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"
49 #include "xfs_inode_item.h"
50 #include "xfs_buf_item.h"
51 #include "xfs_utils.h"
52 #include "xfs_iomap.h"
54 #include <linux/capability.h>
55 #include <linux/writeback.h>
58 #if defined(XFS_RW_TRACE)
68 xfs_inode_t
*ip
= XFS_IO_INODE(io
);
70 if (ip
->i_rwtrace
== NULL
)
72 ktrace_enter(ip
->i_rwtrace
,
73 (void *)(unsigned long)tag
,
75 (void *)((unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff)),
76 (void *)((unsigned long)(ip
->i_d
.di_size
& 0xffffffff)),
78 (void *)((unsigned long)segs
),
79 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
80 (void *)((unsigned long)(offset
& 0xffffffff)),
81 (void *)((unsigned long)ioflags
),
82 (void *)((unsigned long)((io
->io_new_size
>> 32) & 0xffffffff)),
83 (void *)((unsigned long)(io
->io_new_size
& 0xffffffff)),
84 (void *)((unsigned long)current_pid()),
92 xfs_inval_cached_trace(
99 xfs_inode_t
*ip
= XFS_IO_INODE(io
);
101 if (ip
->i_rwtrace
== NULL
)
103 ktrace_enter(ip
->i_rwtrace
,
104 (void *)(__psint_t
)XFS_INVAL_CACHED
,
106 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
107 (void *)((unsigned long)(offset
& 0xffffffff)),
108 (void *)((unsigned long)((len
>> 32) & 0xffffffff)),
109 (void *)((unsigned long)(len
& 0xffffffff)),
110 (void *)((unsigned long)((first
>> 32) & 0xffffffff)),
111 (void *)((unsigned long)(first
& 0xffffffff)),
112 (void *)((unsigned long)((last
>> 32) & 0xffffffff)),
113 (void *)((unsigned long)(last
& 0xffffffff)),
114 (void *)((unsigned long)current_pid()),
126 * xfs_iozero clears the specified range of buffer supplied,
127 * and marks all the affected blocks as valid and modified. If
128 * an affected block is not allocated, it will be allocated. If
129 * an affected block is not completely overwritten, and is not
130 * valid before the operation, it will be read from disk before
131 * being partially zeroed.
135 struct inode
*ip
, /* inode */
136 loff_t pos
, /* offset in file */
137 size_t count
, /* size of data to zero */
138 loff_t end_size
) /* max file size to set */
142 struct address_space
*mapping
;
146 mapping
= ip
->i_mapping
;
148 unsigned long index
, offset
;
150 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
151 index
= pos
>> PAGE_CACHE_SHIFT
;
152 bytes
= PAGE_CACHE_SIZE
- offset
;
157 page
= grab_cache_page(mapping
, index
);
162 status
= mapping
->a_ops
->prepare_write(NULL
, page
, offset
,
168 memset((void *) (kaddr
+ offset
), 0, bytes
);
169 flush_dcache_page(page
);
170 status
= mapping
->a_ops
->commit_write(NULL
, page
, offset
,
175 if (pos
> i_size_read(ip
))
176 i_size_write(ip
, pos
< end_size
? pos
: end_size
);
182 page_cache_release(page
);
190 ssize_t
/* bytes read, or (-) error */
194 const struct iovec
*iovp
,
200 struct file
*file
= iocb
->ki_filp
;
201 struct inode
*inode
= file
->f_mapping
->host
;
210 ip
= XFS_BHVTOI(bdp
);
211 vp
= BHV_TO_VNODE(bdp
);
214 XFS_STATS_INC(xs_read_calls
);
216 /* START copy & waste from filemap.c */
217 for (seg
= 0; seg
< segs
; seg
++) {
218 const struct iovec
*iv
= &iovp
[seg
];
221 * If any segment has a negative length, or the cumulative
222 * length ever wraps negative then return -EINVAL.
225 if (unlikely((ssize_t
)(size
|iv
->iov_len
) < 0))
226 return XFS_ERROR(-EINVAL
);
228 /* END copy & waste from filemap.c */
230 if (unlikely(ioflags
& IO_ISDIRECT
)) {
231 xfs_buftarg_t
*target
=
232 (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ?
