2 * Copyright (c) 2000-2003,2005 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"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
47 #include "xfs_inode_item.h"
48 #include "xfs_buf_item.h"
49 #include "xfs_utils.h"
50 #include "xfs_iomap.h"
52 #include <linux/capability.h>
53 #include <linux/writeback.h>
56 #if defined(XFS_RW_TRACE)
66 xfs_inode_t
*ip
= XFS_IO_INODE(io
);
68 if (ip
->i_rwtrace
== NULL
)
70 ktrace_enter(ip
->i_rwtrace
,
71 (void *)(unsigned long)tag
,
73 (void *)((unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff)),
74 (void *)((unsigned long)(ip
->i_d
.di_size
& 0xffffffff)),
76 (void *)((unsigned long)segs
),
77 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
78 (void *)((unsigned long)(offset
& 0xffffffff)),
79 (void *)((unsigned long)ioflags
),
80 (void *)((unsigned long)((io
->io_new_size
>> 32) & 0xffffffff)),
81 (void *)((unsigned long)(io
->io_new_size
& 0xffffffff)),
82 (void *)((unsigned long)current_pid()),
90 xfs_inval_cached_trace(
97 xfs_inode_t
*ip
= XFS_IO_INODE(io
);
99 if (ip
->i_rwtrace
== NULL
)
101 ktrace_enter(ip
->i_rwtrace
,
102 (void *)(__psint_t
)XFS_INVAL_CACHED
,
104 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
105 (void *)((unsigned long)(offset
& 0xffffffff)),
106 (void *)((unsigned long)((len
>> 32) & 0xffffffff)),
107 (void *)((unsigned long)(len
& 0xffffffff)),
108 (void *)((unsigned long)((first
>> 32) & 0xffffffff)),
109 (void *)((unsigned long)(first
& 0xffffffff)),
110 (void *)((unsigned long)((last
>> 32) & 0xffffffff)),
111 (void *)((unsigned long)(last
& 0xffffffff)),
112 (void *)((unsigned long)current_pid()),
124 * xfs_iozero clears the specified range of buffer supplied,
125 * and marks all the affected blocks as valid and modified. If
126 * an affected block is not allocated, it will be allocated. If
127 * an affected block is not completely overwritten, and is not
128 * valid before the operation, it will be read from disk before
129 * being partially zeroed.
133 struct inode
*ip
, /* inode */
134 loff_t pos
, /* offset in file */
135 size_t count
) /* size of data to zero */
138 struct address_space
*mapping
;
141 mapping
= ip
->i_mapping
;
143 unsigned offset
, bytes
;
146 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
147 bytes
= PAGE_CACHE_SIZE
- offset
;
151 status
= pagecache_write_begin(NULL
, mapping
, pos
, bytes
,
152 AOP_FLAG_UNINTERRUPTIBLE
,
157 zero_user_page(page
, offset
, bytes
, KM_USER0
);
159 status
= pagecache_write_end(NULL
, mapping
, pos
, bytes
, bytes
,
161 WARN_ON(status
<= 0); /* can't return less than zero! */
170 ssize_t
/* bytes read, or (-) error */
174 const struct iovec
*iovp
,
180 struct file
*file
= iocb
->ki_filp
;
181 struct inode
*inode
= file
->f_mapping
->host
;
190 ip
= XFS_BHVTOI(bdp
);
191 vp
= BHV_TO_VNODE(bdp
);
194 XFS_STATS_INC(xs_read_calls
);
196 /* START copy & waste from filemap.c */
197 for (seg
= 0; seg
< segs
; seg
++) {
198 const struct iovec
*iv
= &iovp
[seg
];
201 * If any segment has a negative length, or the cumulative
202 * length ever wraps negative then return -EINVAL.
205 if (unlikely((ssize_t
)(size
|iv
->iov_len
) < 0))
206 return XFS_ERROR(-EINVAL
);
208 /* END copy & waste from filemap.c */
210 if (unlikely(ioflags
& IO_ISDIRECT
)) {
211 xfs_buftarg_t
*target
=
212 (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ?
