Loading...
1// SPDX-License-Identifier: GPL-2.0
2
3#include "misc.h"
4#include "ctree.h"
5#include "block-rsv.h"
6#include "space-info.h"
7#include "transaction.h"
8
9static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
10 struct btrfs_block_rsv *block_rsv,
11 struct btrfs_block_rsv *dest, u64 num_bytes,
12 u64 *qgroup_to_release_ret)
13{
14 struct btrfs_space_info *space_info = block_rsv->space_info;
15 u64 qgroup_to_release = 0;
16 u64 ret;
17
18 spin_lock(&block_rsv->lock);
19 if (num_bytes == (u64)-1) {
20 num_bytes = block_rsv->size;
21 qgroup_to_release = block_rsv->qgroup_rsv_size;
22 }
23 block_rsv->size -= num_bytes;
24 if (block_rsv->reserved >= block_rsv->size) {
25 num_bytes = block_rsv->reserved - block_rsv->size;
26 block_rsv->reserved = block_rsv->size;
27 block_rsv->full = 1;
28 } else {
29 num_bytes = 0;
30 }
31 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
32 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
33 block_rsv->qgroup_rsv_size;
34 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
35 } else {
36 qgroup_to_release = 0;
37 }
38 spin_unlock(&block_rsv->lock);
39
40 ret = num_bytes;
41 if (num_bytes > 0) {
42 if (dest) {
43 spin_lock(&dest->lock);
44 if (!dest->full) {
45 u64 bytes_to_add;
46
47 bytes_to_add = dest->size - dest->reserved;
48 bytes_to_add = min(num_bytes, bytes_to_add);
49 dest->reserved += bytes_to_add;
50 if (dest->reserved >= dest->size)
51 dest->full = 1;
52 num_bytes -= bytes_to_add;
53 }
54 spin_unlock(&dest->lock);
55 }
56 if (num_bytes)
57 btrfs_space_info_free_bytes_may_use(fs_info,
58 space_info,
59 num_bytes);
60 }
61 if (qgroup_to_release_ret)
62 *qgroup_to_release_ret = qgroup_to_release;
63 return ret;
64}
65
66int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
67 struct btrfs_block_rsv *dst, u64 num_bytes,
68 bool update_size)
69{
70 int ret;
71
72 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
73 if (ret)
74 return ret;
75
76 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
77 return 0;
78}
79
80void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
81{
82 memset(rsv, 0, sizeof(*rsv));
83 spin_lock_init(&rsv->lock);
84 rsv->type = type;
85}
86
87void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
88 struct btrfs_block_rsv *rsv,
89 unsigned short type)
90{
91 btrfs_init_block_rsv(rsv, type);
92 rsv->space_info = btrfs_find_space_info(fs_info,
93 BTRFS_BLOCK_GROUP_METADATA);
94}
95
96struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
97 unsigned short type)
98{
99 struct btrfs_block_rsv *block_rsv;
100
101 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
102 if (!block_rsv)
103 return NULL;
104
105 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
106 return block_rsv;
107}
108
109void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
110 struct btrfs_block_rsv *rsv)
111{
112 if (!rsv)
113 return;
114 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
115 kfree(rsv);
116}
117
118int btrfs_block_rsv_add(struct btrfs_root *root,
119 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
120 enum btrfs_reserve_flush_enum flush)
121{
122 int ret;
123
124 if (num_bytes == 0)
125 return 0;
126
127 ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
128 if (!ret)
129 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
130
131 return ret;
132}
133
134int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
135{
136 u64 num_bytes = 0;
137 int ret = -ENOSPC;
138
139 if (!block_rsv)
140 return 0;
141
142 spin_lock(&block_rsv->lock);
143 num_bytes = div_factor(block_rsv->size, min_factor);
144 if (block_rsv->reserved >= num_bytes)
145 ret = 0;
146 spin_unlock(&block_rsv->lock);
147
148 return ret;
149}
150
151int btrfs_block_rsv_refill(struct btrfs_root *root,
152 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
153 enum btrfs_reserve_flush_enum flush)
154{
155 u64 num_bytes = 0;
156 int ret = -ENOSPC;
157
158 if (!block_rsv)
159 return 0;
160
161 spin_lock(&block_rsv->lock);
162 num_bytes = min_reserved;
163 if (block_rsv->reserved >= num_bytes)
164 ret = 0;
165 else
166 num_bytes -= block_rsv->reserved;
167 spin_unlock(&block_rsv->lock);
168
169 if (!ret)
170 return 0;
171
172 ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
173 if (!ret) {
174 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
175 return 0;
176 }
177
178 return ret;
179}
180
181u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
182 struct btrfs_block_rsv *block_rsv,
183 u64 num_bytes, u64 *qgroup_to_release)
184{
185 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
186 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
187 struct btrfs_block_rsv *target = NULL;
188
189 /*
190 * If we are the delayed_rsv then push to the global rsv, otherwise dump
191 * into the delayed rsv if it is not full.
