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 enum btrfs_space_info_sub_group { 102 BTRFS_SUB_GROUP_PRIMARY, 103 BTRFS_SUB_GROUP_DATA_RELOC, 104 BTRFS_SUB_GROUP_TREELOG, 105 }; 106 107 #define BTRFS_SPACE_INFO_SUB_GROUP_MAX 1 108 struct btrfs_space_info { 109 struct btrfs_fs_info *fs_info; 110 struct btrfs_space_info *parent; 111 struct btrfs_space_info *sub_group[BTRFS_SPACE_INFO_SUB_GROUP_MAX]; 112 int subgroup_id; 113 spinlock_t lock; 114 115 u64 total_bytes; /* total bytes in the space, 116 this doesn't take mirrors into account */ 117 u64 bytes_used; /* total bytes used, 118 this doesn't take mirrors into account */ 119 u64 bytes_pinned; /* total bytes pinned, will be freed when the 120 transaction finishes */ 121 u64 bytes_reserved; /* total bytes the allocator has reserved for 122 current allocations */ 123 u64 bytes_may_use; /* number of bytes that may be used for 124 delalloc/allocations */ 125 u64 bytes_readonly; /* total bytes that are read only */ 126 u64 bytes_zone_unusable; /* total bytes that are unusable until 127 resetting the device zone */ 128 129 u64 max_extent_size; /* This will hold the maximum extent size of 130 the space info if we had an ENOSPC in the 131 allocator. */ 132 /* Chunk size in bytes */ 133 u64 chunk_size; 134 135 /* 136 * Once a block group drops below this threshold (percents) we'll 137 * schedule it for reclaim. 138 */ 139 int bg_reclaim_threshold; 140 141 int clamp; /* Used to scale our threshold for preemptive 142 flushing. The value is >> clamp, so turns 143 out to be a 2^clamp divisor. */ 144 145 unsigned int full:1; /* indicates that we cannot allocate any more 146 chunks for this space */ 147 unsigned int chunk_alloc:1; /* set if we are allocating a chunk */ 148 149 unsigned int flush:1; /* set if we are trying to make space */ 150 151 unsigned int force_alloc; /* set if we need to force a chunk 152 alloc for this space */ 153 154 u64 disk_used; /* total bytes used on disk */ 155 u64 disk_total; /* total bytes on disk, takes mirrors into 156 account */ 157 158 u64 flags; 159 160 struct list_head list; 161 /* Protected by the spinlock 'lock'. */ 162 struct list_head ro_bgs; 163 struct list_head priority_tickets; 164 struct list_head tickets; 165 166 /* 167 * Size of space that needs to be reclaimed in order to satisfy pending 168 * tickets 169 */ 170 u64 reclaim_size; 171 172 /* 173 * tickets_id just indicates the next ticket will be handled, so note 174 * it's not stored per ticket. 175 */ 176 u64 tickets_id; 177 178 struct rw_semaphore groups_sem; 179 /* for block groups in our same type */ 180 struct list_head block_groups[BTRFS_NR_RAID_TYPES]; 181 182 struct kobject kobj; 183 struct kobject *block_group_kobjs[BTRFS_NR_RAID_TYPES]; 184 185 /* 186 * Monotonically increasing counter of block group reclaim attempts 187 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_count 188 */ 189 u64 reclaim_count; 190 191 /* 192 * Monotonically increasing counter of reclaimed bytes 193 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_bytes 194 */ 195 u64 reclaim_bytes; 196 197 /* 198 * Monotonically increasing counter of reclaim errors 199 * Exposed in /sys/fs/<uuid>/allocation/<type>/reclaim_errors 200 */ 201 u64 reclaim_errors; 202 203 /* 204 * If true, use the dynamic relocation threshold, instead of the 205 * fixed bg_reclaim_threshold. 206 */ 207 bool dynamic_reclaim; 208 209 /* 210 * Periodically check all block groups against the reclaim 211 * threshold in the cleaner thread. 212 */ 213 bool periodic_reclaim; 214 215 /* 216 * Periodic reclaim should be a no-op if a space_info hasn't 217 * freed any space since the last time we tried. 