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
25 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_error.h"
34 #include "xfs_vnodeops.h"
35 #include "xfs_da_btree.h"
36 #include "xfs_ioctl.h"
37 #include "xfs_trace.h"
39 #include <linux/dcache.h>
41 static const struct vm_operations_struct xfs_file_vm_ops
;
46 * xfs_iozero clears the specified range of buffer supplied,
47 * and marks all the affected blocks as valid and modified. If
48 * an affected block is not allocated, it will be allocated. If
49 * an affected block is not completely overwritten, and is not
50 * valid before the operation, it will be read from disk before
51 * being partially zeroed.
55 struct xfs_inode
*ip
, /* inode */
56 loff_t pos
, /* offset in file */
57 size_t count
) /* size of data to zero */
60 struct address_space
*mapping
;
63 mapping
= VFS_I(ip
)->i_mapping
;
65 unsigned offset
, bytes
;
68 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
69 bytes
= PAGE_CACHE_SIZE
- offset
;
73 status
= pagecache_write_begin(NULL
, mapping
, pos
, bytes
,
74 AOP_FLAG_UNINTERRUPTIBLE
,
79 zero_user(page
, offset
, bytes
);
81 status
= pagecache_write_end(NULL
, mapping
, pos
, bytes
, bytes
,
83 WARN_ON(status
<= 0); /* can't return less than zero! */
97 struct inode
*inode
= file
->f_mapping
->host
;
98 struct xfs_inode
*ip
= XFS_I(inode
);
103 trace_xfs_file_fsync(ip
);
105 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
106 return -XFS_ERROR(EIO
);
108 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
113 * We always need to make sure that the required inode state is safe on
114 * disk. The inode might be clean but we still might need to force the
115 * log because of committed transactions that haven't hit the disk yet.
116 * Likewise, there could be unflushed non-transactional changes to the
117 * inode core that have to go to disk and this requires us to issue
118 * a synchronous transaction to capture these changes correctly.
120 * This code relies on the assumption that if the i_update_core field
121 * of the inode is clear and the inode is unpinned then it is clean
122 * and no action is required.
124 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
127 * First check if the VFS inode is marked dirty. All the dirtying
128 * of non-transactional updates no goes through mark_inode_dirty*,
129 * which allows us to distinguish beteeen pure timestamp updates
130 * and i_size updates which need to be caught for fdatasync.
131 * After that also theck for the dirty state in the XFS inode, which
132 * might gets cleared when the inode gets written out via the AIL
133 * or xfs_iflush_cluster.
135 if (((inode
->i_state
& I_DIRTY_DATASYNC
) ||
136 ((inode
->i_state
& I_DIRTY_SYNC
) && !datasync
)) &&
139 * Kick off a transaction to log the inode core to get the
140 * updates. The sync transaction will also force the log.
142 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
143 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
144 error
= xfs_trans_reserve(tp
, 0,
145 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
), 0, 0, 0);
147 xfs_trans_cancel(tp
, 0);
150 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
153 * Note - it's possible that we might have pushed ourselves out
154 * of the way during trans_reserve which would flush the inode.
155 * But there's no guarantee that the inode buffer has actually
156 * gone out yet (it's delwri). Plus the buffer could be pinned
157 * anyway if it's part of an inode in another recent
158 * transaction. So we play it safe and fire off the
159 * transaction anyway.
161 xfs_trans_ijoin(tp
, ip
);
162 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
163 xfs_trans_set_sync(tp
);
164 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
166 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
169 * Timestamps/size haven't changed since last inode flush or
170 * inode transaction commit. That means either nothing got
171 * written or a transaction committed which caught the updates.
172 * If the latter happened and the transaction hasn't hit the
173 * disk yet, the inode will be still be pinned. If it is,
176 if (xfs_ipincount(ip
)) {
177 error
= _xfs_log_force_lsn(ip
->i_mount
,
178 ip
->i_itemp
->ili_last_lsn
,
179 XFS_LOG_SYNC
, &log_flushed
);
181 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
184 if (ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) {
186 * If the log write didn't issue an ordered tag we need
187 * to flush the disk cache for the data device now.
