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
26 #include "xfs_trans.h"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_alloc.h"
33 #include "xfs_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
40 #include "xfs_error.h"
42 #include "xfs_vnodeops.h"
43 #include "xfs_da_btree.h"
44 #include "xfs_ioctl.h"
45 #include "xfs_trace.h"
47 #include <linux/dcache.h>
49 static const struct vm_operations_struct xfs_file_vm_ops
;
54 * xfs_iozero clears the specified range of buffer supplied,
55 * and marks all the affected blocks as valid and modified. If
56 * an affected block is not allocated, it will be allocated. If
57 * an affected block is not completely overwritten, and is not
58 * valid before the operation, it will be read from disk before
59 * being partially zeroed.
63 struct xfs_inode
*ip
, /* inode */
64 loff_t pos
, /* offset in file */
65 size_t count
) /* size of data to zero */
68 struct address_space
*mapping
;
71 mapping
= VFS_I(ip
)->i_mapping
;
73 unsigned offset
, bytes
;
76 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
77 bytes
= PAGE_CACHE_SIZE
- offset
;
81 status
= pagecache_write_begin(NULL
, mapping
, pos
, bytes
,
82 AOP_FLAG_UNINTERRUPTIBLE
,
87 zero_user(page
, offset
, bytes
);
89 status
= pagecache_write_end(NULL
, mapping
, pos
, bytes
, bytes
,
91 WARN_ON(status
<= 0); /* can't return less than zero! */
103 struct dentry
*dentry
,
106 struct xfs_inode
*ip
= XFS_I(dentry
->d_inode
);
107 struct xfs_trans
*tp
;
111 xfs_itrace_entry(ip
);
113 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
114 return -XFS_ERROR(EIO
);
116 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
119 * We always need to make sure that the required inode state is safe on
120 * disk. The inode might be clean but we still might need to force the
121 * log because of committed transactions that haven't hit the disk yet.
122 * Likewise, there could be unflushed non-transactional changes to the
123 * inode core that have to go to disk and this requires us to issue
124 * a synchronous transaction to capture these changes correctly.
126 * This code relies on the assumption that if the i_update_core field
127 * of the inode is clear and the inode is unpinned then it is clean
128 * and no action is required.
130 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
133 * First check if the VFS inode is marked dirty. All the dirtying
134 * of non-transactional updates no goes through mark_inode_dirty*,
135 * which allows us to distinguish beteeen pure timestamp updates
136 * and i_size updates which need to be caught for fdatasync.
137 * After that also theck for the dirty state in the XFS inode, which
138 * might gets cleared when the inode gets written out via the AIL
139 * or xfs_iflush_cluster.
141 if (((dentry
->d_inode
->i_state
& I_DIRTY_DATASYNC
) ||
142 ((dentry
->d_inode
->i_state
& I_DIRTY_SYNC
) && !datasync
)) &&
145 * Kick off a transaction to log the inode core to get the
146 * updates. The sync transaction will also force the log.
148 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
149 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
150 error
= xfs_trans_reserve(tp
, 0,
151 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
), 0, 0, 0);
153 xfs_trans_cancel(tp
, 0);
156 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
159 * Note - it's possible that we might have pushed ourselves out
160 * of the way during trans_reserve which would flush the inode.
161 * But there's no guarantee that the inode buffer has actually
162 * gone out yet (it's delwri). Plus the buffer could be pinned
163 * anyway if it's part of an inode in another recent
164 * transaction. So we play it safe and fire off the
165 * transaction anyway.
167 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
168 xfs_trans_ihold(tp
, ip
);
169 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
170 xfs_trans_set_sync(tp
);
171 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
173 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
176 * Timestamps/size haven't changed since last inode flush or
177 * inode transaction commit. That means either nothing got
178 * written or a transaction committed which caught the updates.
179 * If the latter happened and the transaction hasn't hit the
180 * disk yet, the inode will be still be pinned. If it is,
183 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
184 if (xfs_ipincount(ip
)) {
185 if (ip
->i_itemp
->ili_last_lsn
) {
186 error
= _xfs_log_force_lsn(ip
->i_mount
,
187 ip
->i_itemp
->ili_last_lsn
,
188 XFS_LOG_SYNC
, &log_flushed
);
190 error
= _xfs_log_force(ip
->i_mount
,
191 XFS_LOG_SYNC
, &log_flushed
);
196 if (ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) {
198 * If the log write didn't issue an ordered tag we need
199 * to flush the disk cache for the data device now.
