1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
9 #include <linux/list.h>
10 #include <linux/types.h>
11 #include <linux/spinlock.h>
14 #include <linux/dax.h>
15 #include <linux/uio.h>
16 #include <linux/list_lru.h>
22 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))
25 XBRW_READ = 1, /* transfer into target memory */
26 XBRW_WRITE = 2, /* transfer from target memory */
27 XBRW_ZERO = 3, /* Zero target memory */
30 #define XBF_READ (1 << 0) /* buffer intended for reading from device */
31 #define XBF_WRITE (1 << 1) /* buffer intended for writing to device */
32 #define XBF_READ_AHEAD (1 << 2) /* asynchronous read-ahead */
33 #define XBF_NO_IOACCT (1 << 3) /* bypass I/O accounting (non-LRU bufs) */
34 #define XBF_ASYNC (1 << 4) /* initiator will not wait for completion */
35 #define XBF_DONE (1 << 5) /* all pages in the buffer uptodate */
36 #define XBF_STALE (1 << 6) /* buffer has been staled, do not find it */
37 #define XBF_WRITE_FAIL (1 << 24)/* async writes have failed on this buffer */
39 /* I/O hints for the BIO layer */
40 #define XBF_SYNCIO (1 << 10)/* treat this buffer as synchronous I/O */
41 #define XBF_FUA (1 << 11)/* force cache write through mode */
42 #define XBF_FLUSH (1 << 12)/* flush the disk cache before a write */
44 /* flags used only as arguments to access routines */
45 #define XBF_TRYLOCK (1 << 16)/* lock requested, but do not wait */
46 #define XBF_UNMAPPED (1 << 17)/* do not map the buffer */
48 /* flags used only internally */
49 #define _XBF_PAGES (1 << 20)/* backed by refcounted pages */
50 #define _XBF_KMEM (1 << 21)/* backed by heap memory */
51 #define _XBF_DELWRI_Q (1 << 22)/* buffer on a delwri queue */
52 #define _XBF_COMPOUND (1 << 23)/* compound buffer */
54 typedef unsigned int xfs_buf_flags_t;
56 #define XFS_BUF_FLAGS \
57 { XBF_READ, "READ" }, \
58 { XBF_WRITE, "WRITE" }, \
59 { XBF_READ_AHEAD, "READ_AHEAD" }, \
60 { XBF_NO_IOACCT, "NO_IOACCT" }, \
61 { XBF_ASYNC, "ASYNC" }, \
62 { XBF_DONE, "DONE" }, \
63 { XBF_STALE, "STALE" }, \
64 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
65 { XBF_SYNCIO, "SYNCIO" }, \
67 { XBF_FLUSH, "FLUSH" }, \
68 { XBF_TRYLOCK, "TRYLOCK" }, /* should never be set */\
69 { XBF_UNMAPPED, "UNMAPPED" }, /* ditto */\
70 { _XBF_PAGES, "PAGES" }, \
71 { _XBF_KMEM, "KMEM" }, \
72 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
73 { _XBF_COMPOUND, "COMPOUND" }
77 * Internal state flags.
79 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */
80 #define XFS_BSTATE_IN_FLIGHT (1 << 1) /* I/O in flight */
83 * The xfs_buftarg contains 2 notions of "sector size" -
85 * 1) The metadata sector size, which is the minimum unit and
86 * alignment of IO which will be performed by metadata operations.
87 * 2) The device logical sector size
89 * The first is specified at mkfs time, and is stored on-disk in the
90 * superblock's sb_sectsize.
92 * The latter is derived from the underlying device, and controls direct IO
93 * alignment constraints.
