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
23 #include "xfs_trans.h"
24 #include "xfs_mount.h"
25 #include "xfs_bmap_btree.h"
26 #include "xfs_alloc.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_error.h"
32 #include "xfs_vnodeops.h"
33 #include "xfs_da_btree.h"
34 #include "xfs_ioctl.h"
35 #include "xfs_trace.h"
37 #include <linux/dcache.h>
38 #include <linux/falloc.h>
40 static const struct vm_operations_struct xfs_file_vm_ops
;
43 * Locking primitives for read and write IO paths to ensure we consistently use
44 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
51 if (type
& XFS_IOLOCK_EXCL
)
52 mutex_lock(&VFS_I(ip
)->i_mutex
);
61 xfs_iunlock(ip
, type
);
62 if (type
& XFS_IOLOCK_EXCL
)
63 mutex_unlock(&VFS_I(ip
)->i_mutex
);
71 xfs_ilock_demote(ip
, type
);
72 if (type
& XFS_IOLOCK_EXCL
)
73 mutex_unlock(&VFS_I(ip
)->i_mutex
);
79 * xfs_iozero clears the specified range of buffer supplied,
80 * and marks all the affected blocks as valid and modified. If
81 * an affected block is not allocated, it will be allocated. If
82 * an affected block is not completely overwritten, and is not
83 * valid before the operation, it will be read from disk before
84 * being partially zeroed.
88 struct xfs_inode
*ip
, /* inode */
89 loff_t pos
, /* offset in file */
90 size_t count
) /* size of data to zero */
93 struct address_space
*mapping
;
96 mapping
= VFS_I(ip
)->i_mapping
;
98 unsigned offset
, bytes
;
101 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
102 bytes
= PAGE_CACHE_SIZE
- offset
;
106 status
= pagecache_write_begin(NULL
, mapping
, pos
, bytes
,
107 AOP_FLAG_UNINTERRUPTIBLE
,
112 zero_user(page
, offset
, bytes
);
114 status
= pagecache_write_end(NULL
, mapping
, pos
, bytes
, bytes
,
116 WARN_ON(status
<= 0); /* can't return less than zero! */
126 * Fsync operations on directories are much simpler than on regular files,
127 * as there is no file data to flush, and thus also no need for explicit
128 * cache flush operations, and there are no non-transaction metadata updates
129 * on directories either.
138 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
139 struct xfs_mount
*mp
= ip
->i_mount
;
142 trace_xfs_dir_fsync(ip
);
144 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
145 if (xfs_ipincount(ip
))
146 lsn
= ip
->i_itemp
->ili_last_lsn
;
147 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
151 return _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
161 struct inode
*inode
= file
->f_mapping
->host
;
162 struct xfs_inode
*ip
= XFS_I(inode
);
163 struct xfs_mount
*mp
= ip
->i_mount
;
168 trace_xfs_file_fsync(ip
);
170 error
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
174 if (XFS_FORCED_SHUTDOWN(mp
))
175 return -XFS_ERROR(EIO
);
177 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
179 if (mp
->m_flags
& XFS_MOUNT_BARRIER
) {
181 * If we have an RT and/or log subvolume we need to make sure
182 * to flush the write cache the device used for file data
183 * first. This is to ensure newly written file data make
184 * it to disk before logging the new inode size in case of
185 * an extending write.
187 if (XFS_IS_REALTIME_INODE(ip
))
188 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
189 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
190 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
194 * All metadata updates are logged, which means that we just have
195 * to flush the log up to the latest LSN that touched the inode.
197 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
198 if (xfs_ipincount(ip
)) {
200 (ip
->i_itemp
->ili_fields
& ~XFS_ILOG_TIMESTAMP
))
201 lsn
= ip
->i_itemp
->ili_last_lsn
;
203 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
206 error
= _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
209 * If we only have a single device, and the log force about was
210 * a no-op we might have to flush the data device cache here.
