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"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_ioctl.h"
36 #include "xfs_trace.h"
38 #include "xfs_icache.h"
41 #include <linux/dcache.h>
42 #include <linux/falloc.h>
43 #include <linux/pagevec.h>
44 #include <linux/backing-dev.h>
46 static const struct vm_operations_struct xfs_file_vm_ops
;
49 * Locking primitives for read and write IO paths to ensure we consistently use
50 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
57 if (type
& XFS_IOLOCK_EXCL
)
58 mutex_lock(&VFS_I(ip
)->i_mutex
);
67 xfs_iunlock(ip
, type
);
68 if (type
& XFS_IOLOCK_EXCL
)
69 mutex_unlock(&VFS_I(ip
)->i_mutex
);
77 xfs_ilock_demote(ip
, type
);
78 if (type
& XFS_IOLOCK_EXCL
)
79 mutex_unlock(&VFS_I(ip
)->i_mutex
);
85 * xfs_iozero clears the specified range of buffer supplied,
86 * and marks all the affected blocks as valid and modified. If
87 * an affected block is not allocated, it will be allocated. If
88 * an affected block is not completely overwritten, and is not
89 * valid before the operation, it will be read from disk before
90 * being partially zeroed.
94 struct xfs_inode
*ip
, /* inode */
95 loff_t pos
, /* offset in file */
96 size_t count
) /* size of data to zero */
99 struct address_space
*mapping
;
102 mapping
= VFS_I(ip
)->i_mapping
;
104 unsigned offset
, bytes
;
107 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
108 bytes
= PAGE_CACHE_SIZE
- offset
;
112 status
= pagecache_write_begin(NULL
, mapping
, pos
, bytes
,
113 AOP_FLAG_UNINTERRUPTIBLE
,
118 zero_user(page
, offset
, bytes
);
120 status
= pagecache_write_end(NULL
, mapping
, pos
, bytes
, bytes
,
122 WARN_ON(status
<= 0); /* can't return less than zero! */
132 xfs_update_prealloc_flags(
133 struct xfs_inode
*ip
,
134 enum xfs_prealloc_flags flags
)
136 struct xfs_trans
*tp
;
139 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_WRITEID
);
140 error
= xfs_trans_reserve(tp
, &M_RES(ip
->i_mount
)->tr_writeid
, 0, 0);
142 xfs_trans_cancel(tp
, 0);
146 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
147 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
149 if (!(flags
& XFS_PREALLOC_INVISIBLE
)) {
150 ip
->i_d
.di_mode
&= ~S_ISUID
;
151 if (ip
->i_d
.di_mode
& S_IXGRP
)
152 ip
->i_d
.di_mode
&= ~S_ISGID
;
153 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
156 if (flags
& XFS_PREALLOC_SET
)
157 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
158 if (flags
& XFS_PREALLOC_CLEAR
)
159 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
161 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
162 if (flags
& XFS_PREALLOC_SYNC
)
163 xfs_trans_set_sync(tp
);
164 return xfs_trans_commit(tp
, 0);
168 * Fsync operations on directories are much simpler than on regular files,
169 * as there is no file data to flush, and thus also no need for explicit
170 * cache flush operations, and there are no non-transaction metadata updates
171 * on directories either.
180 struct xfs_inode
*ip
= XFS_I(file
->f_mapping
->host
);
181 struct xfs_mount
*mp
= ip
->i_mount
;
184 trace_xfs_dir_fsync(ip
);
186 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
187 if (xfs_ipincount(ip
))
188 lsn
= ip
->i_itemp
->ili_last_lsn
;
189 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
193 return _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, NULL
);
203 struct inode
*inode
= file
->f_mapping
->host
;
204 struct xfs_inode
*ip
= XFS_I(inode
);
205 struct xfs_mount
*mp
= ip
->i_mount
;
210 trace_xfs_file_fsync(ip
);
212 error
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
216 if (XFS_FORCED_SHUTDOWN(mp
))
219 xfs_iflags_clear(ip
, XFS_ITRUNCATED
);
221 if (mp
->m_flags
& XFS_MOUNT_BARRIER
) {
223 * If we have an RT and/or log subvolume we need to make sure
224 * to flush the write cache the device used for file data
225 * first. This is to ensure newly written file data make
226 * it to disk before logging the new inode size in case of
227 * an extending write.
229 if (XFS_IS_REALTIME_INODE(ip
))
230 xfs_blkdev_issue_flush(mp
->m_rtdev_targp
);
231 else if (mp
->m_logdev_targp
!= mp
->m_ddev_targp
)
232 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
236 * All metadata updates are logged, which means that we just have
237 * to flush the log up to the latest LSN that touched the inode.
