326 | qgroup = find_qgroup_rb(fs_info, qgroupid); | |
327 | if (!qgroup) | |
328 | return -EINVAL; | |
329 | if (qgroup->rfer != rfer || qgroup->excl != excl) | |
330 | return -EINVAL; | |
331 | return 0; | |
332 | } | |
333 | #endif | |
334 | ||
bed92eae AJ |
335 | /* |
336 | * The full config is read in one go, only called from open_ctree() | |
337 | * It doesn't use any locking, as at this point we're still single-threaded | |
338 | */ | |
339 | int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info) | |
340 | { | |
341 | struct btrfs_key key; | |
342 | struct btrfs_key found_key; | |
343 | struct btrfs_root *quota_root = fs_info->quota_root; | |
344 | struct btrfs_path *path = NULL; | |
345 | struct extent_buffer *l; | |
346 | int slot; | |
347 | int ret = 0; | |
348 | u64 flags = 0; | |
b382a324 | 349 | u64 rescan_progress = 0; |
bed92eae | 350 | |
afcdd129 | 351 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
bed92eae AJ |
352 | return 0; |
353 | ||
323b88f4 | 354 | fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL); |
1e8f9158 WS |
355 | if (!fs_info->qgroup_ulist) { |
356 | ret = -ENOMEM; | |
357 | goto out; | |
358 | } | |
359 | ||
bed92eae AJ |
360 | path = btrfs_alloc_path(); |
361 | if (!path) { | |
362 | ret = -ENOMEM; | |
363 | goto out; | |
364 | } | |
365 | ||
366 | /* default this to quota off, in case no status key is found */ | |
367 | fs_info->qgroup_flags = 0; | |
368 | ||
369 | /* | |
370 | * pass 1: read status, all qgroup infos and limits | |
371 | */ | |
372 | key.objectid = 0; | |
373 | key.type = 0; | |
374 | key.offset = 0; | |
375 | ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1); | |
376 | if (ret) | |
377 | goto out; | |
378 | ||
379 | while (1) { | |
380 | struct btrfs_qgroup *qgroup; | |
381 | ||
382 | slot = path->slots[0]; | |
383 | l = path->nodes[0]; | |
384 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
385 | ||
386 | if (found_key.type == BTRFS_QGROUP_STATUS_KEY) { | |
387 | struct btrfs_qgroup_status_item *ptr; | |
388 | ||
389 | ptr = btrfs_item_ptr(l, slot, | |
390 | struct btrfs_qgroup_status_item); | |
391 | ||
392 | if (btrfs_qgroup_status_version(l, ptr) != | |
393 | BTRFS_QGROUP_STATUS_VERSION) { | |
efe120a0 FH |
394 | btrfs_err(fs_info, |
395 | "old qgroup version, quota disabled"); | |
bed92eae AJ |
396 | goto out; |
397 | } | |
398 | if (btrfs_qgroup_status_generation(l, ptr) != | |
399 | fs_info->generation) { | |
400 | flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
efe120a0 | 401 | btrfs_err(fs_info, |
5d163e0e | 402 | "qgroup generation mismatch, marked as inconsistent"); |
bed92eae AJ |
403 | } |
404 | fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, | |
405 | ptr); | |
b382a324 | 406 | rescan_progress = btrfs_qgroup_status_rescan(l, ptr); |
bed92eae AJ |
407 | goto next1; |
408 | } | |
409 | ||
410 | if (found_key.type != BTRFS_QGROUP_INFO_KEY && | |
411 | found_key.type != BTRFS_QGROUP_LIMIT_KEY) | |
412 | goto next1; | |
413 | ||
414 | qgroup = find_qgroup_rb(fs_info, found_key.offset); | |
415 | if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) || | |
416 | (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) { | |
d41e36a0 | 417 | btrfs_err(fs_info, "inconsistent qgroup config"); |
bed92eae AJ |
418 | flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; |
419 | } | |
420 | if (!qgroup) { | |
421 | qgroup = add_qgroup_rb(fs_info, found_key.offset); | |
422 | if (IS_ERR(qgroup)) { | |
423 | ret = PTR_ERR(qgroup); | |
424 | goto out; | |
425 | } | |
426 | } | |
427 | switch (found_key.type) { | |
428 | case BTRFS_QGROUP_INFO_KEY: { | |
429 | struct btrfs_qgroup_info_item *ptr; | |
430 | ||
431 | ptr = btrfs_item_ptr(l, slot, | |
432 | struct btrfs_qgroup_info_item); | |
433 | qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr); | |
434 | qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr); | |
435 | qgroup->excl = btrfs_qgroup_info_excl(l, ptr); | |
436 | qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr); | |
437 | /* generation currently unused */ | |
438 | break; | |
439 | } | |
440 | case BTRFS_QGROUP_LIMIT_KEY: { | |
441 | struct btrfs_qgroup_limit_item *ptr; | |
442 | ||
443 | ptr = btrfs_item_ptr(l, slot, | |
444 | struct btrfs_qgroup_limit_item); | |
445 | qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr); | |
446 | qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr); | |
447 | qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr); | |
448 | qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr); | |
449 | qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr); | |
450 | break; | |
451 | } | |
452 | } | |
453 | next1: | |
454 | ret = btrfs_next_item(quota_root, path); | |
455 | if (ret < 0) | |
456 | goto out; | |
457 | if (ret) | |
458 | break; | |
459 | } | |
460 | btrfs_release_path(path); | |
461 | ||
462 | /* | |
463 | * pass 2: read all qgroup relations | |
464 | */ | |
465 | key.objectid = 0; | |
466 | key.type = BTRFS_QGROUP_RELATION_KEY; | |
467 | key.offset = 0; | |
468 | ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0); | |
469 | if (ret) | |
470 | goto out; | |
471 | while (1) { | |
472 | slot = path->slots[0]; | |
473 | l = path->nodes[0]; | |
474 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
475 | ||
476 | if (found_key.type != BTRFS_QGROUP_RELATION_KEY) | |
477 | goto next2; | |
478 | ||
479 | if (found_key.objectid > found_key.offset) { | |
480 | /* parent <- member, not needed to build config */ | |
481 | /* FIXME should we omit the key completely? */ | |
482 | goto next2; | |
483 | } | |
484 | ||
485 | ret = add_relation_rb(fs_info, found_key.objectid, | |
486 | found_key.offset); | |
ff24858c | 487 | if (ret == -ENOENT) { |
efe120a0 FH |
488 | btrfs_warn(fs_info, |
489 | "orphan qgroup relation 0x%llx->0x%llx", | |
c1c9ff7c | 490 | found_key.objectid, found_key.offset); |
ff24858c AJ |
491 | ret = 0; /* ignore the error */ |
492 | } | |
bed92eae AJ |
493 | if (ret) |
494 | goto out; | |
495 | next2: | |
496 | ret = btrfs_next_item(quota_root, path); | |
497 | if (ret < 0) | |
498 | goto out; | |
499 | if (ret) | |
500 | break; | |
501 | } | |
502 | out: | |
503 | fs_info->qgroup_flags |= flags; | |
afcdd129 JB |
504 | if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) |
505 | clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); | |
506 | else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN && | |
507 | ret >= 0) | |
b382a324 | 508 | ret = qgroup_rescan_init(fs_info, rescan_progress, 0); |
bed92eae AJ |
509 | btrfs_free_path(path); |
510 | ||
eb1716af | 511 | if (ret < 0) { |
1e8f9158 | 512 | ulist_free(fs_info->qgroup_ulist); |
eb1716af | 513 | fs_info->qgroup_ulist = NULL; |
b382a324 | 514 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; |
eb1716af | 515 | } |
1e8f9158 | 516 | |
bed92eae AJ |
517 | return ret < 0 ? ret : 0; |
518 | } | |
519 | ||
520 | /* | |
e685da14 WS |
521 | * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(), |
522 | * first two are in single-threaded paths.And for the third one, we have set | |
523 | * quota_root to be null with qgroup_lock held before, so it is safe to clean | |
524 | * up the in-memory structures without qgroup_lock held. | |
bed92eae AJ |
525 | */ |
526 | void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info) | |
527 | { | |
528 | struct rb_node *n; | |
529 | struct btrfs_qgroup *qgroup; | |
bed92eae AJ |
530 | |
531 | while ((n = rb_first(&fs_info->qgroup_tree))) { | |
532 | qgroup = rb_entry(n, struct btrfs_qgroup, node); | |
533 | rb_erase(n, &fs_info->qgroup_tree); | |
4082bd3d | 534 | __del_qgroup_rb(qgroup); |
bed92eae | 535 | } |
1e7bac1e WS |
536 | /* |
537 | * we call btrfs_free_qgroup_config() when umounting | |
01327610 | 538 | * filesystem and disabling quota, so we set qgroup_ulist |
1e7bac1e WS |
539 | * to be null here to avoid double free. |
540 | */ | |
1e8f9158 | 541 | ulist_free(fs_info->qgroup_ulist); |
1e7bac1e | 542 | fs_info->qgroup_ulist = NULL; |
bed92eae AJ |
543 | } |
544 | ||
545 | static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, | |
546 | struct btrfs_root *quota_root, | |
547 | u64 src, u64 dst) | |
548 | { | |
549 | int ret; | |
550 | struct btrfs_path *path; | |
551 | struct btrfs_key key; | |
552 | ||
553 | path = btrfs_alloc_path(); | |
554 | if (!path) | |
555 | return -ENOMEM; | |
556 | ||
557 | key.objectid = src; | |
558 | key.type = BTRFS_QGROUP_RELATION_KEY; | |
559 | key.offset = dst; | |
560 | ||
561 | ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0); | |
562 | ||
563 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
564 | ||
565 | btrfs_free_path(path); | |
566 | return ret; | |
567 | } | |
568 | ||
569 | static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, | |
570 | struct btrfs_root *quota_root, | |
571 | u64 src, u64 dst) | |
572 | { | |
573 | int ret; | |
574 | struct btrfs_path *path; | |
575 | struct btrfs_key key; | |
576 | ||
577 | path = btrfs_alloc_path(); | |
578 | if (!path) | |
579 | return -ENOMEM; | |
580 | ||
581 | key.objectid = src; | |
582 | key.type = BTRFS_QGROUP_RELATION_KEY; | |
583 | key.offset = dst; | |
584 | ||
585 | ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1); | |
586 | if (ret < 0) | |
587 | goto out; | |
588 | ||
589 | if (ret > 0) { | |
590 | ret = -ENOENT; | |
591 | goto out; | |
592 | } | |
593 | ||
594 | ret = btrfs_del_item(trans, quota_root, path); | |
595 | out: | |
596 | btrfs_free_path(path); | |
597 | return ret; | |
598 | } | |
599 | ||
600 | static int add_qgroup_item(struct btrfs_trans_handle *trans, | |
601 | struct btrfs_root *quota_root, u64 qgroupid) | |
602 | { | |
603 | int ret; | |
604 | struct btrfs_path *path; | |
605 | struct btrfs_qgroup_info_item *qgroup_info; | |
606 | struct btrfs_qgroup_limit_item *qgroup_limit; | |
607 | struct extent_buffer *leaf; | |
608 | struct btrfs_key key; | |
609 | ||
f5ee5c9a | 610 | if (btrfs_is_testing(quota_root->fs_info)) |
faa2dbf0 | 611 | return 0; |
fccb84c9 | 612 | |
bed92eae AJ |
613 | path = btrfs_alloc_path(); |
614 | if (!path) | |
615 | return -ENOMEM; | |
616 | ||
617 | key.objectid = 0; | |
618 | key.type = BTRFS_QGROUP_INFO_KEY; | |
619 | key.offset = qgroupid; | |
620 | ||
0b4699dc MF |
621 | /* |
622 | * Avoid a transaction abort by catching -EEXIST here. In that | |
623 | * case, we proceed by re-initializing the existing structure | |
624 | * on disk. | |
625 | */ | |
626 | ||
bed92eae AJ |
627 | ret = btrfs_insert_empty_item(trans, quota_root, path, &key, |
628 | sizeof(*qgroup_info)); | |
0b4699dc | 629 | if (ret && ret != -EEXIST) |
bed92eae AJ |
630 | goto out; |
631 | ||
632 | leaf = path->nodes[0]; | |
633 | qgroup_info = btrfs_item_ptr(leaf, path->slots[0], | |
634 | struct btrfs_qgroup_info_item); | |
635 | btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid); | |
636 | btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0); | |
637 | btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0); | |
638 | btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0); | |
639 | btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0); | |
640 | ||
641 | btrfs_mark_buffer_dirty(leaf); | |
642 | ||
643 | btrfs_release_path(path); | |
644 | ||
645 | key.type = BTRFS_QGROUP_LIMIT_KEY; | |
646 | ret = btrfs_insert_empty_item(trans, quota_root, path, &key, | |
647 | sizeof(*qgroup_limit)); | |
0b4699dc | 648 | if (ret && ret != -EEXIST) |
bed92eae AJ |
649 | goto out; |
650 | ||
651 | leaf = path->nodes[0]; | |
652 | qgroup_limit = btrfs_item_ptr(leaf, path->slots[0], | |
653 | struct btrfs_qgroup_limit_item); | |
654 | btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0); | |
655 | btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0); | |
656 | btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0); | |
657 | btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0); | |
658 | btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0); | |
659 | ||
660 | btrfs_mark_buffer_dirty(leaf); | |
661 | ||
662 | ret = 0; | |
663 | out: | |
664 | btrfs_free_path(path); | |
665 | return ret; | |
666 | } | |
667 | ||
668 | static int del_qgroup_item(struct btrfs_trans_handle *trans, | |
669 | struct btrfs_root *quota_root, u64 qgroupid) | |
670 | { | |
671 | int ret; | |
672 | struct btrfs_path *path; | |
673 | struct btrfs_key key; | |
674 | ||
675 | path = btrfs_alloc_path(); | |
676 | if (!path) | |
677 | return -ENOMEM; | |
678 | ||
679 | key.objectid = 0; | |
680 | key.type = BTRFS_QGROUP_INFO_KEY; | |
681 | key.offset = qgroupid; | |
682 | ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1); | |
683 | if (ret < 0) | |
684 | goto out; | |
685 | ||
686 | if (ret > 0) { | |
687 | ret = -ENOENT; | |
688 | goto out; | |
689 | } | |
690 | ||
691 | ret = btrfs_del_item(trans, quota_root, path); | |
692 | if (ret) | |
693 | goto out; | |
694 | ||
695 | btrfs_release_path(path); | |
696 | ||
697 | key.type = BTRFS_QGROUP_LIMIT_KEY; | |
698 | ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1); | |
699 | if (ret < 0) | |
700 | goto out; | |
701 | ||
702 | if (ret > 0) { | |
703 | ret = -ENOENT; | |
704 | goto out; | |
705 | } | |
706 | ||
707 | ret = btrfs_del_item(trans, quota_root, path); | |
708 | ||
709 | out: | |
710 | btrfs_free_path(path); | |
711 | return ret; | |
712 | } | |
713 | ||
714 | static int update_qgroup_limit_item(struct btrfs_trans_handle *trans, | |
1510e71c DY |
715 | struct btrfs_root *root, |
716 | struct btrfs_qgroup *qgroup) | |
bed92eae AJ |
717 | { |
718 | struct btrfs_path *path; | |
719 | struct btrfs_key key; | |
720 | struct extent_buffer *l; | |
721 | struct btrfs_qgroup_limit_item *qgroup_limit; | |
722 | int ret; | |
723 | int slot; | |
724 | ||
