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
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_bmap_util.h"
34 #include "xfs_error.h"
36 #include "xfs_dir2_priv.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_trace.h"
40 #include "xfs_dinode.h"
42 #include <linux/aio.h>
43 #include <linux/dcache.h>
44 #include <linux/falloc.h>
45 #include <linux/pagevec.h>
47 static const struct vm_operations_struct xfs_file_vm_ops
;
50 * Locking primitives for read and write IO paths to ensure we consistently use
51 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
58 if (type
& XFS_IOLOCK_EXCL
)
59 mutex_lock(&VFS_I(ip
)->i_mutex
);
68 xfs_iunlock(ip
, type
);
69 if (type
& XFS_IOLOCK_EXCL
)
70 mutex_unlock(&VFS_I(ip
)->i_mutex
);
78 xfs_ilock_demote(ip
, type
);
79 if (type
& XFS_IOLOCK_EXCL
)
80 mutex_unlock(&VFS_I(ip
)->i_mutex
);
86 * xfs_iozero clears the specified range of buffer supplied,
87 * and marks all the affected blocks as valid and modified. If
88 * an affected block is not allocated, it will be allocated. If
89 * an affected block is not completely overwritten, and is not
90 * valid before the operation, it will be read from disk before
91 * being partially zeroed.
95 struct xfs_inode
*ip
, /* inode */
96 loff_t pos
, /* offset in file */
97 size_t count
) /* size of data to zero */
100 struct address_space
*mapping
;
103 mapping
= VFS_I(ip
)->i_mapping
;
105 unsigned offset
, bytes
;
108 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
109 bytes
= PAGE_CACHE_SIZE
- offset
;
113 status
= pagecache_write_begin(NULL
, mapping
, pos
, bytes
,
114 AOP_FLAG_UNINTERRUPTIBLE
,
119 zero_user(page
, offset
, bytes
);
121 status
= pagecache_write_end(NULL
, mapping
, pos
, bytes
, bytes
,
123 WARN_ON(status
<= 0); /* can't return less than zero! */
133 * Fsync operations on directories are much simpler than on regular files,
134 * as there is no file data to flush, and thus also no need for explicit
135 * cache flush operations, and there are no non-transaction metadata updates
136 * on directories either.
145 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
146 struct xfs_mount
*mp
= ip
->i_mount
;
149 trace_xfs_dir_fsync(ip
);
151 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
152 if (xfs_ipincount(ip
))
153 lsn
= ip
->i_itemp
->ili_last_lsn
;
154 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
158 return _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
168 struct inode
*inode
= file
->f_mapping
->host
;
169 struct xfs_inode
*ip
= XFS_I(inode
);
170 struct xfs_mount
*mp
= ip
->i_mount
;
175 trace_xfs_file_fsync(ip
);
177 error
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
181 if (XFS_FORCED_SHUTDOWN(mp
))
182 return -XFS_ERROR(EIO
);
184 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
186 if (mp
->m_flags
& XFS_MOUNT_BARRIER
) {
188 * If we have an RT and/or log subvolume we need to make sure
189 * to flush the write cache the device used for file data
190 * first. This is to ensure newly written file data make
191 * it to disk before logging the new inode size in case of
192 * an extending write.
194 if (XFS_IS_REALTIME_INODE(ip
))
195 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
196 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
197 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
201 * All metadata updates are logged, which means that we just have
202 * to flush the log up to the latest LSN that touched the inode.
204 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
205 if (xfs_ipincount(ip
)) {
207 (ip
->i_itemp
->ili_fields
& ~XFS_ILOG_TIMESTAMP
))
208 lsn
= ip
->i_itemp
->ili_last_lsn
;
210 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
213 error
= _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
216 * If we only have a single device, and the log force about was
217 * a no-op we might have to flush the data device cache here.
