xref: /openbmc/linux/fs/btrfs/space-info.h (revision 80d0624d)
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  */
13 enum 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 	 * - Committing transaction
31 	 */
32 	BTRFS_RESERVE_FLUSH_EVICT,
33 
34 	/*
35 	 * Flush space by above mentioned methods and by:
36 	 * - Running delayed iputs
37 	 * - Committing transaction
38 	 *
39 	 * Can be interrupted by a fatal signal.
40 	 */
41 	BTRFS_RESERVE_FLUSH_DATA,
42 	BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE,
43 	BTRFS_RESERVE_FLUSH_ALL,
44 
45 	/*
46 	 * Pretty much the same as FLUSH_ALL, but can also steal space from
47 	 * global rsv.
48 	 *
49 	 * Can be interrupted by a fatal signal.
50 	 */
51 	BTRFS_RESERVE_FLUSH_ALL_STEAL,
52 
53 	/*
54 	 * This is for btrfs_use_block_rsv only.  We have exhausted our block
55 	 * rsv and our global block rsv.  This can happen for things like
56 	 * delalloc where we are overwriting a lot of extents with a single
57 	 * extent and didn't reserve enough space.  Alternatively it can happen
58 	 * with delalloc where we reserve 1 extents worth for a large extent but
59 	 * fragmentation leads to multiple extents being created.  This will
60 	 * give us the reservation in the case of
61 	 *
62 	 * if (num_bytes < (space_info->total_bytes -
63 	 *		    btrfs_space_info_used(space_info, false))
64 	 *
65 	 * Which ignores bytes_may_use.  This is potentially dangerous, but our
66 	 * reservation system is generally pessimistic so is able to absorb this
67 	 * style of mistake.
68 	 */
69 	BTRFS_RESERVE_FLUSH_EMERGENCY,
70 };
71 
72 enum btrfs_flush_state {
73 	FLUSH_DELAYED_ITEMS_NR	= 1,
74 	FLUSH_DELAYED_ITEMS	= 2,
75 	FLUSH_DELAYED_REFS_NR	= 3,
76 	FLUSH_DELAYED_REFS	= 4,
77 	FLUSH_DELALLOC		= 5,
78 	FLUSH_DELALLOC_WAIT	= 6,
79 	FLUSH_DELALLOC_FULL	= 7,
80 	ALLOC_CHUNK		= 8,
81 	ALLOC_CHUNK_FORCE	= 9,
82 	RUN_DELAYED_IPUTS	= 10,
83 	COMMIT_TRANS		= 11,
84 };
85 
86 struct btrfs_space_info {
87 	spinlock_t lock;
88 
89 	u64 total_bytes;	/* total bytes in the space,
90 				   this doesn't take mirrors into account */
91 	u64 bytes_used;		/* total bytes used,
92 				   this doesn't take mirrors into account */
93 	u64 bytes_pinned;	/* total bytes pinned, will be freed when the
94 				   transaction finishes */
95 	u64 bytes_reserved;	/* total bytes the allocator has reserved for
96 				   current allocations */
97 	u64 bytes_may_use;	/* number of bytes that may be used for
98 				   delalloc/allocations */
99 	u64 bytes_readonly;	/* total bytes that are read only */
100 	u64 bytes_zone_unusable;	/* total bytes that are unusable until
101 					   resetting the device zone */
102 
103 	u64 max_extent_size;	/* This will hold the maximum extent size of
104 				   the space info if we had an ENOSPC in the
105 				   allocator. */
106 	/* Chunk size in bytes */
107 	u64 chunk_size;
108 
109 	/*
110 	 * Once a block group drops below this threshold (percents) we'll
111 	 * schedule it for reclaim.
