Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2
3#ifndef BTRFS_SPACE_INFO_H
4#define BTRFS_SPACE_INFO_H
5
6#include "volumes.h"
7
8/*
9 * Different levels for to flush space when doing space reservations.
10 *
11 * The higher the level, the more methods we try to reclaim space.
12 */
13enum btrfs_reserve_flush_enum {
14 /* If we are in the transaction, we can't flush anything.*/
15 BTRFS_RESERVE_NO_FLUSH,
16
17 /*
18 * Flush space by:
19 * - Running delayed inode items
20 * - Allocating a new chunk
21 */
22 BTRFS_RESERVE_FLUSH_LIMIT,
23
24 /*
25 * Flush space by:
26 * - Running delayed inode items
27 * - Running delayed refs
28 * - Running delalloc and waiting for ordered extents
29 * - Allocating a new chunk
30 */
31 BTRFS_RESERVE_FLUSH_EVICT,
32
33 /*
34 * Flush space by above mentioned methods and by:
35 * - Running delayed iputs
36 * - Committing transaction
37 *
38 * Can be interrupted by a fatal signal.
39 */
40 BTRFS_RESERVE_FLUSH_DATA,
41 BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE,
42 BTRFS_RESERVE_FLUSH_ALL,
43
44 /*
45 * Pretty much the same as FLUSH_ALL, but can also steal space from
46 * global rsv.
47 *
48 * Can be interrupted by a fatal signal.
49 */
50 BTRFS_RESERVE_FLUSH_ALL_STEAL,
51
52 /*
53 * This is for btrfs_use_block_rsv only. We have exhausted our block
54 * rsv and our global block rsv. This can happen for things like
55 * delalloc where we are overwriting a lot of extents with a single
56 * extent and didn't reserve enough space. Alternatively it can happen
57 * with delalloc where we reserve 1 extents worth for a large extent but
58 * fragmentation leads to multiple extents being created. This will
59 * give us the reservation in the case of
60 *
61 * if (num_bytes < (space_info->total_bytes -
62 * btrfs_space_info_used(space_info, false))
63 *
64 * Which ignores bytes_may_use. This is potentially dangerous, but our
65 * reservation system is generally pessimistic so is able to absorb this
66 * style of mistake.
67 */
68 BTRFS_RESERVE_FLUSH_EMERGENCY,
69};
70
71enum btrfs_flush_state {
72 FLUSH_DELAYED_ITEMS_NR = 1,
73 FLUSH_DELAYED_ITEMS = 2,
74 FLUSH_DELAYED_REFS_NR = 3,
75 FLUSH_DELAYED_REFS = 4,
76 FLUSH_DELALLOC = 5,
77 FLUSH_DELALLOC_WAIT = 6,
78 FLUSH_DELALLOC_FULL = 7,
79 ALLOC_CHUNK = 8,
80 ALLOC_CHUNK_FORCE = 9,
81 RUN_DELAYED_IPUTS = 10,
82 COMMIT_TRANS = 11,
83};
84
85struct btrfs_space_info {
86 spinlock_t lock;
87
88 u64 total_bytes; /* total bytes in the space,
89 this doesn't take mirrors into account */
90 u64 bytes_used; /* total bytes used,
91 this doesn't take mirrors into account */
92 u64 bytes_pinned; /* total bytes pinned, will be freed when the
93 transaction finishes */
94 u64 bytes_reserved; /* total bytes the allocator has reserved for
95 current allocations */
96 u64 bytes_may_use; /* number of bytes that may be used for
97 delalloc/allocations */
98 u64 bytes_readonly; /* total bytes that are read only */
99 /* Total bytes in the space, but only accounts active block groups. */
100 u64 active_total_bytes;
101 u64 bytes_zone_unusable; /* total bytes that are unusable until
102 resetting the device zone */
103
104 u64 max_extent_size; /* This will hold the maximum extent size of
105 the space info if we had an ENOSPC in the
106 allocator. */
107 /* Chunk size in bytes */
108 u64 chunk_size;
109
110 /*
111 * Once a block group drops below this threshold (percents) we'll
112 * schedule it for reclaim.
113 */
114 int bg_reclaim_threshold;
115
116 int clamp; /* Used to scale our threshold for preemptive
117 flushing. The value is >> clamp, so turns
118 out to be a 2^clamp divisor. */
119
120 unsigned int full:1; /* indicates that we cannot allocate any more
121 chunks for this space */
122 unsigned int chunk_alloc:1; /* set if we are allocating a chunk */
123
124 unsigned int flush:1; /* set if we are trying to make space */
125
126 unsigned int force_alloc; /* set if we need to force a chunk
127 alloc for this space */
128
129 u64 disk_used; /* total bytes used on disk */
130 u64 disk_total; /* total bytes on disk, takes mirrors into
131 account */
132
133 u64 flags;
134
135 struct list_head list;
136 /* Protected by the spinlock 'lock'. */
137 struct list_head ro_bgs;
138 struct list_head priority_tickets;
139 struct list_head tickets;
140
141 /*
142 * Size of space that needs to be reclaimed in order to satisfy pending
143 * tickets
144 */
145 u64 reclaim_size;
146
147 /*
148 * tickets_id just indicates the next ticket will be handled, so note
149 * it's not stored per ticket.