725 | key.objectid = 0; | |
726 | key.type = BTRFS_QGROUP_LIMIT_KEY; | |
1510e71c | 727 | key.offset = qgroup->qgroupid; |
bed92eae AJ |
728 | |
729 | path = btrfs_alloc_path(); | |
84cbe2f7 WS |
730 | if (!path) |
731 | return -ENOMEM; | |
732 | ||
bed92eae AJ |
733 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
734 | if (ret > 0) | |
735 | ret = -ENOENT; | |
736 | ||
737 | if (ret) | |
738 | goto out; | |
739 | ||
740 | l = path->nodes[0]; | |
741 | slot = path->slots[0]; | |
a3df41ee | 742 | qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item); |
1510e71c DY |
743 | btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags); |
744 | btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer); | |
745 | btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl); | |
746 | btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer); | |
747 | btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl); | |
bed92eae AJ |
748 | |
749 | btrfs_mark_buffer_dirty(l); | |
750 | ||
751 | out: | |
752 | btrfs_free_path(path); | |
753 | return ret; | |
754 | } | |
755 | ||
756 | static int update_qgroup_info_item(struct btrfs_trans_handle *trans, | |
757 | struct btrfs_root *root, | |
758 | struct btrfs_qgroup *qgroup) | |
759 | { | |
760 | struct btrfs_path *path; | |
761 | struct btrfs_key key; | |
762 | struct extent_buffer *l; | |
763 | struct btrfs_qgroup_info_item *qgroup_info; | |
764 | int ret; | |
765 | int slot; | |
766 | ||
f5ee5c9a | 767 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 768 | return 0; |
fccb84c9 | 769 | |
bed92eae AJ |
770 | key.objectid = 0; |
771 | key.type = BTRFS_QGROUP_INFO_KEY; | |
772 | key.offset = qgroup->qgroupid; | |
773 | ||
774 | path = btrfs_alloc_path(); | |
84cbe2f7 WS |
775 | if (!path) |
776 | return -ENOMEM; | |
777 | ||
bed92eae AJ |
778 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
779 | if (ret > 0) | |
780 | ret = -ENOENT; | |
781 | ||
782 | if (ret) | |
783 | goto out; | |
784 | ||
785 | l = path->nodes[0]; | |
786 | slot = path->slots[0]; | |
a3df41ee | 787 | qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item); |
bed92eae AJ |
788 | btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid); |
789 | btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer); | |
790 | btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr); | |
791 | btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl); | |
792 | btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr); | |
793 | ||
794 | btrfs_mark_buffer_dirty(l); | |
795 | ||
796 | out: | |
797 | btrfs_free_path(path); | |
798 | return ret; | |
799 | } | |
800 | ||
801 | static int update_qgroup_status_item(struct btrfs_trans_handle *trans, | |
802 | struct btrfs_fs_info *fs_info, | |
803 | struct btrfs_root *root) | |
804 | { | |
805 | struct btrfs_path *path; | |
806 | struct btrfs_key key; | |
807 | struct extent_buffer *l; | |
808 | struct btrfs_qgroup_status_item *ptr; | |
809 | int ret; | |
810 | int slot; | |
811 | ||
812 | key.objectid = 0; | |
813 | key.type = BTRFS_QGROUP_STATUS_KEY; | |
814 | key.offset = 0; | |
815 | ||
816 | path = btrfs_alloc_path(); | |
84cbe2f7 WS |
817 | if (!path) |
818 | return -ENOMEM; | |
819 | ||
bed92eae AJ |
820 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
821 | if (ret > 0) | |
822 | ret = -ENOENT; | |
823 | ||
824 | if (ret) | |
825 | goto out; | |
826 | ||
827 | l = path->nodes[0]; | |
828 | slot = path->slots[0]; | |
829 | ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item); | |
830 | btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags); | |
831 | btrfs_set_qgroup_status_generation(l, ptr, trans->transid); | |
2f232036 JS |
832 | btrfs_set_qgroup_status_rescan(l, ptr, |
833 | fs_info->qgroup_rescan_progress.objectid); | |
bed92eae AJ |
834 | |
835 | btrfs_mark_buffer_dirty(l); | |
836 | ||
837 | out: | |
838 | btrfs_free_path(path); | |
839 | return ret; | |
840 | } | |
841 | ||
842 | /* | |
843 | * called with qgroup_lock held | |
844 | */ | |
845 | static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans, | |
846 | struct btrfs_root *root) | |
847 | { | |
848 | struct btrfs_path *path; | |
849 | struct btrfs_key key; | |
06b3a860 | 850 | struct extent_buffer *leaf = NULL; |
bed92eae | 851 | int ret; |
06b3a860 | 852 | int nr = 0; |
bed92eae | 853 | |
bed92eae AJ |
854 | path = btrfs_alloc_path(); |
855 | if (!path) | |
856 | return -ENOMEM; | |
857 | ||
06b3a860 WS |
858 | path->leave_spinning = 1; |
859 | ||
860 | key.objectid = 0; | |
861 | key.offset = 0; | |
862 | key.type = 0; | |
bed92eae | 863 | |
06b3a860 | 864 | while (1) { |
bed92eae | 865 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
06b3a860 WS |
866 | if (ret < 0) |
867 | goto out; | |
868 | leaf = path->nodes[0]; | |
869 | nr = btrfs_header_nritems(leaf); | |
870 | if (!nr) | |
bed92eae | 871 | break; |
06b3a860 WS |
872 | /* |
873 | * delete the leaf one by one | |
874 | * since the whole tree is going | |
875 | * to be deleted. | |
876 | */ | |
877 | path->slots[0] = 0; | |
878 | ret = btrfs_del_items(trans, root, path, 0, nr); | |
bed92eae AJ |
879 | if (ret) |
880 | goto out; | |
06b3a860 | 881 | |
bed92eae AJ |
882 | btrfs_release_path(path); |
883 | } | |
884 | ret = 0; | |
885 | out: | |
bed92eae AJ |
886 | btrfs_free_path(path); |
887 | return ret; | |
888 | } | |
889 | ||
890 | int btrfs_quota_enable(struct btrfs_trans_handle *trans, | |
891 | struct btrfs_fs_info *fs_info) | |
892 | { | |
893 | struct btrfs_root *quota_root; | |
7708f029 | 894 | struct btrfs_root *tree_root = fs_info->tree_root; |
bed92eae AJ |
895 | struct btrfs_path *path = NULL; |
896 | struct btrfs_qgroup_status_item *ptr; | |
897 | struct extent_buffer *leaf; | |
898 | struct btrfs_key key; | |
7708f029 WS |
899 | struct btrfs_key found_key; |
900 | struct btrfs_qgroup *qgroup = NULL; | |
bed92eae | 901 | int ret = 0; |
7708f029 | 902 | int slot; |
bed92eae | 903 | |
f2f6ed3d | 904 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
5d23515b | 905 | if (fs_info->quota_root) |
bed92eae | 906 | goto out; |
bed92eae | 907 | |
52bf8e7a | 908 | fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL); |
1e8f9158 WS |
909 | if (!fs_info->qgroup_ulist) { |
910 | ret = -ENOMEM; | |
911 | goto out; | |
912 | } | |
913 | ||
bed92eae AJ |
914 | /* |
915 | * initially create the quota tree | |
916 | */ | |
917 | quota_root = btrfs_create_tree(trans, fs_info, | |
918 | BTRFS_QUOTA_TREE_OBJECTID); | |
919 | if (IS_ERR(quota_root)) { | |
920 | ret = PTR_ERR(quota_root); | |
921 | goto out; | |
922 | } | |
923 | ||
924 | path = btrfs_alloc_path(); | |
5b7ff5b3 TI |
925 | if (!path) { |
926 | ret = -ENOMEM; | |
927 | goto out_free_root; | |
928 | } | |
bed92eae AJ |
929 | |
930 | key.objectid = 0; | |
931 | key.type = BTRFS_QGROUP_STATUS_KEY; | |
932 | key.offset = 0; | |
933 | ||
934 | ret = btrfs_insert_empty_item(trans, quota_root, path, &key, | |
935 | sizeof(*ptr)); | |
936 | if (ret) | |
5b7ff5b3 | 937 | goto out_free_path; |
bed92eae AJ |
938 | |
939 | leaf = path->nodes[0]; | |
940 | ptr = btrfs_item_ptr(leaf, path->slots[0], | |
941 | struct btrfs_qgroup_status_item); | |
942 | btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid); | |
943 | btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION); | |
944 | fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON | | |
945 | BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
946 | btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags); | |
2f232036 | 947 | btrfs_set_qgroup_status_rescan(leaf, ptr, 0); |
bed92eae AJ |
948 | |
949 | btrfs_mark_buffer_dirty(leaf); | |
950 | ||
7708f029 WS |
951 | key.objectid = 0; |
952 | key.type = BTRFS_ROOT_REF_KEY; | |
953 | key.offset = 0; | |
954 | ||
955 | btrfs_release_path(path); | |
956 | ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0); | |
957 | if (ret > 0) | |
958 | goto out_add_root; | |
959 | if (ret < 0) | |
960 | goto out_free_path; | |
961 | ||
962 | ||
963 | while (1) { | |
964 | slot = path->slots[0]; | |
965 | leaf = path->nodes[0]; | |
966 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
967 | ||
968 | if (found_key.type == BTRFS_ROOT_REF_KEY) { | |
969 | ret = add_qgroup_item(trans, quota_root, | |
970 | found_key.offset); | |
971 | if (ret) | |
972 | goto out_free_path; | |
973 | ||
7708f029 WS |
974 | qgroup = add_qgroup_rb(fs_info, found_key.offset); |
975 | if (IS_ERR(qgroup)) { | |
7708f029 WS |
976 | ret = PTR_ERR(qgroup); |
977 | goto out_free_path; | |
978 | } | |
7708f029 WS |
979 | } |
980 | ret = btrfs_next_item(tree_root, path); | |
981 | if (ret < 0) | |
982 | goto out_free_path; | |
983 | if (ret) | |
984 | break; | |
985 | } | |
986 | ||
987 | out_add_root: | |
988 | btrfs_release_path(path); | |
989 | ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID); | |
990 | if (ret) | |
991 | goto out_free_path; | |
992 | ||
7708f029 WS |
993 | qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID); |
994 | if (IS_ERR(qgroup)) { | |
7708f029 WS |
995 | ret = PTR_ERR(qgroup); |
996 | goto out_free_path; | |
997 | } | |
58400fce | 998 | spin_lock(&fs_info->qgroup_lock); |
bed92eae | 999 | fs_info->quota_root = quota_root; |
5d23515b | 1000 | set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); |
bed92eae | 1001 | spin_unlock(&fs_info->qgroup_lock); |
5d23515b NB |
1002 | ret = qgroup_rescan_init(fs_info, 0, 1); |
1003 | if (!ret) { | |
1004 | qgroup_rescan_zero_tracking(fs_info); | |
1005 | btrfs_queue_work(fs_info->qgroup_rescan_workers, | |
1006 | &fs_info->qgroup_rescan_work); | |
1007 | } | |
1008 | ||
5b7ff5b3 | 1009 | out_free_path: |
bed92eae | 1010 | btrfs_free_path(path); |
5b7ff5b3 TI |
1011 | out_free_root: |
1012 | if (ret) { | |
1013 | free_extent_buffer(quota_root->node); | |
1014 | free_extent_buffer(quota_root->commit_root); | |
1015 | kfree(quota_root); | |
1016 | } | |
1017 | out: | |
eb1716af | 1018 | if (ret) { |
1e8f9158 | 1019 | ulist_free(fs_info->qgroup_ulist); |
eb1716af JS |
1020 | fs_info->qgroup_ulist = NULL; |
1021 | } | |
f2f6ed3d | 1022 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
1023 | return ret; |
1024 | } | |
1025 | ||
1026 | int btrfs_quota_disable(struct btrfs_trans_handle *trans, | |
1027 | struct btrfs_fs_info *fs_info) | |
1028 | { | |
bed92eae AJ |
1029 | struct btrfs_root *quota_root; |
1030 | int ret = 0; | |
1031 | ||
f2f6ed3d | 1032 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
58400fce | 1033 | if (!fs_info->quota_root) |
f2f6ed3d | 1034 | goto out; |
afcdd129 | 1035 | clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); |
d06f23d6 | 1036 | btrfs_qgroup_wait_for_completion(fs_info, false); |
967ef513 | 1037 | spin_lock(&fs_info->qgroup_lock); |
bed92eae AJ |
1038 | quota_root = fs_info->quota_root; |
1039 | fs_info->quota_root = NULL; | |
8ea0ec9e | 1040 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON; |
bed92eae AJ |
1041 | spin_unlock(&fs_info->qgroup_lock); |
1042 | ||
e685da14 WS |
1043 | btrfs_free_qgroup_config(fs_info); |
1044 | ||
bed92eae AJ |
1045 | ret = btrfs_clean_quota_tree(trans, quota_root); |
1046 | if (ret) | |
1047 | goto out; | |
1048 | ||
1cd5447e | 1049 | ret = btrfs_del_root(trans, fs_info, "a_root->root_key); |
bed92eae AJ |
1050 | if (ret) |
1051 | goto out; | |
1052 | ||
1053 | list_del("a_root->dirty_list); | |
1054 | ||
1055 | btrfs_tree_lock(quota_root->node); | |
7c302b49 | 1056 | clean_tree_block(fs_info, quota_root->node); |
bed92eae AJ |
1057 | btrfs_tree_unlock(quota_root->node); |
1058 | btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1); | |
1059 | ||
1060 | free_extent_buffer(quota_root->node); | |
1061 | free_extent_buffer(quota_root->commit_root); | |
1062 | kfree(quota_root); | |
1063 | out: | |
f2f6ed3d | 1064 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
1065 | return ret; |
1066 | } | |
1067 | ||
2f232036 JS |
1068 | static void qgroup_dirty(struct btrfs_fs_info *fs_info, |
1069 | struct btrfs_qgroup *qgroup) | |
bed92eae | 1070 | { |
2f232036 JS |
1071 | if (list_empty(&qgroup->dirty)) |
1072 | list_add(&qgroup->dirty, &fs_info->dirty_qgroups); | |
bed92eae AJ |
1073 | } |
1074 | ||
9c8b35b1 | 1075 | /* |
429d6275 QW |
1076 | * The easy accounting, we're updating qgroup relationship whose child qgroup |
1077 | * only has exclusive extents. | |
1078 | * | |
1079 | * In this case, all exclsuive extents will also be exlusive for parent, so | |
1080 | * excl/rfer just get added/removed. | |
1081 | * | |
1082 | * So is qgroup reservation space, which should also be added/removed to | |
1083 | * parent. | |
1084 | * Or when child tries to release reservation space, parent will underflow its | |
1085 | * reservation (for relationship adding case). | |
9c8b35b1 QW |
1086 | * |
1087 | * Caller should hold fs_info->qgroup_lock. | |
1088 | */ | |
1089 | static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info, | |
1090 | struct ulist *tmp, u64 ref_root, | |
429d6275 | 1091 | struct btrfs_qgroup *src, int sign) |
9c8b35b1 QW |
1092 | { |
1093 | struct btrfs_qgroup *qgroup; | |
1094 | struct btrfs_qgroup_list *glist; | |
1095 | struct ulist_node *unode; | |
1096 | struct ulist_iterator uiter; | |
429d6275 | 1097 | u64 num_bytes = src->excl; |
9c8b35b1 QW |
1098 | int ret = 0; |
1099 | ||
1100 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
1101 | if (!qgroup) | |
1102 | goto out; | |
1103 | ||
1104 | qgroup->rfer += sign * num_bytes; | |
1105 | qgroup->rfer_cmpr += sign * num_bytes; | |
1106 | ||
1107 | WARN_ON(sign < 0 && qgroup->excl < num_bytes); | |
1108 | qgroup->excl += sign * num_bytes; | |
1109 | qgroup->excl_cmpr += sign * num_bytes; | |
429d6275 QW |
1110 | |
1111 | if (sign > 0) | |
64ee4e75 | 1112 | qgroup_rsv_add_by_qgroup(fs_info, qgroup, src); |
