1 /* SPDX-License-Identifier: GPL-2.0 */
2
3 #ifndef BTRFS_EXTENT_IO_H
4 #define BTRFS_EXTENT_IO_H
5
6 #include <linux/rbtree.h>
7 #include <linux/refcount.h>
8 #include <linux/fiemap.h>
9 #include <linux/btrfs_tree.h>
10 #include <linux/spinlock.h>
11 #include <linux/atomic.h>
12 #include <linux/rwsem.h>
13 #include <linux/list.h>
14 #include <linux/slab.h>
15 #include "messages.h"
16 #include "ulist.h"
17 #include "misc.h"
18
19 struct page;
20 struct file;
21 struct folio;
22 struct inode;
23 struct fiemap_extent_info;
24 struct readahead_control;
25 struct address_space;
26 struct writeback_control;
27 struct extent_io_tree;
28 struct extent_map_tree;
29 struct extent_state;
30 struct btrfs_block_group;
31 struct btrfs_fs_info;
32 struct btrfs_inode;
33 struct btrfs_root;
34 struct btrfs_trans_handle;
35 struct btrfs_tree_parent_check;
36
37 enum {
38 EXTENT_BUFFER_UPTODATE,
39 EXTENT_BUFFER_DIRTY,
40 EXTENT_BUFFER_TREE_REF,
41 EXTENT_BUFFER_STALE,
42 EXTENT_BUFFER_WRITEBACK,
43 EXTENT_BUFFER_UNMAPPED,
44 /* write IO error */
45 EXTENT_BUFFER_WRITE_ERR,
46 /* Indicate the extent buffer is written zeroed out (for zoned) */
47 EXTENT_BUFFER_ZONED_ZEROOUT,
48 /* Indicate that extent buffer pages a being read */
49 EXTENT_BUFFER_READING,
50 };
51
52 /* these are flags for __process_pages_contig */
53 enum {
54 ENUM_BIT(PAGE_UNLOCK),
55 /* Page starts writeback, clear dirty bit and set writeback bit */
56 ENUM_BIT(PAGE_START_WRITEBACK),
57 ENUM_BIT(PAGE_END_WRITEBACK),
58 };
59
60 /*
61 * Folio private values. Every page that is controlled by the extent map has
62 * folio private set to this value.
63 */
64 #define EXTENT_FOLIO_PRIVATE 1
65
66 /*
67 * The extent buffer bitmap operations are done with byte granularity instead of
68 * word granularity for two reasons:
69 * 1. The bitmaps must be little-endian on disk.
70 * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a
71 * single word in a bitmap may straddle two pages in the extent buffer.
72 */
73 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
74 #define BYTE_MASK ((1U << BITS_PER_BYTE) - 1)
75 #define BITMAP_FIRST_BYTE_MASK(start) \
76 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
77 #define BITMAP_LAST_BYTE_MASK(nbits) \
78 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
79
80
81 int __init extent_buffer_init_cachep(void);
82 void __cold extent_buffer_free_cachep(void);
83
84 #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE)
85 struct extent_buffer {
86 u64 start;
87 u32 len;
88 u32 folio_size;
89 unsigned long bflags;
90 struct btrfs_fs_info *fs_info;
91
92 /*
93 * The address where the eb can be accessed without any cross-page handling.
94 * This can be NULL if not possible.
95 */
96 void *addr;
97
98 spinlock_t refs_lock;
99 refcount_t refs;
100 int read_mirror;
101 /* Inhibit WB_SYNC_NONE writeback when > 0. */
102 atomic_t writeback_inhibitors;
103 /* >= 0 if eb belongs to a log tree, -1 otherwise */
104 s8 log_index;
105 u8 folio_shift;
106 struct rcu_head rcu_head;
107
108 struct rw_semaphore lock;
109
110 /*
111 * Pointers to all the folios of the extent buffer.
112 *
113 * For now the folio is always order 0 (aka, a single page).
114 */
115 struct folio *folios[INLINE_EXTENT_BUFFER_PAGES];
116 #ifdef CONFIG_BTRFS_DEBUG
117 struct list_head leak_list;
118 pid_t lock_owner;
119 #endif
120 };
121
122 /*
123 * Wrapper struct for managing preallocating an extent_buffer, its folios and a
124 * btrfs_folio_state if needed.
125 *
126 * Only used to mediate allocation, do not refer to the eb directly if not
127 * returned from a successful eb allocating API.
