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_NO_IOACCT (1u << 3) /* bypass I/O accounting (non-LRU bufs) */
31 #define XBF_ASYNC (1u << 4) /* initiator will not wait for completion */
32 #define XBF_DONE (1u << 5) /* all pages in the buffer uptodate */
33 #define XBF_STALE (1u << 6) /* buffer has been staled, do not find it */
34 #define XBF_WRITE_FAIL (1u << 7) /* async writes have failed on this buffer */
35
36 /* buffer type flags for write callbacks */
37 #define _XBF_INODES (1u << 16)/* inode buffer */
38 #define _XBF_DQUOTS (1u << 17)/* dquot buffer */
39 #define _XBF_LOGRECOVERY (1u << 18)/* log recovery buffer */
40
41 /* flags used only internally */
42 #define _XBF_PAGES (1u << 20)/* backed by refcounted pages */
43 #define _XBF_KMEM (1u << 21)/* backed by heap memory */
44 #define _XBF_DELWRI_Q (1u << 22)/* buffer on a delwri queue */
45
46 /* flags used only as arguments to access routines */
47 /*
48 * Online fsck is scanning the buffer cache for live buffers. Do not warn
49 * about length mismatches during lookups and do not return stale buffers.
50 */
51 #define XBF_LIVESCAN (1u << 28)
52 #define XBF_INCORE (1u << 29)/* lookup only, return if found in cache */
53 #define XBF_TRYLOCK (1u << 30)/* lock requested, but do not wait */
54 #define XBF_UNMAPPED (1u << 31)/* do not map the buffer */
55
56
57 typedef unsigned int xfs_buf_flags_t;
58
59 #define XFS_BUF_FLAGS \
60 { XBF_READ, "READ" }, \
61 { XBF_WRITE, "WRITE" }, \
62 { XBF_READ_AHEAD, "READ_AHEAD" }, \
63 { XBF_NO_IOACCT, "NO_IOACCT" }, \
64 { XBF_ASYNC, "ASYNC" }, \
65 { XBF_DONE, "DONE" }, \
66 { XBF_STALE, "STALE" }, \
67 { XBF_WRITE_FAIL, "WRITE_FAIL" }, \
68 { _XBF_INODES, "INODES" }, \
69 { _XBF_DQUOTS, "DQUOTS" }, \
70 { _XBF_LOGRECOVERY, "LOG_RECOVERY" }, \
71 { _XBF_PAGES, "PAGES" }, \
72 { _XBF_KMEM, "KMEM" }, \
73 { _XBF_DELWRI_Q, "DELWRI_Q" }, \
74 /* The following interface flags should never be set */ \
75 { XBF_LIVESCAN, "LIVESCAN" }, \
76 { XBF_INCORE, "INCORE" }, \
77 { XBF_TRYLOCK, "TRYLOCK" }, \
78 { XBF_UNMAPPED, "UNMAPPED" }
79
80 /*
81 * Internal state flags.
82 */
83 #define XFS_BSTATE_DISPOSE (1 << 0) /* buffer being discarded */
84 #define XFS_BSTATE_IN_FLIGHT (1 << 1) /* I/O in flight */
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 typedef struct xfs_buftarg {
100 dev_t bt_dev;
101 struct block_device *bt_bdev;
102 struct dax_device *bt_daxdev;
103 u64 bt_dax_part_off;
104 struct xfs_mount *bt_mount;
105 unsigned int bt_meta_sectorsize;
106 size_t bt_meta_sectormask;
107 size_t bt_logical_sectorsize;
108 size_t bt_logical_sectormask;
109
110 /* LRU control structures */
111 struct shrinker bt_shrinker;
112 struct list_lru bt_lru;
113
114 struct percpu_counter bt_io_count;
115 struct ratelimit_state bt_ioerror_rl;
116 } xfs_buftarg_t;
117
118 #define XB_PAGES 2
119
120 struct xfs_buf_map {
121 xfs_daddr_t bm_bn; /* block number for I/O */
122 int bm_len; /* size of I/O */
123 unsigned int bm_flags;
124 };
125
126 /*
127 * Online fsck is scanning the buffer cache for live buffers. Do not warn
128 * about length mismatches during lookups and do not return stale buffers.
