xref: /linux/fs/xfs/xfs_buf.h (revision f7f0adfe64de08803990dc4cbecd2849c04e314a)
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_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 /*
46  * Online fsck is scanning the buffer cache for live buffers.  Do not warn
47  * about length mismatches during lookups and do not return stale buffers.
48  */
49 #define XBF_LIVESCAN	 (1u << 28)
50 #define XBF_INCORE	 (1u << 29)/* lookup only, return if found in cache */
51 #define XBF_TRYLOCK	 (1u << 30)/* lock requested, but do not wait */
52 #define XBF_UNMAPPED	 (1u << 31)/* do not map the buffer */
53 
54 
55 typedef unsigned int xfs_buf_flags_t;
56 
57 #define XFS_BUF_FLAGS \
58 	{ XBF_READ,		"READ" }, \
59 	{ XBF_WRITE,		"WRITE" }, \
60 	{ XBF_READ_AHEAD,	"READ_AHEAD" }, \
61 	{ XBF_NO_IOACCT,	"NO_IOACCT" }, \
62 	{ XBF_ASYNC,		"ASYNC" }, \
63 	{ XBF_DONE,		"DONE" }, \
64 	{ XBF_STALE,		"STALE" }, \
65 	{ XBF_WRITE_FAIL,	"WRITE_FAIL" }, \
66 	{ _XBF_LOGRECOVERY,	"LOG_RECOVERY" }, \
67 	{ _XBF_PAGES,		"PAGES" }, \
68 	{ _XBF_KMEM,		"KMEM" }, \
69 	{ _XBF_DELWRI_Q,	"DELWRI_Q" }, \
70 	/* The following interface flags should never be set */ \
71 	{ XBF_LIVESCAN,		"LIVESCAN" }, \
72 	{ XBF_INCORE,		"INCORE" }, \
73 	{ XBF_TRYLOCK,		"TRYLOCK" }, \
74 	{ XBF_UNMAPPED,		"UNMAPPED" }
75 
76 /*
77  * Internal state flags.
78  */
79 #define XFS_BSTATE_DISPOSE	 (1 << 0)	/* buffer being discarded */
80 #define XFS_BSTATE_IN_FLIGHT	 (1 << 1)	/* I/O in flight */
81 
82 struct xfs_buf_cache {
83 	struct rhashtable	bc_hash;
84 };
85 
86 int xfs_buf_cache_init(struct xfs_buf_cache *bch);
87 void xfs_buf_cache_destroy(struct xfs_buf_cache *bch);
88 
89 /*
90  * The xfs_buftarg contains 2 notions of "sector size" -
91  *
92  * 1) The metadata sector size, which is the minimum unit and
93  *    alignment of IO which will be performed by metadata operations.
94  * 2) The device logical sector size
95  *
96  * The first is specified at mkfs time, and is stored on-disk in the
97  * superblock's sb_sectsize.
98  *
99  * The latter is derived from the underlying device, and controls direct IO
100  * alignment constraints.
101  */
102 struct xfs_buftarg {
103 	dev_t			bt_dev;
104 	struct file		*bt_bdev_file;
105 	struct block_device	*bt_bdev;
106 	struct dax_device	*bt_daxdev;
107 	struct file		*bt_file;
108 	u64			bt_dax_part_off;
109 	struct xfs_mount	*bt_mount;
110 	unsigned int		bt_meta_sectorsize;
111 	size_t			bt_meta_sectormask;
112 	size_t			bt_logical_sectorsize;
113 	size_t			bt_logical_sectormask;
114 
115 	/* LRU control structures */
116 	struct shrinker		*bt_shrinker;
117 	struct list_lru		bt_lru;
118 
119 	struct percpu_counter	bt_io_count;
120 	struct ratelimit_state	bt_ioerror_rl;
121 
122 	/* Atomic write unit values */
123 	unsigned int		bt_bdev_awu_min;
124 	unsigned int		bt_bdev_awu_max;
125 
126 	/* built-in cache, if we're not using the perag one */
127 	struct xfs_buf_cache	bt_cache[];
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 	unsigned int		bm_flags;
136 };
137 
138 /*
139  * Online fsck is scanning the buffer cache for live buffers.  Do not warn
140  * about length mismatches during lookups and do not return stale buffers.