128 *
129 * The eb folios and bfs should generally not be fully attached, except briefly
130 * before they are NULLed in the struct after successful attachment.
131 */
132 struct btrfs_eb_prealloc {
133 struct extent_buffer *eb;
134 struct btrfs_folio_state *bfs;
135 /* eb alloc may use GFP_NOWAIT; caller can drop locks and retry. */
136 bool supports_nowait;
137 /* GFP_NOWAIT eb alloc failed; preallocate again and retry. */
138 bool needs_prealloc;
139 };
140
141 struct btrfs_eb_write_context {
142 struct writeback_control *wbc;
143 struct extent_buffer *eb;
144 /* Block group @eb resides in. Only used for zoned mode. */
145 struct btrfs_block_group *zoned_bg;
146 };
147
offset_in_eb_folio(const struct extent_buffer * eb,u64 start)148 static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb,
149 u64 start)
150 {
151 ASSERT(eb->folio_size);
152 return start & (eb->folio_size - 1);
153 }
154
155 /*
156 * Get the correct offset inside the page of extent buffer.
157 *
158 * @eb: target extent buffer
159 * @start: offset inside the extent buffer
160 *
161 * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases.
162 */
get_eb_offset_in_folio(const struct extent_buffer * eb,unsigned long offset)163 static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb,
164 unsigned long offset)
165 {
166 /*
167 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
168 * 1.1) One large folio covering the whole eb
169 * The eb->start is aligned to folio size, thus adding it
170 * won't cause any difference.
171 * 1.2) Several page sized folios
172 * The eb->start is aligned to folio (page) size, thus
173 * adding it won't cause any difference.
174 *
175 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
176 * In this case there would only be one page sized folio, and there
177 * may be several different extent buffers in the page/folio.
178 * We need to add eb->start to properly access the offset inside
179 * that eb.
180 */
181 return offset_in_folio(eb->folios[0], offset + eb->start);
182 }
183
get_eb_folio_index(const struct extent_buffer * eb,unsigned long offset)184 static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb,
185 unsigned long offset)
186 {
187 /*
188 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
189 * 1.1) One large folio covering the whole eb.
190 * the folio_shift would be large enough to always make us
191 * return 0 as index.
192 * 1.2) Several page sized folios
193 * The folio_shift would be PAGE_SHIFT, giving us the correct
194 * index.
195 *
196 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
197 * The folio would only be page sized, and always give us 0 as index.
198 */
199 return offset >> eb->folio_shift;
200 }
201
202 /*
203 * Structure to record how many bytes and which ranges are set/cleared
204 */
205 struct extent_changeset {
206 /* How many bytes are set/cleared in this operation */
207 u64 bytes_changed;
208
209 /* Changed ranges */
210 struct ulist range_changed;
211 };
212
extent_changeset_init(struct extent_changeset * changeset)213 static inline void extent_changeset_init(struct extent_changeset *changeset)
214 {
215 changeset->bytes_changed = 0;
216 ulist_init(&changeset->range_changed);
217 }
218
219 /*
220 * Sentinel value for range_changed.prealloc indicating that the changeset
221 * only tracks bytes_changed and does not record individual ranges. This
222 * avoids GFP_ATOMIC allocations inside add_extent_changeset() when the
223 * caller doesn't need to iterate the changed ranges afterwards.