218 * This can only happen for fdatasync/O_DSYNC if we were overwriting
219 * an already allocated file and thus do not have any metadata to
222 if ((mp
->m_flags
& XFS_MOUNT_BARRIER
) &&
223 mp
->m_logdev_targp
== mp
->m_ddev_targp
&&
224 !XFS_IS_REALTIME_INODE(ip
) &&
226 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
234 const struct iovec
*iovp
,
235 unsigned long nr_segs
,
238 struct file
*file
= iocb
->ki_filp
;
239 struct inode
*inode
= file
->f_mapping
->host
;
240 struct xfs_inode
*ip
= XFS_I(inode
);
241 struct xfs_mount
*mp
= ip
->i_mount
;
247 XFS_STATS_INC(xs_read_calls
);
249 BUG_ON(iocb
->ki_pos
!= pos
);
251 if (unlikely(file
->f_flags
& O_DIRECT
))
252 ioflags
|= IO_ISDIRECT
;
253 if (file
->f_mode
& FMODE_NOCMTIME
)
256 ret
= generic_segment_checks(iovp
, &nr_segs
, &size
, VERIFY_WRITE
);
260 if (unlikely(ioflags
& IO_ISDIRECT
)) {
261 xfs_buftarg_t
*target
=
262 XFS_IS_REALTIME_INODE(ip
) ?
263 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
264 /* DIO must be aligned to device logical sector size */
265 if ((pos
| size
) & target
->bt_logical_sectormask
) {
266 if (pos
== i_size_read(inode
))
268 return -XFS_ERROR(EINVAL
);
272 n
= mp
->m_super
->s_maxbytes
- pos
;
273 if (n
<= 0 || size
== 0)
279 if (XFS_FORCED_SHUTDOWN(mp
))
283 * Locking is a bit tricky here. If we take an exclusive lock
284 * for direct IO, we effectively serialise all new concurrent
285 * read IO to this file and block it behind IO that is currently in
286 * progress because IO in progress holds the IO lock shared. We only
287 * need to hold the lock exclusive to blow away the page cache, so
288 * only take lock exclusively if the page cache needs invalidation.
289 * This allows the normal direct IO case of no page cache pages to
290 * proceeed concurrently without serialisation.
292 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
293 if ((ioflags
& IO_ISDIRECT
) && inode
->i_mapping
->nrpages
) {
294 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
295 xfs_rw_ilock(ip
, XFS_IOLOCK_EXCL
);
297 if (inode
->i_mapping
->nrpages
) {
298 ret
= -filemap_write_and_wait_range(
299 VFS_I(ip
)->i_mapping
,
302 xfs_rw_iunlock(ip
, XFS_IOLOCK_EXCL
);
305 truncate_pagecache_range(VFS_I(ip
), pos
, -1);
307 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
310 trace_xfs_file_read(ip
, size
, pos
, ioflags
);
312 ret
= generic_file_aio_read(iocb
, iovp
, nr_segs
, pos
);
314 XFS_STATS_ADD(xs_read_bytes
, ret
);
316 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
321 xfs_file_splice_read(
324 struct pipe_inode_info
*pipe
,
328 struct xfs_inode
*ip
= XFS_I(infilp
->f_mapping
->host
);
332 XFS_STATS_INC(xs_read_calls
);
334 if (infilp
->f_mode
& FMODE_NOCMTIME
)
337 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
340 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
342 trace_xfs_file_splice_read(ip
, count
, *ppos
, ioflags
);
344 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
346 XFS_STATS_ADD(xs_read_bytes
, ret
);
348 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
353 * xfs_file_splice_write() does not use xfs_rw_ilock() because
354 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
355 * couuld cause lock inversions between the aio_write path and the splice path
356 * if someone is doing concurrent splice(2) based writes and write(2) based
357 * writes to the same inode. The only real way to fix this is to re-implement
358 * the generic code here with correct locking orders.
361 xfs_file_splice_write(
362 struct pipe_inode_info
*pipe
,
363 struct file
*outfilp
,
368 struct inode
*inode
= outfilp
->f_mapping
->host
;
369 struct xfs_inode
*ip
= XFS_I(inode
);
373 XFS_STATS_INC(xs_write_calls
);
375 if (outfilp
->f_mode
& FMODE_NOCMTIME
)
378 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
381 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
383 trace_xfs_file_splice_write(ip
, count
, *ppos
, ioflags
);
385 ret
= generic_file_splice_write(pipe
, outfilp
, ppos
, count
, flags
);
387 XFS_STATS_ADD(xs_write_bytes
, ret
);
389 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
394 * This routine is called to handle zeroing any space in the last block of the
395 * file that is beyond the EOF. We do this since the size is being increased
396 * without writing anything to that block and we don't want to read the
397 * garbage on the disk.