141  */
142 #define XBM_LIVESCAN		(1U << 0)
143 
144 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
145 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
146 
147 struct xfs_buf_ops {
148 	char *name;
149 	union {
150 		__be32 magic[2];	/* v4 and v5 on disk magic values */
151 		__be16 magic16[2];	/* v4 and v5 on disk magic values */
152 	};
153 	void (*verify_read)(struct xfs_buf *);
154 	void (*verify_write)(struct xfs_buf *);
155 	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
156 };
157 
158 struct xfs_buf {
159 	/*
160 	 * first cacheline holds all the fields needed for an uncontended cache
161 	 * hit to be fully processed. The semaphore straddles the cacheline
162 	 * boundary, but the counter and lock sits on the first cacheline,
163 	 * which is the only bit that is touched if we hit the semaphore
164 	 * fast-path on locking.
165 	 */
166 	struct rhash_head	b_rhash_head;	/* pag buffer hash node */
167 
168 	xfs_daddr_t		b_rhash_key;	/* buffer cache index */
169 	int			b_length;	/* size of buffer in BBs */
170 	unsigned int		b_hold;		/* reference count */
171 	atomic_t		b_lru_ref;	/* lru reclaim ref count */
172 	xfs_buf_flags_t		b_flags;	/* status flags */
173 	struct semaphore	b_sema;		/* semaphore for lockables */
174 
175 	/*
176 	 * concurrent access to b_lru and b_lru_flags are protected by
177 	 * bt_lru_lock and not by b_sema
178 	 */
179 	struct list_head	b_lru;		/* lru list */
180 	spinlock_t		b_lock;		/* internal state lock */
181 	unsigned int		b_state;	/* internal state flags */
182 	wait_queue_head_t	b_waiters;	/* unpin waiters */
183 	struct list_head	b_list;
184 	struct xfs_perag	*b_pag;
185 	struct xfs_mount	*b_mount;
186 	struct xfs_buftarg	*b_target;	/* buffer target (device) */
187 	void			*b_addr;	/* virtual address of buffer */
188 	struct work_struct	b_ioend_work;
189 	struct completion	b_iowait;	/* queue for I/O waiters */
190 	struct xfs_buf_log_item	*b_log_item;
191 	struct list_head	b_li_list;	/* Log items list head */
192 	struct xfs_trans	*b_transp;
193 	struct page		**b_pages;	/* array of page pointers */
194 	struct page		*b_page_array[XB_PAGES]; /* inline pages */
195 	struct xfs_buf_map	*b_maps;	/* compound buffer map */
196 	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
197 	int			b_map_count;
198 	atomic_t		b_pin_count;	/* pin count */
199 	unsigned int		b_page_count;	/* size of page array */
200 	unsigned int		b_offset;	/* page offset of b_addr,
201 						   only for _XBF_KMEM buffers */
202 	int			b_error;	/* error code on I/O */
203 	void			(*b_iodone)(struct xfs_buf *bp);
204 
205 	/*
206 	 * async write failure retry count. Initialised to zero on the first
207 	 * failure, then when it exceeds the maximum configured without a
208 	 * success the write is considered to be failed permanently and the
209 	 * iodone handler will take appropriate action.
210 	 *
211 	 * For retry timeouts, we record the jiffy of the first failure. This
212 	 * means that we can change the retry timeout for buffers already under
213 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
214 	 *
215 	 * last_error is used to ensure that we are getting repeated errors, not
216 	 * different errors. e.g. a block device might change ENOSPC to EIO when
217 	 * a failure timeout occurs, so we want to re-initialise the error
218 	 * retry behaviour appropriately when that happens.
