xref: /openbmc/linux/fs/xfs/xfs_buf.h (revision 27ab1c1c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #ifndef __XFS_BUF_H__
7 #define __XFS_BUF_H__
8 
9 #include <linux/list.h>
10 #include <linux/types.h>
11 #include <linux/spinlock.h>
12 #include <linux/mm.h>
13 #include <linux/fs.h>
14 #include <linux/dax.h>
15 #include <linux/uio.h>
16 #include <linux/list_lru.h>
17 
18 /*
19  *	Base types
20  */
21 struct xfs_buf;
22 
23 #define XFS_BUF_DADDR_NULL	((xfs_daddr_t) (-1LL))
24 
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 */
33 
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 */
38 
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 */
43 
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 */
47 
48 typedef unsigned int xfs_buf_flags_t;
49 
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" }
68 
69 /*
70  * Internal state flags.
71  */
72 #define XFS_BSTATE_DISPOSE	 (1 << 0)	/* buffer being discarded */
73 #define XFS_BSTATE_IN_FLIGHT	 (1 << 1)	/* I/O in flight */
74 
75 /*
76  * The xfs_buftarg contains 2 notions of "sector size" -
77  *
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
81  *
82  * The first is specified at mkfs time, and is stored on-disk in the
83  * superblock's sb_sectsize.
84  *
85  * The latter is derived from the underlying device, and controls direct IO
86  * alignment constraints.
87  */
88 typedef struct xfs_buftarg {
89 	dev_t			bt_dev;
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;
97 
98 	/* LRU control structures */
99 	struct shrinker		bt_shrinker;
100 	struct list_lru		bt_lru;
101 
102 	struct percpu_counter	bt_io_count;
103 	struct ratelimit_state	bt_ioerror_rl;
104 } xfs_buftarg_t;
105 
106 #define XB_PAGES	2
107 
108 struct xfs_buf_map {
109 	xfs_daddr_t		bm_bn;	/* block number for I/O */
110 	int			bm_len;	/* size of I/O */
111 };
112 
113 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
114 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
115 
116 struct xfs_buf_ops {
117 	char *name;
118 	union {
119 		__be32 magic[2];	/* v4 and v5 on disk magic values */
120 		__be16 magic16[2];	/* v4 and v5 on disk magic values */
121 	};
122 	void (*verify_read)(struct xfs_buf *);
123 	void (*verify_write)(struct xfs_buf *);
124 	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
125 };
126 
127 struct xfs_buf {
128 	/*
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.
134 	 */
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 */
142 
143 	/*
144 	 * concurrent access to b_lru and b_lru_flags are protected by
145 	 * bt_lru_lock and not by b_sema
146 	 */
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 	xfs_buftarg_t		*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 */
166 	int			b_map_count;
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 in first page */
171 	int			b_error;	/* error code on I/O */
172 
173 	/*
174 	 * async write failure retry count. Initialised to zero on the first
175 	 * failure, then when it exceeds the maximum configured without a
176 	 * success the write is considered to be failed permanently and the
177 	 * iodone handler will take appropriate action.
178 	 *
179 	 * For retry timeouts, we record the jiffie of the first failure. This
180 	 * means that we can change the retry timeout for buffers already under
181 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
182 	 *
183 	 * last_error is used to ensure that we are getting repeated errors, not
184 	 * different errors. e.g. a block device might change ENOSPC to EIO when
185 	 * a failure timeout occurs, so we want to re-initialise the error
186 	 * retry behaviour appropriately when that happens.