211 * This can only happen for fdatasync/O_DSYNC if we were overwriting
212 * an already allocated file and thus do not have any metadata to
215 if ((mp
->m_flags
& XFS_MOUNT_BARRIER
) &&
216 mp
->m_logdev_targp
== mp
->m_ddev_targp
&&
217 !XFS_IS_REALTIME_INODE(ip
) &&
219 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
227 const struct iovec
*iovp
,
228 unsigned long nr_segs
,
231 struct file
*file
= iocb
->ki_filp
;
232 struct inode
*inode
= file
->f_mapping
->host
;
233 struct xfs_inode
*ip
= XFS_I(inode
);
234 struct xfs_mount
*mp
= ip
->i_mount
;
240 XFS_STATS_INC(xs_read_calls
);
242 BUG_ON(iocb
->ki_pos
!= pos
);
244 if (unlikely(file
->f_flags
& O_DIRECT
))
245 ioflags
|= IO_ISDIRECT
;
246 if (file
->f_mode
& FMODE_NOCMTIME
)
249 ret
= generic_segment_checks(iovp
, &nr_segs
, &size
, VERIFY_WRITE
);
253 if (unlikely(ioflags
& IO_ISDIRECT
)) {
254 xfs_buftarg_t
*target
=
255 XFS_IS_REALTIME_INODE(ip
) ?
256 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
257 if ((iocb
->ki_pos
& target
->bt_smask
) ||
258 (size
& target
->bt_smask
)) {
259 if (iocb
->ki_pos
== i_size_read(inode
))
261 return -XFS_ERROR(EINVAL
);
265 n
= mp
->m_super
->s_maxbytes
- iocb
->ki_pos
;
266 if (n
<= 0 || size
== 0)
272 if (XFS_FORCED_SHUTDOWN(mp
))
276 * Locking is a bit tricky here. If we take an exclusive lock
277 * for direct IO, we effectively serialise all new concurrent
278 * read IO to this file and block it behind IO that is currently in
279 * progress because IO in progress holds the IO lock shared. We only
280 * need to hold the lock exclusive to blow away the page cache, so
281 * only take lock exclusively if the page cache needs invalidation.
282 * This allows the normal direct IO case of no page cache pages to
283 * proceeed concurrently without serialisation.
285 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
286 if ((ioflags
& IO_ISDIRECT
) && inode
->i_mapping
->nrpages
) {
287 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
288 xfs_rw_ilock(ip
, XFS_IOLOCK_EXCL
);
290 if (inode
->i_mapping
->nrpages
) {
291 ret
= -xfs_flushinval_pages(ip
,
292 (iocb
->ki_pos
& PAGE_CACHE_MASK
),
293 -1, FI_REMAPF_LOCKED
);
295 xfs_rw_iunlock(ip
, XFS_IOLOCK_EXCL
);
299 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
302 trace_xfs_file_read(ip
, size
, iocb
->ki_pos
, ioflags
);
304 ret
= generic_file_aio_read(iocb
, iovp
, nr_segs
, iocb
->ki_pos
);
306 XFS_STATS_ADD(xs_read_bytes
, ret
);
308 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
313 xfs_file_splice_read(
316 struct pipe_inode_info
*pipe
,
320 struct xfs_inode
*ip
= XFS_I(infilp
->f_mapping
->host
);
324 XFS_STATS_INC(xs_read_calls
);
326 if (infilp
->f_mode
& FMODE_NOCMTIME
)
329 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
332 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
334 trace_xfs_file_splice_read(ip
, count
, *ppos
, ioflags
);
336 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
338 XFS_STATS_ADD(xs_read_bytes
, ret
);
340 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
345 * xfs_file_splice_write() does not use xfs_rw_ilock() because
346 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
347 * couuld cause lock inversions between the aio_write path and the splice path
348 * if someone is doing concurrent splice(2) based writes and write(2) based
349 * writes to the same inode. The only real way to fix this is to re-implement
350 * the generic code here with correct locking orders.
