xref: /linux/fs/btrfs/space-info.h (revision 1cbfb828e05171ca2dd77b5988d068e6872480fe)
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 
15 struct btrfs_fs_info;
16 struct 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  */
23 enum 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 
82 /*
83  * Please be aware that the order of enum values will be the order of the reclaim
84  * process in btrfs_async_reclaim_metadata_space().
85  */
86 enum btrfs_flush_state {
87 	FLUSH_DELAYED_ITEMS_NR	= 1,
88 	FLUSH_DELAYED_ITEMS	= 2,
89 	FLUSH_DELAYED_REFS_NR	= 3,
90 	FLUSH_DELAYED_REFS	= 4,
91 	FLUSH_DELALLOC		= 5,
92 	FLUSH_DELALLOC_WAIT	= 6,
93 	FLUSH_DELALLOC_FULL	= 7,
94 	ALLOC_CHUNK		= 8,
95 	ALLOC_CHUNK_FORCE	= 9,
96 	RUN_DELAYED_IPUTS	= 10,
97 	COMMIT_TRANS		= 11,
98 	RESET_ZONES		= 12,
99 };
100 
101 struct btrfs_space_info {
102 	struct btrfs_fs_info *fs_info;
103 	spinlock_t lock;
104 
105 	u64 total_bytes;	/* total bytes in the space,
106 				   this doesn't take mirrors into account */
107 	u64 bytes_used;		/* total bytes used,
108 				   this doesn't take mirrors into account */
109 	u64 bytes_pinned;	/* total bytes pinned, will be freed when the
110 				   transaction finishes */
111 	u64 bytes_reserved;	/* total bytes the allocator has reserved for
112 				   current allocations */
113 	u64 bytes_may_use;	/* number of bytes that may be used for
114 				   delalloc/allocations */
115 	u64 bytes_readonly;	/* total bytes that are read only */
116 	u64 bytes_zone_unusable;	/* total bytes that are unusable until
117 					   resetting the device zone */
118 
119 	u64 max_extent_size;	/* This will hold the maximum extent size of
120 				   the space info if we had an ENOSPC in the
121 				   allocator. */
122 	/* Chunk size in bytes */
123 	u64 chunk_size;
124 
125 	/*
126 	 * Once a block group drops below this threshold (percents) we'll
127 	 * schedule it for reclaim.
128 	 */
129 	int bg_reclaim_threshold;
130 
131 	int clamp;		/* Used to scale our threshold for preemptive
132 				   flushing. The value is >> clamp, so turns
133 				   out to be a 2^clamp divisor. */
134 
135 	unsigned int full:1;	/* indicates that we cannot allocate any more
136 				   chunks for this space */
137 	unsigned int chunk_alloc:1;	/* set if we are allocating a chunk */
138 
139 	unsigned int flush:1;		/* set if we are trying to make space */
140 
141 	unsigned int force_alloc;	/* set if we need to force a chunk
142 					   alloc for this space */
143 
144 	u64 disk_used;		/* total bytes used on disk */
145 	u64 disk_total;		/* total bytes on disk, takes mirrors into
146 				   account */
147 
148 	u64 flags;
149 
150 	struct list_head list;
151 	/* Protected by the spinlock 'lock'. */
152 	struct list_head ro_bgs;
153 	struct list_head priority_tickets;
154 	struct list_head tickets;
155 
156 	/*
157 	 * Size of space that needs to be reclaimed in order to satisfy pending
158 	 * tickets
159 	 */
160 	u64 reclaim_size;
161 
162 	/*
163 	 * tickets_id just indicates the next ticket will be handled, so note
164 	 * it's not stored per ticket.
165 	 */
166 	u64 tickets_id;
167 
168 	struct rw_semaphore groups_sem;
169 	/* for block groups in our same type */
170 	struct list_head block_groups[BTRFS_NR_RAID_TYPES];
171 
172 	struct kobject kobj;
173 	struct kobject *block_group_kobjs[BTRFS_NR_RAID_TYPES];
174 
175 	/*
176 	 * Monotonically increasing counter of block group reclaim attempts
177 	 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_count
178 	 */
179 	u64 reclaim_count;
180 
181 	/*
182 	 * Monotonically increasing counter of reclaimed bytes
183 	 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_bytes
184 	 */
185 	u64 reclaim_bytes;
186 
187 	/*
188 	 * Monotonically increasing counter of reclaim errors
189 	 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_errors
190 	 */
191 	u64 reclaim_errors;
192 
193 	/*
194 	 * If true, use the dynamic relocation threshold, instead of the
195 	 * fixed bg_reclaim_threshold.
