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 struct io_failure_record;
  65
  66 typedef int (extent_submit_bio_hook_t)(struct inode *inode, struct bio *bio,
  67                                        int mirror_num, unsigned long bio_flags,
  68                                        u64 bio_offset);
  69 struct extent_io_ops {
  70         int (*fill_delalloc)(struct inode *inode, struct page *locked_page,
  71                              u64 start, u64 end, int *page_started,
  72                              unsigned long *nr_written);
  73         int (*writepage_start_hook)(struct page *page, u64 start, u64 end);
  74         extent_submit_bio_hook_t *submit_bio_hook;
  75         int (*merge_bio_hook)(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         struct io_failure_record *failrec;
 115
 116 #ifdef CONFIG_BTRFS_DEBUG
 117         struct list_head leak_list;
 118 #endif
 119 };
 120
 121 #define INLINE_EXTENT_BUFFER_PAGES 16
 122 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE)
 123 struct extent_buffer {
 124         u64 start;
 125         unsigned long len;
 126         unsigned long bflags;
 127         struct btrfs_fs_info *fs_info;
 128         spinlock_t refs_lock;
 129         atomic_t refs;
 130         atomic_t io_pages;
 131         int read_mirror;
 132         struct rcu_head rcu_head;
 133         pid_t lock_owner;
 134
 135         /* count of read lock holders on the extent buffer */
 136         atomic_t write_locks;
 137         atomic_t read_locks;
 138         atomic_t blocking_writers;
 139         atomic_t blocking_readers;
 140         atomic_t spinning_readers;
 141         atomic_t spinning_writers;
 142         short lock_nested;
 143         /* >= 0 if eb belongs to a log tree, -1 otherwise */
 144         short log_index;
 145
 146         /* protects write locks */
 147         rwlock_t lock;
 148
 149         /* readers use lock_wq while they wait for the write
 150          * lock holders to unlock
 151          */
 152         wait_queue_head_t write_lock_wq;
 153
 154         /* writers use read_lock_wq while they wait for readers
 155          * to unlock
 156          */
 157         wait_queue_head_t read_lock_wq;
 158         struct page *pages[INLINE_EXTENT_BUFFER_PAGES];
 159 #ifdef CONFIG_BTRFS_DEBUG
 160         struct list_head leak_list;
 161 #endif
 162 };
 163
 164 /*
 165  * Structure to record how many bytes and which ranges are set/cleared
 166  */
 167 struct extent_changeset {
 168         /* How many bytes are set/cleared in this operation */
 169         u64 bytes_changed;
 170
 171         /* Changed ranges */
 172         struct ulist *range_changed;
 173 };
 174
 175 static inline void extent_set_compress_type(unsigned long *bio_flags,
 176                                             int compress_type)
 177 {
 178         *bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT;
 179 }
 180
 181 static inline int extent_compress_type(unsigned long bio_flags)
 182 {
 183         return bio_flags >> EXTENT_BIO_FLAG_SHIFT;
 184 }
 185
 186 struct extent_map_tree;
 187
 188 typedef struct extent_map *(get_extent_t)(struct inode *inode,
 189                                           struct page *page,
 190                                           size_t pg_offset,
 191                                           u64 start, u64 len,
 192                                           int create);
 193
 194 void extent_io_tree_init(struct extent_io_tree *tree,
 195                          struct address_space *mapping);
 196 int try_release_extent_mapping(struct extent_map_tree *map,
 197                                struct extent_io_tree *tree, struct page *page,
 198                                gfp_t mask);
 199 int try_release_extent_buffer(struct page *page);
 200 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 201                      struct extent_state **cached);
 202
 203 static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
 204 {
 205         return lock_extent_bits(tree, start, end, NULL);
 206 }
 207
 208 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
 209 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
 210                           get_extent_t *get_extent, int mirror_num);
 211 int __init extent_io_init(void);
 212 void extent_io_exit(void);
 213
 214 u64 count_range_bits(struct extent_io_tree *tree,
 215                      u64 *start, u64 search_end,
 216                      u64 max_bytes, unsigned bits, int contig);
 217
 218 void free_extent_state(struct extent_state *state);
 219 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
 220                    unsigned bits, int filled,
 221                    struct extent_state *cached_state);
 222 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 223                 unsigned bits, struct extent_changeset *changeset);
 224 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 225                      unsigned bits, int wake, int delete,
 226                      struct extent_state **cached, gfp_t mask);
 227
 228 static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
 229 {
 230         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
 231                                 GFP_NOFS);
 232 }
 233
 234 static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start,
 235                 u64 end, struct extent_state **cached, gfp_t mask)
 236 {
 237         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
 238                                 mask);
 239 }
 240
 241 static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start,
 242                 u64 end, unsigned bits)