399 STATIC
int /* error (positive) */
401 struct xfs_inode
*ip
,
405 struct xfs_mount
*mp
= ip
->i_mount
;
406 xfs_fileoff_t last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
407 int zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
411 struct xfs_bmbt_irec imap
;
413 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
414 error
= xfs_bmapi_read(ip
, last_fsb
, 1, &imap
, &nimaps
, 0);
415 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
422 * If the block underlying isize is just a hole, then there
423 * is nothing to zero.
425 if (imap
.br_startblock
== HOLESTARTBLOCK
)
428 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
429 if (isize
+ zero_len
> offset
)
430 zero_len
= offset
- isize
;
431 return xfs_iozero(ip
, isize
, zero_len
);
435 * Zero any on disk space between the current EOF and the new, larger EOF.
437 * This handles the normal case of zeroing the remainder of the last block in
438 * the file and the unusual case of zeroing blocks out beyond the size of the
439 * file. This second case only happens with fixed size extents and when the
440 * system crashes before the inode size was updated but after blocks were
443 * Expects the iolock to be held exclusive, and will take the ilock internally.
445 int /* error (positive) */
447 struct xfs_inode
*ip
,
448 xfs_off_t offset
, /* starting I/O offset */
449 xfs_fsize_t isize
) /* current inode size */
451 struct xfs_mount
*mp
= ip
->i_mount
;
452 xfs_fileoff_t start_zero_fsb
;
453 xfs_fileoff_t end_zero_fsb
;
454 xfs_fileoff_t zero_count_fsb
;
455 xfs_fileoff_t last_fsb
;
456 xfs_fileoff_t zero_off
;
457 xfs_fsize_t zero_len
;
460 struct xfs_bmbt_irec imap
;
462 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
463 ASSERT(offset
> isize
);
466 * First handle zeroing the block on which isize resides.
468 * We only zero a part of that block so it is handled specially.
470 if (XFS_B_FSB_OFFSET(mp
, isize
) != 0) {
471 error
= xfs_zero_last_block(ip
, offset
, isize
);
477 * Calculate the range between the new size and the old where blocks
478 * needing to be zeroed may exist.
480 * To get the block where the last byte in the file currently resides,
481 * we need to subtract one from the size and truncate back to a block
482 * boundary. We subtract 1 in case the size is exactly on a block
485 last_fsb
= isize
? XFS_B_TO_FSBT(mp
, isize
- 1) : (xfs_fileoff_t
)-1;
486 start_zero_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
487 end_zero_fsb
= XFS_B_TO_FSBT(mp
, offset
- 1);
488 ASSERT((xfs_sfiloff_t
)last_fsb
< (xfs_sfiloff_t
)start_zero_fsb
);
489 if (last_fsb
== end_zero_fsb
) {
491 * The size was only incremented on its last block.
492 * We took care of that above, so just return.
497 ASSERT(start_zero_fsb
<= end_zero_fsb
);
498 while (start_zero_fsb
<= end_zero_fsb
) {
500 zero_count_fsb
= end_zero_fsb
- start_zero_fsb
+ 1;
502 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
503 error
= xfs_bmapi_read(ip
, start_zero_fsb
, zero_count_fsb
,
505 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
511 if (imap
.br_state
== XFS_EXT_UNWRITTEN
||
512 imap
.br_startblock
== HOLESTARTBLOCK
) {
513 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
514 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
519 * There are blocks we need to zero.
521 zero_off
= XFS_FSB_TO_B(mp
, start_zero_fsb
);
522 zero_len
= XFS_FSB_TO_B(mp
, imap
.br_blockcount
);
524 if ((zero_off
+ zero_len
) > offset
)
525 zero_len
= offset
- zero_off
;
527 error
= xfs_iozero(ip
, zero_off
, zero_len
);
531 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
532 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
539 * Common pre-write limit and setup checks.
541 * Called with the iolocked held either shared and exclusive according to
542 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
543 * if called for a direct write beyond i_size.
546 xfs_file_aio_write_checks(
552 struct inode
*inode
= file
->f_mapping
->host
;
553 struct xfs_inode
*ip
= XFS_I(inode
);
557 error
= generic_write_checks(file
, pos
, count
, S_ISBLK(inode
->i_mode
));
562 * If the offset is beyond the size of the file, we need to zero any
563 * blocks that fall between the existing EOF and the start of this
564 * write. If zeroing is needed and we are currently holding the
565 * iolock shared, we need to update it to exclusive which implies
566 * having to redo all checks before.