190 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
193 * If this inode is on the RT dev we need to flush that
196 if (XFS_IS_REALTIME_INODE(ip
))
197 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
206 const struct iovec
*iovp
,
207 unsigned long nr_segs
,
210 struct file
*file
= iocb
->ki_filp
;
211 struct inode
*inode
= file
->f_mapping
->host
;
212 struct xfs_inode
*ip
= XFS_I(inode
);
213 struct xfs_mount
*mp
= ip
->i_mount
;
220 XFS_STATS_INC(xs_read_calls
);
222 BUG_ON(iocb
->ki_pos
!= pos
);
224 if (unlikely(file
->f_flags
& O_DIRECT
))
225 ioflags
|= IO_ISDIRECT
;
226 if (file
->f_mode
& FMODE_NOCMTIME
)
229 /* START copy & waste from filemap.c */
230 for (seg
= 0; seg
< nr_segs
; seg
++) {
231 const struct iovec
*iv
= &iovp
[seg
];
234 * If any segment has a negative length, or the cumulative
235 * length ever wraps negative then return -EINVAL.
238 if (unlikely((ssize_t
)(size
|iv
->iov_len
) < 0))
239 return XFS_ERROR(-EINVAL
);
241 /* END copy & waste from filemap.c */
243 if (unlikely(ioflags
& IO_ISDIRECT
)) {
244 xfs_buftarg_t
*target
=
245 XFS_IS_REALTIME_INODE(ip
) ?
246 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
247 if ((iocb
->ki_pos
& target
->bt_smask
) ||
248 (size
& target
->bt_smask
)) {
249 if (iocb
->ki_pos
== ip
->i_size
)
251 return -XFS_ERROR(EINVAL
);
255 n
= XFS_MAXIOFFSET(mp
) - iocb
->ki_pos
;
256 if (n
<= 0 || size
== 0)
262 if (XFS_FORCED_SHUTDOWN(mp
))
265 if (unlikely(ioflags
& IO_ISDIRECT
))
266 mutex_lock(&inode
->i_mutex
);
267 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
269 if (unlikely(ioflags
& IO_ISDIRECT
)) {
270 if (inode
->i_mapping
->nrpages
) {
271 ret
= -xfs_flushinval_pages(ip
,
272 (iocb
->ki_pos
& PAGE_CACHE_MASK
),
273 -1, FI_REMAPF_LOCKED
);
275 mutex_unlock(&inode
->i_mutex
);
277 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
282 trace_xfs_file_read(ip
, size
, iocb
->ki_pos
, ioflags
);
284 ret
= generic_file_aio_read(iocb
, iovp
, nr_segs
, iocb
->ki_pos
);
286 XFS_STATS_ADD(xs_read_bytes
, ret
);
288 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
293 xfs_file_splice_read(
296 struct pipe_inode_info
*pipe
,
300 struct xfs_inode
*ip
= XFS_I(infilp
->f_mapping
->host
);
304 XFS_STATS_INC(xs_read_calls
);
306 if (infilp
->f_mode
& FMODE_NOCMTIME
)
309 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
312 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
314 trace_xfs_file_splice_read(ip
, count
, *ppos
, ioflags
);
316 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
318 XFS_STATS_ADD(xs_read_bytes
, ret
);
320 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
325 xfs_file_splice_write(
326 struct pipe_inode_info
*pipe
,
327 struct file
*outfilp
,
332 struct inode
*inode
= outfilp
->f_mapping
->host
;
333 struct xfs_inode
*ip
= XFS_I(inode
);
334 xfs_fsize_t isize
, new_size
;
338 XFS_STATS_INC(xs_write_calls
);
340 if (outfilp
->f_mode
& FMODE_NOCMTIME
)
343 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
346 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
348 new_size
= *ppos
+ count
;
350 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
351 if (new_size
> ip
->i_size
)
352 ip
->i_new_size
= new_size
;
353 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
355 trace_xfs_file_splice_write(ip
, count
, *ppos
, ioflags
);
357 ret
= generic_file_splice_write(pipe
, outfilp
, ppos
, count
, flags
);
359 XFS_STATS_ADD(xs_write_bytes
, ret
);
361 isize
= i_size_read(inode
);
362 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& *ppos
> isize
))
365 if (*ppos
> ip
->i_size
) {
366 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
367 if (*ppos
> ip
->i_size
)
369 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
372 if (ip
->i_new_size
) {
373 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
375 if (ip
->i_d
.di_size
> ip
->i_size
)
376 ip
->i_d
.di_size
= ip
->i_size
;
377 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
379 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
384 * This routine is called to handle zeroing any space in the last
385 * block of the file that is beyond the EOF. We do this since the
386 * size is being increased without writing anything to that block
387 * and we don't want anyone to read the garbage on the disk.