202 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
205 * If this inode is on the RT dev we need to flush that
208 if (XFS_IS_REALTIME_INODE(ip
))
209 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
218 const struct iovec
*iovp
,
219 unsigned long nr_segs
,
222 struct file
*file
= iocb
->ki_filp
;
223 struct inode
*inode
= file
->f_mapping
->host
;
224 struct xfs_inode
*ip
= XFS_I(inode
);
225 struct xfs_mount
*mp
= ip
->i_mount
;
232 XFS_STATS_INC(xs_read_calls
);
234 BUG_ON(iocb
->ki_pos
!= pos
);
236 if (unlikely(file
->f_flags
& O_DIRECT
))
237 ioflags
|= IO_ISDIRECT
;
238 if (file
->f_mode
& FMODE_NOCMTIME
)
241 /* START copy & waste from filemap.c */
242 for (seg
= 0; seg
< nr_segs
; seg
++) {
243 const struct iovec
*iv
= &iovp
[seg
];
246 * If any segment has a negative length, or the cumulative
247 * length ever wraps negative then return -EINVAL.
250 if (unlikely((ssize_t
)(size
|iv
->iov_len
) < 0))
251 return XFS_ERROR(-EINVAL
);
253 /* END copy & waste from filemap.c */
255 if (unlikely(ioflags
& IO_ISDIRECT
)) {
256 xfs_buftarg_t
*target
=
257 XFS_IS_REALTIME_INODE(ip
) ?
258 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
259 if ((iocb
->ki_pos
& target
->bt_smask
) ||
260 (size
& target
->bt_smask
)) {
261 if (iocb
->ki_pos
== ip
->i_size
)
263 return -XFS_ERROR(EINVAL
);
267 n
= XFS_MAXIOFFSET(mp
) - iocb
->ki_pos
;
268 if (n
<= 0 || size
== 0)
274 if (XFS_FORCED_SHUTDOWN(mp
))
277 if (unlikely(ioflags
& IO_ISDIRECT
))
278 mutex_lock(&inode
->i_mutex
);
279 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
281 if (DM_EVENT_ENABLED(ip
, DM_EVENT_READ
) && !(ioflags
& IO_INVIS
)) {
282 int dmflags
= FILP_DELAY_FLAG(file
) | DM_SEM_FLAG_RD(ioflags
);
283 int iolock
= XFS_IOLOCK_SHARED
;
285 ret
= -XFS_SEND_DATA(mp
, DM_EVENT_READ
, ip
, iocb
->ki_pos
, size
,
288 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
289 if (unlikely(ioflags
& IO_ISDIRECT
))
290 mutex_unlock(&inode
->i_mutex
);
295 if (unlikely(ioflags
& IO_ISDIRECT
)) {
296 if (inode
->i_mapping
->nrpages
) {
297 ret
= -xfs_flushinval_pages(ip
,
298 (iocb
->ki_pos
& PAGE_CACHE_MASK
),
299 -1, FI_REMAPF_LOCKED
);
301 mutex_unlock(&inode
->i_mutex
);
303 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
308 trace_xfs_file_read(ip
, size
, iocb
->ki_pos
, ioflags
);
310 ret
= generic_file_aio_read(iocb
, iovp
, nr_segs
, iocb
->ki_pos
);
312 XFS_STATS_ADD(xs_read_bytes
, ret
);
314 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
319 xfs_file_splice_read(
322 struct pipe_inode_info
*pipe
,
326 struct xfs_inode
*ip
= XFS_I(infilp
->f_mapping
->host
);
327 struct xfs_mount
*mp
= ip
->i_mount
;
331 XFS_STATS_INC(xs_read_calls
);
333 if (infilp
->f_mode
& FMODE_NOCMTIME
)
336 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
339 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
341 if (DM_EVENT_ENABLED(ip
, DM_EVENT_READ
) && !(ioflags
& IO_INVIS
)) {
342 int iolock
= XFS_IOLOCK_SHARED
;
345 error
= XFS_SEND_DATA(mp
, DM_EVENT_READ
, ip
, *ppos
, count
,
346 FILP_DELAY_FLAG(infilp
), &iolock
);
348 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
353 trace_xfs_file_splice_read(ip
, count
, *ppos
, ioflags
);
355 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
357 XFS_STATS_ADD(xs_read_bytes
, ret
);
359 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
364 xfs_file_splice_write(
365 struct pipe_inode_info
*pipe
,
366 struct file
*outfilp
,
371 struct inode
*inode
= outfilp
->f_mapping
->host
;
372 struct xfs_inode
*ip
= XFS_I(inode
);
373 struct xfs_mount
*mp
= ip
->i_mount
;
374 xfs_fsize_t isize
, new_size
;
378 XFS_STATS_INC(xs_write_calls
);
380 if (outfilp
->f_mode
& FMODE_NOCMTIME
)
383 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
386 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
388 if (DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
) && !(ioflags
& IO_INVIS
)) {
389 int iolock
= XFS_IOLOCK_EXCL
;
392 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, ip
, *ppos
, count
,
393 FILP_DELAY_FLAG(outfilp
