| 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_TREE_REF, |
| 42 | EXTENT_BUFFER_STALE, |
| 43 | EXTENT_BUFFER_WRITEBACK, |
| 44 | EXTENT_BUFFER_UNMAPPED, |
| 45 | /* write IO error */ |
| 46 | EXTENT_BUFFER_WRITE_ERR, |
| 47 | /* Indicate the extent buffer is written zeroed out (for zoned) */ |
| 48 | EXTENT_BUFFER_ZONED_ZEROOUT, |
| 49 | /* Indicate that extent buffer pages a being read */ |
| 50 | EXTENT_BUFFER_READING, |
| 51 | }; |
| 52 | |
| 53 | /* these are flags for __process_pages_contig */ |
| 54 | enum { |
| 55 | ENUM_BIT(PAGE_UNLOCK), |
| 56 | /* Page starts writeback, clear dirty bit and set writeback bit */ |
| 57 | ENUM_BIT(PAGE_START_WRITEBACK), |
| 58 | ENUM_BIT(PAGE_END_WRITEBACK), |
| 59 | ENUM_BIT(PAGE_SET_ORDERED), |
| 60 | }; |
| 61 | |
| 62 | /* |
| 63 | * Folio private values. Every page that is controlled by the extent map has |
| 64 | * folio private set to this value. |
| 65 | */ |
| 66 | #define EXTENT_FOLIO_PRIVATE 1 |
| 67 | |
| 68 | /* |
| 69 | * The extent buffer bitmap operations are done with byte granularity instead of |
| 70 | * word granularity for two reasons: |
| 71 | * 1. The bitmaps must be little-endian on disk. |
| 72 | * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a |
| 73 | * single word in a bitmap may straddle two pages in the extent buffer. |
| 74 | */ |
| 75 | #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE) |
| 76 | #define BYTE_MASK ((1U << BITS_PER_BYTE) - 1) |
| 77 | #define BITMAP_FIRST_BYTE_MASK(start) \ |
| 78 | ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK) |
| 79 | #define BITMAP_LAST_BYTE_MASK(nbits) \ |
| 80 | (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1))) |
| 81 | |
| 82 | |
| 83 | int __init extent_buffer_init_cachep(void); |
| 84 | void __cold extent_buffer_free_cachep(void); |
| 85 | |
| 86 | #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE) |
| 87 | struct extent_buffer { |
| 88 | u64 start; |
| 89 | u32 len; |
| 90 | u32 folio_size; |
| 91 | unsigned long bflags; |
| 92 | struct btrfs_fs_info *fs_info; |
| 93 | |
| 94 | /* |
| 95 | * The address where the eb can be accessed without any cross-page handling. |
| 96 | * This can be NULL if not possible. |
| 97 | */ |
| 98 | void *addr; |
| 99 | |
| 100 | spinlock_t refs_lock; |
| 101 | atomic_t refs; |
| 102 | int read_mirror; |
| 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 | static inline struct extent_changeset *extent_changeset_alloc(void) |
| 201 | { |
| 202 | struct extent_changeset *ret; |
| 203 | |
| 204 | ret = kmalloc(sizeof(*ret), GFP_KERNEL); |
| 205 | if (!ret) |
| 206 | return NULL; |
| 207 | |
| 208 | extent_changeset_init(ret); |
| 209 | return ret; |
| 210 | } |
| 211 | |
| 212 | static inline void extent_changeset_prealloc(struct extent_changeset *changeset, gfp_t gfp_mask) |
| 213 | { |
| 214 | ulist_prealloc(&changeset->range_changed, gfp_mask); |
| 215 | } |
| 216 | |
| 217 | static 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 | |
| 225 | static 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 | |
| 233 | bool try_release_extent_mapping(struct folio *folio, gfp_t mask); |
| 234 | int try_release_extent_buffer(struct folio *folio); |
| 235 | |
| 236 | int btrfs_read_folio(struct file *file, struct folio *folio); |
| 237 | void extent_write_locked_range(struct inode *inode, const struct folio *locked_folio, |
| 238 | u64 start, u64 end, struct writeback_control *wbc, |
| 239 | bool pages_dirty); |
| 240 | int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc); |
| 241 | int btree_write_cache_pages(struct address_space *mapping, |
| 242 | struct writeback_control *wbc); |
| 243 | void btrfs_btree_wait_writeback_range(struct btrfs_fs_info *fs_info, u64 start, u64 end); |
| 244 | void btrfs_readahead(struct readahead_control *rac); |
| 245 | int set_folio_extent_mapped(struct folio *folio); |
| 246 | void clear_folio_extent_mapped(struct folio *folio); |
| 247 | |
| 248 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
| 249 | u64 start, u64 owner_root, int level); |
| 250 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
| 251 | u64 start); |
| 252 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src); |
| 253 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
| 254 | u64 start); |
| 255 | void free_extent_buffer(struct extent_buffer *eb); |
| 256 | void free_extent_buffer_stale(struct extent_buffer *eb); |
| 257 | int read_extent_buffer_pages(struct extent_buffer *eb, int mirror_num, |
| 258 | const struct btrfs_tree_parent_check *parent_check); |
| 259 | int read_extent_buffer_pages_nowait(struct extent_buffer *eb, int mirror_num, |
| 260 | const struct btrfs_tree_parent_check *parent_check); |
| 261 | |
| 262 | static inline void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
| 263 | { |
| 264 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
| 265 | TASK_UNINTERRUPTIBLE); |
| 266 | } |
| 267 | |
| 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); |