239 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
240 if (xfs_ipincount(ip
)) {
242 (ip
->i_itemp
->ili_fields
& ~XFS_ILOG_TIMESTAMP
))
243 lsn
= ip
->i_itemp
->ili_last_lsn
;
245 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
248 error
= _xfs_log_force_lsn(mp
, lsn
, XFS_LOG_SYNC
, &log_flushed
);
251 * If we only have a single device, and the log force about was
252 * a no-op we might have to flush the data device cache here.
253 * This can only happen for fdatasync/O_DSYNC if we were overwriting
254 * an already allocated file and thus do not have any metadata to
257 if ((mp
->m_flags
& XFS_MOUNT_BARRIER
) &&
258 mp
->m_logdev_targp
== mp
->m_ddev_targp
&&
259 !XFS_IS_REALTIME_INODE(ip
) &&
261 xfs_blkdev_issue_flush(mp
->m_ddev_targp
);
271 struct file
*file
= iocb
->ki_filp
;
272 struct inode
*inode
= file
->f_mapping
->host
;
273 struct xfs_inode
*ip
= XFS_I(inode
);
274 struct xfs_mount
*mp
= ip
->i_mount
;
275 size_t size
= iov_iter_count(to
);
279 loff_t pos
= iocb
->ki_pos
;
281 XFS_STATS_INC(xs_read_calls
);
283 if (unlikely(iocb
->ki_flags
& IOCB_DIRECT
))
284 ioflags
|= XFS_IO_ISDIRECT
;
285 if (file
->f_mode
& FMODE_NOCMTIME
)
286 ioflags
|= XFS_IO_INVIS
;
288 if (unlikely(ioflags
& XFS_IO_ISDIRECT
)) {
289 xfs_buftarg_t
*target
=
290 XFS_IS_REALTIME_INODE(ip
) ?
291 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
292 /* DIO must be aligned to device logical sector size */
293 if ((pos
| size
) & target
->bt_logical_sectormask
) {
294 if (pos
== i_size_read(inode
))
300 n
= mp
->m_super
->s_maxbytes
- pos
;
301 if (n
<= 0 || size
== 0)
307 if (XFS_FORCED_SHUTDOWN(mp
))
311 * Locking is a bit tricky here. If we take an exclusive lock
312 * for direct IO, we effectively serialise all new concurrent
313 * read IO to this file and block it behind IO that is currently in
314 * progress because IO in progress holds the IO lock shared. We only
315 * need to hold the lock exclusive to blow away the page cache, so
316 * only take lock exclusively if the page cache needs invalidation.
317 * This allows the normal direct IO case of no page cache pages to
318 * proceeed concurrently without serialisation.
320 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
321 if ((ioflags
& XFS_IO_ISDIRECT
) && inode
->i_mapping
->nrpages
) {
322 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
323 xfs_rw_ilock(ip
, XFS_IOLOCK_EXCL
);
325 if (inode
->i_mapping
->nrpages
) {
326 ret
= filemap_write_and_wait_range(
327 VFS_I(ip
)->i_mapping
,
328 pos
, pos
+ size
- 1);
330 xfs_rw_iunlock(ip
, XFS_IOLOCK_EXCL
);
335 * Invalidate whole pages. This can return an error if
336 * we fail to invalidate a page, but this should never
337 * happen on XFS. Warn if it does fail.
339 ret
= invalidate_inode_pages2_range(VFS_I(ip
)->i_mapping
,
340 pos
>> PAGE_CACHE_SHIFT
,
341 (pos
+ size
- 1) >> PAGE_CACHE_SHIFT
);
345 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
348 trace_xfs_file_read(ip
, size
, pos
, ioflags
);
350 ret
= generic_file_read_iter(iocb
, to
);
352 XFS_STATS_ADD(xs_read_bytes
, ret
);
354 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
359 xfs_file_splice_read(
362 struct pipe_inode_info
*pipe
,
366 struct xfs_inode
*ip
= XFS_I(infilp
->f_mapping
->host
);
370 XFS_STATS_INC(xs_read_calls
);
372 if (infilp
->f_mode
& FMODE_NOCMTIME
)
373 ioflags
|= XFS_IO_INVIS
;
375 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
378 xfs_rw_ilock(ip
, XFS_IOLOCK_SHARED
);
380 trace_xfs_file_splice_read(ip
, count
, *ppos
, ioflags
);
382 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
384 XFS_STATS_ADD(xs_read_bytes
, ret
);
386 xfs_rw_iunlock(ip
, XFS_IOLOCK_SHARED
);
391 * This routine is called to handle zeroing any space in the last block of the
392 * file that is beyond the EOF. We do this since the size is being increased
393 * without writing anything to that block and we don't want to read the
394 * garbage on the disk.