218 */ 219 bool periodic_reclaim_ready; 220 221 /* 222 * Net bytes freed or allocated since the last reclaim pass. 223 */ 224 s64 reclaimable_bytes; 225 }; 226 227 struct reserve_ticket { 228 u64 bytes; 229 int error; 230 bool steal; 231 struct list_head list; 232 wait_queue_head_t wait; 233 }; 234 235 static inline bool btrfs_mixed_space_info(const struct btrfs_space_info *space_info) 236 { 237 return ((space_info->flags & BTRFS_BLOCK_GROUP_METADATA) && 238 (space_info->flags & BTRFS_BLOCK_GROUP_DATA)); 239 } 240 241 /* 242 * 243 * Declare a helper function to detect underflow of various space info members 244 */ 245 #define DECLARE_SPACE_INFO_UPDATE(name, trace_name) \ 246 static inline void \ 247 btrfs_space_info_update_##name(struct btrfs_space_info *sinfo, \ 248 s64 bytes) \ 249 { \ 250 struct btrfs_fs_info *fs_info = sinfo->fs_info; \ 251 const u64 abs_bytes = (bytes < 0) ? -bytes : bytes; \ 252 lockdep_assert_held(&sinfo->lock); \ 253 trace_update_##name(fs_info, sinfo, sinfo->name, bytes); \ 254 trace_btrfs_space_reservation(fs_info, trace_name, \ 255 sinfo->flags, abs_bytes, \ 256 bytes > 0); \ 257 if (bytes < 0 && sinfo->name < -bytes) { \ 258 WARN_ON(1); \ 259 sinfo->name = 0; \ 260 return; \ 261 } \ 262 sinfo->name += bytes; \ 263 } 264 265 DECLARE_SPACE_INFO_UPDATE(bytes_may_use, "space_info"); 266 DECLARE_SPACE_INFO_UPDATE(bytes_pinned, "pinned"); 267 DECLARE_SPACE_INFO_UPDATE(bytes_zone_unusable, "zone_unusable"); 268 269 int btrfs_init_space_info(struct btrfs_fs_info *fs_info); 270 void btrfs_add_bg_to_space_info(struct btrfs_fs_info *info, 271 struct btrfs_block_group *block_group); 272 void btrfs_update_space_info_chunk_size(struct btrfs_space_info *space_info, 273 u64 chunk_size); 274 struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info, 275 u64 flags); 276 u64 __pure btrfs_space_info_used(const struct btrfs_space_info *s_info, 277 bool may_use_included); 278 void btrfs_clear_space_info_full(struct btrfs_fs_info *info); 279 void btrfs_dump_space_info(struct btrfs_fs_info *fs_info, 280 struct btrfs_space_info *info, u64 bytes, 281 bool dump_block_groups); 282 int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info, 283 struct btrfs_space_info *space_info, 284 u64 orig_bytes, 285 enum btrfs_reserve_flush_enum flush); 286 void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info, 287 struct btrfs_space_info *space_info); 288 int btrfs_can_overcommit(struct btrfs_fs_info *fs_info, 289 const struct btrfs_space_info *space_info, u64 bytes, 290 enum btrfs_reserve_flush_enum flush); 291 292 static inline void btrfs_space_info_free_bytes_may_use( 293 struct btrfs_space_info *space_info, 294 u64 num_bytes) 295 { 296 spin_lock(&space_info->lock); 297 btrfs_space_info_update_bytes_may_use(space_info, -num_bytes); 298 btrfs_try_granting_tickets(space_info->fs_info, space_info); 299 spin_unlock(&space_info->lock); 300 } 301 int btrfs_reserve_data_bytes(struct btrfs_space_info *space_info, u64 bytes, 302 enum btrfs_reserve_flush_enum flush); 303 void btrfs_dump_space_info_for_trans_abort(struct btrfs_fs_info *fs_info); 304 void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info); 305 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo); 306 307 void btrfs_space_info_update_reclaimable(struct btrfs_space_info *space_info, s64 bytes); 308 void btrfs_set_periodic_reclaim_ready(struct btrfs_space_info *space_info, bool ready); 309 int btrfs_calc_reclaim_threshold(const struct btrfs_space_info *space_info); 310 void btrfs_reclaim_sweep(const struct btrfs_fs_info *fs_info); 311 void btrfs_return_free_space(struct btrfs_space_info *space_info, u64 len); 312 313 #endif /* BTRFS_SPACE_INFO_H */ 314