Loading...
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 "volumes.h"
8
9/*
10 * Different levels for to flush space when doing space reservations.
11 *
12 * The higher the level, the more methods we try to reclaim space.
13 */
14enum btrfs_reserve_flush_enum {
15 /* If we are in the transaction, we can't flush anything.*/
16 BTRFS_RESERVE_NO_FLUSH,
17
18 /*
19 * Flush space by:
20 * - Running delayed inode items
21 * - Allocating a new chunk
22 */
23 BTRFS_RESERVE_FLUSH_LIMIT,
24
25 /*
26 * Flush space by:
27 * - Running delayed inode items
28 * - Running delayed refs
29 * - Running delalloc and waiting for ordered extents
30 * - Allocating a new chunk
31 * - Committing transaction
32 */
33 BTRFS_RESERVE_FLUSH_EVICT,
34
35 /*
36 * Flush space by above mentioned methods and by:
37 * - Running delayed iputs
38 * - Committing transaction
39 *
40 * Can be interrupted by a fatal signal.
41 */
42 BTRFS_RESERVE_FLUSH_DATA,
43 BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE,
44 BTRFS_RESERVE_FLUSH_ALL,
45
46 /*
47 * Pretty much the same as FLUSH_ALL, but can also steal space from
48 * global rsv.
49 *
50 * Can be interrupted by a fatal signal.
51 */
52 BTRFS_RESERVE_FLUSH_ALL_STEAL,
53
54 /*
55 * This is for btrfs_use_block_rsv only. We have exhausted our block
56 * rsv and our global block rsv. This can happen for things like
57 * delalloc where we are overwriting a lot of extents with a single
58 * extent and didn't reserve enough space. Alternatively it can happen
59 * with delalloc where we reserve 1 extents worth for a large extent but
60 * fragmentation leads to multiple extents being created. This will
61 * give us the reservation in the case of
62 *
63 * if (num_bytes < (space_info->total_bytes -
64 * btrfs_space_info_used(space_info, false))
65 *
66 * Which ignores bytes_may_use. This is potentially dangerous, but our
67 * reservation system is generally pessimistic so is able to absorb this
68 * style of mistake.
69 */
70 BTRFS_RESERVE_FLUSH_EMERGENCY,
71};
72
73enum btrfs_flush_state {
74 FLUSH_DELAYED_ITEMS_NR = 1,
75 FLUSH_DELAYED_ITEMS = 2,
76 FLUSH_DELAYED_REFS_NR = 3,
77 FLUSH_DELAYED_REFS = 4,
78 FLUSH_DELALLOC = 5,
79 FLUSH_DELALLOC_WAIT = 6,
80 FLUSH_DELALLOC_FULL = 7,
81 ALLOC_CHUNK = 8,
82 ALLOC_CHUNK_FORCE = 9,
83 RUN_DELAYED_IPUTS = 10,
84 COMMIT_TRANS = 11,
85};
86
87struct btrfs_space_info {
88 spinlock_t lock;
89
90 u64 total_bytes; /* total bytes in the space,
91 this doesn't take mirrors into account */
92 u64 bytes_used; /* total bytes used,
93 this doesn't take mirrors into account */
94 u64 bytes_pinned; /* total bytes pinned, will be freed when the
95 transaction finishes */
96 u64 bytes_reserved; /* total bytes the allocator has reserved for
97 current allocations */
98 u64 bytes_may_use; /* number of bytes that may be used for
99 delalloc/allocations */
100 u64 bytes_readonly; /* total bytes that are read only */
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_space_info *space_info,
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 spinlock_t lock;
98
99 u64 total_bytes; /* total bytes in the space,
100 this doesn't take mirrors into account */
101 u64 bytes_used; /* total bytes used,
102 this doesn't take mirrors into account */
103 u64 bytes_pinned; /* total bytes pinned, will be freed when the
104 transaction finishes */
105 u64 bytes_reserved; /* total bytes the allocator has reserved for
106 current allocations */
107 u64 bytes_may_use; /* number of bytes that may be used for
108 delalloc/allocations */
109 u64 bytes_readonly; /* total bytes that are read only */
110 u64 bytes_zone_unusable; /* total bytes that are unusable until
111 resetting the device zone */
112
113 u64 max_extent_size; /* This will hold the maximum extent size of
114 the space info if we had an ENOSPC in the
115 allocator. */
116 /* Chunk size in bytes */
117 u64 chunk_size;
118
119 /*
120 * Once a block group drops below this threshold (percents) we'll
121 * schedule it for reclaim.
