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_CLEAR_DATA_RESV  (1U << 17)
  24 #define EXTENT_IOBITS           (EXTENT_LOCKED | EXTENT_WRITEBACK)
  25 #define EXTENT_CTLBITS          (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
  26
  27 /*
  28  * flags for bio submission. The high bits indicate the compression
  29  * type for this bio
  30  */
  31 #define EXTENT_BIO_COMPRESSED 1
  32 #define EXTENT_BIO_TREE_LOG 2
  33 #define EXTENT_BIO_FLAG_SHIFT 16
  34
  35 /* these are bit numbers for test/set bit */
  36 #define EXTENT_BUFFER_UPTODATE 0
  37 #define EXTENT_BUFFER_DIRTY 2
  38 #define EXTENT_BUFFER_CORRUPT 3
  39 #define EXTENT_BUFFER_READAHEAD 4       /* this got triggered by readahead */
  40 #define EXTENT_BUFFER_TREE_REF 5
  41 #define EXTENT_BUFFER_STALE 6
  42 #define EXTENT_BUFFER_WRITEBACK 7
  43 #define EXTENT_BUFFER_READ_ERR 8        /* read IO error */
  44 #define EXTENT_BUFFER_DUMMY 9
  45 #define EXTENT_BUFFER_IN_TREE 10
  46 #define EXTENT_BUFFER_WRITE_ERR 11    /* write IO error */
  47
  48 /* these are flags for __process_pages_contig */
  49 #define PAGE_UNLOCK             (1 << 0)
  50 #define PAGE_CLEAR_DIRTY        (1 << 1)
  51 #define PAGE_SET_WRITEBACK      (1 << 2)
  52 #define PAGE_END_WRITEBACK      (1 << 3)
  53 #define PAGE_SET_PRIVATE2       (1 << 4)
  54 #define PAGE_SET_ERROR          (1 << 5)
  55 #define PAGE_LOCK               (1 << 6)
  56
  57 /*
  58  * page->private values.  Every page that is controlled by the extent
  59  * map has page->private set to one.
  60  */
  61 #define EXTENT_PAGE_PRIVATE 1
  62
  63 /*
  64  * The extent buffer bitmap operations are done with byte granularity instead of
  65  * word granularity for two reasons:
  66  * 1. The bitmaps must be little-endian on disk.
  67  * 2. Bitmap items are not guaranteed to be aligned to a word and therefore a
  68  *    single word in a bitmap may straddle two pages in the extent buffer.
  69  */
  70 #define BIT_BYTE(nr) ((nr) / BITS_PER_BYTE)
  71 #define BYTE_MASK ((1 << BITS_PER_BYTE) - 1)
  72 #define BITMAP_FIRST_BYTE_MASK(start) \
  73         ((BYTE_MASK << ((start) & (BITS_PER_BYTE - 1))) & BYTE_MASK)
  74 #define BITMAP_LAST_BYTE_MASK(nbits) \
  75         (BYTE_MASK >> (-(nbits) & (BITS_PER_BYTE - 1)))
  76
  77 static inline int le_test_bit(int nr, const u8 *addr)
  78 {
  79         return 1U & (addr[BIT_BYTE(nr)] >> (nr & (BITS_PER_BYTE-1)));
  80 }
  81
  82 extern void le_bitmap_set(u8 *map, unsigned int start, int len);
  83 extern void le_bitmap_clear(u8 *map, unsigned int start, int len);
  84
  85 struct extent_state;
  86 struct btrfs_root;
  87 struct btrfs_io_bio;
  88 struct io_failure_record;
  89
  90 typedef int (extent_submit_bio_hook_t)(struct inode *inode, struct bio *bio,
  91                                        int mirror_num, unsigned long bio_flags,
  92                                        u64 bio_offset);
  93 struct extent_io_ops {
  94         int (*fill_delalloc)(struct inode *inode, struct page *locked_page,
  95                              u64 start, u64 end, int *page_started,
  96                              unsigned long *nr_written);
  97         int (*writepage_start_hook)(struct page *page, u64 start, u64 end);
  98         extent_submit_bio_hook_t *submit_bio_hook;
  99         int (*merge_bio_hook)(struct page *page, unsigned long offset,
 100                               size_t size, struct bio *bio,
 101                               unsigned long bio_flags);
 102         int (*readpage_io_failed_hook)(struct page *page, int failed_mirror);
 103         int (*readpage_end_io_hook)(struct btrfs_io_bio *io_bio, u64 phy_offset,
 104                                     struct page *page, u64 start, u64 end,
 105                                     int mirror);
 106         int (*writepage_end_io_hook)(struct page *page, u64 start, u64 end,
 107                                       struct extent_state *state, int uptodate);
 108         void (*set_bit_hook)(struct inode *inode, struct extent_state *state,
 109                              unsigned *bits);
 110         void (*clear_bit_hook)(struct inode *inode, struct extent_state *state,
 111                                unsigned *bits);
 112         void (*merge_extent_hook)(struct inode *inode,
 113                                   struct extent_state *new,
 114                                   struct extent_state *other);
 115         void (*split_extent_hook)(struct inode *inode,
 116                                   struct extent_state *orig, u64 split);
 117 };
 118
 119 struct extent_io_tree {
 120         struct rb_root state;
 121         struct address_space *mapping;
 122         u64 dirty_bytes;
 123         int track_uptodate;
 124         spinlock_t lock;
 125         const struct extent_io_ops *ops;
 126 };
 127
 128 struct extent_state {
 129         u64 start;
 130         u64 end; /* inclusive */
 131         struct rb_node rb_node;
 132
 133         /* ADD NEW ELEMENTS AFTER THIS */
 134         wait_queue_head_t wq;
 135         atomic_t refs;
 136         unsigned state;
 137
 138         struct io_failure_record *failrec;
 139
 140 #ifdef CONFIG_BTRFS_DEBUG
 141         struct list_head leak_list;
 142 #endif
 143 };
 144
 145 #define INLINE_EXTENT_BUFFER_PAGES 16
 146 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE)
 147 struct extent_buffer {
 148         u64 start;
