xref: /linux/fs/xfs/xfs_buf.h (revision 7effbd18e496a954540fb6319db0d87d28a70496)
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 extern struct kmem_cache *xfs_buf_cache;
19 
20 /*
21  *	Base types
22  */
23 struct xfs_buf;
24 
25 #define XFS_BUF_DADDR_NULL	((xfs_daddr_t) (-1LL))
26 
27 #define XBF_READ	 (1u << 0) /* buffer intended for reading from device */
28 #define XBF_WRITE	 (1u << 1) /* buffer intended for writing to device */
29 #define XBF_READ_AHEAD	 (1u << 2) /* asynchronous read-ahead */
30 #define XBF_NO_IOACCT	 (1u << 3) /* bypass I/O accounting (non-LRU bufs) */
31 #define XBF_ASYNC	 (1u << 4) /* initiator will not wait for completion */
32 #define XBF_DONE	 (1u << 5) /* all pages in the buffer uptodate */
33 #define XBF_STALE	 (1u << 6) /* buffer has been staled, do not find it */
34 #define XBF_WRITE_FAIL	 (1u << 7) /* async writes have failed on this buffer */
35 
36 /* buffer type flags for write callbacks */
37 #define _XBF_INODES	 (1u << 16)/* inode buffer */
38 #define _XBF_DQUOTS	 (1u << 17)/* dquot buffer */
39 #define _XBF_LOGRECOVERY (1u << 18)/* log recovery buffer */
40 
41 /* flags used only internally */
42 #define _XBF_PAGES	 (1u << 20)/* backed by refcounted pages */
43 #define _XBF_KMEM	 (1u << 21)/* backed by heap memory */
44 #define _XBF_DELWRI_Q	 (1u << 22)/* buffer on a delwri queue */
45 
46 /* flags used only as arguments to access routines */
47 #define XBF_INCORE	 (1u << 29)/* lookup only, return if found in cache */
48 #define XBF_TRYLOCK	 (1u << 30)/* lock requested, but do not wait */
49 #define XBF_UNMAPPED	 (1u << 31)/* do not map the buffer */
50 
51 
52 typedef unsigned int xfs_buf_flags_t;
53 
54 #define XFS_BUF_FLAGS \
55 	{ XBF_READ,		"READ" }, \
56 	{ XBF_WRITE,		"WRITE" }, \
57 	{ XBF_READ_AHEAD,	"READ_AHEAD" }, \
58 	{ XBF_NO_IOACCT,	"NO_IOACCT" }, \
59 	{ XBF_ASYNC,		"ASYNC" }, \
60 	{ XBF_DONE,		"DONE" }, \
61 	{ XBF_STALE,		"STALE" }, \
62 	{ XBF_WRITE_FAIL,	"WRITE_FAIL" }, \
63 	{ _XBF_INODES,		"INODES" }, \
64 	{ _XBF_DQUOTS,		"DQUOTS" }, \
65 	{ _XBF_LOGRECOVERY,	"LOG_RECOVERY" }, \
66 	{ _XBF_PAGES,		"PAGES" }, \
67 	{ _XBF_KMEM,		"KMEM" }, \
68 	{ _XBF_DELWRI_Q,	"DELWRI_Q" }, \
69 	/* The following interface flags should never be set */ \
70 	{ XBF_INCORE,		"INCORE" }, \
71 	{ XBF_TRYLOCK,		"TRYLOCK" }, \
72 	{ XBF_UNMAPPED,		"UNMAPPED" }
73 
74 /*
75  * Internal state flags.
76  */
77 #define XFS_BSTATE_DISPOSE	 (1 << 0)	/* buffer being discarded */
78 #define XFS_BSTATE_IN_FLIGHT	 (1 << 1)	/* I/O in flight */
79 
80 /*
81  * The xfs_buftarg contains 2 notions of "sector size" -
82  *
83  * 1) The metadata sector size, which is the minimum unit and
84  *    alignment of IO which will be performed by metadata operations.
85  * 2) The device logical sector size
86  *
87  * The first is specified at mkfs time, and is stored on-disk in the
88  * superblock's sb_sectsize.
89  *
90  * The latter is derived from the underlying device, and controls direct IO
91  * alignment constraints.
