1 #ifndef __EXTENTIO__ 2 #define __EXTENTIO__ 3 4 #include <linux/rbtree.h> 5 #include <linux/refcount.h> 6 #include "ulist.h" 7 8 /* bits for the extent state */ 9 #define EXTENT_DIRTY (1U << 0) 10 #define EXTENT_WRITEBACK (1U << 1) 11 #define EXTENT_UPTODATE (1U << 2) 12 #define EXTENT_LOCKED (1U << 3) 13 #define EXTENT_NEW (1U << 4) 14 #define EXTENT_DELALLOC (1U << 5) 15 #define EXTENT_DEFRAG (1U << 6) 16 #define EXTENT_BOUNDARY (1U << 9) 17 #define EXTENT_NODATASUM (1U << 10) 18 #define EXTENT_CLEAR_META_RESV (1U << 11) 19 #define EXTENT_FIRST_DELALLOC (1U << 12) 20 #define EXTENT_NEED_WAIT (1U << 13) 21 #define EXTENT_DAMAGED (1U << 14) 22 #define EXTENT_NORESERVE (1U << 15) 23 #define EXTENT_QGROUP_RESERVED (1U << 16) 24 #define EXTENT_CLEAR_DATA_RESV (1U << 17) 25 #define EXTENT_DELALLOC_NEW (1U << 18) 26 #define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) 27 #define EXTENT_DO_ACCOUNTING (EXTENT_CLEAR_META_RESV | \ 28 EXTENT_CLEAR_DATA_RESV) 29 #define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC) 30 31 /* 32 * flags for bio submission. The high bits indicate the compression 33 * type for this bio 34 */ 35 #define EXTENT_BIO_COMPRESSED 1 36 #define EXTENT_BIO_TREE_LOG 2 37 #define EXTENT_BIO_FLAG_SHIFT 16 38 39 /* these are bit numbers for test/set bit */ 40 #define EXTENT_BUFFER_UPTODATE 0 41 #define EXTENT_BUFFER_DIRTY 2 42 #define EXTENT_BUFFER_CORRUPT 3 43 #define EXTENT_BUFFER_READAHEAD 4 /* this got triggered by readahead */ 44 #define EXTENT_BUFFER_TREE_REF 5 45 #define EXTENT_BUFFER_STALE 6 46 #define EXTENT_BUFFER_WRITEBACK 7 47 #define EXTENT_BUFFER_READ_ERR 8 /* read IO error */ 48 #define EXTENT_BUFFER_DUMMY 9 49 #define EXTENT_BUFFER_IN_TREE 10 50 #define EXTENT_BUFFER_WRITE_ERR 11 /* write IO error */ 51 52 /* these are flags for __process_pages_contig */ 53 #define PAGE_UNLOCK (1 << 0) 54 #define PAGE_CLEAR_DIRTY (1 << 1) 55 #define PAGE_SET_WRITEBACK (1 << 2) 56 #define PAGE_END_WRITEBACK (1 << 3) 57 #define PAGE_SET_PRIVATE2 (1 << 4) 58 #define PAGE_SET_ERROR (1 << 5) 59 #define PAGE_LOCK (1 << 6) 60 61 /* 62 * page->private values. Every page that is controlled by the extent 63 * map has page->private set to one. 64 */ 65 #define EXTENT_PAGE_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 ((1 << 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 static inline int le_test_bit(int nr, const u8 *addr) 82 { 83 return 1U & (addr[BIT_BYTE(nr)] >> (nr & (BITS_PER_BYTE-1))); 84 } 85 86 extern void le_bitmap_set(u8 *map, unsigned int start, int len); 87 extern void le_bitmap_clear(u8 *map, unsigned int start, int len); 88 89 struct extent_state; 90 struct btrfs_root; 91 struct btrfs_inode; 92 struct btrfs_io_bio; 93 struct io_failure_record; 94 95 typedef blk_status_t (extent_submit_bio_hook_t)(void *private_data, struct bio *bio, 96 int mirror_num, unsigned long bio_flags, 97 u64 bio_offset); 98 struct extent_io_ops { 99 /* 100 * The following callbacks must be allways defined, the function 101 * pointer will be called unconditionally. 102 */ 103 extent_submit_bio_hook_t *submit_bio_hook; 104 int (*readpage_end_io_hook)(struct btrfs_io_bio *io_bio, u64 phy_offset, 105 struct page *page, u64 start, u64 end, 106 int mirror); 107 int (*merge_bio_hook)(struct page *page, unsigned long offset, 108 size_t size, struct bio *bio, 109 unsigned long bio_flags); 110 int (*readpage_io_failed_hook)(struct page *page, int failed_mirror); 111 struct btrfs_fs_info *(*tree_fs_info)(void *private_data); 112 void (*set_range_writeback)(void *private_data, u64 start, u64 end); 113 114 /* 115 * Optional hooks, called if the pointer is not NULL 116 */ 117 int (*fill_delalloc)(void *private_data, struct page *locked_page, 118 u64 start, u64 end, int *page_started, 119 unsigned long *nr_written); 120 121 int (*writepage_start_hook)(struct page *page, u64 start, u64 end); 122 void (*writepage_end_io_hook)(struct page *page, u64 start, u64 end, 123 struct extent_state *state, int uptodate); 124 void (*set_bit_hook)(void *private_data, struct extent_state *state, 125 unsigned *bits); 126 void (*clear_bit_hook)(void *private_data, 127 struct extent_state *state, 128 unsigned *bits); 129 void (*merge_extent_hook)(void *private_data, 130 struct extent_state *new, 131 struct extent_state *other); 132 void (*split_extent_hook)(void *private_data, 133 struct extent_state *orig, u64 split); 134 void (*check_extent_io_range)(void *private_data, const char *caller, 135 u64 start, u64 end); 136 }; 137 138 struct extent_io_tree { 139 struct rb_root state; 140 void *private_data; 141 u64 dirty_bytes; 142 int track_uptodate; 143 spinlock_t lock; 144 const struct extent_io_ops *ops; 145 }; 146 147 struct extent_state { 148 u64 start; 149 u64 end; /* inclusive */ 150 struct rb_node rb_node; 151 152 /* ADD NEW ELEMENTS AFTER THIS */ 153 wait_queue_head_t wq; 154 refcount_t refs; 155 unsigned state; 156 157 struct io_failure_record *failrec; 158 159 #ifdef CONFIG_BTRFS_DEBUG 160 struct list_head leak_list; 161 #endif 162 }; 163 164 #define INLINE_EXTENT_BUFFER_PAGES 16 165 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE) 166 struct extent_buffer { 167 u64 start; 168 unsigned long len; 169 unsigned long bflags; 170 struct btrfs_fs_info *fs_info; 171 spinlock_t refs_lock; 172 atomic_t refs; 173 atomic_t io_pages; 174 int read_mirror; 175 struct rcu_head rcu_head; 176 pid_t lock_owner; 177 178 /* count of read lock holders on the extent buffer */ 179 atomic_t write_locks; 180 atomic_t read_locks; 181 atomic_t blocking_writers; 182 atomic_t blocking_readers; 183 atomic_t spinning_readers; 184 atomic_t spinning_writers; 185 short lock_nested; 186 /* >= 0 if eb belongs to a log tree, -1 otherwise */ 187 short log_index; 188 189 /* protects write locks */ 190 rwlock_t lock; 191 192 /* readers use lock_wq while they wait for the write 193 * lock holders to unlock 194 */ 195 wait_queue_head_t write_lock_wq; 196 197 /* writers use read_lock_wq while they wait for readers 198 * to unlock 199 */ 200 wait_queue_head_t read_lock_wq; 201 struct page *pages[INLINE_EXTENT_BUFFER_PAGES]; 202 #ifdef CONFIG_BTRFS_DEBUG 203 struct list_head leak_list; 204 #endif 205 }; 206 207 /* 208 * Structure to record how many bytes and which ranges are set/cleared 209 */ 210 struct extent_changeset { 211 /* How many bytes are set/cleared in this operation */ 212 unsigned int bytes_changed; 213 214 /* Changed ranges */ 215 struct ulist range_changed; 216 }; 217 218 static inline void extent_changeset_init(struct extent_changeset *changeset) 219 { 220 changeset->bytes_changed = 0; 221 ulist_init(&changeset->range_changed); 222 } 223 224 static inline struct extent_changeset *extent_changeset_alloc(void) 225 { 226 struct extent_changeset *ret; 227 228 ret = kmalloc(sizeof(*ret), GFP_KERNEL); 229 if (!ret) 230 return NULL; 231 232 extent_changeset_init(ret); 233 return ret; 234 } 235 236 static inline void extent_changeset_release(struct extent_changeset *changeset) 237 { 238 if (!changeset) 239 return; 240 changeset->bytes_changed = 0; 241 ulist_release(&changeset->range_changed); 242 } 243 244 static inline void extent_changeset_free(struct extent_changeset *changeset) 245 { 246 if (!changeset) 247 return; 248 extent_changeset_release(changeset); 249 kfree(changeset); 250 } 251 252 static inline void extent_set_compress_type(unsigned long *bio_flags, 253 int compress_type) 254 { 255 *bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT; 256 } 257 258 static inline int extent_compress_type(unsigned long bio_flags) 259 { 260 return bio_flags >> EXTENT_BIO_FLAG_SHIFT; 261 } 262 263 struct extent_map_tree; 264 265 typedef struct extent_map *(get_extent_t)(struct btrfs_inode *inode, 266 struct page *page, 267 size_t pg_offset, 268 u64 start, u64 len, 269 int create); 270 271 void extent_io_tree_init(struct extent_io_tree *tree, void *private_data); 272 int try_release_extent_mapping(struct extent_map_tree *map, 273 struct extent_io_tree *tree, struct page *page, 274 gfp_t mask); 275 int try_release_extent_buffer(struct page *page); 276 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 277 struct extent_state **cached); 278 279 static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) 280 { 281 return lock_extent_bits(tree, start, end, NULL); 282 } 283 284 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end); 285 int extent_read_full_page(struct extent_io_tree *tree, struct page *page, 286 get_extent_t *get_extent, int mirror_num); 287 int __init extent_io_init(void); 288 void extent_io_exit(void); 289 290 u64 count_range_bits(struct extent_io_tree *tree, 291 u64 *start, u64 search_end, 292 u64 max_bytes, unsigned bits, int contig); 293 294 void free_extent_state(struct extent_state *state); 295 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, 296 unsigned bits, int filled, 297 struct extent_state *cached_state); 298 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, 