429d6275 | 1113 | else |
64ee4e75 | 1114 | qgroup_rsv_release_by_qgroup(fs_info, qgroup, src); |
9c8b35b1 QW |
1115 | |
1116 | qgroup_dirty(fs_info, qgroup); | |
1117 | ||
1118 | /* Get all of the parent groups that contain this qgroup */ | |
1119 | list_for_each_entry(glist, &qgroup->groups, next_group) { | |
1120 | ret = ulist_add(tmp, glist->group->qgroupid, | |
ef2fff64 | 1121 | qgroup_to_aux(glist->group), GFP_ATOMIC); |
9c8b35b1 QW |
1122 | if (ret < 0) |
1123 | goto out; | |
1124 | } | |
1125 | ||
1126 | /* Iterate all of the parents and adjust their reference counts */ | |
1127 | ULIST_ITER_INIT(&uiter); | |
1128 | while ((unode = ulist_next(tmp, &uiter))) { | |
ef2fff64 | 1129 | qgroup = unode_aux_to_qgroup(unode); |
9c8b35b1 QW |
1130 | qgroup->rfer += sign * num_bytes; |
1131 | qgroup->rfer_cmpr += sign * num_bytes; | |
1132 | WARN_ON(sign < 0 && qgroup->excl < num_bytes); | |
1133 | qgroup->excl += sign * num_bytes; | |
429d6275 | 1134 | if (sign > 0) |
64ee4e75 | 1135 | qgroup_rsv_add_by_qgroup(fs_info, qgroup, src); |
429d6275 | 1136 | else |
64ee4e75 | 1137 | qgroup_rsv_release_by_qgroup(fs_info, qgroup, src); |
9c8b35b1 QW |
1138 | qgroup->excl_cmpr += sign * num_bytes; |
1139 | qgroup_dirty(fs_info, qgroup); | |
1140 | ||
1141 | /* Add any parents of the parents */ | |
1142 | list_for_each_entry(glist, &qgroup->groups, next_group) { | |
1143 | ret = ulist_add(tmp, glist->group->qgroupid, | |
ef2fff64 | 1144 | qgroup_to_aux(glist->group), GFP_ATOMIC); |
9c8b35b1 QW |
1145 | if (ret < 0) |
1146 | goto out; | |
1147 | } | |
1148 | } | |
1149 | ret = 0; | |
1150 | out: | |
1151 | return ret; | |
1152 | } | |
1153 | ||
1154 | ||
1155 | /* | |
1156 | * Quick path for updating qgroup with only excl refs. | |
1157 | * | |
1158 | * In that case, just update all parent will be enough. | |
1159 | * Or we needs to do a full rescan. | |
1160 | * Caller should also hold fs_info->qgroup_lock. | |
1161 | * | |
1162 | * Return 0 for quick update, return >0 for need to full rescan | |
1163 | * and mark INCONSISTENT flag. | |
1164 | * Return < 0 for other error. | |
1165 | */ | |
1166 | static int quick_update_accounting(struct btrfs_fs_info *fs_info, | |
1167 | struct ulist *tmp, u64 src, u64 dst, | |
1168 | int sign) | |
1169 | { | |
1170 | struct btrfs_qgroup *qgroup; | |
1171 | int ret = 1; | |
1172 | int err = 0; | |
1173 | ||
1174 | qgroup = find_qgroup_rb(fs_info, src); | |
1175 | if (!qgroup) | |
1176 | goto out; | |
1177 | if (qgroup->excl == qgroup->rfer) { | |
1178 | ret = 0; | |
1179 | err = __qgroup_excl_accounting(fs_info, tmp, dst, | |
429d6275 | 1180 | qgroup, sign); |
9c8b35b1 QW |
1181 | if (err < 0) { |
1182 | ret = err; | |
1183 | goto out; | |
1184 | } | |
1185 | } | |
1186 | out: | |
1187 | if (ret) | |
1188 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1189 | return ret; | |
1190 | } | |
1191 | ||
bed92eae AJ |
1192 | int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, |
1193 | struct btrfs_fs_info *fs_info, u64 src, u64 dst) | |
1194 | { | |
1195 | struct btrfs_root *quota_root; | |
b7fef4f5 WS |
1196 | struct btrfs_qgroup *parent; |
1197 | struct btrfs_qgroup *member; | |
534e6623 | 1198 | struct btrfs_qgroup_list *list; |
9c8b35b1 | 1199 | struct ulist *tmp; |
bed92eae AJ |
1200 | int ret = 0; |
1201 | ||
8465ecec QW |
1202 | /* Check the level of src and dst first */ |
1203 | if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst)) | |
1204 | return -EINVAL; | |
1205 | ||
6602caf1 | 1206 | tmp = ulist_alloc(GFP_KERNEL); |
ab3680dd CE |
1207 | if (!tmp) |
1208 | return -ENOMEM; | |
1209 | ||
f2f6ed3d | 1210 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1211 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1212 | if (!quota_root) { |
1213 | ret = -EINVAL; | |
1214 | goto out; | |
1215 | } | |
b7fef4f5 WS |
1216 | member = find_qgroup_rb(fs_info, src); |
1217 | parent = find_qgroup_rb(fs_info, dst); | |
1218 | if (!member || !parent) { | |
1219 | ret = -EINVAL; | |
1220 | goto out; | |
1221 | } | |
bed92eae | 1222 | |
534e6623 WS |
1223 | /* check if such qgroup relation exist firstly */ |
1224 | list_for_each_entry(list, &member->groups, next_group) { | |
1225 | if (list->group == parent) { | |
1226 | ret = -EEXIST; | |
1227 | goto out; | |
1228 | } | |
1229 | } | |
1230 | ||
bed92eae AJ |
1231 | ret = add_qgroup_relation_item(trans, quota_root, src, dst); |
1232 | if (ret) | |
f2f6ed3d | 1233 | goto out; |
bed92eae AJ |
1234 | |
1235 | ret = add_qgroup_relation_item(trans, quota_root, dst, src); | |
1236 | if (ret) { | |
1237 | del_qgroup_relation_item(trans, quota_root, src, dst); | |
f2f6ed3d | 1238 | goto out; |
bed92eae AJ |
1239 | } |
1240 | ||
1241 | spin_lock(&fs_info->qgroup_lock); | |
0b246afa | 1242 | ret = add_relation_rb(fs_info, src, dst); |
9c8b35b1 QW |
1243 | if (ret < 0) { |
1244 | spin_unlock(&fs_info->qgroup_lock); | |
1245 | goto out; | |
1246 | } | |
1247 | ret = quick_update_accounting(fs_info, tmp, src, dst, 1); | |
bed92eae | 1248 | spin_unlock(&fs_info->qgroup_lock); |
f2f6ed3d WS |
1249 | out: |
1250 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
9c8b35b1 | 1251 | ulist_free(tmp); |
bed92eae AJ |
1252 | return ret; |
1253 | } | |
1254 | ||
025db916 | 1255 | static int __del_qgroup_relation(struct btrfs_trans_handle *trans, |
bed92eae AJ |
1256 | struct btrfs_fs_info *fs_info, u64 src, u64 dst) |
1257 | { | |
1258 | struct btrfs_root *quota_root; | |
534e6623 WS |
1259 | struct btrfs_qgroup *parent; |
1260 | struct btrfs_qgroup *member; | |
1261 | struct btrfs_qgroup_list *list; | |
9c8b35b1 | 1262 | struct ulist *tmp; |
bed92eae AJ |
1263 | int ret = 0; |
1264 | int err; | |
1265 | ||
6602caf1 | 1266 | tmp = ulist_alloc(GFP_KERNEL); |
9c8b35b1 QW |
1267 | if (!tmp) |
1268 | return -ENOMEM; | |
1269 | ||
bed92eae | 1270 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1271 | if (!quota_root) { |
1272 | ret = -EINVAL; | |
1273 | goto out; | |
1274 | } | |
bed92eae | 1275 | |
534e6623 WS |
1276 | member = find_qgroup_rb(fs_info, src); |
1277 | parent = find_qgroup_rb(fs_info, dst); | |
1278 | if (!member || !parent) { | |
1279 | ret = -EINVAL; | |
1280 | goto out; | |
1281 | } | |
1282 | ||
1283 | /* check if such qgroup relation exist firstly */ | |
1284 | list_for_each_entry(list, &member->groups, next_group) { | |
1285 | if (list->group == parent) | |
1286 | goto exist; | |
1287 | } | |
1288 | ret = -ENOENT; | |
1289 | goto out; | |
1290 | exist: | |
bed92eae AJ |
1291 | ret = del_qgroup_relation_item(trans, quota_root, src, dst); |
1292 | err = del_qgroup_relation_item(trans, quota_root, dst, src); | |
1293 | if (err && !ret) | |
1294 | ret = err; | |
1295 | ||
1296 | spin_lock(&fs_info->qgroup_lock); | |
1297 | del_relation_rb(fs_info, src, dst); | |
9c8b35b1 | 1298 | ret = quick_update_accounting(fs_info, tmp, src, dst, -1); |
bed92eae | 1299 | spin_unlock(&fs_info->qgroup_lock); |
f2f6ed3d | 1300 | out: |
9c8b35b1 | 1301 | ulist_free(tmp); |
f5a6b1c5 DY |
1302 | return ret; |
1303 | } | |
1304 | ||
1305 | int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, | |
1306 | struct btrfs_fs_info *fs_info, u64 src, u64 dst) | |
1307 | { | |
1308 | int ret = 0; | |
1309 | ||
1310 | mutex_lock(&fs_info->qgroup_ioctl_lock); | |
1311 | ret = __del_qgroup_relation(trans, fs_info, src, dst); | |
f2f6ed3d | 1312 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
f5a6b1c5 | 1313 | |
bed92eae AJ |
1314 | return ret; |
1315 | } | |
1316 | ||
1317 | int btrfs_create_qgroup(struct btrfs_trans_handle *trans, | |
4087cf24 | 1318 | struct btrfs_fs_info *fs_info, u64 qgroupid) |
bed92eae AJ |
1319 | { |
1320 | struct btrfs_root *quota_root; | |
1321 | struct btrfs_qgroup *qgroup; | |
1322 | int ret = 0; | |
1323 | ||
f2f6ed3d | 1324 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1325 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1326 | if (!quota_root) { |
1327 | ret = -EINVAL; | |
1328 | goto out; | |
1329 | } | |
534e6623 WS |
1330 | qgroup = find_qgroup_rb(fs_info, qgroupid); |
1331 | if (qgroup) { | |
1332 | ret = -EEXIST; | |
1333 | goto out; | |
1334 | } | |
bed92eae AJ |
1335 | |
1336 | ret = add_qgroup_item(trans, quota_root, qgroupid); | |
534e6623 WS |
1337 | if (ret) |
1338 | goto out; | |
bed92eae AJ |
1339 | |
1340 | spin_lock(&fs_info->qgroup_lock); | |
1341 | qgroup = add_qgroup_rb(fs_info, qgroupid); | |
1342 | spin_unlock(&fs_info->qgroup_lock); | |
1343 | ||
1344 | if (IS_ERR(qgroup)) | |
1345 | ret = PTR_ERR(qgroup); | |
f2f6ed3d WS |
1346 | out: |
1347 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
bed92eae AJ |
1348 | return ret; |
1349 | } | |
1350 | ||
1351 | int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, | |
1352 | struct btrfs_fs_info *fs_info, u64 qgroupid) | |
1353 | { | |
1354 | struct btrfs_root *quota_root; | |
2cf68703 | 1355 | struct btrfs_qgroup *qgroup; |
f5a6b1c5 | 1356 | struct btrfs_qgroup_list *list; |
bed92eae AJ |
1357 | int ret = 0; |
1358 | ||
f2f6ed3d | 1359 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
bed92eae | 1360 | quota_root = fs_info->quota_root; |
f2f6ed3d WS |
1361 | if (!quota_root) { |
1362 | ret = -EINVAL; | |
1363 | goto out; | |
1364 | } | |
bed92eae | 1365 | |
2cf68703 | 1366 | qgroup = find_qgroup_rb(fs_info, qgroupid); |
534e6623 WS |
1367 | if (!qgroup) { |
1368 | ret = -ENOENT; | |
1369 | goto out; | |
1370 | } else { | |
f5a6b1c5 DY |
1371 | /* check if there are no children of this qgroup */ |
1372 | if (!list_empty(&qgroup->members)) { | |
f2f6ed3d WS |
1373 | ret = -EBUSY; |
1374 | goto out; | |
2cf68703 AJ |
1375 | } |
1376 | } | |
bed92eae | 1377 | ret = del_qgroup_item(trans, quota_root, qgroupid); |
36b96fdc SD |
1378 | if (ret && ret != -ENOENT) |
1379 | goto out; | |
bed92eae | 1380 | |
f5a6b1c5 DY |
1381 | while (!list_empty(&qgroup->groups)) { |
1382 | list = list_first_entry(&qgroup->groups, | |
1383 | struct btrfs_qgroup_list, next_group); | |
1384 | ret = __del_qgroup_relation(trans, fs_info, | |
1385 | qgroupid, | |
1386 | list->group->qgroupid); | |
1387 | if (ret) | |
1388 | goto out; | |
1389 | } | |
1390 | ||
bed92eae | 1391 | spin_lock(&fs_info->qgroup_lock); |
0b246afa | 1392 | del_qgroup_rb(fs_info, qgroupid); |
bed92eae | 1393 | spin_unlock(&fs_info->qgroup_lock); |
f2f6ed3d WS |
1394 | out: |
1395 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
bed92eae AJ |
1396 | return ret; |
1397 | } | |
1398 | ||
1399 | int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, | |
1400 | struct btrfs_fs_info *fs_info, u64 qgroupid, | |
1401 | struct btrfs_qgroup_limit *limit) | |
1402 | { | |
f2f6ed3d | 1403 | struct btrfs_root *quota_root; |
bed92eae AJ |
1404 | struct btrfs_qgroup *qgroup; |
1405 | int ret = 0; | |
fe759907 YD |
1406 | /* Sometimes we would want to clear the limit on this qgroup. |
1407 | * To meet this requirement, we treat the -1 as a special value | |
1408 | * which tell kernel to clear the limit on this qgroup. | |
1409 | */ | |
1410 | const u64 CLEAR_VALUE = -1; | |
bed92eae | 1411 | |
f2f6ed3d WS |
1412 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
1413 | quota_root = fs_info->quota_root; | |
1414 | if (!quota_root) { | |
1415 | ret = -EINVAL; | |
1416 | goto out; | |
1417 | } | |
bed92eae | 1418 | |
ddb47afa WS |
1419 | qgroup = find_qgroup_rb(fs_info, qgroupid); |
1420 | if (!qgroup) { | |
1421 | ret = -ENOENT; | |
1422 | goto out; | |
1423 | } | |
bed92eae | 1424 | |
58400fce | 1425 | spin_lock(&fs_info->qgroup_lock); |
fe759907 YD |
1426 | if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) { |
1427 | if (limit->max_rfer == CLEAR_VALUE) { | |
1428 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER; | |
1429 | limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER; | |
1430 | qgroup->max_rfer = 0; | |
1431 | } else { | |
1432 | qgroup->max_rfer = limit->max_rfer; | |
1433 | } | |
1434 | } | |
1435 | if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) { | |
1436 | if (limit->max_excl == CLEAR_VALUE) { | |
1437 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL; | |
1438 | limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL; | |
1439 | qgroup->max_excl = 0; | |
1440 | } else { | |
1441 | qgroup->max_excl = limit->max_excl; | |
1442 | } | |
1443 | } | |
1444 | if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) { | |
1445 | if (limit->rsv_rfer == CLEAR_VALUE) { | |
1446 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER; | |
1447 | limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER; | |
1448 | qgroup->rsv_rfer = 0; | |
1449 | } else { | |
1450 | qgroup->rsv_rfer = limit->rsv_rfer; | |
1451 | } | |
1452 | } | |
1453 | if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) { | |
1454 | if (limit->rsv_excl == CLEAR_VALUE) { | |
1455 | qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL; | |
1456 | limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL; | |
1457 | qgroup->rsv_excl = 0; | |
1458 | } else { | |
1459 | qgroup->rsv_excl = limit->rsv_excl; | |
1460 | } | |
1461 | } | |
03477d94 DY |
1462 | qgroup->lim_flags |= limit->flags; |
1463 | ||
bed92eae | 1464 | spin_unlock(&fs_info->qgroup_lock); |
1510e71c DY |
1465 | |
1466 | ret = update_qgroup_limit_item(trans, quota_root, qgroup); | |
1467 | if (ret) { | |
1468 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1469 | btrfs_info(fs_info, "unable to update quota limit for %llu", | |
1470 | qgroupid); | |
1471 | } | |
1472 | ||
f2f6ed3d WS |
1473 | out: |
1474 | mutex_unlock(&fs_info->qgroup_ioctl_lock); | |
bed92eae AJ |
1475 | return ret; |
1476 | } | |
1152651a | 1477 | |
50b3e040 | 1478 | int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info, |
cb93b52c QW |
1479 | struct btrfs_delayed_ref_root *delayed_refs, |