233 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
234 if ((*offset
& target
->bt_smask
) ||
235 (size
& target
->bt_smask
)) {
236 if (*offset
== ip
->i_d
.di_size
) {
239 return -XFS_ERROR(EINVAL
);
243 n
= XFS_MAXIOFFSET(mp
) - *offset
;
244 if ((n
<= 0) || (size
== 0))
250 if (XFS_FORCED_SHUTDOWN(mp
))
253 if (unlikely(ioflags
& IO_ISDIRECT
))
254 mutex_lock(&inode
->i_mutex
);
255 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
257 if (DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_READ
) &&
258 !(ioflags
& IO_INVIS
)) {
259 bhv_vrwlock_t locktype
= VRWLOCK_READ
;
260 int dmflags
= FILP_DELAY_FLAG(file
) | DM_SEM_FLAG_RD(ioflags
);
262 ret
= -XFS_SEND_DATA(mp
, DM_EVENT_READ
,
263 BHV_TO_VNODE(bdp
), *offset
, size
,
266 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
271 if (unlikely((ioflags
& IO_ISDIRECT
) && VN_CACHED(vp
)))
272 bhv_vop_flushinval_pages(vp
, ctooff(offtoct(*offset
)),
273 -1, FI_REMAPF_LOCKED
);
275 xfs_rw_enter_trace(XFS_READ_ENTER
, &ip
->i_iocore
,
276 (void *)iovp
, segs
, *offset
, ioflags
);
277 ret
= __generic_file_aio_read(iocb
, iovp
, segs
, offset
);
278 if (ret
== -EIOCBQUEUED
&& !(ioflags
& IO_ISAIO
))
279 ret
= wait_on_sync_kiocb(iocb
);
281 XFS_STATS_ADD(xs_read_bytes
, ret
);
283 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
286 if (unlikely(ioflags
& IO_ISDIRECT
))
287 mutex_unlock(&inode
->i_mutex
);
302 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
303 xfs_mount_t
*mp
= ip
->i_mount
;
306 XFS_STATS_INC(xs_read_calls
);
307 if (XFS_FORCED_SHUTDOWN(mp
))
310 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
312 if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp
)->v_vfsp
, ip
, DM_EVENT_READ
) &&
313 (!(ioflags
& IO_INVIS
))) {
314 bhv_vrwlock_t locktype
= VRWLOCK_READ
;
317 error
= XFS_SEND_DATA(mp
, DM_EVENT_READ
, BHV_TO_VNODE(bdp
),
319 FILP_DELAY_FLAG(filp
), &locktype
);
321 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
325 xfs_rw_enter_trace(XFS_SENDFILE_ENTER
, &ip
->i_iocore
,
326 (void *)(unsigned long)target
, count
, *offset
, ioflags
);
327 ret
= generic_file_sendfile(filp
, offset
, count
, actor
, target
);
329 XFS_STATS_ADD(xs_read_bytes
, ret
);
331 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
340 struct pipe_inode_info
*pipe
,
346 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
347 xfs_mount_t
*mp
= ip
->i_mount
;
350 XFS_STATS_INC(xs_read_calls
);
351 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
354 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
356 if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp
)->v_vfsp
, ip
, DM_EVENT_READ
) &&
357 (!(ioflags
& IO_INVIS
))) {
358 bhv_vrwlock_t locktype
= VRWLOCK_READ
;
361 error
= XFS_SEND_DATA(mp
, DM_EVENT_READ
, BHV_TO_VNODE(bdp
),
363 FILP_DELAY_FLAG(infilp
), &locktype
);
365 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
369 xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER
, &ip
->i_iocore
,
370 pipe
, count
, *ppos
, ioflags
);
371 ret
= generic_file_splice_read(infilp
, ppos
, pipe
, count
, flags
);
373 XFS_STATS_ADD(xs_read_bytes
, ret
);
375 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
382 struct pipe_inode_info
*pipe
,
383 struct file
*outfilp
,
390 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
391 xfs_mount_t
*mp
= ip
->i_mount
;
394 XFS_STATS_INC(xs_write_calls
);
395 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
398 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
400 if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp
)->v_vfsp
, ip
, DM_EVENT_WRITE
) &&
401 (!(ioflags
& IO_INVIS
))) {
402 bhv_vrwlock_t locktype
= VRWLOCK_WRITE
;
405 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, BHV_TO_VNODE(bdp
),
407 FILP_DELAY_FLAG(outfilp
), &locktype
);
409 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
413 xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER
, &ip
->i_iocore
,
414 pipe
, count
, *ppos
, ioflags
);
415 ret
= generic_file_splice_write(pipe
, outfilp
, ppos
, count
, flags
);
417 XFS_STATS_ADD(xs_write_bytes
, ret
);
419 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
424 * This routine is called to handle zeroing any space in the last
425 * block of the file that is beyond the EOF. We do this since the
426 * size is being increased without writing anything to that block
427 * and we don't want anyone to read the garbage on the disk.