), &iolock
);
395 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
400 new_size
= *ppos
+ count
;
402 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
403 if (new_size
> ip
->i_size
)
404 ip
->i_new_size
= new_size
;
405 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
407 trace_xfs_file_splice_write(ip
, count
, *ppos
, ioflags
);
409 ret
= generic_file_splice_write(pipe
, outfilp
, ppos
, count
, flags
);
411 XFS_STATS_ADD(xs_write_bytes
, ret
);
413 isize
= i_size_read(inode
);
414 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& *ppos
> isize
))
417 if (*ppos
> ip
->i_size
) {
418 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
419 if (*ppos
> ip
->i_size
)
421 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
424 if (ip
->i_new_size
) {
425 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
427 if (ip
->i_d
.di_size
> ip
->i_size
)
428 ip
->i_d
.di_size
= ip
->i_size
;
429 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
431 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
436 * This routine is called to handle zeroing any space in the last
437 * block of the file that is beyond the EOF. We do this since the
438 * size is being increased without writing anything to that block
439 * and we don't want anyone to read the garbage on the disk.
441 STATIC
int /* error (positive) */
447 xfs_fileoff_t last_fsb
;
448 xfs_mount_t
*mp
= ip
->i_mount
;
453 xfs_bmbt_irec_t imap
;
455 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
457 zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
458 if (zero_offset
== 0) {
460 * There are no extra bytes in the last block on disk to
466 last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
468 error
= xfs_bmapi(NULL
, ip
, last_fsb
, 1, 0, NULL
, 0, &imap
,
469 &nimaps
, NULL
, NULL
);
475 * If the block underlying isize is just a hole, then there
476 * is nothing to zero.
478 if (imap
.br_startblock
== HOLESTARTBLOCK
) {
482 * Zero the part of the last block beyond the EOF, and write it
483 * out sync. We need to drop the ilock while we do this so we
484 * don't deadlock when the buffer cache calls back to us.
486 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
488 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
489 if (isize
+ zero_len
> offset
)
490 zero_len
= offset
- isize
;
491 error
= xfs_iozero(ip
, isize
, zero_len
);
493 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
499 * Zero any on disk space between the current EOF and the new,
500 * larger EOF. This handles the normal case of zeroing the remainder
501 * of the last block in the file and the unusual case of zeroing blocks
502 * out beyond the size of the file. This second case only happens
503 * with fixed size extents and when the system crashes before the inode
504 * size was updated but after blocks were allocated. If fill is set,
505 * then any holes in the range are filled and zeroed. If not, the holes
506 * are left alone as holes.
509 int /* error (positive) */
512 xfs_off_t offset
, /* starting I/O offset */
513 xfs_fsize_t isize
) /* current inode size */
515 xfs_mount_t
*mp
= ip
->i_mount
;
516 xfs_fileoff_t start_zero_fsb
;
517 xfs_fileoff_t end_zero_fsb
;
518 xfs_fileoff_t zero_count_fsb
;
519 xfs_fileoff_t last_fsb
;
520 xfs_fileoff_t zero_off
;
521 xfs_fsize_t zero_len
;
524 xfs_bmbt_irec_t imap
;
526 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
527 ASSERT(offset
> isize
);
530 * First handle zeroing the block on which isize resides.
531 * We only zero a part of that block so it is handled specially.
533 error
= xfs_zero_last_block(ip
, offset
, isize
);
535 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
540 * Calculate the range between the new size and the old
541 * where blocks needing to be zeroed may exist. To get the
542 * block where the last byte in the file currently resides,
543 * we need to subtract one from the size and truncate back
544 * to a block boundary. We subtract 1 in case the size is
545 * exactly on a block boundary.
