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 /* flags used only internally */
38 #define _XBF_KMEM (1u << 21)/* backed by heap memory */
39 #define _XBF_DELWRI_Q (1u << 22)/* buffer on a delwri queue */
40
41 /* flags used only as arguments to access routines */
42 /*
43 * Online fsck is scanning the buffer cache for live buffers. Do not warn
44 * about length mismatches during lookups and do not return stale buffers.
45 */
46 #define XBF_LIVESCAN (1u << 28)
47 #define XBF_INCORE (1u << 29)/* lookup only, return if found in cache */
48 #define XBF_TRYLOCK (1u << 30)/* lock requested, but do not wait */
49
50
51 typedef unsigned int xfs_buf_flags_t;
52
53 #define XFS_BUF_FLAGS \
54 { XBF_READ, "READ" }, \
55 { XBF_WRITE, "WRITE" }, \
56 { XBF_READ_AHEAD, "READ_AHEAD" }, \
57 { XBF_ASYNC, "ASYNC" }, \
58 { XBF_DONE, "DONE" }, \
59 { XBF_STALE, "STALE" }, \
60 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
61 { _XBF_KMEM, "KMEM" }, \
62 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
63 /* The following interface flags should never be set */ \
64 { XBF_LIVESCAN, "LIVESCAN" }, \
65 { XBF_INCORE, "INCORE" }, \
66 { XBF_TRYLOCK, "TRYLOCK" }
67
68 /*
69 * The xfs_buftarg contains 2 notions of "sector size" -
70 *
71 * 1) The metadata sector size, which is the minimum unit and
72 * alignment of IO which will be performed by metadata operations.
73 * 2) The device logical sector size
74 *
75 * The first is specified at mkfs time, and is stored on-disk in the
76 * superblock's sb_sectsize.
77 *
78 * The latter is derived from the underlying device, and controls direct IO
79 * alignment constraints.
80 */
81 struct xfs_buftarg {
82 dev_t bt_dev;
83 struct block_device *bt_bdev;
84 struct dax_device *bt_daxdev;
85 struct file *bt_file;
86 u64 bt_dax_part_off;
87 struct xfs_mount *bt_mount;
88 unsigned int bt_meta_sectorsize;
89 size_t bt_meta_sectormask;
90 size_t bt_logical_sectorsize;
91 size_t bt_logical_sectormask;
92 xfs_daddr_t bt_nr_sectors;
93
94 /* LRU control structures */
95 struct shrinker *bt_shrinker;
96 struct list_lru bt_lru;
97
98 struct percpu_counter bt_readahead_count;
99 struct ratelimit_state bt_ioerror_rl;
100
101 /* Hardware atomic write unit values, bytes */
102 unsigned int bt_awu_min;
103 unsigned int bt_awu_max;
104
105 struct rhashtable bt_hash;
106 };
107
108 struct xfs_buf_map {
109 xfs_daddr_t bm_bn; /* block number for I/O */
110 int bm_len; /* size of I/O */
111 unsigned int bm_flags;
112 };
113
114 /*
115 * Online fsck is scanning the buffer cache for live buffers. Do not warn
116 * about length mismatches during lookups and do not return stale buffers.
117 */
118 #define XBM_LIVESCAN (1U << 0)
119
120 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
121 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
122
123 struct xfs_buf_ops {
124 char *name;
125 union {
126 __be32 magic[2]; /* v4 and v5 on disk magic values */
127 __be16 magic16[2]; /* v4 and v5 on disk magic values */
128 };
129 void (*verify_read)(struct xfs_buf *);
130 void (*verify_write)(struct xfs_buf *);
131 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
132 };
133
134 struct xfs_buf {
135 /*
136 * first cacheline holds all the fields needed for an uncontended cache
137 * hit to be fully processed. The semaphore straddles the cacheline
138 * boundary, but the counter and lock sits on the first cacheline,
139 * which is the only bit that is touched if we hit the semaphore
140 * fast-path on locking.
141 */
142 struct rhash_head b_rhash_head; /* pag buffer hash node */
143
144 xfs_daddr_t b_rhash_key; /* buffer cache index */
145 int b_length; /* size of buffer in BBs */
146 struct lockref b_lockref; /* refcount + lock */
147 atomic_t b_lru_ref; /* lru reclaim ref count */
148 xfs_buf_flags_t b_flags; /* status flags */
149 struct semaphore b_sema; /* semaphore for lockables */
150
151 /*
152 * concurrent access to b_lru and b_lru_flags are protected by
153 * bt_lru_lock and not by b_sema
154 */
155 struct list_head b_lru; /* lru list */
156 wait_queue_head_t b_waiters; /* unpin waiters */
157 struct list_head b_list;
158 struct xfs_perag *b_pag;
159 struct xfs_mount *b_mount;
160 struct xfs_buftarg *b_target; /* buffer target (device) */
161 void *b_addr; /* virtual address of buffer */
162 struct work_struct b_ioend_work;
163 struct completion b_iowait; /* queue for I/O waiters */
164 struct xfs_buf_log_item *b_log_item;
165 struct list_head b_li_list; /* Log items list head */
166 struct xfs_trans *b_transp;
167 struct xfs_buf_map *b_maps; /* compound buffer map */
168 struct xfs_buf_map __b_map; /* inline compound buffer map */
169 int b_map_count;
170 atomic_t b_pin_count; /* pin count */
171 int b_error; /* error code on I/O */
172 void (*b_iodone)(struct xfs_buf *bp);
173
174 /*
175 * async write failure retry count. Initialised to zero on the first
176 * failure, then when it exceeds the maximum configured without a
177 * success the write is considered to be failed permanently and the
178 * iodone handler will take appropriate action.
