xref: /openbmc/linux/fs/xfs/xfs_buf.h (revision ae3473231e77a3f1909d48cd144cebe5e1d049b3)
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_NO_IOACCT	 (1 << 3) /* bypass I/O accounting (non-LRU bufs) */
47 #define XBF_ASYNC	 (1 << 4) /* initiator will not wait for completion */
48 #define XBF_DONE	 (1 << 5) /* all pages in the buffer uptodate */
49 #define XBF_STALE	 (1 << 6) /* buffer has been staled, do not find it */
50 #define XBF_WRITE_FAIL	 (1 << 24)/* async writes have failed on this buffer */
51 
52 /* I/O hints for the BIO layer */
53 #define XBF_SYNCIO	 (1 << 10)/* treat this buffer as synchronous I/O */
54 #define XBF_FUA		 (1 << 11)/* force cache write through mode */
55 #define XBF_FLUSH	 (1 << 12)/* flush the disk cache before a write */
56 
57 /* flags used only as arguments to access routines */
58 #define XBF_TRYLOCK	 (1 << 16)/* lock requested, but do not wait */
59 #define XBF_UNMAPPED	 (1 << 17)/* do not map the buffer */
60 
61 /* flags used only internally */
62 #define _XBF_PAGES	 (1 << 20)/* backed by refcounted pages */
63 #define _XBF_KMEM	 (1 << 21)/* backed by heap memory */
64 #define _XBF_DELWRI_Q	 (1 << 22)/* buffer on a delwri queue */
65 #define _XBF_COMPOUND	 (1 << 23)/* compound buffer */
66 #define _XBF_IN_FLIGHT	 (1 << 25) /* I/O in flight, for accounting purposes */
67 
68 typedef unsigned int xfs_buf_flags_t;
69 
70 #define XFS_BUF_FLAGS \
71 	{ XBF_READ,		"READ" }, \
72 	{ XBF_WRITE,		"WRITE" }, \
73 	{ XBF_READ_AHEAD,	"READ_AHEAD" }, \
74 	{ XBF_NO_IOACCT,	"NO_IOACCT" }, \
75 	{ XBF_ASYNC,		"ASYNC" }, \
76 	{ XBF_DONE,		"DONE" }, \
77 	{ XBF_STALE,		"STALE" }, \
78 	{ XBF_WRITE_FAIL,	"WRITE_FAIL" }, \
79 	{ XBF_SYNCIO,		"SYNCIO" }, \
80 	{ XBF_FUA,		"FUA" }, \
81 	{ XBF_FLUSH,		"FLUSH" }, \
82 	{ XBF_TRYLOCK,		"TRYLOCK" },	/* should never be set */\
83 	{ XBF_UNMAPPED,		"UNMAPPED" },	/* ditto */\
84 	{ _XBF_PAGES,		"PAGES" }, \
85 	{ _XBF_KMEM,		"KMEM" }, \
86 	{ _XBF_DELWRI_Q,	"DELWRI_Q" }, \
87 	{ _XBF_COMPOUND,	"COMPOUND" }, \
88 	{ _XBF_IN_FLIGHT,	"IN_FLIGHT" }
89 
90 
91 /*
92  * Internal state flags.
93  */
94 #define XFS_BSTATE_DISPOSE	 (1 << 0)	/* buffer being discarded */
95 
96 /*
97  * The xfs_buftarg contains 2 notions of "sector size" -
98  *
99  * 1) The metadata sector size, which is the minimum unit and
100  *    alignment of IO which will be performed by metadata operations.
101  * 2) The device logical sector size
102  *
103  * The first is specified at mkfs time, and is stored on-disk in the
104  * superblock's sb_sectsize.
105  *
106  * The latter is derived from the underlying device, and controls direct IO
107  * alignment constraints.
108  */
109 typedef struct xfs_buftarg {
110 	dev_t			bt_dev;
111 	struct block_device	*bt_bdev;
112 	struct backing_dev_info	*bt_bdi;
113 	struct xfs_mount	*bt_mount;
114 	unsigned int		bt_meta_sectorsize;
115 	size_t			bt_meta_sectormask;
116 	size_t			bt_logical_sectorsize;
117 	size_t			bt_logical_sectormask;
118 
119 	/* LRU control structures */
120 	struct shrinker		bt_shrinker;
121 	struct list_lru		bt_lru;
122 
123 	struct percpu_counter	bt_io_count;
124 } xfs_buftarg_t;
125 
126 struct xfs_buf;
127 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *);
128 
129 
130 #define XB_PAGES	2
131 
132 struct xfs_buf_map {
133 	xfs_daddr_t		bm_bn;	/* block number for I/O */
134 	int			bm_len;	/* size of I/O */
135 };
136 
137 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
138 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
139 
140 struct xfs_buf_ops {
141 	char *name;
142 	void (*verify_read)(struct xfs_buf *);
143 	void (*verify_write)(struct xfs_buf *);
144 };
145 
146 typedef struct xfs_buf {
147 	/*
148 	 * first cacheline holds all the fields needed for an uncontended cache
149 	 * hit to be fully processed. The semaphore straddles the cacheline
150 	 * boundary, but the counter and lock sits on the first cacheline,
151 	 * which is the only bit that is touched if we hit the semaphore
152 	 * fast-path on locking.