396 STATIC
int /* error (positive) */
398 struct xfs_inode
*ip
,
403 struct xfs_mount
*mp
= ip
->i_mount
;
404 xfs_fileoff_t last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
405 int zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
409 struct xfs_bmbt_irec imap
;
411 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
412 error
= xfs_bmapi_read(ip
, last_fsb
, 1, &imap
, &nimaps
, 0);
413 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
420 * If the block underlying isize is just a hole, then there
421 * is nothing to zero.
423 if (imap
.br_startblock
== HOLESTARTBLOCK
)
426 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
427 if (isize
+ zero_len
> offset
)
428 zero_len
= offset
- isize
;
430 return xfs_iozero(ip
, isize
, zero_len
);
434 * Zero any on disk space between the current EOF and the new, larger EOF.
436 * This handles the normal case of zeroing the remainder of the last block in
437 * the file and the unusual case of zeroing blocks out beyond the size of the
438 * file. This second case only happens with fixed size extents and when the
439 * system crashes before the inode size was updated but after blocks were
442 * Expects the iolock to be held exclusive, and will take the ilock internally.
444 int /* error (positive) */
446 struct xfs_inode
*ip
,
447 xfs_off_t offset
, /* starting I/O offset */
448 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
, did_zeroing
);
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
);
532 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
533 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
540 * Common pre-write limit and setup checks.
542 * Called with the iolocked held either shared and exclusive according to
543 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
544 * if called for a direct write beyond i_size.
547 xfs_file_aio_write_checks(
549 struct iov_iter
*from
,
552 struct file
*file
= iocb
->ki_filp
;
553 struct inode
*inode
= file
->f_mapping
->host
;
554 struct xfs_inode
*ip
= XFS_I(inode
);
556 size_t count
= iov_iter_count(from
);
559 error
= generic_write_checks(iocb
, from
);
563 error
= xfs_break_layouts(inode
, iolock
, true);
568 * If the offset is beyond the size of the file, we need to zero any
569 * blocks that fall between the existing EOF and the start of this
570 * write. If zeroing is needed and we are currently holding the
571 * iolock shared, we need to update it to exclusive which implies
572 * having to redo all checks before.
574 * We need to serialise against EOF updates that occur in IO
575 * completions here. We want to make sure that nobody is changing the
576 * size while we do this check until we have placed an IO barrier (i.e.
577 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
578 * The spinlock effectively forms a memory barrier once we have the
579 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
580 * and hence be able to correctly determine if we need to run zeroing.
582 spin_lock(&ip
->i_flags_lock
);
583 if (iocb
->ki_pos
> i_size_read(inode
)) {
586 spin_unlock(&ip
->i_flags_lock
);
587 if (*iolock
== XFS_IOLOCK_SHARED
) {
588 xfs_rw_iunlock(ip
, *iolock
);
589 *iolock
= XFS_IOLOCK_EXCL
;
590 xfs_rw_ilock(ip
, *iolock
);
591 iov_iter_reexpand(from
, count
);
594 * We now have an IO submission barrier in place, but
595 * AIO can do EOF updates during IO completion and hence
596 * we now need to wait for all of them to drain. Non-AIO
597 * DIO will have drained before we are given the
598 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
601 inode_dio_wait(inode
);
604 error
= xfs_zero_eof(ip
, iocb
->ki_pos
, i_size_read(inode
), &zero
);
608 spin_unlock(&ip
->i_flags_lock
);
611 * Updating the timestamps will grab the ilock again from
612 * xfs_fs_dirty_inode, so we have to call it after dropping the
613 * lock above. Eventually we should look into a way to avoid
614 * the pointless lock roundtrip.
616 if (likely(!(file
->f_mode
& FMODE_NOCMTIME
))) {
617 error
= file_update_time(file
);
623 * If we're writing the file then make sure to clear the setuid and
624 * setgid bits if the process is not being run by root. This keeps
625 * people from modifying setuid and setgid binaries.
627 return file_remove_suid(file
);
631 * xfs_file_dio_aio_write - handle direct IO writes
633 * Lock the inode appropriately to prepare for and issue a direct IO write.
634 * By separating it from the buffered write path we remove all the tricky to
635 * follow locking changes and looping.
637 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
638 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
639 * pages are flushed out.
641 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
642 * allowing them to be done in parallel with reads and other direct IO writes.