129 */
130 #define XBM_LIVESCAN (1U << 0)
131
132 #define DEFINE_SINGLE_BUF_MAP(map, blkno, numblk) \
133 struct xfs_buf_map (map) = { .bm_bn = (blkno), .bm_len = (numblk) };
134
135 struct xfs_buf_ops {
136 char *name;
137 union {
138 __be32 magic[2]; /* v4 and v5 on disk magic values */
139 __be16 magic16[2]; /* v4 and v5 on disk magic values */
140 };
141 void (*verify_read)(struct xfs_buf *);
142 void (*verify_write)(struct xfs_buf *);
143 xfs_failaddr_t (*verify_struct)(struct xfs_buf *bp);
144 };
145
146 struct xfs_buf {
147 /*
148 * first cacheline holds all the fields needed for an uncontended cache
149 * hit to be fully processed. The semaphore straddles the cacheline
150 * boundary, but the counter and lock sits on the first cacheline,
151 * which is the only bit that is touched if we hit the semaphore
152 * fast-path on locking.
153 */
154 struct rhash_head b_rhash_head; /* pag buffer hash node */
155
156 xfs_daddr_t b_rhash_key; /* buffer cache index */
157 int b_length; /* size of buffer in BBs */
158 atomic_t b_hold; /* reference count */
159 atomic_t b_lru_ref; /* lru reclaim ref count */
160 xfs_buf_flags_t b_flags; /* status flags */
161 struct semaphore b_sema; /* semaphore for lockables */
162
163 /*
164 * concurrent access to b_lru and b_lru_flags are protected by
165 * bt_lru_lock and not by b_sema
166 */
167 struct list_head b_lru; /* lru list */
168 spinlock_t b_lock; /* internal state lock */
169 unsigned int b_state; /* internal state flags */
170 int b_io_error; /* internal IO error state */
171 wait_queue_head_t b_waiters; /* unpin waiters */
172 struct list_head b_list;
173 struct xfs_perag *b_pag; /* contains rbtree root */
174 struct xfs_mount *b_mount;
175 struct xfs_buftarg *b_target; /* buffer target (device) */
176 void *b_addr; /* virtual address of buffer */
177 struct work_struct b_ioend_work;
178 struct completion b_iowait; /* queue for I/O waiters */
179 struct xfs_buf_log_item *b_log_item;
180 struct list_head b_li_list; /* Log items list head */
181 struct xfs_trans *b_transp;
182 struct page **b_pages; /* array of page pointers */
183 struct page *b_page_array[XB_PAGES]; /* inline pages */
184 struct xfs_buf_map *b_maps; /* compound buffer map */
185 struct xfs_buf_map __b_map; /* inline compound buffer map */
186 int b_map_count;
187 atomic_t b_pin_count; /* pin count */
188 atomic_t b_io_remaining; /* #outstanding I/O requests */
189 unsigned int b_page_count; /* size of page array */
190 unsigned int b_offset; /* page offset of b_addr,
191 only for _XBF_KMEM buffers */
192 int b_error; /* error code on I/O */
193
194 /*
195 * async write failure retry count. Initialised to zero on the first
196 * failure, then when it exceeds the maximum configured without a
197 * success the write is considered to be failed permanently and the
198 * iodone handler will take appropriate action.
199 *
200 * For retry timeouts, we record the jiffie of the first failure. This
201 * means that we can change the retry timeout for buffers already under
202 * I/O and thus avoid getting stuck in a retry loop with a long timeout.
203 *
204 * last_error is used to ensure that we are getting repeated errors, not
205 * different errors. e.g. a block device might change ENOSPC to EIO when
206 * a failure timeout occurs, so we want to re-initialise the error
207 * retry behaviour appropriately when that happens.
208 */
209 int b_retries;
210 unsigned long b_first_retry_time; /* in jiffies */
211 int b_last_error;
212
213 const struct xfs_buf_ops *b_ops;
214 struct rcu_head b_rcu;
215 };
216
217 /* Finding and Reading Buffers */
218 int xfs_buf_get_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
219 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp);
220 int xfs_buf_read_map(struct xfs_buftarg *target, struct xfs_buf_map *map,
221 int nmaps, xfs_buf_flags_t flags, struct xfs_buf **bpp,
222 const struct xfs_buf_ops *ops, xfs_failaddr_t fa);
223 void xfs_buf_readahead_map(struct xfs_buftarg *target,
224 struct xfs_buf_map *map, int nmaps,
225 const struct xfs_buf_ops *ops);
226
227 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)228 xfs_buf_incore(
229 struct xfs_buftarg *target,
230 xfs_daddr_t blkno,
231 size_t numblks,
232 xfs_buf_flags_t flags,
233 struct xfs_buf **bpp)
234 {
235 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
236
237 return xfs_buf_get_map(target, &map, 1, XBF_INCORE | flags, bpp);
238 }
239
240 static inline int
xfs_buf_get(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,struct xfs_buf ** bpp)241 xfs_buf_get(
242 struct xfs_buftarg *target,
243 xfs_daddr_t blkno,
244 size_t numblks,
245 struct xfs_buf **bpp)
246 {
247 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
248
249 return xfs_buf_get_map(target, &map, 1, 0, bpp);
250 }
251
252 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)253 xfs_buf_read(
254 struct xfs_buftarg *target,
255 xfs_daddr_t blkno,
256 size_t numblks,
257 xfs_buf_flags_t flags,
258 struct xfs_buf **bpp,
259 const struct xfs_buf_ops *ops)
260 {
261 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
262
263 return xfs_buf_read_map(target, &map, 1, flags, bpp, ops,
264 __builtin_return_address(0));
265 }
266
267 static inline void
xfs_buf_readahead(struct xfs_buftarg * target,xfs_daddr_t blkno,size_t numblks,const struct xfs_buf_ops * ops)268 xfs_buf_readahead(
269 struct xfs_buftarg *target,
270 xfs_daddr_t blkno,
271 size_t numblks,
272 const struct xfs_buf_ops *ops)
273 {
274 DEFINE_SINGLE_BUF_MAP(map, blkno, numblks);
275 return xfs_buf_readahead_map(target, &map, 1, ops);
276 }
277
278 int xfs_buf_get_uncached(struct xfs_buftarg *target, size_t numblks,
279 xfs_buf_flags_t flags, struct xfs_buf **bpp);
280 int xfs_buf_read_uncached(struct xfs_buftarg *target, xfs_daddr_t daddr,
281 size_t numblks, xfs_buf_flags_t flags, struct xfs_buf **bpp,
282 const struct xfs_buf_ops *ops);
283 int _xfs_buf_read(struct xfs_buf *bp, xfs_buf_flags_t flags);
284 void xfs_buf_hold(struct xfs_buf *bp);
285
286 /* Releasing Buffers */
287 extern void xfs_buf_rele(struct xfs_buf *);
288
289 /* Locking and Unlocking Buffers */
290 extern int xfs_buf_trylock(struct xfs_buf *);
291 extern void xfs_buf_lock(struct xfs_buf *);
292 extern void xfs_buf_unlock(struct xfs_buf *);
293 #define xfs_buf_islocked(bp) \
294 ((bp)->b_sema.count <= 0)
295
xfs_buf_relse(struct xfs_buf * bp)296 static inline void xfs_buf_relse(struct xfs_buf *bp)
297 {
298 xfs_buf_unlock(bp);
299 xfs_buf_rele(bp);
300 }
301
302 /* Buffer Read and Write Routines */
303 extern int xfs_bwrite(struct xfs_buf *bp);
304
305 extern void __xfs_buf_ioerror(struct xfs_buf *bp, int error,
306 xfs_failaddr_t failaddr);
307 #define xfs_buf_ioerror(bp, err) __xfs_buf_ioerror((bp), (err), __this_address)
308 extern void xfs_buf_ioerror_alert(struct xfs_buf *bp, xfs_failaddr_t fa);
309 void xfs_buf_ioend_fail(struct xfs_buf *);
310 void xfs_buf_zero(struct xfs_buf *bp, size_t boff, size_t bsize);
311 void __xfs_buf_mark_corrupt(struct xfs_buf *bp, xfs_failaddr_t fa);
312 #define xfs_buf_mark_corrupt(bp) __xfs_buf_mark_corrupt((bp), __this_address)
313
314 /* Buffer Utility Routines */
315 extern void *xfs_buf_offset(struct xfs_buf *, size_t);
316 extern void xfs_buf_stale(struct xfs_buf *bp);
317
318 /* Delayed Write Buffer Routines */
319 extern void xfs_buf_delwri_cancel(struct list_head *);
320 extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *);
321 void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl);
322 extern int xfs_buf_delwri_submit(struct list_head *);
323 extern int xfs_buf_delwri_submit_nowait(struct list_head *);
324 extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
325
xfs_buf_daddr(struct xfs_buf * bp)326 static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)
327 {
328 return bp->b_maps[0].bm_bn;
329 }
330
331 void xfs_buf_set_ref(struct xfs_buf *bp, int lru_ref);
332
333 /*
334 * If the buffer is already on the LRU, do nothing. Otherwise set the buffer
335 * up with a reference count of 0 so it will be tossed from the cache when
336 * released.
337 */
xfs_buf_oneshot(struct xfs_buf * bp)338 static inline void xfs_buf_oneshot(struct xfs_buf *bp)
339 {
340 if (!list_empty(&bp->b_lru) || atomic_read(&bp->b_lru_ref) > 1)
341 return;
342 atomic_set(&bp->b_lru_ref, 0);
343 }
344
xfs_buf_ispinned(struct xfs_buf * bp)345 static inline int xfs_buf_ispinned(struct xfs_buf *bp)
346 {
347 return atomic_read(&bp->b_pin_count);
348 }
349
350 static inline int
xfs_buf_verify_cksum(struct xfs_buf * bp,unsigned long cksum_offset)351 xfs_buf_verify_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
352 {
353 return xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
354 cksum_offset);
355 }
356
357 static inline void
xfs_buf_update_cksum(struct xfs_buf * bp,unsigned long cksum_offset)358 xfs_buf_update_cksum(struct xfs_buf *bp, unsigned long cksum_offset)
359 {
360 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
361 cksum_offset);
362 }
363
364 /*
365 * Handling of buftargs.
366 */
367 struct xfs_buftarg *xfs_alloc_buftarg(struct xfs_mount *mp,
368 struct block_device *bdev);
369 extern void xfs_free_buftarg(struct xfs_buftarg *);
370 extern void xfs_buftarg_wait(struct xfs_buftarg *);
371 extern void xfs_buftarg_drain(struct xfs_buftarg *);
372 extern int xfs_setsize_buftarg(struct xfs_buftarg *, unsigned int);
373
374 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev)
375 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
376
377 int xfs_buf_reverify(struct xfs_buf *bp, const struct xfs_buf_ops *ops);
378 bool xfs_verify_magic(struct xfs_buf *bp, __be32 dmagic);
379 bool xfs_verify_magic16(struct xfs_buf *bp, __be16 dmagic);
380
381 #endif /* __XFS_BUF_H__ */
382