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 #include <linux/lockref.h>
18
19 extern struct kmem_cache *xfs_buf_cache;
20
21 /*
22 * Base types
23 */
24 struct xfs_buf;
25
26 #define XFS_BUF_DADDR_MAX ((xfs_daddr_t) S64_MAX)
27 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))
28
29 #define XBF_READ (1u << 0) /* buffer intended for reading from device */
30 #define XBF_WRITE (1u << 1) /* buffer intended for writing to device */
31 #define XBF_READ_AHEAD (1u << 2) /* asynchronous read-ahead */
32 #define XBF_ASYNC (1u << 4) /* initiator will not wait for completion */
33 #define XBF_DONE (1u << 5) /* all pages in the buffer uptodate */
34 #define XBF_STALE (1u << 6) /* buffer has been staled, do not find it */
35 #define XBF_WRITE_FAIL (1u << 7) /* async writes have failed on this buffer */
36
37 /* buffer type flags for write callbacks */
38 #define _XBF_LOGRECOVERY (1u << 18)/* log recovery buffer */
39
40 /* flags used only internally */
41 #define _XBF_KMEM (1u << 21)/* backed by heap memory */
42 #define _XBF_DELWRI_Q (1u << 22)/* buffer on a delwri queue */
43
44 /* flags used only as arguments to access routines */
45 /*
46 * Online fsck is scanning the buffer cache for live buffers. Do not warn
47 * about length mismatches during lookups and do not return stale buffers.
48 */
49 #define XBF_LIVESCAN (1u << 28)
50 #define XBF_INCORE (1u << 29)/* lookup only, return if found in cache */
51 #define XBF_TRYLOCK (1u << 30)/* lock requested, but do not wait */
52
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_ASYNC, "ASYNC" }, \
61 { XBF_DONE, "DONE" }, \
62 { XBF_STALE, "STALE" }, \
63 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
64 { _XBF_LOGRECOVERY, "LOG_RECOVERY" }, \
65 { _XBF_KMEM, "KMEM" }, \
66 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
67 /* The following interface flags should never be set */ \
68 { XBF_LIVESCAN, "LIVESCAN" }, \
69 { XBF_INCORE, "INCORE" }, \
70 { XBF_TRYLOCK, "TRYLOCK" }
71
72 /*
73 * The xfs_buftarg contains 2 notions of "sector size" -
74 *
75 * 1) The metadata sector size, which is the minimum unit and
76 * alignment of IO which will be performed by metadata operations.
77 * 2) The device logical sector size
78 *
79 * The first is specified at mkfs time, and is stored on-disk in the
80 * superblock's sb_sectsize.
81 *
82 * The latter is derived from the underlying device, and controls direct IO
83 * alignment constraints.
84 */
85 struct xfs_buftarg {
86 dev_t bt_dev;
87 struct block_device *bt_bdev;
88 struct dax_device *bt_daxdev;
89 struct file *bt_file;
90 u64 bt_dax_part_off;
91 struct xfs_mount *bt_mount;
92 unsigned int bt_meta_sectorsize;
93 size_t bt_meta_sectormask;
94 size_t bt_logical_sectorsize;
95 size_t bt_logical_sectormask;
96 xfs_daddr_t bt_nr_sectors;
97
98 /* LRU control structures */
99 struct shrinker *bt_shrinker;
100 struct list_lru bt_lru;
101
102 struct percpu_counter bt_readahead_count;
103 struct ratelimit_state bt_ioerror_rl;
104
105 /* Hardware atomic write unit values, bytes */
106 unsigned int bt_awu_min;
107 unsigned int bt_awu_max;
108
109 struct rhashtable bt_hash;
110 };
111
112 struct xfs_buf_map {
113 xfs_daddr_t bm_bn; /* block number for I/O */
114 int bm_len; /* size of I/O */
115 unsigned int bm_flags;
116 };
117
118 /*
119 * Online fsck is scanning the buffer cache for live buffers. Do not warn
120 * about length mismatches during lookups and do not return stale buffers.