429 STATIC
int /* error (positive) */
434 xfs_fsize_t end_size
)
436 xfs_fileoff_t last_fsb
;
437 xfs_mount_t
*mp
= io
->io_mount
;
442 xfs_bmbt_irec_t imap
;
445 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
) != 0);
447 zero_offset
= XFS_B_FSB_OFFSET(mp
, isize
);
448 if (zero_offset
== 0) {
450 * There are no extra bytes in the last block on disk to
456 last_fsb
= XFS_B_TO_FSBT(mp
, isize
);
458 error
= XFS_BMAPI(mp
, NULL
, io
, last_fsb
, 1, 0, NULL
, 0, &imap
,
459 &nimaps
, NULL
, NULL
);
465 * If the block underlying isize is just a hole, then there
466 * is nothing to zero.
468 if (imap
.br_startblock
== HOLESTARTBLOCK
) {
472 * Zero the part of the last block beyond the EOF, and write it
473 * out sync. We need to drop the ilock while we do this so we
474 * don't deadlock when the buffer cache calls back to us.
476 XFS_IUNLOCK(mp
, io
, XFS_ILOCK_EXCL
| XFS_EXTSIZE_RD
);
478 loff
= XFS_FSB_TO_B(mp
, last_fsb
);
479 zero_len
= mp
->m_sb
.sb_blocksize
- zero_offset
;
480 error
= xfs_iozero(ip
, loff
+ zero_offset
, zero_len
, end_size
);
482 XFS_ILOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
488 * Zero any on disk space between the current EOF and the new,
489 * larger EOF. This handles the normal case of zeroing the remainder
490 * of the last block in the file and the unusual case of zeroing blocks
491 * out beyond the size of the file. This second case only happens
492 * with fixed size extents and when the system crashes before the inode
493 * size was updated but after blocks were allocated. If fill is set,
494 * then any holes in the range are filled and zeroed. If not, the holes
495 * are left alone as holes.
498 int /* error (positive) */
502 xfs_off_t offset
, /* starting I/O offset */
503 xfs_fsize_t isize
, /* current inode size */
504 xfs_fsize_t end_size
) /* terminal inode size */
506 struct inode
*ip
= vn_to_inode(vp
);
507 xfs_fileoff_t start_zero_fsb
;
508 xfs_fileoff_t end_zero_fsb
;
509 xfs_fileoff_t zero_count_fsb
;
510 xfs_fileoff_t last_fsb
;
511 xfs_mount_t
*mp
= io
->io_mount
;
514 xfs_bmbt_irec_t imap
;
516 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
));
517 ASSERT(ismrlocked(io
->io_iolock
, MR_UPDATE
));
518 ASSERT(offset
> isize
);
521 * First handle zeroing the block on which isize resides.
522 * We only zero a part of that block so it is handled specially.
524 error
= xfs_zero_last_block(ip
, io
, isize
, end_size
);
526 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
));
527 ASSERT(ismrlocked(io
->io_iolock
, MR_UPDATE
));
532 * Calculate the range between the new size and the old
533 * where blocks needing to be zeroed may exist. To get the
534 * block where the last byte in the file currently resides,
535 * we need to subtract one from the size and truncate back
536 * to a block boundary. We subtract 1 in case the size is
537 * exactly on a block boundary.
539 last_fsb
= isize
? XFS_B_TO_FSBT(mp
, isize
- 1) : (xfs_fileoff_t
)-1;
540 start_zero_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
541 end_zero_fsb
= XFS_B_TO_FSBT(mp
, offset
- 1);
542 ASSERT((xfs_sfiloff_t
)last_fsb
< (xfs_sfiloff_t
)start_zero_fsb
);
543 if (last_fsb
== end_zero_fsb
) {
545 * The size was only incremented on its last block.
546 * We took care of that above, so just return.
