1 /* SPDX-License-Identifier: GPL-2.0 */
2
3 #ifndef BTRFS_BLOCK_GROUP_H
4 #define BTRFS_BLOCK_GROUP_H
5
6 #include <linux/atomic.h>
7 #include <linux/mutex.h>
8 #include <linux/list.h>
9 #include <linux/spinlock.h>
10 #include <linux/refcount.h>
11 #include <linux/wait.h>
12 #include <linux/sizes.h>
13 #include <linux/rwsem.h>
14 #include <linux/rbtree.h>
15 #include <uapi/linux/btrfs_tree.h>
16 #include "free-space-cache.h"
17
18 struct btrfs_chunk_map;
19 struct btrfs_fs_info;
20 struct btrfs_inode;
21 struct btrfs_trans_handle;
22
23 enum btrfs_disk_cache_state {
24 BTRFS_DC_WRITTEN,
25 BTRFS_DC_ERROR,
26 BTRFS_DC_CLEAR,
27 BTRFS_DC_SETUP,
28 };
29
30 enum btrfs_block_group_size_class {
31 /* Unset */
32 BTRFS_BG_SZ_NONE,
33 /* 0 < size <= 128K */
34 BTRFS_BG_SZ_SMALL,
35 /* 128K < size <= 8M */
36 BTRFS_BG_SZ_MEDIUM,
37 /* 8M < size < BG_LENGTH */
38 BTRFS_BG_SZ_LARGE,
39 };
40
41 /*
42 * This describes the state of the block_group for async discard. This is due
43 * to the two pass nature of it where extent discarding is prioritized over
44 * bitmap discarding. BTRFS_DISCARD_RESET_CURSOR is set when we are resetting
45 * between lists to prevent contention for discard state variables
46 * (eg. discard_cursor).
47 */
48 enum btrfs_discard_state {
49 BTRFS_DISCARD_EXTENTS,
50 BTRFS_DISCARD_BITMAPS,
51 BTRFS_DISCARD_RESET_CURSOR,
52 BTRFS_DISCARD_FULLY_REMAPPED,
53 };
54
55 /*
56 * Control flags for do_chunk_alloc's force field CHUNK_ALLOC_NO_FORCE means to
57 * only allocate a chunk if we really need one.
58 *
59 * CHUNK_ALLOC_LIMITED means to only try and allocate one if we have very few
60 * chunks already allocated. This is used as part of the clustering code to
61 * help make sure we have a good pool of storage to cluster in, without filling
62 * the FS with empty chunks
63 *
64 * CHUNK_ALLOC_FORCE means it must try to allocate one
65 *
66 * CHUNK_ALLOC_FORCE_FOR_EXTENT like CHUNK_ALLOC_FORCE but called from
67 * find_free_extent() that also activates the zone
68 */
69 enum btrfs_chunk_alloc_enum {
70 CHUNK_ALLOC_NO_FORCE,
71 CHUNK_ALLOC_LIMITED,
72 CHUNK_ALLOC_FORCE,
73 CHUNK_ALLOC_FORCE_FOR_EXTENT,
74 };
75
76 /* Block group flags set at runtime */
77 enum btrfs_block_group_flags {
78 BLOCK_GROUP_FLAG_IREF,
79 BLOCK_GROUP_FLAG_REMOVED,
80 BLOCK_GROUP_FLAG_TO_COPY,
81 BLOCK_GROUP_FLAG_RELOCATING_REPAIR,
82 BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED,
83 BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
84 BLOCK_GROUP_FLAG_ZONED_DATA_RELOC,
85 /* Does the block group need to be added to the free space tree? */
86 BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE,
87 /* Set after we add a new block group to the free space tree. */
88 BLOCK_GROUP_FLAG_FREE_SPACE_ADDED,
89 /* Indicate that the block group is placed on a sequential zone */
90 BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE,
91 /*
92 * Indicate that block group is in the list of new block groups of a
93 * transaction.
