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_ASYNC (1u << 4) /* initiator will not wait for completion */
31 #define XBF_DONE (1u << 5) /* all pages in the buffer uptodate */
32 #define XBF_STALE (1u << 6) /* buffer has been staled, do not find it */
33 #define XBF_WRITE_FAIL (1u << 7) /* async writes have failed on this buffer */
34
35 /* buffer type flags for write callbacks */
36 #define _XBF_LOGRECOVERY (1u << 18)/* log recovery buffer */
37
38 /* flags used only internally */
39 #define _XBF_PAGES (1u << 20)/* backed by refcounted pages */
40 #define _XBF_KMEM (1u << 21)/* backed by heap memory */
41 #define _XBF_DELWRI_Q (1u << 22)/* buffer on a delwri queue */
42
43 /* flags used only as arguments to access routines */
44 /*
45 * Online fsck is scanning the buffer cache for live buffers. Do not warn
46 * about length mismatches during lookups and do not return stale buffers.
47 */
48 #define XBF_LIVESCAN (1u << 28)
49 #define XBF_INCORE (1u << 29)/* lookup only, return if found in cache */
50 #define XBF_TRYLOCK (1u << 30)/* lock requested, but do not wait */
51 #define XBF_UNMAPPED (1u << 31)/* do not map the buffer */
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_PAGES, "PAGES" }, \
66 { _XBF_KMEM, "KMEM" }, \
67 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
68 /* The following interface flags should never be set */ \
69 { XBF_LIVESCAN, "LIVESCAN" }, \
70 { XBF_INCORE, "INCORE" }, \
71 { XBF_TRYLOCK, "TRYLOCK" }, \
72 { XBF_UNMAPPED, "UNMAPPED" }
73
74 /*
75 * Internal state flags.
76 */
77 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */
78
79 struct xfs_buf_cache {
80 struct rhashtable bc_hash;
81 };
82
83 int xfs_buf_cache_init(struct xfs_buf_cache *bch);
84 void xfs_buf_cache_destroy(struct xfs_buf_cache *bch);
85
86 /*
87 * The xfs_buftarg contains 2 notions of "sector size" -
88 *
89 * 1) The metadata sector size, which is the minimum unit and
90 * alignment of IO which will be performed by metadata operations.
91 * 2) The device logical sector size
92 *
93 * The first is specified at mkfs time, and is stored on-disk in the
94 * superblock's sb_sectsize.
95 *
96 * The latter is derived from the underlying device, and controls direct IO
97 * alignment constraints.
98 */
99 struct xfs_buftarg {
100 dev_t bt_dev;
101 struct file *bt_bdev_file;
102 struct block_device *bt_bdev;
103 struct dax_device *bt_daxdev;
104 struct file *bt_file;
105 u64 bt_dax_part_off;
106 struct xfs_mount *bt_mount;
107 unsigned int bt_meta_sectorsize;
108 size_t bt_meta_sectormask;
109 size_t bt_logical_sectorsize;
110 size_t bt_logical_sectormask;
111
112 /* LRU control structures */
113 struct shrinker *bt_shrinker;
114 struct list_lru bt_lru;
115
116 struct percpu_counter bt_readahead_count;
117 struct ratelimit_state bt_ioerror_rl;
118
119 /* Atomic write unit values */
120 unsigned int bt_bdev_awu_min;
121 unsigned int bt_bdev_awu_max;
122
123 /* built-in cache, if we're not using the perag one */
124 struct xfs_buf_cache bt_cache[];
125 };
126
127 #define XB_PAGES 2
128
129 struct xfs_buf_map {
130 xfs_daddr_t bm_bn; /* block number for I/O */
131 int bm_len; /* size of I/O */
132 unsigned int bm_flags;
133 };
134
135 /*
136 * Online fsck is scanning the buffer cache for live buffers. Do not warn
137 * about length mismatches during lookups and do not return stale buffers.