389 STATIC
int /* error (positive) */
395 xfs_fileoff_t last_fsb
;
396 xfs_mount_t
*mp
= ip
->i_mount
;
401 xfs_bmbt_irec_t imap
;
403 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
405 zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
406 if (zero_offset
== 0) {
408 * There are no extra bytes in the last block on disk to
414 last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
416 error
= xfs_bmapi(NULL
, ip
, last_fsb
, 1, 0, NULL
, 0, &imap
,
423 * If the block underlying isize is just a hole, then there
424 * is nothing to zero.
426 if (imap
.br_startblock
== HOLESTARTBLOCK
) {
430 * Zero the part of the last block beyond the EOF, and write it
431 * out sync. We need to drop the ilock while we do this so we
432 * don't deadlock when the buffer cache calls back to us.
434 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
436 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
437 if (isize
+ zero_len
> offset
)
438 zero_len
= offset
- isize
;
439 error
= xfs_iozero(ip
, isize
, zero_len
);
441 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
447 * Zero any on disk space between the current EOF and the new,
448 * larger EOF. This handles the normal case of zeroing the remainder
449 * of the last block in the file and the unusual case of zeroing blocks
450 * out beyond the size of the file. This second case only happens
451 * with fixed size extents and when the system crashes before the inode
452 * size was updated but after blocks were allocated. If fill is set,
453 * then any holes in the range are filled and zeroed. If not, the holes
454 * are left alone as holes.
457 int /* error (positive) */
460 xfs_off_t offset
, /* starting I/O offset */
461 xfs_fsize_t isize
) /* current inode size */
463 xfs_mount_t
*mp
= ip
->i_mount
;
464 xfs_fileoff_t start_zero_fsb
;
465 xfs_fileoff_t end_zero_fsb
;
466 xfs_fileoff_t zero_count_fsb
;
467 xfs_fileoff_t last_fsb
;
468 xfs_fileoff_t zero_off
;
469 xfs_fsize_t zero_len
;
472 xfs_bmbt_irec_t imap
;
474 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
475 ASSERT(offset
> isize
);
478 * First handle zeroing the block on which isize resides.
479 * We only zero a part of that block so it is handled specially.
481 error
= xfs_zero_last_block(ip
, offset
, isize
);
483 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
488 * Calculate the range between the new size and the old
489 * where blocks needing to be zeroed may exist. To get the
490 * block where the last byte in the file currently resides,
491 * we need to subtract one from the size and truncate back
492 * to a block boundary. We subtract 1 in case the size is
493 * exactly on a block boundary.