568 if (*pos
> i_size_read(inode
)) {
569 if (*iolock
== XFS_IOLOCK_SHARED
) {
570 xfs_rw_iunlock(ip
, *iolock
);
571 *iolock
= XFS_IOLOCK_EXCL
;
572 xfs_rw_ilock(ip
, *iolock
);
575 error
= -xfs_zero_eof(ip
, *pos
, i_size_read(inode
));
581 * Updating the timestamps will grab the ilock again from
582 * xfs_fs_dirty_inode, so we have to call it after dropping the
583 * lock above. Eventually we should look into a way to avoid
584 * the pointless lock roundtrip.
586 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
587 error
= file_update_time(file
);
593 * If we're writing the file then make sure to clear the setuid and
594 * setgid bits if the process is not being run by root. This keeps
595 * people from modifying setuid and setgid binaries.
597 return file_remove_suid(file
);
601 * xfs_file_dio_aio_write - handle direct IO writes
603 * Lock the inode appropriately to prepare for and issue a direct IO write.
604 * By separating it from the buffered write path we remove all the tricky to
605 * follow locking changes and looping.
607 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
608 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
609 * pages are flushed out.
611 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
612 * allowing them to be done in parallel with reads and other direct IO writes.
613 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
614 * needs to do sub-block zeroing and that requires serialisation against other
615 * direct IOs to the same block. In this case we need to serialise the
616 * submission of the unaligned IOs so that we don't get racing block zeroing in
617 * the dio layer. To avoid the problem with aio, we also need to wait for
618 * outstanding IOs to complete so that unwritten extent conversion is completed
619 * before we try to map the overlapping block. This is currently implemented by
620 * hitting it with a big hammer (i.e. inode_dio_wait()).
622 * Returns with locks held indicated by @iolock and errors indicated by
623 * negative return values.
626 xfs_file_dio_aio_write(
628 const struct iovec
*iovp
,
629 unsigned long nr_segs
,
633 struct file
*file
= iocb
->ki_filp
;
634 struct address_space
*mapping
= file
->f_mapping
;
635 struct inode
*inode
= mapping
->host
;
636 struct xfs_inode
*ip
= XFS_I(inode
);
637 struct xfs_mount
*mp
= ip
->i_mount
;
639 size_t count
= ocount
;
640 int unaligned_io
= 0;
642 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
643 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
645 /* DIO must be aligned to device logical sector size */
646 if ((pos
| count
) & target
->bt_logical_sectormask
)
647 return -XFS_ERROR(EINVAL
);
649 /* "unaligned" here means not aligned to a filesystem block */
650 if ((pos
& mp
->m_blockmask
) || ((pos
+ count
) & mp
->m_blockmask
))
654 * We don't need to take an exclusive lock unless there page cache needs
655 * to be invalidated or unaligned IO is being executed. We don't need to
656 * consider the EOF extension case here because
657 * xfs_file_aio_write_checks() will relock the inode as necessary for
658 * EOF zeroing cases and fill out the new inode size as appropriate.
660 if (unaligned_io
|| mapping
->nrpages
)
661 iolock
= XFS_IOLOCK_EXCL
;
663 iolock
= XFS_IOLOCK_SHARED
;
664 xfs_rw_ilock(ip
, iolock
);
667 * Recheck if there are cached pages that need invalidate after we got
668 * the iolock to protect against other threads adding new pages while
669 * we were waiting for the iolock.