 243 {
 244         int wake = 0;
 245
 246         if (bits & EXTENT_LOCKED)
 247                 wake = 1;
 248
 249         return clear_extent_bit(tree, start, end, bits, wake, 0, NULL,
 250                         GFP_NOFS);
 251 }
 252
 253 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 254                            unsigned bits, struct extent_changeset *changeset);
 255 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 256                    unsigned bits, u64 *failed_start,
 257                    struct extent_state **cached_state, gfp_t mask);
 258
 259 static inline int set_extent_bits(struct extent_io_tree *tree, u64 start,
 260                 u64 end, unsigned bits)
 261 {
 262         return set_extent_bit(tree, start, end, bits, NULL, NULL, GFP_NOFS);
 263 }
 264
 265 static inline int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
 266                 u64 end, struct extent_state **cached_state, gfp_t mask)
 267 {
 268         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
 269                                 cached_state, mask);
 270 }
 271
 272 static inline int set_extent_dirty(struct extent_io_tree *tree, u64 start,
 273                 u64 end, gfp_t mask)
 274 {
 275         return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
 276                               NULL, mask);
 277 }
 278
 279 static inline int clear_extent_dirty(struct extent_io_tree *tree, u64 start,
 280                 u64 end)
 281 {
 282         return clear_extent_bit(tree, start, end,
 283                                 EXTENT_DIRTY | EXTENT_DELALLOC |
 284                                 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
 285 }
 286
 287 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 288                        unsigned bits, unsigned clear_bits,
 289                        struct extent_state **cached_state);
 290
 291 static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
 292                 u64 end, struct extent_state **cached_state)
 293 {
 294         return set_extent_bit(tree, start, end,
 295                               EXTENT_DELALLOC | EXTENT_UPTODATE,
 296                               NULL, cached_state, GFP_NOFS);
 297 }
 298
 299 static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
 300                 u64 end, struct extent_state **cached_state)
 301 {
 302         return set_extent_bit(tree, start, end,
 303                               EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
 304                               NULL, cached_state, GFP_NOFS);
 305 }
 306
 307 static inline int set_extent_new(struct extent_io_tree *tree, u64 start,
 308                 u64 end)
 309 {
 310         return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL,
 311                         GFP_NOFS);
 312 }
 313
 314 static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start,
 315                 u64 end, struct extent_state **cached_state, gfp_t mask)
 316 {
 317         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
 318                               cached_state, mask);
 319 }
 320
 321 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 322                           u64 *start_ret, u64 *end_ret, unsigned bits,
 323                           struct extent_state **cached_state);
 324 int extent_invalidatepage(struct extent_io_tree *tree,
 325                           struct page *page, unsigned long offset);
 326 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
 327                           get_extent_t *get_extent,
 328                           struct writeback_control *wbc);
 329 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
 330                               u64 start, u64 end, get_extent_t *get_extent,
 331                               int mode);
 332 int extent_writepages(struct extent_io_tree *tree,
 333                       struct address_space *mapping,
 334                       get_extent_t *get_extent,
 335                       struct writeback_control *wbc);
 336 int btree_write_cache_pages(struct address_space *mapping,
 337                             struct writeback_control *wbc);
 338 int extent_readpages(struct extent_io_tree *tree,
 339                      struct address_space *mapping,
 340                      struct list_head *pages, unsigned nr_pages,
 341                      get_extent_t get_extent);
 342 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 343                 __u64 start, __u64 len, get_extent_t *get_extent);
 344 void set_page_extent_mapped(struct page *page);
 345
 346 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 347                                           u64 start);
 348 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
 349                                                   u64 start, unsigned long len);
 350 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
 351                                                 u64 start, u32 nodesize);
 352 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src);
 353 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
 354                                          u64 start);
 355 void free_extent_buffer(struct extent_buffer *eb);
 356 void free_extent_buffer_stale(struct extent_buffer *eb);
 357 #define WAIT_NONE       0
 358 #define WAIT_COMPLETE   1
 359 #define WAIT_PAGE_LOCK  2
 360 int read_extent_buffer_pages(struct extent_io_tree *tree,
 361                              struct extent_buffer *eb, u64 start, int wait,
 362                              get_extent_t *get_extent, int mirror_num);
