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
22 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))
23
24 typedef enum {
25 XBRW_READ = 1, /* transfer into target memory */
26 XBRW_WRITE = 2, /* transfer from target memory */
27 XBRW_ZERO = 3, /* Zero target memory */
28 } xfs_buf_rw_t;
29
30 #define XBF_READ (1 << 0) /* buffer intended for reading from device */
31 #define XBF_WRITE (1 << 1) /* buffer intended for writing to device */
32 #define XBF_READ_AHEAD (1 << 2) /* asynchronous read-ahead */
33 #define XBF_NO_IOACCT (1 << 3) /* bypass I/O accounting (non-LRU bufs) */
34 #define XBF_ASYNC (1 << 4) /* initiator will not wait for completion */
35 #define XBF_DONE (1 << 5) /* all pages in the buffer uptodate */
36 #define XBF_STALE (1 << 6) /* buffer has been staled, do not find it */
37 #define XBF_WRITE_FAIL (1 << 24)/* async writes have failed on this buffer */
38
39 /* I/O hints for the BIO layer */
40 #define XBF_SYNCIO (1 << 10)/* treat this buffer as synchronous I/O */
41 #define XBF_FUA (1 << 11)/* force cache write through mode */
42 #define XBF_FLUSH (1 << 12)/* flush the disk cache before a write */
43
44 /* flags used only as arguments to access routines */
45 #define XBF_TRYLOCK (1 << 16)/* lock requested, but do not wait */
46 #define XBF_UNMAPPED (1 << 17)/* do not map the buffer */
47
48 /* flags used only internally */
49 #define _XBF_PAGES (1 << 20)/* backed by refcounted pages */
50 #define _XBF_KMEM (1 << 21)/* backed by heap memory */
51 #define _XBF_DELWRI_Q (1 << 22)/* buffer on a delwri queue */
52 #define _XBF_COMPOUND (1 << 23)/* compound buffer */
53
54 typedef unsigned int xfs_buf_flags_t;
55
56 #define XFS_BUF_FLAGS \
57 { XBF_READ, "READ" }, \
58 { XBF_WRITE, "WRITE" }, \
59 { XBF_READ_AHEAD, "READ_AHEAD" }, \
60 { XBF_NO_IOACCT, "NO_IOACCT" }, \
61 { XBF_ASYNC, "ASYNC" }, \
62 { XBF_DONE, "DONE" }, \
63 { XBF_STALE, "STALE" }, \
64 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
65 { XBF_SYNCIO, "SYNCIO" }, \
66 { XBF_FUA, "FUA" }, \
67 { XBF_FLUSH, "FLUSH" }, \
68 { XBF_TRYLOCK, "TRYLOCK" }, /* should never be set */\
69 { XBF_UNMAPPED, "UNMAPPED" }, /* ditto */\
70 { _XBF_PAGES, "PAGES" }, \
71 { _XBF_KMEM, "KMEM" }, \
72 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
73 { _XBF_COMPOUND, "COMPOUND" }
74
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 /*
83 * The xfs_buftarg contains 2 notions of "sector size" -
84 *
85 * 1) The metadata sector size, which is the minimum unit and
86 * alignment of IO which will be performed by metadata operations.
87 * 2) The device logical sector size
88 *
89 * The first is specified at mkfs time, and is stored on-disk in the
90 * superblock's sb_sectsize.
91 *
92 * The latter is derived from the underlying device, and controls direct IO
93 * alignment constraints.
94 */
95 typedef struct xfs_buftarg {
96 dev_t bt_dev;
97 struct block_device *bt_bdev;
98 struct dax_device *bt_daxdev;
99 struct xfs_mount *bt_mount;
100 unsigned int bt_meta_sectorsize;
101 size_t bt_meta_sectormask;
102 size_t bt_logical_sectorsize;
103 size_t bt_logical_sectormask;
104
105 /* LRU control structures */
106 struct shrinker bt_shrinker;
107 struct list_lru bt_lru;
108
109 struct percpu_counter bt_io_count;
110 } xfs_buftarg_t;
111
112 struct xfs_buf;
113 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *);
114
115
116 #define XB_PAGES 2
117
118 struct xfs_buf_map {
119 xfs_daddr_t bm_bn; /* block number for I/O */
120 int bm_len; /* size of I/O */
121 };
122
123 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
124 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
125
126 struct xfs_buf_ops {
127 char *name;
128 void (*verify_read)(struct xfs_buf *);
129 void (*verify_write)(struct xfs_buf *);
130 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
131 };
132
133 typedef struct xfs_buf {
134 /*
135 * first cacheline holds all the fields needed for an uncontended cache
136 * hit to be fully processed. The semaphore straddles the cacheline
137 * boundary, but the counter and lock sits on the first cacheline,
138 * which is the only bit that is touched if we hit the semaphore
139 * fast-path on locking.