551 ASSERT(start_zero_fsb
<= end_zero_fsb
);
552 while (start_zero_fsb
<= end_zero_fsb
) {
554 zero_count_fsb
= end_zero_fsb
- start_zero_fsb
+ 1;
555 error
= XFS_BMAPI(mp
, NULL
, io
, start_zero_fsb
, zero_count_fsb
,
556 0, NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
558 ASSERT(ismrlocked(io
->io_lock
, MR_UPDATE
));
559 ASSERT(ismrlocked(io
->io_iolock
, MR_UPDATE
));
564 if (imap
.br_state
== XFS_EXT_UNWRITTEN
||
565 imap
.br_startblock
== HOLESTARTBLOCK
) {
567 * This loop handles initializing pages that were
568 * partially initialized by the code below this
569 * loop. It basically zeroes the part of the page
570 * that sits on a hole and sets the page as P_HOLE
571 * and calls remapf if it is a mapped file.
573 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
574 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
579 * There are blocks we need to zero.
580 * Drop the inode lock while we're doing the I/O.
581 * We'll still have the iolock to protect us.
583 XFS_IUNLOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
585 error
= xfs_iozero(ip
,
586 XFS_FSB_TO_B(mp
, start_zero_fsb
),
587 XFS_FSB_TO_B(mp
, imap
.br_blockcount
),
593 start_zero_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
594 ASSERT(start_zero_fsb
<= (end_zero_fsb
+ 1));
596 XFS_ILOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
603 XFS_ILOCK(mp
, io
, XFS_ILOCK_EXCL
|XFS_EXTSIZE_RD
);
608 ssize_t
/* bytes written, or (-) error */
612 const struct iovec
*iovp
,
618 struct file
*file
= iocb
->ki_filp
;
619 struct address_space
*mapping
= file
->f_mapping
;
620 struct inode
*inode
= mapping
->host
;
621 unsigned long segs
= nsegs
;
624 ssize_t ret
= 0, error
= 0;
625 xfs_fsize_t isize
, new_size
;
631 bhv_vrwlock_t locktype
;
632 size_t ocount
= 0, count
;
634 int need_i_mutex
= 1, need_flush
= 0;
636 XFS_STATS_INC(xs_write_calls
);
638 vp
= BHV_TO_VNODE(bdp
);
639 xip
= XFS_BHVTOI(bdp
);
641 for (seg
= 0; seg
< segs
; seg
++) {
642 const struct iovec
*iv
= &iovp
[seg
];
645 * If any segment has a negative length, or the cumulative
646 * length ever wraps negative then return -EINVAL.
648 ocount
+= iv
->iov_len
;
649 if (unlikely((ssize_t
)(ocount
|iv
->iov_len
) < 0))
651 if (access_ok(VERIFY_READ
, iv
->iov_base
, iv
->iov_len
))
656 ocount
-= iv
->iov_len
; /* This segment is no good */
669 vfs_wait_for_freeze(vp
->v_vfsp
, SB_FREEZE_WRITE
);
671 if (XFS_FORCED_SHUTDOWN(mp
))
674 if (ioflags
& IO_ISDIRECT
) {
675 xfs_buftarg_t
*target
=
676 (xip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ?
677 mp
->m_rtdev_targp
: mp
->m_ddev_targp
;
679 if ((pos
& target
->bt_smask
) || (count
& target
->bt_smask
))
680 return XFS_ERROR(-EINVAL
);
682 if (!VN_CACHED(vp
) && pos
< i_size_read(inode
))
691 iolock
= XFS_IOLOCK_EXCL
;
692 locktype
= VRWLOCK_WRITE
;
694 mutex_lock(&inode
->i_mutex
);
696 iolock
= XFS_IOLOCK_SHARED
;
697 locktype
= VRWLOCK_WRITE_DIRECT
;
700 xfs_ilock(xip
, XFS_ILOCK_EXCL
|iolock
);
702 isize
= i_size_read(inode
);
704 if (file
->f_flags
& O_APPEND
)
708 error
= -generic_write_checks(file
, &pos
, &count
,
709 S_ISBLK(inode
->i_mode
));
711 xfs_iunlock(xip
, XFS_ILOCK_EXCL
|iolock
);
712 goto out_unlock_mutex
;
715 new_size
= pos
+ count
;
716 if (new_size
> isize
)
717 io
->io_new_size
= new_size
;
719 if ((DM_EVENT_ENABLED(vp
->v_vfsp
, xip
, DM_EVENT_WRITE
) &&
720 !(ioflags
& IO_INVIS
) && !eventsent
)) {
721 loff_t savedsize
= pos
;
722 int dmflags
= FILP_DELAY_FLAG(file
);
725 dmflags
|= DM_FLAGS_IMUX
;
727 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
728 error
= XFS_SEND_DATA(xip
->i_mount
, DM_EVENT_WRITE
, vp
,
732 xfs_iunlock(xip
, iolock
);
733 goto out_unlock_mutex
;
735 xfs_ilock(xip
, XFS_ILOCK_EXCL
);
739 * The iolock was dropped and reacquired in XFS_SEND_DATA
740 * so we have to recheck the size when appending.