353 xfs_file_splice_write(
354 struct pipe_inode_info
*pipe
,
355 struct file
*outfilp
,
360 struct inode
*inode
= outfilp
->f_mapping
->host
;
361 struct xfs_inode
*ip
= XFS_I(inode
);
365 XFS_STATS_INC(xs_write_calls
);
367 if (outfilp
->f_mode
& FMODE_NOCMTIME
)
370 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
373 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
375 trace_xfs_file_splice_write(ip
, count
, *ppos
, ioflags
);
377 ret
= generic_file_splice_write(pipe
, outfilp
, ppos
, count
, flags
);
379 XFS_STATS_ADD(xs_write_bytes
, ret
);
381 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
386 * This routine is called to handle zeroing any space in the last block of the
387 * file that is beyond the EOF. We do this since the size is being increased
388 * without writing anything to that block and we don't want to read the
389 * garbage on the disk.
391 STATIC
int /* error (positive) */
393 struct xfs_inode
*ip
,
397 struct xfs_mount
*mp
= ip
->i_mount
;
398 xfs_fileoff_t last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
399 int zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
403 struct xfs_bmbt_irec imap
;
405 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
406 error
= xfs_bmapi_read(ip
, last_fsb
, 1, &imap
, &nimaps
, 0);
407 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
414 * If the block underlying isize is just a hole, then there
415 * is nothing to zero.
417 if (imap
.br_startblock
== HOLESTARTBLOCK
)
420 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
421 if (isize
+ zero_len
> offset
)
422 zero_len
= offset
- isize
;
423 return xfs_iozero(ip
, isize
, zero_len
);
427 * Zero any on disk space between the current EOF and the new, larger EOF.
429 * This handles the normal case of zeroing the remainder of the last block in
430 * the file and the unusual case of zeroing blocks out beyond the size of the
431 * file. This second case only happens with fixed size extents and when the
432 * system crashes before the inode size was updated but after blocks were
435 * Expects the iolock to be held exclusive, and will take the ilock internally.
437 int /* error (positive) */
439 struct xfs_inode
*ip
,
440 xfs_off_t offset
, /* starting I/O offset */
441 xfs_fsize_t isize
) /* current inode size */
443 struct xfs_mount
*mp
= ip
->i_mount
;
444 xfs_fileoff_t start_zero_fsb
;
445 xfs_fileoff_t end_zero_fsb
;
446 xfs_fileoff_t zero_count_fsb
;
447 xfs_fileoff_t last_fsb
;
448 xfs_fileoff_t zero_off
;
449 xfs_fsize_t zero_len
;
452 struct xfs_bmbt_irec imap
;
454 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
455 ASSERT(offset
> isize
);
458 * First handle zeroing the block on which isize resides.
460 * We only zero a part of that block so it is handled specially.
462 if (XFS_B_FSB_OFFSET(mp
, isize
) != 0) {
463 error
= xfs_zero_last_block(ip
, offset
, isize
);
469 * Calculate the range between the new size and the old where blocks
470 * needing to be zeroed may exist.
472 * To get the block where the last byte in the file currently resides,
473 * we need to subtract one from the size and truncate back to a block
474 * boundary. We subtract 1 in case the size is exactly on a block
477 last_fsb
= isize
? XFS_B_TO_FSBT(mp
, isize
- 1) : (xfs_fileoff_t
)-1;
478 start_zero_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
479 end_zero_fsb
= XFS_B_TO_FSBT(mp
, offset
- 1);
480 ASSERT((xfs_sfiloff_t
)last_fsb
< (xfs_sfiloff_t
)start_zero_fsb
);
481 if (last_fsb
== end_zero_fsb
) {
483 * The size was only incremented on its last block.
484 * We took care of that above, so just return.