140 */
141 struct rhash_head b_rhash_head; /* pag buffer hash node */
142 xfs_daddr_t b_bn; /* block number of buffer */
143 int b_length; /* size of buffer in BBs */
144 atomic_t b_hold; /* reference count */
145 atomic_t b_lru_ref; /* lru reclaim ref count */
146 xfs_buf_flags_t b_flags; /* status flags */
147 struct semaphore b_sema; /* semaphore for lockables */
148
149 /*
150 * concurrent access to b_lru and b_lru_flags are protected by
151 * bt_lru_lock and not by b_sema
152 */
153 struct list_head b_lru; /* lru list */
154 spinlock_t b_lock; /* internal state lock */
155 unsigned int b_state; /* internal state flags */
156 int b_io_error; /* internal IO error state */
157 wait_queue_head_t b_waiters; /* unpin waiters */
158 struct list_head b_list;
159 struct xfs_perag *b_pag; /* contains rbtree root */
160 xfs_buftarg_t *b_target; /* buffer target (device) */
161 void *b_addr; /* virtual address of buffer */
162 struct work_struct b_ioend_work;
163 struct workqueue_struct *b_ioend_wq; /* I/O completion wq */
164 xfs_buf_iodone_t b_iodone; /* I/O completion function */
165 struct completion b_iowait; /* queue for I/O waiters */
166 void *b_log_item;
167 struct list_head b_li_list; /* Log items list head */
168 struct xfs_trans *b_transp;
169 struct page **b_pages; /* array of page pointers */
170 struct page *b_page_array[XB_PAGES]; /* inline pages */
171 struct xfs_buf_map *b_maps; /* compound buffer map */
172 struct xfs_buf_map __b_map; /* inline compound buffer map */
173 int b_map_count;
174 int b_io_length; /* IO size in BBs */
175 atomic_t b_pin_count; /* pin count */
176 atomic_t b_io_remaining; /* #outstanding I/O requests */
177 unsigned int b_page_count; /* size of page array */
178 unsigned int b_offset; /* page offset in first page */
179 int b_error; /* error code on I/O */
180
181 /*
182 * async write failure retry count. Initialised to zero on the first
183 * failure, then when it exceeds the maximum configured without a
184 * success the write is considered to be failed permanently and the
185 * iodone handler will take appropriate action.
186 *
187 * For retry timeouts, we record the jiffie of the first failure. This
188 * means that we can change the retry timeout for buffers already under
189 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
190 *
191 * last_error is used to ensure that we are getting repeated errors, not
192 * different errors. e.g. a block device might change ENOSPC to EIO when
193 * a failure timeout occurs, so we want to re-initialise the error
194 * retry behaviour appropriately when that happens.
195 */
196 int b_retries;
197 unsigned long b_first_retry_time; /* in jiffies */
198 int b_last_error;
199
200 const struct xfs_buf_ops *b_ops;
201 } xfs_buf_t;
202
203 /* Finding and Reading Buffers */
204 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
205 xfs_daddr_t blkno, size_t numblks,
206 xfs_buf_flags_t flags);
207
208 struct xfs_buf *_xfs_buf_alloc(struct xfs_buftarg *target,
209 struct xfs_buf_map *map, int nmaps,
210 xfs_buf_flags_t flags);
211
212 static inline struct xfs_buf *
xfs_buf_alloc(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,xfs_buf_flags_t flags)213 xfs_buf_alloc(
214 struct xfs_buftarg *target,
215 xfs_daddr_t blkno,
216 size_t numblks,
217 xfs_buf_flags_t flags)
218 {
219 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
220 return _xfs_buf_alloc(target, &map, 1, flags);
221 }
222
223 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target,
224 struct xfs_buf_map *map, int nmaps,
225 xfs_buf_flags_t flags);
226 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target,
227 struct xfs_buf_map *map, int nmaps,
228 xfs_buf_flags_t flags,
229 const struct xfs_buf_ops *ops);
230 void xfs_buf_readahead_map(struct xfs_buftarg *target,
231 struct xfs_buf_map *map, int nmaps,
232 const struct xfs_buf_ops *ops);
233
234 static inline struct xfs_buf *
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,xfs_buf_flags_t flags)235 xfs_buf_get(
236 struct xfs_buftarg *target,
237 xfs_daddr_t blkno,
238 size_t numblks,
239 xfs_buf_flags_t flags)
240 {
241 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
242 return xfs_buf_get_map(target, &map, 1, flags);
243 }
244
245 static inline struct xfs_buf *
xfs_buf_read(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,xfs_buf_flags_t flags,const struct xfs_buf_ops * ops)246 xfs_buf_read(
247 struct xfs_buftarg *target,
248 xfs_daddr_t blkno,
249 size_t numblks,
250 xfs_buf_flags_t flags,
251 const struct xfs_buf_ops *ops)
252 {
253 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
254 return xfs_buf_read_map(target, &map, 1, flags, ops);
255 }
256