94 */
95 BLOCK_GROUP_FLAG_NEW,
96 BLOCK_GROUP_FLAG_FULLY_REMAPPED,
97 BLOCK_GROUP_FLAG_STRIPE_REMOVAL_PENDING,
98 };
99
100 enum btrfs_caching_type {
101 BTRFS_CACHE_NO,
102 BTRFS_CACHE_STARTED,
103 BTRFS_CACHE_FINISHED,
104 BTRFS_CACHE_ERROR,
105 };
106
107 struct btrfs_caching_control {
108 struct list_head list;
109 struct mutex mutex;
110 wait_queue_head_t wait;
111 struct btrfs_work work;
112 struct btrfs_block_group *block_group;
113 /* Track progress of caching during allocation. */
114 atomic_t progress;
115 refcount_t count;
116 };
117
118 /* Once caching_thread() finds this much free space, it will wake up waiters. */
119 #define CACHING_CTL_WAKE_UP SZ_2M
120
121 struct btrfs_block_group {
122 struct btrfs_fs_info *fs_info;
123 struct btrfs_inode *inode;
124 spinlock_t lock;
125 unsigned int ro;
126 u64 start;
127 u64 length;
128 u64 pinned;
129 u64 reserved;
130 u64 used;
131 u64 delalloc_bytes;
132 u64 bytes_super;
133 u64 flags;
134 u64 cache_generation;
135 u64 global_root_id;
136 u64 remap_bytes;
137 u32 identity_remap_count;
138 /* The last commited identity_remap_count value of this block group. */
139 u32 last_identity_remap_count;
140 /*
141 * The last committed used bytes of this block group, if the above @used
142 * is still the same as @last_used, we don't need to update block
143 * group item of this block group.
144 */
145 u64 last_used;
146 /* The last committed remap_bytes value of this block group. */
147 u64 last_remap_bytes;
148 /* The last committed flags value for this block group. */
149 u64 last_flags;
150
151 /*
152 * If the free space extent count exceeds this number, convert the block
153 * group to bitmaps.
154 */
155 u32 bitmap_high_thresh;
156
157 /*
158 * If the free space extent count drops below this number, convert the
159 * block group back to extents.
160 */
161 u32 bitmap_low_thresh;
162
163 /*
164 * It is just used for the delayed data space allocation because
165 * only the data space allocation and the relative metadata update
166 * can be done cross the transaction.
167 */
168 struct rw_semaphore data_rwsem;
169
170 /* For raid56, this is a full stripe, without parity */
171 unsigned long full_stripe_len;
172 unsigned long runtime_flags;
173
174 enum btrfs_disk_cache_state disk_cache_state;
175
176 /* Cache tracking stuff */
177 enum btrfs_caching_type cached;
178 struct btrfs_caching_control *caching_ctl;
179
180 struct btrfs_space_info *space_info;
181
182 /* Free space cache stuff */
183 struct btrfs_free_space_ctl *free_space_ctl;
184
185 /* Block group cache stuff */
186 struct rb_node cache_node;
187
188 /* For block groups in the same raid type */
189 struct list_head list;
190
191 refcount_t refs;
192
193 /*
194 * When non-zero it means the block group's logical address and its
195 * device extents can not be reused for future block group allocations
196 * until the counter goes down to 0. This is to prevent them from being
197 * reused while some task is still using the block group after it was
198 * deleted - we want to make sure they can only be reused for new block
199 * groups after that task is done with the deleted block group.
200 */
201 atomic_t frozen;
202
203 /*
204 * List of struct btrfs_free_clusters for this block group.
205 * Today it will only have one thing on it, but that may change
206 */
207 struct list_head cluster_list;
208
209 /*
210 * Used for several lists:
211 *
212 * 1) struct btrfs_fs_info::unused_bgs
213 * 2) struct btrfs_fs_info::reclaim_bgs
214 * 3) struct btrfs_transaction::deleted_bgs
215 * 4) struct btrfs_trans_handle::new_bgs
216 */
217 struct list_head bg_list;
218
219 /* For read-only block groups */
220 struct list_head ro_list;
221
222 /* For discard operations */
223 struct list_head discard_list;
224 int discard_index;
225 enum btrfs_discard_state discard_state;
226 u64 discard_eligible_time;
227 u64 discard_cursor;
228
229 /* For dirty block groups */
230 struct list_head dirty_list;
231 struct list_head io_list;
232
233 struct btrfs_io_ctl io_ctl;
234
235 /*
236 * Incremented when doing extent allocations and holding a read lock
237 * on the space_info's groups_sem semaphore.