671 if (mapping
->nrpages
&& iolock
== XFS_IOLOCK_SHARED
) {
672 xfs_rw_iunlock(ip
, iolock
);
673 iolock
= XFS_IOLOCK_EXCL
;
674 xfs_rw_ilock(ip
, iolock
);
677 ret
= xfs_file_aio_write_checks(file
, &pos
, &count
, &iolock
);
681 if (mapping
->nrpages
) {
682 ret
= -filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
686 truncate_pagecache_range(VFS_I(ip
), pos
, -1);
690 * If we are doing unaligned IO, wait for all other IO to drain,
691 * otherwise demote the lock if we had to flush cached pages
694 inode_dio_wait(inode
);
695 else if (iolock
== XFS_IOLOCK_EXCL
) {
696 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
697 iolock
= XFS_IOLOCK_SHARED
;
700 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
, 0);
701 ret
= generic_file_direct_write(iocb
, iovp
,
702 &nr_segs
, pos
, &iocb
->ki_pos
, count
, ocount
);
705 xfs_rw_iunlock(ip
, iolock
);
707 /* No fallback to buffered IO on errors for XFS. */
708 ASSERT(ret
< 0 || ret
== count
);
713 xfs_file_buffered_aio_write(
715 const struct iovec
*iovp
,
716 unsigned long nr_segs
,
720 struct file
*file
= iocb
->ki_filp
;
721 struct address_space
*mapping
= file
->f_mapping
;
722 struct inode
*inode
= mapping
->host
;
723 struct xfs_inode
*ip
= XFS_I(inode
);
726 int iolock
= XFS_IOLOCK_EXCL
;
727 size_t count
= ocount
;
729 xfs_rw_ilock(ip
, iolock
);
731 ret
= xfs_file_aio_write_checks(file
, &pos
, &count
, &iolock
);
735 /* We can write back this queue in page reclaim */
736 current
->backing_dev_info
= mapping
->backing_dev_info
;
739 trace_xfs_file_buffered_write(ip
, count
, iocb
->ki_pos
, 0);
740 ret
= generic_file_buffered_write(iocb
, iovp
, nr_segs
,
741 pos
, &iocb
->ki_pos
, count
, 0);
744 * If we just got an ENOSPC, try to write back all dirty inodes to
745 * convert delalloc space to free up some of the excess reserved
748 if (ret
== -ENOSPC
&& !enospc
) {
750 xfs_flush_inodes(ip
->i_mount
);
754 current
->backing_dev_info
= NULL
;
756 xfs_rw_iunlock(ip
, iolock
);
763 const struct iovec
*iovp
,
764 unsigned long nr_segs
,
767 struct file
*file
= iocb
->ki_filp
;
768 struct address_space
*mapping
= file
->f_mapping
;
769 struct inode
*inode
= mapping
->host
;
770 struct xfs_inode
*ip
= XFS_I(inode
);
774 XFS_STATS_INC(xs_write_calls
);
776 BUG_ON(iocb
->ki_pos
!= pos
);
778 ret
= generic_segment_checks(iovp
, &nr_segs
, &ocount
, VERIFY_READ
);
785 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
790 if (unlikely(file
->f_flags
& O_DIRECT
))
791 ret
= xfs_file_dio_aio_write(iocb
, iovp
, nr_segs
, pos
, ocount
);
793 ret
= xfs_file_buffered_aio_write(iocb
, iovp
, nr_segs
, pos
,
799 XFS_STATS_ADD(xs_write_bytes
, ret
);
801 /* Handle various SYNC-type writes */
802 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
818 struct inode
*inode
= file_inode(file
);
819 struct xfs_inode
*ip
= XFS_I(inode
);
820 struct xfs_trans
*tp
;
824 if (!S_ISREG(inode
->i_mode
))
826 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
827 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
))
830 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
831 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
832 error
= xfs_free_file_space(ip
, offset
, len
);
835 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
836 unsigned blksize_mask
= (1 << inode
->i_blkbits
) - 1;
838 if (offset
& blksize_mask
|| len
& blksize_mask
) {
843 ASSERT(offset
+ len
< i_size_read(inode
));
844 new_size
= i_size_read(inode
) - len
;
846 error
= xfs_collapse_file_space(ip
, offset
, len
);
850 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
851 offset
+ len
> i_size_read(inode
)) {
852 new_size
= offset
+ len
;
853 error
= -inode_newsize_ok(inode
, new_size
);
858 if (mode
& FALLOC_FL_ZERO_RANGE
)
859 error
= xfs_zero_file_space(ip
, offset
, len
);
861 error
= xfs_alloc_file_space(ip
, offset
, len
,
867 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_WRITEID
);
868 error
= xfs_trans_reserve(tp
, &M_RES(ip
->i_mount
)->tr_writeid
, 0, 0);
870 xfs_trans_cancel(tp
, 0);
874 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
875 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
876 ip
->i_d
.di_mode
&= ~S_ISUID
;
877 if (ip
->i_d
.di_mode
& S_IXGRP
)
878 ip
->i_d
.di_mode
&= ~S_ISGID
;
880 if (!(mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_COLLAPSE_RANGE
)))
881 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
883 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
884 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
886 if (file
->f_flags
& O_DSYNC
)
887 xfs_trans_set_sync(tp
);
888 error
= xfs_trans_commit(tp
, 0);
892 /* Change file size if needed */
896 iattr
.ia_valid
= ATTR_SIZE
;
897 iattr
.ia_size
= new_size
;
898 error
= xfs_setattr_size(ip
, &iattr
);
902 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
912 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
914 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
924 struct xfs_inode
*ip
= XFS_I(inode
);
928 error
= xfs_file_open(inode
, file
);
933 * If there are any blocks, read-ahead block 0 as we're almost
934 * certain to have the next operation be a read there.