95 typedef struct xfs_buftarg {
97 struct block_device *bt_bdev;
98 struct dax_device *bt_daxdev;
99 struct xfs_mount *bt_mount;
100 unsigned int bt_meta_sectorsize;
101 size_t bt_meta_sectormask;
102 size_t bt_logical_sectorsize;
103 size_t bt_logical_sectormask;
105 /* LRU control structures */
106 struct shrinker bt_shrinker;
107 struct list_lru bt_lru;
109 struct percpu_counter bt_io_count;
113 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *);
119 xfs_daddr_t bm_bn; /* block number for I/O */
120 int bm_len; /* size of I/O */
123 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
124 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
128 uint32_t magic[2]; /* v4 and v5 on disk magic values */
129 void (*verify_read)(struct xfs_buf *);
130 void (*verify_write)(struct xfs_buf *);
131 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
134 typedef struct xfs_buf {
136 * first cacheline holds all the fields needed for an uncontended cache
137 * hit to be fully processed. The semaphore straddles the cacheline
138 * boundary, but the counter and lock sits on the first cacheline,
139 * which is the only bit that is touched if we hit the semaphore
140 * fast-path on locking.
142 struct rhash_head b_rhash_head; /* pag buffer hash node */
143 xfs_daddr_t b_bn; /* block number of buffer */
144 int b_length; /* size of buffer in BBs */
145 atomic_t b_hold; /* reference count */
146 atomic_t b_lru_ref; /* lru reclaim ref count */
147 xfs_buf_flags_t b_flags; /* status flags */
148 struct semaphore b_sema; /* semaphore for lockables */
151 * concurrent access to b_lru and b_lru_flags are protected by
152 * bt_lru_lock and not by b_sema
154 struct list_head b_lru; /* lru list */
155 spinlock_t b_lock; /* internal state lock */
156 unsigned int b_state; /* internal state flags */
157 int b_io_error; /* internal IO error state */
158 wait_queue_head_t b_waiters; /* unpin waiters */
159 struct list_head b_list;
160 struct xfs_perag *b_pag; /* contains rbtree root */
161 xfs_buftarg_t *b_target; /* buffer target (device) */
162 void *b_addr; /* virtual address of buffer */
163 struct work_struct b_ioend_work;
164 struct workqueue_struct *b_ioend_wq; /* I/O completion wq */
165 xfs_buf_iodone_t b_iodone; /* I/O completion function */
166 struct completion b_iowait; /* queue for I/O waiters */
168 struct list_head b_li_list; /* Log items list head */
169 struct xfs_trans *b_transp;
170 struct page **b_pages; /* array of page pointers */
171 struct page *b_page_array[XB_PAGES]; /* inline pages */
172 struct xfs_buf_map *b_maps; /* compound buffer map */
173 struct xfs_buf_map __b_map; /* inline compound buffer map */
175 int b_io_length; /* IO size in BBs */
176 atomic_t b_pin_count; /* pin count */
177 atomic_t b_io_remaining; /* #outstanding I/O requests */
178 unsigned int b_page_count; /* size of page array */
179 unsigned int b_offset; /* page offset in first page */
180 int b_error; /* error code on I/O */
183 * async write failure retry count. Initialised to zero on the first
184 * failure, then when it exceeds the maximum configured without a
185 * success the write is considered to be failed permanently and the
186 * iodone handler will take appropriate action.
188 * For retry timeouts, we record the jiffie of the first failure. This
189 * means that we can change the retry timeout for buffers already under
190 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
192 * last_error is used to ensure that we are getting repeated errors, not
193 * different errors. e.g. a block device might change ENOSPC to EIO when
194 * a failure timeout occurs, so we want to re-initialise the error
195 * retry behaviour appropriately when that happens.
