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	 (1u << 0) /* buffer intended for reading from device */
26 #define XBF_WRITE	 (1u << 1) /* buffer intended for writing to device */
27 #define XBF_READ_AHEAD	 (1u << 2) /* asynchronous read-ahead */
28 #define XBF_NO_IOACCT	 (1u << 3) /* bypass I/O accounting (non-LRU bufs) */
29 #define XBF_ASYNC	 (1u << 4) /* initiator will not wait for completion */
30 #define XBF_DONE	 (1u << 5) /* all pages in the buffer uptodate */
31 #define XBF_STALE	 (1u << 6) /* buffer has been staled, do not find it */
32 #define XBF_WRITE_FAIL	 (1u << 7) /* async writes have failed on this buffer */
33 
34 /* buffer type flags for write callbacks */
35 #define _XBF_INODES	 (1u << 16)/* inode buffer */
36 #define _XBF_DQUOTS	 (1u << 17)/* dquot buffer */
37 #define _XBF_LOGRECOVERY (1u << 18)/* log recovery buffer */
38 
39 /* flags used only internally */
40 #define _XBF_PAGES	 (1u << 20)/* backed by refcounted pages */
41 #define _XBF_KMEM	 (1u << 21)/* backed by heap memory */
42 #define _XBF_DELWRI_Q	 (1u << 22)/* buffer on a delwri queue */
43 
44 /* flags used only as arguments to access routines */
45 #define XBF_TRYLOCK	 (1u << 30)/* lock requested, but do not wait */
46 #define XBF_UNMAPPED	 (1u << 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 	u64			bt_dax_part_off;
93 	struct xfs_mount	*bt_mount;
94 	unsigned int		bt_meta_sectorsize;
95 	size_t			bt_meta_sectormask;
96 	size_t			bt_logical_sectorsize;
97 	size_t			bt_logical_sectormask;
98 
99 	/* LRU control structures */
100 	struct shrinker		bt_shrinker;
101 	struct list_lru		bt_lru;
102 
103 	struct percpu_counter	bt_io_count;
104 	struct ratelimit_state	bt_ioerror_rl;
105 } xfs_buftarg_t;
106 
107 #define XB_PAGES	2
108 
109 struct xfs_buf_map {
110 	xfs_daddr_t		bm_bn;	/* block number for I/O */
111 	int			bm_len;	/* size of I/O */
112 };
113 
114 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
115 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
116 
117 struct xfs_buf_ops {
118 	char *name;
119 	union {
120 		__be32 magic[2];	/* v4 and v5 on disk magic values */
121 		__be16 magic16[2];	/* v4 and v5 on disk magic values */
122 	};
123 	void (*verify_read)(struct xfs_buf *);
124 	void (*verify_write)(struct xfs_buf *);
125 	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
126 };
127 
128 struct xfs_buf {
129 	/*
130 	 * first cacheline holds all the fields needed for an uncontended cache
131 	 * hit to be fully processed. The semaphore straddles the cacheline
132 	 * boundary, but the counter and lock sits on the first cacheline,
133 	 * which is the only bit that is touched if we hit the semaphore
134 	 * fast-path on locking.
135 	 */
136 	struct rhash_head	b_rhash_head;	/* pag buffer hash node */
137 
138 	xfs_daddr_t		b_rhash_key;	/* buffer cache index */
139 	int			b_length;	/* size of buffer in BBs */
140 	atomic_t		b_hold;		/* reference count */
141 	atomic_t		b_lru_ref;	/* lru reclaim ref count */
142 	xfs_buf_flags_t		b_flags;	/* status flags */
143 	struct semaphore	b_sema;		/* semaphore for lockables */
144 
145 	/*
146 	 * concurrent access to b_lru and b_lru_flags are protected by
147 	 * bt_lru_lock and not by b_sema
148 	 */
149 	struct list_head	b_lru;		/* lru list */
150 	spinlock_t		b_lock;		/* internal state lock */
151 	unsigned int		b_state;	/* internal state flags */
152 	int			b_io_error;	/* internal IO error state */
153 	wait_queue_head_t	b_waiters;	/* unpin waiters */
154 	struct list_head	b_list;
155 	struct xfs_perag	*b_pag;		/* contains rbtree root */
156 	struct xfs_mount	*b_mount;
157 	struct xfs_buftarg	*b_target;	/* buffer target (device) */
158 	void			*b_addr;	/* virtual address of buffer */
159 	struct work_struct	b_ioend_work;
160 	struct completion	b_iowait;	/* queue for I/O waiters */
161 	struct xfs_buf_log_item	*b_log_item;
162 	struct list_head	b_li_list;	/* Log items list head */
163 	struct xfs_trans	*b_transp;
164 	struct page		**b_pages;	/* array of page pointers */
165 	struct page		*b_page_array[XB_PAGES]; /* inline pages */
166 	struct xfs_buf_map	*b_maps;	/* compound buffer map */
167 	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
168 	int			b_map_count;
169 	atomic_t		b_pin_count;	/* pin count */
170 	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
171 	unsigned int		b_page_count;	/* size of page array */
172 	unsigned int		b_offset;	/* page offset of b_addr,
173 						   only for _XBF_KMEM buffers */
174 	int			b_error;	/* error code on I/O */
175 
176 	/*
177 	 * async write failure retry count. Initialised to zero on the first
178 	 * failure, then when it exceeds the maximum configured without a
179 	 * success the write is considered to be failed permanently and the
180 	 * iodone handler will take appropriate action.