 149         unsigned long len;
 150         unsigned long bflags;
 151         struct btrfs_fs_info *fs_info;
 152         spinlock_t refs_lock;
 153         atomic_t refs;
 154         atomic_t io_pages;
 155         int read_mirror;
 156         struct rcu_head rcu_head;
 157         pid_t lock_owner;
 158
 159         /* count of read lock holders on the extent buffer */
 160         atomic_t write_locks;
 161         atomic_t read_locks;
 162         atomic_t blocking_writers;
 163         atomic_t blocking_readers;
 164         atomic_t spinning_readers;
 165         atomic_t spinning_writers;
 166         short lock_nested;
 167         /* >= 0 if eb belongs to a log tree, -1 otherwise */
 168         short log_index;
 169
 170         /* protects write locks */
 171         rwlock_t lock;
 172
 173         /* readers use lock_wq while they wait for the write
 174          * lock holders to unlock
 175          */
 176         wait_queue_head_t write_lock_wq;
 177
 178         /* writers use read_lock_wq while they wait for readers
 179          * to unlock
 180          */
 181         wait_queue_head_t read_lock_wq;
 182         struct page *pages[INLINE_EXTENT_BUFFER_PAGES];
 183 #ifdef CONFIG_BTRFS_DEBUG
 184         struct list_head leak_list;
 185 #endif
 186 };
 187
 188 /*
 189  * Structure to record how many bytes and which ranges are set/cleared
 190  */
 191 struct extent_changeset {
 192         /* How many bytes are set/cleared in this operation */
 193         u64 bytes_changed;
 194
 195         /* Changed ranges */
 196         struct ulist range_changed;
 197 };
 198
 199 static inline void extent_set_compress_type(unsigned long *bio_flags,
 200                                             int compress_type)
 201 {
 202         *bio_flags |= compress_type << EXTENT_BIO_FLAG_SHIFT;
 203 }
 204
 205 static inline int extent_compress_type(unsigned long bio_flags)
 206 {
 207         return bio_flags >> EXTENT_BIO_FLAG_SHIFT;
 208 }
 209
 210 struct extent_map_tree;
 211
 212 typedef struct extent_map *(get_extent_t)(struct inode *inode,
 213                                           struct page *page,
 214                                           size_t pg_offset,
 215                                           u64 start, u64 len,
 216                                           int create);
 217
 218 void extent_io_tree_init(struct extent_io_tree *tree,
 219                          struct address_space *mapping);
 220 int try_release_extent_mapping(struct extent_map_tree *map,
 221                                struct extent_io_tree *tree, struct page *page,
 222                                gfp_t mask);
 223 int try_release_extent_buffer(struct page *page);
 224 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 225                      struct extent_state **cached);
 226
 227 static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
 228 {
 229         return lock_extent_bits(tree, start, end, NULL);
 230 }
 231
 232 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end);
 233 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
 234                           get_extent_t *get_extent, int mirror_num);
 235 int __init extent_io_init(void);
 236 void extent_io_exit(void);
 237
 238 u64 count_range_bits(struct extent_io_tree *tree,
 239                      u64 *start, u64 search_end,
 240                      u64 max_bytes, unsigned bits, int contig);
 241
 242 void free_extent_state(struct extent_state *state);
 243 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
 244                    unsigned bits, int filled,
 245                    struct extent_state *cached_state);
 246 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 247                 unsigned bits, struct extent_changeset *changeset);
 248 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 249                      unsigned bits, int wake, int delete,
 250                      struct extent_state **cached, gfp_t mask);
 251
 252 static inline int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
 253 {
 254         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
 255                                 GFP_NOFS);
 256 }
 257
 258 static inline int unlock_extent_cached(struct extent_io_tree *tree, u64 start,
 259                 u64 end, struct extent_state **cached, gfp_t mask)
 260 {
 261         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
 262                                 mask);
 263 }
 264
 265 static inline int clear_extent_bits(struct extent_io_tree *tree, u64 start,
 266                 u64 end, unsigned bits)
 267 {
 268         int wake = 0;
 269
 270         if (bits & EXTENT_LOCKED)
 271                 wake = 1;
 272
 273         return clear_extent_bit(tree, start, end, bits, wake, 0, NULL,
 274                         GFP_NOFS);
 275 }
 276
 277 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
 278                            unsigned bits, struct extent_changeset *changeset);
 279 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 280                    unsigned bits, u64 *failed_start,
 281                    struct extent_state **cached_state, gfp_t mask);
 282
 283 static inline int set_extent_bits(struct extent_io_tree *tree, u64 start,
 284                 u64 end, unsigned bits)
 285 {
 286         return set_extent_bit(tree, start, end, bits, NULL, NULL, GFP_NOFS);
 287 }
 288
