xref: /linux/fs/btrfs/extent_io.h (revision 23fd95663b070cf781f37b8058a8c055f168110b)
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 struct btrfs_eb_write_context {
123 	struct writeback_control *wbc;
124 	struct extent_buffer *eb;
125 	/* Block group @eb resides in. Only used for zoned mode. */
126 	struct btrfs_block_group *zoned_bg;
127 };
128 
129 static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb,
130 					       u64 start)
131 {
132 	ASSERT(eb->folio_size);
133 	return start & (eb->folio_size - 1);
134 }
135 
136 /*
137  * Get the correct offset inside the page of extent buffer.
138  *
139  * @eb:		target extent buffer
140  * @start:	offset inside the extent buffer
141  *
142  * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases.
143  */
144 static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb,
145 					    unsigned long offset)
146 {
147 	/*
148 	 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
149 	 *    1.1) One large folio covering the whole eb
150 	 *	   The eb->start is aligned to folio size, thus adding it
151 	 *	   won't cause any difference.
152 	 *    1.2) Several page sized folios
153 	 *	   The eb->start is aligned to folio (page) size, thus
154 	 *	   adding it won't cause any difference.
155 	 *
156 	 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
157 	 *    In this case there would only be one page sized folio, and there
158 	 *    may be several different extent buffers in the page/folio.
159 	 *    We need to add eb->start to properly access the offset inside
160 	 *    that eb.
161 	 */
162 	return offset_in_folio(eb->folios[0], offset + eb->start);
163 }
164 
165 static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb,
166 					       unsigned long offset)
167 {
168 	/*
169 	 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
170 	 *    1.1) One large folio covering the whole eb.
171 	 *	   the folio_shift would be large enough to always make us
172 	 *	   return 0 as index.
173 	 *    1.2) Several page sized folios
174 	 *         The folio_shift would be PAGE_SHIFT, giving us the correct
175 	 *         index.
176 	 *
177 	 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
178 	 *    The folio would only be page sized, and always give us 0 as index.
179 	 */
180 	return offset >> eb->folio_shift;
181 }
182 
183 /*
184  * Structure to record how many bytes and which ranges are set/cleared
185  */
186 struct extent_changeset {
187 	/* How many bytes are set/cleared in this operation */
188 	u64 bytes_changed;
189 
190 	/* Changed ranges */
191 	struct ulist range_changed;
192 };
193 
194 static inline void extent_changeset_init(struct extent_changeset *changeset)
195 {
196 	changeset->bytes_changed = 0;
197 	ulist_init(&changeset->range_changed);
198 }
199 
200 /*
201  * Sentinel value for range_changed.prealloc indicating that the changeset
202  * only tracks bytes_changed and does not record individual ranges. This
203  * avoids GFP_ATOMIC allocations inside add_extent_changeset() when the
204  * caller doesn't need to iterate the changed ranges afterwards.
205  */
206 #define EXTENT_CHANGESET_BYTES_ONLY	((struct ulist_node *)1)
207 
208 static inline void extent_changeset_init_bytes_only(struct extent_changeset *changeset)
209 {
210 	changeset->bytes_changed = 0;
211 	changeset->range_changed.prealloc = EXTENT_CHANGESET_BYTES_ONLY;
212 }
213 
214 static inline bool extent_changeset_tracks_ranges(const struct extent_changeset *changeset)
215 {
216 	return changeset->range_changed.prealloc != EXTENT_CHANGESET_BYTES_ONLY;
217 }
218 
219 static inline struct extent_changeset *extent_changeset_alloc(void)
220 {
221 	struct extent_changeset *ret;
222 
223 	ret = kmalloc_obj(*ret);
224 	if (!ret)
225 		return NULL;
226 
227 	extent_changeset_init(ret);
228 	return ret;
229 }
230 
231 static inline void extent_changeset_prealloc(struct extent_changeset *changeset, gfp_t gfp_mask)
232 {
233 	ASSERT(extent_changeset_tracks_ranges(changeset));
234 	ulist_prealloc(&changeset->range_changed, gfp_mask);
235 }
236 
237 static inline void extent_changeset_release(struct extent_changeset *changeset)
238 {
239 	if (!changeset)
240 		return;
241 	changeset->bytes_changed = 0;
242 	if (extent_changeset_tracks_ranges(changeset))
243 		ulist_release(&changeset->range_changed);
244 }
245 
246 static inline void extent_changeset_free(struct extent_changeset *changeset)
247 {
248 	if (!changeset)
249 		return;
250 	extent_changeset_release(changeset);
251 	kfree(changeset);
252 }
253 
254 bool try_release_extent_mapping(struct folio *folio, gfp_t mask);
255 int try_release_extent_buffer(struct folio *folio);
256 
257 int btrfs_read_folio(struct file *file, struct folio *folio);
258 void extent_write_locked_range(struct inode *inode, const struct folio *locked_folio,
259 			       u64 start, u64 end, struct writeback_control *wbc,
260 			       bool pages_dirty);
261 int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc);
262 int btree_writepages(struct address_space *mapping, struct writeback_control *wbc);
263 void btrfs_btree_wait_writeback_range(struct btrfs_fs_info *fs_info, u64 start, u64 end);
264 void btrfs_readahead(struct readahead_control *rac);
265 int set_folio_extent_mapped(struct folio *folio);
266 void clear_folio_extent_mapped(struct folio *folio);
267 
268 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
269 					  u64 start, u64 owner_root, int level);
270 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
271 						u64 start);
272 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src);
273 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
274 					 u64 start);
275 void free_extent_buffer(struct extent_buffer *eb);
276 void free_extent_buffer_stale(struct extent_buffer *eb);
277 int read_extent_buffer_pages(struct extent_buffer *eb, int mirror_num,
278 			     const struct btrfs_tree_parent_check *parent_check);
279 int read_extent_buffer_pages_nowait(struct extent_buffer *eb, int mirror_num,
280 				    const struct btrfs_tree_parent_check *parent_check);
281 
282 static inline void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
283 {
284 	wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
285 		       TASK_UNINTERRUPTIBLE);
286 }
287 
288 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
289 				u64 bytenr, u64 owner_root, u64 gen, int level,
290 				const struct btrfs_key *first_key);
291 void btrfs_readahead_node_child(struct extent_buffer *node, int slot);
292 
293 /* Note: this can be used in for loops without caching the value in a variable. */
294 static inline int __pure num_extent_pages(const struct extent_buffer *eb)
295 {
296 	/*
297 	 * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to
298 	 * sectorsize, it's just eb->len >> PAGE_SHIFT.
