fs/xfs/xfs_buf.h

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #ifndef __XFS_BUF_H__
  19 #define __XFS_BUF_H__
  20
  21 #include <linux/list.h>
  22 #include <linux/types.h>
  23 #include <linux/spinlock.h>
  24 #include <linux/mm.h>
  25 #include <linux/fs.h>
  26 #include <linux/dax.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/uio.h>
  29 #include <linux/list_lru.h>
  30
  31 /*
  32  *      Base types
  33  */
  34
  35 #define XFS_BUF_DADDR_NULL      ((xfs_daddr_t) (-1LL))
  36
  37 typedef enum {
  38         XBRW_READ = 1,                  /* transfer into target memory */
  39         XBRW_WRITE = 2,                 /* transfer from target memory */
  40         XBRW_ZERO = 3,                  /* Zero target memory */
  41 } xfs_buf_rw_t;
  42
  43 #define XBF_READ         (1 << 0) /* buffer intended for reading from device */
  44 #define XBF_WRITE        (1 << 1) /* buffer intended for writing to device */
  45 #define XBF_READ_AHEAD   (1 << 2) /* asynchronous read-ahead */
  46 #define XBF_ASYNC        (1 << 4) /* initiator will not wait for completion */
  47 #define XBF_DONE         (1 << 5) /* all pages in the buffer uptodate */
  48 #define XBF_STALE        (1 << 6) /* buffer has been staled, do not find it */
  49 #define XBF_WRITE_FAIL   (1 << 24)/* async writes have failed on this buffer */
  50
  51 /* I/O hints for the BIO layer */
  52 #define XBF_SYNCIO       (1 << 10)/* treat this buffer as synchronous I/O */
  53 #define XBF_FUA          (1 << 11)/* force cache write through mode */
  54 #define XBF_FLUSH        (1 << 12)/* flush the disk cache before a write */
  55
  56 /* flags used only as arguments to access routines */
  57 #define XBF_TRYLOCK      (1 << 16)/* lock requested, but do not wait */
  58 #define XBF_UNMAPPED     (1 << 17)/* do not map the buffer */
  59
  60 /* flags used only internally */
  61 #define _XBF_PAGES       (1 << 20)/* backed by refcounted pages */
  62 #define _XBF_KMEM        (1 << 21)/* backed by heap memory */
  63 #define _XBF_DELWRI_Q    (1 << 22)/* buffer on a delwri queue */
  64 #define _XBF_COMPOUND    (1 << 23)/* compound buffer */
  65
  66 typedef unsigned int xfs_buf_flags_t;
  67
  68 #define XFS_BUF_FLAGS \
  69         { XBF_READ,             "READ" }, \
  70         { XBF_WRITE,            "WRITE" }, \
  71         { XBF_READ_AHEAD,       "READ_AHEAD" }, \
  72         { XBF_ASYNC,            "ASYNC" }, \
  73         { XBF_DONE,             "DONE" }, \
  74         { XBF_STALE,            "STALE" }, \
  75         { XBF_WRITE_FAIL,       "WRITE_FAIL" }, \
  76         { XBF_SYNCIO,           "SYNCIO" }, \
  77         { XBF_FUA,              "FUA" }, \
  78         { XBF_FLUSH,            "FLUSH" }, \
  79         { XBF_TRYLOCK,          "TRYLOCK" },    /* should never be set */\
  80         { XBF_UNMAPPED,         "UNMAPPED" },   /* ditto */\
  81         { _XBF_PAGES,           "PAGES" }, \
  82         { _XBF_KMEM,            "KMEM" }, \
  83         { _XBF_DELWRI_Q,        "DELWRI_Q" }, \
  84         { _XBF_COMPOUND,        "COMPOUND" }
  85
  86
  87 /*
  88  * Internal state flags.
  89  */
  90 #define XFS_BSTATE_DISPOSE       (1 << 0)       /* buffer being discarded */
  91
  92 /*
  93  * The xfs_buftarg contains 2 notions of "sector size" -
  94  *
  95  * 1) The metadata sector size, which is the minimum unit and
  96  *    alignment of IO which will be performed by metadata operations.
