fs/btrfs/extent_io.h

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