547 last_fsb
= isize
? XFS_B_TO_FSBT(mp
, isize
- 1) : (xfs_fileoff_t
)-1;
548 start_zero_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
549 end_zero_fsb
= XFS_B_TO_FSBT(mp
, offset
- 1);
550 ASSERT((xfs_sfiloff_t
)last_fsb
< (xfs_sfiloff_t
)start_zero_fsb
);
551 if (last_fsb
== end_zero_fsb
) {
553 * The size was only incremented on its last block.
554 * We took care of that above, so just return.
559 ASSERT(start_zero_fsb
<= end_zero_fsb
);
560 while (start_zero_fsb
<= end_zero_fsb
) {
562 zero_count_fsb
= end_zero_fsb
- start_zero_fsb
+ 1;
563 error
= xfs_bmapi(NULL
, ip
, start_zero_fsb
, zero_count_fsb
,
564 0, NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
566 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
571 if (imap
.br_state
== XFS_EXT_UNWRITTEN
||
572 imap
.br_startblock
== HOLESTARTBLOCK
) {
574 * This loop handles initializing pages that were
575 * partially initialized by the code below this
576 * loop. It basically zeroes the part of the page
577 * that sits on a hole and sets the page as P_HOLE
578 * and calls remapf if it is a mapped file.
580 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
581 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
586 * There are blocks we need to zero.
587 * Drop the inode lock while we're doing the I/O.
588 * We'll still have the iolock to protect us.
590 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
592 zero_off
= XFS_FSB_TO_B(mp
, start_zero_fsb
);
593 zero_len
= XFS_FSB_TO_B(mp
, imap
.br_blockcount
);
595 if ((zero_off
+ zero_len
) > offset
)
596 zero_len
= offset
- zero_off
;
598 error
= xfs_iozero(ip
, zero_off
, zero_len
);
603 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
604 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
606 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
612 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
620 const struct iovec
*iovp
,
621 unsigned long nr_segs
,
624 struct file
*file
= iocb
->ki_filp
;
625 struct address_space
*mapping
= file
->f_mapping
;
626 struct inode
*inode
= mapping
->host
;
627 struct xfs_inode
*ip
= XFS_I(inode
);
628 struct xfs_mount
*mp
= ip
->i_mount
;
629 ssize_t ret
= 0, error
= 0;
631 xfs_fsize_t isize
, new_size
;
634 size_t ocount
= 0, count
;
637 XFS_STATS_INC(xs_write_calls
);
639 BUG_ON(iocb
->ki_pos
!= pos
);
641 if (unlikely(file
->f_flags
& O_DIRECT
))
642 ioflags
|= IO_ISDIRECT
;
643 if (file
->f_mode
& FMODE_NOCMTIME
)
646 error
= generic_segment_checks(iovp
, &nr_segs
, &ocount
, VERIFY_READ
);
654 xfs_wait_for_freeze(mp
, SB_FREEZE_WRITE
);
656 if (XFS_FORCED_SHUTDOWN(mp
))
660 if (ioflags
& IO_ISDIRECT
) {
661 iolock
= XFS_IOLOCK_SHARED
;
664 iolock
= XFS_IOLOCK_EXCL
;
666 mutex_lock(&inode
->i_mutex
);
669 xfs_ilock(ip
, XFS_ILOCK_EXCL
|iolock
);
672 error
= -generic_write_checks(file
, &pos
, &count
,
673 S_ISBLK(inode
->i_mode
));
675 xfs_iunlock(ip
, XFS_ILOCK_EXCL
|iolock
);
676 goto out_unlock_mutex
;
679 if ((DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
) &&
680 !(ioflags
& IO_INVIS
) && !eventsent
)) {
681 int dmflags
= FILP_DELAY_FLAG(file
);
684 dmflags
|= DM_FLAGS_IMUX
;
686 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
687 error
= XFS_SEND_DATA(ip
->i_mount
, DM_EVENT_WRITE
, ip
,
688 pos
, count
, dmflags
, &iolock
);
690 goto out_unlock_internal
;
692 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
696 * The iolock was dropped and reacquired in XFS_SEND_DATA
697 * so we have to recheck the size when appending.
698 * We will only "goto start;" once, since having sent the
699 * event prevents another call to XFS_SEND_DATA, which is
700 * what allows the size to change in the first place.
702 if ((file
->f_flags
& O_APPEND
) && pos
!= ip
->i_size
)
706 if (ioflags
& IO_ISDIRECT
) {
707 xfs_buftarg_t
*target
=
708 XFS_IS_REALTIME_INODE(ip
) ?
