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 "ulist.h" 9 10 /* bits for the extent state */ 11 #define EXTENT_DIRTY (1U << 0) 12 #define EXTENT_WRITEBACK (1U << 1) 13 #define EXTENT_UPTODATE (1U << 2) 14 #define EXTENT_LOCKED (1U << 3) 15 #define EXTENT_NEW (1U << 4) 16 #define EXTENT_DELALLOC (1U << 5) 17 #define EXTENT_DEFRAG (1U << 6) 18 #define EXTENT_BOUNDARY (1U << 9) 19 #define EXTENT_NODATASUM (1U << 10) 20 #define EXTENT_CLEAR_META_RESV (1U << 11) 21 #define EXTENT_FIRST_DELALLOC (1U << 12) 22 #define EXTENT_NEED_WAIT (1U << 13) 23 #define EXTENT_DAMAGED (1U << 14) 24 #define EXTENT_NORESERVE (1U << 15) 25 #define EXTENT_QGROUP_RESERVED (1U << 16) 26 #define EXTENT_CLEAR_DATA_RESV (1U << 17) 27 #define EXTENT_DELALLOC_NEW (1U << 18) 28 #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) 29 #define EXTENT_DO_ACCOUNTING (EXTENT_CLEAR_META_RESV | \ 30 EXTENT_CLEAR_DATA_RESV) 31 #define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC) 32 33 /* 34 * flags for bio submission. The high bits indicate the compression 35 * type for this bio 36 */ 37 #define EXTENT_BIO_COMPRESSED 1 38 #define EXTENT_BIO_FLAG_SHIFT 16 39 40 /* these are bit numbers for test/set bit */ 41 #define EXTENT_BUFFER_UPTODATE 0 42 #define EXTENT_BUFFER_DIRTY 2 43 #define EXTENT_BUFFER_CORRUPT 3 44 #define EXTENT_BUFFER_READAHEAD 4 /* this got triggered by readahead */ 45 #define EXTENT_BUFFER_TREE_REF 5 46 #define EXTENT_BUFFER_STALE 6 47 #define EXTENT_BUFFER_WRITEBACK 7 48 #define EXTENT_BUFFER_READ_ERR 8 /* read IO error */ 49 #define EXTENT_BUFFER_UNMAPPED 9 50 #define EXTENT_BUFFER_IN_TREE 10 51 #define EXTENT_BUFFER_WRITE_ERR 11 /* write IO error */ 52 53 /* these are flags for __process_pages_contig */ 54 #define PAGE_UNLOCK (1 << 0) 55 #define PAGE_CLEAR_DIRTY (1 << 1) 56 #define PAGE_SET_WRITEBACK (1 << 2) 57 #define PAGE_END_WRITEBACK (1 << 3) 58 #define PAGE_SET_PRIVATE2 (1 << 4) 59 #define PAGE_SET_ERROR (1 << 5) 60 #define PAGE_LOCK (1 << 6) 61 62 /* 63 * page->private values. Every page that is controlled by the extent 64 * map has page->private set to one. 65 */ 66 #define EXTENT_PAGE_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 ((1 << 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 struct extent_state; 83 struct btrfs_root; 84 struct btrfs_inode; 85 struct btrfs_io_bio; 86 struct io_failure_record; 87 88 typedef blk_status_t (extent_submit_bio_hook_t)(void *private_data, struct bio *bio, 89 int mirror_num, unsigned long bio_flags, 90 u64 bio_offset); 91 92 typedef blk_status_t (extent_submit_bio_start_t)(void *private_data, 93 struct bio *bio, u64 bio_offset); 94 95 struct extent_io_ops { 96 /* 97 * The following callbacks must be allways defined, the function 98 * pointer will be called unconditionally. 99 */ 100 extent_submit_bio_hook_t *submit_bio_hook; 101 int (*readpage_end_io_hook)(struct btrfs_io_bio *io_bio, u64 phy_offset, 102 struct page *page, u64 start, u64 end, 103 int mirror); 104 int (*readpage_io_failed_hook)(struct page *page, int failed_mirror); 105 106 /* 107 * Optional hooks, called if the pointer is not NULL 108 */ 109 int (*fill_delalloc)(void *private_data, struct page *locked_page, 110 u64 start, u64 end, int *page_started, 111 unsigned long *nr_written, 112 struct writeback_control *wbc); 113 114 int (*writepage_start_hook)(struct page *page, u64 start, u64 end); 115 void (*writepage_end_io_hook)(struct page *page, u64 start, u64 end, 116 struct extent_state *state, int uptodate); 117 void (*set_bit_hook)(void *private_data, struct extent_state *state, 118 unsigned *bits); 119 void (*clear_bit_hook)(void *private_data, 120 struct extent_state *state, 121 unsigned *bits); 122 void (*merge_extent_hook)(void *private_data, 123 struct extent_state *new, 124 struct extent_state *other); 125 void (*split_extent_hook)(void *private_data, 126 struct extent_state *orig, u64 split); 127 void (*check_extent_io_range)(void *private_data, const char *caller, 128 u64 start, u64 end); 129 }; 130 131 struct extent_io_tree { 132 struct rb_root state; 133 void *private_data; 134 u64 dirty_bytes; 135 int track_uptodate; 136 spinlock_t lock; 137 const struct extent_io_ops *ops; 138 }; 139 140 struct extent_state { 141 u64 start; 142 u64 end; /* inclusive */ 143 struct rb_node rb_node; 144 145 /* ADD NEW ELEMENTS AFTER THIS */ 146 wait_queue_head_t wq; 147 refcount_t refs; 148 unsigned state; 149 150 struct io_failure_record *failrec; 151 152 #ifdef CONFIG_BTRFS_DEBUG 153 struct list_head leak_list; 154 #endif 155 }; 156 157 #define INLINE_EXTENT_BUFFER_PAGES 16 158 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE) 159 struct extent_buffer { 160 