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 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 82 int __init extent_buffer_init_cachep(void); 83 void __cold extent_buffer_free_cachep(void); 84 85 #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE) 86 struct 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 /* Inhibit WB_SYNC_NONE writeback when > 0. */ 103 atomic_t writeback_inhibitors; 104 /* >= 0 if eb belongs to a log tree, -1 otherwise */ 105 s8 log_index; 106 u8 folio_shift; 107 struct rcu_head rcu_head; 108 109 struct rw_semaphore lock; 110 111 /* 112 * Pointers to all the folios of the extent buffer. 113 * 114 * For now the folio is always order 0 (aka, a single page). 115 */ 116 struct folio *folios[INLINE_EXTENT_BUFFER_PAGES]; 117 #ifdef CONFIG_BTRFS_DEBUG 118 struct list_head leak_list; 119 pid_t lock_owner; 120 #endif 121 }; 122 123 struct btrfs_eb_write_context { 124 struct writeback_control *wbc; 125 struct extent_buffer *eb; 126 /* Block group @eb resides in. Only used for zoned mode. */ 127 struct btrfs_block_group *zoned_bg; 128 }; 129 130 static inline unsigned long offset_in_eb_folio(const struct extent_buffer *eb, 131 u64 start) 132 { 133 ASSERT(eb->folio_size); 134 return start & (eb->folio_size - 1); 135 } 136 137 /* 138 * Get the correct offset inside the page of extent buffer. 139 * 140 * @eb: target extent buffer 141 * @start: offset inside the extent buffer 142 * 143 * Will handle both sectorsize == PAGE_SIZE and sectorsize < PAGE_SIZE cases. 144 */ 145 static inline size_t get_eb_offset_in_folio(const struct extent_buffer *eb, 146 unsigned long offset) 147 { 148 /* 149 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case 150 * 1.1) One large folio covering the whole eb 151 * The eb->start is aligned to folio size, thus adding it 152 * won't cause any difference. 153 * 1.2) Several page sized folios 154 * The eb->start is aligned to folio (page) size, thus 155 * adding it won't cause any difference. 156 * 157 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case 158 * In this case there would only be one page sized folio, and there 159 * may be several different extent buffers in the page/folio. 160 * We need to add eb->start to properly access the offset inside 161 * that eb. 162 */ 163 return offset_in_folio(eb->folios[0], offset + eb->start); 164 } 165 166 static inline unsigned long get_eb_folio_index(const struct extent_buffer *eb, 167 unsigned long offset) 168 { 169 /* 170 * 1) sectorsize == PAGE_SIZE and nodesize >= PAGE_SIZE case 171 * 1.1) One large folio covering the whole eb. 172 * the folio_shift would be large enough to always make us 173 * return 0 as index. 174 * 1.2) Several page sized folios 175 * The folio_shift would be PAGE_SHIFT, giving us the correct 176 * index. 177 * 178 * 2) sectorsize < PAGE_SIZE and nodesize < PAGE_SIZE case 179 * The folio would only be page sized, and always give us 0 as index. 180 */ 181 return offset >> eb->folio_shift; 182 } 183 184 /* 185 * Structure to record how many bytes and which ranges are set/cleared 186 */ 187 struct extent_changeset { 188 /* How many bytes are set/cleared in this operation */ 189 u64 bytes_changed; 190 191 /* Changed ranges */ 192 struct ulist range_changed; 193 }; 194 195 static inline void extent_changeset_init(struct extent_changeset *changeset) 196 { 197 changeset->bytes_changed = 0; 198 ulist_init(&changeset->range_changed); 199 } 200 201 /* 202 * Sentinel value for range_changed.prealloc indicating that the changeset 203 * only tracks bytes_changed and does not record individual ranges. This 204 * avoids GFP_ATOMIC allocations inside add_extent_changeset() when the 205 * caller doesn't need to iterate the changed ranges afterwards. 206 */ 207 #define EXTENT_CHANGESET_BYTES_ONLY ((struct ulist_node *)1) 208 209 static inline void extent_changeset_init_bytes_only(struct extent_changeset *changeset) 210 { 211 changeset->bytes_changed = 0; 212 changeset->range_changed.prealloc = EXTENT_CHANGESET_BYTES_ONLY; 213 } 214 215 static inline bool extent_changeset_tracks_ranges(const struct extent_changeset *changeset) 216 { 217 return changeset->range_changed.prealloc != EXTENT_CHANGESET_BYTES_ONLY; 218 } 219 220 static inline struct extent_changeset *extent_changeset_alloc(void) 221 { 222 struct extent_changeset *ret; 223 224 ret = kmalloc_obj(*ret); 225 if (!ret) 226 return NULL; 227 228 extent_changeset_init(ret); 229 return ret; 230 } 231 232 static inline void extent_changeset_prealloc(struct extent_changeset *changeset, gfp_t gfp_mask) 233 { 234 ASSERT(extent_changeset_tracks_ranges(changeset)); 235 ulist_prealloc(&changeset->range_changed, gfp_mask); 236 } 237 238 static inline void extent_changeset_release(struct extent_changeset *changeset) 239 { 240 if (!changeset) 241 return; 242 changeset->bytes_changed = 0; 243 if (extent_changeset_tracks_ranges(changeset)) 244 ulist_release(&changeset->range_changed); 245 } 246 247 static inline void extent_changeset_free(struct extent_changeset *changeset) 248 { 249 if (!changeset) 250 return; 251 extent_changeset_release(changeset); 252 kfree(changeset); 253 } 254 255 bool try_release_extent_mapping(struct