138 */
139 #define XBM_LIVESCAN (1U << 0)
140
141 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
142 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
143
144 struct xfs_buf_ops {
145 char *name;
146 union {
147 __be32 magic[2]; /* v4 and v5 on disk magic values */
148 __be16 magic16[2]; /* v4 and v5 on disk magic values */
149 };
150 void (*verify_read)(struct xfs_buf *);
151 void (*verify_write)(struct xfs_buf *);
152 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
153 };
154
155 struct xfs_buf {
156 /*
157 * first cacheline holds all the fields needed for an uncontended cache
158 * hit to be fully processed. The semaphore straddles the cacheline
159 * boundary, but the counter and lock sits on the first cacheline,
160 * which is the only bit that is touched if we hit the semaphore
161 * fast-path on locking.
162 */
163 struct rhash_head b_rhash_head; /* pag buffer hash node */
164
165 xfs_daddr_t b_rhash_key; /* buffer cache index */
166 int b_length; /* size of buffer in BBs */
167 unsigned int b_hold; /* reference count */
168 atomic_t b_lru_ref; /* lru reclaim ref count */
169 xfs_buf_flags_t b_flags; /* status flags */
170 struct semaphore b_sema; /* semaphore for lockables */
171
172 /*
173 * concurrent access to b_lru and b_lru_flags are protected by
174 * bt_lru_lock and not by b_sema
175 */
176 struct list_head b_lru; /* lru list */
177 spinlock_t b_lock; /* internal state lock */
178 unsigned int b_state; /* internal state flags */
179 wait_queue_head_t b_waiters; /* unpin waiters */
180 struct list_head b_list;
181 struct xfs_perag *b_pag;
182 struct xfs_mount *b_mount;
183 struct xfs_buftarg *b_target; /* buffer target (device) */
184 void *b_addr; /* virtual address of buffer */
185 struct work_struct b_ioend_work;
186 struct completion b_iowait; /* queue for I/O waiters */
187 struct xfs_buf_log_item *b_log_item;
188 struct list_head b_li_list; /* Log items list head */
189 struct xfs_trans *b_transp;
190 struct page **b_pages; /* array of page pointers */
191 struct page *b_page_array[XB_PAGES]; /* inline pages */
192 struct xfs_buf_map *b_maps; /* compound buffer map */
193 struct xfs_buf_map __b_map; /* inline compound buffer map */
194 int b_map_count;
195 atomic_t b_pin_count; /* pin count */
196 unsigned int b_page_count; /* size of page array */
197 unsigned int b_offset; /* page offset of b_addr,
198 only for _XBF_KMEM buffers */
199 int b_error; /* error code on I/O */
200 void (*b_iodone)(struct xfs_buf *bp);
201
202 /*
203 * async write failure retry count. Initialised to zero on the first
204 * failure, then when it exceeds the maximum configured without a
205 * success the write is considered to be failed permanently and the
206 * iodone handler will take appropriate action.
207 *
208 * For retry timeouts, we record the jiffy of the first failure. This
209 * means that we can change the retry timeout for buffers already under
210 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
211 *
212 * last_error is used to ensure that we are getting repeated errors, not
213 * different errors. e.g. a block device might change ENOSPC to EIO when
214 * a failure timeout occurs, so we want to re-initialise the error
215 * retry behaviour appropriately when that happens.