u64 start; 161 unsigned long len; 162 unsigned long bflags; 163 struct btrfs_fs_info *fs_info; 164 spinlock_t refs_lock; 165 atomic_t refs; 166 atomic_t io_pages; 167 int read_mirror; 168 struct rcu_head rcu_head; 169 pid_t lock_owner; 170 171 /* count of read lock holders on the extent buffer */ 172 atomic_t write_locks; 173 atomic_t read_locks; 174 atomic_t blocking_writers; 175 atomic_t blocking_readers; 176 atomic_t spinning_readers; 177 atomic_t spinning_writers; 178 short lock_nested; 179 /* >= 0 if eb belongs to a log tree, -1 otherwise */ 180 short log_index; 181 182 /* protects write locks */ 183 rwlock_t lock; 184 185 /* readers use lock_wq while they wait for the write 186 * lock holders to unlock 187 */ 188 wait_queue_head_t write_lock_wq; 189 190 /* writers use read_lock_wq while they wait for readers 191 * to unlock 192 */ 193 wait_queue_head_t read_lock_wq; 194 struct page *pages[INLINE_EXTENT_BUFFER_PAGES]; 195 #ifdef CONFIG_BTRFS_DEBUG 196 struct list_head leak_list; 197 #endif 198 }; 199 200 /* 201 * Structure to record how many bytes and which ranges are set/cleared 202 */ 203 struct extent_changeset { 204 /* How many bytes are set/cleared in this operation */ 205 unsigned int bytes_changed; 206 207 /* Changed ranges */ 208 struct ulist range_changed; 209 }; 210 211 static inline void extent_changeset_init(struct extent_changeset *changeset) 212 { 213 changeset->bytes_changed = 0; 214 ulist_init(&changeset->range_changed); 215 } 216 217 static inline struct extent_changeset *extent_changeset_alloc(void) 218 { 219 struct extent_changeset *ret; 220 221 ret = kmalloc(sizeof(*ret), GFP_KERNEL); 222 if (!ret) 223 return NULL; 224 225 extent_changeset_init(ret); 226 return ret; 227 } 228 229 static inline void extent_changeset_release(struct extent_changeset *changeset) 230 { 231 if (!changeset) 232 return; 233 changeset->bytes_changed = 0; 234 ulist_release(&changeset->range_changed); 235 } 236 237 static inline void extent_changeset_free(struct extent_changeset *changeset) 238 { 239 if (!changeset) 240 return; 241 extent_changeset_release(changeset); 242 kfree(changeset); 243 } 244 245 static inline void extent_set_compress_type(unsigned long *bio_flags, 246 int compress_type) 247 { 248 *bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT; 249 } 250 251 static inline int extent_compress_type(unsigned long bio_flags) 252 { 253 return bio_flags >> EXTENT_BIO_FLAG_SHIFT; 254 } 255 256 struct extent_map_tree; 257 258 typedef struct extent_map *(get_extent_t)(struct btrfs_inode *inode, 259 struct page *page, 260 size_t pg_offset, 261 u64 start, u64 len, 262 int create); 263 264 void extent_io_tree_init(struct extent_io_tree *tree, void *private_data); 265 int try_release_extent_mapping(struct page *page, gfp_t mask); 266 int try_release_extent_buffer(struct page *page); 267 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 268 struct extent_state **cached); 269 270 static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) 271 { 272 return lock_extent_bits(tree, start, end, NULL); 273 } 274 275 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end); 276 int extent_read_full_page(struct extent_io_tree *tree, struct page *page, 277 get_extent_t *get_extent, int mirror_num); 278 int __init extent_io_init(void); 279 void __cold extent_io_exit(void); 280 281 u64 count_range_bits(struct extent_io_tree *tree, 282 u64 *start, u64 search_end, 283 u64 max_bytes, unsigned bits, int contig); 284 285 void free_extent_state(struct extent_state *state); 286 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, 287 unsigned bits, int filled, 288 struct extent_state *cached_state); 289 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 290 unsigned bits, struct extent_changeset *changeset); 291 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 292 unsigned bits, int wake, int delete, 293 struct extent_state **cached); 294 int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 295 unsigned bits, int wake, int delete, 296 struct extent_state **cached, gfp_t mask, 297 struct extent_changeset *changeset); 298 299 static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) 300 { 301 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL); 302 } 303 304 static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start, 305 u64 end, struct extent_state **cached) 306 { 307 return __clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, 308 GFP_NOFS, NULL); 309 } 310 311 static inline int unlock_extent_cached_atomic(struct extent_io_tree *tree, 312 u64 start, u64 end, struct extent_state **cached) 313 { 314 return __clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, 