196 	 */
197 	bool dynamic_reclaim;
198 
199 	/*
200 	 * Periodically check all block groups against the reclaim
201 	 * threshold in the cleaner thread.
202 	 */
203 	bool periodic_reclaim;
204 
205 	/*
206 	 * Periodic reclaim should be a no-op if a space_info hasn't
207 	 * freed any space since the last time we tried.
208 	 */
209 	bool periodic_reclaim_ready;
210 
211 	/*
212 	 * Net bytes freed or allocated since the last reclaim pass.
213 	 */
214 	s64 reclaimable_bytes;
215 };
216 
217 struct reserve_ticket {
218 	u64 bytes;
219 	int error;
220 	bool steal;
221 	struct list_head list;
222 	wait_queue_head_t wait;
223 };
224 
225 static inline bool btrfs_mixed_space_info(const struct btrfs_space_info *space_info)
226 {
227 	return ((space_info->flags & BTRFS_BLOCK_GROUP_METADATA) &&
228 		(space_info->flags & BTRFS_BLOCK_GROUP_DATA));
229 }
230 
231 /*
232  *
233  * Declare a helper function to detect underflow of various space info members
234  */
235 #define DECLARE_SPACE_INFO_UPDATE(name, trace_name)			\
236 static inline void							\
237 btrfs_space_info_update_##name(struct btrfs_space_info *sinfo,		\
238 			       s64 bytes)				\
239 {									\
240 	struct btrfs_fs_info *fs_info = sinfo->fs_info;			\
241 	const u64 abs_bytes = (bytes < 0) ? -bytes : bytes;		\
242 	lockdep_assert_held(&sinfo->lock);				\
243 	trace_update_##name(fs_info, sinfo, sinfo->name, bytes);	\
244 	trace_btrfs_space_reservation(fs_info, trace_name,		\
245 				      sinfo->flags, abs_bytes,		\
246 				      bytes > 0);			\
247 	if (bytes < 0 && sinfo->name < -bytes) {			\
248 		WARN_ON(1);						\
249 		sinfo->name = 0;					\
250 		return;							\
251 	}								\
252 	sinfo->name += bytes;						\
253 }
254 
255 DECLARE_SPACE_INFO_UPDATE(bytes_may_use, "space_info");
256 DECLARE_SPACE_INFO_UPDATE(bytes_pinned, "pinned");
257 DECLARE_SPACE_INFO_UPDATE(bytes_zone_unusable, "zone_unusable");
258 
259 int btrfs_init_space_info(struct btrfs_fs_info *fs_info);
260 void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info,
261 				struct btrfs_block_group *block_group);
262 void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info,
263 					u64 chunk_size);
264 struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
265 					       u64 flags);
266 u64 __pure btrfs_space_info_used(const struct btrfs_space_info *s_info,
267 			  bool may_use_included);
268 void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
269 void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
270 			   struct btrfs_space_info *info, u64 bytes,
271 			   int dump_block_groups);
272 int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
273 				 struct btrfs_space_info *space_info,
274 				 u64 orig_bytes,
275 				 enum btrfs_reserve_flush_enum flush);
276 void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
277 				struct btrfs_space_info *space_info);
278 int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
279 			 const struct btrfs_space_info *space_info, u64 bytes,
280 			 enum btrfs_reserve_flush_enum flush);
281 
282 static inline void btrfs_space_info_free_bytes_may_use(
283 				struct btrfs_space_info *space_info,
284 				u64 num_bytes)
285 {
286 	spin_lock(&space_info->lock);
287 	btrfs_space_info_update_bytes_may_use(space_info, -num_bytes);
288 	btrfs_try_granting_tickets(space_info->fs_info, space_info);
289 	spin_unlock(&space_info->lock);
290 }
291 int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
292 			     enum btrfs_reserve_flush_enum flush);
293 void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info);
294 void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info);
295 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo);
296 
297 void btrfs_space_info_update_reclaimable(struct btrfs_space_info *space_info, s64 bytes);
298 void btrfs_set_periodic_reclaim_ready(struct btrfs_space_info *space_info, bool ready);
299 bool btrfs_should_periodic_reclaim(struct btrfs_space_info *space_info);
300 int btrfs_calc_reclaim_threshold(const struct btrfs_space_info *space_info);
301 void btrfs_reclaim_sweep(const struct btrfs_fs_info *fs_info);
302 void btrfs_return_free_space(struct btrfs_space_info *space_info, u64 len);
303 
304 #endif /* BTRFS_SPACE_INFO_H */
305