489 ASSERT(start_zero_fsb
<= end_zero_fsb
);
490 while (start_zero_fsb
<= end_zero_fsb
) {
492 zero_count_fsb
= end_zero_fsb
- start_zero_fsb
+ 1;
494 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
495 error
= xfs_bmapi_read(ip
, start_zero_fsb
, zero_count_fsb
,
497 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
503 if (imap
.br_state
== XFS_EXT_UNWRITTEN
||
504 imap
.br_startblock
== HOLESTARTBLOCK
) {
505 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
506 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
511 * There are blocks we need to zero.
513 zero_off
= XFS_FSB_TO_B(mp
, start_zero_fsb
);
514 zero_len
= XFS_FSB_TO_B(mp
, imap
.br_blockcount
);
516 if ((zero_off
+ zero_len
) > offset
)
517 zero_len
= offset
- zero_off
;
519 error
= xfs_iozero(ip
, zero_off
, zero_len
);
523 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
524 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
531 * Common pre-write limit and setup checks.
533 * Called with the iolocked held either shared and exclusive according to
534 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
535 * if called for a direct write beyond i_size.
538 xfs_file_aio_write_checks(
544 struct inode
*inode
= file
->f_mapping
->host
;
545 struct xfs_inode
*ip
= XFS_I(inode
);
549 error
= generic_write_checks(file
, pos
, count
, S_ISBLK(inode
->i_mode
));
554 * If the offset is beyond the size of the file, we need to zero any
555 * blocks that fall between the existing EOF and the start of this
556 * write. If zeroing is needed and we are currently holding the
557 * iolock shared, we need to update it to exclusive which implies
558 * having to redo all checks before.
560 if (*pos
> i_size_read(inode
)) {
561 if (*iolock
== XFS_IOLOCK_SHARED
) {
562 xfs_rw_iunlock(ip
, *iolock
);
563 *iolock
= XFS_IOLOCK_EXCL
;
564 xfs_rw_ilock(ip
, *iolock
);
567 error
= -xfs_zero_eof(ip
, *pos
, i_size_read(inode
));
573 * Updating the timestamps will grab the ilock again from
574 * xfs_fs_dirty_inode, so we have to call it after dropping the
575 * lock above. Eventually we should look into a way to avoid
576 * the pointless lock roundtrip.
578 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
579 error
= file_update_time(file
);
585 * If we're writing the file then make sure to clear the setuid and
586 * setgid bits if the process is not being run by root. This keeps
587 * people from modifying setuid and setgid binaries.
589 return file_remove_suid(file
);
593 * xfs_file_dio_aio_write - handle direct IO writes
595 * Lock the inode appropriately to prepare for and issue a direct IO write.
596 * By separating it from the buffered write path we remove all the tricky to
597 * follow locking changes and looping.
599 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
600 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
601 * pages are flushed out.
603 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
604 * allowing them to be done in parallel with reads and other direct IO writes.
605 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
606 * needs to do sub-block zeroing and that requires serialisation against other
607 * direct IOs to the same block. In this case we need to serialise the
608 * submission of the unaligned IOs so that we don't get racing block zeroing in
609 * the dio layer. To avoid the problem with aio, we also need to wait for
610 * outstanding IOs to complete so that unwritten extent conversion is completed
611 * before we try to map the overlapping block. This is currently implemented by
612 * hitting it with a big hammer (i.e. inode_dio_wait()).
614 * Returns with locks held indicated by @iolock and errors indicated by
615 * negative return values.
618 xfs_file_dio_aio_write(
620 const struct iovec
*iovp
,
621 unsigned long nr_segs
,
625 struct file
*file
= iocb
->ki_filp
;
626 struct address_space
*mapping
= file
->f_mapping
;
627 struct inode
*inode
= mapping
->host
;
628 struct xfs_inode
*ip
= XFS_I(inode
);
629 struct xfs_mount
*mp
= ip
->i_mount
;
631 size_t count
= ocount
;
632 int unaligned_io
= 0;
634 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
635 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
637 if ((pos
& target
->bt_smask
) || (count
& target
->bt_smask
))
638 return -XFS_ERROR(EINVAL
);
640 if ((pos
& mp
->m_blockmask
) || ((pos
+ count
) & mp
->m_blockmask
))
644 * We don't need to take an exclusive lock unless there page cache needs
645 * to be invalidated or unaligned IO is being executed. We don't need to
646 * consider the EOF extension case here because
647 * xfs_file_aio_write_checks() will relock the inode as necessary for
648 * EOF zeroing cases and fill out the new inode size as appropriate.