92  */
93 typedef struct xfs_buftarg {
94 	dev_t			bt_dev;
95 	struct block_device	*bt_bdev;
96 	struct dax_device	*bt_daxdev;
97 	u64			bt_dax_part_off;
98 	struct xfs_mount	*bt_mount;
99 	unsigned int		bt_meta_sectorsize;
100 	size_t			bt_meta_sectormask;
101 	size_t			bt_logical_sectorsize;
102 	size_t			bt_logical_sectormask;
103 
104 	/* LRU control structures */
105 	struct shrinker		bt_shrinker;
106 	struct list_lru		bt_lru;
107 
108 	struct percpu_counter	bt_io_count;
109 	struct ratelimit_state	bt_ioerror_rl;
110 } xfs_buftarg_t;
111 
112 #define XB_PAGES	2
113 
114 struct xfs_buf_map {
115 	xfs_daddr_t		bm_bn;	/* block number for I/O */
116 	int			bm_len;	/* size of I/O */
117 };
118 
119 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
120 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
121 
122 struct xfs_buf_ops {
123 	char *name;
124 	union {
125 		__be32 magic[2];	/* v4 and v5 on disk magic values */
126 		__be16 magic16[2];	/* v4 and v5 on disk magic values */
127 	};
128 	void (*verify_read)(struct xfs_buf *);
129 	void (*verify_write)(struct xfs_buf *);
130 	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
131 };
132 
133 struct xfs_buf {
134 	/*
135 	 * first cacheline holds all the fields needed for an uncontended cache
136 	 * hit to be fully processed. The semaphore straddles the cacheline
137 	 * boundary, but the counter and lock sits on the first cacheline,
138 	 * which is the only bit that is touched if we hit the semaphore
139 	 * fast-path on locking.
140 	 */
141 	struct rhash_head	b_rhash_head;	/* pag buffer hash node */
142 
143 	xfs_daddr_t		b_rhash_key;	/* buffer cache index */
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 */
149 
150 	/*
151 	 * concurrent access to b_lru and b_lru_flags are protected by
152 	 * bt_lru_lock and not by b_sema
153 	 */
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 	struct xfs_mount	*b_mount;
162 	struct xfs_buftarg	*b_target;	/* buffer target (device) */
163 	void			*b_addr;	/* virtual address of buffer */
164 	struct work_struct	b_ioend_work;
165 	struct completion	b_iowait;	/* queue for I/O waiters */
166 	struct xfs_buf_log_item	*b_log_item;
167 	struct list_head	b_li_list;	/* Log items list head */
168 	struct xfs_trans	*b_transp;
169 	struct page		**b_pages;	/* array of page pointers */
170 	struct page		*b_page_array[XB_PAGES]; /* inline pages */
171 	struct xfs_buf_map	*b_maps;	/* compound buffer map */
172 	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
173 	int			b_map_count;
174 	atomic_t		b_pin_count;	/* pin count */
175 	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
176 	unsigned int		b_page_count;	/* size of page array */
177 	unsigned int		b_offset;	/* page offset of b_addr,
178 						   only for _XBF_KMEM buffers */
179 	int			b_error;	/* error code on I/O */
180 
181 	/*
182 	 * async write failure retry count. Initialised to zero on the first
183 	 * failure, then when it exceeds the maximum configured without a
184 	 * success the write is considered to be failed permanently and the
185 	 * iodone handler will take appropriate action.
186 	 *
187 	 * For retry timeouts, we record the jiffie of the first failure. This
188 	 * means that we can change the retry timeout for buffers already under
189 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
190 	 *
191 	 * last_error is used to ensure that we are getting repeated errors, not
192 	 * different errors. e.g. a block device might change ENOSPC to EIO when
193 	 * a failure timeout occurs, so we want to re-initialise the error
194 	 * retry behaviour appropriately when that happens.