150 */
151 u64 tickets_id;
152
153 struct rw_semaphore groups_sem;
154 /* for block groups in our same type */
155 struct list_head block_groups[BTRFS_NR_RAID_TYPES];
156
157 struct kobject kobj;
158 struct kobject *block_group_kobjs[BTRFS_NR_RAID_TYPES];
159};
160
161struct reserve_ticket {
162 u64 bytes;
163 int error;
164 bool steal;
165 struct list_head list;
166 wait_queue_head_t wait;
167};
168
169static inline bool btrfs_mixed_space_info(struct btrfs_space_info *space_info)
170{
171 return ((space_info->flags & BTRFS_BLOCK_GROUP_METADATA) &&
172 (space_info->flags & BTRFS_BLOCK_GROUP_DATA));
173}
174
175/*
176 *
177 * Declare a helper function to detect underflow of various space info members
178 */
179#define DECLARE_SPACE_INFO_UPDATE(name, trace_name) \
180static inline void \
181btrfs_space_info_update_##name(struct btrfs_fs_info *fs_info, \
182 struct btrfs_space_info *sinfo, \
183 s64 bytes) \
184{ \
185 const u64 abs_bytes = (bytes < 0) ? -bytes : bytes; \
186 lockdep_assert_held(&sinfo->lock); \
187 trace_update_##name(fs_info, sinfo, sinfo->name, bytes); \
188 trace_btrfs_space_reservation(fs_info, trace_name, \
189 sinfo->flags, abs_bytes, \
190 bytes > 0); \
191 if (bytes < 0 && sinfo->name < -bytes) { \
192 WARN_ON(1); \
193 sinfo->name = 0; \
194 return; \
195 } \
196 sinfo->name += bytes; \
197}
198
199DECLARE_SPACE_INFO_UPDATE(bytes_may_use, "space_info");
200DECLARE_SPACE_INFO_UPDATE(bytes_pinned, "pinned");
201
202int btrfs_init_space_info(struct btrfs_fs_info *fs_info);
203void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info,
204 struct btrfs_block_group *block_group);
205void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info,
206 u64 chunk_size);
207struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
208 u64 flags);
209u64 __pure btrfs_space_info_used(struct btrfs_space_info *s_info,
210 bool may_use_included);
211void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
212void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
213 struct btrfs_space_info *info, u64 bytes,
214 int dump_block_groups);
215int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
216 struct btrfs_block_rsv *block_rsv,
217 u64 orig_bytes,
218 enum btrfs_reserve_flush_enum flush);
219void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
220 struct btrfs_space_info *space_info);
221int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
222 struct btrfs_space_info *space_info, u64 bytes,
223 enum btrfs_reserve_flush_enum flush);
224
225static inline void btrfs_space_info_free_bytes_may_use(
226 struct btrfs_fs_info *fs_info,
227 struct btrfs_space_info *space_info,
228 u64 num_bytes)
229{
230 spin_lock(&space_info->lock);
231 btrfs_space_info_update_bytes_may_use(fs_info, space_info, -num_bytes);
232 btrfs_try_granting_tickets(fs_info, space_info);
233 spin_unlock(&space_info->lock);
234}
235int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
236 enum btrfs_reserve_flush_enum flush);
237void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info);
238void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info);
239u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo);
240
241#endif /* BTRFS_SPACE_INFO_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2
3#ifndef BTRFS_SPACE_INFO_H
4#define BTRFS_SPACE_INFO_H
5
6#include <trace/events/btrfs.h>
7#include <linux/spinlock.h>
8#include <linux/list.h>
9#include <linux/kobject.h>
10#include <linux/lockdep.h>
11#include <linux/wait.h>
12#include <linux/rwsem.h>
13#include "volumes.h"
14
15struct btrfs_fs_info;
16struct btrfs_block_group;
17
18/*
19 * Different levels for to flush space when doing space reservations.
20 *
21 * The higher the level, the more methods we try to reclaim space.