198 unsigned long b_first_retry_time; /* in jiffies */
201 const struct xfs_buf_ops *b_ops;
204 /* Finding and Reading Buffers */
205 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
206 xfs_daddr_t blkno, size_t numblks,
207 xfs_buf_flags_t flags);
209 struct xfs_buf *_xfs_buf_alloc(struct xfs_buftarg *target,
210 struct xfs_buf_map *map, int nmaps,
211 xfs_buf_flags_t flags);
213 static inline struct xfs_buf *
215 struct xfs_buftarg *target,
218 xfs_buf_flags_t flags)
220 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
221 return _xfs_buf_alloc(target, &map, 1, flags);
224 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target,
225 struct xfs_buf_map *map, int nmaps,
226 xfs_buf_flags_t flags);
227 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target,
228 struct xfs_buf_map *map, int nmaps,
229 xfs_buf_flags_t flags,
230 const struct xfs_buf_ops *ops);
231 void xfs_buf_readahead_map(struct xfs_buftarg *target,
232 struct xfs_buf_map *map, int nmaps,
233 const struct xfs_buf_ops *ops);
235 static inline struct xfs_buf *
237 struct xfs_buftarg *target,
240 xfs_buf_flags_t flags)
242 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
243 return xfs_buf_get_map(target, &map, 1, flags);
246 static inline struct xfs_buf *
248 struct xfs_buftarg *target,
251 xfs_buf_flags_t flags,
252 const struct xfs_buf_ops *ops)
254 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
255 return xfs_buf_read_map(target, &map, 1, flags, ops);
260 struct xfs_buftarg *target,
263 const struct xfs_buf_ops *ops)
265 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
266 return xfs_buf_readahead_map(target, &map, 1, ops);
269 void xfs_buf_set_empty(struct xfs_buf *bp, size_t numblks);
270 int xfs_buf_associate_memory(struct xfs_buf *bp, void *mem, size_t length);
272 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
274 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
275 size_t numblks, int flags, struct xfs_buf **bpp,
276 const struct xfs_buf_ops *ops);
277 void xfs_buf_hold(struct xfs_buf *bp);
279 /* Releasing Buffers */
280 extern void xfs_buf_free(xfs_buf_t *);
281 extern void xfs_buf_rele(xfs_buf_t *);
283 /* Locking and Unlocking Buffers */
284 extern int xfs_buf_trylock(xfs_buf_t *);
285 extern void xfs_buf_lock(xfs_buf_t *);
286 extern void xfs_buf_unlock(xfs_buf_t *);
287 #define xfs_buf_islocked(bp) \
288 ((bp)->b_sema.count <= 0)
290 /* Buffer Read and Write Routines */
291 extern int xfs_bwrite(struct xfs_buf *bp);
292 extern void xfs_buf_ioend(struct xfs_buf *bp);
293 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
294 xfs_failaddr_t failaddr);
295 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
296 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func);
298 extern int __xfs_buf_submit(struct xfs_buf *bp, bool);
299 static inline int xfs_buf_submit(struct xfs_buf *bp)
301 bool wait = bp->b_flags & XBF_ASYNC ? false : true;
302 return __xfs_buf_submit(bp, wait);
305 extern void xfs_buf_iomove(xfs_buf_t *, size_t, size_t, void *,
307 #define xfs_buf_zero(bp, off, len) \
308 xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO)
310 /* Buffer Utility Routines */
311 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
312 extern void xfs_buf_stale(struct xfs_buf *bp);
314 /* Delayed Write Buffer Routines */
315 extern void xfs_buf_delwri_cancel(struct list_head *);
316 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
317 extern int xfs_buf_delwri_submit(struct list_head *);
318 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
319 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
321 /* Buffer Daemon Setup Routines */
322 extern int xfs_buf_init(void);
323 extern void xfs_buf_terminate(void);
326 * These macros use the IO block map rather than b_bn. b_bn is now really
327 * just for the buffer cache index for cached buffers. As IO does not use b_bn
328 * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
329 * map directly. Uncached buffers are not allowed to be discontiguous, so this
332 * In future, uncached buffers will pass the block number directly to the io
333 * request function and hence these macros will go away at that point.
335 #define XFS_BUF_ADDR(bp) ((bp)->b_maps[0].bm_bn)
336 #define XFS_BUF_SET_ADDR(bp, bno) ((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
338 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
341 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
342 * up with a reference count of 0 so it will be tossed from the cache when
345 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
347 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
349 atomic_set(&bp->b_lru_ref, 0);
352 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
354 return atomic_read(&bp->b_pin_count);
357 static inline void xfs_buf_relse(xfs_buf_t *bp)
364 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
366 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
371 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
373 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
378 * Handling of buftargs.
380 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
381 struct block_device *, struct dax_device *);
382 extern void xfs_free_buftarg(struct xfs_buftarg *);
383 extern void xfs_wait_buftarg(xfs_buftarg_t *);
384 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int);
386 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev)
387 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
389 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
390 bool xfs_verify_magic(struct xfs_buf *bp, uint32_t dmagic);
392 #endif /* __XFS_BUF_H__ */