213 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
214 if ((*offset
& target
->bt_smask
) ||
215 (size
& target
->bt_smask
)) {
216 if (*offset
== ip
->i_size
) {
219 return -XFS_ERROR(EINVAL
);
223 n
= XFS_MAXIOFFSET(mp
) - *offset
;
224 if ((n
<= 0) || (size
== 0))
230 if (XFS_FORCED_SHUTDOWN(mp
))
233 if (unlikely(ioflags
& IO_ISDIRECT
))
234 mutex_lock(&inode
->i_mutex
);
235 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
237 if (DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_READ
) &&
238 !(ioflags
& IO_INVIS
)) {
239 bhv_vrwlock_t locktype
= VRWLOCK_READ
;
240 int dmflags
= FILP_DELAY_FLAG(file
) | DM_SEM_FLAG_RD(ioflags
);
242 ret
= -XFS_SEND_DATA(mp
, DM_EVENT_READ
,
243 BHV_TO_VNODE(bdp
), *offset
, size
,
246 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
247 if (unlikely(ioflags
& IO_ISDIRECT
))
248 mutex_unlock(&inode
->i_mutex
);
253 if (unlikely(ioflags
& IO_ISDIRECT
)) {
255 ret
= bhv_vop_flushinval_pages(vp
, ctooff(offtoct(*offset
)),
256 -1, FI_REMAPF_LOCKED
);
257 mutex_unlock(&inode
->i_mutex
);
259 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
264 xfs_rw_enter_trace(XFS_READ_ENTER
, &ip
->i_iocore
,
265 (void *)iovp
, segs
, *offset
, ioflags
);
267 iocb
->ki_pos
= *offset
;
268 ret
= generic_file_aio_read(iocb
, iovp
, segs
, *offset
);
269 if (ret
== -EIOCBQUEUED
&& !(ioflags
& IO_ISAIO
))
270 ret
= wait_on_sync_kiocb(iocb
);
272 XFS_STATS_ADD(xs_read_bytes
, ret
);
274 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
283 struct pipe_inode_info
*pipe
,
289 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
290 xfs_mount_t
*mp
= ip
->i_mount
;
293 XFS_STATS_INC(xs_read_calls
);
294 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
297 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
299 if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp
)->v_vfsp
, ip
, DM_EVENT_READ
) &&
300 (!(ioflags
& IO_INVIS
))) {
301 bhv_vrwlock_t locktype
= VRWLOCK_READ
;
304 error
= XFS_SEND_DATA(mp
, DM_EVENT_READ
, BHV_TO_VNODE(bdp
),
306 FILP_DELAY_FLAG(infilp
), &locktype
);
308 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
312 xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER
, &ip
->i_iocore
,
313 pipe
, count
, *ppos
, ioflags
);
314 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
316 XFS_STATS_ADD(xs_read_bytes
, ret
);
318 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
325 struct pipe_inode_info
*pipe
,
326 struct file
*outfilp
,
333 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
334 xfs_mount_t
*mp
= ip
->i_mount
;
335 xfs_iocore_t
*io
= &ip
->i_iocore
;
337 struct inode
*inode
= outfilp
->f_mapping
->host
;
338 xfs_fsize_t isize
, new_size
;
340 XFS_STATS_INC(xs_write_calls
);
341 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
344 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
346 if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp
)->v_vfsp
, ip
, DM_EVENT_WRITE
) &&
347 (!(ioflags
& IO_INVIS
))) {
348 bhv_vrwlock_t locktype
= VRWLOCK_WRITE
;
351 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, BHV_TO_VNODE(bdp
),
353 FILP_DELAY_FLAG(outfilp
), &locktype
);
355 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
360 new_size
= *ppos
+ count
;
362 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
363 if (new_size
> ip
->i_size
)
364 io
->io_new_size
= new_size
;
365 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
367 xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER
, &ip
->i_iocore
,
368 pipe
, count
, *ppos
, ioflags
);
369 ret
= generic_file_splice_write(pipe
, outfilp
, ppos
, count
, flags
);
371 XFS_STATS_ADD(xs_write_bytes
, ret
);
373 isize
= i_size_read(inode
);
374 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& *ppos
> isize
))
377 if (*ppos
> ip
->i_size
) {
378 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
379 if (*ppos
> ip
->i_size
)
381 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
384 if (io
->io_new_size
) {
385 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
387 if (ip
->i_d
.di_size
> ip
->i_size
)
388 ip
->i_d
.di_size
= ip
->i_size
;
389 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
391 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
396 * This routine is called to handle zeroing any space in the last
397 * block of the file that is beyond the EOF. We do this since the
398 * size is being increased without writing anything to that block
399 * and we don't want anyone to read the garbage on the disk.