22 */
23enum btrfs_reserve_flush_enum {
24 /* If we are in the transaction, we can't flush anything.*/
25 BTRFS_RESERVE_NO_FLUSH,
26
27 /*
28 * Flush space by:
29 * - Running delayed inode items
30 * - Allocating a new chunk
31 */
32 BTRFS_RESERVE_FLUSH_LIMIT,
33
34 /*
35 * Flush space by:
36 * - Running delayed inode items
37 * - Running delayed refs
38 * - Running delalloc and waiting for ordered extents
39 * - Allocating a new chunk
40 * - Committing transaction
41 */
42 BTRFS_RESERVE_FLUSH_EVICT,
43
44 /*
45 * Flush space by above mentioned methods and by:
46 * - Running delayed iputs
47 * - Committing transaction
48 *
49 * Can be interrupted by a fatal signal.
50 */
51 BTRFS_RESERVE_FLUSH_DATA,
52 BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE,
53 BTRFS_RESERVE_FLUSH_ALL,
54
55 /*
56 * Pretty much the same as FLUSH_ALL, but can also steal space from
57 * global rsv.
58 *
59 * Can be interrupted by a fatal signal.
60 */
61 BTRFS_RESERVE_FLUSH_ALL_STEAL,
62
63 /*
64 * This is for btrfs_use_block_rsv only. We have exhausted our block
65 * rsv and our global block rsv. This can happen for things like
66 * delalloc where we are overwriting a lot of extents with a single
67 * extent and didn't reserve enough space. Alternatively it can happen
68 * with delalloc where we reserve 1 extents worth for a large extent but
69 * fragmentation leads to multiple extents being created. This will
70 * give us the reservation in the case of
71 *
72 * if (num_bytes < (space_info->total_bytes -
73 * btrfs_space_info_used(space_info, false))
74 *
75 * Which ignores bytes_may_use. This is potentially dangerous, but our
76 * reservation system is generally pessimistic so is able to absorb this
77 * style of mistake.
78 */
79 BTRFS_RESERVE_FLUSH_EMERGENCY,
80};
81
82enum btrfs_flush_state {
83 FLUSH_DELAYED_ITEMS_NR = 1,
84 FLUSH_DELAYED_ITEMS = 2,
85 FLUSH_DELAYED_REFS_NR = 3,
86 FLUSH_DELAYED_REFS = 4,
87 FLUSH_DELALLOC = 5,
88 FLUSH_DELALLOC_WAIT = 6,
89 FLUSH_DELALLOC_FULL = 7,
90 ALLOC_CHUNK = 8,
91 ALLOC_CHUNK_FORCE = 9,
92 RUN_DELAYED_IPUTS = 10,
93 COMMIT_TRANS = 11,
94};
95
96struct btrfs_space_info {
97 struct btrfs_fs_info *fs_info;
98 spinlock_t lock;
99
100 u64 total_bytes; /* total bytes in the space,
101 this doesn't take mirrors into account */
102 u64 bytes_used; /* total bytes used,
103 this doesn't take mirrors into account */
104 u64 bytes_pinned; /* total bytes pinned, will be freed when the
105 transaction finishes */
106 u64 bytes_reserved; /* total bytes the allocator has reserved for
107 current allocations */
108 u64 bytes_may_use; /* number of bytes that may be used for
109 delalloc/allocations */
110 u64 bytes_readonly; /* total bytes that are read only */
111 u64 bytes_zone_unusable; /* total bytes that are unusable until
112 resetting the device zone */
113
114 u64 max_extent_size; /* This will hold the maximum extent size of
115 the space info if we had an ENOSPC in the
116 allocator. */
117 /* Chunk size in bytes */
118 u64 chunk_size;
119
120 /*
121 * Once a block group drops below this threshold (percents) we'll
122 * schedule it for reclaim.
123 */
124 int bg_reclaim_threshold;
125
126 int clamp; /* Used to scale our threshold for preemptive
127 flushing. The value is >> clamp, so turns
128 out to be a 2^clamp divisor. */
129
130 unsigned int full:1; /* indicates that we cannot allocate any more
131 chunks for this space */
132 unsigned int chunk_alloc:1; /* set if we are allocating a chunk */
133
134 unsigned int flush:1; /* set if we are trying to make space */
135
136 unsigned int force_alloc; /* set if we need to force a chunk
137 alloc for this space */
138
139 u64 disk_used; /* total bytes used on disk */
140 u64 disk_total; /* total bytes on disk, takes mirrors into
141 account */
142
143 u64 flags;
144
145 struct list_head list;
146 /* Protected by the spinlock 'lock'. */
147 struct list_head ro_bgs;
148 struct list_head priority_tickets;
149 struct list_head tickets;
150
151 /*
152 * Size of space that needs to be reclaimed in order to satisfy pending
153 * tickets
154 */
155 u64 reclaim_size;
156
157 /*
158 * tickets_id just indicates the next ticket will be handled, so note
159 * it's not stored per ticket.