495 last_fsb
= isize
? XFS_B_TO_FSBT(mp
, isize
- 1) : (xfs_fileoff_t
)-1;
496 start_zero_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
497 end_zero_fsb
= XFS_B_TO_FSBT(mp
, offset
- 1);
498 ASSERT((xfs_sfiloff_t
)last_fsb
< (xfs_sfiloff_t
)start_zero_fsb
);
499 if (last_fsb
== end_zero_fsb
) {
501 * The size was only incremented on its last block.
502 * We took care of that above, so just return.
507 ASSERT(start_zero_fsb
<= end_zero_fsb
);
508 while (start_zero_fsb
<= end_zero_fsb
) {
510 zero_count_fsb
= end_zero_fsb
- start_zero_fsb
+ 1;
511 error
= xfs_bmapi(NULL
, ip
, start_zero_fsb
, zero_count_fsb
,
512 0, NULL
, 0, &imap
, &nimaps
, NULL
);
514 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
519 if (imap
.br_state
== XFS_EXT_UNWRITTEN
||
520 imap
.br_startblock
== HOLESTARTBLOCK
) {
522 * This loop handles initializing pages that were
523 * partially initialized by the code below this
524 * loop. It basically zeroes the part of the page
525 * that sits on a hole and sets the page as P_HOLE
526 * and calls remapf if it is a mapped file.
528 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
529 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
534 * There are blocks we need to zero.
535 * Drop the inode lock while we're doing the I/O.
536 * We'll still have the iolock to protect us.
538 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
540 zero_off
= XFS_FSB_TO_B(mp
, start_zero_fsb
);
541 zero_len
= XFS_FSB_TO_B(mp
, imap
.br_blockcount
);
543 if ((zero_off
+ zero_len
) > offset
)
544 zero_len
= offset
- zero_off
;
546 error
= xfs_iozero(ip
, zero_off
, zero_len
);
551 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
552 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
554 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
560 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
568 const struct iovec
*iovp
,
569 unsigned long nr_segs
,
572 struct file
*file
= iocb
->ki_filp
;
573 struct address_space
*mapping
= file
->f_mapping
;
574 struct inode
*inode
= mapping
->host
;
575 struct xfs_inode
*ip
= XFS_I(inode
);
576 struct xfs_mount
*mp
= ip
->i_mount
;
577 ssize_t ret
= 0, error
= 0;
579 xfs_fsize_t isize
, new_size
;
581 size_t ocount
= 0, count
;
584 XFS_STATS_INC(xs_write_calls
);
586 BUG_ON(iocb
->ki_pos
!= pos
);
588 if (unlikely(file
->f_flags
& O_DIRECT
))
589 ioflags
|= IO_ISDIRECT
;
590 if (file
->f_mode
& FMODE_NOCMTIME
)
593 error
= generic_segment_checks(iovp
, &nr_segs
, &ocount
, VERIFY_READ
);
601 xfs_wait_for_freeze(mp
, SB_FREEZE_WRITE
);
603 if (XFS_FORCED_SHUTDOWN(mp
))
607 if (ioflags
& IO_ISDIRECT
) {
608 iolock
= XFS_IOLOCK_SHARED
;
611 iolock
= XFS_IOLOCK_EXCL
;
613 mutex_lock(&inode
->i_mutex
);
616 xfs_ilock(ip
, XFS_ILOCK_EXCL
|iolock
);
619 error
= -generic_write_checks(file
, &pos
, &count
,
620 S_ISBLK(inode
->i_mode
));
622 xfs_iunlock(ip
, XFS_ILOCK_EXCL
|iolock
);
623 goto out_unlock_mutex
;
626 if (ioflags
& IO_ISDIRECT
) {
627 xfs_buftarg_t
*target
=
628 XFS_IS_REALTIME_INODE(ip
) ?