187 	 */
188 	int			b_retries;
189 	unsigned long		b_first_retry_time; /* in jiffies */
190 	int			b_last_error;
191 
192 	const struct xfs_buf_ops	*b_ops;
193 };
194 
195 /* Finding and Reading Buffers */
196 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
197 			   xfs_daddr_t blkno, size_t numblks,
198 			   xfs_buf_flags_t flags);
199 
200 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
201 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
202 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
203 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
204 		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
205 void xfs_buf_readahead_map(struct xfs_buftarg *target,
206 			       struct xfs_buf_map *map, int nmaps,
207 			       const struct xfs_buf_ops *ops);
208 
209 static inline int
210 xfs_buf_get(
211 	struct xfs_buftarg	*target,
212 	xfs_daddr_t		blkno,
213 	size_t			numblks,
214 	struct xfs_buf		**bpp)
215 {
216 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
217 
218 	return xfs_buf_get_map(target, &map, 1, 0, bpp);
219 }
220 
221 static inline int
222 xfs_buf_read(
223 	struct xfs_buftarg	*target,
224 	xfs_daddr_t		blkno,
225 	size_t			numblks,
226 	xfs_buf_flags_t		flags,
227 	struct xfs_buf		**bpp,
228 	const struct xfs_buf_ops *ops)
229 {
230 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
231 
232 	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
233 			__builtin_return_address(0));
234 }
235 
236 static inline void
237 xfs_buf_readahead(
238 	struct xfs_buftarg	*target,
239 	xfs_daddr_t		blkno,
240 	size_t			numblks,
241 	const struct xfs_buf_ops *ops)
242 {
243 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
244 	return xfs_buf_readahead_map(target, &map, 1, ops);
245 }
246 
247 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks, int flags,
248 		struct xfs_buf **bpp);
249 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
250 			  size_t numblks, int flags, struct xfs_buf **bpp,
251 			  const struct xfs_buf_ops *ops);
252 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
253 void xfs_buf_hold(struct xfs_buf *bp);
254 
255 /* Releasing Buffers */
256 extern void xfs_buf_rele(struct xfs_buf *);
257 
258 /* Locking and Unlocking Buffers */
259 extern int xfs_buf_trylock(struct xfs_buf *);
260 extern void xfs_buf_lock(struct xfs_buf *);
261 extern void xfs_buf_unlock(struct xfs_buf *);
262 #define xfs_buf_islocked(bp) \
263 	((bp)->b_sema.count <= 0)
264 
265 static inline void xfs_buf_relse(struct xfs_buf *bp)
266 {
267 	xfs_buf_unlock(bp);
268 	xfs_buf_rele(bp);
269 }
270 
271 /* Buffer Read and Write Routines */
272 extern int xfs_bwrite(struct xfs_buf *bp);
273 
274 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
275 		xfs_failaddr_t failaddr);
276 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
277 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
278 void xfs_buf_ioend_fail(struct xfs_buf *);
279 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
280 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
281 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
282 
283 /* Buffer Utility Routines */
284 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
285 extern void xfs_buf_stale(struct xfs_buf *bp);
286 
287 /* Delayed Write Buffer Routines */
288 extern void xfs_buf_delwri_cancel(struct list_head *);
289 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
290 extern int xfs_buf_delwri_submit(struct list_head *);
291 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
292 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
293 
294 /* Buffer Daemon Setup Routines */
295 extern int xfs_buf_init(void);
296 extern void xfs_buf_terminate(void);
297 
298 /*
299  * These macros use the IO block map rather than b_bn. b_bn is now really
300  * just for the buffer cache index for cached buffers. As IO does not use b_bn
301  * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
302  * map directly. Uncached buffers are not allowed to be discontiguous, so this
303  * is safe to do.
304  *
305  * In future, uncached buffers will pass the block number directly to the io
306  * request function and hence these macros will go away at that point.
307  */
308 #define XFS_BUF_ADDR(bp)		((bp)->b_maps[0].bm_bn)
309 #define XFS_BUF_SET_ADDR(bp, bno)	((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
310 
311 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
312 
313 /*
314  * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
315  * up with a reference count of 0 so it will be tossed from the cache when
316  * released.
317  */
318 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
319 {
320 	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
321 		return;
322 	atomic_set(&bp->b_lru_ref, 0);
323 }
324 
325 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
326 {
327 	return atomic_read(&bp->b_pin_count);
328 }
329 
330 static inline int
331 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
332 {
333 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
334 				cksum_offset);
335 }
336 
337 static inline void
338 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
339 {
340 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
341 			 cksum_offset);
342 }
343 
344 /*
345  *	Handling of buftargs.
346  */
347 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
348 			struct block_device *, struct dax_device *);
349 extern void xfs_free_buftarg(struct xfs_buftarg *);
350 extern void xfs_wait_buftarg(xfs_buftarg_t *);
351 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int);
352 
353 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
354 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
355 
356 static inline int
357 xfs_buftarg_dma_alignment(struct xfs_buftarg *bt)
358 {
359 	return queue_dma_alignment(bt->bt_bdev->bd_disk->queue);
360 }
361 
362 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
363 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
364 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
365 
366 #endif	/* __XFS_BUF_H__ */
367