160 */
161 u64 tickets_id;
162
163 struct rw_semaphore groups_sem;
164 /* for block groups in our same type */
165 struct list_head block_groups[BTRFS_NR_RAID_TYPES];
166
167 struct kobject kobj;
168 struct kobject *block_group_kobjs[BTRFS_NR_RAID_TYPES];
169
170 /*
171 * Monotonically increasing counter of block group reclaim attempts
172 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_count
173 */
174 u64 reclaim_count;
175
176 /*
177 * Monotonically increasing counter of reclaimed bytes
178 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_bytes
179 */
180 u64 reclaim_bytes;
181
182 /*
183 * Monotonically increasing counter of reclaim errors
184 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_errors
185 */
186 u64 reclaim_errors;
187
188 /*
189 * If true, use the dynamic relocation threshold, instead of the
190 * fixed bg_reclaim_threshold.
191 */
192 bool dynamic_reclaim;
193
194 /*
195 * Periodically check all block groups against the reclaim
196 * threshold in the cleaner thread.
197 */
198 bool periodic_reclaim;
199
200 /*
201 * Periodic reclaim should be a no-op if a space_info hasn't
202 * freed any space since the last time we tried.
203 */
204 bool periodic_reclaim_ready;
205
206 /*
207 * Net bytes freed or allocated since the last reclaim pass.
208 */
209 s64 reclaimable_bytes;
210};
211
212struct reserve_ticket {
213 u64 bytes;
214 int error;
215 bool steal;
216 struct list_head list;
217 wait_queue_head_t wait;
218};
219
220static inline bool btrfs_mixed_space_info(const struct btrfs_space_info *space_info)
221{
222 return ((space_info->flags & BTRFS_BLOCK_GROUP_METADATA) &&
223 (space_info->flags & BTRFS_BLOCK_GROUP_DATA));
224}
225
226/*
227 *
228 * Declare a helper function to detect underflow of various space info members
229 */
230#define DECLARE_SPACE_INFO_UPDATE(name, trace_name) \
231static inline void \
232btrfs_space_info_update_##name(struct btrfs_fs_info *fs_info, \
233 struct btrfs_space_info *sinfo, \
234 s64 bytes) \
235{ \
236 const u64 abs_bytes = (bytes < 0) ? -bytes : bytes; \
237 lockdep_assert_held(&sinfo->lock); \
238 trace_update_##name(fs_info, sinfo, sinfo->name, bytes); \
239 trace_btrfs_space_reservation(fs_info, trace_name, \
240 sinfo->flags, abs_bytes, \
241 bytes > 0); \
242 if (bytes < 0 && sinfo->name < -bytes) { \
243 WARN_ON(1); \
244 sinfo->name = 0; \
245 return; \
246 } \
247 sinfo->name += bytes; \
248}
249
250DECLARE_SPACE_INFO_UPDATE(bytes_may_use, "space_info");
251DECLARE_SPACE_INFO_UPDATE(bytes_pinned, "pinned");
252DECLARE_SPACE_INFO_UPDATE(bytes_zone_unusable, "zone_unusable");
253
254int btrfs_init_space_info(struct btrfs_fs_info *fs_info);
255void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info,
256 struct btrfs_block_group *block_group);
257void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info,
258 u64 chunk_size);
259struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
260 u64 flags);
261u64 __pure btrfs_space_info_used(const struct btrfs_space_info *s_info,
262 bool may_use_included);
263void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
264void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
265 struct btrfs_space_info *info, u64 bytes,
266 int dump_block_groups);
267int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
268 struct btrfs_space_info *space_info,
269 u64 orig_bytes,
270 enum btrfs_reserve_flush_enum flush);
271void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
272 struct btrfs_space_info *space_info);
273int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
274 const struct btrfs_space_info *space_info, u64 bytes,
275 enum btrfs_reserve_flush_enum flush);
276
277static inline void btrfs_space_info_free_bytes_may_use(
278 struct btrfs_fs_info *fs_info,
279 struct btrfs_space_info *space_info,
280 u64 num_bytes)
281{
282 spin_lock(&space_info->lock);
283 btrfs_space_info_update_bytes_may_use(fs_info, space_info, -num_bytes);
284 btrfs_try_granting_tickets(fs_info, space_info);
285 spin_unlock(&space_info->lock);
286}
287int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
288 enum btrfs_reserve_flush_enum flush);
289void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info);
290void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info);
291u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo);
292
293void btrfs_space_info_update_reclaimable(struct btrfs_space_info *space_info, s64 bytes);
294void btrfs_set_periodic_reclaim_ready(struct btrfs_space_info *space_info, bool ready);
295bool btrfs_should_periodic_reclaim(struct btrfs_space_info *space_info);
296int btrfs_calc_reclaim_threshold(const struct btrfs_space_info *space_info);
297void btrfs_reclaim_sweep(const struct btrfs_fs_info *fs_info);
298
299#endif /* BTRFS_SPACE_INFO_H */