299 unsigned bits, struct extent_changeset *changeset); 300 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, 301 unsigned bits, int wake, int delete, 302 struct extent_state **cached, gfp_t mask); 303 304 static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end) 305 { 306 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL, 307 GFP_NOFS); 308 } 309 310 static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start, 311 u64 end, struct extent_state **cached, gfp_t mask) 312 { 313 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached, 314 mask); 315 } 316 317 static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start, 318 u64 end, unsigned bits) 319 { 320 int wake = 0; 321 322 if (bits & EXTENT_LOCKED) 323 wake = 1; 324 325 return clear_extent_bit(tree, start, end, bits, wake, 0, NULL, 326 GFP_NOFS); 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, gfp_t mask) 343 { 344 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, 345 cached_state, mask); 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, GFP_NOFS); 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, struct extent_state **cached_state) 369 { 370 return set_extent_bit(tree, start, end, 371 EXTENT_DELALLOC | EXTENT_UPTODATE, 372 NULL, cached_state, GFP_NOFS); 373 } 374 375 static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start, 376 u64 end, struct extent_state **cached_state) 377 { 378 return set_extent_bit(tree, start, end, 379 EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, 380 NULL, cached_state, GFP_NOFS); 381 } 382 383 static inline int set_extent_new(struct extent_io_tree *tree, u64 start, 384 u64 end) 385 { 386 return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL, 387 GFP_NOFS); 388 } 389 390 static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start, 391 u64 end, struct extent_state **cached_state, gfp_t mask) 392 { 393 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL, 394 cached_state, mask); 395 } 396 397 int find_first_extent_bit(struct extent_io_tree *tree, u64 start, 398 u64 *start_ret, u64 *end_ret, unsigned bits, 399 struct extent_state **cached_state); 400 int extent_invalidatepage(struct extent_io_tree *tree, 401 struct page *page, unsigned long offset); 402 int extent_write_full_page(struct extent_io_tree *tree, struct page *page, 403 get_extent_t *get_extent, 404 struct writeback_control *wbc); 405 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode, 406 u64 start, u64 end, get_extent_t *get_extent, 407 int mode); 408 int extent_writepages(struct extent_io_tree *tree, 409 struct address_space *mapping, 410 get_extent_t *get_extent, 411 struct writeback_control *wbc); 412 int btree_write_cache_pages(struct address_space *mapping, 413 struct writeback_control *wbc); 414 int extent_readpages(struct extent_io_tree *tree, 415 struct address_space *mapping, 416 struct list_head *pages, unsigned nr_pages, 417 get_extent_t get_extent); 418 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 419 __u64 start, __u64 len, get_extent_t *get_extent); 420 void set_page_extent_mapped(struct page *page); 421 422 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, 423 u64 start); 424 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, 425 u64 start, unsigned long len); 426 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, 427 u64 start); 428 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src); 429 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, 430 u64 start); 431 void free_extent_buffer(struct extent_buffer *eb); 432 void free_extent_buffer_stale(struct extent_buffer *eb); 433 #define WAIT_NONE 0 434 #define WAIT_COMPLETE 1 435 #define WAIT_PAGE_LOCK 2 436 int read_extent_buffer_pages(struct extent_io_tree *tree, 437 struct extent_buffer *eb, int wait, 438 get_extent_t *get_extent, int mirror_num); 439 void wait_on_extent_buffer_writeback(struct extent_buffer *eb); 440 441 static inline unsigned long num_extent_pages(u64 start, u64 len) 442 { 443 return ((start + len + PAGE_SIZE - 1) >> PAGE_SHIFT) - 444 (start >> PAGE_SHIFT); 445 } 446 447 static inline void extent_buffer_get(struct extent_buffer *eb) 448 { 449 atomic_inc(&eb->refs); 450 } 451 452 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, 453 unsigned long start, 454 unsigned long len); 455 void read_extent_buffer(struct extent_buffer *eb, void *dst, 456 unsigned long start, 457 unsigned long len); 458 