192 */
193 if (block_rsv == delayed_rsv)
194 target = global_rsv;
195 else if (block_rsv != global_rsv && !delayed_rsv->full)
196 target = delayed_rsv;
197
198 if (target && block_rsv->space_info != target->space_info)
199 target = NULL;
200
201 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
202 qgroup_to_release);
203}
204
205int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
206{
207 int ret = -ENOSPC;
208
209 spin_lock(&block_rsv->lock);
210 if (block_rsv->reserved >= num_bytes) {
211 block_rsv->reserved -= num_bytes;
212 if (block_rsv->reserved < block_rsv->size)
213 block_rsv->full = 0;
214 ret = 0;
215 }
216 spin_unlock(&block_rsv->lock);
217 return ret;
218}
219
220void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
221 u64 num_bytes, bool update_size)
222{
223 spin_lock(&block_rsv->lock);
224 block_rsv->reserved += num_bytes;
225 if (update_size)
226 block_rsv->size += num_bytes;
227 else if (block_rsv->reserved >= block_rsv->size)
228 block_rsv->full = 1;
229 spin_unlock(&block_rsv->lock);
230}
231
232int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
233 struct btrfs_block_rsv *dest, u64 num_bytes,
234 int min_factor)
235{
236 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
237 u64 min_bytes;
238
239 if (global_rsv->space_info != dest->space_info)
240 return -ENOSPC;
241
242 spin_lock(&global_rsv->lock);
243 min_bytes = div_factor(global_rsv->size, min_factor);
244 if (global_rsv->reserved < min_bytes + num_bytes) {
245 spin_unlock(&global_rsv->lock);
246 return -ENOSPC;
247 }
248 global_rsv->reserved -= num_bytes;
249 if (global_rsv->reserved < global_rsv->size)
250 global_rsv->full = 0;
251 spin_unlock(&global_rsv->lock);
252
253 btrfs_block_rsv_add_bytes(dest, num_bytes, true);
254 return 0;
255}
256
257void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
258{
259 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
260 struct btrfs_space_info *sinfo = block_rsv->space_info;
261 u64 num_bytes;
262 unsigned min_items;
263
264 /*
265 * The global block rsv is based on the size of the extent tree, the
266 * checksum tree and the root tree. If the fs is empty we want to set
267 * it to a minimal amount for safety.
268 */
269 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
270 btrfs_root_used(&fs_info->csum_root->root_item) +
271 btrfs_root_used(&fs_info->tree_root->root_item);
272
273 /*
274 * We at a minimum are going to modify the csum root, the tree root, and
275 * the extent root.
276 */
277 min_items = 3;
278
279 /*
280 * But we also want to reserve enough space so we can do the fallback
281 * global reserve for an unlink, which is an additional 5 items (see the
282 * comment in __unlink_start_trans for what we're modifying.)
283 *
284 * But we also need space for the delayed ref updates from the unlink,
285 * so its 10, 5 for the actual operation, and 5 for the delayed ref
286 * updates.
