xref: /openbmc/linux/fs/xfs/xfs_buf.h (revision 6396bb221514d2876fd6dc0aa2a1f240d99b37bb)
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 
67 typedef unsigned int xfs_buf_flags_t;
68 
69 #define XFS_BUF_FLAGS \
70 	{ XBF_READ,		"READ" }, \
71 	{ XBF_WRITE,		"WRITE" }, \
72 	{ XBF_READ_AHEAD,	"READ_AHEAD" }, \
73 	{ XBF_NO_IOACCT,	"NO_IOACCT" }, \
74 	{ XBF_ASYNC,		"ASYNC" }, \
75 	{ XBF_DONE,		"DONE" }, \
76 	{ XBF_STALE,		"STALE" }, \
77 	{ XBF_WRITE_FAIL,	"WRITE_FAIL" }, \
78 	{ XBF_SYNCIO,		"SYNCIO" }, \
79 	{ XBF_FUA,		"FUA" }, \
80 	{ XBF_FLUSH,		"FLUSH" }, \
81 	{ XBF_TRYLOCK,		"TRYLOCK" },	/* should never be set */\
82 	{ XBF_UNMAPPED,		"UNMAPPED" },	/* ditto */\
83 	{ _XBF_PAGES,		"PAGES" }, \
84 	{ _XBF_KMEM,		"KMEM" }, \
85 	{ _XBF_DELWRI_Q,	"DELWRI_Q" }, \
86 	{ _XBF_COMPOUND,	"COMPOUND" }
87 
88 
89 /*
90  * Internal state flags.
91  */
92 #define XFS_BSTATE_DISPOSE	 (1 << 0)	/* buffer being discarded */
93 #define XFS_BSTATE_IN_FLIGHT	 (1 << 1)	/* I/O in flight */
94 
95 /*
96  * The xfs_buftarg contains 2 notions of "sector size" -
97  *
98  * 1) The metadata sector size, which is the minimum unit and
99  *    alignment of IO which will be performed by metadata operations.
100  * 2) The device logical sector size
101  *
102  * The first is specified at mkfs time, and is stored on-disk in the
103  * superblock's sb_sectsize.
104  *
105  * The latter is derived from the underlying device, and controls direct IO
106  * alignment constraints.
107  */
108 typedef struct xfs_buftarg {
109 	dev_t			bt_dev;
110 	struct block_device	*bt_bdev;
111 	struct dax_device	*bt_daxdev;
112 	struct xfs_mount	*bt_mount;
113 	unsigned int		bt_meta_sectorsize;
114 	size_t			bt_meta_sectormask;
115 	size_t			bt_logical_sectorsize;
116 	size_t			bt_logical_sectormask;
117 
118 	/* LRU control structures */
119 	struct shrinker		bt_shrinker;
120 	struct list_lru		bt_lru;
121 
122 	struct percpu_counter	bt_io_count;
123 } xfs_buftarg_t;
124 
125 struct xfs_buf;
126 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *);
127 
128 
129 #define XB_PAGES	2
130 
131 struct xfs_buf_map {
132 	xfs_daddr_t		bm_bn;	/* block number for I/O */
133 	int			bm_len;	/* size of I/O */
134 };
135 
136 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
137 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
138 
139 struct xfs_buf_ops {
140 	char *name;
141 	void (*verify_read)(struct xfs_buf *);
142 	void (*verify_write)(struct xfs_buf *);
143 	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
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_log_item;
180 	struct list_head	b_li_list;	/* Log items list head */
181 	struct xfs_trans	*b_transp;
182 	struct page		**b_pages;	/* array of page pointers */
183 	struct page		*b_page_array[XB_PAGES]; /* inline pages */
184 	struct xfs_buf_map	*b_maps;	/* compound buffer map */
185 	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
186 	int			b_map_count;
187 	int			b_io_length;	/* IO size in BBs */
188 	atomic_t		b_pin_count;	/* pin count */
189 	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
190 	unsigned int		b_page_count;	/* size of page array */
191 	unsigned int		b_offset;	/* page offset in first page */
192 	int			b_error;	/* error code on I/O */
193 
194 	/*
195 	 * async write failure retry count. Initialised to zero on the first
196 	 * failure, then when it exceeds the maximum configured without a
197 	 * success the write is considered to be failed permanently and the
198 	 * iodone handler will take appropriate action.
199 	 *
200 	 * For retry timeouts, we record the jiffie of the first failure. This
201 	 * means that we can change the retry timeout for buffers already under
202 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
203 	 *
204 	 * last_error is used to ensure that we are getting repeated errors, not
205 	 * different errors. e.g. a block device might change ENOSPC to EIO when
206 	 * a failure timeout occurs, so we want to re-initialise the error
207 	 * retry behaviour appropriately when that happens.