643 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
644 * needs to do sub-block zeroing and that requires serialisation against other
645 * direct IOs to the same block. In this case we need to serialise the
646 * submission of the unaligned IOs so that we don't get racing block zeroing in
647 * the dio layer. To avoid the problem with aio, we also need to wait for
648 * outstanding IOs to complete so that unwritten extent conversion is completed
649 * before we try to map the overlapping block. This is currently implemented by
650 * hitting it with a big hammer (i.e. inode_dio_wait()).
652 * Returns with locks held indicated by @iolock and errors indicated by
653 * negative return values.
656 xfs_file_dio_aio_write(
658 struct iov_iter
*from
)
660 struct file
*file
= iocb
->ki_filp
;
661 struct address_space
*mapping
= file
->f_mapping
;
662 struct inode
*inode
= mapping
->host
;
663 struct xfs_inode
*ip
= XFS_I(inode
);
664 struct xfs_mount
*mp
= ip
->i_mount
;
666 int unaligned_io
= 0;
668 size_t count
= iov_iter_count(from
);
669 loff_t pos
= iocb
->ki_pos
;
671 struct iov_iter data
;
672 struct xfs_buftarg
*target
= XFS_IS_REALTIME_INODE(ip
) ?
673 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
675 /* DIO must be aligned to device logical sector size */
676 if ((pos
| count
) & target
->bt_logical_sectormask
)
679 /* "unaligned" here means not aligned to a filesystem block */
680 if ((pos
& mp
->m_blockmask
) || ((pos
+ count
) & mp
->m_blockmask
))
684 * We don't need to take an exclusive lock unless there page cache needs
685 * to be invalidated or unaligned IO is being executed. We don't need to
686 * consider the EOF extension case here because
687 * xfs_file_aio_write_checks() will relock the inode as necessary for
688 * EOF zeroing cases and fill out the new inode size as appropriate.
690 if (unaligned_io
|| mapping
->nrpages
)
691 iolock
= XFS_IOLOCK_EXCL
;
693 iolock
= XFS_IOLOCK_SHARED
;
694 xfs_rw_ilock(ip
, iolock
);
697 * Recheck if there are cached pages that need invalidate after we got
698 * the iolock to protect against other threads adding new pages while
699 * we were waiting for the iolock.
701 if (mapping
->nrpages
&& iolock
== XFS_IOLOCK_SHARED
) {
702 xfs_rw_iunlock(ip
, iolock
);
703 iolock
= XFS_IOLOCK_EXCL
;
704 xfs_rw_ilock(ip
, iolock
);
707 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
710 count
= iov_iter_count(from
);
712 end
= pos
+ count
- 1;
714 if (mapping
->nrpages
) {
715 ret
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
720 * Invalidate whole pages. This can return an error if
721 * we fail to invalidate a page, but this should never
722 * happen on XFS. Warn if it does fail.
724 ret
= invalidate_inode_pages2_range(VFS_I(ip
)->i_mapping
,
725 pos
>> PAGE_CACHE_SHIFT
,
726 end
>> PAGE_CACHE_SHIFT
);
732 * If we are doing unaligned IO, wait for all other IO to drain,
733 * otherwise demote the lock if we had to flush cached pages
736 inode_dio_wait(inode
);
737 else if (iolock
== XFS_IOLOCK_EXCL
) {
738 xfs_rw_ilock_demote(ip
, XFS_IOLOCK_EXCL
);
739 iolock
= XFS_IOLOCK_SHARED
;
742 trace_xfs_file_direct_write(ip
, count
, iocb
->ki_pos
, 0);
745 ret
= mapping
->a_ops
->direct_IO(iocb
, &data
, pos
);
747 /* see generic_file_direct_write() for why this is necessary */
748 if (mapping
->nrpages
) {
749 invalidate_inode_pages2_range(mapping
,
750 pos
>> PAGE_CACHE_SHIFT
,
751 end
>> PAGE_CACHE_SHIFT
);
756 iov_iter_advance(from
, ret
);
760 xfs_rw_iunlock(ip
, iolock
);
762 /* No fallback to buffered IO on errors for XFS. */
763 ASSERT(ret
< 0 || ret
== count
);
768 xfs_file_buffered_aio_write(
770 struct iov_iter
*from
)
772 struct file
*file
= iocb
->ki_filp
;
773 struct address_space
*mapping
= file
->f_mapping
;
774 struct inode
*inode
= mapping
->host
;
775 struct xfs_inode
*ip
= XFS_I(inode
);
778 int iolock
= XFS_IOLOCK_EXCL
;
780 xfs_rw_ilock(ip
, iolock
);
782 ret
= xfs_file_aio_write_checks(iocb
, from
, &iolock
);
786 /* We can write back this queue in page reclaim */
787 current
->backing_dev_info
= inode_to_bdi(inode
);
790 trace_xfs_file_buffered_write(ip
, iov_iter_count(from
),
792 ret
= generic_perform_write(file
, from
, iocb
->ki_pos
);
793 if (likely(ret
>= 0))
797 * If we hit a space limit, try to free up some lingering preallocated
798 * space before returning an error. In the case of ENOSPC, first try to
799 * write back all dirty inodes to free up some of the excess reserved
800 * metadata space. This reduces the chances that the eofblocks scan
801 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
802 * also behaves as a filter to prevent too many eofblocks scans from
803 * running at the same time.