287 */
288 min_items += 10;
289
290 num_bytes = max_t(u64, num_bytes,
291 btrfs_calc_insert_metadata_size(fs_info, min_items));
292
293 spin_lock(&sinfo->lock);
294 spin_lock(&block_rsv->lock);
295
296 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
297
298 if (block_rsv->reserved < block_rsv->size) {
299 num_bytes = block_rsv->size - block_rsv->reserved;
300 block_rsv->reserved += num_bytes;
301 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
302 num_bytes);
303 } else if (block_rsv->reserved > block_rsv->size) {
304 num_bytes = block_rsv->reserved - block_rsv->size;
305 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
306 -num_bytes);
307 block_rsv->reserved = block_rsv->size;
308 btrfs_try_granting_tickets(fs_info, sinfo);
309 }
310
311 if (block_rsv->reserved == block_rsv->size)
312 block_rsv->full = 1;
313 else
314 block_rsv->full = 0;
315
316 spin_unlock(&block_rsv->lock);
317 spin_unlock(&sinfo->lock);
318}
319
320void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
321{
322 struct btrfs_space_info *space_info;
323
324 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
325 fs_info->chunk_block_rsv.space_info = space_info;
326
327 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
328 fs_info->global_block_rsv.space_info = space_info;
329 fs_info->trans_block_rsv.space_info = space_info;
330 fs_info->empty_block_rsv.space_info = space_info;
331 fs_info->delayed_block_rsv.space_info = space_info;
332 fs_info->delayed_refs_rsv.space_info = space_info;
333
334 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
335 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
336 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
337 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
338 if (fs_info->quota_root)
339 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
340 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
341
342 btrfs_update_global_block_rsv(fs_info);
343}
344
345void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
346{
347 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1);
348 WARN_ON(fs_info->trans_block_rsv.size > 0);
349 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
350 WARN_ON(fs_info->chunk_block_rsv.size > 0);
351 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
352 WARN_ON(fs_info->delayed_block_rsv.size > 0);
353 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
354 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
355 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
356}
357
358static struct btrfs_block_rsv *get_block_rsv(
359 const struct btrfs_trans_handle *trans,
360 const struct btrfs_root *root)
361{
362 struct btrfs_fs_info *fs_info = root->fs_info;
363 struct btrfs_block_rsv *block_rsv = NULL;
364
365 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
366 (root == fs_info->csum_root && trans->adding_csums) ||
367 (root == fs_info->uuid_root))
368 block_rsv = trans->block_rsv;
369
370 if (!block_rsv)
371 block_rsv = root->block_rsv;
372
373 if (!block_rsv)
374 block_rsv = &fs_info->empty_block_rsv;
375
376 return block_rsv;
377}
378
379struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
380 struct btrfs_root *root,
381 u32 blocksize)
382{
383 struct btrfs_fs_info *fs_info = root->fs_info;
384 struct btrfs_block_rsv *block_rsv;
385 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
386 int ret;
387 bool global_updated = false;
388
389 block_rsv = get_block_rsv(trans, root);
390
391 if (unlikely(block_rsv->size == 0))
392 goto try_reserve;
393again:
394 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
395 if (!ret)
396 return block_rsv;
397
398 if (block_rsv->failfast)
399 return ERR_PTR(ret);
400
401 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
402 global_updated = true;
403 btrfs_update_global_block_rsv(fs_info);
404 goto again;
405 }
406
407 /*
408 * The global reserve still exists to save us from ourselves, so don't
409 * warn_on if we are short on our delayed refs reserve.
410 */
411 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
412 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
413 static DEFINE_RATELIMIT_STATE(_rs,
414 DEFAULT_RATELIMIT_INTERVAL * 10,
415 /*DEFAULT_RATELIMIT_BURST*/ 1);
416 if (__ratelimit(&_rs))
417 WARN(1, KERN_DEBUG
418 "BTRFS: block rsv returned %d\n", ret);
419 }
420try_reserve:
421 ret = btrfs_reserve_metadata_bytes(root, block_rsv, blocksize,
422 BTRFS_RESERVE_NO_FLUSH);
423 if (!ret)
424 return block_rsv;
425 /*
426 * If we couldn't reserve metadata bytes try and use some from
427 * the global reserve if its space type is the same as the global
428 * reservation.
429 */
430 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
431 block_rsv->space_info == global_rsv->space_info) {
432 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
433 if (!ret)
434 return global_rsv;
435 }
436 return ERR_PTR(ret);
437}
1// SPDX-License-Identifier: GPL-2.0
2
3#include "misc.h"
4#include "ctree.h"
5#include "block-rsv.h"
6#include "space-info.h"
7#include "transaction.h"
8#include "block-group.h"
9
10/*
11 * HOW DO BLOCK RESERVES WORK
12 *
13 * Think of block_rsv's as buckets for logically grouped metadata
14 * reservations. Each block_rsv has a ->size and a ->reserved. ->size is
15 * how large we want our block rsv to be, ->reserved is how much space is
16 * currently reserved for this block reserve.