936 mode
= xfs_ilock_data_map_shared(ip
);
937 if (ip
->i_d
.di_nextents
> 0)
938 xfs_dir3_data_readahead(NULL
, ip
, 0, -1);
939 xfs_iunlock(ip
, mode
);
948 return -xfs_release(XFS_I(inode
));
954 struct dir_context
*ctx
)
956 struct inode
*inode
= file_inode(file
);
957 xfs_inode_t
*ip
= XFS_I(inode
);
962 * The Linux API doesn't pass down the total size of the buffer
963 * we read into down to the filesystem. With the filldir concept
964 * it's not needed for correct information, but the XFS dir2 leaf
965 * code wants an estimate of the buffer size to calculate it's
966 * readahead window and size the buffers used for mapping to
969 * Try to give it an estimate that's good enough, maybe at some
970 * point we can change the ->readdir prototype to include the
971 * buffer size. For now we use the current glibc buffer size.
973 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
975 error
= xfs_readdir(ip
, ctx
, bufsize
);
984 struct vm_area_struct
*vma
)
986 vma
->vm_ops
= &xfs_file_vm_ops
;
993 * mmap()d file has taken write protection fault and is being made
994 * writable. We can set the page state up correctly for a writable
995 * page, which means we can do correct delalloc accounting (ENOSPC
996 * checking!) and unwritten extent mapping.
1000 struct vm_area_struct
*vma
,
1001 struct vm_fault
*vmf
)
1003 return block_page_mkwrite(vma
, vmf
, xfs_get_blocks
);
1007 * This type is designed to indicate the type of offset we would like
1008 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
1016 * Lookup the desired type of offset from the given page.
1018 * On success, return true and the offset argument will point to the
1019 * start of the region that was found. Otherwise this function will
1020 * return false and keep the offset argument unchanged.
1023 xfs_lookup_buffer_offset(
1028 loff_t lastoff
= page_offset(page
);
1030 struct buffer_head
*bh
, *head
;
1032 bh
= head
= page_buffers(page
);
1035 * Unwritten extents that have data in the page
1036 * cache covering them can be identified by the
1037 * BH_Unwritten state flag. Pages with multiple
1038 * buffers might have a mix of holes, data and
1039 * unwritten extents - any buffer with valid
1040 * data in it should have BH_Uptodate flag set
1043 if (buffer_unwritten(bh
) ||
1044 buffer_uptodate(bh
)) {
1045 if (type
== DATA_OFF
)
1048 if (type
== HOLE_OFF
)
1056 lastoff
+= bh
->b_size
;
1057 } while ((bh
= bh
->b_this_page
) != head
);
1063 * This routine is called to find out and return a data or hole offset
1064 * from the page cache for unwritten extents according to the desired
1065 * type for xfs_seek_data() or xfs_seek_hole().
1067 * The argument offset is used to tell where we start to search from the
1068 * page cache. Map is used to figure out the end points of the range to
1071 * Return true if the desired type of offset was found, and the argument
1072 * offset is filled with that address. Otherwise, return false and keep
1076 xfs_find_get_desired_pgoff(
1077 struct inode
*inode
,
1078 struct xfs_bmbt_irec
*map
,
1082 struct xfs_inode
*ip
= XFS_I(inode
);
1083 struct xfs_mount
*mp
= ip
->i_mount
;
1084 struct pagevec pvec
;
1088 loff_t startoff
= *offset
;
1089 loff_t lastoff
= startoff
;
1092 pagevec_init(&pvec
, 0);
1094 index
= startoff
>> PAGE_CACHE_SHIFT
;
1095 endoff
= XFS_FSB_TO_B(mp
, map
->br_startoff
+ map
->br_blockcount
);
1096 end
= endoff
>> PAGE_CACHE_SHIFT
;
1102 want
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
);
1103 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
1106 * No page mapped into given range. If we are searching holes
1107 * and if this is the first time we got into the loop, it means
1108 * that the given offset is landed in a hole, return it.
