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 "compression.h"
16 #include "messages.h"
17 #include "ulist.h"
18 #include "misc.h"
19
20 struct page;
21 struct file;
22 struct folio;
23 struct inode;
24 struct fiemap_extent_info;
25 struct readahead_control;
26 struct address_space;
27 struct writeback_control;
28 struct extent_io_tree;
29 struct extent_map_tree;
30 struct extent_state;
31 struct btrfs_block_group;
32 struct btrfs_fs_info;
33 struct btrfs_inode;
34 struct btrfs_root;
35 struct btrfs_trans_handle;
36 struct btrfs_tree_parent_check;
37
38 enum {
39 EXTENT_BUFFER_UPTODATE,
40 EXTENT_BUFFER_DIRTY,
41 EXTENT_BUFFER_CORRUPT,
42 /* this got triggered by readahead */
43 EXTENT_BUFFER_READAHEAD,
44 EXTENT_BUFFER_TREE_REF,
45 EXTENT_BUFFER_STALE,
46 EXTENT_BUFFER_WRITEBACK,
47 /* read IO error */
48 EXTENT_BUFFER_READ_ERR,
49 EXTENT_BUFFER_UNMAPPED,
50 EXTENT_BUFFER_IN_TREE,
51 /* write IO error */
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,
57 };
58
59 /* these are flags for __process_pages_contig */
60 enum {
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),
66 };
67
68 /*
69 * Folio private values. Every page that is controlled by the extent map has
70 * folio private set to this value.
71 */
72 #define EXTENT_FOLIO_PRIVATE 1
73
74 /*
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.
80 */
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)))
87
88
89 int __init extent_buffer_init_cachep(void);
90 void __cold extent_buffer_free_cachep(void);
91
92 #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE)
93 struct extent_buffer {
94 u64 start;
95 u32 len;
96 u32 folio_size;
97 unsigned long bflags;
98 struct btrfs_fs_info *fs_info;
99
100 /*
101 * The address where the eb can be accessed without any cross-page handling.
102 * This can be NULL if not possible.
103 */
104 void *addr;
105
106 spinlock_t refs_lock;
107 atomic_t refs;
108 int read_mirror;
109 /* >= 0 if eb belongs to a log tree, -1 otherwise */
110 s8 log_index;
111 u8 folio_shift;
112 struct rcu_head rcu_head;
113
114 struct rw_semaphore lock;
115
116 /*
117 * Pointers to all the folios of the extent buffer.
118 *
119 * For now the folio is always order 0 (aka, a single page).
120 */
121 struct folio *folios[INLINE_EXTENT_BUFFER_PAGES];
122 #ifdef CONFIG_BTRFS_DEBUG
123 struct list_head leak_list;
124 pid_t lock_owner;
125 #endif
126 };
127
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;
133 };
134
135 static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb,
136 u64 start)
137 {
138 ASSERT(eb->folio_size);
139 return start & (eb->folio_size - 1);
140 }
141
142 /*
143 * Get the correct offset inside the page of extent buffer.
144 *
145 * @eb: target extent buffer
146 * @start: offset inside the extent buffer
147 *
148 * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases.
149 */
150 static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb,
151 unsigned long offset)
152 {
153 /*
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.
161 *
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
166 * that eb.
167 */
168 return offset_in_folio(eb->folios[0], offset + eb->start);
169 }
170
171 static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb,
172 unsigned long offset)
173 {
174 /*
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
178 * return 0 as index.
179 * 1.2) Several page sized folios
180 * The folio_shift would be PAGE_SHIFT, giving us the correct
181 * index.
182 *
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.