401 STATIC
int /* error (positive) */
408 xfs_fileoff_t last_fsb
;
409 xfs_mount_t
*mp
= io
->io_mount
;
414 xfs_bmbt_irec_t imap
;
416 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
) != 0);
418 zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
419 if (zero_offset
== 0) {
421 * There are no extra bytes in the last block on disk to
427 last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
429 error
= XFS_BMAPI(mp
, NULL
, io
, last_fsb
, 1, 0, NULL
, 0, &imap
,
430 &nimaps
, NULL
, NULL
);
436 * If the block underlying isize is just a hole, then there
437 * is nothing to zero.
439 if (imap
.br_startblock
== HOLESTARTBLOCK
) {
443 * Zero the part of the last block beyond the EOF, and write it
444 * out sync. We need to drop the ilock while we do this so we
445 * don't deadlock when the buffer cache calls back to us.
447 XFS_IUNLOCK(mp
, io
, XFS_ILOCK_EXCL
| XFS_EXTSIZE_RD
);
449 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
450 if (isize
+ zero_len
> offset
)
451 zero_len
= offset
- isize
;
452 error
= xfs_iozero(ip
, isize
, zero_len
);
454 XFS_ILOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
460 * Zero any on disk space between the current EOF and the new,
461 * larger EOF. This handles the normal case of zeroing the remainder
462 * of the last block in the file and the unusual case of zeroing blocks
463 * out beyond the size of the file. This second case only happens
464 * with fixed size extents and when the system crashes before the inode
465 * size was updated but after blocks were allocated. If fill is set,
466 * then any holes in the range are filled and zeroed. If not, the holes
467 * are left alone as holes.
470 int /* error (positive) */
474 xfs_off_t offset
, /* starting I/O offset */
475 xfs_fsize_t isize
) /* current inode size */
477 struct inode
*ip
= vn_to_inode(vp
);
478 xfs_fileoff_t start_zero_fsb
;
479 xfs_fileoff_t end_zero_fsb
;
480 xfs_fileoff_t zero_count_fsb
;
481 xfs_fileoff_t last_fsb
;
482 xfs_fileoff_t zero_off
;
483 xfs_fsize_t zero_len
;
484 xfs_mount_t
*mp
= io
->io_mount
;
487 xfs_bmbt_irec_t imap
;
489 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
));
490 ASSERT(ismrlocked(io
->io_iolock
, MR_UPDATE
));
491 ASSERT(offset
> isize
);
494 * First handle zeroing the block on which isize resides.
495 * We only zero a part of that block so it is handled specially.
497 error
= xfs_zero_last_block(ip
, io
, offset
, isize
);
499 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
));
500 ASSERT(ismrlocked(io
->io_iolock
, MR_UPDATE
));
505 * Calculate the range between the new size and the old
506 * where blocks needing to be zeroed may exist. To get the
507 * block where the last byte in the file currently resides,
508 * we need to subtract one from the size and truncate back
509 * to a block boundary. We subtract 1 in case the size is
510 * exactly on a block boundary.
512 last_fsb
= isize
? XFS_B_TO_FSBT(mp
, isize
- 1) : (xfs_fileoff_t
)-1;
513 start_zero_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
514 end_zero_fsb
= XFS_B_TO_FSBT(mp
, offset
- 1);
515 ASSERT((xfs_sfiloff_t
)last_fsb
< (xfs_sfiloff_t
)start_zero_fsb
);
516 if (last_fsb
== end_zero_fsb
) {
518 * The size was only incremented on its last block.
519 * We took care of that above, so just return.