179 *
180 * For retry timeouts, we record the jiffy of the first failure. This
181 * means that we can change the retry timeout for buffers already under
182 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
183 *
184 * last_error is used to ensure that we are getting repeated errors, not
185 * different errors. e.g. a block device might change ENOSPC to EIO when
186 * a failure timeout occurs, so we want to re-initialise the error
187 * retry behaviour appropriately when that happens.
188 */
189 int b_retries;
190 unsigned long b_first_retry_time; /* in jiffies */
191 int b_last_error;
192
193 const struct xfs_buf_ops *b_ops;
194 struct rcu_head b_rcu;
195 };
196
197 /* Finding and Reading Buffers */
198 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
199 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
200 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
201 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
202 const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
203 void xfs_buf_readahead_map(struct xfs_buftarg *target,
204 struct xfs_buf_map *map, int nmaps,
205 const struct xfs_buf_ops *ops);
206
207 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)208 xfs_buf_incore(
209 struct xfs_buftarg *target,
210 xfs_daddr_t blkno,
211 size_t numblks,
212 xfs_buf_flags_t flags,
213 struct xfs_buf **bpp)
214 {
215 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
216
217 return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
218 }
219
220 static inline int
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp)221 xfs_buf_get(
222 struct xfs_buftarg *target,
223 xfs_daddr_t blkno,
224 size_t numblks,
225 struct xfs_buf **bpp)
226 {
227 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
228
229 return xfs_buf_get_map(target, &map, 1, 0, bpp);
230 }
231
232 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)233 xfs_buf_read(
234 struct xfs_buftarg *target,
235 xfs_daddr_t blkno,
236 size_t numblks,
237 xfs_buf_flags_t flags,
238 struct xfs_buf **bpp,
239 const struct xfs_buf_ops *ops)
240 {
241 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
242
243 return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
244 __builtin_return_address(0));
245 }
246
247 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)248 xfs_buf_readahead(
249 struct xfs_buftarg *target,
250 xfs_daddr_t blkno,
251 size_t numblks,
252 const struct xfs_buf_ops *ops)
253 {
254 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
255 return xfs_buf_readahead_map(target, &map, 1, ops);
256 }
257
258 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
259 struct xfs_buf **bpp);
260 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
261 size_t numblks, struct xfs_buf **bpp,
262 const struct xfs_buf_ops *ops);
263 int _xfs_buf_read(struct xfs_buf *bp);
264 void xfs_buf_hold(struct xfs_buf *bp);
265
266 /* Releasing Buffers */
267 extern void xfs_buf_rele(struct xfs_buf *);
268
269 /* Locking and Unlocking Buffers */
270 extern int xfs_buf_trylock(struct xfs_buf *);
271 extern void xfs_buf_lock(struct xfs_buf *);
272 extern void xfs_buf_unlock(struct xfs_buf *);
273 #define xfs_buf_islocked(bp) \
274 ((bp)->b_sema.count <= 0)
275
xfs_buf_relse(struct xfs_buf * bp)276 static inline void xfs_buf_relse(struct xfs_buf *bp)
277 {
278 xfs_buf_unlock(bp);
279 xfs_buf_rele(bp);
280 }
281
282 /* Buffer Read and Write Routines */
283 extern int xfs_bwrite(struct xfs_buf *bp);
284
285 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
286 xfs_failaddr_t failaddr);
287 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
288 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
289 void xfs_buf_fail(struct xfs_buf *bp);
290 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
291 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
292
293 /* Buffer Utility Routines */
xfs_buf_offset(struct xfs_buf * bp,size_t offset)294 static inline void *xfs_buf_offset(struct xfs_buf *bp, size_t offset)
295 {
296 return bp->b_addr + offset;
297 }
298
xfs_buf_zero(struct xfs_buf * bp,size_t boff,size_t bsize)299 static inline void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize)
300 {
301 memset(bp->b_addr + boff, 0, bsize);
302 }
303
304 void xfs_buf_set_uptodate(struct xfs_buf *bp);
305 void xfs_buf_stale(struct xfs_buf *bp);
306 void xfs_buf_clear_stale(struct xfs_buf *bp);
307
308 /* Delayed Write Buffer Routines */
309 extern void xfs_buf_delwri_cancel(struct list_head *);
310 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
311 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
312 extern int xfs_buf_delwri_submit(struct list_head *);
313 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
314
xfs_buf_daddr(struct xfs_buf * bp)315 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
316 {
317 return bp->b_maps[0].bm_bn;
318 }
319
320 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
321
322 /*
323 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
324 * up with a reference count of 0 so it will be tossed from the cache when
325 * released.
326 */
xfs_buf_oneshot(struct xfs_buf * bp)327 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
328 {
329 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
330 return;
331 atomic_set(&bp->b_lru_ref, 0);
332 }
333
xfs_buf_ispinned(struct xfs_buf * bp)334 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
335 {
336 return atomic_read(&bp->b_pin_count);
337 }
338
339 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)340 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
341 {
342 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
343 cksum_offset);
344 }
345
346 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)347 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
348 {
349 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
350 cksum_offset);
351 }
352
353 /*
354 * Handling of buftargs.
355 */
356 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
357 struct file *bdev_file);
358 extern void xfs_free_buftarg(struct xfs_buftarg *);
359 extern void xfs_buftarg_wait(struct xfs_buftarg *);
360 extern void xfs_buftarg_drain(struct xfs_buftarg *);
361 int xfs_configure_buftarg(struct xfs_buftarg *btp, unsigned int sectorsize,
362 xfs_fsblock_t nr_blocks);
363
364 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
365
366 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
367 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
368
369 /* for xfs_buf_mem.c only: */
370 int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize,
371 const char *descr);
372 void xfs_destroy_buftarg(struct xfs_buftarg *btp);
373
374 #endif /* __XFS_BUF_H__ */
375