185 */
186 return offset >> eb->folio_shift;
187 }
188
189 /*
190 * Structure to record how many bytes and which ranges are set/cleared
191 */
192 struct extent_changeset {
193 /* How many bytes are set/cleared in this operation */
194 u64 bytes_changed;
195
196 /* Changed ranges */
197 struct ulist range_changed;
198 };
199
200 static inline void extent_changeset_init(struct extent_changeset *changeset)
201 {
202 changeset->bytes_changed = 0;
203 ulist_init(&changeset->range_changed);
204 }
205
206 static inline struct extent_changeset *extent_changeset_alloc(void)
207 {
208 struct extent_changeset *ret;
209
210 ret = kmalloc(sizeof(*ret), GFP_KERNEL);
211 if (!ret)
212 return NULL;
213
214 extent_changeset_init(ret);
215 return ret;
216 }
217
218 static inline void extent_changeset_prealloc(struct extent_changeset *changeset, gfp_t gfp_mask)
219 {
220 ulist_prealloc(&changeset->range_changed, gfp_mask);
221 }
222
223 static inline void extent_changeset_release(struct extent_changeset *changeset)
224 {
225 if (!changeset)
226 return;
227 changeset->bytes_changed = 0;
228 ulist_release(&changeset->range_changed);
229 }
230
231 static inline void extent_changeset_free(struct extent_changeset *changeset)
232 {
233 if (!changeset)
234 return;
235 extent_changeset_release(changeset);
236 kfree(changeset);
237 }
238
239 bool try_release_extent_mapping(struct page *page, gfp_t mask);
240 int try_release_extent_buffer(struct page *page);
241
242 int btrfs_read_folio(struct file *file, struct folio *folio);
243 void extent_write_locked_range(struct inode *inode, const struct page *locked_page,
244 u64 start, u64 end, struct writeback_control *wbc,
245 bool pages_dirty);
246 int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc);
247 int btree_write_cache_pages(struct address_space *mapping,
248 struct writeback_control *wbc);
249 void btrfs_readahead(struct readahead_control *rac);
250 int set_folio_extent_mapped(struct folio *folio);
251 int set_page_extent_mapped(struct page *page);
252 void clear_page_extent_mapped(struct page *page);
253
254 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
255 u64 start, u64 owner_root, int level);
256 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
257 u64 start, unsigned long len);
258 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
259 u64 start);
260 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src);
261 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
262 u64 start);
263 void free_extent_buffer(struct extent_buffer *eb);
264 void free_extent_buffer_stale(struct extent_buffer *eb);
265 #define WAIT_NONE 0
266 #define WAIT_COMPLETE 1
267 #define WAIT_PAGE_LOCK 2
268 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
269 const struct btrfs_tree_parent_check *parent_check);
270 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
271 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
272 u64 bytenr, u64 owner_root, u64 gen, int level);
273 void btrfs_readahead_node_child(struct extent_buffer *node, int slot);
274
275 static inline int num_extent_pages(const struct extent_buffer *eb)
276 {
277 /*
278 * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to
279 * sectorsize, it's just eb->len >> PAGE_SHIFT.
280 *
281 * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE,
282 * thus have to ensure we get at least one page.
283 */
284 return (eb->len >> PAGE_SHIFT) ?: 1;
285 }
286
287 /*
288 * This can only be determined at runtime by checking eb::folios[0].
289 *
290 * As we can have either one large folio covering the whole eb
291 * (either nodesize <= PAGE_SIZE, or high order folio), or multiple
292 * single-paged folios.
293 */
294 static inline int num_extent_folios(const struct extent_buffer *eb)
295 {
296 if (folio_order(eb->folios[0]))
297 return 1;
298 return num_extent_pages(eb);
299 }
300
301 static inline int extent_buffer_uptodate(const struct extent_buffer *eb)
302 {
303 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
304 }
305
306 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
307 unsigned long start, unsigned long len);
308 void read_extent_buffer(const struct extent_buffer *eb, void *dst,
309 unsigned long start,
310 unsigned long len);
311 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
312 void __user *dst, unsigned long start,
313 unsigned long len);
314 void write_extent_buffer(const struct extent_buffer *eb, const void *src,
315 unsigned long start, unsigned long len);
316
317 static inline void write_extent_buffer_chunk_tree_uuid(
318 const struct extent_buffer *eb, const void *chunk_tree_uuid)
319 {
320 write_extent_buffer(eb, chunk_tree_uuid,
321 offsetof(struct btrfs_header, chunk_tree_uuid),
322 BTRFS_FSID_SIZE);
323 }
324
325 static inline void write_extent_buffer_fsid(const struct extent_buffer *eb,
326 const void *fsid)
327 {
328 write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
329 BTRFS_FSID_SIZE);
330 }
331
332 void copy_extent_buffer_full(const struct extent_buffer *dst,
333 const struct extent_buffer *src);
334 void copy_extent_buffer(const struct extent_buffer *dst,
335 const struct extent_buffer *src,
336 unsigned long dst_offset, unsigned long src_offset,
337 unsigned long len);
338 void memcpy_extent_buffer(const struct extent_buffer *dst,
339 unsigned long dst_offset, unsigned long src_offset,
340 unsigned long len);
341 void memmove_extent_buffer(const struct extent_buffer *dst,
342 unsigned long dst_offset, unsigned long src_offset,
343 unsigned long len);
344 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
345 unsigned long len);
346 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
347 unsigned long pos);
348 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
349 unsigned long pos, unsigned long len);
350 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
351 unsigned long start, unsigned long pos,
352 unsigned long len);
353 void set_extent_buffer_dirty(struct extent_buffer *eb);
354 void set_extent_buffer_uptodate(struct extent_buffer *eb);
355 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
356 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
357 const struct page *locked_page,
358 struct extent_state **cached,
359 u32 bits_to_clear, unsigned long page_ops);
360 int extent_invalidate_folio(struct extent_io_tree *tree,
361 struct folio *folio, size_t offset);
362 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
363 struct extent_buffer *buf);
364
365 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
366 bool nofail);
367 int btrfs_alloc_folio_array(unsigned int nr_folios, struct folio **folio_array);
368
369 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
370 bool find_lock_delalloc_range(struct inode *inode,
371 struct page *locked_page, u64 *start,
372 u64 *end);
373 #endif
374 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
375 u64 start);
376
377 #ifdef CONFIG_BTRFS_DEBUG
378 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info);
379 #else
380 #define btrfs_extent_buffer_leak_debug_check(fs_info) do {} while (0)
381 #endif
382
383 #endif
384
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.