524 ASSERT(start_zero_fsb
<= end_zero_fsb
);
525 while (start_zero_fsb
<= end_zero_fsb
) {
527 zero_count_fsb
= end_zero_fsb
- start_zero_fsb
+ 1;
528 error
= XFS_BMAPI(mp
, NULL
, io
, start_zero_fsb
, zero_count_fsb
,
529 0, NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
531 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
));
532 ASSERT(ismrlocked(io
->io_iolock
, MR_UPDATE
));
537 if (imap
.br_state
== XFS_EXT_UNWRITTEN
||
538 imap
.br_startblock
== HOLESTARTBLOCK
) {
540 * This loop handles initializing pages that were
541 * partially initialized by the code below this
542 * loop. It basically zeroes the part of the page
543 * that sits on a hole and sets the page as P_HOLE
544 * and calls remapf if it is a mapped file.
546 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
547 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
552 * There are blocks we need to zero.
553 * Drop the inode lock while we're doing the I/O.
554 * We'll still have the iolock to protect us.
556 XFS_IUNLOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
558 zero_off
= XFS_FSB_TO_B(mp
, start_zero_fsb
);
559 zero_len
= XFS_FSB_TO_B(mp
, imap
.br_blockcount
);
561 if ((zero_off
+ zero_len
) > offset
)
562 zero_len
= offset
- zero_off
;
564 error
= xfs_iozero(ip
, zero_off
, zero_len
);
569 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
570 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
572 XFS_ILOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
579 XFS_ILOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
584 ssize_t
/* bytes written, or (-) error */
588 const struct iovec
*iovp
,
594 struct file
*file
= iocb
->ki_filp
;
595 struct address_space
*mapping
= file
->f_mapping
;
596 struct inode
*inode
= mapping
->host
;
597 unsigned long segs
= nsegs
;
600 ssize_t ret
= 0, error
= 0;
601 xfs_fsize_t isize
, new_size
;
606 bhv_vrwlock_t locktype
;
607 size_t ocount
= 0, count
;
611 XFS_STATS_INC(xs_write_calls
);
613 vp
= BHV_TO_VNODE(bdp
);
614 xip
= XFS_BHVTOI(bdp
);
616 error
= generic_segment_checks(iovp
, &segs
, &ocount
, VERIFY_READ
);
629 vfs_wait_for_freeze(vp
->v_vfsp
, SB_FREEZE_WRITE
);
631 if (XFS_FORCED_SHUTDOWN(mp
))
635 if (ioflags
& IO_ISDIRECT
) {
636 iolock
= XFS_IOLOCK_SHARED
;
637 locktype
= VRWLOCK_WRITE_DIRECT
;
640 iolock
= XFS_IOLOCK_EXCL
;
641 locktype
= VRWLOCK_WRITE
;
643 mutex_lock(&inode
->i_mutex
);
646 xfs_ilock(xip
, XFS_ILOCK_EXCL
|iolock
);
649 error
= -generic_write_checks(file
, &pos
, &count
,
650 S_ISBLK(inode
->i_mode
));
652 xfs_iunlock(xip
, XFS_ILOCK_EXCL
|iolock
);
653 goto out_unlock_mutex
;
656 if ((DM_EVENT_ENABLED(vp
->v_vfsp
, xip
, DM_EVENT_WRITE
) &&
657 !(ioflags
& IO_INVIS
) && !eventsent
)) {
658 int dmflags
= FILP_DELAY_FLAG(file
);
661 dmflags
|= DM_FLAGS_IMUX
;
663 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
664 error
= XFS_SEND_DATA(xip
->i_mount
, DM_EVENT_WRITE
, vp
,
668 goto out_unlock_internal
;
670 xfs_ilock(xip
, XFS_ILOCK_EXCL
);
674 * The iolock was dropped and reacquired in XFS_SEND_DATA
675 * so we have to recheck the size when appending.
676 * We will only "goto start;" once, since having sent the
677 * event prevents another call to XFS_SEND_DATA, which is
678 * what allows the size to change in the first place.
680 if ((file
->f_flags
& O_APPEND
) && pos
!= xip
->i_size
)
684 if (ioflags
& IO_ISDIRECT
) {
685 xfs_buftarg_t
*target
=
686 (xip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ?