224 */
225 #define EXTENT_CHANGESET_BYTES_ONLY ((struct ulist_node *)1)
226
extent_changeset_init_bytes_only(struct extent_changeset * changeset)227 static inline void extent_changeset_init_bytes_only(struct extent_changeset *changeset)
228 {
229 changeset->bytes_changed = 0;
230 changeset->range_changed.prealloc = EXTENT_CHANGESET_BYTES_ONLY;
231 }
232
extent_changeset_tracks_ranges(const struct extent_changeset * changeset)233 static inline bool extent_changeset_tracks_ranges(const struct extent_changeset *changeset)
234 {
235 return changeset->range_changed.prealloc != EXTENT_CHANGESET_BYTES_ONLY;
236 }
237
extent_changeset_alloc(void)238 static inline struct extent_changeset *extent_changeset_alloc(void)
239 {
240 struct extent_changeset *ret;
241
242 ret = kmalloc_obj(*ret);
243 if (!ret)
244 return NULL;
245
246 extent_changeset_init(ret);
247 return ret;
248 }
249
extent_changeset_prealloc(struct extent_changeset * changeset,gfp_t gfp_mask)250 static inline void extent_changeset_prealloc(struct extent_changeset *changeset, gfp_t gfp_mask)
251 {
252 ASSERT(extent_changeset_tracks_ranges(changeset));
253 ulist_prealloc(&changeset->range_changed, gfp_mask);
254 }
255
extent_changeset_release(struct extent_changeset * changeset)256 static inline void extent_changeset_release(struct extent_changeset *changeset)
257 {
258 if (!changeset)
259 return;
260 changeset->bytes_changed = 0;
261 if (extent_changeset_tracks_ranges(changeset))
262 ulist_release(&changeset->range_changed);
263 }
264
extent_changeset_free(struct extent_changeset * changeset)265 static inline void extent_changeset_free(struct extent_changeset *changeset)
266 {
267 if (!changeset)
268 return;
269 extent_changeset_release(changeset);
270 kfree(changeset);
271 }
272
273 bool try_release_extent_mapping(struct folio *folio, gfp_t mask);
274 int try_release_extent_buffer(struct folio *folio);
275
276 int btrfs_read_folio(struct file *file, struct folio *folio);
277 #ifdef CONFIG_BTRFS_DEBUG
278 void btrfs_check_folio_write_protected(struct folio *folio);
279 #else
btrfs_check_folio_write_protected(struct folio * folio)280 static inline void btrfs_check_folio_write_protected(struct folio *folio) { }
281 #endif
282 void extent_write_locked_range(struct inode *inode, const struct folio *locked_folio,
283 u64 start, u64 end, struct writeback_control *wbc,
284 bool pages_dirty);
285 int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc);
286 int btree_writepages(struct address_space *mapping, struct writeback_control *wbc);
287 void btrfs_btree_wait_writeback_range(struct btrfs_fs_info *fs_info, u64 start, u64 end);
288 void btrfs_readahead(struct readahead_control *rac);
289 int set_folio_extent_mapped(struct folio *folio);
290 void clear_folio_extent_mapped(struct folio *folio);
291
292 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
293 struct btrfs_eb_prealloc *pa,
294 u64 start, u64 owner_root, int level);
295 int btrfs_init_eb_prealloc(struct btrfs_fs_info *fs_info,
296 struct btrfs_eb_prealloc *pa, bool nowait);
297 void btrfs_free_eb_prealloc(struct btrfs_eb_prealloc *pa);
298 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
299 u64 start);
300 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src);
301 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
302 u64 start);
303 void free_extent_buffer(struct extent_buffer *eb);
304 void free_extent_buffer_stale(struct extent_buffer *eb);
305 int read_extent_buffer_pages(struct extent_buffer *eb, int mirror_num,
306 const struct btrfs_tree_parent_check *parent_check);
307 int read_extent_buffer_pages_nowait(struct extent_buffer *eb, int mirror_num,
308 const struct btrfs_tree_parent_check *parent_check);
309
wait_on_extent_buffer_writeback(struct extent_buffer * eb)310 static inline void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
311 {
312 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
313 TASK_UNINTERRUPTIBLE);
314 }
315
316 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
317 u64 bytenr, u64 owner_root, u64 gen, int level,
318 const struct btrfs_key *first_key);
319 void btrfs_readahead_node_child(struct extent_buffer *node, int slot);
320
321 /* Note: this can be used in for loops without caching the value in a variable. */
num_extent_pages(const struct extent_buffer * eb)322 static inline int __pure num_extent_pages(const struct extent_buffer *eb)
323 {
324 /*
325 * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to
326 * sectorsize, it's just eb->len >> PAGE_SHIFT.
327 *
328 * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE,
329 * thus have to ensure we get at least one page.
330 */
331 return (eb->len >> PAGE_SHIFT) ?: 1;
332 }
333
334 /*
335 * This can only be determined at runtime by checking eb::folios[0].
336 *
337 * As we can have either one large folio covering the whole eb
338 * (either nodesize <= PAGE_SIZE, or high order folio), or multiple
339 * single-paged folios.
340 *
341 * Note: this can be used in for loops without caching the value in a variable.