1110 * If we have already stepped through some block buffers to find
1111 * holes but they all contains data. In this case, the last
1112 * offset is already updated and pointed to the end of the last
1113 * mapped page, if it does not reach the endpoint to search,
1114 * that means there should be a hole between them.
1116 if (nr_pages
== 0) {
1117 /* Data search found nothing */
1118 if (type
== DATA_OFF
)
1121 ASSERT(type
== HOLE_OFF
);
1122 if (lastoff
== startoff
|| lastoff
< endoff
) {
1130 * At lease we found one page. If this is the first time we
1131 * step into the loop, and if the first page index offset is
1132 * greater than the given search offset, a hole was found.
1134 if (type
== HOLE_OFF
&& lastoff
== startoff
&&
1135 lastoff
< page_offset(pvec
.pages
[0])) {
1140 for (i
= 0; i
< nr_pages
; i
++) {
1141 struct page
*page
= pvec
.pages
[i
];
1145 * At this point, the page may be truncated or
1146 * invalidated (changing page->mapping to NULL),
1147 * or even swizzled back from swapper_space to tmpfs
1148 * file mapping. However, page->index will not change
1149 * because we have a reference on the page.
1151 * Searching done if the page index is out of range.
1152 * If the current offset is not reaches the end of
1153 * the specified search range, there should be a hole
1156 if (page
->index
> end
) {
1157 if (type
== HOLE_OFF
&& lastoff
< endoff
) {
1166 * Page truncated or invalidated(page->mapping == NULL).
1167 * We can freely skip it and proceed to check the next
1170 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
1175 if (!page_has_buffers(page
)) {
1180 found
= xfs_lookup_buffer_offset(page
, &b_offset
, type
);
1183 * The found offset may be less than the start
1184 * point to search if this is the first time to
1187 *offset
= max_t(loff_t
, startoff
, b_offset
);
1193 * We either searching data but nothing was found, or
1194 * searching hole but found a data buffer. In either
1195 * case, probably the next page contains the desired
1196 * things, update the last offset to it so.
1198 lastoff
= page_offset(page
) + PAGE_SIZE
;
1203 * The number of returned pages less than our desired, search
1204 * done. In this case, nothing was found for searching data,
1205 * but we found a hole behind the last offset.
1207 if (nr_pages
< want
) {
1208 if (type
== HOLE_OFF
) {
1215 index
= pvec
.pages
[i
- 1]->index
+ 1;
1216 pagevec_release(&pvec
);
1217 } while (index
<= end
);
1220 pagevec_release(&pvec
);
1229 struct inode
*inode
= file
->f_mapping
->host
;
1230 struct xfs_inode
*ip
= XFS_I(inode
);
1231 struct xfs_mount
*mp
= ip
->i_mount
;
1232 loff_t
uninitialized_var(offset
);
1234 xfs_fileoff_t fsbno
;
1239 lock
= xfs_ilock_data_map_shared(ip
);
1241 isize
= i_size_read(inode
);
1242 if (start
>= isize
) {
1248 * Try to read extents from the first block indicated
1249 * by fsbno to the end block of the file.
1251 fsbno
= XFS_B_TO_FSBT(mp
, start
);
1252 end
= XFS_B_TO_FSB(mp
, isize
);
1254 struct xfs_bmbt_irec map
[2];
1258 error
= xfs_bmapi_read(ip
, fsbno
, end
- fsbno
, map
, &nmap
,
1263 /* No extents at given offset, must be beyond EOF */
1269 for (i
= 0; i
< nmap
; i
++) {
1270 offset
= max_t(loff_t
, start
,
1271 XFS_FSB_TO_B(mp
, map
[i
].br_startoff
));
1273 /* Landed in a data extent */
1274 if (map
[i
].br_startblock
== DELAYSTARTBLOCK
||
1275 (map
[i
].br_state
== XFS_EXT_NORM
&&
1276 !isnullstartblock(map
[i
].br_startblock
)))
1280 * Landed in an unwritten extent, try to search data
1283 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
) {
1284 if (xfs_find_get_desired_pgoff(inode
, &map
[i
],
1291 * map[0] is hole or its an unwritten extent but
1292 * without data in page cache. Probably means that
1293 * we are reading after EOF if nothing in map[1].
1303 * Nothing was found, proceed to the next round of search
1304 * if reading offset not beyond or hit EOF.