805 if (ret
== -EDQUOT
&& !enospc
) {
806 enospc
= xfs_inode_free_quota_eofblocks(ip
);
809 } else if (ret
== -ENOSPC
&& !enospc
) {
810 struct xfs_eofblocks eofb
= {0};
813 xfs_flush_inodes(ip
->i_mount
);
814 eofb
.eof_scan_owner
= ip
->i_ino
; /* for locking */
815 eofb
.eof_flags
= XFS_EOF_FLAGS_SYNC
;
816 xfs_icache_free_eofblocks(ip
->i_mount
, &eofb
);
820 current
->backing_dev_info
= NULL
;
822 xfs_rw_iunlock(ip
, iolock
);
829 struct iov_iter
*from
)
831 struct file
*file
= iocb
->ki_filp
;
832 struct address_space
*mapping
= file
->f_mapping
;
833 struct inode
*inode
= mapping
->host
;
834 struct xfs_inode
*ip
= XFS_I(inode
);
836 size_t ocount
= iov_iter_count(from
);
838 XFS_STATS_INC(xs_write_calls
);
843 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
846 if (unlikely(iocb
->ki_flags
& IOCB_DIRECT
))
847 ret
= xfs_file_dio_aio_write(iocb
, from
);
849 ret
= xfs_file_buffered_aio_write(iocb
, from
);
854 XFS_STATS_ADD(xs_write_bytes
, ret
);
856 /* Handle various SYNC-type writes */
857 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
864 #define XFS_FALLOC_FL_SUPPORTED \
865 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
866 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
867 FALLOC_FL_INSERT_RANGE)
876 struct inode
*inode
= file_inode(file
);
877 struct xfs_inode
*ip
= XFS_I(inode
);
879 enum xfs_prealloc_flags flags
= 0;
880 uint iolock
= XFS_IOLOCK_EXCL
;
882 bool do_file_insert
= 0;
884 if (!S_ISREG(inode
->i_mode
))
886 if (mode
& ~XFS_FALLOC_FL_SUPPORTED
)
889 xfs_ilock(ip
, iolock
);
890 error
= xfs_break_layouts(inode
, &iolock
, false);
894 xfs_ilock(ip
, XFS_MMAPLOCK_EXCL
);
895 iolock
|= XFS_MMAPLOCK_EXCL
;
897 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
898 error
= xfs_free_file_space(ip
, offset
, len
);
901 } else if (mode
& FALLOC_FL_COLLAPSE_RANGE
) {
902 unsigned blksize_mask
= (1 << inode
->i_blkbits
) - 1;
904 if (offset
& blksize_mask
|| len
& blksize_mask
) {
910 * There is no need to overlap collapse range with EOF,
911 * in which case it is effectively a truncate operation
913 if (offset
+ len
>= i_size_read(inode
)) {
918 new_size
= i_size_read(inode
) - len
;
920 error
= xfs_collapse_file_space(ip
, offset
, len
);
923 } else if (mode
& FALLOC_FL_INSERT_RANGE
) {
924 unsigned blksize_mask
= (1 << inode
->i_blkbits
) - 1;
926 new_size
= i_size_read(inode
) + len
;
927 if (offset
& blksize_mask
|| len
& blksize_mask
) {
932 /* check the new inode size does not wrap through zero */
933 if (new_size
> inode
->i_sb
->s_maxbytes
) {
938 /* Offset should be less than i_size */
939 if (offset
>= i_size_read(inode
)) {
945 flags
|= XFS_PREALLOC_SET
;
947 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
948 offset
+ len
> i_size_read(inode
)) {
949 new_size
= offset
+ len
;
950 error
= inode_newsize_ok(inode
, new_size
);
955 if (mode
& FALLOC_FL_ZERO_RANGE
)
956 error
= xfs_zero_file_space(ip
, offset
, len
);
958 error
= xfs_alloc_file_space(ip
, offset
, len
,
964 if (file
->f_flags
& O_DSYNC
)
965 flags
|= XFS_PREALLOC_SYNC
;
967 error
= xfs_update_prealloc_flags(ip
, flags
);
971 /* Change file size if needed */
975 iattr
.ia_valid
= ATTR_SIZE
;
976 iattr
.ia_size
= new_size
;
977 error
= xfs_setattr_size(ip
, &iattr
);
983 * Perform hole insertion now that the file size has been
984 * updated so that if we crash during the operation we don't
985 * leave shifted extents past EOF and hence losing access to
986 * the data that is contained within them.