709 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
711 if ((pos
& target
->bt_smask
) || (count
& target
->bt_smask
)) {
712 xfs_iunlock(ip
, XFS_ILOCK_EXCL
|iolock
);
713 return XFS_ERROR(-EINVAL
);
716 if (!need_i_mutex
&& (mapping
->nrpages
|| pos
> ip
->i_size
)) {
717 xfs_iunlock(ip
, XFS_ILOCK_EXCL
|iolock
);
718 iolock
= XFS_IOLOCK_EXCL
;
720 mutex_lock(&inode
->i_mutex
);
721 xfs_ilock(ip
, XFS_ILOCK_EXCL
|iolock
);
726 new_size
= pos
+ count
;
727 if (new_size
> ip
->i_size
)
728 ip
->i_new_size
= new_size
;
730 if (likely(!(ioflags
& IO_INVIS
)))
731 file_update_time(file
);
734 * If the offset is beyond the size of the file, we have a couple
735 * of things to do. First, if there is already space allocated
736 * we need to either create holes or zero the disk or ...
738 * If there is a page where the previous size lands, we need
739 * to zero it out up to the new size.
742 if (pos
> ip
->i_size
) {
743 error
= xfs_zero_eof(ip
, pos
, ip
->i_size
);
745 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
746 goto out_unlock_internal
;
749 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
752 * If we're writing the file then make sure to clear the
753 * setuid and setgid bits if the process is not being run
754 * by root. This keeps people from modifying setuid and
757 error
= -file_remove_suid(file
);
759 goto out_unlock_internal
;
761 /* We can write back this queue in page reclaim */
762 current
->backing_dev_info
= mapping
->backing_dev_info
;
764 if ((ioflags
& IO_ISDIRECT
)) {
765 if (mapping
->nrpages
) {
766 WARN_ON(need_i_mutex
== 0);
767 error
= xfs_flushinval_pages(ip
,
768 (pos
& PAGE_CACHE_MASK
),
769 -1, FI_REMAPF_LOCKED
);
771 goto out_unlock_internal
;
775 /* demote the lock now the cached pages are gone */
776 xfs_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
777 mutex_unlock(&inode
->i_mutex
);
779 iolock
= XFS_IOLOCK_SHARED
;
783 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
, ioflags
);
784 ret
= generic_file_direct_write(iocb
, iovp
,
785 &nr_segs
, pos
, &iocb
->ki_pos
, count
, ocount
);
788 * direct-io write to a hole: fall through to buffered I/O
789 * for completing the rest of the request.
791 if (ret
>= 0 && ret
!= count
) {
792 XFS_STATS_ADD(xs_write_bytes
, ret
);
797 ioflags
&= ~IO_ISDIRECT
;
798 xfs_iunlock(ip
, iolock
);
806 trace_xfs_file_buffered_write(ip
, count
, iocb
->ki_pos
, ioflags
);
807 ret2
= generic_file_buffered_write(iocb
, iovp
, nr_segs
,
808 pos
, &iocb
->ki_pos
, count
, ret
);
810 * if we just got an ENOSPC, flush the inode now we
811 * aren't holding any page locks and retry *once*
813 if (ret2
== -ENOSPC
&& !enospc
) {
814 error
= xfs_flush_pages(ip
, 0, -1, 0, FI_NONE
);
816 goto out_unlock_internal
;
823 current
->backing_dev_info
= NULL
;
825 isize
= i_size_read(inode
);
826 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& iocb
->ki_pos
> isize
))
827 iocb
->ki_pos
= isize
;
829 if (iocb
->ki_pos
> ip
->i_size
) {
830 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
831 if (iocb
->ki_pos
> ip
->i_size
)
832 ip
->i_size
= iocb
->ki_pos
;
833 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
836 if (ret
== -ENOSPC
&&
837 DM_EVENT_ENABLED(ip
, DM_EVENT_NOSPACE
) && !(ioflags
& IO_INVIS
)) {
838 xfs_iunlock(ip
, iolock
);
840 mutex_unlock(&inode
->i_mutex
);
841 error
= XFS_SEND_NAMESP(ip
->i_mount
, DM_EVENT_NOSPACE
, ip
,
842 DM_RIGHT_NULL
, ip
, DM_RIGHT_NULL
, NULL
, NULL
,
843 0, 0, 0); /* Delay flag intentionally unused */
845 mutex_lock(&inode
->i_mutex
);
846 xfs_ilock(ip
, iolock
);
848 goto out_unlock_internal
;
854 goto out_unlock_internal
;
856 XFS_STATS_ADD(xs_write_bytes
, ret
);
858 /* Handle various SYNC-type writes */
859 if ((file
->f_flags
& O_DSYNC
) || IS_SYNC(inode
)) {
860 loff_t end
= pos
+ ret
- 1;
863 xfs_iunlock(ip
, iolock
);
865 mutex_unlock(&inode
->i_mutex
);
867 error2
= filemap_write_and_wait_range(mapping
, pos
, end
);
871 mutex_lock(&inode
->i_mutex
);
872 xfs_ilock(ip
, iolock
);
874 error2
= -xfs_file_fsync(file
, file
->f_path
.dentry
,
875 (file
->f_flags
& __O_SYNC
) ? 0 : 1);
881 if (ip
->i_new_size
) {
882 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
885 * If this was a direct or synchronous I/O that failed (such
886 * as ENOSPC) then part of the I/O may have been written to
887 * disk before the error occured. In this case the on-disk
888 * file size may have been adjusted beyond the in-memory file
889 * size and now needs to be truncated back.