17 *
18 * ->failfast exists for the truncate case, and is described below.
19 *
20 * NORMAL OPERATION
21 *
22 * -> Reserve
23 * Entrance: btrfs_block_rsv_add, btrfs_block_rsv_refill
24 *
25 * We call into btrfs_reserve_metadata_bytes() with our bytes, which is
26 * accounted for in space_info->bytes_may_use, and then add the bytes to
27 * ->reserved, and ->size in the case of btrfs_block_rsv_add.
28 *
29 * ->size is an over-estimation of how much we may use for a particular
30 * operation.
31 *
32 * -> Use
33 * Entrance: btrfs_use_block_rsv
34 *
35 * When we do a btrfs_alloc_tree_block() we call into btrfs_use_block_rsv()
36 * to determine the appropriate block_rsv to use, and then verify that
37 * ->reserved has enough space for our tree block allocation. Once
38 * successful we subtract fs_info->nodesize from ->reserved.
39 *
40 * -> Finish
41 * Entrance: btrfs_block_rsv_release
42 *
43 * We are finished with our operation, subtract our individual reservation
44 * from ->size, and then subtract ->size from ->reserved and free up the
45 * excess if there is any.
46 *
47 * There is some logic here to refill the delayed refs rsv or the global rsv
48 * as needed, otherwise the excess is subtracted from
49 * space_info->bytes_may_use.
50 *
51 * TYPES OF BLOCK RESERVES
52 *
53 * BLOCK_RSV_TRANS, BLOCK_RSV_DELOPS, BLOCK_RSV_CHUNK
54 * These behave normally, as described above, just within the confines of the
55 * lifetime of their particular operation (transaction for the whole trans
56 * handle lifetime, for example).
57 *
58 * BLOCK_RSV_GLOBAL
59 * It is impossible to properly account for all the space that may be required
60 * to make our extent tree updates. This block reserve acts as an overflow
61 * buffer in case our delayed refs reserve does not reserve enough space to
62 * update the extent tree.
63 *
64 * We can steal from this in some cases as well, notably on evict() or
65 * truncate() in order to help users recover from ENOSPC conditions.
66 *
67 * BLOCK_RSV_DELALLOC
68 * The individual item sizes are determined by the per-inode size
69 * calculations, which are described with the delalloc code. This is pretty
70 * straightforward, it's just the calculation of ->size encodes a lot of
71 * different items, and thus it gets used when updating inodes, inserting file
72 * extents, and inserting checksums.
73 *
74 * BLOCK_RSV_DELREFS
75 * We keep a running tally of how many delayed refs we have on the system.
76 * We assume each one of these delayed refs are going to use a full
77 * reservation. We use the transaction items and pre-reserve space for every
78 * operation, and use this reservation to refill any gap between ->size and
79 * ->reserved that may exist.
80 *
81 * From there it's straightforward, removing a delayed ref means we remove its
82 * count from ->size and free up reservations as necessary. Since this is
83 * the most dynamic block reserve in the system, we will try to refill this
84 * block reserve first with any excess returned by any other block reserve.
85 *
86 * BLOCK_RSV_EMPTY
87 * This is the fallback block reserve to make us try to reserve space if we
88 * don't have a specific bucket for this allocation. It is mostly used for
89 * updating the device tree and such, since that is a separate pool we're
90 * content to just reserve space from the space_info on demand.
91 *
92 * BLOCK_RSV_TEMP
93 * This is used by things like truncate and iput. We will temporarily
94 * allocate a block reserve, set it to some size, and then truncate bytes
95 * until we have no space left. With ->failfast set we'll simply return
96 * ENOSPC from btrfs_use_block_rsv() to signal that we need to unwind and try
97 * to make a new reservation. This is because these operations are
98 * unbounded, so we want to do as much work as we can, and then back off and
99 * re-reserve.