989 error
= xfs_insert_file_space(ip
, offset
, len
);
992 xfs_iunlock(ip
, iolock
);
1002 if (!(file
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1004 if (XFS_FORCED_SHUTDOWN(XFS_M(inode
->i_sb
)))
1011 struct inode
*inode
,
1014 struct xfs_inode
*ip
= XFS_I(inode
);
1018 error
= xfs_file_open(inode
, file
);
1023 * If there are any blocks, read-ahead block 0 as we're almost
1024 * certain to have the next operation be a read there.
1026 mode
= xfs_ilock_data_map_shared(ip
);
1027 if (ip
->i_d
.di_nextents
> 0)
1028 xfs_dir3_data_readahead(ip
, 0, -1);
1029 xfs_iunlock(ip
, mode
);
1035 struct inode
*inode
,
1038 return xfs_release(XFS_I(inode
));
1044 struct dir_context
*ctx
)
1046 struct inode
*inode
= file_inode(file
);
1047 xfs_inode_t
*ip
= XFS_I(inode
);
1051 * The Linux API doesn't pass down the total size of the buffer
1052 * we read into down to the filesystem. With the filldir concept
1053 * it's not needed for correct information, but the XFS dir2 leaf
1054 * code wants an estimate of the buffer size to calculate it's
1055 * readahead window and size the buffers used for mapping to
1058 * Try to give it an estimate that's good enough, maybe at some
1059 * point we can change the ->readdir prototype to include the
1060 * buffer size. For now we use the current glibc buffer size.
1062 bufsize
= (size_t)min_t(loff_t
, 32768, ip
->i_d
.di_size
);
1064 return xfs_readdir(ip
, ctx
, bufsize
);
1070 struct vm_area_struct
*vma
)
1072 vma
->vm_ops
= &xfs_file_vm_ops
;
1074 file_accessed(filp
);
1079 * This type is designed to indicate the type of offset we would like
1080 * to search from page cache for xfs_seek_hole_data().
1088 * Lookup the desired type of offset from the given page.
1090 * On success, return true and the offset argument will point to the
1091 * start of the region that was found. Otherwise this function will
1092 * return false and keep the offset argument unchanged.
1095 xfs_lookup_buffer_offset(
1100 loff_t lastoff
= page_offset(page
);
1102 struct buffer_head
*bh
, *head
;
1104 bh
= head
= page_buffers(page
);
1107 * Unwritten extents that have data in the page
1108 * cache covering them can be identified by the
1109 * BH_Unwritten state flag. Pages with multiple
1110 * buffers might have a mix of holes, data and
1111 * unwritten extents - any buffer with valid
1112 * data in it should have BH_Uptodate flag set
1115 if (buffer_unwritten(bh
) ||
1116 buffer_uptodate(bh
)) {
1117 if (type
== DATA_OFF
)
1120 if (type
== HOLE_OFF
)
1128 lastoff
+= bh
->b_size
;
1129 } while ((bh
= bh
->b_this_page
) != head
);
1135 * This routine is called to find out and return a data or hole offset
1136 * from the page cache for unwritten extents according to the desired
1137 * type for xfs_seek_hole_data().
1139 * The argument offset is used to tell where we start to search from the
1140 * page cache. Map is used to figure out the end points of the range to
1143 * Return true if the desired type of offset was found, and the argument
1144 * offset is filled with that address. Otherwise, return false and keep
1148 xfs_find_get_desired_pgoff(
1149 struct inode
*inode
,
1150 struct xfs_bmbt_irec
*map
,
1154 struct xfs_inode
*ip
= XFS_I(inode
);
1155 struct xfs_mount
*mp
= ip
->i_mount
;
1156 struct pagevec pvec
;
1160 loff_t startoff
= *offset
;
1161 loff_t lastoff
= startoff
;
1164 pagevec_init(&pvec
, 0);
1166 index
= startoff
>> PAGE_CACHE_SHIFT
;
1167 endoff
= XFS_FSB_TO_B(mp
, map
->br_startoff
+ map
->br_blockcount
);
1168 end
= endoff
>> PAGE_CACHE_SHIFT
;
1174 want
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
);
1175 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
1178 * No page mapped into given range. If we are searching holes
1179 * and if this is the first time we got into the loop, it means
1180 * that the given offset is landed in a hole, return it.