181 	 *
182 	 * For retry timeouts, we record the jiffie of the first failure. This
183 	 * means that we can change the retry timeout for buffers already under
184 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
185 	 *
186 	 * last_error is used to ensure that we are getting repeated errors, not
187 	 * different errors. e.g. a block device might change ENOSPC to EIO when
188 	 * a failure timeout occurs, so we want to re-initialise the error
189 	 * retry behaviour appropriately when that happens.
190 	 */
191 	int			b_retries;
192 	unsigned long		b_first_retry_time; /* in jiffies */
193 	int			b_last_error;
194 
195 	const struct xfs_buf_ops	*b_ops;
196 };
197 
198 /* Finding and Reading Buffers */
199 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
200 			   xfs_daddr_t blkno, size_t numblks,
201 			   xfs_buf_flags_t flags);
202 
203 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
204 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
205 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
206 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
207 		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
208 void xfs_buf_readahead_map(struct xfs_buftarg *target,
209 			       struct xfs_buf_map *map, int nmaps,
210 			       const struct xfs_buf_ops *ops);
211 
212 static inline int
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp)213 xfs_buf_get(
214 	struct xfs_buftarg	*target,
215 	xfs_daddr_t		blkno,
216 	size_t			numblks,
217 	struct xfs_buf		**bpp)
218 {
219 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
220 
221 	return xfs_buf_get_map(target, &map, 1, 0, bpp);
222 }
223 
224 static inline int
xfs_buf_read(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,xfs_buf_flags_t flags,struct xfs_buf ** bpp,const struct xfs_buf_ops * ops)225 xfs_buf_read(
226 	struct xfs_buftarg	*target,
227 	xfs_daddr_t		blkno,
228 	size_t			numblks,
229 	xfs_buf_flags_t		flags,
230 	struct xfs_buf		**bpp,
231 	const struct xfs_buf_ops *ops)
232 {
233 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
234 
235 	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
236 			__builtin_return_address(0));
237 }
238 
239 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)240 xfs_buf_readahead(
241 	struct xfs_buftarg	*target,
242 	xfs_daddr_t		blkno,
243 	size_t			numblks,
244 	const struct xfs_buf_ops *ops)
245 {
246 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
247 	return xfs_buf_readahead_map(target, &map, 1, ops);
248 }
249 
250 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
251 		xfs_buf_flags_t flags, struct xfs_buf **bpp);
252 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
253 		size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
254 		const struct xfs_buf_ops *ops);
255 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
256 void xfs_buf_hold(struct xfs_buf *bp);
257 
258 /* Releasing Buffers */
259 extern void xfs_buf_rele(struct xfs_buf *);
260 
261 /* Locking and Unlocking Buffers */
262 extern int xfs_buf_trylock(struct xfs_buf *);
263 extern void xfs_buf_lock(struct xfs_buf *);
264 extern void xfs_buf_unlock(struct xfs_buf *);
265 #define xfs_buf_islocked(bp) \
266 	((bp)->b_sema.count <= 0)
267 
xfs_buf_relse(struct xfs_buf * bp)268 static inline void xfs_buf_relse(struct xfs_buf *bp)
269 {
270 	xfs_buf_unlock(bp);
271 	xfs_buf_rele(bp);
272 }
273 
274 /* Buffer Read and Write Routines */
275 extern int xfs_bwrite(struct xfs_buf *bp);
276 
277 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
278 		xfs_failaddr_t failaddr);
279 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
280 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
281 void xfs_buf_ioend_fail(struct xfs_buf *);
282 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
283 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
284 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
285 
286 /* Buffer Utility Routines */
287 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
288 extern void xfs_buf_stale(struct xfs_buf *bp);
289 
290 /* Delayed Write Buffer Routines */
291 extern void xfs_buf_delwri_cancel(struct list_head *);
292 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
293 extern int xfs_buf_delwri_submit(struct list_head *);
294 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
295 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
296 
297 /* Buffer Daemon Setup Routines */
298 extern int xfs_buf_init(void);
299 extern void xfs_buf_terminate(void);
300 
xfs_buf_daddr(struct xfs_buf * bp)301 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
302 {
303 	return bp->b_maps[0].bm_bn;
304 }
305 
306 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
307 
308 /*
309  * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
310  * up with a reference count of 0 so it will be tossed from the cache when
311  * released.
312  */
xfs_buf_oneshot(struct xfs_buf * bp)313 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
314 {
315 	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
316 		return;
317 	atomic_set(&bp->b_lru_ref, 0);
318 }
319 
xfs_buf_ispinned(struct xfs_buf * bp)320 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
321 {
322 	return atomic_read(&bp->b_pin_count);
323 }
324 
325 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)326 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
327 {
328 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
329 				cksum_offset);
330 }
331 
332 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)333 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
334 {
335 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
336 			 cksum_offset);
337 }
338 
339 /*
340  *	Handling of buftargs.
341  */
342 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
343 		struct block_device *bdev);
344 extern void xfs_free_buftarg(struct xfs_buftarg *);
345 extern void xfs_buftarg_wait(struct xfs_buftarg *);
346 extern void xfs_buftarg_drain(struct xfs_buftarg *);
347 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
348 
349 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
350 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
351 
352 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
353 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
354 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
355 
356 #endif	/* __XFS_BUF_H__ */
357