121 */
122 #define XBM_LIVESCAN (1U << 0)
123
124 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
125 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
126
127 struct xfs_buf_ops {
128 char *name;
129 union {
130 __be32 magic[2]; /* v4 and v5 on disk magic values */
131 __be16 magic16[2]; /* v4 and v5 on disk magic values */
132 };
133 void (*verify_read)(struct xfs_buf *);
134 void (*verify_write)(struct xfs_buf *);
135 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
136 };
137
138 struct xfs_buf {
139 /*
140 * first cacheline holds all the fields needed for an uncontended cache
141 * hit to be fully processed. The semaphore straddles the cacheline
142 * boundary, but the counter and lock sits on the first cacheline,
143 * which is the only bit that is touched if we hit the semaphore
144 * fast-path on locking.
145 */
146 struct rhash_head b_rhash_head; /* pag buffer hash node */
147
148 xfs_daddr_t b_rhash_key; /* buffer cache index */
149 int b_length; /* size of buffer in BBs */
150 struct lockref b_lockref; /* refcount + lock */
151 atomic_t b_lru_ref; /* lru reclaim ref count */
152 xfs_buf_flags_t b_flags; /* status flags */
153 struct semaphore b_sema; /* semaphore for lockables */
154
155 /*
156 * concurrent access to b_lru and b_lru_flags are protected by
157 * bt_lru_lock and not by b_sema
158 */
159 struct list_head b_lru; /* lru list */
160 wait_queue_head_t b_waiters; /* unpin waiters */
161 struct list_head b_list;
162 struct xfs_perag *b_pag;
163 struct xfs_mount *b_mount;
164 struct xfs_buftarg *b_target; /* buffer target (device) */
165 void *b_addr; /* virtual address of buffer */
166 struct work_struct b_ioend_work;
167 struct completion b_iowait; /* queue for I/O waiters */
168 struct xfs_buf_log_item *b_log_item;
169 struct list_head b_li_list; /* Log items list head */
170 struct xfs_trans *b_transp;
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 atomic_t b_pin_count; /* pin count */
175 int b_error; /* error code on I/O */
176 void (*b_iodone)(struct xfs_buf *bp);
177
178 /*
179 * async write failure retry count. Initialised to zero on the first
180 * failure, then when it exceeds the maximum configured without a
181 * success the write is considered to be failed permanently and the
182 * iodone handler will take appropriate action.
183 *
184 * For retry timeouts, we record the jiffy of the first failure. This
185 * means that we can change the retry timeout for buffers already under
186 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
187 *
188 * last_error is used to ensure that we are getting repeated errors, not
189 * different errors. e.g. a block device might change ENOSPC to EIO when
190 * a failure timeout occurs, so we want to re-initialise the error
191 * retry behaviour appropriately when that happens.
192 */
193 int b_retries;
194 unsigned long b_first_retry_time; /* in jiffies */
195 int b_last_error;
196
197 const struct xfs_buf_ops *b_ops;
198 struct rcu_head b_rcu;
199 };
200
201 /* Finding and Reading Buffers */
202 int 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);
204 int 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);
207 void xfs_buf_readahead_map(struct xfs_buftarg *target,
208 struct xfs_buf_map *map, int nmaps,
209 const struct xfs_buf_ops *ops);
210
211 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)212 xfs_buf_incore(
213 struct xfs_buftarg *target,
214 xfs_daddr_t blkno,
215 size_t numblks,
216 xfs_buf_flags_t flags,
217 struct xfs_buf **bpp)
218 {
219 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
220
221 return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
222 }
223
224 static inline int
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp)225 xfs_buf_get(
226 struct xfs_buftarg *target,
227 xfs_daddr_t blkno,
228 size_t numblks,
229 struct xfs_buf **bpp)
230 {
231 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
232
233 return xfs_buf_get_map(target, &map, 1, 0, bpp);
234 }
235
236 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)237 xfs_buf_read(
238 struct xfs_buftarg *target,
239 xfs_daddr_t blkno,
240 size_t numblks,
241 xfs_buf_flags_t flags,
242 struct xfs_buf **bpp,
243 const struct xfs_buf_ops *ops)
244 {
245 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
246
247 return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
248 __builtin_return_address(0));
249 }
250
