2016-05-24 13:24:36

by Eryu Guan

Subject: [PATCH] generic/250: ignore xfs_io IO errors when writing to dmerror device

When testing with data=journal ext4, direct write to dmerror device
doesn't return EIO, because ext4 turns direct write to buffered write in
data=journal mode and all data is written to journal buffer. The write
only fails later when commiting journal and error messages can be seen
in dmesg.

As the test is checking on the md5 checksum of the test file, it's ok to
ignore the IO error returned by xfs_io, as long as the checksums match
the golden image.

Signed-off-by: Eryu Guan <[email protected]>
---

Not returning errors to userspace properly is kind of an ext4 bug in journal
mode, but ext4 always behaves this way. So I think instead of fixing ext4, this
is the easiest way to make generic/250 work with journal mode ext4.

tests/generic/250 | 3 ++-
tests/generic/250.out | 1 -
2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/generic/250 b/tests/generic/250
index 6d15de2..891b8b0 100755
--- a/tests/generic/250
+++ b/tests/generic/250
@@ -83,7 +83,8 @@ echo "CoW and unmount"
$XFS_IO_PROG -f -c "pwrite -S 0x63 $bufsize 1" $testdir/file2 >> $seqres.full
sync
_dmerror_load_error_table
-$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 >> $seqres.full
+$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 \
+ >> $seqres.full 2>&1
_dmerror_load_working_table
_dmerror_unmount
_dmerror_mount
diff --git a/tests/generic/250.out b/tests/generic/250.out
index 710c80e..0ff87d0 100644
--- a/tests/generic/250.out
+++ b/tests/generic/250.out
@@ -4,7 +4,6 @@ Create the original files
Compare files
ec8bb3b24d5b0f1b5bdf8c8f0f541ee6 SCRATCH_MNT/test-250/file2
CoW and unmount
-pwrite64: Input/output error
Compare files
3ed86318f4ff8da26c1c2a6e3041f9be SCRATCH_MNT/test-250/file2
Check for damage
--
2.5.5

2016-05-24 18:21:36

by Darrick J. Wong

Subject: Re: [PATCH] generic/250: ignore xfs_io IO errors when writing to dmerror device

On Tue, May 24, 2016 at 09:24:36PM +0800, Eryu Guan wrote:
> When testing with data=journal ext4, direct write to dmerror device
> doesn't return EIO, because ext4 turns direct write to buffered write in
> data=journal mode and all data is written to journal buffer. The write
> only fails later when commiting journal and error messages can be seen
> in dmesg.
>
> As the test is checking on the md5 checksum of the test file, it's ok to
> ignore the IO error returned by xfs_io, as long as the checksums match
> the golden image.
>
> Signed-off-by: Eryu Guan <[email protected]>
> ---
>
> Not returning errors to userspace properly is kind of an ext4 bug in journal
> mode, but ext4 always behaves this way. So I think instead of fixing ext4, this
> is the easiest way to make generic/250 work with journal mode ext4.

/me wonders why _require_odirect doesn't _notrun in situations where we
*know* ahead of time that we're not going to get O_DIRECT semantics, but
whatever. It's no big deal if we shuffle the -EIO message to seqres.full.

> tests/generic/250 | 3 ++-
> tests/generic/250.out | 1 -
> 2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/tests/generic/250 b/tests/generic/250
> index 6d15de2..891b8b0 100755
> --- a/tests/generic/250
> +++ b/tests/generic/250
> @@ -83,7 +83,8 @@ echo "CoW and unmount"

Er... that should be 'echo "Write and unmount"', oops, sorry about that.
There's no COW going on here.