219 	 */
220 	int			b_retries;
221 	unsigned long		b_first_retry_time; /* in jiffies */
222 	int			b_last_error;
223 
224 	const struct xfs_buf_ops	*b_ops;
225 	struct rcu_head		b_rcu;
226 };
227 
228 /* Finding and Reading Buffers */
229 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
230 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
231 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
232 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
233 		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
234 void xfs_buf_readahead_map(struct xfs_buftarg *target,
235 			       struct xfs_buf_map *map, int nmaps,
236 			       const struct xfs_buf_ops *ops);
237 
238 static inline int
239 xfs_buf_incore(
240 	struct xfs_buftarg	*target,
241 	xfs_daddr_t		blkno,
242 	size_t			numblks,
243 	xfs_buf_flags_t		flags,
244 	struct xfs_buf		**bpp)
245 {
246 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
247 
248 	return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
249 }
250 
251 static inline int
252 xfs_buf_get(
253 	struct xfs_buftarg	*target,
254 	xfs_daddr_t		blkno,
255 	size_t			numblks,
256 	struct xfs_buf		**bpp)
257 {
258 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
259 
260 	return xfs_buf_get_map(target, &map, 1, 0, bpp);
261 }
262 
263 static inline int
264 xfs_buf_read(
265 	struct xfs_buftarg	*target,
266 	xfs_daddr_t		blkno,
267 	size_t			numblks,
268 	xfs_buf_flags_t		flags,
269 	struct xfs_buf		**bpp,
270 	const struct xfs_buf_ops *ops)
271 {
272 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
273 
274 	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
275 			__builtin_return_address(0));
276 }
277 
278 static inline void
279 xfs_buf_readahead(
280 	struct xfs_buftarg	*target,
281 	xfs_daddr_t		blkno,
282 	size_t			numblks,
283 	const struct xfs_buf_ops *ops)
284 {
285 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
286 	return xfs_buf_readahead_map(target, &map, 1, ops);
287 }
288 
289 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
290 		xfs_buf_flags_t flags, struct xfs_buf **bpp);
291 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
292 		size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
293 		const struct xfs_buf_ops *ops);
294 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
295 void xfs_buf_hold(struct xfs_buf *bp);
296 
297 /* Releasing Buffers */
298 extern void xfs_buf_rele(struct xfs_buf *);
299 
300 /* Locking and Unlocking Buffers */
301 extern int xfs_buf_trylock(struct xfs_buf *);
302 extern void xfs_buf_lock(struct xfs_buf *);
303 extern void xfs_buf_unlock(struct xfs_buf *);
304 #define xfs_buf_islocked(bp) \
305 	((bp)->b_sema.count <= 0)
306 
307 static inline void xfs_buf_relse(struct xfs_buf *bp)
308 {
309 	xfs_buf_unlock(bp);
310 	xfs_buf_rele(bp);
311 }
312 
313 /* Buffer Read and Write Routines */
314 extern int xfs_bwrite(struct xfs_buf *bp);
315 
316 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
317 		xfs_failaddr_t failaddr);
318 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
319 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
320 void xfs_buf_ioend_fail(struct xfs_buf *);
321 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
322 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
323 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
324 
325 /* Buffer Utility Routines */
326 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
327 extern void xfs_buf_stale(struct xfs_buf *bp);
328 
329 /* Delayed Write Buffer Routines */
330 extern void xfs_buf_delwri_cancel(struct list_head *);
331 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
332 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
333 extern int xfs_buf_delwri_submit(struct list_head *);
334 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
335 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
336 
337 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
338 {
339 	return bp->b_maps[0].bm_bn;
340 }
341 
342 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
343 
344 /*
345  * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
346  * up with a reference count of 0 so it will be tossed from the cache when
347  * released.
348  */
349 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
350 {
351 	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
352 		return;
353 	atomic_set(&bp->b_lru_ref, 0);
354 }
355 
356 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
357 {
358 	return atomic_read(&bp->b_pin_count);
359 }
360 
361 static inline int
362 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
363 {
364 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
365 				cksum_offset);
366 }
367 
368 static inline void
369 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
370 {
371 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
372 			 cksum_offset);
373 }
374 
375 /*
376  *	Handling of buftargs.
377  */
378 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
379 		struct file *bdev_file);
380 extern void xfs_free_buftarg(struct xfs_buftarg *);
381 extern void xfs_buftarg_wait(struct xfs_buftarg *);
382 extern void xfs_buftarg_drain(struct xfs_buftarg *);
383 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
384 
385 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
386 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
387 
388 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
389 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
390 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
391 
392 /* for xfs_buf_mem.c only: */
393 int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize,
394 		const char *descr);
395 void xfs_destroy_buftarg(struct xfs_buftarg *btp);
396 
397 #endif	/* __XFS_BUF_H__ */
398