251 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)252 xfs_buf_readahead(
253 struct xfs_buftarg *target,
254 xfs_daddr_t blkno,
255 size_t numblks,
256 const struct xfs_buf_ops *ops)
257 {
258 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
259 return xfs_buf_readahead_map(target, &map, 1, ops);
260 }
261
262 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
263 struct xfs_buf **bpp);
264 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
265 size_t numblks, struct xfs_buf **bpp,
266 const struct xfs_buf_ops *ops);
267 int _xfs_buf_read(struct xfs_buf *bp);
268 void xfs_buf_hold(struct xfs_buf *bp);
269
270 /* Releasing Buffers */
271 extern void xfs_buf_rele(struct xfs_buf *);
272
273 /* Locking and Unlocking Buffers */
274 extern int xfs_buf_trylock(struct xfs_buf *);
275 extern void xfs_buf_lock(struct xfs_buf *);
276 extern void xfs_buf_unlock(struct xfs_buf *);
277 #define xfs_buf_islocked(bp) \
278 ((bp)->b_sema.count <= 0)
279
xfs_buf_relse(struct xfs_buf * bp)280 static inline void xfs_buf_relse(struct xfs_buf *bp)
281 {
282 xfs_buf_unlock(bp);
283 xfs_buf_rele(bp);
284 }
285
286 /* Buffer Read and Write Routines */
287 extern int xfs_bwrite(struct xfs_buf *bp);
288
289 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
290 xfs_failaddr_t failaddr);
291 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
292 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
293 void xfs_buf_ioend_fail(struct xfs_buf *);
294 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
295 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
296
297 /* Buffer Utility Routines */
xfs_buf_offset(struct xfs_buf * bp,size_t offset)298 static inline void *xfs_buf_offset(struct xfs_buf *bp, size_t offset)
299 {
300 return bp->b_addr + offset;
301 }
302
xfs_buf_zero(struct xfs_buf * bp,size_t boff,size_t bsize)303 static inline void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize)
304 {
305 memset(bp->b_addr + boff, 0, bsize);
306 }
307
308 extern void xfs_buf_stale(struct xfs_buf *bp);
309
310 /* Delayed Write Buffer Routines */
311 extern void xfs_buf_delwri_cancel(struct list_head *);
312 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
313 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
314 extern int xfs_buf_delwri_submit(struct list_head *);
315 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
316
xfs_buf_daddr(struct xfs_buf * bp)317 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
318 {
319 return bp->b_maps[0].bm_bn;
320 }
321
322 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
323
324 /*
325 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
326 * up with a reference count of 0 so it will be tossed from the cache when
327 * released.
328 */
xfs_buf_oneshot(struct xfs_buf * bp)329 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
330 {
331 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
332 return;
333 atomic_set(&bp->b_lru_ref, 0);
334 }
335
xfs_buf_ispinned(struct xfs_buf * bp)336 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
337 {
338 return atomic_read(&bp->b_pin_count);
339 }
340
341 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)342 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
343 {
344 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
345 cksum_offset);
346 }
347
348 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)349 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
350 {
351 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
352 cksum_offset);
353 }
354
355 /*
356 * Handling of buftargs.
357 */
358 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
359 struct file *bdev_file);
360 extern void xfs_free_buftarg(struct xfs_buftarg *);
361 extern void xfs_buftarg_wait(struct xfs_buftarg *);
362 extern void xfs_buftarg_drain(struct xfs_buftarg *);
363 int xfs_configure_buftarg(struct xfs_buftarg *btp, unsigned int sectorsize,
364 xfs_fsblock_t nr_blocks);
365
366 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
367
368 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
369 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
370 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
371
372 /* for xfs_buf_mem.c only: */
373 int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize,
374 const char *descr);
375 void xfs_destroy_buftarg(struct xfs_buftarg *btp);
376
377 #endif /* __XFS_BUF_H__ */
378