1480 | struct btrfs_qgroup_extent_record *record) | |
3368d001 QW |
1481 | { |
1482 | struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node; | |
1483 | struct rb_node *parent_node = NULL; | |
1484 | struct btrfs_qgroup_extent_record *entry; | |
1485 | u64 bytenr = record->bytenr; | |
1486 | ||
a4666e68 | 1487 | lockdep_assert_held(&delayed_refs->lock); |
50b3e040 | 1488 | trace_btrfs_qgroup_trace_extent(fs_info, record); |
82bd101b | 1489 | |
3368d001 QW |
1490 | while (*p) { |
1491 | parent_node = *p; | |
1492 | entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record, | |
1493 | node); | |
1494 | if (bytenr < entry->bytenr) | |
1495 | p = &(*p)->rb_left; | |
1496 | else if (bytenr > entry->bytenr) | |
1497 | p = &(*p)->rb_right; | |
1498 | else | |
cb93b52c | 1499 | return 1; |
3368d001 QW |
1500 | } |
1501 | ||
1502 | rb_link_node(&record->node, parent_node, p); | |
1503 | rb_insert_color(&record->node, &delayed_refs->dirty_extent_root); | |
cb93b52c QW |
1504 | return 0; |
1505 | } | |
1506 | ||
fb235dc0 QW |
1507 | int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info, |
1508 | struct btrfs_qgroup_extent_record *qrecord) | |
1509 | { | |
1510 | struct ulist *old_root; | |
1511 | u64 bytenr = qrecord->bytenr; | |
1512 | int ret; | |
1513 | ||
c995ab3c | 1514 | ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false); |
952bd3db NB |
1515 | if (ret < 0) { |
1516 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
1517 | btrfs_warn(fs_info, | |
1518 | "error accounting new delayed refs extent (err code: %d), quota inconsistent", | |
1519 | ret); | |
1520 | return 0; | |
1521 | } | |
fb235dc0 QW |
1522 | |
1523 | /* | |
1524 | * Here we don't need to get the lock of | |
1525 | * trans->transaction->delayed_refs, since inserted qrecord won't | |
1526 | * be deleted, only qrecord->node may be modified (new qrecord insert) | |
1527 | * | |
1528 | * So modifying qrecord->old_roots is safe here | |
1529 | */ | |
1530 | qrecord->old_roots = old_root; | |
1531 | return 0; | |
1532 | } | |
1533 | ||
50b3e040 | 1534 | int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, |
cb93b52c QW |
1535 | struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes, |
1536 | gfp_t gfp_flag) | |
1537 | { | |
1538 | struct btrfs_qgroup_extent_record *record; | |
1539 | struct btrfs_delayed_ref_root *delayed_refs; | |
1540 | int ret; | |
1541 | ||
afcdd129 JB |
1542 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) |
1543 | || bytenr == 0 || num_bytes == 0) | |
cb93b52c QW |
1544 | return 0; |
1545 | if (WARN_ON(trans == NULL)) | |
1546 | return -EINVAL; | |
1547 | record = kmalloc(sizeof(*record), gfp_flag); | |
1548 | if (!record) | |
1549 | return -ENOMEM; | |
1550 | ||
1551 | delayed_refs = &trans->transaction->delayed_refs; | |
1552 | record->bytenr = bytenr; | |
1553 | record->num_bytes = num_bytes; | |
1554 | record->old_roots = NULL; | |
1555 | ||
1556 | spin_lock(&delayed_refs->lock); | |
2ff7e61e | 1557 | ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record); |
cb93b52c | 1558 | spin_unlock(&delayed_refs->lock); |
fb235dc0 | 1559 | if (ret > 0) { |
cb93b52c | 1560 | kfree(record); |
fb235dc0 QW |
1561 | return 0; |
1562 | } | |
1563 | return btrfs_qgroup_trace_extent_post(fs_info, record); | |
3368d001 QW |
1564 | } |
1565 | ||
33d1f05c | 1566 | int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans, |
2ff7e61e | 1567 | struct btrfs_fs_info *fs_info, |
33d1f05c QW |
1568 | struct extent_buffer *eb) |
1569 | { | |
1570 | int nr = btrfs_header_nritems(eb); | |
1571 | int i, extent_type, ret; | |
1572 | struct btrfs_key key; | |
1573 | struct btrfs_file_extent_item *fi; | |
1574 | u64 bytenr, num_bytes; | |
1575 | ||
1576 | /* We can be called directly from walk_up_proc() */ | |
0b246afa | 1577 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
33d1f05c QW |
1578 | return 0; |
1579 | ||
1580 | for (i = 0; i < nr; i++) { | |
1581 | btrfs_item_key_to_cpu(eb, &key, i); | |
1582 | ||
1583 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
1584 | continue; | |
1585 | ||
1586 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
1587 | /* filter out non qgroup-accountable extents */ | |
1588 | extent_type = btrfs_file_extent_type(eb, fi); | |
1589 | ||
1590 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
1591 | continue; | |
1592 | ||
1593 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
1594 | if (!bytenr) | |
1595 | continue; | |
1596 | ||
1597 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
1598 | ||
0b246afa JM |
1599 | ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr, |
1600 | num_bytes, GFP_NOFS); | |
33d1f05c QW |
1601 | if (ret) |
1602 | return ret; | |
1603 | } | |
cddf3b2c | 1604 | cond_resched(); |
33d1f05c QW |
1605 | return 0; |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * Walk up the tree from the bottom, freeing leaves and any interior | |
1610 | * nodes which have had all slots visited. If a node (leaf or | |
1611 | * interior) is freed, the node above it will have it's slot | |
1612 | * incremented. The root node will never be freed. | |
1613 | * | |
1614 | * At the end of this function, we should have a path which has all | |
1615 | * slots incremented to the next position for a search. If we need to | |
1616 | * read a new node it will be NULL and the node above it will have the | |
1617 | * correct slot selected for a later read. | |
1618 | * | |
1619 | * If we increment the root nodes slot counter past the number of | |
1620 | * elements, 1 is returned to signal completion of the search. | |
1621 | */ | |
15b34517 | 1622 | static int adjust_slots_upwards(struct btrfs_path *path, int root_level) |
33d1f05c QW |
1623 | { |
1624 | int level = 0; | |
1625 | int nr, slot; | |
1626 | struct extent_buffer *eb; | |
1627 | ||
1628 | if (root_level == 0) | |
1629 | return 1; | |
1630 | ||
1631 | while (level <= root_level) { | |
1632 | eb = path->nodes[level]; | |
1633 | nr = btrfs_header_nritems(eb); | |
1634 | path->slots[level]++; | |
1635 | slot = path->slots[level]; | |
1636 | if (slot >= nr || level == 0) { | |
1637 | /* | |
1638 | * Don't free the root - we will detect this | |
1639 | * condition after our loop and return a | |
1640 | * positive value for caller to stop walking the tree. | |
1641 | */ | |
1642 | if (level != root_level) { | |
1643 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
1644 | path->locks[level] = 0; | |
1645 | ||
1646 | free_extent_buffer(eb); | |
1647 | path->nodes[level] = NULL; | |
1648 | path->slots[level] = 0; | |
1649 | } | |
1650 | } else { | |
1651 | /* | |
1652 | * We have a valid slot to walk back down | |
1653 | * from. Stop here so caller can process these | |
1654 | * new nodes. | |
1655 | */ | |
1656 | break; | |
1657 | } | |
1658 | ||
1659 | level++; | |
1660 | } | |
1661 | ||
1662 | eb = path->nodes[root_level]; | |
1663 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
1664 | return 1; | |
1665 | ||
1666 | return 0; | |
1667 | } | |
1668 | ||
1669 | int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans, | |
1670 | struct btrfs_root *root, | |
1671 | struct extent_buffer *root_eb, | |
1672 | u64 root_gen, int root_level) | |
1673 | { | |
0b246afa | 1674 | struct btrfs_fs_info *fs_info = root->fs_info; |
33d1f05c QW |
1675 | int ret = 0; |
1676 | int level; | |
1677 | struct extent_buffer *eb = root_eb; | |
1678 | struct btrfs_path *path = NULL; | |
1679 | ||
b6e6bca5 | 1680 | BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL); |
33d1f05c QW |
1681 | BUG_ON(root_eb == NULL); |
1682 | ||
0b246afa | 1683 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
33d1f05c QW |
1684 | return 0; |
1685 | ||
1686 | if (!extent_buffer_uptodate(root_eb)) { | |
581c1760 | 1687 | ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL); |
33d1f05c QW |
1688 | if (ret) |
1689 | goto out; | |
1690 | } | |
1691 | ||
1692 | if (root_level == 0) { | |
2ff7e61e | 1693 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb); |
33d1f05c QW |
1694 | goto out; |
1695 | } | |
1696 | ||
1697 | path = btrfs_alloc_path(); | |
1698 | if (!path) | |
1699 | return -ENOMEM; | |
1700 | ||
1701 | /* | |
1702 | * Walk down the tree. Missing extent blocks are filled in as | |
1703 | * we go. Metadata is accounted every time we read a new | |
1704 | * extent block. | |
1705 | * | |
1706 | * When we reach a leaf, we account for file extent items in it, | |
1707 | * walk back up the tree (adjusting slot pointers as we go) | |
1708 | * and restart the search process. | |
1709 | */ | |
1710 | extent_buffer_get(root_eb); /* For path */ | |
1711 | path->nodes[root_level] = root_eb; | |
1712 | path->slots[root_level] = 0; | |
1713 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
1714 | walk_down: | |
1715 | level = root_level; | |
1716 | while (level >= 0) { | |
1717 | if (path->nodes[level] == NULL) { | |
581c1760 | 1718 | struct btrfs_key first_key; |
33d1f05c QW |
1719 | int parent_slot; |
1720 | u64 child_gen; | |
1721 | u64 child_bytenr; | |
1722 | ||
1723 | /* | |
1724 | * We need to get child blockptr/gen from parent before | |
1725 | * we can read it. | |
1726 | */ | |
1727 | eb = path->nodes[level + 1]; | |
1728 | parent_slot = path->slots[level + 1]; | |
1729 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
1730 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
581c1760 | 1731 | btrfs_node_key_to_cpu(eb, &first_key, parent_slot); |
33d1f05c | 1732 | |
581c1760 QW |
1733 | eb = read_tree_block(fs_info, child_bytenr, child_gen, |
1734 | level, &first_key); | |
33d1f05c QW |
1735 | if (IS_ERR(eb)) { |
1736 | ret = PTR_ERR(eb); | |
1737 | goto out; | |
1738 | } else if (!extent_buffer_uptodate(eb)) { | |
1739 | free_extent_buffer(eb); | |
1740 | ret = -EIO; | |
1741 | goto out; | |
1742 | } | |
1743 | ||
1744 | path->nodes[level] = eb; | |
1745 | path->slots[level] = 0; | |
1746 | ||
1747 | btrfs_tree_read_lock(eb); | |
1748 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1749 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
1750 | ||
0b246afa JM |
1751 | ret = btrfs_qgroup_trace_extent(trans, fs_info, |
1752 | child_bytenr, | |
1753 | fs_info->nodesize, | |
1754 | GFP_NOFS); | |
33d1f05c QW |
1755 | if (ret) |
1756 | goto out; | |
1757 | } | |
1758 | ||
1759 | if (level == 0) { | |
2ff7e61e JM |
1760 | ret = btrfs_qgroup_trace_leaf_items(trans,fs_info, |
1761 | path->nodes[level]); | |
33d1f05c QW |
1762 | if (ret) |
1763 | goto out; | |
1764 | ||
1765 | /* Nonzero return here means we completed our search */ | |
15b34517 | 1766 | ret = adjust_slots_upwards(path, root_level); |
33d1f05c QW |
1767 | if (ret) |
1768 | break; | |
1769 | ||
1770 | /* Restart search with new slots */ | |
1771 | goto walk_down; | |
1772 | } | |
1773 | ||
1774 | level--; | |
1775 | } | |
1776 | ||
1777 | ret = 0; | |
1778 | out: | |
1779 | btrfs_free_path(path); | |
1780 | ||
1781 | return ret; | |
1782 | } | |
1783 | ||
d810ef2b QW |
1784 | #define UPDATE_NEW 0 |
1785 | #define UPDATE_OLD 1 | |
1786 | /* | |
1787 | * Walk all of the roots that points to the bytenr and adjust their refcnts. | |
1788 | */ | |
1789 | static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info, | |
1790 | struct ulist *roots, struct ulist *tmp, | |
1791 | struct ulist *qgroups, u64 seq, int update_old) | |
1792 | { | |
1793 | struct ulist_node *unode; | |
1794 | struct ulist_iterator uiter; | |
1795 | struct ulist_node *tmp_unode; | |
1796 | struct ulist_iterator tmp_uiter; | |
1797 | struct btrfs_qgroup *qg; | |
1798 | int ret = 0; | |
1799 | ||
1800 | if (!roots) | |
1801 | return 0; | |
1802 | ULIST_ITER_INIT(&uiter); | |
1803 | while ((unode = ulist_next(roots, &uiter))) { | |
1804 | qg = find_qgroup_rb(fs_info, unode->val); | |
1805 | if (!qg) | |
1806 | continue; | |
1807 | ||
1808 | ulist_reinit(tmp); | |
ef2fff64 | 1809 | ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg), |
d810ef2b QW |
1810 | GFP_ATOMIC); |
1811 | if (ret < 0) | |
1812 | return ret; | |
ef2fff64 | 1813 | ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC); |
d810ef2b QW |
1814 | if (ret < 0) |
1815 | return ret; | |
1816 | ULIST_ITER_INIT(&tmp_uiter); | |
1817 | while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) { | |
1818 | struct btrfs_qgroup_list *glist; | |
1819 | ||
ef2fff64 | 1820 | qg = unode_aux_to_qgroup(tmp_unode); |
d810ef2b QW |
1821 | if (update_old) |
1822 | btrfs_qgroup_update_old_refcnt(qg, seq, 1); | |
1823 | else | |
1824 | btrfs_qgroup_update_new_refcnt(qg, seq, 1); | |
1825 | list_for_each_entry(glist, &qg->groups, next_group) { | |
1826 | ret = ulist_add(qgroups, glist->group->qgroupid, | |
ef2fff64 | 1827 | qgroup_to_aux(glist->group), |
d810ef2b QW |
1828 | GFP_ATOMIC); |
1829 | if (ret < 0) | |
1830 | return ret; | |
1831 | ret = ulist_add(tmp, glist->group->qgroupid, | |
ef2fff64 | 1832 | qgroup_to_aux(glist->group), |
d810ef2b QW |
1833 | GFP_ATOMIC); |
1834 | if (ret < 0) | |
1835 | return ret; | |
1836 | } | |
1837 | } | |
1838 | } | |
1839 | return 0; | |
1840 | } | |
1841 | ||
823ae5b8 QW |
1842 | /* |
1843 | * Update qgroup rfer/excl counters. | |
1844 | * Rfer update is easy, codes can explain themselves. | |
e69bcee3 | 1845 | * |
823ae5b8 QW |
1846 | * Excl update is tricky, the update is split into 2 part. |
1847 | * Part 1: Possible exclusive <-> sharing detect: | |
1848 | * | A | !A | | |
1849 | * ------------------------------------- | |
1850 | * B | * | - | | |
1851 | * ------------------------------------- | |
1852 | * !B | + | ** | | |
1853 | * ------------------------------------- | |
1854 | * | |
1855 | * Conditions: | |
1856 | * A: cur_old_roots < nr_old_roots (not exclusive before) | |