 289 static inline int clear_extent_uptodate(struct extent_io_tree *tree, u64 start,
 290                 u64 end, struct extent_state **cached_state, gfp_t mask)
 291 {
 292         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
 293                                 cached_state, mask);
 294 }
 295
 296 static inline int set_extent_dirty(struct extent_io_tree *tree, u64 start,
 297                 u64 end, gfp_t mask)
 298 {
 299         return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
 300                               NULL, mask);
 301 }
 302
 303 static inline int clear_extent_dirty(struct extent_io_tree *tree, u64 start,
 304                 u64 end)
 305 {
 306         return clear_extent_bit(tree, start, end,
 307                                 EXTENT_DIRTY | EXTENT_DELALLOC |
 308                                 EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
 309 }
 310
 311 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 312                        unsigned bits, unsigned clear_bits,
 313                        struct extent_state **cached_state);
 314
 315 static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
 316                 u64 end, struct extent_state **cached_state)
 317 {
 318         return set_extent_bit(tree, start, end,
 319                               EXTENT_DELALLOC | EXTENT_UPTODATE,
 320                               NULL, cached_state, GFP_NOFS);
 321 }
 322
 323 static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
 324                 u64 end, struct extent_state **cached_state)
 325 {
 326         return set_extent_bit(tree, start, end,
 327                               EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
 328                               NULL, cached_state, GFP_NOFS);
 329 }
 330
 331 static inline int set_extent_new(struct extent_io_tree *tree, u64 start,
 332                 u64 end)
 333 {
 334         return set_extent_bit(tree, start, end, EXTENT_NEW, NULL, NULL,
 335                         GFP_NOFS);
 336 }
 337
 338 static inline int set_extent_uptodate(struct extent_io_tree *tree, u64 start,
 339                 u64 end, struct extent_state **cached_state, gfp_t mask)
 340 {
 341         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
 342                               cached_state, mask);
 343 }
 344
 345 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
 346                           u64 *start_ret, u64 *end_ret, unsigned bits,
 347                           struct extent_state **cached_state);
 348 int extent_invalidatepage(struct extent_io_tree *tree,
 349                           struct page *page, unsigned long offset);
 350 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
 351                           get_extent_t *get_extent,
 352                           struct writeback_control *wbc);
 353 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
 354                               u64 start, u64 end, get_extent_t *get_extent,
 355                               int mode);
 356 int extent_writepages(struct extent_io_tree *tree,
 357                       struct address_space *mapping,
 358                       get_extent_t *get_extent,
 359                       struct writeback_control *wbc);
 360 int btree_write_cache_pages(struct address_space *mapping,
 361                             struct writeback_control *wbc);
 362 int extent_readpages(struct extent_io_tree *tree,
 363                      struct address_space *mapping,
 364                      struct list_head *pages, unsigned nr_pages,
 365                      get_extent_t get_extent);
 366 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 367                 __u64 start, __u64 len, get_extent_t *get_extent);
 368 void set_page_extent_mapped(struct page *page);
 369
 370 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 371                                           u64 start);
 372 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
 373                                                   u64 start, unsigned long len);
 374 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
 375                                                 u64 start);
 376 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src);
 377 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
 378                                          u64 start);
 379 void free_extent_buffer(struct extent_buffer *eb);
 380 void free_extent_buffer_stale(struct extent_buffer *eb);
 381 #define WAIT_NONE       0
 382 #define WAIT_COMPLETE   1
 383 #define WAIT_PAGE_LOCK  2
 384 int read_extent_buffer_pages(struct extent_io_tree *tree,
 385                              struct extent_buffer *eb, int wait,
 386                              get_extent_t *get_extent, int mirror_num);
 387 void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
 388
 389 static inline unsigned long num_extent_pages(u64 start, u64 len)
 390 {
 391         return ((start + len + PAGE_SIZE - 1) >> PAGE_SHIFT) -
 392                 (start >> PAGE_SHIFT);
 393 }
 394
 395 static inline void extent_buffer_get(struct extent_buffer *eb)
 396 {
 397         atomic_inc(&eb->refs);
 398 }
 399
 400 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
 401                           unsigned long start,
 402                           unsigned long len);
 403 void read_extent_buffer(struct extent_buffer *eb, void *dst,
