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
18extern struct kmem_cache *xfs_buf_cache;
19
20/*
21 *	Base types
22 */
23struct 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/*
48 * Online fsck is scanning the buffer cache for live buffers.  Do not warn
49 * about length mismatches during lookups and do not return stale buffers.
50 */
51#define XBF_LIVESCAN	 (1u << 28)
52#define XBF_INCORE	 (1u << 29)/* lookup only, return if found in cache */
53#define XBF_TRYLOCK	 (1u << 30)/* lock requested, but do not wait */
54#define XBF_UNMAPPED	 (1u << 31)/* do not map the buffer */
55
56
57typedef unsigned int xfs_buf_flags_t;
58
59#define XFS_BUF_FLAGS \
60	{ XBF_READ,		"READ" }, \
61	{ XBF_WRITE,		"WRITE" }, \
62	{ XBF_READ_AHEAD,	"READ_AHEAD" }, \
63	{ XBF_NO_IOACCT,	"NO_IOACCT" }, \
64	{ XBF_ASYNC,		"ASYNC" }, \
65	{ XBF_DONE,		"DONE" }, \
66	{ XBF_STALE,		"STALE" }, \
67	{ XBF_WRITE_FAIL,	"WRITE_FAIL" }, \
68	{ _XBF_INODES,		"INODES" }, \
69	{ _XBF_DQUOTS,		"DQUOTS" }, \
70	{ _XBF_LOGRECOVERY,	"LOG_RECOVERY" }, \
71	{ _XBF_PAGES,		"PAGES" }, \
72	{ _XBF_KMEM,		"KMEM" }, \
73	{ _XBF_DELWRI_Q,	"DELWRI_Q" }, \
74	/* The following interface flags should never be set */ \
75	{ XBF_LIVESCAN,		"LIVESCAN" }, \
76	{ XBF_INCORE,		"INCORE" }, \
77	{ XBF_TRYLOCK,		"TRYLOCK" }, \
78	{ XBF_UNMAPPED,		"UNMAPPED" }
79
80/*
81 * Internal state flags.
82 */
83#define XFS_BSTATE_DISPOSE	 (1 << 0)	/* buffer being discarded */
84#define XFS_BSTATE_IN_FLIGHT	 (1 << 1)	/* I/O in flight */
85
86struct xfs_buf_cache {
87	spinlock_t		bc_lock;
88	struct rhashtable	bc_hash;
89};
90
91int xfs_buf_cache_init(struct xfs_buf_cache *bch);
92void xfs_buf_cache_destroy(struct xfs_buf_cache *bch);
93
94/*
95 * The xfs_buftarg contains 2 notions of "sector size" -
96 *
97 * 1) The metadata sector size, which is the minimum unit and
98 *    alignment of IO which will be performed by metadata operations.
99 * 2) The device logical sector size
100 *
101 * The first is specified at mkfs time, and is stored on-disk in the
102 * superblock's sb_sectsize.
103 *
104 * The latter is derived from the underlying device, and controls direct IO
105 * alignment constraints.
106 */
107struct xfs_buftarg {
108	dev_t			bt_dev;
109	struct file		*bt_bdev_file;
110	struct block_device	*bt_bdev;
111	struct dax_device	*bt_daxdev;
112	struct file		*bt_file;
113	u64			bt_dax_part_off;
114	struct xfs_mount	*bt_mount;
115	unsigned int		bt_meta_sectorsize;
116	size_t			bt_meta_sectormask;
117	size_t			bt_logical_sectorsize;
118	size_t			bt_logical_sectormask;
119
120	/* LRU control structures */
121	struct shrinker		*bt_shrinker;
122	struct list_lru		bt_lru;
123
124	struct percpu_counter	bt_io_count;
125	struct ratelimit_state	bt_ioerror_rl;
126
127	/* built-in cache, if we're not using the perag one */
128	struct xfs_buf_cache	bt_cache[];
129};
130
131#define XB_PAGES	2
132
133struct xfs_buf_map {
134	xfs_daddr_t		bm_bn;	/* block number for I/O */
135	int			bm_len;	/* size of I/O */
136	unsigned int		bm_flags;
137};
138
139/*
140 * Online fsck is scanning the buffer cache for live buffers.  Do not warn
141 * about length mismatches during lookups and do not return stale buffers.