100 */
101
102static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
103 struct btrfs_block_rsv *block_rsv,
104 struct btrfs_block_rsv *dest, u64 num_bytes,
105 u64 *qgroup_to_release_ret)
106{
107 struct btrfs_space_info *space_info = block_rsv->space_info;
108 u64 qgroup_to_release = 0;
109 u64 ret;
110
111 spin_lock(&block_rsv->lock);
112 if (num_bytes == (u64)-1) {
113 num_bytes = block_rsv->size;
114 qgroup_to_release = block_rsv->qgroup_rsv_size;
115 }
116 block_rsv->size -= num_bytes;
117 if (block_rsv->reserved >= block_rsv->size) {
118 num_bytes = block_rsv->reserved - block_rsv->size;
119 block_rsv->reserved = block_rsv->size;
120 block_rsv->full = 1;
121 } else {
122 num_bytes = 0;
123 }
124 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
125 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
126 block_rsv->qgroup_rsv_size;
127 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
128 } else {
129 qgroup_to_release = 0;
130 }
131 spin_unlock(&block_rsv->lock);
132
133 ret = num_bytes;
134 if (num_bytes > 0) {
135 if (dest) {
136 spin_lock(&dest->lock);
137 if (!dest->full) {
138 u64 bytes_to_add;
139
140 bytes_to_add = dest->size - dest->reserved;
141 bytes_to_add = min(num_bytes, bytes_to_add);
142 dest->reserved += bytes_to_add;
143 if (dest->reserved >= dest->size)
144 dest->full = 1;
145 num_bytes -= bytes_to_add;
146 }
147 spin_unlock(&dest->lock);
148 }
149 if (num_bytes)
150 btrfs_space_info_free_bytes_may_use(fs_info,
151 space_info,
152 num_bytes);
153 }
154 if (qgroup_to_release_ret)
155 *qgroup_to_release_ret = qgroup_to_release;
156 return ret;
157}
158
159int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
160 struct btrfs_block_rsv *dst, u64 num_bytes,
161 bool update_size)
162{
163 int ret;
164
165 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
166 if (ret)
167 return ret;
168
169 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
170 return 0;
171}
172
173void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
174{
175 memset(rsv, 0, sizeof(*rsv));
176 spin_lock_init(&rsv->lock);
177 rsv->type = type;
178}
179
180void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
181 struct btrfs_block_rsv *rsv,
182 unsigned short type)
183{
184 btrfs_init_block_rsv(rsv, type);
185 rsv->space_info = btrfs_find_space_info(fs_info,
186 BTRFS_BLOCK_GROUP_METADATA);
187}
188
189struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
190 unsigned short type)
191{
192 struct btrfs_block_rsv *block_rsv;
193
194 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
195 if (!block_rsv)
196 return NULL;
197
198 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
199 return block_rsv;
200}
201
202void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
203 struct btrfs_block_rsv *rsv)
204{
205 if (!rsv)
206 return;
207 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
208 kfree(rsv);
209}
210
211int btrfs_block_rsv_add(struct btrfs_root *root,
212 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
213 enum btrfs_reserve_flush_enum flush)
214{
215 int ret;
216
217 if (num_bytes == 0)
218 return 0;
219
220 ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
221 if (!ret)
222 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
223
224 return ret;
225}
226
227int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
228{
229 u64 num_bytes = 0;
230 int ret = -ENOSPC;
231
232 if (!block_rsv)
233 return 0;
234
235 spin_lock(&block_rsv->lock);
236 num_bytes = div_factor(block_rsv->size, min_factor);
237 if (block_rsv->reserved >= num_bytes)
238 ret = 0;
239 spin_unlock(&block_rsv->lock);
240
241 return ret;
242}
243
244int btrfs_block_rsv_refill(struct btrfs_root *root,
245 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
246 enum btrfs_reserve_flush_enum flush)
247{
248 u64 num_bytes = 0;
249 int ret = -ENOSPC;
250
251 if (!block_rsv)
252 return 0;
253
254 spin_lock(&block_rsv->lock);
255 num_bytes = min_reserved;
256 if (block_rsv->reserved >= num_bytes)
257 ret = 0;
258 else
259 num_bytes -= block_rsv->reserved;
260 spin_unlock(&block_rsv->lock);
261
262 if (!ret)
263 return 0;
264
265 ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
266 if (!ret) {
267 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
268 return 0;
269 }
270
271 return ret;
272}
273
274u64 btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
275 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
276 u64 *qgroup_to_release)
277{
278 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
279 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
280 struct btrfs_block_rsv *target = NULL;
281
282 /*
283 * If we are the delayed_rsv then push to the global rsv, otherwise dump
284 * into the delayed rsv if it is not full.