1857 | * !A: cur_old_roots == nr_old_roots (possible exclusive before) | |
1858 | * B: cur_new_roots < nr_new_roots (not exclusive now) | |
01327610 | 1859 | * !B: cur_new_roots == nr_new_roots (possible exclusive now) |
823ae5b8 QW |
1860 | * |
1861 | * Results: | |
1862 | * +: Possible sharing -> exclusive -: Possible exclusive -> sharing | |
1863 | * *: Definitely not changed. **: Possible unchanged. | |
1864 | * | |
1865 | * For !A and !B condition, the exception is cur_old/new_roots == 0 case. | |
1866 | * | |
1867 | * To make the logic clear, we first use condition A and B to split | |
1868 | * combination into 4 results. | |
1869 | * | |
1870 | * Then, for result "+" and "-", check old/new_roots == 0 case, as in them | |
1871 | * only on variant maybe 0. | |
1872 | * | |
1873 | * Lastly, check result **, since there are 2 variants maybe 0, split them | |
1874 | * again(2x2). | |
1875 | * But this time we don't need to consider other things, the codes and logic | |
1876 | * is easy to understand now. | |
1877 | */ | |
1878 | static int qgroup_update_counters(struct btrfs_fs_info *fs_info, | |
1879 | struct ulist *qgroups, | |
1880 | u64 nr_old_roots, | |
1881 | u64 nr_new_roots, | |
1882 | u64 num_bytes, u64 seq) | |
1883 | { | |
1884 | struct ulist_node *unode; | |
1885 | struct ulist_iterator uiter; | |
1886 | struct btrfs_qgroup *qg; | |
1887 | u64 cur_new_count, cur_old_count; | |
1888 | ||
1889 | ULIST_ITER_INIT(&uiter); | |
1890 | while ((unode = ulist_next(qgroups, &uiter))) { | |
1891 | bool dirty = false; | |
1892 | ||
ef2fff64 | 1893 | qg = unode_aux_to_qgroup(unode); |
823ae5b8 QW |
1894 | cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq); |
1895 | cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq); | |
1896 | ||
bc074524 JM |
1897 | trace_qgroup_update_counters(fs_info, qg->qgroupid, |
1898 | cur_old_count, cur_new_count); | |
0f5dcf8d | 1899 | |
823ae5b8 QW |
1900 | /* Rfer update part */ |
1901 | if (cur_old_count == 0 && cur_new_count > 0) { | |
1902 | qg->rfer += num_bytes; | |
1903 | qg->rfer_cmpr += num_bytes; | |
1904 | dirty = true; | |
1905 | } | |
1906 | if (cur_old_count > 0 && cur_new_count == 0) { | |
1907 | qg->rfer -= num_bytes; | |
1908 | qg->rfer_cmpr -= num_bytes; | |
1909 | dirty = true; | |
1910 | } | |
1911 | ||
1912 | /* Excl update part */ | |
1913 | /* Exclusive/none -> shared case */ | |
1914 | if (cur_old_count == nr_old_roots && | |
1915 | cur_new_count < nr_new_roots) { | |
1916 | /* Exclusive -> shared */ | |
1917 | if (cur_old_count != 0) { | |
1918 | qg->excl -= num_bytes; | |
1919 | qg->excl_cmpr -= num_bytes; | |
1920 | dirty = true; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | /* Shared -> exclusive/none case */ | |
1925 | if (cur_old_count < nr_old_roots && | |
1926 | cur_new_count == nr_new_roots) { | |
1927 | /* Shared->exclusive */ | |
1928 | if (cur_new_count != 0) { | |
1929 | qg->excl += num_bytes; | |
1930 | qg->excl_cmpr += num_bytes; | |
1931 | dirty = true; | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | /* Exclusive/none -> exclusive/none case */ | |
1936 | if (cur_old_count == nr_old_roots && | |
1937 | cur_new_count == nr_new_roots) { | |
1938 | if (cur_old_count == 0) { | |
1939 | /* None -> exclusive/none */ | |
1940 | ||
1941 | if (cur_new_count != 0) { | |
1942 | /* None -> exclusive */ | |
1943 | qg->excl += num_bytes; | |
1944 | qg->excl_cmpr += num_bytes; | |
1945 | dirty = true; | |
1946 | } | |
1947 | /* None -> none, nothing changed */ | |
1948 | } else { | |
1949 | /* Exclusive -> exclusive/none */ | |
1950 | ||
1951 | if (cur_new_count == 0) { | |
1952 | /* Exclusive -> none */ | |
1953 | qg->excl -= num_bytes; | |
1954 | qg->excl_cmpr -= num_bytes; | |
1955 | dirty = true; | |
1956 | } | |
1957 | /* Exclusive -> exclusive, nothing changed */ | |
1958 | } | |
1959 | } | |
c05f9429 | 1960 | |
823ae5b8 QW |
1961 | if (dirty) |
1962 | qgroup_dirty(fs_info, qg); | |
1963 | } | |
1964 | return 0; | |
1965 | } | |
1966 | ||
5edfd9fd QW |
1967 | /* |
1968 | * Check if the @roots potentially is a list of fs tree roots | |
1969 | * | |
1970 | * Return 0 for definitely not a fs/subvol tree roots ulist | |
1971 | * Return 1 for possible fs/subvol tree roots in the list (considering an empty | |
1972 | * one as well) | |
1973 | */ | |
1974 | static int maybe_fs_roots(struct ulist *roots) | |
1975 | { | |
1976 | struct ulist_node *unode; | |
1977 | struct ulist_iterator uiter; | |
1978 | ||
1979 | /* Empty one, still possible for fs roots */ | |
1980 | if (!roots || roots->nnodes == 0) | |
1981 | return 1; | |
1982 | ||
1983 | ULIST_ITER_INIT(&uiter); | |
1984 | unode = ulist_next(roots, &uiter); | |
1985 | if (!unode) | |
1986 | return 1; | |
1987 | ||
1988 | /* | |
1989 | * If it contains fs tree roots, then it must belong to fs/subvol | |
1990 | * trees. | |
1991 | * If it contains a non-fs tree, it won't be shared with fs/subvol trees. | |
1992 | */ | |
1993 | return is_fstree(unode->val); | |
1994 | } | |
1995 | ||
442244c9 | 1996 | int |
550d7a2e QW |
1997 | btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, |
1998 | struct btrfs_fs_info *fs_info, | |
1999 | u64 bytenr, u64 num_bytes, | |
2000 | struct ulist *old_roots, struct ulist *new_roots) | |
2001 | { | |
2002 | struct ulist *qgroups = NULL; | |
2003 | struct ulist *tmp = NULL; | |
2004 | u64 seq; | |
2005 | u64 nr_new_roots = 0; | |
2006 | u64 nr_old_roots = 0; | |
2007 | int ret = 0; | |
2008 | ||
81353d50 DS |
2009 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
2010 | return 0; | |
2011 | ||
5edfd9fd QW |
2012 | if (new_roots) { |
2013 | if (!maybe_fs_roots(new_roots)) | |
2014 | goto out_free; | |
550d7a2e | 2015 | nr_new_roots = new_roots->nnodes; |
5edfd9fd QW |
2016 | } |
2017 | if (old_roots) { | |
2018 | if (!maybe_fs_roots(old_roots)) | |
2019 | goto out_free; | |
550d7a2e | 2020 | nr_old_roots = old_roots->nnodes; |
5edfd9fd QW |
2021 | } |
2022 | ||
2023 | /* Quick exit, either not fs tree roots, or won't affect any qgroup */ | |
2024 | if (nr_old_roots == 0 && nr_new_roots == 0) | |
2025 | goto out_free; | |
550d7a2e | 2026 | |
550d7a2e QW |
2027 | BUG_ON(!fs_info->quota_root); |
2028 | ||
bc074524 JM |
2029 | trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes, |
2030 | nr_old_roots, nr_new_roots); | |
0f5dcf8d | 2031 | |
550d7a2e QW |
2032 | qgroups = ulist_alloc(GFP_NOFS); |
2033 | if (!qgroups) { | |
2034 | ret = -ENOMEM; | |
2035 | goto out_free; | |
2036 | } | |
2037 | tmp = ulist_alloc(GFP_NOFS); | |
2038 | if (!tmp) { | |
2039 | ret = -ENOMEM; | |
2040 | goto out_free; | |
2041 | } | |
2042 | ||
2043 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
2044 | if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { | |
2045 | if (fs_info->qgroup_rescan_progress.objectid <= bytenr) { | |
2046 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2047 | ret = 0; | |
2048 | goto out_free; | |
2049 | } | |
2050 | } | |
2051 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2052 | ||
2053 | spin_lock(&fs_info->qgroup_lock); | |
2054 | seq = fs_info->qgroup_seq; | |
2055 | ||
2056 | /* Update old refcnts using old_roots */ | |
2057 | ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq, | |
2058 | UPDATE_OLD); | |
2059 | if (ret < 0) | |
2060 | goto out; | |
2061 | ||
2062 | /* Update new refcnts using new_roots */ | |
2063 | ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq, | |
2064 | UPDATE_NEW); | |
2065 | if (ret < 0) | |
2066 | goto out; | |
2067 | ||
2068 | qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots, | |
2069 | num_bytes, seq); | |
2070 | ||
2071 | /* | |
2072 | * Bump qgroup_seq to avoid seq overlap | |
2073 | */ | |
2074 | fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1; | |
2075 | out: | |
2076 | spin_unlock(&fs_info->qgroup_lock); | |
2077 | out_free: | |
2078 | ulist_free(tmp); | |
2079 | ulist_free(qgroups); | |
2080 | ulist_free(old_roots); | |
2081 | ulist_free(new_roots); | |
2082 | return ret; | |
2083 | } | |
2084 | ||
460fb20a | 2085 | int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans) |
550d7a2e | 2086 | { |
460fb20a | 2087 | struct btrfs_fs_info *fs_info = trans->fs_info; |
550d7a2e QW |
2088 | struct btrfs_qgroup_extent_record *record; |
2089 | struct btrfs_delayed_ref_root *delayed_refs; | |
2090 | struct ulist *new_roots = NULL; | |
2091 | struct rb_node *node; | |
9086db86 | 2092 | u64 qgroup_to_skip; |
550d7a2e QW |
2093 | int ret = 0; |
2094 | ||
2095 | delayed_refs = &trans->transaction->delayed_refs; | |
9086db86 | 2096 | qgroup_to_skip = delayed_refs->qgroup_to_skip; |
550d7a2e QW |
2097 | while ((node = rb_first(&delayed_refs->dirty_extent_root))) { |
2098 | record = rb_entry(node, struct btrfs_qgroup_extent_record, | |
2099 | node); | |
2100 | ||
bc074524 | 2101 | trace_btrfs_qgroup_account_extents(fs_info, record); |
0f5dcf8d | 2102 | |
550d7a2e | 2103 | if (!ret) { |
d1b8b94a QW |
2104 | /* |
2105 | * Old roots should be searched when inserting qgroup | |
2106 | * extent record | |
2107 | */ | |
2108 | if (WARN_ON(!record->old_roots)) { | |
2109 | /* Search commit root to find old_roots */ | |
2110 | ret = btrfs_find_all_roots(NULL, fs_info, | |
2111 | record->bytenr, 0, | |
c995ab3c | 2112 | &record->old_roots, false); |
d1b8b94a QW |
2113 | if (ret < 0) |
2114 | goto cleanup; | |
2115 | } | |
2116 | ||
550d7a2e | 2117 | /* |
de47c9d3 | 2118 | * Use SEQ_LAST as time_seq to do special search, which |
550d7a2e QW |
2119 | * doesn't lock tree or delayed_refs and search current |
2120 | * root. It's safe inside commit_transaction(). | |
2121 | */ | |
2122 | ret = btrfs_find_all_roots(trans, fs_info, | |
c995ab3c | 2123 | record->bytenr, SEQ_LAST, &new_roots, false); |
550d7a2e QW |
2124 | if (ret < 0) |
2125 | goto cleanup; | |
d1b8b94a | 2126 | if (qgroup_to_skip) { |
9086db86 | 2127 | ulist_del(new_roots, qgroup_to_skip, 0); |
d1b8b94a QW |
2128 | ulist_del(record->old_roots, qgroup_to_skip, |
2129 | 0); | |
2130 | } | |
550d7a2e QW |
2131 | ret = btrfs_qgroup_account_extent(trans, fs_info, |
2132 | record->bytenr, record->num_bytes, | |
2133 | record->old_roots, new_roots); | |
2134 | record->old_roots = NULL; | |
2135 | new_roots = NULL; | |
2136 | } | |
2137 | cleanup: | |
2138 | ulist_free(record->old_roots); | |
2139 | ulist_free(new_roots); | |
2140 | new_roots = NULL; | |
2141 | rb_erase(node, &delayed_refs->dirty_extent_root); | |
2142 | kfree(record); | |
2143 | ||
2144 | } | |
2145 | return ret; | |
2146 | } | |
2147 | ||
bed92eae AJ |
2148 | /* |
2149 | * called from commit_transaction. Writes all changed qgroups to disk. | |
2150 | */ | |
2151 | int btrfs_run_qgroups(struct btrfs_trans_handle *trans, | |
2152 | struct btrfs_fs_info *fs_info) | |
2153 | { | |
2154 | struct btrfs_root *quota_root = fs_info->quota_root; | |
2155 | int ret = 0; | |
2156 | ||
2157 | if (!quota_root) | |
5d23515b | 2158 | return ret; |
bed92eae AJ |
2159 | |
2160 | spin_lock(&fs_info->qgroup_lock); | |
2161 | while (!list_empty(&fs_info->dirty_qgroups)) { | |
2162 | struct btrfs_qgroup *qgroup; | |
2163 | qgroup = list_first_entry(&fs_info->dirty_qgroups, | |
2164 | struct btrfs_qgroup, dirty); | |
2165 | list_del_init(&qgroup->dirty); | |
2166 | spin_unlock(&fs_info->qgroup_lock); | |
2167 | ret = update_qgroup_info_item(trans, quota_root, qgroup); | |
d3001ed3 DY |
2168 | if (ret) |
2169 | fs_info->qgroup_flags |= | |
2170 | BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2171 | ret = update_qgroup_limit_item(trans, quota_root, qgroup); | |
bed92eae AJ |
2172 | if (ret) |
2173 | fs_info->qgroup_flags |= | |
2174 | BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2175 | spin_lock(&fs_info->qgroup_lock); | |
2176 | } | |
afcdd129 | 2177 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
bed92eae AJ |
2178 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON; |
2179 | else | |
2180 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON; | |
2181 | spin_unlock(&fs_info->qgroup_lock); | |
2182 | ||
2183 | ret = update_qgroup_status_item(trans, fs_info, quota_root); | |
2184 | if (ret) | |
2185 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2186 | ||
bed92eae AJ |
2187 | return ret; |
2188 | } | |
2189 | ||
2190 | /* | |
01327610 | 2191 | * Copy the accounting information between qgroups. This is necessary |
918c2ee1 MF |
2192 | * when a snapshot or a subvolume is created. Throwing an error will |
2193 | * cause a transaction abort so we take extra care here to only error | |
2194 | * when a readonly fs is a reasonable outcome. | |
bed92eae AJ |
2195 | */ |
2196 | int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, | |
2197 | struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid, | |
2198 | struct btrfs_qgroup_inherit *inherit) | |
2199 | { | |
2200 | int ret = 0; | |
2201 | int i; | |
2202 | u64 *i_qgroups; | |
2203 | struct btrfs_root *quota_root = fs_info->quota_root; | |
2204 | struct btrfs_qgroup *srcgroup; | |
2205 | struct btrfs_qgroup *dstgroup; | |
2206 | u32 level_size = 0; | |
3f5e2d3b | 2207 | u64 nums; |
bed92eae | 2208 | |
f2f6ed3d | 2209 | mutex_lock(&fs_info->qgroup_ioctl_lock); |
afcdd129 | 2210 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
f2f6ed3d | 2211 | goto out; |
bed92eae | 2212 | |
f2f6ed3d WS |
2213 | if (!quota_root) { |
2214 | ret = -EINVAL; | |
2215 | goto out; | |
2216 | } | |
bed92eae | 2217 | |
3f5e2d3b WS |
2218 | if (inherit) { |
2219 | i_qgroups = (u64 *)(inherit + 1); | |
2220 | nums = inherit->num_qgroups + 2 * inherit->num_ref_copies + | |
2221 | 2 * inherit->num_excl_copies; | |
2222 | for (i = 0; i < nums; ++i) { | |
2223 | srcgroup = find_qgroup_rb(fs_info, *i_qgroups); | |
09870d27 | 2224 | |
918c2ee1 MF |
2225 | /* |
2226 | * Zero out invalid groups so we can ignore | |