142 */
143#define XBM_LIVESCAN		(1U << 0)
144
145#define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
146	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
147
148struct xfs_buf_ops {
149	char *name;
150	union {
151		__be32 magic[2];	/* v4 and v5 on disk magic values */
152		__be16 magic16[2];	/* v4 and v5 on disk magic values */
153	};
154	void (*verify_read)(struct xfs_buf *);
155	void (*verify_write)(struct xfs_buf *);
156	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
157};
158
159struct xfs_buf {
160	/*
161	 * first cacheline holds all the fields needed for an uncontended cache
162	 * hit to be fully processed. The semaphore straddles the cacheline
163	 * boundary, but the counter and lock sits on the first cacheline,
164	 * which is the only bit that is touched if we hit the semaphore
165	 * fast-path on locking.
166	 */
167	struct rhash_head	b_rhash_head;	/* pag buffer hash node */
168
169	xfs_daddr_t		b_rhash_key;	/* buffer cache index */
170	int			b_length;	/* size of buffer in BBs */
171	atomic_t		b_hold;		/* reference count */
172	atomic_t		b_lru_ref;	/* lru reclaim ref count */
173	xfs_buf_flags_t		b_flags;	/* status flags */
174	struct semaphore	b_sema;		/* semaphore for lockables */
175
176	/*
177	 * concurrent access to b_lru and b_lru_flags are protected by
178	 * bt_lru_lock and not by b_sema
179	 */
180	struct list_head	b_lru;		/* lru list */
181	spinlock_t		b_lock;		/* internal state lock */
182	unsigned int		b_state;	/* internal state flags */
183	int			b_io_error;	/* internal IO error state */
184	wait_queue_head_t	b_waiters;	/* unpin waiters */
185	struct list_head	b_list;
186	struct xfs_perag	*b_pag;		/* contains rbtree root */
187	struct xfs_mount	*b_mount;
188	struct xfs_buftarg	*b_target;	/* buffer target (device) */
189	void			*b_addr;	/* virtual address of buffer */
190	struct work_struct	b_ioend_work;
191	struct completion	b_iowait;	/* queue for I/O waiters */
192	struct xfs_buf_log_item	*b_log_item;
193	struct list_head	b_li_list;	/* Log items list head */
194	struct xfs_trans	*b_transp;
195	struct page		**b_pages;	/* array of page pointers */
196	struct page		*b_page_array[XB_PAGES]; /* inline pages */
197	struct xfs_buf_map	*b_maps;	/* compound buffer map */
198	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
199	int			b_map_count;
200	atomic_t		b_pin_count;	/* pin count */
201	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
202	unsigned int		b_page_count;	/* size of page array */
203	unsigned int		b_offset;	/* page offset of b_addr,
204						   only for _XBF_KMEM buffers */
205	int			b_error;	/* error code on I/O */
206
207	/*
208	 * async write failure retry count. Initialised to zero on the first
209	 * failure, then when it exceeds the maximum configured without a
210	 * success the write is considered to be failed permanently and the
211	 * iodone handler will take appropriate action.
212	 *
213	 * For retry timeouts, we record the jiffie of the first failure. This
214	 * means that we can change the retry timeout for buffers already under
215	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
216	 *
217	 * last_error is used to ensure that we are getting repeated errors, not
218	 * different errors. e.g. a block device might change ENOSPC to EIO when
219	 * a failure timeout occurs, so we want to re-initialise the error
220	 * retry behaviour appropriately when that happens.
