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 #define XBF_READ (1 << 0) /* buffer intended for reading from device */
25 #define XBF_WRITE (1 << 1) /* buffer intended for writing to device */
26 #define XBF_READ_AHEAD (1 << 2) /* asynchronous read-ahead */
27 #define XBF_NO_IOACCT (1 << 3) /* bypass I/O accounting (non-LRU bufs) */
28 #define XBF_ASYNC (1 << 4) /* initiator will not wait for completion */
29 #define XBF_DONE (1 << 5) /* all pages in the buffer uptodate */
30 #define XBF_STALE (1 << 6) /* buffer has been staled, do not find it */
31 #define XBF_WRITE_FAIL (1 << 7) /* async writes have failed on this buffer */
32
33 /* flags used only as arguments to access routines */
34 #define XBF_TRYLOCK (1 << 16)/* lock requested, but do not wait */
35 #define XBF_UNMAPPED (1 << 17)/* do not map the buffer */
36
37 /* flags used only internally */
38 #define _XBF_PAGES (1 << 20)/* backed by refcounted pages */
39 #define _XBF_KMEM (1 << 21)/* backed by heap memory */
40 #define _XBF_DELWRI_Q (1 << 22)/* buffer on a delwri queue */
41
42 typedef unsigned int xfs_buf_flags_t;
43
44 #define XFS_BUF_FLAGS \
45 { XBF_READ, "READ" }, \
46 { XBF_WRITE, "WRITE" }, \
47 { XBF_READ_AHEAD, "READ_AHEAD" }, \
48 { XBF_NO_IOACCT, "NO_IOACCT" }, \
49 { XBF_ASYNC, "ASYNC" }, \
50 { XBF_DONE, "DONE" }, \
51 { XBF_STALE, "STALE" }, \
52 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
53 { XBF_TRYLOCK, "TRYLOCK" }, /* should never be set */\
54 { XBF_UNMAPPED, "UNMAPPED" }, /* ditto */\
55 { _XBF_PAGES, "PAGES" }, \
56 { _XBF_KMEM, "KMEM" }, \
57 { _XBF_DELWRI_Q, "DELWRI_Q" }
58
59
60 /*
61 * Internal state flags.
62 */
63 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */
64 #define XFS_BSTATE_IN_FLIGHT (1 << 1) /* I/O in flight */
65
66 /*
67 * The xfs_buftarg contains 2 notions of "sector size" -
68 *
69 * 1) The metadata sector size, which is the minimum unit and
70 * alignment of IO which will be performed by metadata operations.
71 * 2) The device logical sector size
72 *
73 * The first is specified at mkfs time, and is stored on-disk in the
74 * superblock's sb_sectsize.
75 *
76 * The latter is derived from the underlying device, and controls direct IO
77 * alignment constraints.
78 */
79 typedef struct xfs_buftarg {
80 dev_t bt_dev;
81 struct block_device *bt_bdev;
82 struct dax_device *bt_daxdev;
83 struct xfs_mount *bt_mount;
84 unsigned int bt_meta_sectorsize;
85 size_t bt_meta_sectormask;
86 size_t bt_logical_sectorsize;
87 size_t bt_logical_sectormask;
88
89 /* LRU control structures */
90 struct shrinker bt_shrinker;
91 struct list_lru bt_lru;
92
93 struct percpu_counter bt_io_count;
94 } xfs_buftarg_t;
95
96 struct xfs_buf;
97 typedef void (*xfs_buf_iodone_t)(struct xfs_buf *);
98
99
100 #define XB_PAGES 2
101
102 struct xfs_buf_map {
103 xfs_daddr_t bm_bn; /* block number for I/O */
104 int bm_len; /* size of I/O */
105 };
106
107 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
108 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
109
110 struct xfs_buf_ops {
111 char *name;
112 union {
113 __be32 magic[2]; /* v4 and v5 on disk magic values */
114 __be16 magic16[2]; /* v4 and v5 on disk magic values */
115 };
116 void (*verify_read)(struct xfs_buf *);
117 void (*verify_write)(struct xfs_buf *);
118 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
119 };
120
121 typedef struct xfs_buf {
122 /*
123 * first cacheline holds all the fields needed for an uncontended cache
124 * hit to be fully processed. The semaphore straddles the cacheline
125 * boundary, but the counter and lock sits on the first cacheline,
126 * which is the only bit that is touched if we hit the semaphore
127 * fast-path on locking.