  97  * 2) The device logical sector size
  98  *
  99  * The first is specified at mkfs time, and is stored on-disk in the
 100  * superblock's sb_sectsize.
 101  *
 102  * The latter is derived from the underlying device, and controls direct IO
 103  * alignment constraints.
 104  */
 105 typedef struct xfs_buftarg {
 106         dev_t                   bt_dev;
 107         struct block_device     *bt_bdev;
 108         struct backing_dev_info *bt_bdi;
 109         struct xfs_mount        *bt_mount;
 110         unsigned int            bt_meta_sectorsize;
 111         size_t                  bt_meta_sectormask;
 112         size_t                  bt_logical_sectorsize;
 113         size_t                  bt_logical_sectormask;
 114
 115         /* LRU control structures */
 116         struct shrinker         bt_shrinker;
 117         struct list_lru         bt_lru;
 118 } xfs_buftarg_t;
 119
 120 struct xfs_buf;
 121 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *);
 122
 123
 124 #define XB_PAGES        2
 125
 126 struct xfs_buf_map {
 127         xfs_daddr_t             bm_bn;  /* block number for I/O */
 128         int                     bm_len; /* size of I/O */
 129 };
 130
 131 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
 132         struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
 133
 134 struct xfs_buf_ops {
 135         char *name;
 136         void (*verify_read)(struct xfs_buf *);
 137         void (*verify_write)(struct xfs_buf *);
 138 };
 139
 140 typedef struct xfs_buf {
 141         /*
 142          * first cacheline holds all the fields needed for an uncontended cache
 143          * hit to be fully processed. The semaphore straddles the cacheline
 144          * boundary, but the counter and lock sits on the first cacheline,
 145          * which is the only bit that is touched if we hit the semaphore
 146          * fast-path on locking.
 147          */
 148         struct rb_node          b_rbnode;       /* rbtree node */
 149         xfs_daddr_t             b_bn;           /* block number of buffer */
 150         int                     b_length;       /* size of buffer in BBs */
 151         atomic_t                b_hold;         /* reference count */
 152         atomic_t                b_lru_ref;      /* lru reclaim ref count */
 153         xfs_buf_flags_t         b_flags;        /* status flags */
 154         struct semaphore        b_sema;         /* semaphore for lockables */
 155
 156         /*
 157          * concurrent access to b_lru and b_lru_flags are protected by
 158          * bt_lru_lock and not by b_sema
 159          */
 160         struct list_head        b_lru;          /* lru list */
 161         spinlock_t              b_lock;         /* internal state lock */
 162         unsigned int            b_state;        /* internal state flags */
 163         int                     b_io_error;     /* internal IO error state */
 164         wait_queue_head_t       b_waiters;      /* unpin waiters */
 165         struct list_head        b_list;
 166         struct xfs_perag        *b_pag;         /* contains rbtree root */
 167         xfs_buftarg_t           *b_target;      /* buffer target (device) */
 168         void                    *b_addr;        /* virtual address of buffer */
 169         struct work_struct      b_ioend_work;
 170         struct workqueue_struct *b_ioend_wq;    /* I/O completion wq */
 171         xfs_buf_iodone_t        b_iodone;       /* I/O completion function */
 172         struct completion       b_iowait;       /* queue for I/O waiters */
 173         void                    *b_fspriv;
 174         struct xfs_trans        *b_transp;
 175         struct page             **b_pages;      /* array of page pointers */
 176         struct page             *b_page_array[XB_PAGES]; /* inline pages */
 177         struct xfs_buf_map      *b_maps;        /* compound buffer map */
 178         struct xfs_buf_map      __b_map;        /* inline compound buffer map */
 179         int                     b_map_count;
 180         int                     b_io_length;    /* IO size in BBs */
 181         atomic_t                b_pin_count;    /* pin count */
 182         atomic_t                b_io_remaining; /* #outstanding I/O requests */
 183         unsigned int            b_page_count;   /* size of page array */
 184         unsigned int            b_offset;       /* page offset in first page */
 185         int                     b_error;        /* error code on I/O */
 186
 187         /*
 188          * async write failure retry count. Initialised to zero on the first
 189          * failure, then when it exceeds the maximum configured without a
 190          * success the write is considered to be failed permanently and the
 191          * iodone handler will take appropriate action.