208 	 */
209 	int			b_retries;
210 	unsigned long		b_first_retry_time; /* in jiffies */
211 	int			b_last_error;
212 
213 	const struct xfs_buf_ops	*b_ops;
214 
215 #ifdef XFS_BUF_LOCK_TRACKING
216 	int			b_last_holder;
217 #endif
218 } xfs_buf_t;
219 
220 /* Finding and Reading Buffers */
221 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
222 			   xfs_daddr_t blkno, size_t numblks,
223 			   xfs_buf_flags_t flags);
224 
225 struct xfs_buf *_xfs_buf_alloc(struct xfs_buftarg *target,
226 			       struct xfs_buf_map *map, int nmaps,
227 			       xfs_buf_flags_t flags);
228 
229 static inline struct xfs_buf *
230 xfs_buf_alloc(
231 	struct xfs_buftarg	*target,
232 	xfs_daddr_t		blkno,
233 	size_t			numblks,
234 	xfs_buf_flags_t		flags)
235 {
236 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
237 	return _xfs_buf_alloc(target, &map, 1, flags);
238 }
239 
240 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target,
241 			       struct xfs_buf_map *map, int nmaps,
242 			       xfs_buf_flags_t flags);
243 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target,
244 			       struct xfs_buf_map *map, int nmaps,
245 			       xfs_buf_flags_t flags,
246 			       const struct xfs_buf_ops *ops);
247 void xfs_buf_readahead_map(struct xfs_buftarg *target,
248 			       struct xfs_buf_map *map, int nmaps,
249 			       const struct xfs_buf_ops *ops);
250 
251 static inline struct xfs_buf *
252 xfs_buf_get(
253 	struct xfs_buftarg	*target,
254 	xfs_daddr_t		blkno,
255 	size_t			numblks,
256 	xfs_buf_flags_t		flags)
257 {
258 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
259 	return xfs_buf_get_map(target, &map, 1, flags);
260 }
261 
262 static inline struct xfs_buf *
263 xfs_buf_read(
264 	struct xfs_buftarg	*target,
265 	xfs_daddr_t		blkno,
266 	size_t			numblks,
267 	xfs_buf_flags_t		flags,
268 	const struct xfs_buf_ops *ops)
269 {
270 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
271 	return xfs_buf_read_map(target, &map, 1, flags, ops);
272 }
273 
274 static inline void
275 xfs_buf_readahead(
276 	struct xfs_buftarg	*target,
277 	xfs_daddr_t		blkno,
278 	size_t			numblks,
279 	const struct xfs_buf_ops *ops)
280 {
281 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
282 	return xfs_buf_readahead_map(target, &map, 1, ops);
283 }
284 
285 void xfs_buf_set_empty(struct xfs_buf *bp, size_t numblks);
286 int xfs_buf_associate_memory(struct xfs_buf *bp, void *mem, size_t length);
287 
288 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
289 				int flags);
290 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
291 			  size_t numblks, int flags, struct xfs_buf **bpp,
292 			  const struct xfs_buf_ops *ops);
293 void xfs_buf_hold(struct xfs_buf *bp);
294 
295 /* Releasing Buffers */
296 extern void xfs_buf_free(xfs_buf_t *);
297 extern void xfs_buf_rele(xfs_buf_t *);
298 
299 /* Locking and Unlocking Buffers */
300 extern int xfs_buf_trylock(xfs_buf_t *);
301 extern void xfs_buf_lock(xfs_buf_t *);
302 extern void xfs_buf_unlock(xfs_buf_t *);
303 #define xfs_buf_islocked(bp) \
304 	((bp)->b_sema.count <= 0)
305 
306 /* Buffer Read and Write Routines */
307 extern int xfs_bwrite(struct xfs_buf *bp);
308 extern void xfs_buf_ioend(struct xfs_buf *bp);
309 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
310 		xfs_failaddr_t failaddr);
311 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
312 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func);
313 extern void xfs_buf_submit(struct xfs_buf *bp);
314 extern int xfs_buf_submit_wait(struct xfs_buf *bp);
315 extern void xfs_buf_iomove(xfs_buf_t *, size_t, size_t, void *,
316 				xfs_buf_rw_t);
317 #define xfs_buf_zero(bp, off, len) \
318 	    xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO)
319 
320 /* Buffer Utility Routines */
321 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
322 extern void xfs_buf_stale(struct xfs_buf *bp);
323 
324 /* Delayed Write Buffer Routines */
325 extern void xfs_buf_delwri_cancel(struct list_head *);
326 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
327 extern int xfs_buf_delwri_submit(struct list_head *);
328 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
329 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
330 
331 /* Buffer Daemon Setup Routines */
332 extern int xfs_buf_init(void);
333 extern void xfs_buf_terminate(void);
334 
335 /*
336  * These macros use the IO block map rather than b_bn. b_bn is now really
337  * just for the buffer cache index for cached buffers. As IO does not use b_bn
338  * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
339  * map directly. Uncached buffers are not allowed to be discontiguous, so this
340  * is safe to do.
341  *
342  * In future, uncached buffers will pass the block number directly to the io
343  * request function and hence these macros will go away at that point.
344  */
345 #define XFS_BUF_ADDR(bp)		((bp)->b_maps[0].bm_bn)
346 #define XFS_BUF_SET_ADDR(bp, bno)	((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
347 
348 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
349 
350 /*
351  * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
352  * up with a reference count of 0 so it will be tossed from the cache when
353  * released.
354  */
355 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
356 {
357 	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
358 		return;
359 	atomic_set(&bp->b_lru_ref, 0);
360 }
361 
362 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
363 {
364 	return atomic_read(&bp->b_pin_count);
365 }
366 
367 static inline void xfs_buf_relse(xfs_buf_t *bp)
368 {
369 	xfs_buf_unlock(bp);
370 	xfs_buf_rele(bp);
371 }
372 
373 static inline int
374 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
375 {
376 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
377 				cksum_offset);
378 }
379 
380 static inline void
381 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
382 {
383 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
384 			 cksum_offset);
385 }
386 
387 /*
388  *	Handling of buftargs.
389  */
390 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
391 			struct block_device *, struct dax_device *);
392 extern void xfs_free_buftarg(struct xfs_buftarg *);
393 extern void xfs_wait_buftarg(xfs_buftarg_t *);
394 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int);
395 
396 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
397 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
398 
399 #endif	/* __XFS_BUF_H__ */
400