629 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
631 if ((pos
& target
->bt_smask
) || (count
& target
->bt_smask
)) {
632 xfs_iunlock(ip
, XFS_ILOCK_EXCL
|iolock
);
633 return XFS_ERROR(-EINVAL
);
636 if (!need_i_mutex
&& (mapping
->nrpages
|| pos
> ip
->i_size
)) {
637 xfs_iunlock(ip
, XFS_ILOCK_EXCL
|iolock
);
638 iolock
= XFS_IOLOCK_EXCL
;
640 mutex_lock(&inode
->i_mutex
);
641 xfs_ilock(ip
, XFS_ILOCK_EXCL
|iolock
);
646 new_size
= pos
+ count
;
647 if (new_size
> ip
->i_size
)
648 ip
->i_new_size
= new_size
;
650 if (likely(!(ioflags
& IO_INVIS
)))
651 file_update_time(file
);
654 * If the offset is beyond the size of the file, we have a couple
655 * of things to do. First, if there is already space allocated
656 * we need to either create holes or zero the disk or ...
658 * If there is a page where the previous size lands, we need
659 * to zero it out up to the new size.
662 if (pos
> ip
->i_size
) {
663 error
= xfs_zero_eof(ip
, pos
, ip
->i_size
);
665 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
666 goto out_unlock_internal
;
669 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
672 * If we're writing the file then make sure to clear the
673 * setuid and setgid bits if the process is not being run
674 * by root. This keeps people from modifying setuid and
677 error
= -file_remove_suid(file
);
679 goto out_unlock_internal
;
681 /* We can write back this queue in page reclaim */
682 current
->backing_dev_info
= mapping
->backing_dev_info
;
684 if ((ioflags
& IO_ISDIRECT
)) {
685 if (mapping
->nrpages
) {
686 WARN_ON(need_i_mutex
== 0);
687 error
= xfs_flushinval_pages(ip
,
688 (pos
& PAGE_CACHE_MASK
),
689 -1, FI_REMAPF_LOCKED
);
691 goto out_unlock_internal
;
695 /* demote the lock now the cached pages are gone */
696 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
697 mutex_unlock(&inode
->i_mutex
);
699 iolock
= XFS_IOLOCK_SHARED
;
703 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
, ioflags
);
704 ret
= generic_file_direct_write(iocb
, iovp
,
705 &nr_segs
, pos
, &iocb
->ki_pos
, count
, ocount
);
708 * direct-io write to a hole: fall through to buffered I/O
709 * for completing the rest of the request.
711 if (ret
>= 0 && ret
!= count
) {
712 XFS_STATS_ADD(xs_write_bytes
, ret
);
717 ioflags
&= ~IO_ISDIRECT
;
718 xfs_iunlock(ip
, iolock
);
726 trace_xfs_file_buffered_write(ip
, count
, iocb
->ki_pos
, ioflags
);
727 ret2
= generic_file_buffered_write(iocb
, iovp
, nr_segs
,
728 pos
, &iocb
->ki_pos
, count
, ret
);
730 * if we just got an ENOSPC, flush the inode now we
731 * aren't holding any page locks and retry *once*
733 if (ret2
== -ENOSPC
&& !enospc
) {
734 error
= xfs_flush_pages(ip
, 0, -1, 0, FI_NONE
);
736 goto out_unlock_internal
;
743 current
->backing_dev_info
= NULL
;
745 isize
= i_size_read(inode
);
746 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& iocb
->ki_pos
> isize
))
747 iocb
->ki_pos
= isize
;
749 if (iocb
->ki_pos
> ip
->i_size
) {
750 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
751 if (iocb
->ki_pos
> ip
->i_size
)
752 ip
->i_size
= iocb
->ki_pos
;
753 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
758 goto out_unlock_internal
;
760 XFS_STATS_ADD(xs_write_bytes
, ret
);
762 /* Handle various SYNC-type writes */
763 if ((file
->f_flags
& O_DSYNC
) || IS_SYNC(inode
)) {
764 loff_t end
= pos
+ ret
- 1;
767 xfs_iunlock(ip
, iolock
);
769 mutex_unlock(&inode
->i_mutex
);
771 error2
= filemap_write_and_wait_range(mapping
, pos
, end
);
775 mutex_lock(&inode
->i_mutex
);
776 xfs_ilock(ip
, iolock
);
778 error2
= -xfs_file_fsync(file
,
779 (file
->f_flags
& __O_SYNC
) ? 0 : 1);
785 if (ip
->i_new_size
) {
786 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
789 * If this was a direct or synchronous I/O that failed (such
790 * as ENOSPC) then part of the I/O may have been written to
791 * disk before the error occured. In this case the on-disk
792 * file size may have been adjusted beyond the in-memory file
793 * size and now needs to be truncated back.