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dst, 459 unsigned long start, 460 unsigned long len); 461 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *src); 462 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb, 463 const void *src); 464 void write_extent_buffer(struct extent_buffer *eb, const void *src, 465 unsigned long start, unsigned long len); 466 void copy_extent_buffer_full(struct extent_buffer *dst, 467 struct extent_buffer *src); 468 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, 469 unsigned long dst_offset, unsigned long src_offset, 470 unsigned long len); 471 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, 472 unsigned long src_offset, unsigned long len); 473 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, 474 unsigned long src_offset, unsigned long len); 475 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start, 476 unsigned long len); 477 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start, 478 unsigned long pos); 479 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start, 480 unsigned long pos, unsigned long len); 481 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start, 482 unsigned long pos, unsigned long len); 483 void clear_extent_buffer_dirty(struct extent_buffer *eb); 484 int set_extent_buffer_dirty(struct extent_buffer *eb); 485 void set_extent_buffer_uptodate(struct extent_buffer *eb); 486 void clear_extent_buffer_uptodate(struct extent_buffer *eb); 487 int extent_buffer_uptodate(struct extent_buffer *eb); 488 int extent_buffer_under_io(struct extent_buffer *eb); 489 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset, 490 unsigned long min_len, char **map, 491 unsigned long *map_start, 492 unsigned long *map_len); 493 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end); 494 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end); 495 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end, 496 u64 delalloc_end, struct page *locked_page, 497 unsigned bits_to_clear, 498 unsigned long page_ops); 499 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte); 500 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs); 501 struct bio *btrfs_bio_clone(struct bio *bio); 502 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size); 503 504 struct btrfs_fs_info; 505 struct btrfs_inode; 506 507 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, 508 u64 length, u64 logical, struct page *page, 509 unsigned int pg_offset, int mirror_num); 510 int clean_io_failure(struct btrfs_fs_info *fs_info, 511 struct extent_io_tree *failure_tree, 512 struct extent_io_tree *io_tree, u64 start, 513 struct page *page, u64 ino, unsigned int pg_offset); 514 void end_extent_writepage(struct page *page, int err, u64 start, u64 end); 515 int repair_eb_io_failure(struct btrfs_fs_info *fs_info, 516 struct extent_buffer *eb, int mirror_num); 517 518 /* 519 * When IO fails, either with EIO or csum verification fails, we 520 * try other mirrors that might have a good copy of the data. This 521 * io_failure_record is used to record state as we go through all the 522 * mirrors. If another mirror has good data, the page is set up to date 523 * and things continue. If a good mirror can't be found, the original 524 * bio end_io callback is called to indicate things have failed. 525 */ 526 struct io_failure_record { 527 struct page *page; 528 u64 start; 529 u64 len; 530 u64 logical; 531 unsigned long bio_flags; 532 int this_mirror; 533 int failed_mirror; 534 int in_validation; 535 }; 536 537 538 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, 539 u64 end); 540 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end, 541 struct io_failure_record **failrec_ret); 542 bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio, 543 struct io_failure_record *failrec, int fail_mirror); 544 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio, 545 struct io_failure_record *failrec, 546 struct page *page, int pg_offset, int icsum, 547 bio_end_io_t *endio_func, void *data); 548 int free_io_failure(struct extent_io_tree *failure_tree, 549 struct extent_io_tree *io_tree, 550 struct io_failure_record *rec); 551 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 552 noinline u64 find_lock_delalloc_range(struct inode *inode, 553 struct extent_io_tree *tree, 554 struct page *locked_page, u64 *start, 555 u64 *end, u64 max_bytes); 556 #endif 557 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, 558 u64 start); 559 #endif 560