741 * We will only "goto start;" once, since having sent the
742 * event prevents another call to XFS_SEND_DATA, which is
743 * what allows the size to change in the first place.
745 if ((file
->f_flags
& O_APPEND
) && savedsize
!= isize
) {
746 pos
= isize
= xip
->i_d
.di_size
;
751 if (likely(!(ioflags
& IO_INVIS
))) {
752 file_update_time(file
);
753 xfs_ichgtime_fast(xip
, inode
,
754 XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
758 * If the offset is beyond the size of the file, we have a couple
759 * of things to do. First, if there is already space allocated
760 * we need to either create holes or zero the disk or ...
762 * If there is a page where the previous size lands, we need
763 * to zero it out up to the new size.
767 error
= xfs_zero_eof(BHV_TO_VNODE(bdp
), io
, pos
,
770 xfs_iunlock(xip
, XFS_ILOCK_EXCL
|iolock
);
771 goto out_unlock_mutex
;
774 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
777 * If we're writing the file then make sure to clear the
778 * setuid and setgid bits if the process is not being run
779 * by root. This keeps people from modifying setuid and
783 if (((xip
->i_d
.di_mode
& S_ISUID
) ||
784 ((xip
->i_d
.di_mode
& (S_ISGID
| S_IXGRP
)) ==
785 (S_ISGID
| S_IXGRP
))) &&
786 !capable(CAP_FSETID
)) {
787 error
= xfs_write_clear_setuid(xip
);
789 error
= -remove_suid(file
->f_dentry
);
790 if (unlikely(error
)) {
791 xfs_iunlock(xip
, iolock
);
792 goto out_unlock_mutex
;
797 /* We can write back this queue in page reclaim */
798 current
->backing_dev_info
= mapping
->backing_dev_info
;
800 if ((ioflags
& IO_ISDIRECT
)) {
802 xfs_inval_cached_trace(io
, pos
, -1,
803 ctooff(offtoct(pos
)), -1);
804 bhv_vop_flushinval_pages(vp
, ctooff(offtoct(pos
)),
805 -1, FI_REMAPF_LOCKED
);
809 /* demote the lock now the cached pages are gone */
810 XFS_ILOCK_DEMOTE(mp
, io
, XFS_IOLOCK_EXCL
);
811 mutex_unlock(&inode
->i_mutex
);
813 iolock
= XFS_IOLOCK_SHARED
;
814 locktype
= VRWLOCK_WRITE_DIRECT
;
818 xfs_rw_enter_trace(XFS_DIOWR_ENTER
, io
, (void *)iovp
, segs
,
820 ret
= generic_file_direct_write(iocb
, iovp
,
821 &segs
, pos
, offset
, count
, ocount
);
824 * direct-io write to a hole: fall through to buffered I/O
825 * for completing the rest of the request.