128 */
129 struct rhash_head b_rhash_head; /* pag buffer hash node */
130 xfs_daddr_t b_bn; /* block number of buffer */
131 int b_length; /* size of buffer in BBs */
132 atomic_t b_hold; /* reference count */
133 atomic_t b_lru_ref; /* lru reclaim ref count */
134 xfs_buf_flags_t b_flags; /* status flags */
135 struct semaphore b_sema; /* semaphore for lockables */
136
137 /*
138 * concurrent access to b_lru and b_lru_flags are protected by
139 * bt_lru_lock and not by b_sema
140 */
141 struct list_head b_lru; /* lru list */
142 spinlock_t b_lock; /* internal state lock */
143 unsigned int b_state; /* internal state flags */
144 int b_io_error; /* internal IO error state */
145 wait_queue_head_t b_waiters; /* unpin waiters */
146 struct list_head b_list;
147 struct xfs_perag *b_pag; /* contains rbtree root */
148 struct xfs_mount *b_mount;
149 xfs_buftarg_t *b_target; /* buffer target (device) */
150 void *b_addr; /* virtual address of buffer */
151 struct work_struct b_ioend_work;
152 xfs_buf_iodone_t b_iodone; /* I/O completion function */
153 struct completion b_iowait; /* queue for I/O waiters */
154 struct xfs_buf_log_item *b_log_item;
155 struct list_head b_li_list; /* Log items list head */
156 struct xfs_trans *b_transp;
157 struct page **b_pages; /* array of page pointers */
158 struct page *b_page_array[XB_PAGES]; /* inline pages */
159 struct xfs_buf_map *b_maps; /* compound buffer map */
160 struct xfs_buf_map __b_map; /* inline compound buffer map */
161 int b_map_count;
162 atomic_t b_pin_count; /* pin count */
163 atomic_t b_io_remaining; /* #outstanding I/O requests */
164 unsigned int b_page_count; /* size of page array */
165 unsigned int b_offset; /* page offset in first page */
166 int b_error; /* error code on I/O */
167
168 /*
169 * async write failure retry count. Initialised to zero on the first
170 * failure, then when it exceeds the maximum configured without a
171 * success the write is considered to be failed permanently and the
172 * iodone handler will take appropriate action.
173 *
174 * For retry timeouts, we record the jiffie of the first failure. This
175 * means that we can change the retry timeout for buffers already under
176 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
177 *
178 * last_error is used to ensure that we are getting repeated errors, not
179 * different errors. e.g. a block device might change ENOSPC to EIO when
180 * a failure timeout occurs, so we want to re-initialise the error
181 * retry behaviour appropriately when that happens.
182 */
183 int b_retries;
184 unsigned long b_first_retry_time; /* in jiffies */
185 int b_last_error;
186
187 const struct xfs_buf_ops *b_ops;
188 } xfs_buf_t;
189
190 /* Finding and Reading Buffers */
191 struct xfs_buf *xfs_buf_incore(struct xfs_buftarg *target,
192 xfs_daddr_t blkno, size_t numblks,
193 xfs_buf_flags_t flags);
194
195 struct xfs_buf *xfs_buf_get_map(struct xfs_buftarg *target,
196 struct xfs_buf_map *map, int nmaps,
197 xfs_buf_flags_t flags);
198 struct xfs_buf *xfs_buf_read_map(struct xfs_buftarg *target,
199 struct xfs_buf_map *map, int nmaps,
200 xfs_buf_flags_t flags,
201 const struct xfs_buf_ops *ops);
202 void xfs_buf_readahead_map(struct xfs_buftarg *target,
203 struct xfs_buf_map *map, int nmaps,
204 const struct xfs_buf_ops *ops);
205
206 static inline struct xfs_buf *
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks)207 xfs_buf_get(
208 struct xfs_buftarg *target,
209 xfs_daddr_t blkno,
210 size_t numblks)
211 {
212 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
213 return xfs_buf_get_map(target, &map, 1, 0);
214 }
215
216 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)217 xfs_buf_read(
218 struct xfs_buftarg *target,
219 xfs_daddr_t blkno,
220 size_t numblks,
221 xfs_buf_flags_t flags,
222 const struct xfs_buf_ops *ops)
223 {
224 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
225 return xfs_buf_read_map(target, &map, 1, flags, ops);
226 }
227
228 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)229 xfs_buf_readahead(
230 struct xfs_buftarg *target,
231 xfs_daddr_t blkno,
232 size_t numblks,
233 const struct xfs_buf_ops *ops)
234 {
235 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
236 return xfs_buf_readahead_map(target, &map, 1, ops);
237 }
238
239 struct xfs_buf *xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
240 int flags);
241 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
242 size_t numblks, int flags, struct xfs_buf **bpp,
243 const struct xfs_buf_ops *ops);