2227 | * them later. | |
2228 | */ | |
2229 | if (!srcgroup || | |
2230 | ((srcgroup->qgroupid >> 48) <= (objectid >> 48))) | |
2231 | *i_qgroups = 0ULL; | |
2232 | ||
3f5e2d3b WS |
2233 | ++i_qgroups; |
2234 | } | |
2235 | } | |
2236 | ||
bed92eae AJ |
2237 | /* |
2238 | * create a tracking group for the subvol itself | |
2239 | */ | |
2240 | ret = add_qgroup_item(trans, quota_root, objectid); | |
2241 | if (ret) | |
2242 | goto out; | |
2243 | ||
bed92eae AJ |
2244 | if (srcid) { |
2245 | struct btrfs_root *srcroot; | |
2246 | struct btrfs_key srckey; | |
bed92eae AJ |
2247 | |
2248 | srckey.objectid = srcid; | |
2249 | srckey.type = BTRFS_ROOT_ITEM_KEY; | |
2250 | srckey.offset = (u64)-1; | |
2251 | srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey); | |
2252 | if (IS_ERR(srcroot)) { | |
2253 | ret = PTR_ERR(srcroot); | |
2254 | goto out; | |
2255 | } | |
2256 | ||
0b246afa | 2257 | level_size = fs_info->nodesize; |
bed92eae AJ |
2258 | } |
2259 | ||
2260 | /* | |
2261 | * add qgroup to all inherited groups | |
2262 | */ | |
2263 | if (inherit) { | |
2264 | i_qgroups = (u64 *)(inherit + 1); | |
918c2ee1 MF |
2265 | for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) { |
2266 | if (*i_qgroups == 0) | |
2267 | continue; | |
bed92eae AJ |
2268 | ret = add_qgroup_relation_item(trans, quota_root, |
2269 | objectid, *i_qgroups); | |
918c2ee1 | 2270 | if (ret && ret != -EEXIST) |
bed92eae AJ |
2271 | goto out; |
2272 | ret = add_qgroup_relation_item(trans, quota_root, | |
2273 | *i_qgroups, objectid); | |
918c2ee1 | 2274 | if (ret && ret != -EEXIST) |
bed92eae | 2275 | goto out; |
bed92eae | 2276 | } |
918c2ee1 | 2277 | ret = 0; |
bed92eae AJ |
2278 | } |
2279 | ||
2280 | ||
2281 | spin_lock(&fs_info->qgroup_lock); | |
2282 | ||
2283 | dstgroup = add_qgroup_rb(fs_info, objectid); | |
57a5a882 DC |
2284 | if (IS_ERR(dstgroup)) { |
2285 | ret = PTR_ERR(dstgroup); | |
bed92eae | 2286 | goto unlock; |
57a5a882 | 2287 | } |
bed92eae | 2288 | |
e8c8541a | 2289 | if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) { |
e8c8541a DY |
2290 | dstgroup->lim_flags = inherit->lim.flags; |
2291 | dstgroup->max_rfer = inherit->lim.max_rfer; | |
2292 | dstgroup->max_excl = inherit->lim.max_excl; | |
2293 | dstgroup->rsv_rfer = inherit->lim.rsv_rfer; | |
2294 | dstgroup->rsv_excl = inherit->lim.rsv_excl; | |
1510e71c DY |
2295 | |
2296 | ret = update_qgroup_limit_item(trans, quota_root, dstgroup); | |
2297 | if (ret) { | |
2298 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
5d163e0e JM |
2299 | btrfs_info(fs_info, |
2300 | "unable to update quota limit for %llu", | |
2301 | dstgroup->qgroupid); | |
1510e71c DY |
2302 | goto unlock; |
2303 | } | |
e8c8541a DY |
2304 | } |
2305 | ||
bed92eae AJ |
2306 | if (srcid) { |
2307 | srcgroup = find_qgroup_rb(fs_info, srcid); | |
f3a87f1b | 2308 | if (!srcgroup) |
bed92eae | 2309 | goto unlock; |
fcebe456 JB |
2310 | |
2311 | /* | |
2312 | * We call inherit after we clone the root in order to make sure | |
2313 | * our counts don't go crazy, so at this point the only | |
2314 | * difference between the two roots should be the root node. | |
2315 | */ | |
2316 | dstgroup->rfer = srcgroup->rfer; | |
2317 | dstgroup->rfer_cmpr = srcgroup->rfer_cmpr; | |
2318 | dstgroup->excl = level_size; | |
2319 | dstgroup->excl_cmpr = level_size; | |
bed92eae AJ |
2320 | srcgroup->excl = level_size; |
2321 | srcgroup->excl_cmpr = level_size; | |
3eeb4d59 DY |
2322 | |
2323 | /* inherit the limit info */ | |
2324 | dstgroup->lim_flags = srcgroup->lim_flags; | |
2325 | dstgroup->max_rfer = srcgroup->max_rfer; | |
2326 | dstgroup->max_excl = srcgroup->max_excl; | |
2327 | dstgroup->rsv_rfer = srcgroup->rsv_rfer; | |
2328 | dstgroup->rsv_excl = srcgroup->rsv_excl; | |
2329 | ||
bed92eae AJ |
2330 | qgroup_dirty(fs_info, dstgroup); |
2331 | qgroup_dirty(fs_info, srcgroup); | |
2332 | } | |
2333 | ||
f3a87f1b | 2334 | if (!inherit) |
bed92eae AJ |
2335 | goto unlock; |
2336 | ||
2337 | i_qgroups = (u64 *)(inherit + 1); | |
2338 | for (i = 0; i < inherit->num_qgroups; ++i) { | |
918c2ee1 | 2339 | if (*i_qgroups) { |
0b246afa | 2340 | ret = add_relation_rb(fs_info, objectid, *i_qgroups); |
918c2ee1 MF |
2341 | if (ret) |
2342 | goto unlock; | |
2343 | } | |
bed92eae AJ |
2344 | ++i_qgroups; |
2345 | } | |
2346 | ||
918c2ee1 | 2347 | for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) { |
bed92eae AJ |
2348 | struct btrfs_qgroup *src; |
2349 | struct btrfs_qgroup *dst; | |
2350 | ||
918c2ee1 MF |
2351 | if (!i_qgroups[0] || !i_qgroups[1]) |
2352 | continue; | |
2353 | ||
bed92eae AJ |
2354 | src = find_qgroup_rb(fs_info, i_qgroups[0]); |
2355 | dst = find_qgroup_rb(fs_info, i_qgroups[1]); | |
2356 | ||
2357 | if (!src || !dst) { | |
2358 | ret = -EINVAL; | |
2359 | goto unlock; | |
2360 | } | |
2361 | ||
2362 | dst->rfer = src->rfer - level_size; | |
2363 | dst->rfer_cmpr = src->rfer_cmpr - level_size; | |
bed92eae | 2364 | } |
918c2ee1 | 2365 | for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) { |
bed92eae AJ |
2366 | struct btrfs_qgroup *src; |
2367 | struct btrfs_qgroup *dst; | |
2368 | ||
918c2ee1 MF |
2369 | if (!i_qgroups[0] || !i_qgroups[1]) |
2370 | continue; | |
2371 | ||
bed92eae AJ |
2372 | src = find_qgroup_rb(fs_info, i_qgroups[0]); |
2373 | dst = find_qgroup_rb(fs_info, i_qgroups[1]); | |
2374 | ||
2375 | if (!src || !dst) { | |
2376 | ret = -EINVAL; | |
2377 | goto unlock; | |
2378 | } | |
2379 | ||
2380 | dst->excl = src->excl + level_size; | |
2381 | dst->excl_cmpr = src->excl_cmpr + level_size; | |
bed92eae AJ |
2382 | } |
2383 | ||
2384 | unlock: | |
2385 | spin_unlock(&fs_info->qgroup_lock); | |
2386 | out: | |
f2f6ed3d | 2387 | mutex_unlock(&fs_info->qgroup_ioctl_lock); |
bed92eae AJ |
2388 | return ret; |
2389 | } | |
2390 | ||
003d7c59 JM |
2391 | static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes) |
2392 | { | |
2393 | if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) && | |
dba21324 | 2394 | qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer) |
003d7c59 JM |
2395 | return false; |
2396 | ||
2397 | if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) && | |
dba21324 | 2398 | qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl) |
003d7c59 JM |
2399 | return false; |
2400 | ||
2401 | return true; | |
2402 | } | |
2403 | ||
dba21324 QW |
2404 | static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce, |
2405 | enum btrfs_qgroup_rsv_type type) | |
bed92eae AJ |
2406 | { |
2407 | struct btrfs_root *quota_root; | |
2408 | struct btrfs_qgroup *qgroup; | |
2409 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2410 | u64 ref_root = root->root_key.objectid; | |
2411 | int ret = 0; | |
bed92eae AJ |
2412 | struct ulist_node *unode; |
2413 | struct ulist_iterator uiter; | |
2414 | ||
2415 | if (!is_fstree(ref_root)) | |
2416 | return 0; | |
2417 | ||
2418 | if (num_bytes == 0) | |
2419 | return 0; | |
f29efe29 SD |
2420 | |
2421 | if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) && | |
2422 | capable(CAP_SYS_RESOURCE)) | |
2423 | enforce = false; | |
2424 | ||
bed92eae AJ |
2425 | spin_lock(&fs_info->qgroup_lock); |
2426 | quota_root = fs_info->quota_root; | |
2427 | if (!quota_root) | |
2428 | goto out; | |
2429 | ||
2430 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
2431 | if (!qgroup) | |
2432 | goto out; | |
2433 | ||
2434 | /* | |
2435 | * in a first step, we check all affected qgroups if any limits would | |
2436 | * be exceeded | |
2437 | */ | |
1e8f9158 WS |
2438 | ulist_reinit(fs_info->qgroup_ulist); |
2439 | ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid, | |
a1840b50 | 2440 | qgroup_to_aux(qgroup), GFP_ATOMIC); |
3c97185c WS |
2441 | if (ret < 0) |
2442 | goto out; | |
bed92eae | 2443 | ULIST_ITER_INIT(&uiter); |
1e8f9158 | 2444 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { |
bed92eae AJ |
2445 | struct btrfs_qgroup *qg; |
2446 | struct btrfs_qgroup_list *glist; | |
2447 | ||
ef2fff64 | 2448 | qg = unode_aux_to_qgroup(unode); |
bed92eae | 2449 | |
003d7c59 | 2450 | if (enforce && !qgroup_check_limits(qg, num_bytes)) { |
bed92eae | 2451 | ret = -EDQUOT; |
720f1e20 WS |
2452 | goto out; |
2453 | } | |
bed92eae AJ |
2454 | |
2455 | list_for_each_entry(glist, &qg->groups, next_group) { | |
1e8f9158 WS |
2456 | ret = ulist_add(fs_info->qgroup_ulist, |
2457 | glist->group->qgroupid, | |
a1840b50 | 2458 | qgroup_to_aux(glist->group), GFP_ATOMIC); |
3c97185c WS |
2459 | if (ret < 0) |
2460 | goto out; | |
bed92eae AJ |
2461 | } |
2462 | } | |
3c97185c | 2463 | ret = 0; |
bed92eae AJ |
2464 | /* |
2465 | * no limits exceeded, now record the reservation into all qgroups | |
2466 | */ | |
2467 | ULIST_ITER_INIT(&uiter); | |
1e8f9158 | 2468 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { |
bed92eae AJ |
2469 | struct btrfs_qgroup *qg; |
2470 | ||
ef2fff64 | 2471 | qg = unode_aux_to_qgroup(unode); |
bed92eae | 2472 | |
64ee4e75 QW |
2473 | trace_qgroup_update_reserve(fs_info, qg, num_bytes, type); |
2474 | qgroup_rsv_add(fs_info, qg, num_bytes, type); | |
bed92eae AJ |
2475 | } |
2476 | ||
2477 | out: | |
2478 | spin_unlock(&fs_info->qgroup_lock); | |
bed92eae AJ |
2479 | return ret; |
2480 | } | |
2481 | ||
e1211d0e QW |
2482 | /* |
2483 | * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0 | |
2484 | * qgroup). | |
2485 | * | |
2486 | * Will handle all higher level qgroup too. | |
2487 | * | |
2488 | * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup. | |
2489 | * This special case is only used for META_PERTRANS type. | |
2490 | */ | |
297d750b | 2491 | void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info, |
d4e5c920 QW |
2492 | u64 ref_root, u64 num_bytes, |
2493 | enum btrfs_qgroup_rsv_type type) | |
bed92eae AJ |
2494 | { |
2495 | struct btrfs_root *quota_root; | |
2496 | struct btrfs_qgroup *qgroup; | |
bed92eae AJ |
2497 | struct ulist_node *unode; |
2498 | struct ulist_iterator uiter; | |
3c97185c | 2499 | int ret = 0; |
bed92eae AJ |
2500 | |
2501 | if (!is_fstree(ref_root)) | |
2502 | return; | |
2503 | ||
2504 | if (num_bytes == 0) | |
2505 | return; | |
2506 | ||
e1211d0e QW |
2507 | if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) { |
2508 | WARN(1, "%s: Invalid type to free", __func__); | |
2509 | return; | |
2510 | } | |
bed92eae AJ |
2511 | spin_lock(&fs_info->qgroup_lock); |
2512 | ||
2513 | quota_root = fs_info->quota_root; | |
2514 | if (!quota_root) | |
2515 | goto out; | |
2516 | ||
2517 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
2518 | if (!qgroup) | |
2519 | goto out; | |
2520 | ||
e1211d0e | 2521 | if (num_bytes == (u64)-1) |
8287475a QW |
2522 | /* |
2523 | * We're freeing all pertrans rsv, get reserved value from | |
2524 | * level 0 qgroup as real num_bytes to free. | |
2525 | */ | |
e1211d0e QW |
2526 | num_bytes = qgroup->rsv.values[type]; |
2527 | ||
1e8f9158 WS |
2528 | ulist_reinit(fs_info->qgroup_ulist); |
2529 | ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid, | |
a1840b50 | 2530 | qgroup_to_aux(qgroup), GFP_ATOMIC); |
3c97185c WS |
2531 | if (ret < 0) |
2532 | goto out; | |
bed92eae | 2533 | ULIST_ITER_INIT(&uiter); |
1e8f9158 | 2534 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { |
bed92eae AJ |
2535 | struct btrfs_qgroup *qg; |
2536 | struct btrfs_qgroup_list *glist; | |
2537 | ||
ef2fff64 | 2538 | qg = unode_aux_to_qgroup(unode); |
bed92eae | 2539 | |
64ee4e75 QW |
2540 | trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type); |
2541 | qgroup_rsv_release(fs_info, qg, num_bytes, type); | |
bed92eae AJ |
2542 | |
2543 | list_for_each_entry(glist, &qg->groups, next_group) { | |
1e8f9158 WS |
2544 | ret = ulist_add(fs_info->qgroup_ulist, |
2545 | glist->group->qgroupid, | |
a1840b50 | 2546 | qgroup_to_aux(glist->group), GFP_ATOMIC); |
3c97185c WS |
2547 | if (ret < 0) |
2548 | goto out; | |
bed92eae AJ |
2549 | } |
2550 | } | |
2551 | ||
2552 | out: | |
2553 | spin_unlock(&fs_info->qgroup_lock); | |
bed92eae AJ |
2554 | } |
2555 | ||
2f232036 JS |
2556 | /* |
2557 | * returns < 0 on error, 0 when more leafs are to be scanned. | |
3393168d | 2558 | * returns 1 when done. |
2f232036 JS |
2559 | */ |
2560 | static int | |
b382a324 | 2561 | qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
0a0e8b89 | 2562 | struct btrfs_trans_handle *trans) |
2f232036 JS |
2563 | { |
2564 | struct btrfs_key found; | |
0a0e8b89 | 2565 | struct extent_buffer *scratch_leaf = NULL; |
2f232036 | 2566 | struct ulist *roots = NULL; |
3284da7b | 2567 | struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem); |
fcebe456 | 2568 | u64 num_bytes; |
2f232036 JS |
2569 | int slot; |
2570 | int ret; | |
2571 | ||
2f232036 JS |
2572 | mutex_lock(&fs_info->qgroup_rescan_lock); |
2573 | ret = btrfs_search_slot_for_read(fs_info->extent_root, | |
2574 | &fs_info->qgroup_rescan_progress, | |
2575 | path, 1, 0); | |
2576 | ||
ab8d0fc4 JM |
2577 | btrfs_debug(fs_info, |
2578 | "current progress key (%llu %u %llu), search_slot ret %d", | |
2579 | fs_info->qgroup_rescan_progress.objectid, | |
2580 | fs_info->qgroup_rescan_progress.type, | |
2581 | fs_info->qgroup_rescan_progress.offset, ret); | |
2f232036 JS |
2582 | |
2583 | if (ret) { | |
2584 | /* | |
2585 | * The rescan is about to end, we will not be scanning any | |
2586 | * further blocks. We cannot unset the RESCAN flag here, because | |
2587 | * we want to commit the transaction if everything went well. | |
2588 | * To make the live accounting work in this phase, we set our | |
2589 | * scan progress pointer such that every real extent objectid | |
2590 | * will be smaller. | |
2591 | */ | |
2592 | fs_info->qgroup_rescan_progress.objectid = (u64)-1; | |
2593 | btrfs_release_path(path); | |
