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>
23 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))
25 #define XBF_READ (1 << 0) /* buffer intended for reading from device */
26 #define XBF_WRITE (1 << 1) /* buffer intended for writing to device */
27 #define XBF_READ_AHEAD (1 << 2) /* asynchronous read-ahead */
28 #define XBF_NO_IOACCT (1 << 3) /* bypass I/O accounting (non-LRU bufs) */
29 #define XBF_ASYNC (1 << 4) /* initiator will not wait for completion */
30 #define XBF_DONE (1 << 5) /* all pages in the buffer uptodate */
31 #define XBF_STALE (1 << 6) /* buffer has been staled, do not find it */
32 #define XBF_WRITE_FAIL (1 << 7) /* async writes have failed on this buffer */
34 /* buffer type flags for write callbacks */
35 #define _XBF_INODES (1 << 16)/* inode buffer */
36 #define _XBF_DQUOTS (1 << 17)/* dquot buffer */
37 #define _XBF_LOGRECOVERY (1 << 18)/* log recovery buffer */
39 /* flags used only internally */
40 #define _XBF_PAGES (1 << 20)/* backed by refcounted pages */
41 #define _XBF_KMEM (1 << 21)/* backed by heap memory */
42 #define _XBF_DELWRI_Q (1 << 22)/* buffer on a delwri queue */
44 /* flags used only as arguments to access routines */
45 #define XBF_TRYLOCK (1 << 30)/* lock requested, but do not wait */
46 #define XBF_UNMAPPED (1 << 31)/* do not map the buffer */
48 typedef unsigned int xfs_buf_flags_t;
50 #define XFS_BUF_FLAGS \
51 { XBF_READ, "READ" }, \
52 { XBF_WRITE, "WRITE" }, \
53 { XBF_READ_AHEAD, "READ_AHEAD" }, \
54 { XBF_NO_IOACCT, "NO_IOACCT" }, \
55 { XBF_ASYNC, "ASYNC" }, \
56 { XBF_DONE, "DONE" }, \
57 { XBF_STALE, "STALE" }, \
58 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
59 { _XBF_INODES, "INODES" }, \
60 { _XBF_DQUOTS, "DQUOTS" }, \
61 { _XBF_LOGRECOVERY, "LOG_RECOVERY" }, \
62 { _XBF_PAGES, "PAGES" }, \
63 { _XBF_KMEM, "KMEM" }, \
64 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
65 /* The following interface flags should never be set */ \
66 { XBF_TRYLOCK, "TRYLOCK" }, \
67 { XBF_UNMAPPED, "UNMAPPED" }
70 * Internal state flags.
72 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */
73 #define XFS_BSTATE_IN_FLIGHT (1 << 1) /* I/O in flight */
76 * The xfs_buftarg contains 2 notions of "sector size" -
78 * 1) The metadata sector size, which is the minimum unit and
79 * alignment of IO which will be performed by metadata operations.
80 * 2) The device logical sector size
82 * The first is specified at mkfs time, and is stored on-disk in the
83 * superblock's sb_sectsize.
85 * The latter is derived from the underlying device, and controls direct IO
86 * alignment constraints.
88 typedef struct xfs_buftarg {
90 struct block_device *bt_bdev;
91 struct dax_device *bt_daxdev;
92 struct xfs_mount *bt_mount;
93 unsigned int bt_meta_sectorsize;
94 size_t bt_meta_sectormask;
95 size_t bt_logical_sectorsize;
96 size_t bt_logical_sectormask;
98 /* LRU control structures */
99 struct shrinker bt_shrinker;
100 struct list_lru bt_lru;
102 struct percpu_counter bt_io_count;
103 struct ratelimit_state bt_ioerror_rl;
109 xfs_daddr_t bm_bn; /* block number for I/O */
110 int bm_len; /* size of I/O */
113 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
114 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
119 __be32 magic[2]; /* v4 and v5 on disk magic values */
120 __be16 magic16[2]; /* v4 and v5 on disk magic values */
122 void (*verify_read)(struct xfs_buf *);
123 void (*verify_write)(struct xfs_buf *);
124 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
129 * first cacheline holds all the fields needed for an uncontended cache
130 * hit to be fully processed. The semaphore straddles the cacheline
131 * boundary, but the counter and lock sits on the first cacheline,
132 * which is the only bit that is touched if we hit the semaphore
133 * fast-path on locking.
135 struct rhash_head b_rhash_head; /* pag buffer hash node */
136 xfs_daddr_t b_bn; /* block number of buffer */
137 int b_length; /* size of buffer in BBs */
138 atomic_t b_hold; /* reference count */
139 atomic_t b_lru_ref; /* lru reclaim ref count */
140 xfs_buf_flags_t b_flags; /* status flags */
141 struct semaphore b_sema; /* semaphore for lockables */
144 * concurrent access to b_lru and b_lru_flags are protected by
145 * bt_lru_lock and not by b_sema
147 struct list_head b_lru; /* lru list */
148 spinlock_t b_lock; /* internal state lock */
149 unsigned int b_state; /* internal state flags */
150 int b_io_error; /* internal IO error state */
151 wait_queue_head_t b_waiters; /* unpin waiters */
152 struct list_head b_list;
153 struct xfs_perag *b_pag; /* contains rbtree root */
154 struct xfs_mount *b_mount;
155 struct xfs_buftarg *b_target; /* buffer target (device) */
156 void *b_addr; /* virtual address of buffer */
157 struct work_struct b_ioend_work;
158 struct completion b_iowait; /* queue for I/O waiters */
159 struct xfs_buf_log_item *b_log_item;
160 struct list_head b_li_list; /* Log items list head */
161 struct xfs_trans *b_transp;
162 struct page **b_pages; /* array of page pointers */
163 struct page *b_page_array[XB_PAGES]; /* inline pages */
164 struct xfs_buf_map *b_maps; /* compound buffer map */
165 struct xfs_buf_map __b_map; /* inline compound buffer map */
167 atomic_t b_pin_count; /* pin count */
168 atomic_t b_io_remaining; /* #outstanding I/O requests */
169 unsigned int b_page_count; /* size of page array */
170 unsigned int b_offset; /* page offset of b_addr,
171 only for _XBF_KMEM buffers */
172 int b_error; /* error code on I/O */
175 * async write failure retry count. Initialised to zero on the first
176 * failure, then when it exceeds the maximum configured without a
177 * success the write is considered to be failed permanently and the
178 * iodone handler will take appropriate action.