221	 */
222	int			b_retries;
223	unsigned long		b_first_retry_time; /* in jiffies */
224	int			b_last_error;
225
226	const struct xfs_buf_ops	*b_ops;
227	struct rcu_head		b_rcu;
228};
229
230/* Finding and Reading Buffers */
231int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
232		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
233int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
234		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
235		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
236void xfs_buf_readahead_map(struct xfs_buftarg *target,
237			       struct xfs_buf_map *map, int nmaps,
238			       const struct xfs_buf_ops *ops);
239
240static inline int
241xfs_buf_incore(
242	struct xfs_buftarg	*target,
243	xfs_daddr_t		blkno,
244	size_t			numblks,
245	xfs_buf_flags_t		flags,
246	struct xfs_buf		**bpp)
247{
248	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
249
250	return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
251}
252
253static inline int
254xfs_buf_get(
255	struct xfs_buftarg	*target,
256	xfs_daddr_t		blkno,
257	size_t			numblks,
258	struct xfs_buf		**bpp)
259{
260	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
261
262	return xfs_buf_get_map(target, &map, 1, 0, bpp);
263}
264
265static inline int
266xfs_buf_read(
267	struct xfs_buftarg	*target,
268	xfs_daddr_t		blkno,
269	size_t			numblks,
270	xfs_buf_flags_t		flags,
271	struct xfs_buf		**bpp,
272	const struct xfs_buf_ops *ops)
273{
274	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
275
276	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
277			__builtin_return_address(0));
278}
279
280static inline void
281xfs_buf_readahead(
282	struct xfs_buftarg	*target,
283	xfs_daddr_t		blkno,
284	size_t			numblks,
285	const struct xfs_buf_ops *ops)
286{
287	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
288	return xfs_buf_readahead_map(target, &map, 1, ops);
289}
290
291int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
292		xfs_buf_flags_t flags, struct xfs_buf **bpp);
293int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
294		size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
295		const struct xfs_buf_ops *ops);
296int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
297void xfs_buf_hold(struct xfs_buf *bp);
298
299/* Releasing Buffers */
300extern void xfs_buf_rele(struct xfs_buf *);
301
302/* Locking and Unlocking Buffers */
303extern int xfs_buf_trylock(struct xfs_buf *);
304extern void xfs_buf_lock(struct xfs_buf *);
305extern void xfs_buf_unlock(struct xfs_buf *);
306#define xfs_buf_islocked(bp) \
307	((bp)->b_sema.count <= 0)
308
309static inline void xfs_buf_relse(struct xfs_buf *bp)
310{
311	xfs_buf_unlock(bp);
312	xfs_buf_rele(bp);
313}
314
315/* Buffer Read and Write Routines */
316extern int xfs_bwrite(struct xfs_buf *bp);
317
318extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
319		xfs_failaddr_t failaddr);
320#define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
321extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
322void xfs_buf_ioend_fail(struct xfs_buf *);
323void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
324void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
325#define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
326
327/* Buffer Utility Routines */
328extern void *xfs_buf_offset(struct xfs_buf *, size_t);
329extern void xfs_buf_stale(struct xfs_buf *bp);
330
331/* Delayed Write Buffer Routines */
332extern void xfs_buf_delwri_cancel(struct list_head *);
333extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
334void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
335extern int xfs_buf_delwri_submit(struct list_head *);
336extern int xfs_buf_delwri_submit_nowait(struct list_head *);
337extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
338
339static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
340{
341	return bp->b_maps[0].bm_bn;
342}
343
344void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
345
346/*
347 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
348 * up with a reference count of 0 so it will be tossed from the cache when
349 * released.
350 */
351static inline void xfs_buf_oneshot(struct xfs_buf *bp)
352{
353	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
354		return;
355	atomic_set(&bp->b_lru_ref, 0);
356}
357
358static inline int xfs_buf_ispinned(struct xfs_buf *bp)
359{
360	return atomic_read(&bp->b_pin_count);
361}
362
363static inline int
364xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
365{
366	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
367				cksum_offset);
368}
369
370static inline void
371xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
372{
373	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
374			 cksum_offset);
375}
376
377/*
378 *	Handling of buftargs.
379 */
380struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
381		struct file *bdev_file);
382extern void xfs_free_buftarg(struct xfs_buftarg *);
383extern void xfs_buftarg_wait(struct xfs_buftarg *);
384extern void xfs_buftarg_drain(struct xfs_buftarg *);
385extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
386
387#define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
388#define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
389
390int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
391bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
392bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
393
394/* for xfs_buf_mem.c only: */
395int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize,
396		const char *descr);
397void xfs_destroy_buftarg(struct xfs_buftarg *btp);
398
399#endif	/* __XFS_BUF_H__ */
400