257 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)258 xfs_buf_readahead(
259 struct xfs_buftarg *target,
260 xfs_daddr_t blkno,
261 size_t numblks,
262 const struct xfs_buf_ops *ops)
263 {
264 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
265 return xfs_buf_readahead_map(target, &map, 1, ops);
266 }
267
268 void xfs_buf_set_empty(struct xfs_buf *bp, size_t numblks);
269 int xfs_buf_associate_memory(struct xfs_buf *bp, void *mem, size_t length);
270
271 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
272 int flags);
273 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
274 size_t numblks, int flags, struct xfs_buf **bpp,
275 const struct xfs_buf_ops *ops);
276 void xfs_buf_hold(struct xfs_buf *bp);
277
278 /* Releasing Buffers */
279 extern void xfs_buf_free(xfs_buf_t *);
280 extern void xfs_buf_rele(xfs_buf_t *);
281
282 /* Locking and Unlocking Buffers */
283 extern int xfs_buf_trylock(xfs_buf_t *);
284 extern void xfs_buf_lock(xfs_buf_t *);
285 extern void xfs_buf_unlock(xfs_buf_t *);
286 #define xfs_buf_islocked(bp) \
287 ((bp)->b_sema.count <= 0)
288
289 /* Buffer Read and Write Routines */
290 extern int xfs_bwrite(struct xfs_buf *bp);
291 extern void xfs_buf_ioend(struct xfs_buf *bp);
292 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
293 xfs_failaddr_t failaddr);
294 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
295 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func);
296
297 extern int __xfs_buf_submit(struct xfs_buf *bp, bool);
xfs_buf_submit(struct xfs_buf * bp)298 static inline int xfs_buf_submit(struct xfs_buf *bp)
299 {
300 bool wait = bp->b_flags & XBF_ASYNC ? false : true;
301 return __xfs_buf_submit(bp, wait);
302 }
303
304 extern void xfs_buf_iomove(xfs_buf_t *, size_t, size_t, void *,
305 xfs_buf_rw_t);
306 #define xfs_buf_zero(bp, off, len) \
307 xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO)
308
309 /* Buffer Utility Routines */
310 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
311 extern void xfs_buf_stale(struct xfs_buf *bp);
312
313 /* Delayed Write Buffer Routines */
314 extern void xfs_buf_delwri_cancel(struct list_head *);
315 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
316 extern int xfs_buf_delwri_submit(struct list_head *);
317 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
318 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
319
320 /* Buffer Daemon Setup Routines */
321 extern int xfs_buf_init(void);
322 extern void xfs_buf_terminate(void);
323
324 /*
325 * These macros use the IO block map rather than b_bn. b_bn is now really
326 * just for the buffer cache index for cached buffers. As IO does not use b_bn
327 * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
328 * map directly. Uncached buffers are not allowed to be discontiguous, so this
329 * is safe to do.
330 *
331 * In future, uncached buffers will pass the block number directly to the io
332 * request function and hence these macros will go away at that point.
333 */
334 #define XFS_BUF_ADDR(bp) ((bp)->b_maps[0].bm_bn)
335 #define XFS_BUF_SET_ADDR(bp, bno) ((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
336
337 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
338
339 /*
340 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
341 * up with a reference count of 0 so it will be tossed from the cache when
342 * released.
343 */
xfs_buf_oneshot(struct xfs_buf * bp)344 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
345 {
346 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
347 return;
348 atomic_set(&bp->b_lru_ref, 0);
349 }
350
xfs_buf_ispinned(struct xfs_buf * bp)351 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
352 {
353 return atomic_read(&bp->b_pin_count);
354 }
355
xfs_buf_relse(xfs_buf_t * bp)356 static inline void xfs_buf_relse(xfs_buf_t *bp)
357 {
358 xfs_buf_unlock(bp);
359 xfs_buf_rele(bp);
360 }
361
362 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)363 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
364 {
365 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
366 cksum_offset);
367 }
368
369 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)370 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
371 {
372 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
373 cksum_offset);
374 }
375
376 /*
377 * Handling of buftargs.
378 */
379 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
380 struct block_device *, struct dax_device *);
381 extern void xfs_free_buftarg(struct xfs_buftarg *);
382 extern void xfs_wait_buftarg(xfs_buftarg_t *);
383 extern int xfs_setsize_buftarg(xfs_buftarg_t *, 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 #endif /* __XFS_BUF_H__ */
389