687 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
689 if ((pos
& target
->bt_smask
) || (count
& target
->bt_smask
)) {
690 xfs_iunlock(xip
, XFS_ILOCK_EXCL
|iolock
);
691 return XFS_ERROR(-EINVAL
);
694 if (!need_i_mutex
&& (VN_CACHED(vp
) || pos
> xip
->i_size
)) {
695 xfs_iunlock(xip
, XFS_ILOCK_EXCL
|iolock
);
696 iolock
= XFS_IOLOCK_EXCL
;
697 locktype
= VRWLOCK_WRITE
;
699 mutex_lock(&inode
->i_mutex
);
700 xfs_ilock(xip
, XFS_ILOCK_EXCL
|iolock
);
705 new_size
= pos
+ count
;
706 if (new_size
> xip
->i_size
)
707 io
->io_new_size
= new_size
;
709 if (likely(!(ioflags
& IO_INVIS
))) {
710 file_update_time(file
);
711 xfs_ichgtime_fast(xip
, inode
,
712 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
716 * If the offset is beyond the size of the file, we have a couple
717 * of things to do. First, if there is already space allocated
718 * we need to either create holes or zero the disk or ...
720 * If there is a page where the previous size lands, we need
721 * to zero it out up to the new size.
724 if (pos
> xip
->i_size
) {
725 error
= xfs_zero_eof(BHV_TO_VNODE(bdp
), io
, pos
, xip
->i_size
);
727 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
728 goto out_unlock_internal
;
731 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
734 * If we're writing the file then make sure to clear the
735 * setuid and setgid bits if the process is not being run
736 * by root. This keeps people from modifying setuid and
740 if (((xip
->i_d
.di_mode
& S_ISUID
) ||
741 ((xip
->i_d
.di_mode
& (S_ISGID
| S_IXGRP
)) ==
742 (S_ISGID
| S_IXGRP
))) &&
743 !capable(CAP_FSETID
)) {
744 error
= xfs_write_clear_setuid(xip
);
746 error
= -remove_suid(file
->f_path
.dentry
);
747 if (unlikely(error
)) {
748 goto out_unlock_internal
;
753 /* We can write back this queue in page reclaim */
754 current
->backing_dev_info
= mapping
->backing_dev_info
;
756 if ((ioflags
& IO_ISDIRECT
)) {
758 WARN_ON(need_i_mutex
== 0);
759 xfs_inval_cached_trace(io
, pos
, -1,
760 ctooff(offtoct(pos
)), -1);
761 error
= bhv_vop_flushinval_pages(vp
, ctooff(offtoct(pos
)),
762 -1, FI_REMAPF_LOCKED
);
764 goto out_unlock_internal
;
768 /* demote the lock now the cached pages are gone */
769 XFS_ILOCK_DEMOTE(mp
, io
, XFS_IOLOCK_EXCL
);
770 mutex_unlock(&inode
->i_mutex
);
772 iolock
= XFS_IOLOCK_SHARED
;
773 locktype
= VRWLOCK_WRITE_DIRECT
;
777 xfs_rw_enter_trace(XFS_DIOWR_ENTER
, io
, (void *)iovp
, segs
,
779 ret
= generic_file_direct_write(iocb
, iovp
,
780 &segs
, pos
, offset
, count
, ocount
);
783 * direct-io write to a hole: fall through to buffered I/O
784 * for completing the rest of the request.