216 */
217 int b_retries;
218 unsigned long b_first_retry_time; /* in jiffies */
219 int b_last_error;
220
221 const struct xfs_buf_ops *b_ops;
222 struct rcu_head b_rcu;
223 };
224
225 /* Finding and Reading Buffers */
226 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
227 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
228 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
229 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
230 const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
231 void xfs_buf_readahead_map(struct xfs_buftarg *target,
232 struct xfs_buf_map *map, int nmaps,
233 const struct xfs_buf_ops *ops);
234
235 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)236 xfs_buf_incore(
237 struct xfs_buftarg *target,
238 xfs_daddr_t blkno,
239 size_t numblks,
240 xfs_buf_flags_t flags,
241 struct xfs_buf **bpp)
242 {
243 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
244
245 return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
246 }
247
248 static inline int
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp)249 xfs_buf_get(
250 struct xfs_buftarg *target,
251 xfs_daddr_t blkno,
252 size_t numblks,
253 struct xfs_buf **bpp)
254 {
255 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
256
257 return xfs_buf_get_map(target, &map, 1, 0, bpp);
258 }
259
260 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)261 xfs_buf_read(
262 struct xfs_buftarg *target,
263 xfs_daddr_t blkno,
264 size_t numblks,
265 xfs_buf_flags_t flags,
266 struct xfs_buf **bpp,
267 const struct xfs_buf_ops *ops)
268 {
269 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
270
271 return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
272 __builtin_return_address(0));
273 }
274
275 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)276 xfs_buf_readahead(
277 struct xfs_buftarg *target,
278 xfs_daddr_t blkno,
279 size_t numblks,
280 const struct xfs_buf_ops *ops)
281 {
282 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
283 return xfs_buf_readahead_map(target, &map, 1, ops);
284 }
285
286 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
287 xfs_buf_flags_t flags, struct xfs_buf **bpp);
288 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
289 size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
290 const struct xfs_buf_ops *ops);
291 int _xfs_buf_read(struct xfs_buf *bp);
292 void xfs_buf_hold(struct xfs_buf *bp);
293
294 /* Releasing Buffers */
295 extern void xfs_buf_rele(struct xfs_buf *);
296
297 /* Locking and Unlocking Buffers */
298 extern int xfs_buf_trylock(struct xfs_buf *);
299 extern void xfs_buf_lock(struct xfs_buf *);
300 extern void xfs_buf_unlock(struct xfs_buf *);
301 #define xfs_buf_islocked(bp) \
302 ((bp)->b_sema.count <= 0)
303
xfs_buf_relse(struct xfs_buf * bp)304 static inline void xfs_buf_relse(struct xfs_buf *bp)
305 {
306 xfs_buf_unlock(bp);
307 xfs_buf_rele(bp);
308 }
309
310 /* Buffer Read and Write Routines */
311 extern int xfs_bwrite(struct xfs_buf *bp);
312
313 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
314 xfs_failaddr_t failaddr);
315 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
316 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
317 void xfs_buf_ioend_fail(struct xfs_buf *);
318 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
319 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
320 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
321
322 /* Buffer Utility Routines */
323 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
324 extern void xfs_buf_stale(struct xfs_buf *bp);
325
326 /* Delayed Write Buffer Routines */
327 extern void xfs_buf_delwri_cancel(struct list_head *);
328 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
329 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
330 extern int xfs_buf_delwri_submit(struct list_head *);
331 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
332 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
333
xfs_buf_daddr(struct xfs_buf * bp)334 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
335 {
336 return bp->b_maps[0].bm_bn;
337 }
338
339 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
340
341 /*
342 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
343 * up with a reference count of 0 so it will be tossed from the cache when
344 * released.
345 */
xfs_buf_oneshot(struct xfs_buf * bp)346 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
347 {
348 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
349 return;
350 atomic_set(&bp->b_lru_ref, 0);
351 }
352
xfs_buf_ispinned(struct xfs_buf * bp)353 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
354 {
355 return atomic_read(&bp->b_pin_count);
356 }
357
358 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)359 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
360 {
361 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
362 cksum_offset);
363 }
364
365 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)366 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
367 {
368 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
369 cksum_offset);
370 }
371
372 /*
373 * Handling of buftargs.
374 */
375 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
376 struct file *bdev_file);
377 extern void xfs_free_buftarg(struct xfs_buftarg *);
378 extern void xfs_buftarg_wait(struct xfs_buftarg *);
379 extern void xfs_buftarg_drain(struct xfs_buftarg *);
380 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
381
382 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev)
383 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
384
385 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
386 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
387 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
388
389 /* for xfs_buf_mem.c only: */
390 int xfs_init_buftarg(struct xfs_buftarg *btp, size_t logical_sectorsize,
391 const char *descr);
392 void xfs_destroy_buftarg(struct xfs_buftarg *btp);
393
394 #endif /* __XFS_BUF_H__ */
395