195 	 */
196 	int			b_retries;
197 	unsigned long		b_first_retry_time; /* in jiffies */
198 	int			b_last_error;
199 
200 	const struct xfs_buf_ops	*b_ops;
201 	struct rcu_head		b_rcu;
202 };
203 
204 /* Finding and Reading Buffers */
205 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
206 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
207 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
208 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
209 		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
210 void xfs_buf_readahead_map(struct xfs_buftarg *target,
211 			       struct xfs_buf_map *map, int nmaps,
212 			       const struct xfs_buf_ops *ops);
213 
214 static inline int
215 xfs_buf_incore(
216 	struct xfs_buftarg	*target,
217 	xfs_daddr_t		blkno,
218 	size_t			numblks,
219 	xfs_buf_flags_t		flags,
220 	struct xfs_buf		**bpp)
221 {
222 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
223 
224 	return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
225 }
226 
227 static inline int
228 xfs_buf_get(
229 	struct xfs_buftarg	*target,
230 	xfs_daddr_t		blkno,
231 	size_t			numblks,
232 	struct xfs_buf		**bpp)
233 {
234 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
235 
236 	return xfs_buf_get_map(target, &map, 1, 0, bpp);
237 }
238 
239 static inline int
240 xfs_buf_read(
241 	struct xfs_buftarg	*target,
242 	xfs_daddr_t		blkno,
243 	size_t			numblks,
244 	xfs_buf_flags_t		flags,
245 	struct xfs_buf		**bpp,
246 	const struct xfs_buf_ops *ops)
247 {
248 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
249 
250 	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
251 			__builtin_return_address(0));
252 }
253 
254 static inline void
255 xfs_buf_readahead(
256 	struct xfs_buftarg	*target,
257 	xfs_daddr_t		blkno,
258 	size_t			numblks,
259 	const struct xfs_buf_ops *ops)
260 {
261 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
262 	return xfs_buf_readahead_map(target, &map, 1, ops);
263 }
264 
265 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
266 		xfs_buf_flags_t flags, struct xfs_buf **bpp);
267 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
268 		size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
269 		const struct xfs_buf_ops *ops);
270 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
271 void xfs_buf_hold(struct xfs_buf *bp);
272 
273 /* Releasing Buffers */
274 extern void xfs_buf_rele(struct xfs_buf *);
275 
276 /* Locking and Unlocking Buffers */
277 extern int xfs_buf_trylock(struct xfs_buf *);
278 extern void xfs_buf_lock(struct xfs_buf *);
279 extern void xfs_buf_unlock(struct xfs_buf *);
280 #define xfs_buf_islocked(bp) \
281 	((bp)->b_sema.count <= 0)
282 
283 static inline void xfs_buf_relse(struct xfs_buf *bp)
284 {
285 	xfs_buf_unlock(bp);
286 	xfs_buf_rele(bp);
287 }
288 
289 /* Buffer Read and Write Routines */
290 extern int xfs_bwrite(struct xfs_buf *bp);
291 
292 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
293 		xfs_failaddr_t failaddr);
294 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
295 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
296 void xfs_buf_ioend_fail(struct xfs_buf *);
297 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
298 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
299 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
300 
301 /* Buffer Utility Routines */
302 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
303 extern void xfs_buf_stale(struct xfs_buf *bp);
304 
305 /* Delayed Write Buffer Routines */
306 extern void xfs_buf_delwri_cancel(struct list_head *);
307 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
308 extern int xfs_buf_delwri_submit(struct list_head *);
309 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
310 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
311 
312 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
313 {
314 	return bp->b_maps[0].bm_bn;
315 }
316 
317 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
318 
319 /*
320  * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
321  * up with a reference count of 0 so it will be tossed from the cache when
322  * released.
323  */
324 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
325 {
326 	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
327 		return;
328 	atomic_set(&bp->b_lru_ref, 0);
329 }
330 
331 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
332 {
333 	return atomic_read(&bp->b_pin_count);
334 }
335 
336 static inline int
337 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
338 {
339 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
340 				cksum_offset);
341 }
342 
343 static inline void
344 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
345 {
346 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
347 			 cksum_offset);
348 }
349 
350 /*
351  *	Handling of buftargs.
352  */
353 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
354 		struct block_device *bdev);
355 extern void xfs_free_buftarg(struct xfs_buftarg *);
356 extern void xfs_buftarg_wait(struct xfs_buftarg *);
357 extern void xfs_buftarg_drain(struct xfs_buftarg *);
358 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
359 
360 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
361 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
362 
363 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
364 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
365 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
366 
367 #endif	/* __XFS_BUF_H__ */
368