650 if (unaligned_io
|| mapping
->nrpages
)
651 iolock
= XFS_IOLOCK_EXCL
;
653 iolock
= XFS_IOLOCK_SHARED
;
654 xfs_rw_ilock(ip
, iolock
);
657 * Recheck if there are cached pages that need invalidate after we got
658 * the iolock to protect against other threads adding new pages while
659 * we were waiting for the iolock.
661 if (mapping
->nrpages
&& iolock
== XFS_IOLOCK_SHARED
) {
662 xfs_rw_iunlock(ip
, iolock
);
663 iolock
= XFS_IOLOCK_EXCL
;
664 xfs_rw_ilock(ip
, iolock
);
667 ret
= xfs_file_aio_write_checks(file
, &pos
, &count
, &iolock
);
671 if (mapping
->nrpages
) {
672 ret
= -xfs_flushinval_pages(ip
, (pos
& PAGE_CACHE_MASK
), -1,
679 * If we are doing unaligned IO, wait for all other IO to drain,
680 * otherwise demote the lock if we had to flush cached pages
683 inode_dio_wait(inode
);
684 else if (iolock
== XFS_IOLOCK_EXCL
) {
685 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
686 iolock
= XFS_IOLOCK_SHARED
;
689 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
, 0);
690 ret
= generic_file_direct_write(iocb
, iovp
,
691 &nr_segs
, pos
, &iocb
->ki_pos
, count
, ocount
);
694 xfs_rw_iunlock(ip
, iolock
);
696 /* No fallback to buffered IO on errors for XFS. */
697 ASSERT(ret
< 0 || ret
== count
);
702 xfs_file_buffered_aio_write(
704 const struct iovec
*iovp
,
705 unsigned long nr_segs
,
709 struct file
*file
= iocb
->ki_filp
;
710 struct address_space
*mapping
= file
->f_mapping
;
711 struct inode
*inode
= mapping
->host
;
712 struct xfs_inode
*ip
= XFS_I(inode
);
715 int iolock
= XFS_IOLOCK_EXCL
;
716 size_t count
= ocount
;
718 xfs_rw_ilock(ip
, iolock
);
720 ret
= xfs_file_aio_write_checks(file
, &pos
, &count
, &iolock
);
724 /* We can write back this queue in page reclaim */
725 current
->backing_dev_info
= mapping
->backing_dev_info
;
728 trace_xfs_file_buffered_write(ip
, count
, iocb
->ki_pos
, 0);
729 ret
= generic_file_buffered_write(iocb
, iovp
, nr_segs
,
730 pos
, &iocb
->ki_pos
, count
, ret
);
732 * if we just got an ENOSPC, flush the inode now we aren't holding any
733 * page locks and retry *once*
735 if (ret
== -ENOSPC
&& !enospc
) {
737 ret
= -xfs_flush_pages(ip
, 0, -1, 0, FI_NONE
);
742 current
->backing_dev_info
= NULL
;
744 xfs_rw_iunlock(ip
, iolock
);
751 const struct iovec
*iovp
,
752 unsigned long nr_segs
,
755 struct file
*file
= iocb
->ki_filp
;
756 struct address_space
*mapping
= file
->f_mapping
;
757 struct inode
*inode
= mapping
->host
;
758 struct xfs_inode
*ip
= XFS_I(inode
);
762 XFS_STATS_INC(xs_write_calls
);
764 BUG_ON(iocb
->ki_pos
!= pos
);
766 ret
= generic_segment_checks(iovp
, &nr_segs
, &ocount
, VERIFY_READ
);
773 sb_start_write(inode
->i_sb
);
775 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
780 if (unlikely(file
->f_flags
& O_DIRECT
))
781 ret
= xfs_file_dio_aio_write(iocb
, iovp
, nr_segs
, pos
, ocount
);
783 ret
= xfs_file_buffered_aio_write(iocb
, iovp
, nr_segs
, pos
,
789 XFS_STATS_ADD(xs_write_bytes
, ret
);
791 /* Handle various SYNC-type writes */
792 err
= generic_write_sync(file
, pos
, ret
);
798 sb_end_write(inode
->i_sb
);
809 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
813 xfs_inode_t