238 * Decremented when an ordered extent that represents an IO against this
239 * block group's range is created (after it's added to its inode's
240 * root's list of ordered extents) or immediately after the allocation
241 * if it's a metadata extent or fallocate extent (for these cases we
242 * don't create ordered extents).
243 */
244 atomic_t reservations;
245
246 /*
247 * Incremented while holding the spinlock *lock* by a task checking if
248 * it can perform a nocow write (incremented if the value for the *ro*
249 * field is 0). Decremented by such tasks once they create an ordered
250 * extent or before that if some error happens before reaching that step.
251 * This is to prevent races between block group relocation and nocow
252 * writes through direct IO.
253 */
254 atomic_t nocow_writers;
255
256 /* Lock for free space tree operations. */
257 struct mutex free_space_lock;
258
259 /* Protected by @free_space_lock. */
260 bool using_free_space_bitmaps;
261 /* Protected by @free_space_lock. */
262 bool using_free_space_bitmaps_cached;
263
264 enum btrfs_block_group_size_class size_class:8;
265
266 /* If set, this blockgroup is not used for allocation between two reclaim sweeps. */
267 bool reclaim_mark;
268
269 /*
270 * Number of extents in this block group used for swap files.
271 * All accesses protected by the spinlock 'lock'.
272 */
273 int swap_extents;
274
275 /*
276 * Allocation offset for the block group to implement sequential
277 * allocation. This is used only on a zoned filesystem.
278 */
279 u64 alloc_offset;
280 u64 zone_unusable;
281 u64 zone_capacity;
282 u64 meta_write_pointer;
283 struct btrfs_chunk_map *physical_map;
284 struct list_head active_bg_list;
285 struct work_struct zone_finish_work;
286 struct extent_buffer *last_eb;
287 };
288
btrfs_block_group_end(const struct btrfs_block_group * block_group)289 static inline u64 btrfs_block_group_end(const struct btrfs_block_group *block_group)
290 {
291 return (block_group->start + block_group->length);
292 }
293
btrfs_is_block_group_used(const struct btrfs_block_group * bg)294 static inline bool btrfs_is_block_group_used(const struct btrfs_block_group *bg)
295 {
296 lockdep_assert_held(&bg->lock);
297
298 return (bg->used > 0 || bg->reserved > 0 || bg->pinned > 0 ||
299 bg->remap_bytes > 0);
300 }
301
btrfs_is_block_group_data_only(const struct btrfs_block_group * block_group)302 static inline bool btrfs_is_block_group_data_only(const struct btrfs_block_group *block_group)
303 {
304 /*
305 * In mixed mode the fragmentation is expected to be high, lowering the
306 * efficiency, so only proper data block groups are considered.
307 */
308 return (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
309 !(block_group->flags & BTRFS_BLOCK_GROUP_METADATA);
310 }
311
btrfs_block_group_available_space(const struct btrfs_block_group * bg)312 static inline u64 btrfs_block_group_available_space(const struct btrfs_block_group *bg)
313 {
314 lockdep_assert_held(&bg->lock);
315
316 return (bg->length - bg->used - bg->pinned - bg->reserved -
317 bg->bytes_super - bg->zone_unusable);
318 }
319
320 #ifdef CONFIG_BTRFS_DEBUG
321 int btrfs_should_fragment_free_space(const struct btrfs_block_group *block_group);
322 #endif
323
324 int __init btrfs_init_block_group(void);
325 void __cold btrfs_exit_block_group(void);
326
327 struct btrfs_block_group *btrfs_lookup_first_block_group(
328 struct btrfs_fs_info *info, u64 bytenr);
329 struct btrfs_block_group *btrfs_lookup_block_group(
330 struct btrfs_fs_info *info, u64 bytenr);
331 struct btrfs_block_group *btrfs_next_block_group(
332 struct btrfs_block_group *cache);
333 void btrfs_get_block_group(struct btrfs_block_group *cache);
334 void btrfs_put_block_group(struct btrfs_block_group *cache);
335 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
336 const u64 start);
337 void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg);
338 struct btrfs_block_group *btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info,
339 u64 bytenr);
340 void btrfs_dec_nocow_writers(struct btrfs_block_group *bg);
341 void btrfs_wait_nocow_writers(struct btrfs_block_group *bg);
342 void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache,
343 u64 num_bytes);
344 int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait);