 363 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
 364
 365 static inline unsigned long num_extent_pages(u64 start, u64 len)
 366 {
 367         return ((start + len + PAGE_SIZE - 1) >> PAGE_SHIFT) -
 368                 (start >> PAGE_SHIFT);
 369 }
 370
 371 static inline void extent_buffer_get(struct extent_buffer *eb)
 372 {
 373         atomic_inc(&eb->refs);
 374 }
 375
 376 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
 377                           unsigned long start,
 378                           unsigned long len);
 379 void read_extent_buffer(struct extent_buffer *eb, void *dst,
 380                         unsigned long start,
 381                         unsigned long len);
 382 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dst,
 383                                unsigned long start,
 384                                unsigned long len);
 385 void write_extent_buffer(struct extent_buffer *eb, const void *src,
 386                          unsigned long start, unsigned long len);
 387 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
 388                         unsigned long dst_offset, unsigned long src_offset,
 389                         unsigned long len);
 390 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 391                            unsigned long src_offset, unsigned long len);
 392 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 393                            unsigned long src_offset, unsigned long len);
 394 void memset_extent_buffer(struct extent_buffer *eb, char c,
 395                           unsigned long start, unsigned long len);
 396 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
 397                            unsigned long pos);
 398 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
 399                               unsigned long pos, unsigned long len);
 400 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
 401                                 unsigned long pos, unsigned long len);
 402 void clear_extent_buffer_dirty(struct extent_buffer *eb);
 403 int set_extent_buffer_dirty(struct extent_buffer *eb);
 404 void set_extent_buffer_uptodate(struct extent_buffer *eb);
 405 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
 406 int extent_buffer_uptodate(struct extent_buffer *eb);
 407 int extent_buffer_under_io(struct extent_buffer *eb);
 408 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset,
 409                       unsigned long min_len, char **map,
 410                       unsigned long *map_start,
 411                       unsigned long *map_len);
 412 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
 413 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
 414 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
 415                                  struct page *locked_page,
 416                                  unsigned bits_to_clear,
 417                                  unsigned long page_ops);
 418 struct bio *
 419 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
 420                 gfp_t gfp_flags);
 421 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs);
 422 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask);
 423
 424 struct btrfs_fs_info;
 425
 426 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
 427                       struct page *page, unsigned int pg_offset,
 428                       int mirror_num);
 429 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
 430                      unsigned int pg_offset);
 431 void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
 432 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
 433                          int mirror_num);
 434
 435 /*
 436  * When IO fails, either with EIO or csum verification fails, we
 437  * try other mirrors that might have a good copy of the data.  This
 438  * io_failure_record is used to record state as we go through all the
 439  * mirrors.  If another mirror has good data, the page is set up to date
 440  * and things continue.  If a good mirror can't be found, the original
 441  * bio end_io callback is called to indicate things have failed.
 442  */
 443 struct io_failure_record {
 444         struct page *page;
 445         u64 start;
 446         u64 len;
 447         u64 logical;
 448         unsigned long bio_flags;
 449         int this_mirror;
 450         int failed_mirror;
 451         int in_validation;
 452 };
 453
 454 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end);
 455 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
 456                                 struct io_failure_record **failrec_ret);
 457 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
 458                            struct io_failure_record *failrec, int fail_mirror);
 459 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
 460                                     struct io_failure_record *failrec,
 461                                     struct page *page, int pg_offset, int icsum,
 462                                     bio_end_io_t *endio_func, void *data);
 463 int free_io_failure(struct inode *inode, struct io_failure_record *rec);
 464 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 465 noinline u64 find_lock_delalloc_range(struct inode *inode,
 466                                       struct extent_io_tree *tree,
 467                                       struct page *locked_page, u64 *start,
 468                                       u64 *end, u64 max_bytes);
 469 #endif
 470 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
 471                                                u64 start, u32 nodesize);
 472 #endif