--D

> $XFS_IO_PROG -f -c "pwrite -S 0x63 $bufsize 1" $testdir/file2 >> $seqres.full
> sync
> _dmerror_load_error_table
> -$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 >> $seqres.full
> +$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 \
> + >> $seqres.full 2>&1
> _dmerror_load_working_table
> _dmerror_unmount
> _dmerror_mount
> diff --git a/tests/generic/250.out b/tests/generic/250.out
> index 710c80e..0ff87d0 100644
> --- a/tests/generic/250.out
> +++ b/tests/generic/250.out
> @@ -4,7 +4,6 @@ Create the original files
> Compare files
> ec8bb3b24d5b0f1b5bdf8c8f0f541ee6 SCRATCH_MNT/test-250/file2
> CoW and unmount
> -pwrite64: Input/output error
> Compare files
> 3ed86318f4ff8da26c1c2a6e3041f9be SCRATCH_MNT/test-250/file2
> Check for damage
> --
> 2.5.5
>
> --
> To unsubscribe from this list: send the line "unsubscribe fstests" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html

2016-05-25 04:43:07

by Eryu Guan

Subject: Re: [PATCH] generic/250: ignore xfs_io IO errors when writing to dmerror device

On Tue, May 24, 2016 at 11:21:36AM -0700, Darrick J. Wong wrote:
> On Tue, May 24, 2016 at 09:24:36PM +0800, Eryu Guan wrote:
> > When testing with data=journal ext4, direct write to dmerror device
> > doesn't return EIO, because ext4 turns direct write to buffered write in
> > data=journal mode and all data is written to journal buffer. The write
> > only fails later when commiting journal and error messages can be seen
> > in dmesg.
> >
> > As the test is checking on the md5 checksum of the test file, it's ok to
> > ignore the IO error returned by xfs_io, as long as the checksums match
> > the golden image.
> >
> > Signed-off-by: Eryu Guan <[email protected]>
> > ---
> >
> > Not returning errors to userspace properly is kind of an ext4 bug in journal
> > mode, but ext4 always behaves this way. So I think instead of fixing ext4, this
> > is the easiest way to make generic/250 work with journal mode ext4.
>
> /me wonders why _require_odirect doesn't _notrun in situations where we
> *know* ahead of time that we're not going to get O_DIRECT semantics, but
> whatever. It's no big deal if we shuffle the -EIO message to seqres.full.

I think that's because direct write falls back to buffered write
silently on journal mode ext4, so the _require_odirect check returns
pass.

>
> > tests/generic/250 | 3 ++-
> > tests/generic/250.out | 1 -
> > 2 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/tests/generic/250 b/tests/generic/250
> > index 6d15de2..891b8b0 100755
> > --- a/tests/generic/250
> > +++ b/tests/generic/250
> > @@ -83,7 +83,8 @@ echo "CoW and unmount"
>
> Er... that should be 'echo "Write and unmount"', oops, sorry about that.
> There's no COW going on here.

I can fix that in v2 :)

Thanks,
Eryu
>
> --D
>
> > $XFS_IO_PROG -f -c "pwrite -S 0x63 $bufsize 1" $testdir/file2 >> $seqres.full
> > sync
> > _dmerror_load_error_table
> > -$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 >> $seqres.full
> > +$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 \
> > + >> $seqres.full 2>&1
> > _dmerror_load_working_table
> > _dmerror_unmount
> > _dmerror_mount
> > diff --git a/tests/generic/250.out b/tests/generic/250.out
> > index 710c80e..0ff87d0 100644
> > --- a/tests/generic/250.out
> > +++ b/tests/generic/250.out
> > @@ -4,7 +4,6 @@ Create the original files
> > Compare files
> > ec8bb3b24d5b0f1b5bdf8c8f0f541ee6 SCRATCH_MNT/test-250/file2
> > CoW and unmount
> > -pwrite64: Input/output error
> > Compare files
> > 3ed86318f4ff8da26c1c2a6e3041f9be SCRATCH_MNT/test-250/file2
> > Check for damage
> > --
> > 2.5.5
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe fstests" in
> > the body of a message to [email protected]
> > More majordomo info at http://vger.kernel.org/majordomo-info.html

2016-05-26 03:26:24

by Eryu Guan

Subject: [PATCH v2] generic/250: ignore xfs_io IO errors when writing to dmerror device

When testing with data=journal ext4, direct write to dmerror device
doesn't return EIO, because ext4 turns direct write to buffered write in
data=journal mode and all data is written to journal buffer. The write
only fails later when commiting journal and error messages can be seen
in dmesg.