 404                         unsigned long start,
 405                         unsigned long len);
 406 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dst,
 407                                unsigned long start,
 408                                unsigned long len);
 409 void write_extent_buffer_fsid(struct extent_buffer *eb, const void *src);
 410 void write_extent_buffer_chunk_tree_uuid(struct extent_buffer *eb,
 411                 const void *src);
 412 void write_extent_buffer(struct extent_buffer *eb, const void *src,
 413                          unsigned long start, unsigned long len);
 414 void copy_extent_buffer_full(struct extent_buffer *dst,
 415                              struct extent_buffer *src);
 416 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
 417                         unsigned long dst_offset, unsigned long src_offset,
 418                         unsigned long len);
 419 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 420                            unsigned long src_offset, unsigned long len);
 421 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 422                            unsigned long src_offset, unsigned long len);
 423 void memzero_extent_buffer(struct extent_buffer *eb, unsigned long start,
 424                            unsigned long len);
 425 int extent_buffer_test_bit(struct extent_buffer *eb, unsigned long start,
 426                            unsigned long pos);
 427 void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
 428                               unsigned long pos, unsigned long len);
 429 void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
 430                                 unsigned long pos, unsigned long len);
 431 void clear_extent_buffer_dirty(struct extent_buffer *eb);
 432 int set_extent_buffer_dirty(struct extent_buffer *eb);
 433 void set_extent_buffer_uptodate(struct extent_buffer *eb);
 434 void clear_extent_buffer_uptodate(struct extent_buffer *eb);
 435 int extent_buffer_uptodate(struct extent_buffer *eb);
 436 int extent_buffer_under_io(struct extent_buffer *eb);
 437 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset,
 438                       unsigned long min_len, char **map,
 439                       unsigned long *map_start,
 440                       unsigned long *map_len);
 441 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end);
 442 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end);
 443 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
 444                                  u64 delalloc_end, struct page *locked_page,
 445                                  unsigned bits_to_clear,
 446                                  unsigned long page_ops);
 447 struct bio *
 448 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
 449                 gfp_t gfp_flags);
 450 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs);
 451 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask);
 452
 453 struct btrfs_fs_info;
 454
 455 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
 456                       struct page *page, unsigned int pg_offset,
 457                       int mirror_num);
 458 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
 459                      unsigned int pg_offset);
 460 void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
 461 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
 462                          struct extent_buffer *eb, int mirror_num);
 463
 464 /*
 465  * When IO fails, either with EIO or csum verification fails, we
 466  * try other mirrors that might have a good copy of the data.  This
 467  * io_failure_record is used to record state as we go through all the
 468  * mirrors.  If another mirror has good data, the page is set up to date
 469  * and things continue.  If a good mirror can't be found, the original
 470  * bio end_io callback is called to indicate things have failed.
 471  */
 472 struct io_failure_record {
 473         struct page *page;
 474         u64 start;
 475         u64 len;
 476         u64 logical;
 477         unsigned long bio_flags;
 478         int this_mirror;
 479         int failed_mirror;
 480         int in_validation;
 481 };
 482
 483 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end);
 484 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
 485                                 struct io_failure_record **failrec_ret);
 486 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
 487                            struct io_failure_record *failrec, int fail_mirror);
 488 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
 489                                     struct io_failure_record *failrec,
 490                                     struct page *page, int pg_offset, int icsum,
 491                                     bio_end_io_t *endio_func, void *data);
 492 int free_io_failure(struct inode *inode, struct io_failure_record *rec);
 493 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 494 noinline u64 find_lock_delalloc_range(struct inode *inode,
 495                                       struct extent_io_tree *tree,
 496                                       struct page *locked_page, u64 *start,
 497                                       u64 *end, u64 max_bytes);
 498 #endif
 499 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
 500                                                u64 start);
 501 #endif