*ip
= XFS_I(inode
);
814 int cmd
= XFS_IOC_RESVSP
;
815 int attr_flags
= XFS_ATTR_NOLOCK
;
817 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
824 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
826 if (mode
& FALLOC_FL_PUNCH_HOLE
)
827 cmd
= XFS_IOC_UNRESVSP
;
829 /* check the new inode size is valid before allocating */
830 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
831 offset
+ len
> i_size_read(inode
)) {
832 new_size
= offset
+ len
;
833 error
= inode_newsize_ok(inode
, new_size
);
838 if (file
->f_flags
& O_DSYNC
)
839 attr_flags
|= XFS_ATTR_SYNC
;
841 error
= -xfs_change_file_space(ip
, cmd
, &bf
, 0, attr_flags
);
845 /* Change file size if needed */
849 iattr
.ia_valid
= ATTR_SIZE
;
850 iattr
.ia_size
= new_size
;
851 error
= -xfs_setattr_size(ip
, &iattr
, XFS_ATTR_NOLOCK
);
855 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
865 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
867 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
877 struct xfs_inode
*ip
= XFS_I(inode
);
881 error
= xfs_file_open(inode
, file
);
886 * If there are any blocks, read-ahead block 0 as we're almost
887 * certain to have the next operation be a read there.
889 mode
= xfs_ilock_map_shared(ip
);
890 if (ip
->i_d
.di_nextents
> 0)
891 xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
892 xfs_iunlock(ip
, mode
);
901 return -xfs_release(XFS_I(inode
));
910 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
911 xfs_inode_t
*ip
= XFS_I(inode
);
916 * The Linux API doesn't pass down the total size of the buffer
917 * we read into down to the filesystem. With the filldir concept
918 * it's not needed for correct information, but the XFS dir2 leaf
919 * code wants an estimate of the buffer size to calculate it's
920 * readahead window and size the buffers used for mapping to
923 * Try to give it an estimate that's good enough, maybe at some
924 * point we can change the ->readdir prototype to include the
925 * buffer size. For now we use the current glibc buffer size.
927 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
929 error
= xfs_readdir(ip
, dirent
, bufsize
,
930 (xfs_off_t
*)&filp
->f_pos
, filldir
);
939 struct vm_area_struct
*vma
)
941 vma
->vm_ops
= &xfs_file_vm_ops
;
942 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
949 * mmap()d file has taken write protection fault and is being made
950 * writable. We can set the page state up correctly for a writable
951 * page, which means we can do correct delalloc accounting (ENOSPC
952 * checking!) and unwritten extent mapping.
956 struct vm_area_struct
*vma
,
957 struct vm_fault
*vmf
)
959 return block_page_mkwrite(vma
, vmf
, xfs_get_blocks
);
968 struct inode
*inode
= file
->f_mapping
->host
;
969 struct xfs_inode
*ip
= XFS_I(inode
);
970 struct xfs_mount
*mp
= ip
->i_mount
;
971 struct xfs_bmbt_irec map
[2];
973 loff_t
uninitialized_var(offset
);
980 lock
= xfs_ilock_map_shared(ip
);
982 isize
= i_size_read(inode
);
983 if (start
>= isize
) {
988 fsbno
= XFS_B_TO_FSBT(mp
, start
);
991 * Try to read extents from the first block indicated
992 * by fsbno to the end block of the file.
