xref: /linux/fs/xfs/xfs_buf.h (revision 3fd6c59042dbba50391e30862beac979491145fe)
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 /*
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 
57 typedef 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 
86 struct xfs_buf_cache {
87 	spinlock_t		bc_lock;
88 	struct rhashtable	bc_hash;
89 };
90 
91 int xfs_buf_cache_init(struct xfs_buf_cache *bch);
92 void 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  */
107 struct 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 	/* Atomic write unit values */
128 	unsigned int		bt_bdev_awu_min;
129 	unsigned int		bt_bdev_awu_max;
130 
131 	/* built-in cache, if we're not using the perag one */
132 	struct xfs_buf_cache	bt_cache[];
133 };
134 
135 #define XB_PAGES	2
136 
137 struct xfs_buf_map {
138 	xfs_daddr_t		bm_bn;	/* block number for I/O */
139 	int			bm_len;	/* size of I/O */
140 	unsigned int		bm_flags;
141 };
142 
143 /*
144  * Online fsck is scanning the buffer cache for live buffers.  Do not warn
145  * about length mismatches during lookups and do not return stale buffers.
146  */
147 #define XBM_LIVESCAN		(1U << 0)
148 
149 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
150 	struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
151 
152 struct xfs_buf_ops {
153 	char *name;
154 	union {
155 		__be32 magic[2];	/* v4 and v5 on disk magic values */
156 		__be16 magic16[2];	/* v4 and v5 on disk magic values */
157 	};
158 	void (*verify_read)(struct xfs_buf *);
159 	void (*verify_write)(struct xfs_buf *);
160 	xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
161 };
162 
163 struct xfs_buf {
164 	/*
165 	 * first cacheline holds all the fields needed for an uncontended cache
166 	 * hit to be fully processed. The semaphore straddles the cacheline
167 	 * boundary, but the counter and lock sits on the first cacheline,
168 	 * which is the only bit that is touched if we hit the semaphore
169 	 * fast-path on locking.
170 	 */
171 	struct rhash_head	b_rhash_head;	/* pag buffer hash node */
172 
173 	xfs_daddr_t		b_rhash_key;	/* buffer cache index */
174 	int			b_length;	/* size of buffer in BBs */
175 	atomic_t		b_hold;		/* reference count */
176 	atomic_t		b_lru_ref;	/* lru reclaim ref count */
177 	xfs_buf_flags_t		b_flags;	/* status flags */
178 	struct semaphore	b_sema;		/* semaphore for lockables */
179 
180 	/*
181 	 * concurrent access to b_lru and b_lru_flags are protected by
182 	 * bt_lru_lock and not by b_sema
183 	 */
184 	struct list_head	b_lru;		/* lru list */
185 	spinlock_t		b_lock;		/* internal state lock */
186 	unsigned int		b_state;	/* internal state flags */
187 	int			b_io_error;	/* internal IO error state */
188 	wait_queue_head_t	b_waiters;	/* unpin waiters */
189 	struct list_head	b_list;
190 	struct xfs_perag	*b_pag;		/* contains rbtree root */
191 	struct xfs_mount	*b_mount;
192 	struct xfs_buftarg	*b_target;	/* buffer target (device) */
193 	void			*b_addr;	/* virtual address of buffer */
194 	struct work_struct	b_ioend_work;
195 	struct completion	b_iowait;	/* queue for I/O waiters */
196 	struct xfs_buf_log_item	*b_log_item;
197 	struct list_head	b_li_list;	/* Log items list head */
198 	struct xfs_trans	*b_transp;
199 	struct page		**b_pages;	/* array of page pointers */
200 	struct page		*b_page_array[XB_PAGES]; /* inline pages */
201 	struct xfs_buf_map	*b_maps;	/* compound buffer map */
202 	struct xfs_buf_map	__b_map;	/* inline compound buffer map */
203 	int			b_map_count;
204 	atomic_t		b_pin_count;	/* pin count */
205 	atomic_t		b_io_remaining;	/* #outstanding I/O requests */
206 	unsigned int		b_page_count;	/* size of page array */
207 	unsigned int		b_offset;	/* page offset of b_addr,
208 						   only for _XBF_KMEM buffers */
209 	int			b_error;	/* error code on I/O */
210 
211 	/*
212 	 * async write failure retry count. Initialised to zero on the first
213 	 * failure, then when it exceeds the maximum configured without a
214 	 * success the write is considered to be failed permanently and the
215 	 * iodone handler will take appropriate action.
216 	 *
217 	 * For retry timeouts, we record the jiffy of the first failure. This
218 	 * means that we can change the retry timeout for buffers already under
219 	 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
220 	 *
221 	 * last_error is used to ensure that we are getting repeated errors, not
222 	 * different errors. e.g. a block device might change ENOSPC to EIO when
223 	 * a failure timeout occurs, so we want to re-initialise the error
224 	 * retry behaviour appropriately when that happens.
