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
20struct page;
21struct file;
22struct folio;
23struct inode;
24struct fiemap_extent_info;
25struct readahead_control;
26struct address_space;
27struct writeback_control;
28struct extent_io_tree;
29struct extent_map_tree;
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_CORRUPT,
41	/* this got triggered by readahead */
42	EXTENT_BUFFER_READAHEAD,
43	EXTENT_BUFFER_TREE_REF,
44	EXTENT_BUFFER_STALE,
45	EXTENT_BUFFER_WRITEBACK,
46	/* read IO error */
47	EXTENT_BUFFER_READ_ERR,
48	EXTENT_BUFFER_UNMAPPED,
49	EXTENT_BUFFER_IN_TREE,
50	/* write IO error */
51	EXTENT_BUFFER_WRITE_ERR,
52	/* Indicate the extent buffer is written zeroed out (for zoned) */
53	EXTENT_BUFFER_ZONED_ZEROOUT,
54	/* Indicate that extent buffer pages a being read */
55	EXTENT_BUFFER_READING,
56};
57
58/* these are flags for __process_pages_contig */
59enum {
60	ENUM_BIT(PAGE_UNLOCK),
61	/* Page starts writeback, clear dirty bit and set writeback bit */
62	ENUM_BIT(PAGE_START_WRITEBACK),
63	ENUM_BIT(PAGE_END_WRITEBACK),
64	ENUM_BIT(PAGE_SET_ORDERED),
65};
66
67/*
68 * Folio private values.  Every page that is controlled by the extent map has
69 * folio private set to this value.
70 */
71#define EXTENT_FOLIO_PRIVATE			1
72
73/*
74 * The extent buffer bitmap operations are done with byte granularity instead of
75 * word granularity for two reasons:
76 * 1. The bitmaps must be little-endian on disk.
77 * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a
78 *    single word in a bitmap may straddle two pages in the extent buffer.
79 */
80#define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
81#define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
82#define BITMAP_FIRST_BYTE_MASK(start) \
83	((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
84#define BITMAP_LAST_BYTE_MASK(nbits) \
85	(BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
86
87
88int __init extent_buffer_init_cachep(void);
89void __cold extent_buffer_free_cachep(void);
90
91#define INLINE_EXTENT_BUFFER_PAGES     (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE)
92struct extent_buffer {
93	u64 start;
94	u32 len;
95	u32 folio_size;
96	unsigned long bflags;
97	struct btrfs_fs_info *fs_info;
98
99	/*
100	 * The address where the eb can be accessed without any cross-page handling.
101	 * This can be NULL if not possible.
102	 */
103	void *addr;
104
105	spinlock_t refs_lock;
106	atomic_t refs;
107	int read_mirror;
108	/* >= 0 if eb belongs to a log tree, -1 otherwise */
109	s8 log_index;
110	u8 folio_shift;
111	struct rcu_head rcu_head;
112
113	struct rw_semaphore lock;
114
115	/*
116	 * Pointers to all the folios of the extent buffer.
117	 *
118	 * For now the folio is always order 0 (aka, a single page).
119	 */
120	struct folio *folios[INLINE_EXTENT_BUFFER_PAGES];
121#ifdef CONFIG_BTRFS_DEBUG
122	struct list_head leak_list;
123	pid_t lock_owner;
124#endif
125};
126
127struct btrfs_eb_write_context {
128	struct writeback_control *wbc;
129	struct extent_buffer *eb;
130	/* Block group @eb resides in. Only used for zoned mode. */
131	struct btrfs_block_group *zoned_bg;
132};
133
134static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb,
135					       u64 start)
136{
137	ASSERT(eb->folio_size);
138	return start & (eb->folio_size - 1);
139}
140
141/*
142 * Get the correct offset inside the page of extent buffer.
143 *
144 * @eb:		target extent buffer
145 * @start:	offset inside the extent buffer
146 *
147 * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases.
148 */
149static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb,
150					    unsigned long offset)
151{
152	/*
153	 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
154	 *    1.1) One large folio covering the whole eb
155	 *	   The eb->start is aligned to folio size, thus adding it
156	 *	   won't cause any difference.
157	 *    1.2) Several page sized folios
158	 *	   The eb->start is aligned to folio (page) size, thus
159	 *	   adding it won't cause any difference.
160	 *
161	 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
162	 *    In this case there would only be one page sized folio, and there
163	 *    may be several different extent buffers in the page/folio.
164	 *    We need to add eb->start to properly access the offset inside
165	 *    that eb.
166	 */
167	return offset_in_folio(eb->folios[0], offset + eb->start);
168}
169
170static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb,
171					       unsigned long offset)
172{
173	/*
174	 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case
175	 *    1.1) One large folio covering the whole eb.
176	 *	   the folio_shift would be large enough to always make us
177	 *	   return 0 as index.
178	 *    1.2) Several page sized folios
179	 *         The folio_shift would be PAGE_SHIFT, giving us the correct
180	 *         index.
181	 *
182	 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case
183	 *    The folio would only be page sized, and always give us 0 as index.