244 void xfs_buf_hold(struct xfs_buf *bp);
245
246 /* Releasing Buffers */
247 extern void xfs_buf_free(xfs_buf_t *);
248 extern void xfs_buf_rele(xfs_buf_t *);
249
250 /* Locking and Unlocking Buffers */
251 extern int xfs_buf_trylock(xfs_buf_t *);
252 extern void xfs_buf_lock(xfs_buf_t *);
253 extern void xfs_buf_unlock(xfs_buf_t *);
254 #define xfs_buf_islocked(bp) \
255 ((bp)->b_sema.count <= 0)
256
257 /* Buffer Read and Write Routines */
258 extern int xfs_bwrite(struct xfs_buf *bp);
259 extern void xfs_buf_ioend(struct xfs_buf *bp);
260 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
261 xfs_failaddr_t failaddr);
262 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
263 extern void xfs_buf_ioerror_alert(struct xfs_buf *, const char *func);
264
265 extern int __xfs_buf_submit(struct xfs_buf *bp, bool);
xfs_buf_submit(struct xfs_buf * bp)266 static inline int xfs_buf_submit(struct xfs_buf *bp)
267 {
268 bool wait = bp->b_flags & XBF_ASYNC ? false : true;
269 return __xfs_buf_submit(bp, wait);
270 }
271
272 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
273 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
274 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
275
276 /* Buffer Utility Routines */
277 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
278 extern void xfs_buf_stale(struct xfs_buf *bp);
279
280 /* Delayed Write Buffer Routines */
281 extern void xfs_buf_delwri_cancel(struct list_head *);
282 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
283 extern int xfs_buf_delwri_submit(struct list_head *);
284 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
285 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
286
287 /* Buffer Daemon Setup Routines */
288 extern int xfs_buf_init(void);
289 extern void xfs_buf_terminate(void);
290
291 /*
292 * These macros use the IO block map rather than b_bn. b_bn is now really
293 * just for the buffer cache index for cached buffers. As IO does not use b_bn
294 * anymore, uncached buffers do not use b_bn at all and hence must modify the IO
295 * map directly. Uncached buffers are not allowed to be discontiguous, so this
296 * is safe to do.
297 *
298 * In future, uncached buffers will pass the block number directly to the io
299 * request function and hence these macros will go away at that point.
300 */
301 #define XFS_BUF_ADDR(bp) ((bp)->b_maps[0].bm_bn)
302 #define XFS_BUF_SET_ADDR(bp, bno) ((bp)->b_maps[0].bm_bn = (xfs_daddr_t)(bno))
303
304 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
305
306 /*
307 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
308 * up with a reference count of 0 so it will be tossed from the cache when
309 * released.
310 */
xfs_buf_oneshot(struct xfs_buf * bp)311 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
312 {
313 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
314 return;
315 atomic_set(&bp->b_lru_ref, 0);
316 }
317
xfs_buf_ispinned(struct xfs_buf * bp)318 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
319 {
320 return atomic_read(&bp->b_pin_count);
321 }
322
xfs_buf_relse(xfs_buf_t * bp)323 static inline void xfs_buf_relse(xfs_buf_t *bp)
324 {
325 xfs_buf_unlock(bp);
326 xfs_buf_rele(bp);
327 }
328
329 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)330 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
331 {
332 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
333 cksum_offset);
334 }
335
336 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)337 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
338 {
339 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
340 cksum_offset);
341 }
342
343 /*
344 * Handling of buftargs.
345 */
346 extern xfs_buftarg_t *xfs_alloc_buftarg(struct xfs_mount *,
347 struct block_device *, struct dax_device *);
348 extern void xfs_free_buftarg(struct xfs_buftarg *);
349 extern void xfs_wait_buftarg(xfs_buftarg_t *);
350 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int);
351
352 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev)
353 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
354
355 static inline int
xfs_buftarg_dma_alignment(struct xfs_buftarg * bt)356 xfs_buftarg_dma_alignment(struct xfs_buftarg *bt)
357 {
358 return queue_dma_alignment(bt->bt_bdev->bd_disk->queue);
359 }
360
361 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
362 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
363 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
364
365 #endif /* __XFS_BUF_H__ */
366