315 GFP_ATOMIC, NULL); 316 } 317 318 static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start, 319 u64 end, unsigned bits) 320 { 321 int wake = 0; 322 323 if (bits & EXTENT_LOCKED) 324 wake = 1; 325 326 return clear_extent_bit(tree, start, end, bits, wake, 0, NULL); 327 } 328 329 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 330 unsigned bits, struct extent_changeset *changeset); 331 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 332 unsigned bits, u64 *failed_start, 333 struct extent_state **cached_state, gfp_t mask); 334 335 static inline int set_extent_bits(struct extent_io_tree *tree, u64 start, 336 u64 end, unsigned bits) 337 { 338 return set_extent_bit(tree, start, end, bits, NULL, NULL, GFP_NOFS); 339 } 340 341 static inline int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, 342 u64 end, struct extent_state **cached_state) 343 { 344 return __clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, 345 cached_state, GFP_NOFS, NULL); 346 } 347 348 static inline int set_extent_dirty(struct extent_io_tree *tree, u64 start, 349 u64 end, gfp_t mask) 350 { 351 return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL, 352 NULL, mask); 353 } 354 355 static inline int clear_extent_dirty(struct extent_io_tree *tree, u64 start, 356 u64 end) 357 { 358 return clear_extent_bit(tree, start, end, 359 EXTENT_DIRTY | EXTENT_DELALLOC | 360 EXTENT_DO_ACCOUNTING, 0, 0, NULL); 361 } 362 363 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 364 unsigned bits, unsigned clear_bits, 365 struct extent_state **cached_state); 366 367 static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start, 368 u64 end, unsigned int extra_bits, 369 struct extent_state **cached_state) 370 { 371 return set_extent_bit(tree, start, end, 372 EXTENT_DELALLOC | EXTENT_UPTODATE | extra_bits, 373 NULL, cached_state, GFP_NOFS); 374 } 375 376 static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start, 377 u64 end, struct extent_state **cached_state) 378 { 379 return set_extent_bit(tree, start, end, 380 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, 381 NULL, cached_state, GFP_NOFS); 382 } 383 384 static inline int set_extent_new(struct extent_io_tree *tree, u64 start, 385 u64 end) 386 { 387 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL, 388 GFP_NOFS); 389 } 390 391 static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start, 392 u64 end, struct extent_state **cached_state, gfp_t mask) 393 { 394 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, 395 cached_state, mask); 396 } 397 398 int find_first_extent_bit(struct extent_io_tree *tree, u64 start, 399 u64 *start_ret, u64 *end_ret, unsigned bits, 400 struct extent_state **cached_state); 401 int extent_invalidatepage(struct extent_io_tree *tree, 402 struct page *page, unsigned long offset); 403 int extent_write_full_page(struct page *page, struct writeback_control *wbc); 404 int extent_write_locked_range(struct inode *inode, u64 start, u64 end, 405 int mode); 406 int extent_writepages(struct address_space *mapping, 407 struct writeback_control *wbc); 408 int btree_write_cache_pages(struct address_space *mapping, 409 struct writeback_control *wbc); 410 int extent_readpages(struct address_space *mapping, struct list_head *pages, 411 unsigned nr_pages); 412 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 413 __u64 start, __u64 len); 414 void set_page_extent_mapped(struct page *page); 415 416 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, 417 u64 start); 418 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, 419 u64 start, unsigned long len); 420 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, 421 u64 start); 422 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src); 423 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, 424 u64 start); 425 void free_extent_buffer(struct extent_buffer *eb); 426 void free_extent_buffer_stale(struct extent_buffer *eb); 427 #define WAIT_NONE 0 428 #define WAIT_COMPLETE 1 429 #define WAIT_PAGE_LOCK 2 430 int read_extent_buffer_pages(struct extent_io_tree *tree, 431 struct extent_buffer *eb, int wait, 432 int mirror_num); 433 void wait_on_extent_buffer_writeback(struct extent_buffer *eb); 434 435 static inline int num_extent_pages(const struct extent_buffer *eb) 436 { 437 return (round_up(eb->start + eb->len, PAGE_SIZE) >> PAGE_SHIFT) - 438 (eb->start >> PAGE_SHIFT); 439 } 440 441 static inline void extent_buffer_get(struct extent_buffer *eb) 442 { 443 atomic_inc(&eb->refs); 444 } 445 446 static inline int extent_buffer_uptodate(struct extent_buffer *eb) 447 { 448 return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); 449 } 450 451 