1 /* SPDX-License-Identifier: GPL-2.0 */
3 #ifndef BTRFS_EXTENT_IO_H
4 #define BTRFS_EXTENT_IO_H
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 "compression.h"
24 struct fiemap_extent_info;
25 struct readahead_control;
27 struct writeback_control;
28 struct extent_io_tree;
29 struct extent_map_tree;
31 struct btrfs_block_group;
35 struct btrfs_trans_handle;
36 struct btrfs_tree_parent_check;
39 EXTENT_BUFFER_UPTODATE,
41 EXTENT_BUFFER_CORRUPT,
42 /* this got triggered by readahead */
43 EXTENT_BUFFER_READAHEAD,
44 EXTENT_BUFFER_TREE_REF,
46 EXTENT_BUFFER_WRITEBACK,
48 EXTENT_BUFFER_READ_ERR,
49 EXTENT_BUFFER_UNMAPPED,
50 EXTENT_BUFFER_IN_TREE,
52 EXTENT_BUFFER_WRITE_ERR,
53 /* Indicate the extent buffer is written zeroed out (for zoned) */
54 EXTENT_BUFFER_ZONED_ZEROOUT,
55 /* Indicate that extent buffer pages a being read */
56 EXTENT_BUFFER_READING,
59 /* these are flags for __process_pages_contig */
61 ENUM_BIT(PAGE_UNLOCK),
62 /* Page starts writeback, clear dirty bit and set writeback bit */
63 ENUM_BIT(PAGE_START_WRITEBACK),
64 ENUM_BIT(PAGE_END_WRITEBACK),
65 ENUM_BIT(PAGE_SET_ORDERED),
69 * Folio private values. Every page that is controlled by the extent map has
70 * folio private set to this value.
72 #define EXTENT_FOLIO_PRIVATE 1
75 * The extent buffer bitmap operations are done with byte granularity instead of
76 * word granularity for two reasons:
77 * 1. The bitmaps must be little-endian on disk.
78 * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a
79 * single word in a bitmap may straddle two pages in the extent buffer.
81 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
82 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
83 #define BITMAP_FIRST_BYTE_MASK(start) \
84 ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
85 #define BITMAP_LAST_BYTE_MASK(nbits) \
86 (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
89 int __init extent_buffer_init_cachep(void);
90 void __cold extent_buffer_free_cachep(void);
92 #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE)
93 struct extent_buffer {
98 struct btrfs_fs_info *fs_info;
101 * The address where the eb can be accessed without any cross-page handling.
102 * This can be NULL if not possible.
106 spinlock_t refs_lock;
109 /* >= 0 if eb belongs to a log tree, -1 otherwise */
112 struct rcu_head rcu_head;
114 struct rw_semaphore lock;
117 * Pointers to all the folios of the extent buffer.
119 * For now the folio is always order 0 (aka, a single page).
121 struct folio *folios[INLINE_EXTENT_BUFFER_PAGES];
122 #ifdef CONFIG_BTRFS_DEBUG
123 struct list_head leak_list;
128 struct btrfs_eb_write_context {
129 struct writeback_control *wbc;
130 struct extent_buffer *eb;
131 /* Block group @eb resides in. Only used for zoned mode. */
132 struct btrfs_block_group *zoned_bg;
135 static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb,
138 ASSERT(eb->folio_size);
139 return start & (eb->folio_size - 1);
143 * Get the correct offset inside the page of extent buffer.
145 * @eb: target extent buffer
146 * @start: offset inside the extent buffer
148 * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases.
150 static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb,
151 unsigned long offset)
154 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
155 * 1.1) One large folio covering the whole eb
156 * The eb->start is aligned to folio size, thus adding it
157 * won't cause any difference.
158 * 1.2) Several page sized folios
159 * The eb->start is aligned to folio (page) size, thus
160 * adding it won't cause any difference.
162 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
163 * In this case there would only be one page sized folio, and there
164 * may be several different extent buffers in the page/folio.
165 * We need to add eb->start to properly access the offset inside
168 return offset_in_folio(eb->folios[0], offset + eb->start);
171 static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb,
172 unsigned long offset)
175 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
176 * 1.1) One large folio covering the whole eb.
177 * the folio_shift would be large enough to always make us
179 * 1.2) Several page sized folios
180 * The folio_shift would be PAGE_SHIFT, giving us the correct
183 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
184 * The folio would only be page sized, and always give us 0 as index.