1182 * If we have already stepped through some block buffers to find
1183 * holes but they all contains data. In this case, the last
1184 * offset is already updated and pointed to the end of the last
1185 * mapped page, if it does not reach the endpoint to search,
1186 * that means there should be a hole between them.
1188 if (nr_pages
== 0) {
1189 /* Data search found nothing */
1190 if (type
== DATA_OFF
)
1193 ASSERT(type
== HOLE_OFF
);
1194 if (lastoff
== startoff
|| lastoff
< endoff
) {
1202 * At lease we found one page. If this is the first time we
1203 * step into the loop, and if the first page index offset is
1204 * greater than the given search offset, a hole was found.
1206 if (type
== HOLE_OFF
&& lastoff
== startoff
&&
1207 lastoff
< page_offset(pvec
.pages
[0])) {
1212 for (i
= 0; i
< nr_pages
; i
++) {
1213 struct page
*page
= pvec
.pages
[i
];
1217 * At this point, the page may be truncated or
1218 * invalidated (changing page->mapping to NULL),
1219 * or even swizzled back from swapper_space to tmpfs
1220 * file mapping. However, page->index will not change
1221 * because we have a reference on the page.
1223 * Searching done if the page index is out of range.
1224 * If the current offset is not reaches the end of
1225 * the specified search range, there should be a hole
1228 if (page
->index
> end
) {
1229 if (type
== HOLE_OFF
&& lastoff
< endoff
) {
1238 * Page truncated or invalidated(page->mapping == NULL).
1239 * We can freely skip it and proceed to check the next
1242 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
1247 if (!page_has_buffers(page
)) {
1252 found
= xfs_lookup_buffer_offset(page
, &b_offset
, type
);
1255 * The found offset may be less than the start
1256 * point to search if this is the first time to
1259 *offset
= max_t(loff_t
, startoff
, b_offset
);
1265 * We either searching data but nothing was found, or
1266 * searching hole but found a data buffer. In either
1267 * case, probably the next page contains the desired
1268 * things, update the last offset to it so.
1270 lastoff
= page_offset(page
) + PAGE_SIZE
;
1275 * The number of returned pages less than our desired, search
1276 * done. In this case, nothing was found for searching data,
1277 * but we found a hole behind the last offset.
1279 if (nr_pages
< want
) {
1280 if (type
== HOLE_OFF
) {
1287 index
= pvec
.pages
[i
- 1]->index
+ 1;
1288 pagevec_release(&pvec
);
1289 } while (index
<= end
);
1292 pagevec_release(&pvec
);
1302 struct inode
*inode
= file
->f_mapping
->host
;
1303 struct xfs_inode
*ip
= XFS_I(inode
);
1304 struct xfs_mount
*mp
= ip
->i_mount
;
1305 loff_t
uninitialized_var(offset
);
1307 xfs_fileoff_t fsbno
;
1312 if (XFS_FORCED_SHUTDOWN(mp
))
1315 lock
= xfs_ilock_data_map_shared(ip
);
1317 isize
= i_size_read(inode
);
1318 if (start
>= isize
) {
1324 * Try to read extents from the first block indicated
1325 * by fsbno to the end block of the file.
1327 fsbno
= XFS_B_TO_FSBT(mp
, start
);
1328 end
= XFS_B_TO_FSB(mp
, isize
);
1331 struct xfs_bmbt_irec map
[2];
1335 error
= xfs_bmapi_read(ip
, fsbno
, end
- fsbno
, map
, &nmap
,
1340 /* No extents at given offset, must be beyond EOF */
1346 for (i
= 0; i
< nmap
; i
++) {
1347 offset
= max_t(loff_t
, start
,
1348 XFS_FSB_TO_B(mp
, map
[i
].br_startoff
));
1350 /* Landed in the hole we wanted? */
1351 if (whence
== SEEK_HOLE
&&
1352 map
[i
].br_startblock
== HOLESTARTBLOCK
)
1355 /* Landed in the data extent we wanted? */
1356 if (whence
== SEEK_DATA
&&
1357 (map
[i
].br_startblock
== DELAYSTARTBLOCK
||
1358 (map
[i
].br_state
== XFS_EXT_NORM
&&
1359 !isnullstartblock(map
[i
].br_startblock
))))
1363 * Landed in an unwritten extent, try to search
1364 * for hole or data from page cache.