285 */
286 if (block_rsv == delayed_rsv)
287 target = global_rsv;
288 else if (block_rsv != global_rsv && !delayed_rsv->full)
289 target = delayed_rsv;
290
291 if (target && block_rsv->space_info != target->space_info)
292 target = NULL;
293
294 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
295 qgroup_to_release);
296}
297
298int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
299{
300 int ret = -ENOSPC;
301
302 spin_lock(&block_rsv->lock);
303 if (block_rsv->reserved >= num_bytes) {
304 block_rsv->reserved -= num_bytes;
305 if (block_rsv->reserved < block_rsv->size)
306 block_rsv->full = 0;
307 ret = 0;
308 }
309 spin_unlock(&block_rsv->lock);
310 return ret;
311}
312
313void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
314 u64 num_bytes, bool update_size)
315{
316 spin_lock(&block_rsv->lock);
317 block_rsv->reserved += num_bytes;
318 if (update_size)
319 block_rsv->size += num_bytes;
320 else if (block_rsv->reserved >= block_rsv->size)
321 block_rsv->full = 1;
322 spin_unlock(&block_rsv->lock);
323}
324
325int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
326 struct btrfs_block_rsv *dest, u64 num_bytes,
327 int min_factor)
328{
329 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
330 u64 min_bytes;
331
332 if (global_rsv->space_info != dest->space_info)
333 return -ENOSPC;
334
335 spin_lock(&global_rsv->lock);
336 min_bytes = div_factor(global_rsv->size, min_factor);
337 if (global_rsv->reserved < min_bytes + num_bytes) {
338 spin_unlock(&global_rsv->lock);
339 return -ENOSPC;
340 }
341 global_rsv->reserved -= num_bytes;
342 if (global_rsv->reserved < global_rsv->size)
343 global_rsv->full = 0;
344 spin_unlock(&global_rsv->lock);
345
346 btrfs_block_rsv_add_bytes(dest, num_bytes, true);
347 return 0;
348}
349
350void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
351{
352 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
353 struct btrfs_space_info *sinfo = block_rsv->space_info;
354 u64 num_bytes;
355 unsigned min_items;
356
357 /*
358 * The global block rsv is based on the size of the extent tree, the
359 * checksum tree and the root tree. If the fs is empty we want to set
360 * it to a minimal amount for safety.
361 */
362 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
363 btrfs_root_used(&fs_info->csum_root->root_item) +
364 btrfs_root_used(&fs_info->tree_root->root_item);
365
366 /*
367 * We at a minimum are going to modify the csum root, the tree root, and
368 * the extent root.
369 */
370 min_items = 3;
371
372 /*
373 * But we also want to reserve enough space so we can do the fallback
374 * global reserve for an unlink, which is an additional 5 items (see the
375 * comment in __unlink_start_trans for what we're modifying.)
376 *
377 * But we also need space for the delayed ref updates from the unlink,
378 * so its 10, 5 for the actual operation, and 5 for the delayed ref
379 * updates.