1306 fsbno
= map
[i
- 1].br_startoff
+ map
[i
- 1].br_blockcount
;
1307 start
= XFS_FSB_TO_B(mp
, fsbno
);
1308 if (start
>= isize
) {
1315 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1318 xfs_iunlock(ip
, lock
);
1330 struct inode
*inode
= file
->f_mapping
->host
;
1331 struct xfs_inode
*ip
= XFS_I(inode
);
1332 struct xfs_mount
*mp
= ip
->i_mount
;
1333 loff_t
uninitialized_var(offset
);
1335 xfs_fileoff_t fsbno
;
1340 if (XFS_FORCED_SHUTDOWN(mp
))
1341 return -XFS_ERROR(EIO
);
1343 lock
= xfs_ilock_data_map_shared(ip
);
1345 isize
= i_size_read(inode
);
1346 if (start
>= isize
) {
1351 fsbno
= XFS_B_TO_FSBT(mp
, start
);
1352 end
= XFS_B_TO_FSB(mp
, isize
);
1355 struct xfs_bmbt_irec map
[2];
1359 error
= xfs_bmapi_read(ip
, fsbno
, end
- fsbno
, map
, &nmap
,
1364 /* No extents at given offset, must be beyond EOF */
1370 for (i
= 0; i
< nmap
; i
++) {
1371 offset
= max_t(loff_t
, start
,
1372 XFS_FSB_TO_B(mp
, map
[i
].br_startoff
));
1374 /* Landed in a hole */
1375 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1379 * Landed in an unwritten extent, try to search hole
1382 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
) {
1383 if (xfs_find_get_desired_pgoff(inode
, &map
[i
],
1390 * map[0] contains data or its unwritten but contains
1391 * data in page cache, probably means that we are
1392 * reading after EOF. We should fix offset to point
1393 * to the end of the file(i.e., there is an implicit
1394 * hole at the end of any file).
1404 * Both mappings contains data, proceed to the next round of
1405 * search if the current reading offset not beyond or hit EOF.
1407 fsbno
= map
[i
- 1].br_startoff
+ map
[i
- 1].br_blockcount
;
1408 start
= XFS_FSB_TO_B(mp
, fsbno
);
1409 if (start
>= isize
) {
1417 * At this point, we must have found a hole. However, the returned
1418 * offset may be bigger than the file size as it may be aligned to
1419 * page boundary for unwritten extents, we need to deal with this
1420 * situation in particular.
1422 offset
= min_t(loff_t
, offset
, isize
);
1423 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1426 xfs_iunlock(ip
, lock
);
1443 return generic_file_llseek(file
, offset
, origin
);
1445 return xfs_seek_data(file
, offset
);
1447 return xfs_seek_hole(file
, offset
);
1453 const struct file_operations xfs_file_operations
= {
1454 .llseek
= xfs_file_llseek
,
1455 .read
= do_sync_read
,
1456 .write
= do_sync_write
,
1457 .aio_read
= xfs_file_aio_read
,
1458 .aio_write
= xfs_file_aio_write
,
1459 .splice_read
= xfs_file_splice_read
,
1460 .splice_write
= xfs_file_splice_write
,
1461 .unlocked_ioctl
= xfs_file_ioctl
,
1462 #ifdef CONFIG_COMPAT
1463 .compat_ioctl
= xfs_file_compat_ioctl
,
1465 .mmap
= xfs_file_mmap
,
1466 .open
= xfs_file_open
,
1467 .release
= xfs_file_release
,
1468 .fsync
= xfs_file_fsync
,
1469 .fallocate
= xfs_file_fallocate
,
1472 const struct file_operations xfs_dir_file_operations
= {
1473 .open
= xfs_dir_open
,
1474 .read
= generic_read_dir
,
1475 .iterate
= xfs_file_readdir
,
1476 .llseek
= generic_file_llseek
,
1477 .unlocked_ioctl
= xfs_file_ioctl
,
1478 #ifdef CONFIG_COMPAT
1479 .compat_ioctl
= xfs_file_compat_ioctl
,
1481 .fsync
= xfs_dir_fsync
,
1484 static const struct vm_operations_struct xfs_file_vm_ops
= {
1485 .fault
= filemap_fault
,
1486 .page_mkwrite
= xfs_vm_page_mkwrite
,
1487 .remap_pages
= generic_file_remap_pages
,