xfs_buf.h revision 4c7f65ae
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 */
21struct 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
48typedef 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 */
88typedef 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
108struct 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
116struct 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
127struct 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
137	xfs_daddr_t		b_rhash_key;	/* buffer cache index */
138	int			b_length;	/* size of buffer in BBs */
139	atomic_t		b_hold;		/* reference count */
140	atomic_t		b_lru_ref;	/* lru reclaim ref count */
141	xfs_buf_flags_t		b_flags;	/* status flags */
142	struct semaphore	b_sema;		/* semaphore for lockables */
143
144	/*
145	 * concurrent access to b_lru and b_lru_flags are protected by
146	 * bt_lru_lock and not by b_sema
147	 */
148	struct list_head	b_lru;		/* lru list */
149	spinlock_t		b_lock;		/* internal state lock */
150	unsigned int		b_state;	/* internal state flags */
151	int			b_io_error;	/* internal IO error state */
152	wait_queue_head_t	b_waiters;	/* unpin waiters */
153	struct list_head	b_list;
154	struct xfs_perag	*b_pag;		/* contains rbtree root */
155	struct xfs_mount	*b_mount;
156	struct xfs_buftarg	*b_target;	/* buffer target (device) */
157	void			*b_addr;	/* virtual address of buffer */
158	struct work_struct	b_ioend_work;
159	struct completion	b_iowait;	/* queue for I/O waiters */
160	struct xfs_buf_log_item	*b_log_item;
161	struct list_head	b_li_list;	/* Log items list head */
162	struct xfs_trans	*b_transp;
163	struct page		**b_pages;	/* array of page pointers */
164	struct page		*b_page_array[XB_PAGES]; /* inline pages */
165	struct xfs_buf_map	*b_maps;	/* compound buffer map */
166	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
167	int			b_map_count;
168	atomic_t		b_pin_count;	/* pin count */
169	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
170	unsigned int		b_page_count;	/* size of page array */
171	unsigned int		b_offset;	/* page offset of b_addr,
172						   only for _XBF_KMEM buffers */
173	int			b_error;	/* error code on I/O */
174
175	/*
176	 * async write failure retry count. Initialised to zero on the first
177	 * failure, then when it exceeds the maximum configured without a
178	 * success the write is considered to be failed permanently and the
179	 * iodone handler will take appropriate action.
180	 *
181	 * For retry timeouts, we record the jiffie of the first failure. This
182	 * means that we can change the retry timeout for buffers already under
183	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
184	 *
185	 * last_error is used to ensure that we are getting repeated errors, not
186	 * different errors. e.g. a block device might change ENOSPC to EIO when
187	 * a failure timeout occurs, so we want to re-initialise the error
188	 * retry behaviour appropriately when that happens.
189	 */
190	int			b_retries;
191	unsigned long		b_first_retry_time; /* in jiffies */
192	int			b_last_error;
193
194	const struct xfs_buf_ops	*b_ops;
195};
196
197/* Finding and Reading Buffers */
198struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
199			   xfs_daddr_t blkno, size_t numblks,
200			   xfs_buf_flags_t flags);
201
202int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
203		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
204int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
205		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
206		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
207void xfs_buf_readahead_map(struct xfs_buftarg *target,
208			       struct xfs_buf_map *map, int nmaps,
209			       const struct xfs_buf_ops *ops);
210
211static inline int
212xfs_buf_get(
213	struct xfs_buftarg	*target,
214	xfs_daddr_t		blkno,
215	size_t			numblks,
216	struct xfs_buf		**bpp)
217{
218	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
219
220	return xfs_buf_get_map(target, &map, 1, 0, bpp);
221}
222
223static inline int
224xfs_buf_read(
225	struct xfs_buftarg	*target,
226	xfs_daddr_t		blkno,
227	size_t			numblks,
228	xfs_buf_flags_t		flags,
229	struct xfs_buf		**bpp,
230	const struct xfs_buf_ops *ops)
231{
232	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
233
234	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
235			__builtin_return_address(0));
236}
237
238static inline void
239xfs_buf_readahead(
240	struct xfs_buftarg	*target,
241	xfs_daddr_t		blkno,
242	size_t			numblks,
243	const struct xfs_buf_ops *ops)
244{
245	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
246	return xfs_buf_readahead_map(target, &map, 1, ops);
247}
248
249int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks, int flags,
250		struct xfs_buf **bpp);
251int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
252			  size_t numblks, int flags, struct xfs_buf **bpp,
253			  const struct xfs_buf_ops *ops);
254int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
255void xfs_buf_hold(struct xfs_buf *bp);
256
257/* Releasing Buffers */
258extern void xfs_buf_rele(struct xfs_buf *);
259
260/* Locking and Unlocking Buffers */
261extern int xfs_buf_trylock(struct xfs_buf *);
262extern void xfs_buf_lock(struct xfs_buf *);
263extern void xfs_buf_unlock(struct xfs_buf *);
264#define xfs_buf_islocked(bp) \
265	((bp)->b_sema.count <= 0)
266
267static inline void xfs_buf_relse(struct xfs_buf *bp)
268{
269	xfs_buf_unlock(bp);
270	xfs_buf_rele(bp);
271}
272
273/* Buffer Read and Write Routines */
274extern int xfs_bwrite(struct xfs_buf *bp);
275
276extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
277		xfs_failaddr_t failaddr);
278#define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
279extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
280void xfs_buf_ioend_fail(struct xfs_buf *);
281void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
282void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
283#define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
284
285/* Buffer Utility Routines */
286extern void *xfs_buf_offset(struct xfs_buf *, size_t);
287extern void xfs_buf_stale(struct xfs_buf *bp);
288
289/* Delayed Write Buffer Routines */
290extern void xfs_buf_delwri_cancel(struct list_head *);
291extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
292extern int xfs_buf_delwri_submit(struct list_head *);
293extern int xfs_buf_delwri_submit_nowait(struct list_head *);
294extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
295
296/* Buffer Daemon Setup Routines */
297extern int xfs_buf_init(void);
298extern void xfs_buf_terminate(void);
299
300static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
301{
302	return bp->b_maps[0].bm_bn;
303}
304
305void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
306
307/*
308 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
309 * up with a reference count of 0 so it will be tossed from the cache when
310 * released.
311 */
312static inline void xfs_buf_oneshot(struct xfs_buf *bp)
313{
314	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
315		return;
316	atomic_set(&bp->b_lru_ref, 0);
317}
318
319static inline int xfs_buf_ispinned(struct xfs_buf *bp)
320{
321	return atomic_read(&bp->b_pin_count);
322}
323
324static inline int
325xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
326{
327	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
328				cksum_offset);
329}
330
331static inline void
332xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
333{
334	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
335			 cksum_offset);
336}
337
338/*
339 *	Handling of buftargs.
340 */
341extern struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *,
342		struct block_device *, struct dax_device *);
343extern void xfs_free_buftarg(struct xfs_buftarg *);
344extern void xfs_buftarg_wait(struct xfs_buftarg *);
345extern void xfs_buftarg_drain(struct xfs_buftarg *);
346extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
347
348#define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
349#define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
350
351int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
352bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
353bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
354
355#endif	/* __XFS_BUF_H__ */
356