345 struct btrfs_caching_control *btrfs_get_caching_control(
346 struct btrfs_block_group *cache);
347 int btrfs_add_new_free_space(struct btrfs_block_group *block_group,
348 u64 start, u64 end, u64 *total_added_ret);
349 struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
350 struct btrfs_fs_info *fs_info,
351 const u64 chunk_offset);
352 void btrfs_remove_bg_from_sinfo(struct btrfs_block_group *bg);
353 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
354 struct btrfs_chunk_map *map);
355 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info);
356 void btrfs_mark_bg_unused(struct btrfs_block_group *bg);
357 void btrfs_reclaim_block_groups(struct btrfs_fs_info *fs_info, unsigned int limit);
358 void btrfs_reclaim_bgs_work(struct work_struct *work);
359 void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info);
360 void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg);
361 int btrfs_read_block_groups(struct btrfs_fs_info *info);
362 struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans,
363 struct btrfs_space_info *space_info,
364 u64 type, u64 chunk_offset, u64 size);
365 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans);
366 int btrfs_inc_block_group_ro(struct btrfs_block_group *cache,
367 bool do_chunk_alloc);
368 void btrfs_dec_block_group_ro(struct btrfs_block_group *cache);
369 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans);
370 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans);
371 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans);
372 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
373 u64 bytenr, u64 num_bytes, bool alloc);
374 int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
375 u64 ram_bytes, u64 num_bytes, bool delalloc,
376 bool force_wrong_size_class);
377 void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, u64 num_bytes,
378 bool is_delalloc);
379 int btrfs_chunk_alloc(struct btrfs_trans_handle *trans,
380 struct btrfs_space_info *space_info, u64 flags,
381 enum btrfs_chunk_alloc_enum force);
382 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type);
383 void check_system_chunk(struct btrfs_trans_handle *trans, const u64 type);
384 void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans,
385 bool is_item_insertion);
386 u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags);
387 void btrfs_put_block_group_cache(struct btrfs_fs_info *info);
388 int btrfs_free_block_groups(struct btrfs_fs_info *info);
389 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
390 u64 physical, u64 **logical, int *naddrs, int *stripe_len);
391
btrfs_data_alloc_profile(struct btrfs_fs_info * fs_info)392 static inline u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
393 {
394 return btrfs_get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
395 }
396
btrfs_metadata_alloc_profile(struct btrfs_fs_info * fs_info)397 static inline u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
398 {
399 return btrfs_get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
400 }
401
btrfs_system_alloc_profile(struct btrfs_fs_info * fs_info)402 static inline u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
403 {
404 return btrfs_get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
405 }
406
btrfs_block_group_done(const struct btrfs_block_group * cache)407 static inline int btrfs_block_group_done(const struct btrfs_block_group *cache)
408 {
409 smp_mb();
410 return cache->cached == BTRFS_CACHE_FINISHED ||
411 cache->cached == BTRFS_CACHE_ERROR;
412 }
413
414 void btrfs_freeze_block_group(struct btrfs_block_group *cache);
415 void btrfs_unfreeze_block_group(struct btrfs_block_group *cache);
416
417 bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg);
418 void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount);
419
420 enum btrfs_block_group_size_class btrfs_calc_block_group_size_class(u64 size);
421 int btrfs_use_block_group_size_class(struct btrfs_block_group *bg,
422 enum btrfs_block_group_size_class size_class,
423 bool force_wrong_size_class);
424 bool btrfs_block_group_should_use_size_class(const struct btrfs_block_group *bg);
425 void btrfs_mark_bg_fully_remapped(struct btrfs_block_group *bg,
426 struct btrfs_trans_handle *trans);
427 int btrfs_populate_fully_remapped_bgs_list(struct btrfs_fs_info *fs_info);
428
429 #endif /* BTRFS_BLOCK_GROUP_H */
430