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, 452 unsigned long start, unsigned long len); 453 void read_extent_buffer(const struct extent_buffer *eb, void *dst, 454 unsigned long start, 455 unsigned long len); 456 int read_extent_buffer_to_user(const struct extent_buffer *eb, 457 void __user *dst, unsigned long start, 458 unsigned long len); 459 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *src); 460 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, 461 const void *src); 462 void write_extent_buffer(struct extent_buffer *eb, const void *src, 463 unsigned long start, unsigned long len); 464 void copy_extent_buffer_full(struct extent_buffer *dst, 465 struct extent_buffer *src); 466 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, 467 unsigned long dst_offset, unsigned long src_offset, 468 unsigned long len); 469 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, 470 unsigned long src_offset, unsigned long len); 471 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, 472 unsigned long src_offset, unsigned long len); 473 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, 474 unsigned long len); 475 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, 476 unsigned long pos); 477 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, 478 unsigned long pos, unsigned long len); 479 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, 480 unsigned long pos, unsigned long len); 481 void clear_extent_buffer_dirty(struct extent_buffer *eb); 482 bool set_extent_buffer_dirty(struct extent_buffer *eb); 483 void set_extent_buffer_uptodate(struct extent_buffer *eb); 484 void clear_extent_buffer_uptodate(struct extent_buffer *eb); 485 int extent_buffer_under_io(struct extent_buffer *eb); 486 int map_private_extent_buffer(const struct extent_buffer *eb, 487 unsigned long offset, unsigned long min_len, 488 char **map, unsigned long *map_start, 489 unsigned long *map_len); 490 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end); 491 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end); 492 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, 493 u64 delalloc_end, struct page *locked_page, 494 unsigned bits_to_clear, 495 unsigned long page_ops); 496 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte); 497 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs); 498 struct bio *btrfs_bio_clone(struct bio *bio); 499 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size); 500 501 struct btrfs_fs_info; 502 struct btrfs_inode; 503 504 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, 505 u64 length, u64 logical, struct page *page, 506 unsigned int pg_offset, int mirror_num); 507 int clean_io_failure(struct btrfs_fs_info *fs_info, 508 struct extent_io_tree *failure_tree, 509 struct extent_io_tree *io_tree, u64 start, 510 struct page *page, u64 ino, unsigned int pg_offset); 511 void end_extent_writepage(struct page *page, int err, u64 start, u64 end); 512 int repair_eb_io_failure(struct btrfs_fs_info *fs_info, 513 struct extent_buffer *eb, int mirror_num); 514 515 /* 516 * When IO fails, either with EIO or csum verification fails, we 517 * try other mirrors that might have a good copy of the data. This 518 * io_failure_record is used to record state as we go through all the 519 * mirrors. If another mirror has good data, the page is set up to date 520 * and things continue. If a good mirror can't be found, the original 521 * bio end_io callback is called to indicate things have failed. 522 */ 523 struct io_failure_record { 524 struct page *page; 525 u64 start; 526 u64 len; 527 u64 logical; 528 unsigned long bio_flags; 529 int this_mirror; 530 int failed_mirror; 531 int in_validation; 532 }; 533 534 535 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, 536 u64 end); 537 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, 538 struct io_failure_record **failrec_ret); 539 bool btrfs_check_repairable(struct inode *inode, unsigned failed_bio_pages, 540 struct io_failure_record *failrec, int fail_mirror); 541 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, 542 struct io_failure_record *failrec, 543 struct page *page, int pg_offset, int icsum, 544 bio_end_io_t *endio_func, void *data); 545 int free_io_failure(struct extent_io_tree *failure_tree, 546 struct extent_io_tree *io_tree, 547 struct io_failure_record *rec); 548 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 549 u64 btrfs_find_lock_delalloc_range(struct inode *inode, 550 struct extent_io_tree *tree, 551 struct page *locked_page, u64 *start, 552 u64 *end, u64 max_bytes); 553 #endif 554 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, 555 u64 start); 556 557 #endif 558