folio *folio, gfp_t mask); 256 int try_release_extent_buffer(struct folio *folio); 257 258 int btrfs_read_folio(struct file *file, struct folio *folio); 259 void extent_write_locked_range(struct inode *inode, const struct folio *locked_folio, 260 u64 start, u64 end, struct writeback_control *wbc, 261 bool pages_dirty); 262 int btrfs_writepages(struct address_space *mapping, struct writeback_control *wbc); 263 int btree_writepages(struct address_space *mapping, struct writeback_control *wbc); 264 void btrfs_btree_wait_writeback_range(struct btrfs_fs_info *fs_info, u64 start, u64 end); 265 void btrfs_readahead(struct readahead_control *rac); 266 int set_folio_extent_mapped(struct folio *folio); 267 void clear_folio_extent_mapped(struct folio *folio); 268 269 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, 270 u64 start, u64 owner_root, int level); 271 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, 272 u64 start); 273 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src); 274 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, 275 u64 start); 276 void free_extent_buffer(struct extent_buffer *eb); 277 void free_extent_buffer_stale(struct extent_buffer *eb); 278 int read_extent_buffer_pages(struct extent_buffer *eb, int mirror_num, 279 const struct btrfs_tree_parent_check *parent_check); 280 int read_extent_buffer_pages_nowait(struct extent_buffer *eb, int mirror_num, 281 const struct btrfs_tree_parent_check *parent_check); 282 283 static inline void wait_on_extent_buffer_writeback(struct extent_buffer *eb) 284 { 285 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, 286 TASK_UNINTERRUPTIBLE); 287 } 288 289 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, 290 u64 bytenr, u64 owner_root, u64 gen, int level); 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 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, 330 unsigned long start, unsigned long len); 331 void read_extent_buffer(const struct extent_buffer *eb, void *dst, 332 unsigned long start, 333 unsigned long len); 334 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, 335 void __user *dst, unsigned long start, 336 unsigned long len); 337 void write_extent_buffer(const struct extent_buffer *eb, const void *src, 338 unsigned long start, unsigned long len); 339 340 static inline void write_extent_buffer_chunk_tree_uuid( 341 const struct extent_buffer *eb, const void *chunk_tree_uuid) 342 { 343 write_extent_buffer(eb, chunk_tree_uuid, 344 offsetof(struct btrfs_header, chunk_tree_uuid), 345 BTRFS_FSID_SIZE); 346 } 347 348 static inline void write_extent_buffer_fsid(const struct extent_buffer *eb, 349 const void *fsid) 350 { 351 write_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 352 BTRFS_FSID_SIZE); 353 } 354 355 void copy_extent_buffer_full(const struct extent_buffer *dst, 356 const struct extent_buffer *src); 357 void copy_extent_buffer(const struct extent_buffer *dst, 358 const struct extent_buffer *src, 359 unsigned long dst_offset, unsigned long src_offset, 360 unsigned long len); 361 void memcpy_extent_buffer(const struct extent_buffer *dst, 362 unsigned long dst_offset, unsigned long src_offset, 363 unsigned long len); 364 void memmove_extent_buffer(const struct extent_buffer *dst, 365 unsigned long dst_offset, unsigned long src_offset, 366 unsigned long len); 367 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, 368 unsigned long len); 369 bool extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, 370 unsigned long pos); 371 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, 372 unsigned long pos, unsigned long len); 373 void extent_buffer_bitmap_clear(const struct extent_buffer *eb, 374 unsigned long start, unsigned long pos, 375 unsigned long len); 376 void set_extent_buffer_dirty(struct extent_buffer *eb); 377 void set_extent_buffer_uptodate(struct extent_buffer *eb); 378 void clear_extent_buffer_uptodate(struct extent_buffer *eb); 379 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, 380 const struct folio *locked_folio, 381 struct extent_state **cached, 382 u32 bits_to_clear, unsigned long page_ops); 383 int extent_invalidate_folio(struct extent_io_tree *tree, 384 struct folio *folio, size_t offset); 385 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans, 386 struct extent_buffer *buf); 387 388 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array, 389 bool nofail); 390 int btrfs_alloc_folio_array(unsigned int nr_folios, unsigned int order, 391 struct folio **folio_array); 392 393 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 394 bool find_lock_delalloc_range(struct inode *inode, 395 struct folio *locked_folio, u64 *start, 396 u64 *end); 397 #endif 398 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, 399 u64 start); 400 401 #ifdef CONFIG_BTRFS_DEBUG 402 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info); 403 #else 404 #define btrfs_extent_buffer_leak_debug_check(fs_info) do {} while (0) 405 #endif 406 407 void btrfs_inhibit_eb_writeback(struct btrfs_trans_handle *trans, 408 struct extent_buffer *eb); 409 void btrfs_uninhibit_all_eb_writeback(struct btrfs_trans_handle *trans); 410 411 #endif 412