153 	 */
154 	struct rhash_head	b_rhash_head;	/* pag buffer hash node */
155 	xfs_daddr_t		b_bn;		/* block number of buffer */
156 	int			b_length;	/* size of buffer in BBs */
157 	atomic_t		b_hold;		/* reference count */
158 	atomic_t		b_lru_ref;	/* lru reclaim ref count */
159 	xfs_buf_flags_t		b_flags;	/* status flags */
160 	struct semaphore	b_sema;		/* semaphore for lockables */
161 
162 	/*
163 	 * concurrent access to b_lru and b_lru_flags are protected by
164 	 * bt_lru_lock and not by b_sema
165 	 */
166 	struct list_head	b_lru;		/* lru list */
167 	spinlock_t		b_lock;		/* internal state lock */
168 	unsigned int		b_state;	/* internal state flags */
169 	int			b_io_error;	/* internal IO error state */
170 	wait_queue_head_t	b_waiters;	/* unpin waiters */
171 	struct list_head	b_list;
172 	struct xfs_perag	*b_pag;		/* contains rbtree root */
173 	xfs_buftarg_t		*b_target;	/* buffer target (device) */
174 	void			*b_addr;	/* virtual address of buffer */
175 	struct work_struct	b_ioend_work;
176 	struct workqueue_struct	*b_ioend_wq;	/* I/O completion wq */
177 	xfs_buf_iodone_t	b_iodone;	/* I/O completion function */
178 	struct completion	b_iowait;	/* queue for I/O waiters */
179 	void			*b_fspriv;
180 	struct xfs_trans	*b_transp;
181 	struct page		**b_pages;	/* array of page pointers */
182 	struct page		*b_page_array[XB_PAGES]; /* inline pages */
183 	struct xfs_buf_map	*b_maps;	/* compound buffer map */
184 	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
185 	int			b_map_count;
186 	int			b_io_length;	/* IO size in BBs */
187 	atomic_t		b_pin_count;	/* pin count */
188 	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
189 	unsigned int		b_page_count;	/* size of page array */
190 	unsigned int		b_offset;	/* page offset in first page */
191 	int			b_error;	/* error code on I/O */
192 
193 	/*
194 	 * async write failure retry count. Initialised to zero on the first
195 	 * failure, then when it exceeds the maximum configured without a
196 	 * success the write is considered to be failed permanently and the
197 	 * iodone handler will take appropriate action.
198 	 *
199 	 * For retry timeouts, we record the jiffie of the first failure. This
200 	 * means that we can change the retry timeout for buffers already under
201 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
202 	 *
203 	 * last_error is used to ensure that we are getting repeated errors, not
204 	 * different errors. e.g. a block device might change ENOSPC to EIO when
205 	 * a failure timeout occurs, so we want to re-initialise the error
206 	 * retry behaviour appropriately when that happens.
207 	 */
208 	int			b_retries;
209 	unsigned long		b_first_retry_time; /* in jiffies */
210 	int			b_last_error;
211 
212 	const struct xfs_buf_ops	*b_ops;
213 
214 #ifdef XFS_BUF_LOCK_TRACKING
215 	int			b_last_holder;
216 #endif
217 } xfs_buf_t;
218 
219 /* Finding and Reading Buffers */
220 struct xfs_buf *_xfs_buf_find(struct xfs_buftarg *target,
221 			      struct xfs_buf_map *map, int nmaps,
222 			      xfs_buf_flags_t flags, struct xfs_buf *new_bp);
223 
224 static inline struct xfs_buf *
225 xfs_incore(
226 	struct xfs_buftarg	*target,
227 	xfs_daddr_t		blkno,
228 	size_t			numblks,
229 	xfs_buf_flags_t		flags)
230 {
231 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
232 	return _xfs_buf_find(target, &map, 1, flags, NULL);
233 }
234 
235 struct xfs_buf *_xfs_buf_alloc(struct xfs_buftarg *target,
236 			       struct xfs_buf_map *map, int nmaps,
237 			       xfs_buf_flags_t flags);
238 
239 static inline struct xfs_buf *
240 xfs_buf_alloc(
241 	struct xfs_buftarg	*target,
242 	xfs_daddr_t		blkno,
243 	size_t			numblks,
244 	xfs_buf_flags_t		flags)
245 {
246 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
247 	return _xfs_buf_alloc(target, &map, 1, flags);
248 }
249 
250 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target,
251 			       struct xfs_buf_map *map, int nmaps,
252 			       xfs_buf_flags_t flags);
253 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target,
254 			       struct xfs_buf_map *map, int nmaps,
255 			       xfs_buf_flags_t flags,
256 			       const struct xfs_buf_ops *ops);
257 void xfs_buf_readahead_map(struct xfs_buftarg *target,
258 			       struct xfs_buf_map *map, int nmaps,
259 			       const struct xfs_buf_ops *ops);
260 
261 static inline struct xfs_buf *
262 xfs_buf_get(
263 	struct xfs_buftarg	*target,
264 	xfs_daddr_t		blkno,
265 	size_t			numblks,
266 	xfs_buf_flags_t		flags)
267 {