 192          *
 193          * For retry timeouts, we record the jiffie of the first failure. This
 194          * means that we can change the retry timeout for buffers already under
 195          * I/O and thus avoid getting stuck in a retry loop with a long timeout.
 196          *
 197          * last_error is used to ensure that we are getting repeated errors, not
 198          * different errors. e.g. a block device might change ENOSPC to EIO when
 199          * a failure timeout occurs, so we want to re-initialise the error
 200          * retry behaviour appropriately when that happens.
 201          */
 202         int                     b_retries;
 203         unsigned long           b_first_retry_time; /* in jiffies */
 204         int                     b_last_error;
 205
 206         const struct xfs_buf_ops        *b_ops;
 207
 208 #ifdef XFS_BUF_LOCK_TRACKING
 209         int                     b_last_holder;
 210 #endif
 211 } xfs_buf_t;
 212
 213 /* Finding and Reading Buffers */
 214 struct xfs_buf *_xfs_buf_find(struct xfs_buftarg *target,
 215                               struct xfs_buf_map *map, int nmaps,
 216                               xfs_buf_flags_t flags, struct xfs_buf *new_bp);
 217
 218 static inline struct xfs_buf *
 219 xfs_incore(
 220         struct xfs_buftarg      *target,
 221         xfs_daddr_t             blkno,
 222         size_t                  numblks,
 223         xfs_buf_flags_t         flags)
 224 {
 225         DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
 226         return _xfs_buf_find(target, &map, 1, flags, NULL);
 227 }
 228
 229 struct xfs_buf *_xfs_buf_alloc(struct xfs_buftarg *target,
 230                                struct xfs_buf_map *map, int nmaps,
 231                                xfs_buf_flags_t flags);
 232
 233 static inline struct xfs_buf *
 234 xfs_buf_alloc(
 235         struct xfs_buftarg      *target,
 236         xfs_daddr_t             blkno,
 237         size_t                  numblks,
 238         xfs_buf_flags_t         flags)
 239 {
 240         DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
 241         return _xfs_buf_alloc(target, &map, 1, flags);
 242 }
 243
 244 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target,
 245                                struct xfs_buf_map *map, int nmaps,
 246                                xfs_buf_flags_t flags);
 247 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target,
 248                                struct xfs_buf_map *map, int nmaps,
 249                                xfs_buf_flags_t flags,
 250                                const struct xfs_buf_ops *ops);
 251 void xfs_buf_readahead_map(struct xfs_buftarg *target,
 252                                struct xfs_buf_map *map, int nmaps,
 253                                const struct xfs_buf_ops *ops);
 254
 255 static inline struct xfs_buf *
 256 xfs_buf_get(
 257         struct xfs_buftarg      *target,
 258         xfs_daddr_t             blkno,
 259         size_t                  numblks,
 260         xfs_buf_flags_t         flags)
 261 {
 262         DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
 263         return xfs_buf_get_map(target, &map, 1, flags);
 264 }
 265
 266 static inline struct xfs_buf *
 267 xfs_buf_read(
 268         struct xfs_buftarg      *target,
 269         xfs_daddr_t             blkno,
 270         size_t                  numblks,
 271         xfs_buf_flags_t         flags,
 272         const struct xfs_buf_ops *ops)
 273 {
 274         DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
 275         return xfs_buf_read_map(target, &map, 1, flags, ops);
 276 }
 277
 278 static inline void
 279 xfs_buf_readahead(
 280         struct xfs_buftarg      *target,
 281         xfs_daddr_t             blkno,
 282         size_t                  numblks,
 283         const struct xfs_buf_ops *ops)
 284 {
 285         DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
 286         return xfs_buf_readahead_map(target, &map, 1, ops);
 287 }
 288
 289 struct xfs_buf *xfs_buf_get_empty(struct xfs_buftarg *target, size_t numblks);