180 * For retry timeouts, we record the jiffie of the first failure. This
181 * means that we can change the retry timeout for buffers already under
182 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
184 * last_error is used to ensure that we are getting repeated errors, not
185 * different errors. e.g. a block device might change ENOSPC to EIO when
186 * a failure timeout occurs, so we want to re-initialise the error
187 * retry behaviour appropriately when that happens.
190 unsigned long b_first_retry_time; /* in jiffies */
193 const struct xfs_buf_ops *b_ops;
196 /* Finding and Reading Buffers */
197 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
198 xfs_daddr_t blkno, size_t numblks,
199 xfs_buf_flags_t flags);
201 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
202 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
203 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
204 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
205 const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
206 void xfs_buf_readahead_map(struct xfs_buftarg *target,
207 struct xfs_buf_map *map, int nmaps,
208 const struct xfs_buf_ops *ops);
212 struct xfs_buftarg *target,
215 struct xfs_buf **bpp)
217 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
219 return xfs_buf_get_map(target, &map, 1, 0, bpp);
224 struct xfs_buftarg *target,
227 xfs_buf_flags_t flags,
228 struct xfs_buf **bpp,
229 const struct xfs_buf_ops *ops)
231 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
233 return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
234 __builtin_return_address(0));
239 struct xfs_buftarg *target,
242 const struct xfs_buf_ops *ops)
244 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
245 return xfs_buf_readahead_map(target, &map, 1, ops);
248 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks, int flags,
249 struct xfs_buf **bpp);
250 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
251 size_t numblks, int flags, struct xfs_buf **bpp,
252 const struct xfs_buf_ops *ops);
253 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
254 void xfs_buf_hold(struct xfs_buf *bp);
256 /* Releasing Buffers */
257 extern void xfs_buf_rele(struct xfs_buf *);
259 /* Locking and Unlocking Buffers */
260 extern int xfs_buf_trylock(struct xfs_buf *);
261 extern void xfs_buf_lock(struct xfs_buf *);
262 extern void xfs_buf_unlock(struct xfs_buf *);
263 #define xfs_buf_islocked(bp) \
264 ((bp)->b_sema.count <= 0)
266 static inline void xfs_buf_relse(struct xfs_buf *bp)
272 /* Buffer Read and Write Routines */
273 extern int xfs_bwrite(struct xfs_buf *bp);
275 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
276 xfs_failaddr_t failaddr);
277 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
278 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
279 void xfs_buf_ioend_fail(struct xfs_buf *);
280 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
281 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
282 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
284 /* Buffer Utility Routines */
285 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
286 extern void xfs_buf_stale(struct xfs_buf *bp);
288 /* Delayed Write Buffer Routines */
289 extern void xfs_buf_delwri_cancel(struct list_head *);
290 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
291 extern int xfs_buf_delwri_submit(struct list_head *);
292 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
293 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
295 /* Buffer Daemon Setup Routines */
296 extern int xfs_buf_init(void);
297 extern void xfs_buf_terminate(void);
300 * These macros use the IO block map rather than b_bn. b_bn is now really
301 * just for the buffer cache index for cached buffers. As IO does not use b_bn
302 * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
303 * map directly. Uncached buffers are not allowed to be discontiguous, so this
306 * In future, uncached buffers will pass the block number directly to the io
307 * request function and hence these macros will go away at that point.
309 #define XFS_BUF_ADDR(bp) ((bp)->b_maps[0].bm_bn)
310 #define XFS_BUF_SET_ADDR(bp, bno) ((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
312 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
315 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
316 * up with a reference count of 0 so it will be tossed from the cache when
319 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
321 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
323 atomic_set(&bp->b_lru_ref, 0);
326 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
328 return atomic_read(&bp->b_pin_count);
332 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
334 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
339 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
341 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
346 * Handling of buftargs.
348 extern struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *,
349 struct block_device *, struct dax_device *);
350 extern void xfs_free_buftarg(struct xfs_buftarg *);
351 extern void xfs_buftarg_wait(struct xfs_buftarg *);
352 extern void xfs_buftarg_drain(struct xfs_buftarg *);
353 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
355 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev)
356 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
359 xfs_buftarg_dma_alignment(struct xfs_buftarg *bt)
361 return queue_dma_alignment(bt->bt_bdev->bd_disk->queue);
364 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
365 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
366 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
368 #endif /* __XFS_BUF_H__ */