2594 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2595 | return ret; | |
2596 | } | |
2597 | ||
2598 | btrfs_item_key_to_cpu(path->nodes[0], &found, | |
2599 | btrfs_header_nritems(path->nodes[0]) - 1); | |
2600 | fs_info->qgroup_rescan_progress.objectid = found.objectid + 1; | |
2601 | ||
2602 | btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); | |
0a0e8b89 QW |
2603 | scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]); |
2604 | if (!scratch_leaf) { | |
2605 | ret = -ENOMEM; | |
2606 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2607 | goto out; | |
2608 | } | |
2609 | extent_buffer_get(scratch_leaf); | |
2610 | btrfs_tree_read_lock(scratch_leaf); | |
2611 | btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK); | |
2f232036 JS |
2612 | slot = path->slots[0]; |
2613 | btrfs_release_path(path); | |
2614 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2615 | ||
2616 | for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) { | |
2617 | btrfs_item_key_to_cpu(scratch_leaf, &found, slot); | |
3a6d75e8 JB |
2618 | if (found.type != BTRFS_EXTENT_ITEM_KEY && |
2619 | found.type != BTRFS_METADATA_ITEM_KEY) | |
2f232036 | 2620 | continue; |
3a6d75e8 | 2621 | if (found.type == BTRFS_METADATA_ITEM_KEY) |
da17066c | 2622 | num_bytes = fs_info->nodesize; |
3a6d75e8 JB |
2623 | else |
2624 | num_bytes = found.offset; | |
2625 | ||
fcebe456 | 2626 | ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0, |
c995ab3c | 2627 | &roots, false); |
2f232036 JS |
2628 | if (ret < 0) |
2629 | goto out; | |
9d220c95 QW |
2630 | /* For rescan, just pass old_roots as NULL */ |
2631 | ret = btrfs_qgroup_account_extent(trans, fs_info, | |
2632 | found.objectid, num_bytes, NULL, roots); | |
2633 | if (ret < 0) | |
fcebe456 | 2634 | goto out; |
2f232036 | 2635 | } |
2f232036 | 2636 | out: |
0a0e8b89 QW |
2637 | if (scratch_leaf) { |
2638 | btrfs_tree_read_unlock_blocking(scratch_leaf); | |
2639 | free_extent_buffer(scratch_leaf); | |
2640 | } | |
2f232036 JS |
2641 | btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); |
2642 | ||
2643 | return ret; | |
2644 | } | |
2645 | ||
d458b054 | 2646 | static void btrfs_qgroup_rescan_worker(struct btrfs_work *work) |
2f232036 | 2647 | { |
b382a324 JS |
2648 | struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info, |
2649 | qgroup_rescan_work); | |
2f232036 JS |
2650 | struct btrfs_path *path; |
2651 | struct btrfs_trans_handle *trans = NULL; | |
2f232036 | 2652 | int err = -ENOMEM; |
53b7cde9 | 2653 | int ret = 0; |
2f232036 JS |
2654 | |
2655 | path = btrfs_alloc_path(); | |
2656 | if (!path) | |
2657 | goto out; | |
2f232036 JS |
2658 | |
2659 | err = 0; | |
7343dd61 | 2660 | while (!err && !btrfs_fs_closing(fs_info)) { |
2f232036 JS |
2661 | trans = btrfs_start_transaction(fs_info->fs_root, 0); |
2662 | if (IS_ERR(trans)) { | |
2663 | err = PTR_ERR(trans); | |
2664 | break; | |
2665 | } | |
afcdd129 | 2666 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
2f232036 JS |
2667 | err = -EINTR; |
2668 | } else { | |
0a0e8b89 | 2669 | err = qgroup_rescan_leaf(fs_info, path, trans); |
2f232036 JS |
2670 | } |
2671 | if (err > 0) | |
3a45bb20 | 2672 | btrfs_commit_transaction(trans); |
2f232036 | 2673 | else |
3a45bb20 | 2674 | btrfs_end_transaction(trans); |
2f232036 JS |
2675 | } |
2676 | ||
2677 | out: | |
2f232036 | 2678 | btrfs_free_path(path); |
2f232036 JS |
2679 | |
2680 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
7343dd61 JM |
2681 | if (!btrfs_fs_closing(fs_info)) |
2682 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2f232036 | 2683 | |
3393168d | 2684 | if (err > 0 && |
2f232036 JS |
2685 | fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) { |
2686 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2687 | } else if (err < 0) { | |
2688 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT; | |
2689 | } | |
2690 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2691 | ||
53b7cde9 | 2692 | /* |
01327610 | 2693 | * only update status, since the previous part has already updated the |
53b7cde9 QW |
2694 | * qgroup info. |
2695 | */ | |
2696 | trans = btrfs_start_transaction(fs_info->quota_root, 1); | |
2697 | if (IS_ERR(trans)) { | |
2698 | err = PTR_ERR(trans); | |
2699 | btrfs_err(fs_info, | |
913e1535 | 2700 | "fail to start transaction for status update: %d", |
53b7cde9 QW |
2701 | err); |
2702 | goto done; | |
2703 | } | |
2704 | ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root); | |
2705 | if (ret < 0) { | |
2706 | err = ret; | |
ab8d0fc4 | 2707 | btrfs_err(fs_info, "fail to update qgroup status: %d", err); |
53b7cde9 | 2708 | } |
3a45bb20 | 2709 | btrfs_end_transaction(trans); |
53b7cde9 | 2710 | |
7343dd61 JM |
2711 | if (btrfs_fs_closing(fs_info)) { |
2712 | btrfs_info(fs_info, "qgroup scan paused"); | |
2713 | } else if (err >= 0) { | |
efe120a0 | 2714 | btrfs_info(fs_info, "qgroup scan completed%s", |
3393168d | 2715 | err > 0 ? " (inconsistency flag cleared)" : ""); |
2f232036 | 2716 | } else { |
efe120a0 | 2717 | btrfs_err(fs_info, "qgroup scan failed with %d", err); |
2f232036 | 2718 | } |
57254b6e | 2719 | |
53b7cde9 | 2720 | done: |
d2c609b8 JM |
2721 | mutex_lock(&fs_info->qgroup_rescan_lock); |
2722 | fs_info->qgroup_rescan_running = false; | |
2723 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
57254b6e | 2724 | complete_all(&fs_info->qgroup_rescan_completion); |
2f232036 JS |
2725 | } |
2726 | ||
b382a324 JS |
2727 | /* |
2728 | * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all | |
2729 | * memory required for the rescan context. | |
2730 | */ | |
2731 | static int | |
2732 | qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid, | |
2733 | int init_flags) | |
2f232036 JS |
2734 | { |
2735 | int ret = 0; | |
2f232036 | 2736 | |
b382a324 JS |
2737 | if (!init_flags && |
2738 | (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) || | |
2739 | !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) { | |
2740 | ret = -EINVAL; | |
2741 | goto err; | |
2742 | } | |
2f232036 JS |
2743 | |
2744 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
2745 | spin_lock(&fs_info->qgroup_lock); | |
b382a324 JS |
2746 | |
2747 | if (init_flags) { | |
2748 | if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) | |
2749 | ret = -EINPROGRESS; | |
2750 | else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) | |
2751 | ret = -EINVAL; | |
2752 | ||
2753 | if (ret) { | |
2754 | spin_unlock(&fs_info->qgroup_lock); | |
2755 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2756 | goto err; | |
2757 | } | |
b382a324 | 2758 | fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN; |
2f232036 JS |
2759 | } |
2760 | ||
2f232036 JS |
2761 | memset(&fs_info->qgroup_rescan_progress, 0, |
2762 | sizeof(fs_info->qgroup_rescan_progress)); | |
b382a324 | 2763 | fs_info->qgroup_rescan_progress.objectid = progress_objectid; |
190631f1 | 2764 | init_completion(&fs_info->qgroup_rescan_completion); |
8d9eddad | 2765 | fs_info->qgroup_rescan_running = true; |
b382a324 JS |
2766 | |
2767 | spin_unlock(&fs_info->qgroup_lock); | |
2768 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2769 | ||
b382a324 JS |
2770 | memset(&fs_info->qgroup_rescan_work, 0, |
2771 | sizeof(fs_info->qgroup_rescan_work)); | |
fc97fab0 | 2772 | btrfs_init_work(&fs_info->qgroup_rescan_work, |
9e0af237 | 2773 | btrfs_qgroup_rescan_helper, |
fc97fab0 | 2774 | btrfs_qgroup_rescan_worker, NULL, NULL); |
b382a324 JS |
2775 | |
2776 | if (ret) { | |
2777 | err: | |
efe120a0 | 2778 | btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret); |
b382a324 JS |
2779 | return ret; |
2780 | } | |
2781 | ||
2782 | return 0; | |
2783 | } | |
2784 | ||
2785 | static void | |
2786 | qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info) | |
2787 | { | |
2788 | struct rb_node *n; | |
2789 | struct btrfs_qgroup *qgroup; | |
2790 | ||
2791 | spin_lock(&fs_info->qgroup_lock); | |
2f232036 JS |
2792 | /* clear all current qgroup tracking information */ |
2793 | for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) { | |
2794 | qgroup = rb_entry(n, struct btrfs_qgroup, node); | |
2795 | qgroup->rfer = 0; | |
2796 | qgroup->rfer_cmpr = 0; | |
2797 | qgroup->excl = 0; | |
2798 | qgroup->excl_cmpr = 0; | |
2799 | } | |
2800 | spin_unlock(&fs_info->qgroup_lock); | |
b382a324 | 2801 | } |
2f232036 | 2802 | |
b382a324 JS |
2803 | int |
2804 | btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info) | |
2805 | { | |
2806 | int ret = 0; | |
2807 | struct btrfs_trans_handle *trans; | |
2808 | ||
2809 | ret = qgroup_rescan_init(fs_info, 0, 1); | |
2810 | if (ret) | |
2811 | return ret; | |
2812 | ||
2813 | /* | |
2814 | * We have set the rescan_progress to 0, which means no more | |
2815 | * delayed refs will be accounted by btrfs_qgroup_account_ref. | |
2816 | * However, btrfs_qgroup_account_ref may be right after its call | |
2817 | * to btrfs_find_all_roots, in which case it would still do the | |
2818 | * accounting. | |
2819 | * To solve this, we're committing the transaction, which will | |
2820 | * ensure we run all delayed refs and only after that, we are | |
2821 | * going to clear all tracking information for a clean start. | |
2822 | */ | |
2823 | ||
2824 | trans = btrfs_join_transaction(fs_info->fs_root); | |
2825 | if (IS_ERR(trans)) { | |
2826 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2827 | return PTR_ERR(trans); | |
2828 | } | |
3a45bb20 | 2829 | ret = btrfs_commit_transaction(trans); |
b382a324 JS |
2830 | if (ret) { |
2831 | fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN; | |
2832 | return ret; | |
2833 | } | |
2834 | ||
2835 | qgroup_rescan_zero_tracking(fs_info); | |
2836 | ||
fc97fab0 QW |
2837 | btrfs_queue_work(fs_info->qgroup_rescan_workers, |
2838 | &fs_info->qgroup_rescan_work); | |
2f232036 JS |
2839 | |
2840 | return 0; | |
2841 | } | |
57254b6e | 2842 | |
d06f23d6 JM |
2843 | int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info, |
2844 | bool interruptible) | |
57254b6e JS |
2845 | { |
2846 | int running; | |
2847 | int ret = 0; | |
2848 | ||
2849 | mutex_lock(&fs_info->qgroup_rescan_lock); | |
2850 | spin_lock(&fs_info->qgroup_lock); | |
d2c609b8 | 2851 | running = fs_info->qgroup_rescan_running; |
57254b6e JS |
2852 | spin_unlock(&fs_info->qgroup_lock); |
2853 | mutex_unlock(&fs_info->qgroup_rescan_lock); | |
2854 | ||
d06f23d6 JM |
2855 | if (!running) |
2856 | return 0; | |
2857 | ||
2858 | if (interruptible) | |
57254b6e JS |
2859 | ret = wait_for_completion_interruptible( |
2860 | &fs_info->qgroup_rescan_completion); | |
d06f23d6 JM |
2861 | else |
2862 | wait_for_completion(&fs_info->qgroup_rescan_completion); | |
57254b6e JS |
2863 | |
2864 | return ret; | |
2865 | } | |
b382a324 JS |
2866 | |
2867 | /* | |
2868 | * this is only called from open_ctree where we're still single threaded, thus | |
2869 | * locking is omitted here. | |
2870 | */ | |
2871 | void | |
2872 | btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info) | |
2873 | { | |
2874 | if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) | |
fc97fab0 QW |
2875 | btrfs_queue_work(fs_info->qgroup_rescan_workers, |
2876 | &fs_info->qgroup_rescan_work); | |
b382a324 | 2877 | } |
52472553 QW |
2878 | |
2879 | /* | |
2880 | * Reserve qgroup space for range [start, start + len). | |
2881 | * | |
2882 | * This function will either reserve space from related qgroups or doing | |
2883 | * nothing if the range is already reserved. | |
2884 | * | |
2885 | * Return 0 for successful reserve | |
2886 | * Return <0 for error (including -EQUOT) | |
2887 | * | |
2888 | * NOTE: this function may sleep for memory allocation. | |
364ecf36 QW |
2889 | * if btrfs_qgroup_reserve_data() is called multiple times with |
2890 | * same @reserved, caller must ensure when error happens it's OK | |
2891 | * to free *ALL* reserved space. | |
52472553 | 2892 | */ |
364ecf36 QW |
2893 | int btrfs_qgroup_reserve_data(struct inode *inode, |
2894 | struct extent_changeset **reserved_ret, u64 start, | |
2895 | u64 len) | |
52472553 QW |
2896 | { |
2897 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
52472553 QW |
2898 | struct ulist_node *unode; |
2899 | struct ulist_iterator uiter; | |
364ecf36 QW |
2900 | struct extent_changeset *reserved; |
2901 | u64 orig_reserved; | |
2902 | u64 to_reserve; | |
52472553 QW |
2903 | int ret; |
2904 | ||
afcdd129 JB |
2905 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) || |
2906 | !is_fstree(root->objectid) || len == 0) | |
52472553 QW |
2907 | return 0; |
2908 | ||
364ecf36 QW |
2909 | /* @reserved parameter is mandatory for qgroup */ |
2910 | if (WARN_ON(!reserved_ret)) | |
2911 | return -EINVAL; | |
2912 | if (!*reserved_ret) { | |
2913 | *reserved_ret = extent_changeset_alloc(); | |
2914 | if (!*reserved_ret) | |
2915 | return -ENOMEM; | |
2916 | } | |
2917 | reserved = *reserved_ret; | |
2918 | /* Record already reserved space */ | |
2919 | orig_reserved = reserved->bytes_changed; | |
52472553 | 2920 | ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start, |
364ecf36 QW |
2921 | start + len -1, EXTENT_QGROUP_RESERVED, reserved); |
2922 | ||
2923 | /* Newly reserved space */ | |
2924 | to_reserve = reserved->bytes_changed - orig_reserved; | |
81fb6f77 | 2925 | trace_btrfs_qgroup_reserve_data(inode, start, len, |
364ecf36 | 2926 | to_reserve, QGROUP_RESERVE); |
52472553 QW |
2927 | if (ret < 0) |
2928 | goto cleanup; | |
dba21324 | 2929 | ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA); |
52472553 QW |
2930 | if (ret < 0) |
2931 | goto cleanup; | |
2932 | ||
52472553 QW |
2933 | return ret; |
2934 | ||
2935 | cleanup: | |
364ecf36 | 2936 | /* cleanup *ALL* already reserved ranges */ |
52472553 | 2937 | ULIST_ITER_INIT(&uiter); |