As the test is checking on the md5 checksum of the test file, it's ok to
ignore the IO error returned by xfs_io, as long as the checksums match
the golden image.

Signed-off-by: Eryu Guan <[email protected]>
---

v2:
- Replace "CoW" with "Write" and update .out accordingly

tests/generic/250 | 5 +++--
tests/generic/250.out | 3 +--
2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/tests/generic/250 b/tests/generic/250
index 6d15de2..3c4fe6d 100755
--- a/tests/generic/250
+++ b/tests/generic/250
@@ -79,11 +79,12 @@ _dmerror_mount
echo "Compare files"
md5sum $testdir/file2 | _filter_scratch

-echo "CoW and unmount"
+echo "Write and unmount"
$XFS_IO_PROG -f -c "pwrite -S 0x63 $bufsize 1" $testdir/file2 >> $seqres.full
sync
_dmerror_load_error_table
-$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 >> $seqres.full
+$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 \
+ >> $seqres.full 2>&1
_dmerror_load_working_table
_dmerror_unmount
_dmerror_mount
diff --git a/tests/generic/250.out b/tests/generic/250.out
index 710c80e..67b006b 100644
--- a/tests/generic/250.out
+++ b/tests/generic/250.out
@@ -3,8 +3,7 @@ Format and mount
Create the original files
Compare files
ec8bb3b24d5b0f1b5bdf8c8f0f541ee6 SCRATCH_MNT/test-250/file2
-CoW and unmount
-pwrite64: Input/output error
+Write and unmount
Compare files
3ed86318f4ff8da26c1c2a6e3041f9be SCRATCH_MNT/test-250/file2
Check for damage
--
2.5.5

2016-05-26 06:00:44

by Darrick J. Wong

Subject: Re: [PATCH v2] generic/250: ignore xfs_io IO errors when writing to dmerror device

On Thu, May 26, 2016 at 11:26:24AM +0800, Eryu Guan wrote:
> When testing with data=journal ext4, direct write to dmerror device
> doesn't return EIO, because ext4 turns direct write to buffered write in
> data=journal mode and all data is written to journal buffer. The write
> only fails later when commiting journal and error messages can be seen
> in dmesg.
>
> As the test is checking on the md5 checksum of the test file, it's ok to
> ignore the IO error returned by xfs_io, as long as the checksums match
> the golden image.

Looks fine to me,
Reviewed-by: Darrick J. Wong <[email protected]>

--D

>
> Signed-off-by: Eryu Guan <[email protected]>
> ---
>
> v2:
> - Replace "CoW" with "Write" and update .out accordingly
>
> tests/generic/250 | 5 +++--
> tests/generic/250.out | 3 +--
> 2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/tests/generic/250 b/tests/generic/250
> index 6d15de2..3c4fe6d 100755
> --- a/tests/generic/250
> +++ b/tests/generic/250
> @@ -79,11 +79,12 @@ _dmerror_mount
> echo "Compare files"
> md5sum $testdir/file2 | _filter_scratch
>
> -echo "CoW and unmount"
> +echo "Write and unmount"
> $XFS_IO_PROG -f -c "pwrite -S 0x63 $bufsize 1" $testdir/file2 >> $seqres.full
> sync
> _dmerror_load_error_table
> -$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 >> $seqres.full
> +$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $bufsize 0 $filesize" $testdir/file2 \
> + >> $seqres.full 2>&1
> _dmerror_load_working_table
> _dmerror_unmount
> _dmerror_mount
> diff --git a/tests/generic/250.out b/tests/generic/250.out
> index 710c80e..67b006b 100644
> --- a/tests/generic/250.out
> +++ b/tests/generic/250.out
> @@ -3,8 +3,7 @@ Format and mount
> Create the original files
> Compare files
> ec8bb3b24d5b0f1b5bdf8c8f0f541ee6 SCRATCH_MNT/test-250/file2
> -CoW and unmount
> -pwrite64: Input/output error
> +Write and unmount
> Compare files
> 3ed86318f4ff8da26c1c2a6e3041f9be SCRATCH_MNT/test-250/file2
> Check for damage
> --
> 2.5.5
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html