380 */
381 min_items += 10;
382
383 num_bytes = max_t(u64, num_bytes,
384 btrfs_calc_insert_metadata_size(fs_info, min_items));
385
386 spin_lock(&sinfo->lock);
387 spin_lock(&block_rsv->lock);
388
389 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
390
391 if (block_rsv->reserved < block_rsv->size) {
392 num_bytes = block_rsv->size - block_rsv->reserved;
393 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
394 num_bytes);
395 block_rsv->reserved = block_rsv->size;
396 } else if (block_rsv->reserved > block_rsv->size) {
397 num_bytes = block_rsv->reserved - block_rsv->size;
398 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
399 -num_bytes);
400 block_rsv->reserved = block_rsv->size;
401 btrfs_try_granting_tickets(fs_info, sinfo);
402 }
403
404 if (block_rsv->reserved == block_rsv->size)
405 block_rsv->full = 1;
406 else
407 block_rsv->full = 0;
408
409 if (block_rsv->size >= sinfo->total_bytes)
410 sinfo->force_alloc = CHUNK_ALLOC_FORCE;
411 spin_unlock(&block_rsv->lock);
412 spin_unlock(&sinfo->lock);
413}
414
415void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
416{
417 struct btrfs_space_info *space_info;
418
419 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
420 fs_info->chunk_block_rsv.space_info = space_info;
421
422 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
423 fs_info->global_block_rsv.space_info = space_info;
424 fs_info->trans_block_rsv.space_info = space_info;
425 fs_info->empty_block_rsv.space_info = space_info;
426 fs_info->delayed_block_rsv.space_info = space_info;
427 fs_info->delayed_refs_rsv.space_info = space_info;
428
429 /*
430 * Our various recovery options can leave us with NULL roots, so check
431 * here and just bail before we go dereferencing NULLs everywhere.
432 */
433 if (!fs_info->extent_root || !fs_info->csum_root ||
434 !fs_info->dev_root || !fs_info->chunk_root || !fs_info->tree_root)
435 return;
436
437 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
438 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
439 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
440 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
441 if (fs_info->quota_root)
442 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
443 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
444
445 btrfs_update_global_block_rsv(fs_info);
446}
447
448void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
449{
450 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
451 NULL);
452 WARN_ON(fs_info->trans_block_rsv.size > 0);
453 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
454 WARN_ON(fs_info->chunk_block_rsv.size > 0);
455 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
456 WARN_ON(fs_info->delayed_block_rsv.size > 0);
457 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
458 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
459 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
460}
461
462static struct btrfs_block_rsv *get_block_rsv(
463 const struct btrfs_trans_handle *trans,
464 const struct btrfs_root *root)
465{
466 struct btrfs_fs_info *fs_info = root->fs_info;
467 struct btrfs_block_rsv *block_rsv = NULL;
468
469 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
470 (root == fs_info->csum_root && trans->adding_csums) ||
471 (root == fs_info->uuid_root))
472 block_rsv = trans->block_rsv;
473
474 if (!block_rsv)
475 block_rsv = root->block_rsv;
476
477 if (!block_rsv)
478 block_rsv = &fs_info->empty_block_rsv;
479
480 return block_rsv;
481}
482
483struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
484 struct btrfs_root *root,
485 u32 blocksize)
486{
487 struct btrfs_fs_info *fs_info = root->fs_info;
488 struct btrfs_block_rsv *block_rsv;
489 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
490 int ret;
491 bool global_updated = false;
492
493 block_rsv = get_block_rsv(trans, root);
494
495 if (unlikely(block_rsv->size == 0))
496 goto try_reserve;
497again:
498 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
499 if (!ret)
500 return block_rsv;
501
502 if (block_rsv->failfast)
503 return ERR_PTR(ret);
504
505 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
506 global_updated = true;
507 btrfs_update_global_block_rsv(fs_info);
508 goto again;
509 }
510
511 /*
512 * The global reserve still exists to save us from ourselves, so don't
513 * warn_on if we are short on our delayed refs reserve.
514 */
515 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
516 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
517 static DEFINE_RATELIMIT_STATE(_rs,
518 DEFAULT_RATELIMIT_INTERVAL * 10,
519 /*DEFAULT_RATELIMIT_BURST*/ 1);
520 if (__ratelimit(&_rs))
521 WARN(1, KERN_DEBUG
522 "BTRFS: block rsv %d returned %d\n",
523 block_rsv->type, ret);
524 }
525try_reserve:
526 ret = btrfs_reserve_metadata_bytes(root, block_rsv, blocksize,
527 BTRFS_RESERVE_NO_FLUSH);
528 if (!ret)
529 return block_rsv;
530 /*
531 * If we couldn't reserve metadata bytes try and use some from
532 * the global reserve if its space type is the same as the global
533 * reservation.
534 */
535 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
536 block_rsv->space_info == global_rsv->space_info) {
537 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
538 if (!ret)
539 return global_rsv;
540 }
541 return ERR_PTR(ret);
542}