1366 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
) {
1367 if (xfs_find_get_desired_pgoff(inode
, &map
[i
],
1368 whence
== SEEK_HOLE
? HOLE_OFF
: DATA_OFF
,
1375 * We only received one extent out of the two requested. This
1376 * means we've hit EOF and didn't find what we are looking for.
1380 * If we were looking for a hole, set offset to
1381 * the end of the file (i.e., there is an implicit
1382 * hole at the end of any file).
1384 if (whence
== SEEK_HOLE
) {
1389 * If we were looking for data, it's nowhere to be found
1391 ASSERT(whence
== SEEK_DATA
);
1399 * Nothing was found, proceed to the next round of search
1400 * if the next reading offset is not at or beyond EOF.
1402 fsbno
= map
[i
- 1].br_startoff
+ map
[i
- 1].br_blockcount
;
1403 start
= XFS_FSB_TO_B(mp
, fsbno
);
1404 if (start
>= isize
) {
1405 if (whence
== SEEK_HOLE
) {
1409 ASSERT(whence
== SEEK_DATA
);
1417 * If at this point we have found the hole we wanted, 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 if (whence
== SEEK_HOLE
)
1423 offset
= min_t(loff_t
, offset
, isize
);
1424 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
1427 xfs_iunlock(ip
, lock
);
1444 return generic_file_llseek(file
, offset
, whence
);
1447 return xfs_seek_hole_data(file
, offset
, whence
);
1454 * Locking for serialisation of IO during page faults. This results in a lock
1458 * i_mmap_lock (XFS - truncate serialisation)
1460 * i_lock (XFS - extent map serialisation)
1464 struct vm_area_struct
*vma
,
1465 struct vm_fault
*vmf
)
1467 struct xfs_inode
*ip
= XFS_I(vma
->vm_file
->f_mapping
->host
);
1470 trace_xfs_filemap_fault(ip
);
1472 xfs_ilock(ip
, XFS_MMAPLOCK_SHARED
);
1473 error
= filemap_fault(vma
, vmf
);
1474 xfs_iunlock(ip
, XFS_MMAPLOCK_SHARED
);
1480 * mmap()d file has taken write protection fault and is being made writable. We
1481 * can set the page state up correctly for a writable page, which means we can
1482 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1486 xfs_filemap_page_mkwrite(
1487 struct vm_area_struct
*vma
,
1488 struct vm_fault
*vmf
)
1490 struct xfs_inode
*ip
= XFS_I(vma
->vm_file
->f_mapping
->host
);
1493 trace_xfs_filemap_page_mkwrite(ip
);
1495 xfs_ilock(ip
, XFS_MMAPLOCK_SHARED
);
1496 error
= block_page_mkwrite(vma
, vmf
, xfs_get_blocks
);
1497 xfs_iunlock(ip
, XFS_MMAPLOCK_SHARED
);
1502 const struct file_operations xfs_file_operations
= {
1503 .llseek
= xfs_file_llseek
,
1504 .read_iter
= xfs_file_read_iter
,
1505 .write_iter
= xfs_file_write_iter
,
1506 .splice_read
= xfs_file_splice_read
,
1507 .splice_write
= iter_file_splice_write
,
1508 .unlocked_ioctl
= xfs_file_ioctl
,
1509 #ifdef CONFIG_COMPAT
1510 .compat_ioctl
= xfs_file_compat_ioctl
,
1512 .mmap
= xfs_file_mmap
,
1513 .open
= xfs_file_open
,
1514 .release
= xfs_file_release
,
1515 .fsync
= xfs_file_fsync
,
1516 .fallocate
= xfs_file_fallocate
,
1519 const struct file_operations xfs_dir_file_operations
= {
1520 .open
= xfs_dir_open
,
1521 .read
= generic_read_dir
,
1522 .iterate
= xfs_file_readdir
,
1523 .llseek
= generic_file_llseek
,
1524 .unlocked_ioctl
= xfs_file_ioctl
,
1525 #ifdef CONFIG_COMPAT
1526 .compat_ioctl
= xfs_file_compat_ioctl
,
1528 .fsync
= xfs_dir_fsync
,
1531 static const struct vm_operations_struct xfs_file_vm_ops
= {
1532 .fault
= xfs_filemap_fault
,
1533 .map_pages
= filemap_map_pages
,
1534 .page_mkwrite
= xfs_filemap_page_mkwrite
,