891 if (ip
->i_d
.di_size
> ip
->i_size
)
892 ip
->i_d
.di_size
= ip
->i_size
;
893 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
895 xfs_iunlock(ip
, iolock
);
898 mutex_unlock(&inode
->i_mutex
);
907 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
909 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
919 struct xfs_inode
*ip
= XFS_I(inode
);
923 error
= xfs_file_open(inode
, file
);
928 * If there are any blocks, read-ahead block 0 as we're almost
929 * certain to have the next operation be a read there.
931 mode
= xfs_ilock_map_shared(ip
);
932 if (ip
->i_d
.di_nextents
> 0)
933 xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
934 xfs_iunlock(ip
, mode
);
943 return -xfs_release(XFS_I(inode
));
952 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
953 xfs_inode_t
*ip
= XFS_I(inode
);
958 * The Linux API doesn't pass down the total size of the buffer
959 * we read into down to the filesystem. With the filldir concept
960 * it's not needed for correct information, but the XFS dir2 leaf
961 * code wants an estimate of the buffer size to calculate it's
962 * readahead window and size the buffers used for mapping to
965 * Try to give it an estimate that's good enough, maybe at some
966 * point we can change the ->readdir prototype to include the
967 * buffer size. For now we use the current glibc buffer size.
969 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
971 error
= xfs_readdir(ip
, dirent
, bufsize
,
972 (xfs_off_t
*)&filp
->f_pos
, filldir
);
981 struct vm_area_struct
*vma
)
983 vma
->vm_ops
= &xfs_file_vm_ops
;
984 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
991 * mmap()d file has taken write protection fault and is being made
992 * writable. We can set the page state up correctly for a writable
993 * page, which means we can do correct delalloc accounting (ENOSPC
994 * checking!) and unwritten extent mapping.
998 struct vm_area_struct
*vma
,
999 struct vm_fault
*vmf
)
1001 return block_page_mkwrite(vma
, vmf
, xfs_get_blocks
);
1004 const struct file_operations xfs_file_operations
= {
1005 .llseek
= generic_file_llseek
,
1006 .read
= do_sync_read
,
1007 .write
= do_sync_write
,
1008 .aio_read
= xfs_file_aio_read
,
1009 .aio_write
= xfs_file_aio_write
,
1010 .splice_read
= xfs_file_splice_read
,
1011 .splice_write
= xfs_file_splice_write
,
1012 .unlocked_ioctl
= xfs_file_ioctl
,
1013 #ifdef CONFIG_COMPAT
1014 .compat_ioctl
= xfs_file_compat_ioctl
,
1016 .mmap
= xfs_file_mmap
,
1017 .open
= xfs_file_open
,
1018 .release
= xfs_file_release
,
1019 .fsync
= xfs_file_fsync
,
1020 #ifdef HAVE_FOP_OPEN_EXEC
1021 .open_exec
= xfs_file_open_exec
,
1025 const struct file_operations xfs_dir_file_operations
= {
1026 .open
= xfs_dir_open
,
1027 .read
= generic_read_dir
,
1028 .readdir
= xfs_file_readdir
,
1029 .llseek
= generic_file_llseek
,
1030 .unlocked_ioctl
= xfs_file_ioctl
,
1031 #ifdef CONFIG_COMPAT
1032 .compat_ioctl
= xfs_file_compat_ioctl
,
1034 .fsync
= xfs_file_fsync
,
1037 static const struct vm_operations_struct xfs_file_vm_ops
= {
1038 .fault
= filemap_fault
,
1039 .page_mkwrite
= xfs_vm_page_mkwrite
,