268 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
269 	return xfs_buf_get_map(target, &map, 1, flags);
270 }
271 
272 static inline struct xfs_buf *
273 xfs_buf_read(
274 	struct xfs_buftarg	*target,
275 	xfs_daddr_t		blkno,
276 	size_t			numblks,
277 	xfs_buf_flags_t		flags,
278 	const struct xfs_buf_ops *ops)
279 {
280 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
281 	return xfs_buf_read_map(target, &map, 1, flags, ops);
282 }
283 
284 static inline void
285 xfs_buf_readahead(
286 	struct xfs_buftarg	*target,
287 	xfs_daddr_t		blkno,
288 	size_t			numblks,
289 	const struct xfs_buf_ops *ops)
290 {
291 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
292 	return xfs_buf_readahead_map(target, &map, 1, ops);
293 }
294 
295 struct xfs_buf *xfs_buf_get_empty(struct xfs_buftarg *target, size_t numblks);
296 void xfs_buf_set_empty(struct xfs_buf *bp, size_t numblks);
297 int xfs_buf_associate_memory(struct xfs_buf *bp, void *mem, size_t length);
298 
299 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
300 				int flags);
301 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
302 			  size_t numblks, int flags, struct xfs_buf **bpp,
303 			  const struct xfs_buf_ops *ops);
304 void xfs_buf_hold(struct xfs_buf *bp);
305 
306 /* Releasing Buffers */
307 extern void xfs_buf_free(xfs_buf_t *);
308 extern void xfs_buf_rele(xfs_buf_t *);
309 
310 /* Locking and Unlocking Buffers */
311 extern int xfs_buf_trylock(xfs_buf_t *);
312 extern void xfs_buf_lock(xfs_buf_t *);
313 extern void xfs_buf_unlock(xfs_buf_t *);
314 #define xfs_buf_islocked(bp) \
315 	((bp)->b_sema.count <= 0)
316 
317 /* Buffer Read and Write Routines */
318 extern int xfs_bwrite(struct xfs_buf *bp);
319 extern void xfs_buf_ioend(struct xfs_buf *bp);
320 extern void xfs_buf_ioerror(xfs_buf_t *, int);
321 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func);
322 extern void xfs_buf_submit(struct xfs_buf *bp);
323 extern int xfs_buf_submit_wait(struct xfs_buf *bp);
324 extern void xfs_buf_iomove(xfs_buf_t *, size_t, size_t, void *,
325 				xfs_buf_rw_t);
326 #define xfs_buf_zero(bp, off, len) \
327 	    xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO)
328 
329 /* Buffer Utility Routines */
330 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
331 extern void xfs_buf_stale(struct xfs_buf *bp);
332 
333 /* Delayed Write Buffer Routines */
334 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
335 extern int xfs_buf_delwri_submit(struct list_head *);
336 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
337 
338 /* Buffer Daemon Setup Routines */
339 extern int xfs_buf_init(void);
340 extern void xfs_buf_terminate(void);
341 
342 /*
343  * These macros use the IO block map rather than b_bn. b_bn is now really
344  * just for the buffer cache index for cached buffers. As IO does not use b_bn
345  * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
346  * map directly. Uncached buffers are not allowed to be discontiguous, so this
347  * is safe to do.
348  *
349  * In future, uncached buffers will pass the block number directly to the io
350  * request function and hence these macros will go away at that point.
351  */
352 #define XFS_BUF_ADDR(bp)		((bp)->b_maps[0].bm_bn)
353 #define XFS_BUF_SET_ADDR(bp, bno)	((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
354 
355 static inline void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref)
356 {
357 	atomic_set(&bp->b_lru_ref, lru_ref);
358 }
359 
360 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
361 {
362 	return atomic_read(&bp->b_pin_count);
363 }
364 
365 static inline void xfs_buf_relse(xfs_buf_t *bp)
366 {
367 	xfs_buf_unlock(bp);
368 	xfs_buf_rele(bp);
369 }
370 
371 static inline int
372 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
373 {
374 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
375 				cksum_offset);
376 }
377 
378 static inline void
379 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
380 {
381 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
382 			 cksum_offset);
383 }
384 
385 /*
386  *	Handling of buftargs.
387  */
388 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
389 			struct block_device *);
390 extern void xfs_free_buftarg(struct xfs_mount *, struct xfs_buftarg *);
391 extern void xfs_wait_buftarg(xfs_buftarg_t *);
392 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int);
393 
394 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
395 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
396 
397 #endif	/* __XFS_BUF_H__ */
398