994 end
= XFS_B_TO_FSB(mp
, isize
);
996 error
= xfs_bmapi_read(ip
, fsbno
, end
- fsbno
, map
, &nmap
,
1002 * Treat unwritten extent as data extent since it might
1003 * contains dirty data in page cache.
1005 if (map
[0].br_startblock
!= HOLESTARTBLOCK
) {
1006 offset
= max_t(loff_t
, start
,
1007 XFS_FSB_TO_B(mp
, map
[0].br_startoff
));
1014 offset
= max_t(loff_t
, start
,
1015 XFS_FSB_TO_B(mp
, map
[1].br_startoff
));
1018 if (offset
!= file
->f_pos
)
1019 file
->f_pos
= offset
;
1022 xfs_iunlock_map_shared(ip
, lock
);
1035 struct inode
*inode
= file
->f_mapping
->host
;
1036 struct xfs_inode
*ip
= XFS_I(inode
);
1037 struct xfs_mount
*mp
= ip
->i_mount
;
1038 loff_t
uninitialized_var(offset
);
1041 xfs_fileoff_t fsbno
;
1045 if (XFS_FORCED_SHUTDOWN(mp
))
1046 return -XFS_ERROR(EIO
);
1048 lock
= xfs_ilock_map_shared(ip
);
1050 isize
= i_size_read(inode
);
1051 if (start
>= isize
) {
1056 fsbno
= XFS_B_TO_FSBT(mp
, start
);
1057 error
= xfs_bmap_first_unused(NULL
, ip
, 1, &fsbno
, XFS_DATA_FORK
);
1061 holeoff
= XFS_FSB_TO_B(mp
, fsbno
);
1062 if (holeoff
<= start
)
1066 * xfs_bmap_first_unused() could return a value bigger than
1067 * isize if there are no more holes past the supplied offset.
1069 offset
= min_t(loff_t
, holeoff
, isize
);
1072 if (offset
!= file
->f_pos
)
1073 file
->f_pos
= offset
;
1076 xfs_iunlock_map_shared(ip
, lock
);
1093 return generic_file_llseek(file
, offset
, origin
);
1095 return xfs_seek_data(file
, offset
, origin
);
1097 return xfs_seek_hole(file
, offset
, origin
);
1103 const struct file_operations xfs_file_operations
= {
1104 .llseek
= xfs_file_llseek
,
1105 .read
= do_sync_read
,
1106 .write
= do_sync_write
,
1107 .aio_read
= xfs_file_aio_read
,
1108 .aio_write
= xfs_file_aio_write
,
1109 .splice_read
= xfs_file_splice_read
,
1110 .splice_write
= xfs_file_splice_write
,
1111 .unlocked_ioctl
= xfs_file_ioctl
,
1112 #ifdef CONFIG_COMPAT
1113 .compat_ioctl
= xfs_file_compat_ioctl
,
1115 .mmap
= xfs_file_mmap
,
1116 .open
= xfs_file_open
,
1117 .release
= xfs_file_release
,
1118 .fsync
= xfs_file_fsync
,
1119 .fallocate
= xfs_file_fallocate
,
1122 const struct file_operations xfs_dir_file_operations
= {
1123 .open
= xfs_dir_open
,
1124 .read
= generic_read_dir
,
1125 .readdir
= xfs_file_readdir
,
1126 .llseek
= generic_file_llseek
,
1127 .unlocked_ioctl
= xfs_file_ioctl
,
1128 #ifdef CONFIG_COMPAT
1129 .compat_ioctl
= xfs_file_compat_ioctl
,
1131 .fsync
= xfs_dir_fsync
,
1134 static const struct vm_operations_struct xfs_file_vm_ops
= {
1135 .fault
= filemap_fault
,
1136 .page_mkwrite
= xfs_vm_page_mkwrite
,