342 */
num_extent_folios(const struct extent_buffer * eb)343 static inline int __pure num_extent_folios(const struct extent_buffer *eb)
344 {
345 if (!eb->folios[0])
346 return 0;
347 if (folio_order(eb->folios[0]))
348 return 1;
349 return num_extent_pages(eb);
350 }
351
extent_buffer_uptodate(const struct extent_buffer * eb)352 static inline bool extent_buffer_uptodate(const struct extent_buffer *eb)
353 {
354 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
355 }
356
extent_buffer_under_io(const struct extent_buffer * eb)357 static inline bool extent_buffer_under_io(const struct extent_buffer *eb)
358 {
359 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
360 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
361 }
362
363 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
364 unsigned long start, unsigned long len);
365 void read_extent_buffer(const struct extent_buffer *eb, void *dst,
366 unsigned long start,
367 unsigned long len);
368 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
369 void __user *dst, unsigned long start,
370 unsigned long len);
371 void write_extent_buffer(const struct extent_buffer *eb, const void *src,
372 unsigned long start, unsigned long len);
373
write_extent_buffer_chunk_tree_uuid(const struct extent_buffer * eb,const void * chunk_tree_uuid)374 static inline void write_extent_buffer_chunk_tree_uuid(
375 const struct extent_buffer *eb, const void *chunk_tree_uuid)
376 {
377 write_extent_buffer(eb, chunk_tree_uuid,
378 offsetof(struct btrfs_header, chunk_tree_uuid),
379 BTRFS_FSID_SIZE);
380 }
381
write_extent_buffer_fsid(const struct extent_buffer * eb,const void * fsid)382 static inline void write_extent_buffer_fsid(const struct extent_buffer *eb,
383 const void *fsid)
384 {
385 write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
386 BTRFS_FSID_SIZE);
387 }
388
389 void copy_extent_buffer_full(const struct extent_buffer *dst,
390 const struct extent_buffer *src);
391 void copy_extent_buffer(const struct extent_buffer *dst,
392 const struct extent_buffer *src,
393 unsigned long dst_offset, unsigned long src_offset,
394 unsigned long len);
395 void memcpy_extent_buffer(const struct extent_buffer *dst,
396 unsigned long dst_offset, unsigned long src_offset,
397 unsigned long len);
398 void memmove_extent_buffer(const struct extent_buffer *dst,
399 unsigned long dst_offset, unsigned long src_offset,
400 unsigned long len);
401 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
402 unsigned long len);
403 bool extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
404 unsigned long pos);
405 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
406 unsigned long pos, unsigned long len);
407 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
408 unsigned long start, unsigned long pos,
409 unsigned long len);
410 void set_extent_buffer_dirty(struct extent_buffer *eb);
411 void set_extent_buffer_uptodate(struct extent_buffer *eb);
412 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
413 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
414 const struct folio *locked_folio,
415 struct extent_state **cached,
416 u32 bits_to_clear, unsigned long page_ops);
417 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
418 struct extent_buffer *buf);
419 void btrfs_zoned_release_dirty_metadata(struct btrfs_fs_info *fs_info);
420
btrfs_clear_folio_dirty_tag(struct folio * folio)421 static inline void btrfs_clear_folio_dirty_tag(struct folio *folio)
422 {
423 ASSERT(!folio_test_dirty(folio));
424 ASSERT(folio_test_locked(folio));
425 ASSERT(folio->mapping);
426 xa_lock_irq(&folio->mapping->i_pages);
427 __xa_clear_mark(&folio->mapping->i_pages, folio->index,
428 PAGECACHE_TAG_DIRTY);
429 xa_unlock_irq(&folio->mapping->i_pages);
430 }
431
432 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array, gfp_t gfp);
433 int btrfs_alloc_folio_array(unsigned int nr_folios, unsigned int order,
434 struct folio **folio_array, gfp_t gfp);
435
436 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
437 bool find_lock_delalloc_range(struct inode *inode,
438 struct folio *locked_folio, u64 *start,
439 u64 *end);
440 #endif
441 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
442 u64 start);
443
444 #ifdef CONFIG_BTRFS_DEBUG
445 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info);
446 #else
447 #define btrfs_extent_buffer_leak_debug_check(fs_info) do {} while (0)
448 #endif
449
450 void btrfs_inhibit_eb_writeback(struct btrfs_trans_handle *trans,
451 struct extent_buffer *eb);
452 void btrfs_uninhibit_all_eb_writeback(struct btrfs_trans_handle *trans);
453
454 #endif
455