795 if (ip
->i_d
.di_size
> ip
->i_size
)
796 ip
->i_d
.di_size
= ip
->i_size
;
797 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
799 xfs_iunlock(ip
, iolock
);
802 mutex_unlock(&inode
->i_mutex
);
811 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
813 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
823 struct xfs_inode
*ip
= XFS_I(inode
);
827 error
= xfs_file_open(inode
, file
);
832 * If there are any blocks, read-ahead block 0 as we're almost
833 * certain to have the next operation be a read there.
835 mode
= xfs_ilock_map_shared(ip
);
836 if (ip
->i_d
.di_nextents
> 0)
837 xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
838 xfs_iunlock(ip
, mode
);
847 return -xfs_release(XFS_I(inode
));
856 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
857 xfs_inode_t
*ip
= XFS_I(inode
);
862 * The Linux API doesn't pass down the total size of the buffer
863 * we read into down to the filesystem. With the filldir concept
864 * it's not needed for correct information, but the XFS dir2 leaf
865 * code wants an estimate of the buffer size to calculate it's
866 * readahead window and size the buffers used for mapping to
869 * Try to give it an estimate that's good enough, maybe at some
870 * point we can change the ->readdir prototype to include the
871 * buffer size. For now we use the current glibc buffer size.
873 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
875 error
= xfs_readdir(ip
, dirent
, bufsize
,
876 (xfs_off_t
*)&filp
->f_pos
, filldir
);
885 struct vm_area_struct
*vma
)
887 vma
->vm_ops
= &xfs_file_vm_ops
;
888 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
895 * mmap()d file has taken write protection fault and is being made
896 * writable. We can set the page state up correctly for a writable
897 * page, which means we can do correct delalloc accounting (ENOSPC
898 * checking!) and unwritten extent mapping.
902 struct vm_area_struct
*vma
,
903 struct vm_fault
*vmf
)
905 return block_page_mkwrite(vma
, vmf
, xfs_get_blocks
);
908 const struct file_operations xfs_file_operations
= {
909 .llseek
= generic_file_llseek
,
910 .read
= do_sync_read
,
911 .write
= do_sync_write
,
912 .aio_read
= xfs_file_aio_read
,
913 .aio_write
= xfs_file_aio_write
,
914 .splice_read
= xfs_file_splice_read
,
915 .splice_write
= xfs_file_splice_write
,
916 .unlocked_ioctl
= xfs_file_ioctl
,
918 .compat_ioctl
= xfs_file_compat_ioctl
,
920 .mmap
= xfs_file_mmap
,
921 .open
= xfs_file_open
,
922 .release
= xfs_file_release
,
923 .fsync
= xfs_file_fsync
,
926 const struct file_operations xfs_dir_file_operations
= {
927 .open
= xfs_dir_open
,
928 .read
= generic_read_dir
,
929 .readdir
= xfs_file_readdir
,
930 .llseek
= generic_file_llseek
,
931 .unlocked_ioctl
= xfs_file_ioctl
,
933 .compat_ioctl
= xfs_file_compat_ioctl
,
935 .fsync
= xfs_file_fsync
,
938 static const struct vm_operations_struct xfs_file_vm_ops
= {
939 .fault
= filemap_fault
,
940 .page_mkwrite
= xfs_vm_page_mkwrite
,