299 	 *
300 	 * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE,
301 	 * thus have to ensure we get at least one page.
302 	 */
303 	return (eb->len >> PAGE_SHIFT) ?: 1;
304 }
305 
306 /*
307  * This can only be determined at runtime by checking eb::folios[0].
308  *
309  * As we can have either one large folio covering the whole eb
310  * (either nodesize <= PAGE_SIZE, or high order folio), or multiple
311  * single-paged folios.
312  *
313  * Note: this can be used in for loops without caching the value in a variable.
314  */
315 static inline int __pure num_extent_folios(const struct extent_buffer *eb)
316 {
317 	if (!eb->folios[0])
318 		return 0;
319 	if (folio_order(eb->folios[0]))
320 		return 1;
321 	return num_extent_pages(eb);
322 }
323 
324 static inline bool extent_buffer_uptodate(const struct extent_buffer *eb)
325 {
326 	return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
327 }
328 
329 static inline bool extent_buffer_under_io(const struct extent_buffer *eb)
330 {
331 	return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
332 		test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
333 }
334 
335 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
336 			 unsigned long start, unsigned long len);
337 void read_extent_buffer(const struct extent_buffer *eb, void *dst,
338 			unsigned long start,
339 			unsigned long len);
340 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
341 				       void __user *dst, unsigned long start,
342 				       unsigned long len);
343 void write_extent_buffer(const struct extent_buffer *eb, const void *src,
344 			 unsigned long start, unsigned long len);
345 
346 static inline void write_extent_buffer_chunk_tree_uuid(
347 		const struct extent_buffer *eb, const void *chunk_tree_uuid)
348 {
349 	write_extent_buffer(eb, chunk_tree_uuid,
350 			    offsetof(struct btrfs_header, chunk_tree_uuid),
351 			    BTRFS_FSID_SIZE);
352 }
353 
354 static inline void write_extent_buffer_fsid(const struct extent_buffer *eb,
355 					    const void *fsid)
356 {
357 	write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
358 			    BTRFS_FSID_SIZE);
359 }
360 
361 void copy_extent_buffer_full(const struct extent_buffer *dst,
362 			     const struct extent_buffer *src);
363 void copy_extent_buffer(const struct extent_buffer *dst,
364 			const struct extent_buffer *src,
365 			unsigned long dst_offset, unsigned long src_offset,
366 			unsigned long len);
367 void memcpy_extent_buffer(const struct extent_buffer *dst,
368 			  unsigned long dst_offset, unsigned long src_offset,
369 			  unsigned long len);
370 void memmove_extent_buffer(const struct extent_buffer *dst,
371 			   unsigned long dst_offset, unsigned long src_offset,
372 			   unsigned long len);
373 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
374 			   unsigned long len);
375 bool extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
376 			    unsigned long pos);
377 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
378 			      unsigned long pos, unsigned long len);
379 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
380 				unsigned long start, unsigned long pos,
381 				unsigned long len);
382 void set_extent_buffer_dirty(struct extent_buffer *eb);
383 void set_extent_buffer_uptodate(struct extent_buffer *eb);
384 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
385 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
386 				  const struct folio *locked_folio,
387 				  struct extent_state **cached,
388 				  u32 bits_to_clear, unsigned long page_ops);
389 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
390 			      struct extent_buffer *buf);
391 
392 static inline void btrfs_clear_folio_dirty_tag(struct folio *folio)
393 {
394 	ASSERT(!folio_test_dirty(folio));
395 	ASSERT(folio_test_locked(folio));
396 	ASSERT(folio->mapping);
397 	xa_lock_irq(&folio->mapping->i_pages);
398 	__xa_clear_mark(&folio->mapping->i_pages, folio->index,
399 			PAGECACHE_TAG_DIRTY);
400 	xa_unlock_irq(&folio->mapping->i_pages);
401 }
402 
403 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array, gfp_t gfp);
404 int btrfs_alloc_folio_array(unsigned int nr_folios, unsigned int order,
405 			    struct folio **folio_array, gfp_t gfp);
406 
407 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
408 bool find_lock_delalloc_range(struct inode *inode,
409 			      struct folio *locked_folio, u64 *start,
410 			     u64 *end);
411 #endif
412 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
413 					       u64 start);
414 
415 #ifdef CONFIG_BTRFS_DEBUG
416 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info);
417 #else
418 #define btrfs_extent_buffer_leak_debug_check(fs_info)	do {} while (0)
419 #endif
420 
421 void btrfs_inhibit_eb_writeback(struct btrfs_trans_handle *trans,
422 			       struct extent_buffer *eb);
423 void btrfs_uninhibit_all_eb_writeback(struct btrfs_trans_handle *trans);
424 
425 #endif
426