786 if (ret
>= 0 && ret
!= count
) {
787 XFS_STATS_ADD(xs_write_bytes
, ret
);
792 ioflags
&= ~IO_ISDIRECT
;
793 xfs_iunlock(xip
, iolock
);
797 xfs_rw_enter_trace(XFS_WRITE_ENTER
, io
, (void *)iovp
, segs
,
799 ret
= generic_file_buffered_write(iocb
, iovp
, segs
,
800 pos
, offset
, count
, ret
);
803 current
->backing_dev_info
= NULL
;
805 if (ret
== -EIOCBQUEUED
&& !(ioflags
& IO_ISAIO
))
806 ret
= wait_on_sync_kiocb(iocb
);
808 if ((ret
== -ENOSPC
) &&
809 DM_EVENT_ENABLED(vp
->v_vfsp
, xip
, DM_EVENT_NOSPACE
) &&
810 !(ioflags
& IO_INVIS
)) {
812 xfs_rwunlock(bdp
, locktype
);
814 mutex_unlock(&inode
->i_mutex
);
815 error
= XFS_SEND_NAMESP(xip
->i_mount
, DM_EVENT_NOSPACE
, vp
,
816 DM_RIGHT_NULL
, vp
, DM_RIGHT_NULL
, NULL
, NULL
,
817 0, 0, 0); /* Delay flag intentionally unused */
819 mutex_lock(&inode
->i_mutex
);
820 xfs_rwlock(bdp
, locktype
);
822 goto out_unlock_internal
;
828 isize
= i_size_read(inode
);
829 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& *offset
> isize
))
832 if (*offset
> xip
->i_size
) {
833 xfs_ilock(xip
, XFS_ILOCK_EXCL
);
834 if (*offset
> xip
->i_size
)
835 xip
->i_size
= *offset
;
836 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
841 goto out_unlock_internal
;
843 XFS_STATS_ADD(xs_write_bytes
, ret
);
845 /* Handle various SYNC-type writes */
846 if ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
)) {
847 error
= xfs_write_sync_logforce(mp
, xip
);
849 goto out_unlock_internal
;
851 xfs_rwunlock(bdp
, locktype
);
853 mutex_unlock(&inode
->i_mutex
);
855 error
= sync_page_range(inode
, mapping
, pos
, ret
);
859 mutex_lock(&inode
->i_mutex
);
860 xfs_rwlock(bdp
, locktype
);
864 if (io
->io_new_size
) {
865 xfs_ilock(xip
, XFS_ILOCK_EXCL
);
868 * If this was a direct or synchronous I/O that failed (such
869 * as ENOSPC) then part of the I/O may have been written to
870 * disk before the error occured. In this case the on-disk
871 * file size may have been adjusted beyond the in-memory file
872 * size and now needs to be truncated back.
874 if (xip
->i_d
.di_size
> xip
->i_size
)
875 xip
->i_d
.di_size
= xip
->i_size
;
876 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
878 xfs_rwunlock(bdp
, locktype
);
881 mutex_unlock(&inode
->i_mutex
);
886 * All xfs metadata buffers except log state machine buffers
887 * get this attached as their b_bdstrat callback function.
888 * This is so that we can catch a buffer
889 * after prematurely unpinning it to forcibly shutdown the filesystem.
892 xfs_bdstrat_cb(struct xfs_buf
*bp
)
896 mp
= XFS_BUF_FSPRIVATE3(bp
, xfs_mount_t
*);
897 if (!XFS_FORCED_SHUTDOWN(mp
)) {
898 xfs_buf_iorequest(bp
);
901 xfs_buftrace("XFS__BDSTRAT IOERROR", bp
);
903 * Metadata write that didn't get logged but
904 * written delayed anyway. These aren't associated
905 * with a transaction, and can be ignored.
907 if (XFS_BUF_IODONE_FUNC(bp
) == NULL
&&
908 (XFS_BUF_ISREAD(bp
)) == 0)
909 return (xfs_bioerror_relse(bp
));
911 return (xfs_bioerror(bp
));
917 xfs_bmap(bhv_desc_t
*bdp
,
924 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
925 xfs_iocore_t
*io
= &ip
->i_iocore
;
927 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
);
928 ASSERT(((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) != 0) ==
929 ((ip
->i_iocore
.io_flags
& XFS_IOCORE_RT
) != 0));
931 return xfs_iomap(io
, offset
, count
, flags
, iomapp
, niomaps
);
935 * Wrapper around bdstrat so that we can stop data
936 * from going to disk in case we are shutting down the filesystem.
937 * Typically user data goes thru this path; one of the exceptions
942 struct xfs_mount
*mp
,
946 if (!XFS_FORCED_SHUTDOWN(mp
)) {
947 /* Grio redirection would go here
948 * if (XFS_BUF_IS_GRIO(bp)) {
951 xfs_buf_iorequest(bp
);
955 xfs_buftrace("XFSBDSTRAT IOERROR", bp
);
956 return (xfs_bioerror_relse(bp
));
960 * If the underlying (data/log/rt) device is readonly, there are some
961 * operations that cannot proceed.
964 xfs_dev_is_read_only(
968 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
969 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
970 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
972 "XFS: %s required on read-only device.", message
);
974 "XFS: write access unavailable, cannot proceed.");