112 	 */
113 	int bg_reclaim_threshold;
114 
115 	int clamp;		/* Used to scale our threshold for preemptive
116 				   flushing. The value is >> clamp, so turns
117 				   out to be a 2^clamp divisor. */
118 
119 	unsigned int full:1;	/* indicates that we cannot allocate any more
120 				   chunks for this space */
121 	unsigned int chunk_alloc:1;	/* set if we are allocating a chunk */
122 
123 	unsigned int flush:1;		/* set if we are trying to make space */
124 
125 	unsigned int force_alloc;	/* set if we need to force a chunk
126 					   alloc for this space */
127 
128 	u64 disk_used;		/* total bytes used on disk */
129 	u64 disk_total;		/* total bytes on disk, takes mirrors into
130 				   account */
131 
132 	u64 flags;
133 
134 	struct list_head list;
135 	/* Protected by the spinlock 'lock'. */
136 	struct list_head ro_bgs;
137 	struct list_head priority_tickets;
138 	struct list_head tickets;
139 
140 	/*
141 	 * Size of space that needs to be reclaimed in order to satisfy pending
142 	 * tickets
143 	 */
144 	u64 reclaim_size;
145 
146 	/*
147 	 * tickets_id just indicates the next ticket will be handled, so note
148 	 * it's not stored per ticket.
149 	 */
150 	u64 tickets_id;
151 
152 	struct rw_semaphore groups_sem;
153 	/* for block groups in our same type */
154 	struct list_head block_groups[BTRFS_NR_RAID_TYPES];
155 
156 	struct kobject kobj;
157 	struct kobject *block_group_kobjs[BTRFS_NR_RAID_TYPES];
158 };
159 
160 struct reserve_ticket {
161 	u64 bytes;
162 	int error;
163 	bool steal;
164 	struct list_head list;
165 	wait_queue_head_t wait;
166 };
167 
168 static inline bool btrfs_mixed_space_info(struct btrfs_space_info *space_info)
169 {
170 	return ((space_info->flags & BTRFS_BLOCK_GROUP_METADATA) &&
171 		(space_info->flags & BTRFS_BLOCK_GROUP_DATA));
172 }
173 
174 /*
175  *
176  * Declare a helper function to detect underflow of various space info members
177  */
178 #define DECLARE_SPACE_INFO_UPDATE(name, trace_name)			\
179 static inline void							\
180 btrfs_space_info_update_##name(struct btrfs_fs_info *fs_info,		\
181 			       struct btrfs_space_info *sinfo,		\
182 			       s64 bytes)				\
183 {									\
184 	const u64 abs_bytes = (bytes < 0) ? -bytes : bytes;		\
185 	lockdep_assert_held(&sinfo->lock);				\
186 	trace_update_##name(fs_info, sinfo, sinfo->name, bytes);	\
187 	trace_btrfs_space_reservation(fs_info, trace_name,		\
188 				      sinfo->flags, abs_bytes,		\
189 				      bytes > 0);			\
190 	if (bytes < 0 && sinfo->name < -bytes) {			\
191 		WARN_ON(1);						\
192 		sinfo->name = 0;					\
193 		return;							\
194 	}								\
195 	sinfo->name += bytes;						\
196 }
197 
198 DECLARE_SPACE_INFO_UPDATE(bytes_may_use, "space_info");
199 DECLARE_SPACE_INFO_UPDATE(bytes_pinned, "pinned");
200 DECLARE_SPACE_INFO_UPDATE(bytes_zone_unusable, "zone_unusable");
201 
202 int btrfs_init_space_info(struct btrfs_fs_info *fs_info);
203 void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info,
204 				struct btrfs_block_group *block_group);
205 void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info,
206 					u64 chunk_size);
207 struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
208 					       u64 flags);
209 u64 __pure btrfs_space_info_used(struct btrfs_space_info *s_info,
210 			  bool may_use_included);
211 void btrfs_clear_space_info_full(struct btrfs_fs_info *info);
212 void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
213 			   struct btrfs_space_info *info, u64 bytes,
214 			   int dump_block_groups);
215 int 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);
219 void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
220 				struct btrfs_space_info *space_info);
221 int 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 
225 static 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 }
235 int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
236 			     enum btrfs_reserve_flush_enum flush);
237 void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info);
238 void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info);
239 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo);
240 
241 #endif /* BTRFS_SPACE_INFO_H */
242