184	 */
185	return offset >> eb->folio_shift;
186}
187
188/*
189 * Structure to record how many bytes and which ranges are set/cleared
190 */
191struct extent_changeset {
192	/* How many bytes are set/cleared in this operation */
193	u64 bytes_changed;
194
195	/* Changed ranges */
196	struct ulist range_changed;
197};
198
199static inline void extent_changeset_init(struct extent_changeset *changeset)
200{
201	changeset->bytes_changed = 0;
202	ulist_init(&changeset->range_changed);
203}
204
205static inline struct extent_changeset *extent_changeset_alloc(void)
206{
207	struct extent_changeset *ret;
208
209	ret = kmalloc(sizeof(*ret), GFP_KERNEL);
210	if (!ret)
211		return NULL;
212
213	extent_changeset_init(ret);
214	return ret;
215}
216
217static inline void extent_changeset_release(struct extent_changeset *changeset)
218{
219	if (!changeset)
220		return;
221	changeset->bytes_changed = 0;
222	ulist_release(&changeset->range_changed);
223}
224
225static inline void extent_changeset_free(struct extent_changeset *changeset)
226{
227	if (!changeset)
228		return;
229	extent_changeset_release(changeset);
230	kfree(changeset);
231}
232
233int try_release_extent_mapping(struct page *page, gfp_t mask);
234int try_release_extent_buffer(struct page *page);
235
236int btrfs_read_folio(struct file *file, struct folio *folio);
237void extent_write_locked_range(struct inode *inode, struct page *locked_page,
238			       u64 start, u64 end, struct writeback_control *wbc,
239			       bool pages_dirty);
240int extent_writepages(struct address_space *mapping,
241		      struct writeback_control *wbc);
242int btree_write_cache_pages(struct address_space *mapping,
243			    struct writeback_control *wbc);
244void extent_readahead(struct readahead_control *rac);
245int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
246		  u64 start, u64 len);
247int set_folio_extent_mapped(struct folio *folio);
248int set_page_extent_mapped(struct page *page);
249void clear_page_extent_mapped(struct page *page);
250
251struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
252					  u64 start, u64 owner_root, int level);
253struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
254						  u64 start, unsigned long len);
255struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
256						u64 start);
257struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src);
258struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
259					 u64 start);
260void free_extent_buffer(struct extent_buffer *eb);
261void free_extent_buffer_stale(struct extent_buffer *eb);
262#define WAIT_NONE	0
263#define WAIT_COMPLETE	1
264#define WAIT_PAGE_LOCK	2
265int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
266			     struct btrfs_tree_parent_check *parent_check);
267void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
268void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
269				u64 bytenr, u64 owner_root, u64 gen, int level);
270void btrfs_readahead_node_child(struct extent_buffer *node, int slot);
271
272static inline int num_extent_pages(const struct extent_buffer *eb)
273{
274	/*
275	 * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to
276	 * sectorsize, it's just eb->len >> PAGE_SHIFT.
277	 *
278	 * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE,
279	 * thus have to ensure we get at least one page.
280	 */
281	return (eb->len >> PAGE_SHIFT) ?: 1;
282}
283
284/*
285 * This can only be determined at runtime by checking eb::folios[0].
286 *
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.
290 */
291static inline int num_extent_folios(const struct extent_buffer *eb)
292{
293	if (folio_order(eb->folios[0]))
294		return 1;
295	return num_extent_pages(eb);
296}
297
298static inline int extent_buffer_uptodate(const struct extent_buffer *eb)
299{
300	return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
301}
302
303int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
304			 unsigned long start, unsigned long len);
305void read_extent_buffer(const struct extent_buffer *eb, void *dst,
306			unsigned long start,
307			unsigned long len);
308int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
309				       void __user *dst, unsigned long start,
310				       unsigned long len);
311void write_extent_buffer(const struct extent_buffer *eb, const void *src,
312			 unsigned long start, unsigned long len);
313
314static inline void write_extent_buffer_chunk_tree_uuid(
315		const struct extent_buffer *eb, const void *chunk_tree_uuid)
316{
317	write_extent_buffer(eb, chunk_tree_uuid,
318			    offsetof(struct btrfs_header, chunk_tree_uuid),
319			    BTRFS_FSID_SIZE);
320}
321
322static inline void write_extent_buffer_fsid(const struct extent_buffer *eb,
323					    const void *fsid)
324{
325	write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
326			    BTRFS_FSID_SIZE);
327}
328
329void copy_extent_buffer_full(const struct extent_buffer *dst,
330			     const struct extent_buffer *src);
331void copy_extent_buffer(const struct extent_buffer *dst,
332			const struct extent_buffer *src,
333			unsigned long dst_offset, unsigned long src_offset,
334			unsigned long len);
335void memcpy_extent_buffer(const struct extent_buffer *dst,
336			  unsigned long dst_offset, unsigned long src_offset,
337			  unsigned long len);
338void memmove_extent_buffer(const struct extent_buffer *dst,
339			   unsigned long dst_offset, unsigned long src_offset,
340			   unsigned long len);
341void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
342			   unsigned long len);
343int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
344			   unsigned long pos);
345void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
346			      unsigned long pos, unsigned long len);
347void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
348				unsigned long start, unsigned long pos,
349				unsigned long len);
350void set_extent_buffer_dirty(struct extent_buffer *eb);
351void set_extent_buffer_uptodate(struct extent_buffer *eb);
352void clear_extent_buffer_uptodate(struct extent_buffer *eb);
353void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
354void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
355				  struct page *locked_page,
356				  u32 bits_to_clear, unsigned long page_ops);
357int extent_invalidate_folio(struct extent_io_tree *tree,
358			    struct folio *folio, size_t offset);
359void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
360			      struct extent_buffer *buf);
361
362int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
363			   gfp_t extra_gfp);
364
365#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
366bool find_lock_delalloc_range(struct inode *inode,
367			     struct page *locked_page, u64 *start,
368			     u64 *end);
369#endif
370struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
371					       u64 start);
372
373#ifdef CONFIG_BTRFS_DEBUG
374void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info);
375#else
376#define btrfs_extent_buffer_leak_debug_check(fs_info)	do {} while (0)
377#endif
378
379#endif
380