 290 void xfs_buf_set_empty(struct xfs_buf *bp, size_t numblks);
 291 int xfs_buf_associate_memory(struct xfs_buf *bp, void *mem, size_t length);
 292
 293 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
 294                                 int flags);
 295 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
 296                           size_t numblks, int flags, struct xfs_buf **bpp,
 297                           const struct xfs_buf_ops *ops);
 298 void xfs_buf_hold(struct xfs_buf *bp);
 299
 300 /* Releasing Buffers */
 301 extern void xfs_buf_free(xfs_buf_t *);
 302 extern void xfs_buf_rele(xfs_buf_t *);
 303
 304 /* Locking and Unlocking Buffers */
 305 extern int xfs_buf_trylock(xfs_buf_t *);
 306 extern void xfs_buf_lock(xfs_buf_t *);
 307 extern void xfs_buf_unlock(xfs_buf_t *);
 308 #define xfs_buf_islocked(bp) \
 309         ((bp)->b_sema.count <= 0)
 310
 311 /* Buffer Read and Write Routines */
 312 extern int xfs_bwrite(struct xfs_buf *bp);
 313 extern void xfs_buf_ioend(struct xfs_buf *bp);
 314 extern void xfs_buf_ioerror(xfs_buf_t *, int);
 315 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func);
 316 extern void xfs_buf_submit(struct xfs_buf *bp);
 317 extern int xfs_buf_submit_wait(struct xfs_buf *bp);
 318 extern void xfs_buf_iomove(xfs_buf_t *, size_t, size_t, void *,
 319                                 xfs_buf_rw_t);
 320 #define xfs_buf_zero(bp, off, len) \
 321             xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO)
 322
 323 /* Buffer Utility Routines */
 324 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
 325 extern void xfs_buf_stale(struct xfs_buf *bp);
 326
 327 /* Delayed Write Buffer Routines */
 328 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
 329 extern int xfs_buf_delwri_submit(struct list_head *);
 330 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
 331
 332 /* Buffer Daemon Setup Routines */
 333 extern int xfs_buf_init(void);
 334 extern void xfs_buf_terminate(void);
 335
 336 /*
 337  * These macros use the IO block map rather than b_bn. b_bn is now really
 338  * just for the buffer cache index for cached buffers. As IO does not use b_bn
 339  * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
 340  * map directly. Uncached buffers are not allowed to be discontiguous, so this
 341  * is safe to do.
 342  *
 343  * In future, uncached buffers will pass the block number directly to the io
 344  * request function and hence these macros will go away at that point.
 345  */
 346 #define XFS_BUF_ADDR(bp)                ((bp)->b_maps[0].bm_bn)
 347 #define XFS_BUF_SET_ADDR(bp, bno)       ((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
 348
 349 static inline void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref)
 350 {
 351         atomic_set(&bp->b_lru_ref, lru_ref);
 352 }
 353
 354 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
 355 {
 356         return atomic_read(&bp->b_pin_count);
 357 }
 358
 359 static inline void xfs_buf_relse(xfs_buf_t *bp)
 360 {
 361         xfs_buf_unlock(bp);
 362         xfs_buf_rele(bp);
 363 }
 364
 365 static inline int
 366 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
 367 {
 368         return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 369                                 cksum_offset);
 370 }
 371
 372 static inline void
 373 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
 374 {
 375         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
 376                          cksum_offset);
 377 }
 378
 379 /*
 380  *      Handling of buftargs.
 381  */
 382 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
 383                         struct block_device *);
 384 extern void xfs_free_buftarg(struct xfs_mount *, struct xfs_buftarg *);
 385 extern void xfs_wait_buftarg(xfs_buftarg_t *);
 386 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int);
 387
 388 #define xfs_getsize_buftarg(buftarg)    block_size((buftarg)->bt_bdev)
 389 #define xfs_readonly_buftarg(buftarg)   bdev_read_only((buftarg)->bt_bdev)
 390
 391 #endif  /* __XFS_BUF_H__ */