225 	 */
226 	int			b_retries;
227 	unsigned long		b_first_retry_time; /* in jiffies */
228 	int			b_last_error;
229 
230 	const struct xfs_buf_ops	*b_ops;
231 	struct rcu_head		b_rcu;
232 };
233 
234 /* Finding and Reading Buffers */
235 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
236 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
237 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
238 		int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
239 		const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
240 void xfs_buf_readahead_map(struct xfs_buftarg *target,
241 			       struct xfs_buf_map *map, int nmaps,
242 			       const struct xfs_buf_ops *ops);
243 
244 static inline int
xfs_buf_incore(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,xfs_buf_flags_t flags,struct xfs_buf ** bpp)245 xfs_buf_incore(
246 	struct xfs_buftarg	*target,
247 	xfs_daddr_t		blkno,
248 	size_t			numblks,
249 	xfs_buf_flags_t		flags,
250 	struct xfs_buf		**bpp)
251 {
252 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
253 
254 	return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
255 }
256 
257 static inline int
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp)258 xfs_buf_get(
259 	struct xfs_buftarg	*target,
260 	xfs_daddr_t		blkno,
261 	size_t			numblks,
262 	struct xfs_buf		**bpp)
263 {
264 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
265 
266 	return xfs_buf_get_map(target, &map, 1, 0, bpp);
267 }
268 
269 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)270 xfs_buf_read(
271 	struct xfs_buftarg	*target,
272 	xfs_daddr_t		blkno,
273 	size_t			numblks,
274 	xfs_buf_flags_t		flags,
275 	struct xfs_buf		**bpp,
276 	const struct xfs_buf_ops *ops)
277 {
278 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
279 
280 	return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
281 			__builtin_return_address(0));
282 }
283 
284 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)285 xfs_buf_readahead(
286 	struct xfs_buftarg	*target,
287 	xfs_daddr_t		blkno,
288 	size_t			numblks,
289 	const struct xfs_buf_ops *ops)
290 {
291 	DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
292 	return xfs_buf_readahead_map(target, &map, 1, ops);
293 }
294 
295 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
296 		xfs_buf_flags_t flags, struct xfs_buf **bpp);
297 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
298 		size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
299 		const struct xfs_buf_ops *ops);
300 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
301 void xfs_buf_hold(struct xfs_buf *bp);
302 
303 /* Releasing Buffers */
304 extern void xfs_buf_rele(struct xfs_buf *);
305 
306 /* Locking and Unlocking Buffers */
307 extern int xfs_buf_trylock(struct xfs_buf *);
308 extern void xfs_buf_lock(struct xfs_buf *);
309 extern void xfs_buf_unlock(struct xfs_buf *);
310 #define xfs_buf_islocked(bp) \
311 	((bp)->b_sema.count <= 0)
312 
xfs_buf_relse(struct xfs_buf * bp)313 static inline void xfs_buf_relse(struct xfs_buf *bp)
314 {
315 	xfs_buf_unlock(bp);
316 	xfs_buf_rele(bp);
317 }
318 
319 /* Buffer Read and Write Routines */
320 extern int xfs_bwrite(struct xfs_buf *bp);
321 
322 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
323 		xfs_failaddr_t failaddr);
324 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
325 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
326 void xfs_buf_ioend_fail(struct xfs_buf *);
327 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
328 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
329 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
330 
331 /* Buffer Utility Routines */
332 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
333 extern void xfs_buf_stale(struct xfs_buf *bp);
334 
335 /* Delayed Write Buffer Routines */
336 extern void xfs_buf_delwri_cancel(struct list_head *);
337 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
338 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
339 extern int xfs_buf_delwri_submit(struct list_head *);
340 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
341 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
342 
xfs_buf_daddr(struct xfs_buf * bp)343 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
344 {
345 	return bp->b_maps[0].bm_bn;
346 }
347 
348 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
349 
350 /*
351  * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
352  * up with a reference count of 0 so it will be tossed from the cache when
353  * released.
354  */
xfs_buf_oneshot(struct xfs_buf * bp)355 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
356 {
357 	if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
358 		return;
359 	atomic_set(&bp->b_lru_ref, 0);
360 }
361 
xfs_buf_ispinned(struct xfs_buf * bp)362 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
363 {
364 	return atomic_read(&bp->b_pin_count);
365 }
366 
367 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)368 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
369 {
370 	return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
371 				cksum_offset);
372 }
373 
374 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)375 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
376 {
377 	xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
378 			 cksum_offset);
379 }
380 
381 /*
382  *	Handling of buftargs.
383  */
384 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
385 		struct file *bdev_file);
386 extern void xfs_free_buftarg(struct xfs_buftarg *);
387 extern void xfs_buftarg_wait(struct xfs_buftarg *);
388 extern void xfs_buftarg_drain(struct xfs_buftarg *);
389 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
390 
391 #define xfs_getsize_buftarg(buftarg)	block_size((buftarg)->bt_bdev)
392 #define xfs_readonly_buftarg(buftarg)	bdev_read_only((buftarg)->bt_bdev)
393 
394 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
395 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
396 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
397 
398 /* for xfs_buf_mem.c only: */
399 int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize,
400 		const char *descr);
401 void xfs_destroy_buftarg(struct xfs_buftarg *btp);
402 
403 #endif	/* __XFS_BUF_H__ */
404