extent_io.h

  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
 19struct page;
 20struct file;
 21struct folio;
 22struct inode;
 23struct fiemap_extent_info;
 24struct readahead_control;
 25struct address_space;
 26struct writeback_control;
 27struct extent_io_tree;
 28struct extent_map_tree;
 29struct extent_state;
 30struct btrfs_block_group;
 31struct btrfs_fs_info;
 32struct btrfs_inode;
 33struct btrfs_root;
 34struct btrfs_trans_handle;
 35struct btrfs_tree_parent_check;
 36
 37enum {
 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 */
 53enum {
 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	ENUM_BIT(PAGE_SET_ORDERED),
 59};
 60
 61/*
 62 * Folio private values.  Every page that is controlled by the extent map has
 63 * folio private set to this value.
 64 */
 65#define EXTENT_FOLIO_PRIVATE			1
 66
 67/*
 68 * The extent buffer bitmap operations are done with byte granularity instead of
 69 * word granularity for two reasons:
 70 * 1. The bitmaps must be little-endian on disk.
 71 * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a
 72 *    single word in a bitmap may straddle two pages in the extent buffer.
 73 */
 74#define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
 75#define BYTE_MASK ((1U << BITS_PER_BYTE) - 1)
 76#define BITMAP_FIRST_BYTE_MASK(start) \
 77	((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
 78#define BITMAP_LAST_BYTE_MASK(nbits) \
 79	(BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
 80
 81
 82int __init extent_buffer_init_cachep(void);
 83void __cold extent_buffer_free_cachep(void);
 84
 85#define INLINE_EXTENT_BUFFER_PAGES     (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE)
 86struct extent_buffer {
 87	u64 start;
 88	u32 len;
 89	u32 folio_size;
 90	unsigned long bflags;
 91	struct btrfs_fs_info *fs_info;
 92
 93	/*
 94	 * The address where the eb can be accessed without any cross-page handling.
 95	 * This can be NULL if not possible.
 96	 */
 97	void *addr;
 98
 99	spinlock_t refs_lock;
100	refcount_t refs;
101	int read_mirror;
102	/* >= 0 if eb belongs to a log tree, -1 otherwise */
103	s8 log_index;
104	u8 folio_shift;
105	struct rcu_head rcu_head;
106
107	struct rw_semaphore lock;
108
109	/*
110	 * Pointers to all the folios of the extent buffer.
111	 *
112	 * For now the folio is always order 0 (aka, a single page).
113	 */
114	struct folio *folios[INLINE_EXTENT_BUFFER_PAGES];
115#ifdef CONFIG_BTRFS_DEBUG
116	struct list_head leak_list;
117	pid_t lock_owner;
118#endif
119};
120
121struct btrfs_eb_write_context {
122	struct writeback_control *wbc;
123	struct extent_buffer *eb;
124	/* Block group @eb resides in. Only used for zoned mode. */
125	struct btrfs_block_group *zoned_bg;
126};
127
128static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb,
129					       u64 start)
130{
131	ASSERT(eb->folio_size);
132	return start & (eb->folio_size - 1);
133}
134
135/*
136 * Get the correct offset inside the page of extent buffer.
137 *
138 * @eb:		target extent buffer
139 * @start:	offset inside the extent buffer
140 *
141 * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases.
142 */
143static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb,
144					    unsigned long offset)
145{
146	/*
147	 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
148	 *    1.1) One large folio covering the whole eb
149	 *	   The eb->start is aligned to folio size, thus adding it
150	 *	   won't cause any difference.
151	 *    1.2) Several page sized folios
152	 *	   The eb->start is aligned to folio (page) size, thus
153	 *	   adding it won't cause any difference.
154	 *
155	 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
156	 *    In this case there would only be one page sized folio, and there
157	 *    may be several different extent buffers in the page/folio.
158	 *    We need to add eb->start to properly access the offset inside
159	 *    that eb.
160	 */
161	return offset_in_folio(eb->folios[0], offset + eb->start);
162}
163
164static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb,
165					       unsigned long offset)
166{
167	/*
168	 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
169	 *    1.1) One large folio covering the whole eb.
170	 *	   the folio_shift would be large enough to always make us
171	 *	   return 0 as index.
172	 *    1.2) Several page sized folios
173	 *         The folio_shift would be PAGE_SHIFT, giving us the correct
174	 *         index.
175	 *
176	 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
177	 *    The folio would only be page sized, and always give us 0 as index.