364ecf36 | 2938 | while ((unode = ulist_next(&reserved->range_changed, &uiter))) |
52472553 | 2939 | clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val, |
ae0f1625 | 2940 | unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL); |
364ecf36 | 2941 | extent_changeset_release(reserved); |
52472553 QW |
2942 | return ret; |
2943 | } | |
f695fdce | 2944 | |
bc42bda2 QW |
2945 | /* Free ranges specified by @reserved, normally in error path */ |
2946 | static int qgroup_free_reserved_data(struct inode *inode, | |
2947 | struct extent_changeset *reserved, u64 start, u64 len) | |
2948 | { | |
2949 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2950 | struct ulist_node *unode; | |
2951 | struct ulist_iterator uiter; | |
2952 | struct extent_changeset changeset; | |
2953 | int freed = 0; | |
2954 | int ret; | |
2955 | ||
2956 | extent_changeset_init(&changeset); | |
2957 | len = round_up(start + len, root->fs_info->sectorsize); | |
2958 | start = round_down(start, root->fs_info->sectorsize); | |
2959 | ||
2960 | ULIST_ITER_INIT(&uiter); | |
2961 | while ((unode = ulist_next(&reserved->range_changed, &uiter))) { | |
2962 | u64 range_start = unode->val; | |
2963 | /* unode->aux is the inclusive end */ | |
2964 | u64 range_len = unode->aux - range_start + 1; | |
2965 | u64 free_start; | |
2966 | u64 free_len; | |
2967 | ||
2968 | extent_changeset_release(&changeset); | |
2969 | ||
2970 | /* Only free range in range [start, start + len) */ | |
2971 | if (range_start >= start + len || | |
2972 | range_start + range_len <= start) | |
2973 | continue; | |
2974 | free_start = max(range_start, start); | |
2975 | free_len = min(start + len, range_start + range_len) - | |
2976 | free_start; | |
2977 | /* | |
2978 | * TODO: To also modify reserved->ranges_reserved to reflect | |
2979 | * the modification. | |
2980 | * | |
2981 | * However as long as we free qgroup reserved according to | |
2982 | * EXTENT_QGROUP_RESERVED, we won't double free. | |
2983 | * So not need to rush. | |
2984 | */ | |
2985 | ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
2986 | free_start, free_start + free_len - 1, | |
2987 | EXTENT_QGROUP_RESERVED, &changeset); | |
2988 | if (ret < 0) | |
2989 | goto out; | |
2990 | freed += changeset.bytes_changed; | |
2991 | } | |
d4e5c920 QW |
2992 | btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed, |
2993 | BTRFS_QGROUP_RSV_DATA); | |
bc42bda2 QW |
2994 | ret = freed; |
2995 | out: | |
2996 | extent_changeset_release(&changeset); | |
2997 | return ret; | |
2998 | } | |
2999 | ||
3000 | static int __btrfs_qgroup_release_data(struct inode *inode, | |
3001 | struct extent_changeset *reserved, u64 start, u64 len, | |
3002 | int free) | |
f695fdce QW |
3003 | { |
3004 | struct extent_changeset changeset; | |
81fb6f77 | 3005 | int trace_op = QGROUP_RELEASE; |
f695fdce QW |
3006 | int ret; |
3007 | ||
bc42bda2 QW |
3008 | /* In release case, we shouldn't have @reserved */ |
3009 | WARN_ON(!free && reserved); | |
3010 | if (free && reserved) | |
3011 | return qgroup_free_reserved_data(inode, reserved, start, len); | |
364ecf36 | 3012 | extent_changeset_init(&changeset); |
f695fdce | 3013 | ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start, |
f734c44a | 3014 | start + len -1, EXTENT_QGROUP_RESERVED, &changeset); |
f695fdce QW |
3015 | if (ret < 0) |
3016 | goto out; | |
3017 | ||
d51ea5dd | 3018 | if (free) |
81fb6f77 | 3019 | trace_op = QGROUP_FREE; |
81fb6f77 QW |
3020 | trace_btrfs_qgroup_release_data(inode, start, len, |
3021 | changeset.bytes_changed, trace_op); | |
d51ea5dd QW |
3022 | if (free) |
3023 | btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info, | |
3024 | BTRFS_I(inode)->root->objectid, | |
d4e5c920 | 3025 | changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA); |
7bc329c1 | 3026 | ret = changeset.bytes_changed; |
f695fdce | 3027 | out: |
364ecf36 | 3028 | extent_changeset_release(&changeset); |
f695fdce QW |
3029 | return ret; |
3030 | } | |
3031 | ||
3032 | /* | |
3033 | * Free a reserved space range from io_tree and related qgroups | |
3034 | * | |
3035 | * Should be called when a range of pages get invalidated before reaching disk. | |
3036 | * Or for error cleanup case. | |
bc42bda2 QW |
3037 | * if @reserved is given, only reserved range in [@start, @start + @len) will |
3038 | * be freed. | |
f695fdce QW |
3039 | * |
3040 | * For data written to disk, use btrfs_qgroup_release_data(). | |
3041 | * | |
3042 | * NOTE: This function may sleep for memory allocation. | |
3043 | */ | |
bc42bda2 QW |
3044 | int btrfs_qgroup_free_data(struct inode *inode, |
3045 | struct extent_changeset *reserved, u64 start, u64 len) | |
f695fdce | 3046 | { |
bc42bda2 | 3047 | return __btrfs_qgroup_release_data(inode, reserved, start, len, 1); |
f695fdce QW |
3048 | } |
3049 | ||
3050 | /* | |
3051 | * Release a reserved space range from io_tree only. | |
3052 | * | |
3053 | * Should be called when a range of pages get written to disk and corresponding | |
3054 | * FILE_EXTENT is inserted into corresponding root. | |
3055 | * | |
3056 | * Since new qgroup accounting framework will only update qgroup numbers at | |
3057 | * commit_transaction() time, its reserved space shouldn't be freed from | |
3058 | * related qgroups. | |
3059 | * | |
3060 | * But we should release the range from io_tree, to allow further write to be | |
3061 | * COWed. | |
3062 | * | |
3063 | * NOTE: This function may sleep for memory allocation. | |
3064 | */ | |
3065 | int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len) | |
3066 | { | |
bc42bda2 | 3067 | return __btrfs_qgroup_release_data(inode, NULL, start, len, 0); |
f695fdce | 3068 | } |
55eeaf05 | 3069 | |
8287475a QW |
3070 | static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes, |
3071 | enum btrfs_qgroup_rsv_type type) | |
3072 | { | |
3073 | if (type != BTRFS_QGROUP_RSV_META_PREALLOC && | |
3074 | type != BTRFS_QGROUP_RSV_META_PERTRANS) | |
3075 | return; | |
3076 | if (num_bytes == 0) | |
3077 | return; | |
3078 | ||
3079 | spin_lock(&root->qgroup_meta_rsv_lock); | |
3080 | if (type == BTRFS_QGROUP_RSV_META_PREALLOC) | |
3081 | root->qgroup_meta_rsv_prealloc += num_bytes; | |
3082 | else | |
3083 | root->qgroup_meta_rsv_pertrans += num_bytes; | |
3084 | spin_unlock(&root->qgroup_meta_rsv_lock); | |
3085 | } | |
3086 | ||
3087 | static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes, | |
3088 | enum btrfs_qgroup_rsv_type type) | |
3089 | { | |
3090 | if (type != BTRFS_QGROUP_RSV_META_PREALLOC && | |
3091 | type != BTRFS_QGROUP_RSV_META_PERTRANS) | |
3092 | return 0; | |
3093 | if (num_bytes == 0) | |
3094 | return 0; | |
3095 | ||
3096 | spin_lock(&root->qgroup_meta_rsv_lock); | |
3097 | if (type == BTRFS_QGROUP_RSV_META_PREALLOC) { | |
3098 | num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc, | |
3099 | num_bytes); | |
3100 | root->qgroup_meta_rsv_prealloc -= num_bytes; | |
3101 | } else { | |
3102 | num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans, | |
3103 | num_bytes); | |
3104 | root->qgroup_meta_rsv_pertrans -= num_bytes; | |
3105 | } | |
3106 | spin_unlock(&root->qgroup_meta_rsv_lock); | |
3107 | return num_bytes; | |
3108 | } | |
3109 | ||
733e03a0 QW |
3110 | int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes, |
3111 | enum btrfs_qgroup_rsv_type type, bool enforce) | |
55eeaf05 | 3112 | { |
0b246afa | 3113 | struct btrfs_fs_info *fs_info = root->fs_info; |
55eeaf05 QW |
3114 | int ret; |
3115 | ||
0b246afa | 3116 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) || |
afcdd129 | 3117 | !is_fstree(root->objectid) || num_bytes == 0) |
55eeaf05 QW |
3118 | return 0; |
3119 | ||
0b246afa | 3120 | BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize)); |
4ee0d883 | 3121 | trace_qgroup_meta_reserve(root, type, (s64)num_bytes); |
733e03a0 | 3122 | ret = qgroup_reserve(root, num_bytes, enforce, type); |
55eeaf05 QW |
3123 | if (ret < 0) |
3124 | return ret; | |
8287475a QW |
3125 | /* |
3126 | * Record what we have reserved into root. | |
3127 | * | |
3128 | * To avoid quota disabled->enabled underflow. | |
3129 | * In that case, we may try to free space we haven't reserved | |
3130 | * (since quota was disabled), so record what we reserved into root. | |
3131 | * And ensure later release won't underflow this number. | |
3132 | */ | |
3133 | add_root_meta_rsv(root, num_bytes, type); | |
55eeaf05 QW |
3134 | return ret; |
3135 | } | |
3136 | ||
733e03a0 | 3137 | void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root) |
55eeaf05 | 3138 | { |
0b246afa | 3139 | struct btrfs_fs_info *fs_info = root->fs_info; |
55eeaf05 | 3140 | |
0b246afa | 3141 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) || |
afcdd129 | 3142 | !is_fstree(root->objectid)) |
55eeaf05 QW |
3143 | return; |
3144 | ||
e1211d0e | 3145 | /* TODO: Update trace point to handle such free */ |
4ee0d883 | 3146 | trace_qgroup_meta_free_all_pertrans(root); |
e1211d0e QW |
3147 | /* Special value -1 means to free all reserved space */ |
3148 | btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1, | |
733e03a0 | 3149 | BTRFS_QGROUP_RSV_META_PERTRANS); |
55eeaf05 QW |
3150 | } |
3151 | ||
733e03a0 QW |
3152 | void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes, |
3153 | enum btrfs_qgroup_rsv_type type) | |
55eeaf05 | 3154 | { |
0b246afa JM |
3155 | struct btrfs_fs_info *fs_info = root->fs_info; |
3156 | ||
3157 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) || | |
afcdd129 | 3158 | !is_fstree(root->objectid)) |
55eeaf05 QW |
3159 | return; |
3160 | ||
8287475a QW |
3161 | /* |
3162 | * reservation for META_PREALLOC can happen before quota is enabled, | |
3163 | * which can lead to underflow. | |
3164 | * Here ensure we will only free what we really have reserved. | |
3165 | */ | |
3166 | num_bytes = sub_root_meta_rsv(root, num_bytes, type); | |
0b246afa | 3167 | BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize)); |
4ee0d883 | 3168 | trace_qgroup_meta_reserve(root, type, -(s64)num_bytes); |
733e03a0 | 3169 | btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type); |
55eeaf05 | 3170 | } |
56fa9d07 | 3171 | |
64cfaef6 QW |
3172 | static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root, |
3173 | int num_bytes) | |
3174 | { | |
3175 | struct btrfs_root *quota_root = fs_info->quota_root; | |
3176 | struct btrfs_qgroup *qgroup; | |
3177 | struct ulist_node *unode; | |
3178 | struct ulist_iterator uiter; | |
3179 | int ret = 0; | |
3180 | ||
3181 | if (num_bytes == 0) | |
3182 | return; | |
3183 | if (!quota_root) | |
3184 | return; | |
3185 | ||
3186 | spin_lock(&fs_info->qgroup_lock); | |
3187 | qgroup = find_qgroup_rb(fs_info, ref_root); | |
3188 | if (!qgroup) | |
3189 | goto out; | |
3190 | ulist_reinit(fs_info->qgroup_ulist); | |
3191 | ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid, | |
a1840b50 | 3192 | qgroup_to_aux(qgroup), GFP_ATOMIC); |
64cfaef6 QW |
3193 | if (ret < 0) |
3194 | goto out; | |
3195 | ULIST_ITER_INIT(&uiter); | |
3196 | while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) { | |
3197 | struct btrfs_qgroup *qg; | |
3198 | struct btrfs_qgroup_list *glist; | |
3199 | ||
3200 | qg = unode_aux_to_qgroup(unode); | |
3201 | ||
3202 | qgroup_rsv_release(fs_info, qg, num_bytes, | |
3203 | BTRFS_QGROUP_RSV_META_PREALLOC); | |
3204 | qgroup_rsv_add(fs_info, qg, num_bytes, | |
3205 | BTRFS_QGROUP_RSV_META_PERTRANS); | |
3206 | list_for_each_entry(glist, &qg->groups, next_group) { | |
3207 | ret = ulist_add(fs_info->qgroup_ulist, | |
3208 | glist->group->qgroupid, | |
a1840b50 | 3209 | qgroup_to_aux(glist->group), GFP_ATOMIC); |
64cfaef6 QW |
3210 | if (ret < 0) |
3211 | goto out; | |
3212 | } | |
3213 | } | |
3214 | out: | |
3215 | spin_unlock(&fs_info->qgroup_lock); | |
3216 | } | |
3217 | ||
3218 | void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes) | |
3219 | { | |
3220 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3221 | ||
3222 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) || | |
3223 | !is_fstree(root->objectid)) | |
3224 | return; | |
8287475a QW |
3225 | /* Same as btrfs_qgroup_free_meta_prealloc() */ |
3226 | num_bytes = sub_root_meta_rsv(root, num_bytes, | |
3227 | BTRFS_QGROUP_RSV_META_PREALLOC); | |
4ee0d883 | 3228 | trace_qgroup_meta_convert(root, num_bytes); |
64cfaef6 QW |
3229 | qgroup_convert_meta(fs_info, root->objectid, num_bytes); |
3230 | } | |
3231 | ||
56fa9d07 | 3232 | /* |
01327610 | 3233 | * Check qgroup reserved space leaking, normally at destroy inode |
56fa9d07 QW |
3234 | * time |
3235 | */ | |
3236 | void btrfs_qgroup_check_reserved_leak(struct inode *inode) | |
3237 | { | |
3238 | struct extent_changeset changeset; | |
3239 | struct ulist_node *unode; | |
3240 | struct ulist_iterator iter; | |
3241 | int ret; | |
3242 | ||
364ecf36 | 3243 | extent_changeset_init(&changeset); |
56fa9d07 | 3244 | ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1, |
f734c44a | 3245 | EXTENT_QGROUP_RESERVED, &changeset); |
56fa9d07 QW |
3246 | |
3247 | WARN_ON(ret < 0); | |
3248 | if (WARN_ON(changeset.bytes_changed)) { | |
3249 | ULIST_ITER_INIT(&iter); | |
53d32359 | 3250 | while ((unode = ulist_next(&changeset.range_changed, &iter))) { |
56fa9d07 QW |
3251 | btrfs_warn(BTRFS_I(inode)->root->fs_info, |
3252 | "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu", | |
3253 | inode->i_ino, unode->val, unode->aux); | |
3254 | } | |
0b08e1f4 DS |
3255 | btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info, |
3256 | BTRFS_I(inode)->root->objectid, | |
d4e5c920 | 3257 | changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA); |
0b08e1f4 | 3258 | |
56fa9d07 | 3259 | } |
364ecf36 | 3260 | extent_changeset_release(&changeset); |
56fa9d07 | 3261 | } |