827 if (ret
>= 0 && ret
!= count
) {
828 XFS_STATS_ADD(xs_write_bytes
, ret
);
834 ioflags
&= ~IO_ISDIRECT
;
835 xfs_iunlock(xip
, iolock
);
839 xfs_rw_enter_trace(XFS_WRITE_ENTER
, io
, (void *)iovp
, segs
,
841 ret
= generic_file_buffered_write(iocb
, iovp
, segs
,
842 pos
, offset
, count
, ret
);
845 current
->backing_dev_info
= NULL
;
847 if (ret
== -EIOCBQUEUED
&& !(ioflags
& IO_ISAIO
))
848 ret
= wait_on_sync_kiocb(iocb
);
850 if ((ret
== -ENOSPC
) &&
851 DM_EVENT_ENABLED(vp
->v_vfsp
, xip
, DM_EVENT_NOSPACE
) &&
852 !(ioflags
& IO_INVIS
)) {
854 xfs_rwunlock(bdp
, locktype
);
856 mutex_unlock(&inode
->i_mutex
);
857 error
= XFS_SEND_NAMESP(xip
->i_mount
, DM_EVENT_NOSPACE
, vp
,
858 DM_RIGHT_NULL
, vp
, DM_RIGHT_NULL
, NULL
, NULL
,
859 0, 0, 0); /* Delay flag intentionally unused */
863 mutex_lock(&inode
->i_mutex
);
864 xfs_rwlock(bdp
, locktype
);
865 pos
= xip
->i_d
.di_size
;
870 isize
= i_size_read(inode
);
871 if (unlikely(ret
< 0 && ret
!= -EFAULT
&& *offset
> isize
))
874 if (*offset
> xip
->i_d
.di_size
) {
875 xfs_ilock(xip
, XFS_ILOCK_EXCL
);
876 if (*offset
> xip
->i_d
.di_size
) {
877 xip
->i_d
.di_size
= *offset
;
878 i_size_write(inode
, *offset
);
879 xip
->i_update_core
= 1;
880 xip
->i_update_size
= 1;
882 xfs_iunlock(xip
, XFS_ILOCK_EXCL
);
887 goto out_unlock_internal
;
889 XFS_STATS_ADD(xs_write_bytes
, ret
);
891 /* Handle various SYNC-type writes */
892 if ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
)) {
893 error
= xfs_write_sync_logforce(mp
, xip
);
895 goto out_unlock_internal
;
897 xfs_rwunlock(bdp
, locktype
);
899 mutex_unlock(&inode
->i_mutex
);
901 error
= sync_page_range(inode
, mapping
, pos
, ret
);
908 xfs_rwunlock(bdp
, locktype
);
911 mutex_unlock(&inode
->i_mutex
);
917 * All xfs metadata buffers except log state machine buffers
918 * get this attached as their b_bdstrat callback function.
919 * This is so that we can catch a buffer
920 * after prematurely unpinning it to forcibly shutdown the filesystem.
923 xfs_bdstrat_cb(struct xfs_buf
*bp
)
927 mp
= XFS_BUF_FSPRIVATE3(bp
, xfs_mount_t
*);
928 if (!XFS_FORCED_SHUTDOWN(mp
)) {
929 xfs_buf_iorequest(bp
);
932 xfs_buftrace("XFS__BDSTRAT IOERROR", bp
);
934 * Metadata write that didn't get logged but
935 * written delayed anyway. These aren't associated
936 * with a transaction, and can be ignored.
938 if (XFS_BUF_IODONE_FUNC(bp
) == NULL
&&
939 (XFS_BUF_ISREAD(bp
)) == 0)
940 return (xfs_bioerror_relse(bp
));
942 return (xfs_bioerror(bp
));
948 xfs_bmap(bhv_desc_t
*bdp
,
955 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
956 xfs_iocore_t
*io
= &ip
->i_iocore
;
958 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
);
959 ASSERT(((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) != 0) ==
960 ((ip
->i_iocore
.io_flags
& XFS_IOCORE_RT
) != 0));
962 return xfs_iomap(io
, offset
, count
, flags
, iomapp
, niomaps
);
966 * Wrapper around bdstrat so that we can stop data
967 * from going to disk in case we are shutting down the filesystem.
968 * Typically user data goes thru this path; one of the exceptions
973 struct xfs_mount
*mp
,
977 if (!XFS_FORCED_SHUTDOWN(mp
)) {
978 /* Grio redirection would go here
979 * if (XFS_BUF_IS_GRIO(bp)) {
982 xfs_buf_iorequest(bp
);
986 xfs_buftrace("XFSBDSTRAT IOERROR", bp
);
987 return (xfs_bioerror_relse(bp
));
991 * If the underlying (data/log/rt) device is readonly, there are some
992 * operations that cannot proceed.
995 xfs_dev_is_read_only(
999 if (xfs_readonly_buftarg(mp
->m_ddev_targp
) ||
1000 xfs_readonly_buftarg(mp
->m_logdev_targp
) ||
1001 (mp
->m_rtdev_targp
&& xfs_readonly_buftarg(mp
->m_rtdev_targp
))) {
1003 "XFS: %s required on read-only device.", message
);
1005 "XFS: write access unavailable, cannot proceed.");