178	 */
179	return offset >> eb->folio_shift;
180}
181
182/*
183 * Structure to record how many bytes and which ranges are set/cleared
184 */
185struct extent_changeset {
186	/* How many bytes are set/cleared in this operation */
187	u64 bytes_changed;
188
189	/* Changed ranges */
190	struct ulist range_changed;
191};
192
193static inline void extent_changeset_init(struct extent_changeset *changeset)
194{
195	changeset->bytes_changed = 0;
196	ulist_init(&changeset->range_changed);
197}
198
199static inline struct extent_changeset *extent_changeset_alloc(void)
200{
201	struct extent_changeset *ret;
202
203	ret = kmalloc(sizeof(*ret), GFP_KERNEL);
204	if (!ret)
205		return NULL;
206
207	extent_changeset_init(ret);
208	return ret;
209}
210
211static inline void extent_changeset_prealloc(struct extent_changeset *changeset, gfp_t gfp_mask)
212{
213	ulist_prealloc(&changeset->range_changed, gfp_mask);
214}
215
216static inline void extent_changeset_release(struct extent_changeset *changeset)
217{
218	if (!changeset)
219		return;
220	changeset->bytes_changed = 0;
221	ulist_release(&changeset->range_changed);
222}
223
224static inline void extent_changeset_free(struct extent_changeset *changeset)
225{
226	if (!changeset)
227		return;
228	extent_changeset_release(changeset);
229	kfree(changeset);
230}
231
232bool try_release_extent_mapping(struct folio *folio, gfp_t mask);
233int try_release_extent_buffer(struct folio *folio);
234
235int btrfs_read_folio(struct file *file, struct folio *folio);
236void extent_write_locked_range(struct inode *inode, const struct folio *locked_folio,
237			       u64 start, u64 end, struct writeback_control *wbc,
238			       bool pages_dirty);
239int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc);
240int btree_writepages(struct address_space *mapping, struct writeback_control *wbc);
241void btrfs_btree_wait_writeback_range(struct btrfs_fs_info *fs_info, u64 start, u64 end);
242void btrfs_readahead(struct readahead_control *rac);
243int set_folio_extent_mapped(struct folio *folio);
244void clear_folio_extent_mapped(struct folio *folio);
245
246struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
247					  u64 start, u64 owner_root, int level);
248struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
249						u64 start);
250struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src);
251struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
252					 u64 start);
253void free_extent_buffer(struct extent_buffer *eb);
254void free_extent_buffer_stale(struct extent_buffer *eb);
255int read_extent_buffer_pages(struct extent_buffer *eb, int mirror_num,
256			     const struct btrfs_tree_parent_check *parent_check);
257int read_extent_buffer_pages_nowait(struct extent_buffer *eb, int mirror_num,
258				    const struct btrfs_tree_parent_check *parent_check);
259
260static inline void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
261{
262	wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
263		       TASK_UNINTERRUPTIBLE);
264}
265
266void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
267				u64 bytenr, u64 owner_root, u64 gen, int level);
268void btrfs_readahead_node_child(struct extent_buffer *node, int slot);
269
270/* Note: this can be used in for loops without caching the value in a variable. */
271static inline int __pure num_extent_pages(const struct extent_buffer *eb)
272{
273	/*
274	 * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to
275	 * sectorsize, it's just eb->len >> PAGE_SHIFT.
276	 *
277	 * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE,
278	 * thus have to ensure we get at least one page.
279	 */
280	return (eb->len >> PAGE_SHIFT) ?: 1;
281}
282
283/*
284 * This can only be determined at runtime by checking eb::folios[0].
285 *
286 * As we can have either one large folio covering the whole eb
287 * (either nodesize <= PAGE_SIZE, or high order folio), or multiple
288 * single-paged folios.
289 *
290 * Note: this can be used in for loops without caching the value in a variable.
291 */
292static inline int __pure num_extent_folios(const struct extent_buffer *eb)
293{
294	if (!eb->folios[0])
295		return 0;
296	if (folio_order(eb->folios[0]))
297		return 1;
298	return num_extent_pages(eb);
299}
300
301static inline int extent_buffer_uptodate(const struct extent_buffer *eb)
302{
303	return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
304}
305
306int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
307			 unsigned long start, unsigned long len);
308void read_extent_buffer(const struct extent_buffer *eb, void *dst,
309			unsigned long start,
310			unsigned long len);
311int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
312				       void __user *dst, unsigned long start,
313				       unsigned long len);
314void write_extent_buffer(const struct extent_buffer *eb, const void *src,
315			 unsigned long start, unsigned long len);
316
317static 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
325static 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
332void copy_extent_buffer_full(const struct extent_buffer *dst,
333			     const struct extent_buffer *src);
334void 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);
338void memcpy_extent_buffer(const struct extent_buffer *dst,
339			  unsigned long dst_offset, unsigned long src_offset,
340			  unsigned long len);
341void memmove_extent_buffer(const struct extent_buffer *dst,
342			   unsigned long dst_offset, unsigned long src_offset,
343			   unsigned long len);
344void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
345			   unsigned long len);
346bool extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
347			    unsigned long pos);
348void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
349			      unsigned long pos, unsigned long len);
350void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
351				unsigned long start, unsigned long pos,
352				unsigned long len);
353void set_extent_buffer_dirty(struct extent_buffer *eb);
354void set_extent_buffer_uptodate(struct extent_buffer *eb);
355void clear_extent_buffer_uptodate(struct extent_buffer *eb);
356void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
357				  const struct folio *locked_folio,
358				  struct extent_state **cached,
359				  u32 bits_to_clear, unsigned long page_ops);
360int extent_invalidate_folio(struct extent_io_tree *tree,
361			    struct folio *folio, size_t offset);
362void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
363			      struct extent_buffer *buf);
364
365int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
366			   bool nofail);
367int btrfs_alloc_folio_array(unsigned int nr_folios, unsigned int order,
368			    struct folio **folio_array);
369
370#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
371bool find_lock_delalloc_range(struct inode *inode,
372			      struct folio *locked_folio, u64 *start,
373			     u64 *end);
374#endif
375struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
376					       u64 start);
377
378#ifdef CONFIG_BTRFS_DEBUG
379void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info);
380#else
381#define btrfs_extent_buffer_leak_debug_check(fs_info)	do {} while (0)
382#endif
383
384#endif