122 */
123 int bg_reclaim_threshold;
124
125 int clamp; /* Used to scale our threshold for preemptive
126 flushing. The value is >> clamp, so turns
127 out to be a 2^clamp divisor. */
128
129 unsigned int full:1; /* indicates that we cannot allocate any more
130 chunks for this space */
131 unsigned int chunk_alloc:1; /* set if we are allocating a chunk */
132
133 unsigned int flush:1; /* set if we are trying to make space */
134
135 unsigned int force_alloc; /* set if we need to force a chunk
136 alloc for this space */
137
138 u64 disk_used; /* total bytes used on disk */
139 u64 disk_total; /* total bytes on disk, takes mirrors into
140 account */
141
142 u64 flags;
143
144 struct list_head list;
145 /* Protected by the spinlock 'lock'. */
146 struct list_head ro_bgs;
147 struct list_head priority_tickets;
148 struct list_head tickets;
149
150 /*
151 * Size of space that needs to be reclaimed in order to satisfy pending
152 * tickets
153 */
154 u64 reclaim_size;
155
156 /*
157 * tickets_id just indicates the next ticket will be handled, so note
158 * it's not stored per ticket.
159 */
160 u64 tickets_id;
161
162 struct rw_semaphore groups_sem;
163 /* for block groups in our same type */
164 struct list_head block_groups[BTRFS_NR_RAID_TYPES];
165
166 struct kobject kobj;
167 struct kobject *block_group_kobjs[BTRFS_NR_RAID_TYPES];
168};
169
170struct reserve_ticket {
171 u64 bytes;
172 int error;
173 bool steal;
174 struct list_head list;
175 wait_queue_head_t wait;
176};
177
178static inline bool btrfs_mixed_space_info(struct btrfs_space_info *space_info)
179{
180 return ((space_info->flags & BTRFS_BLOCK_GROUP_METADATA) &&
181 (space_info->flags & BTRFS_BLOCK_GROUP_DATA));
182}
183
184/*
185 *
186 * Declare a helper function to detect underflow of various space info members
187 */
188#define DECLARE_SPACE_INFO_UPDATE(name, trace_name) \
189static inline void \
190btrfs_space_info_update_##name(struct btrfs_fs_info *fs_info, \
191 struct btrfs_space_info *sinfo, \
192 s64 bytes) \
193{ \
194 const u64 abs_bytes = (bytes < 0) ? -bytes : bytes; \
195 lockdep_assert_held(&sinfo->lock); \
196 trace_update_##name(fs_info, sinfo, sinfo->name, bytes); \
197 trace_btrfs_space_reservation(fs_info, trace_name, \
198 sinfo->flags, abs_bytes, \
199 bytes > 0); \
200 if (bytes < 0 && sinfo->name < -bytes) { \
201 WARN_ON(1); \
202 sinfo->name = 0; \
203 return; \
204 } \
205 sinfo->name += bytes; \
206}
207
208DECLARE_SPACE_INFO_UPDATE(bytes_may_use, "space_info");
209DECLARE_SPACE_INFO_UPDATE(bytes_pinned, "pinned");
210
211int btrfs_init_space_info(struct btrfs_fs_info *fs_info);
212void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info,
213 struct btrfs_block_group *block_group);
214void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info,
215 u64 chunk_size);
216struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
217 u64 flags);
218u64 __pure btrfs_space_info_used(struct btrfs_space_info *s_info,
219 bool may_use_included);
220void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
221void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
222 struct btrfs_space_info *info, u64 bytes,
223 int dump_block_groups);
224int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
225 struct btrfs_space_info *space_info,
226 u64 orig_bytes,
227 enum btrfs_reserve_flush_enum flush);
228void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
229 struct btrfs_space_info *space_info);
230int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
231 struct btrfs_space_info *space_info, u64 bytes,
232 enum btrfs_reserve_flush_enum flush);
233
234static inline void btrfs_space_info_free_bytes_may_use(
235 struct btrfs_fs_info *fs_info,
236 struct btrfs_space_info *space_info,
237 u64 num_bytes)
238{
239 spin_lock(&space_info->lock);
240 btrfs_space_info_update_bytes_may_use(fs_info, space_info, -num_bytes);
241 btrfs_try_granting_tickets(fs_info, space_info);
242 spin_unlock(&space_info->lock);
243}
244int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
245 enum btrfs_reserve_flush_enum flush);
246void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info);
247void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info);
248u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo);
249
250#endif /* BTRFS_SPACE_INFO_H */