[PATCH RESEND] xfstests: _filter_mkfs should consume input from stdin first to avoid EPIPE

[PATCH RESEND] xfstests: _filter_mkfs should consume input from stdin first to avoid EPIPE

[Eryu Guan]

_filter_mkfs is a filter so that it should read from stdin first
before printing anything out. Otherwise the command prior to the
pipeline may get EPIPE.

I saw this when testing extN with generic/204, _scratch_mkfs_sized was
unable to create fs because of EPIPE, then _scratch_mount failed.

generic/204 12s ... [failed, exit status 1] - output mismatch (see /root/xfstests/results//generic/204.out.bad)
    --- tests/generic/204.out   2013-11-01 16:47:56.728591856 +0800
    +++ /root/xfstests/results//generic/204.out.bad     2013-11-01 22:52:53.207828779 +0800
    @@ -1,2 +1,7 @@
     QA output created by 204
    -*** done
    +mount: wrong fs type, bad option, bad superblock on /dev/sda6,
    +       missing codepage or helper program, or other error
    +       In some cases useful info is found in syslog - try
    +       dmesg | tail  or so
    +

Signed-off-by: Eryu Guan <eguan@redhat.com>
---
 common/filter | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/common/filter b/common/filter
index e37ce69..faf33e3 100644
--- a/common/filter
+++ b/common/filter
@@ -134,16 +134,19 @@ _filter_date()
 # Non XFS filesystems always return a 4k block size and a 256 byte inode.
 _filter_mkfs()
 {
+    local tmp=/tmp/mkfs_filter
+    cat - >$tmp
+
     case $FSTYP in
     xfs)
 	;;
     *)
 	perl -e 'print STDERR "dbsize=4096\nisize=256\n"'
+	rm -f $tmp
 	return ;;
     esac
 
-    set -
-    perl -ne '
+    cat $tmp | perl -ne '
     if (/^meta-data=([\w,|\/.-]+)\s+isize=(\d+)\s+agcount=(\d+), agsize=(\d+) blks/) {
 	print STDERR "ddev=$1\nisize=$2\nagcount=$3\nagsize=$4\n";
 	print STDOUT "meta-data=DDEV isize=XXX agcount=N, agsize=XXX blks\n";
@@ -175,6 +178,7 @@ _filter_mkfs()
 	print STDERR "rtdev=$1\nrtextsz=$2\nrtblocks=$3\nrtextents=$4\n";
 	print STDOUT "realtime =RDEV extsz=XXX blocks=XXX, rtextents=XXX\n";
     }'
+    rm -f $tmp
 }
 
 
-- 
1.8.3.1

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH RESEND] xfstests: _filter_mkfs should consume input from stdin first to avoid EPIPE

[Dave Chinner]

On Tue, Nov 12, 2013 at 12:17:54PM +0800, Eryu Guan wrote:
> _filter_mkfs is a filter so that it should read from stdin first
> before printing anything out. Otherwise the command prior to the
> pipeline may get EPIPE.
> 
> I saw this when testing extN with generic/204, _scratch_mkfs_sized was
> unable to create fs because of EPIPE, then _scratch_mount failed.

Ok, so that's a problem for filesystem types that don't actually
filter the output.

> generic/204 12s ... [failed, exit status 1] - output mismatch (see /root/xfstests/results//generic/204.out.bad)
>     --- tests/generic/204.out   2013-11-01 16:47:56.728591856 +0800
>     +++ /root/xfstests/results//generic/204.out.bad     2013-11-01 22:52:53.207828779 +0800
>     @@ -1,2 +1,7 @@
>      QA output created by 204
>     -*** done
>     +mount: wrong fs type, bad option, bad superblock on /dev/sda6,
>     +       missing codepage or helper program, or other error
>     +       In some cases useful info is found in syslog - try
>     +       dmesg | tail  or so
>     +
> 
> Signed-off-by: Eryu Guan <eguan@redhat.com>
> ---
>  common/filter | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/common/filter b/common/filter
> index e37ce69..faf33e3 100644
> --- a/common/filter
> +++ b/common/filter
> @@ -134,16 +134,19 @@ _filter_date()
>  # Non XFS filesystems always return a 4k block size and a 256 byte inode.
>  _filter_mkfs()
>  {
> +    local tmp=/tmp/mkfs_filter
> +    cat - >$tmp
> +

tmp is a global scope variable, so shouldn't be used as a local lik
ethis. Indeed, a file is not needed at all here, because

>      case $FSTYP in
>      xfs)
>  	;;
>      *)
>  	perl -e 'print STDERR "dbsize=4096\nisize=256\n"'
> +	rm -f $tmp
>  	return ;;

for filesystems that don't consume stdin, you could just pipe it to
/dev/null here.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs