2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include <linux/list.h>
22 #include <linux/types.h>
23 #include <linux/spinlock.h>
24 #include <asm/system.h>
27 #include <linux/buffer_head.h>
28 #include <linux/uio.h>
34 #define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))
36 #define xfs_buf_ctob(pp) ((pp) * PAGE_CACHE_SIZE)
37 #define xfs_buf_btoc(dd) (((dd) + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT)
38 #define xfs_buf_btoct(dd) ((dd) >> PAGE_CACHE_SHIFT)
39 #define xfs_buf_poff(aa) ((aa) & ~PAGE_CACHE_MASK)
42 XBRW_READ
= 1, /* transfer into target memory */
43 XBRW_WRITE
= 2, /* transfer from target memory */
44 XBRW_ZERO
= 3, /* Zero target memory */
48 XBF_READ
= (1 << 0), /* buffer intended for reading from device */
49 XBF_WRITE
= (1 << 1), /* buffer intended for writing to device */
50 XBF_MAPPED
= (1 << 2), /* buffer mapped (b_addr valid) */
51 XBF_ASYNC
= (1 << 4), /* initiator will not wait for completion */
52 XBF_DONE
= (1 << 5), /* all pages in the buffer uptodate */
53 XBF_DELWRI
= (1 << 6), /* buffer has dirty pages */
54 XBF_STALE
= (1 << 7), /* buffer has been staled, do not find it */
55 XBF_FS_MANAGED
= (1 << 8), /* filesystem controls freeing memory */
56 XBF_ORDERED
= (1 << 11), /* use ordered writes */
57 XBF_READ_AHEAD
= (1 << 12), /* asynchronous read-ahead */
59 /* flags used only as arguments to access routines */
60 XBF_LOCK
= (1 << 14), /* lock requested */
61 XBF_TRYLOCK
= (1 << 15), /* lock requested, but do not wait */
62 XBF_DONT_BLOCK
= (1 << 16), /* do not block in current thread */
64 /* flags used only internally */
65 _XBF_PAGE_CACHE
= (1 << 17),/* backed by pagecache */
66 _XBF_PAGES
= (1 << 18), /* backed by refcounted pages */
67 _XBF_RUN_QUEUES
= (1 << 19),/* run block device task queue */
68 _XBF_DELWRI_Q
= (1 << 21), /* buffer on delwri queue */
71 * Special flag for supporting metadata blocks smaller than a FSB.
73 * In this case we can have multiple xfs_buf_t on a single page and
74 * need to lock out concurrent xfs_buf_t readers as they only
75 * serialise access to the buffer.
77 * If the FSB size >= PAGE_CACHE_SIZE case, we have no serialisation
78 * between reads of the page. Hence we can have one thread read the
79 * page and modify it, but then race with another thread that thinks
80 * the page is not up-to-date and hence reads it again.
82 * The result is that the first modifcation to the page is lost.
83 * This sort of AGF/AGI reading race can happen when unlinking inodes
84 * that require truncation and results in the AGI unlinked list
85 * modifications being lost.
87 _XBF_PAGE_LOCKED
= (1 << 22),
93 } xfs_buftarg_flags_t
;
95 typedef struct xfs_bufhash
{
96 struct list_head bh_list
;
100 typedef struct xfs_buftarg
{
102 struct block_device
*bt_bdev
;
103 struct address_space
*bt_mapping
;
104 unsigned int bt_bsize
;
105 unsigned int bt_sshift
;
108 /* per device buffer hash table */
111 xfs_bufhash_t
*bt_hash
;
113 /* per device delwri queue */
114 struct task_struct
*bt_task
;
115 struct list_head bt_list
;
116 struct list_head bt_delwrite_queue
;
117 spinlock_t bt_delwrite_lock
;
118 unsigned long bt_flags
;
122 * xfs_buf_t: Buffer structure for pagecache-based buffers
124 * This buffer structure is used by the pagecache buffer management routines
125 * to refer to an assembly of pages forming a logical buffer.
127 * The buffer structure is used on a temporary basis only, and discarded when
128 * released. The real data storage is recorded in the pagecache. Buffers are
129 * hashed to the block device on which the file system resides.
133 typedef void (*xfs_buf_iodone_t
)(struct xfs_buf
*);
134 typedef void (*xfs_buf_relse_t
)(struct xfs_buf
*);
135 typedef int (*xfs_buf_bdstrat_t
)(struct xfs_buf
*);
139 typedef struct xfs_buf
{
140 struct semaphore b_sema
; /* semaphore for lockables */
141 unsigned long b_queuetime
; /* time buffer was queued */
142 atomic_t b_pin_count
; /* pin count */
143 wait_queue_head_t b_waiters
; /* unpin waiters */
144 struct list_head b_list
;
145 xfs_buf_flags_t b_flags
; /* status flags */
146 struct list_head b_hash_list
; /* hash table list */
147 xfs_bufhash_t
*b_hash
; /* hash table list start */
148 xfs_buftarg_t
*b_target
; /* buffer target (device) */
149 atomic_t b_hold
; /* reference count */
150 xfs_daddr_t b_bn
; /* block number for I/O */
151 xfs_off_t b_file_offset
; /* offset in file */
152 size_t b_buffer_length
;/* size of buffer in bytes */
153 size_t b_count_desired
;/* desired transfer size */
154 void *b_addr
; /* virtual address of buffer */
155 struct work_struct b_iodone_work
;
156 atomic_t b_io_remaining
; /* #outstanding I/O requests */
157 xfs_buf_iodone_t b_iodone
; /* I/O completion function */
158 xfs_buf_relse_t b_relse
; /* releasing function */
159 xfs_buf_bdstrat_t b_strat
; /* pre-write function */
160 struct semaphore b_iodonesema
; /* Semaphore for I/O waiters */
164 unsigned short b_error
; /* error code on I/O */
165 unsigned int b_page_count
; /* size of page array */
166 unsigned int b_offset
; /* page offset in first page */
167 struct page
**b_pages
; /* array of page pointers */
168 struct page
*b_page_array
[XB_PAGES
]; /* inline pages */
169 #ifdef XFS_BUF_LOCK_TRACKING
175 /* Finding and Reading Buffers */
176 extern xfs_buf_t
*_xfs_buf_find(xfs_buftarg_t
*, xfs_off_t
, size_t,
177 xfs_buf_flags_t
, xfs_buf_t
*);
178 #define xfs_incore(buftarg,blkno,len,lockit) \
179 _xfs_buf_find(buftarg, blkno ,len, lockit, NULL)
181 extern xfs_buf_t
*xfs_buf_get_flags(xfs_buftarg_t
*, xfs_off_t
, size_t,
183 #define xfs_buf_get(target, blkno, len, flags) \
184 xfs_buf_get_flags((target), (blkno), (len), XBF_LOCK | XBF_MAPPED)
186 extern xfs_buf_t
*xfs_buf_read_flags(xfs_buftarg_t
*, xfs_off_t
, size_t,
188 #define xfs_buf_read(target, blkno, len, flags) \
189 xfs_buf_read_flags((target), (blkno), (len), XBF_LOCK | XBF_MAPPED)
191 extern xfs_buf_t
*xfs_buf_get_empty(size_t, xfs_buftarg_t
*);
192 extern xfs_buf_t
*xfs_buf_get_noaddr(size_t, xfs_buftarg_t
*);
193 extern int xfs_buf_associate_memory(xfs_buf_t
*, void *, size_t);
194 extern void xfs_buf_hold(xfs_buf_t
*);
195 extern void xfs_buf_readahead(xfs_buftarg_t
*, xfs_off_t
, size_t,
198 /* Releasing Buffers */
199 extern void xfs_buf_free(xfs_buf_t
*);
200 extern void xfs_buf_rele(xfs_buf_t
*);
202 /* Locking and Unlocking Buffers */
203 extern int xfs_buf_cond_lock(xfs_buf_t
*);
204 extern int xfs_buf_lock_value(xfs_buf_t
*);
205 extern void xfs_buf_lock(xfs_buf_t
*);
206 extern void xfs_buf_unlock(xfs_buf_t
*);
208 /* Buffer Read and Write Routines */
209 extern void xfs_buf_ioend(xfs_buf_t
*, int);
210 extern void xfs_buf_ioerror(xfs_buf_t
*, int);
211 extern int xfs_buf_iostart(xfs_buf_t
*, xfs_buf_flags_t
);
212 extern int xfs_buf_iorequest(xfs_buf_t
*);
213 extern int xfs_buf_iowait(xfs_buf_t
*);
214 extern void xfs_buf_iomove(xfs_buf_t
*, size_t, size_t, xfs_caddr_t
,
217 static inline int xfs_buf_iostrategy(xfs_buf_t
*bp
)
219 return bp
->b_strat
? bp
->b_strat(bp
) : xfs_buf_iorequest(bp
);
222 static inline int xfs_buf_geterror(xfs_buf_t
*bp
)
224 return bp
? bp
->b_error
: ENOMEM
;
227 /* Buffer Utility Routines */
228 extern xfs_caddr_t
xfs_buf_offset(xfs_buf_t
*, size_t);
230 /* Pinning Buffer Storage in Memory */
231 extern void xfs_buf_pin(xfs_buf_t
*);
232 extern void xfs_buf_unpin(xfs_buf_t
*);
233 extern int xfs_buf_ispin(xfs_buf_t
*);
235 /* Delayed Write Buffer Routines */
236 extern void xfs_buf_delwri_dequeue(xfs_buf_t
*);
238 /* Buffer Daemon Setup Routines */
239 extern int xfs_buf_init(void);
240 extern void xfs_buf_terminate(void);
243 extern ktrace_t
*xfs_buf_trace_buf
;
244 extern void xfs_buf_trace(xfs_buf_t
*, char *, void *, void *);
246 #define xfs_buf_trace(bp,id,ptr,ra) do { } while (0)
249 #define xfs_buf_target_name(target) \
250 ({ char __b[BDEVNAME_SIZE]; bdevname((target)->bt_bdev, __b); __b; })
253 #define XFS_B_ASYNC XBF_ASYNC
254 #define XFS_B_DELWRI XBF_DELWRI
255 #define XFS_B_READ XBF_READ
256 #define XFS_B_WRITE XBF_WRITE
257 #define XFS_B_STALE XBF_STALE
259 #define XFS_BUF_TRYLOCK XBF_TRYLOCK
260 #define XFS_INCORE_TRYLOCK XBF_TRYLOCK
261 #define XFS_BUF_LOCK XBF_LOCK
262 #define XFS_BUF_MAPPED XBF_MAPPED
264 #define BUF_BUSY XBF_DONT_BLOCK
266 #define XFS_BUF_BFLAGS(bp) ((bp)->b_flags)
267 #define XFS_BUF_ZEROFLAGS(bp) ((bp)->b_flags &= \
268 ~(XBF_READ|XBF_WRITE|XBF_ASYNC|XBF_DELWRI|XBF_ORDERED))
270 #define XFS_BUF_STALE(bp) ((bp)->b_flags |= XFS_B_STALE)
271 #define XFS_BUF_UNSTALE(bp) ((bp)->b_flags &= ~XFS_B_STALE)
272 #define XFS_BUF_ISSTALE(bp) ((bp)->b_flags & XFS_B_STALE)
273 #define XFS_BUF_SUPER_STALE(bp) do { \
275 xfs_buf_delwri_dequeue(bp); \
279 #define XFS_BUF_MANAGE XBF_FS_MANAGED
280 #define XFS_BUF_UNMANAGE(bp) ((bp)->b_flags &= ~XBF_FS_MANAGED)
282 #define XFS_BUF_DELAYWRITE(bp) ((bp)->b_flags |= XBF_DELWRI)
283 #define XFS_BUF_UNDELAYWRITE(bp) xfs_buf_delwri_dequeue(bp)
284 #define XFS_BUF_ISDELAYWRITE(bp) ((bp)->b_flags & XBF_DELWRI)
286 #define XFS_BUF_ERROR(bp,no) xfs_buf_ioerror(bp,no)
287 #define XFS_BUF_GETERROR(bp) xfs_buf_geterror(bp)
288 #define XFS_BUF_ISERROR(bp) (xfs_buf_geterror(bp) ? 1 : 0)
290 #define XFS_BUF_DONE(bp) ((bp)->b_flags |= XBF_DONE)
291 #define XFS_BUF_UNDONE(bp) ((bp)->b_flags &= ~XBF_DONE)
292 #define XFS_BUF_ISDONE(bp) ((bp)->b_flags & XBF_DONE)
294 #define XFS_BUF_BUSY(bp) do { } while (0)
295 #define XFS_BUF_UNBUSY(bp) do { } while (0)
296 #define XFS_BUF_ISBUSY(bp) (1)
298 #define XFS_BUF_ASYNC(bp) ((bp)->b_flags |= XBF_ASYNC)
299 #define XFS_BUF_UNASYNC(bp) ((bp)->b_flags &= ~XBF_ASYNC)
300 #define XFS_BUF_ISASYNC(bp) ((bp)->b_flags & XBF_ASYNC)
302 #define XFS_BUF_ORDERED(bp) ((bp)->b_flags |= XBF_ORDERED)
303 #define XFS_BUF_UNORDERED(bp) ((bp)->b_flags &= ~XBF_ORDERED)
304 #define XFS_BUF_ISORDERED(bp) ((bp)->b_flags & XBF_ORDERED)
306 #define XFS_BUF_SHUT(bp) do { } while (0)
307 #define XFS_BUF_UNSHUT(bp) do { } while (0)
308 #define XFS_BUF_ISSHUT(bp) (0)
310 #define XFS_BUF_HOLD(bp) xfs_buf_hold(bp)
311 #define XFS_BUF_READ(bp) ((bp)->b_flags |= XBF_READ)
312 #define XFS_BUF_UNREAD(bp) ((bp)->b_flags &= ~XBF_READ)
313 #define XFS_BUF_ISREAD(bp) ((bp)->b_flags & XBF_READ)
315 #define XFS_BUF_WRITE(bp) ((bp)->b_flags |= XBF_WRITE)
316 #define XFS_BUF_UNWRITE(bp) ((bp)->b_flags &= ~XBF_WRITE)
317 #define XFS_BUF_ISWRITE(bp) ((bp)->b_flags & XBF_WRITE)
319 #define XFS_BUF_IODONE_FUNC(bp) ((bp)->b_iodone)
320 #define XFS_BUF_SET_IODONE_FUNC(bp, func) ((bp)->b_iodone = (func))
321 #define XFS_BUF_CLR_IODONE_FUNC(bp) ((bp)->b_iodone = NULL)
322 #define XFS_BUF_SET_BDSTRAT_FUNC(bp, func) ((bp)->b_strat = (func))
323 #define XFS_BUF_CLR_BDSTRAT_FUNC(bp) ((bp)->b_strat = NULL)
325 #define XFS_BUF_FSPRIVATE(bp, type) ((type)(bp)->b_fspriv)
326 #define XFS_BUF_SET_FSPRIVATE(bp, val) ((bp)->b_fspriv = (void*)(val))
327 #define XFS_BUF_FSPRIVATE2(bp, type) ((type)(bp)->b_fspriv2)
328 #define XFS_BUF_SET_FSPRIVATE2(bp, val) ((bp)->b_fspriv2 = (void*)(val))
329 #define XFS_BUF_FSPRIVATE3(bp, type) ((type)(bp)->b_fspriv3)
330 #define XFS_BUF_SET_FSPRIVATE3(bp, val) ((bp)->b_fspriv3 = (void*)(val))
331 #define XFS_BUF_SET_START(bp) do { } while (0)
332 #define XFS_BUF_SET_BRELSE_FUNC(bp, func) ((bp)->b_relse = (func))
334 #define XFS_BUF_PTR(bp) (xfs_caddr_t)((bp)->b_addr)
335 #define XFS_BUF_SET_PTR(bp, val, cnt) xfs_buf_associate_memory(bp, val, cnt)
336 #define XFS_BUF_ADDR(bp) ((bp)->b_bn)
337 #define XFS_BUF_SET_ADDR(bp, bno) ((bp)->b_bn = (xfs_daddr_t)(bno))
338 #define XFS_BUF_OFFSET(bp) ((bp)->b_file_offset)
339 #define XFS_BUF_SET_OFFSET(bp, off) ((bp)->b_file_offset = (off))
340 #define XFS_BUF_COUNT(bp) ((bp)->b_count_desired)
341 #define XFS_BUF_SET_COUNT(bp, cnt) ((bp)->b_count_desired = (cnt))
342 #define XFS_BUF_SIZE(bp) ((bp)->b_buffer_length)
343 #define XFS_BUF_SET_SIZE(bp, cnt) ((bp)->b_buffer_length = (cnt))
345 #define XFS_BUF_SET_VTYPE_REF(bp, type, ref) do { } while (0)
346 #define XFS_BUF_SET_VTYPE(bp, type) do { } while (0)
347 #define XFS_BUF_SET_REF(bp, ref) do { } while (0)
349 #define XFS_BUF_ISPINNED(bp) xfs_buf_ispin(bp)
351 #define XFS_BUF_VALUSEMA(bp) xfs_buf_lock_value(bp)
352 #define XFS_BUF_CPSEMA(bp) (xfs_buf_cond_lock(bp) == 0)
353 #define XFS_BUF_VSEMA(bp) xfs_buf_unlock(bp)
354 #define XFS_BUF_PSEMA(bp,x) xfs_buf_lock(bp)
355 #define XFS_BUF_V_IODONESEMA(bp) up(&bp->b_iodonesema);
357 #define XFS_BUF_SET_TARGET(bp, target) ((bp)->b_target = (target))
358 #define XFS_BUF_TARGET(bp) ((bp)->b_target)
359 #define XFS_BUFTARG_NAME(target) xfs_buf_target_name(target)
361 static inline int xfs_bawrite(void *mp
, xfs_buf_t
*bp
)
364 bp
->b_strat
= xfs_bdstrat_cb
;
365 xfs_buf_delwri_dequeue(bp
);
366 return xfs_buf_iostart(bp
, XBF_WRITE
| XBF_ASYNC
| _XBF_RUN_QUEUES
);
369 static inline void xfs_buf_relse(xfs_buf_t
*bp
)
376 #define xfs_bpin(bp) xfs_buf_pin(bp)
377 #define xfs_bunpin(bp) xfs_buf_unpin(bp)
379 #define xfs_buftrace(id, bp) \
380 xfs_buf_trace(bp, id, NULL, (void *)__builtin_return_address(0))
382 #define xfs_biodone(bp) xfs_buf_ioend(bp, 0)
384 #define xfs_biomove(bp, off, len, data, rw) \
385 xfs_buf_iomove((bp), (off), (len), (data), \
386 ((rw) == XFS_B_WRITE) ? XBRW_WRITE : XBRW_READ)
388 #define xfs_biozero(bp, off, len) \
389 xfs_buf_iomove((bp), (off), (len), NULL, XBRW_ZERO)
392 static inline int XFS_bwrite(xfs_buf_t
*bp
)
394 int iowait
= (bp
->b_flags
& XBF_ASYNC
) == 0;
398 bp
->b_flags
|= _XBF_RUN_QUEUES
;
400 xfs_buf_delwri_dequeue(bp
);
401 xfs_buf_iostrategy(bp
);
403 error
= xfs_buf_iowait(bp
);
410 * No error can be returned from xfs_buf_iostart for delwri
411 * buffers as they are queued and no I/O is issued.
413 static inline void xfs_bdwrite(void *mp
, xfs_buf_t
*bp
)
415 bp
->b_strat
= xfs_bdstrat_cb
;
417 (void)xfs_buf_iostart(bp
, XBF_DELWRI
| XBF_ASYNC
);
420 #define XFS_bdstrat(bp) xfs_buf_iorequest(bp)
422 #define xfs_iowait(bp) xfs_buf_iowait(bp)
424 #define xfs_baread(target, rablkno, ralen) \
425 xfs_buf_readahead((target), (rablkno), (ralen), XBF_DONT_BLOCK)
429 * Handling of buftargs.
431 extern xfs_buftarg_t
*xfs_alloc_buftarg(struct block_device
*, int);
432 extern void xfs_free_buftarg(xfs_buftarg_t
*, int);
433 extern void xfs_wait_buftarg(xfs_buftarg_t
*);
434 extern int xfs_setsize_buftarg(xfs_buftarg_t
*, unsigned int, unsigned int);
435 extern int xfs_flush_buftarg(xfs_buftarg_t
*, int);
436 #ifdef CONFIG_KDB_MODULES
437 extern struct list_head
*xfs_get_buftarg_list(void);
440 #define xfs_getsize_buftarg(buftarg) block_size((buftarg)->bt_bdev)
441 #define xfs_readonly_buftarg(buftarg) bdev_read_only((buftarg)->bt_bdev)
443 #define xfs_binval(buftarg) xfs_flush_buftarg(buftarg, 1)
444 #define XFS_bflush(buftarg) xfs_flush_buftarg(buftarg, 1)
446 #endif /* __XFS_BUF_H__ */