186 return offset >> eb->folio_shift;
190 * Structure to record how many bytes and which ranges are set/cleared
192 struct extent_changeset {
193 /* How many bytes are set/cleared in this operation */
197 struct ulist range_changed;
200 static inline void extent_changeset_init(struct extent_changeset *changeset)
202 changeset->bytes_changed = 0;
203 ulist_init(&changeset->range_changed);
206 static inline struct extent_changeset *extent_changeset_alloc(void)
208 struct extent_changeset *ret;
210 ret = kmalloc(sizeof(*ret), GFP_KERNEL);
214 extent_changeset_init(ret);
218 static inline void extent_changeset_release(struct extent_changeset *changeset)
222 changeset->bytes_changed = 0;
223 ulist_release(&changeset->range_changed);
226 static inline void extent_changeset_free(struct extent_changeset *changeset)
230 extent_changeset_release(changeset);
234 bool try_release_extent_mapping(struct page *page, gfp_t mask);
235 int try_release_extent_buffer(struct page *page);
237 int btrfs_read_folio(struct file *file, struct folio *folio);
238 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
239 u64 start, u64 end, struct writeback_control *wbc,
241 int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc);
242 int btree_write_cache_pages(struct address_space *mapping,
243 struct writeback_control *wbc);
244 void btrfs_readahead(struct readahead_control *rac);
245 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
247 int set_folio_extent_mapped(struct folio *folio);
248 int set_page_extent_mapped(struct page *page);
249 void clear_page_extent_mapped(struct page *page);
251 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
252 u64 start, u64 owner_root, int level);
253 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
254 u64 start, unsigned long len);
255 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
257 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src);
258 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
260 void free_extent_buffer(struct extent_buffer *eb);
261 void free_extent_buffer_stale(struct extent_buffer *eb);
263 #define WAIT_COMPLETE 1
264 #define WAIT_PAGE_LOCK 2
265 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
266 struct btrfs_tree_parent_check *parent_check);
267 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
268 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
269 u64 bytenr, u64 owner_root, u64 gen, int level);
270 void btrfs_readahead_node_child(struct extent_buffer *node, int slot);
272 static inline int num_extent_pages(const struct extent_buffer *eb)
275 * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to
276 * sectorsize, it's just eb->len >> PAGE_SHIFT.
278 * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE,
279 * thus have to ensure we get at least one page.
281 return (eb->len >> PAGE_SHIFT) ?: 1;
285 * This can only be determined at runtime by checking eb::folios[0].
287 * As we can have either one large folio covering the whole eb
288 * (either nodesize <= PAGE_SIZE, or high order folio), or multiple
289 * single-paged folios.
291 static inline int num_extent_folios(const struct extent_buffer *eb)
293 if (folio_order(eb->folios[0]))
295 return num_extent_pages(eb);
298 static inline int extent_buffer_uptodate(const struct extent_buffer *eb)
300 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
303 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
304 unsigned long start, unsigned long len);
305 void read_extent_buffer(const struct extent_buffer *eb, void *dst,
308 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
309 void __user *dst, unsigned long start,
311 void write_extent_buffer(const struct extent_buffer *eb, const void *src,
312 unsigned long start, unsigned long len);
314 static inline void write_extent_buffer_chunk_tree_uuid(
315 const struct extent_buffer *eb, const void *chunk_tree_uuid)
317 write_extent_buffer(eb, chunk_tree_uuid,
318 offsetof(struct btrfs_header, chunk_tree_uuid),
322 static inline void write_extent_buffer_fsid(const struct extent_buffer *eb,
325 write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
329 void copy_extent_buffer_full(const struct extent_buffer *dst,
330 const struct extent_buffer *src);
331 void copy_extent_buffer(const struct extent_buffer *dst,
332 const struct extent_buffer *src,
333 unsigned long dst_offset, unsigned long src_offset,
335 void memcpy_extent_buffer(const struct extent_buffer *dst,
336 unsigned long dst_offset, unsigned long src_offset,
338 void memmove_extent_buffer(const struct extent_buffer *dst,
339 unsigned long dst_offset, unsigned long src_offset,
341 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
343 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
345 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
346 unsigned long pos, unsigned long len);
347 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
348 unsigned long start, unsigned long pos,
350 void set_extent_buffer_dirty(struct extent_buffer *eb);
351 void set_extent_buffer_uptodate(struct extent_buffer *eb);
352 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
353 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
354 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
355 struct page *locked_page,
356 struct extent_state **cached,
357 u32 bits_to_clear, unsigned long page_ops);
358 int extent_invalidate_folio(struct extent_io_tree *tree,
359 struct folio *folio, size_t offset);
360 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
361 struct extent_buffer *buf);
363 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
365 int btrfs_alloc_folio_array(unsigned int nr_folios, struct folio **folio_array,
368 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
369 bool find_lock_delalloc_range(struct inode *inode,
370 struct page *locked_page, u64 *start,
373 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
376 #ifdef CONFIG_BTRFS_DEBUG
377 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info);
379 #define btrfs_extent_buffer_leak_debug_check(fs_info) do {} while (0)