| 271 | |
| 272 | /* Note: this can be used in for loops without caching the value in a variable. */ |
| 273 | static inline int __pure num_extent_pages(const struct extent_buffer *eb) |
| 274 | { |
| 275 | /* |
| 276 | * For sectorsize == PAGE_SIZE case, since nodesize is always aligned to |
| 277 | * sectorsize, it's just eb->len >> PAGE_SHIFT. |
| 278 | * |
| 279 | * For sectorsize < PAGE_SIZE case, we could have nodesize < PAGE_SIZE, |
| 280 | * thus have to ensure we get at least one page. |
| 281 | */ |
| 282 | return (eb->len >> PAGE_SHIFT) ?: 1; |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * This can only be determined at runtime by checking eb::folios[0]. |
| 287 | * |
| 288 | * As we can have either one large folio covering the whole eb |
| 289 | * (either nodesize <= PAGE_SIZE, or high order folio), or multiple |
| 290 | * single-paged folios. |
| 291 | * |
| 292 | * Note: this can be used in for loops without caching the value in a variable. |
| 293 | */ |
| 294 | static inline int __pure num_extent_folios(const struct extent_buffer *eb) |
| 295 | { |
| 296 | if (!eb->folios[0]) |
| 297 | return 0; |
| 298 | if (folio_order(eb->folios[0])) |
| 299 | return 1; |
| 300 | return num_extent_pages(eb); |
| 301 | } |
| 302 | |
| 303 | static inline int extent_buffer_uptodate(const struct extent_buffer *eb) |
| 304 | { |
| 305 | return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
| 306 | } |
| 307 | |
| 308 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
| 309 | unsigned long start, unsigned long len); |
| 310 | void read_extent_buffer(const struct extent_buffer *eb, void *dst, |
| 311 | unsigned long start, |
| 312 | unsigned long len); |
| 313 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
| 314 | void __user *dst, unsigned long start, |
| 315 | unsigned long len); |
| 316 | void write_extent_buffer(const struct extent_buffer *eb, const void *src, |
| 317 | unsigned long start, unsigned long len); |
| 318 | |
| 319 | static inline void write_extent_buffer_chunk_tree_uuid( |
| 320 | const struct extent_buffer *eb, const void *chunk_tree_uuid) |
| 321 | { |
| 322 | write_extent_buffer(eb, chunk_tree_uuid, |
| 323 | offsetof(struct btrfs_header, chunk_tree_uuid), |
| 324 | BTRFS_FSID_SIZE); |
| 325 | } |
| 326 | |
| 327 | static inline void write_extent_buffer_fsid(const struct extent_buffer *eb, |
| 328 | const void *fsid) |
| 329 | { |
| 330 | write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), |
| 331 | BTRFS_FSID_SIZE); |
| 332 | } |
| 333 | |
| 334 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
| 335 | const struct extent_buffer *src); |
| 336 | void copy_extent_buffer(const struct extent_buffer *dst, |
| 337 | const struct extent_buffer *src, |
| 338 | unsigned long dst_offset, unsigned long src_offset, |
| 339 | unsigned long len); |
| 340 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
| 341 | unsigned long dst_offset, unsigned long src_offset, |
| 342 | unsigned long len); |
| 343 | void memmove_extent_buffer(const struct extent_buffer *dst, |
| 344 | unsigned long dst_offset, unsigned long src_offset, |
| 345 | unsigned long len); |
| 346 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
| 347 | unsigned long len); |
| 348 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
| 349 | unsigned long pos); |
| 350 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
| 351 | unsigned long pos, unsigned long len); |
| 352 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
| 353 | unsigned long start, unsigned long pos, |
| 354 | unsigned long len); |
| 355 | void set_extent_buffer_dirty(struct extent_buffer *eb); |
| 356 | void set_extent_buffer_uptodate(struct extent_buffer *eb); |
| 357 | void clear_extent_buffer_uptodate(struct extent_buffer *eb); |
| 358 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
| 359 | const struct folio *locked_folio, |
| 360 | struct extent_state **cached, |
| 361 | u32 bits_to_clear, unsigned long page_ops); |
| 362 | int extent_invalidate_folio(struct extent_io_tree *tree, |
| 363 | struct folio *folio, size_t offset); |
| 364 | void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans, |
| 365 | struct extent_buffer *buf); |
| 366 | |
| 367 | int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array, |
| 368 | bool nofail); |
| 369 | int btrfs_alloc_folio_array(unsigned int nr_folios, struct folio **folio_array); |
| 370 | |
| 371 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
| 372 | bool find_lock_delalloc_range(struct inode *inode, |
| 373 | struct folio *locked_folio, u64 *start, |
| 374 | u64 *end); |
| 375 | #endif |
| 376 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, |
| 377 | u64 start); |
| 378 | |
| 379 | #ifdef CONFIG_BTRFS_DEBUG |
| 380 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info); |
| 381 | #else |
| 382 | #define btrfs_extent_buffer_leak_debug_check(fs_info) do {} while (0) |
| 383 | #endif |
| 384 | |
| 385 | #endif |