[PATCH] xfs_admin: pick up log arguments correctly

[PATCH] xfs_admin: pick up log arguments correctly

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

In commit ab9d8d69, we added support to xfs_admin to pass an external
log to xfs_db and xfs_repair.  Unfortunately, we didn't do this
correctly -- by appending the log arguments to DB_OPTS, we now guarantee
an invocation of xfs_db when we don't have any work for it to do.

Brian Foster noticed that this results in xfs/764 hanging fstests
because xfs_db (when not compiled with libeditline) will wait for input
on stdin.  I didn't notice because my build includes libeditline and my
test runner script does silly things with pipes such that xfs_db would
exit immediately.

Reported-by: Brian Foster <bfoster@redhat.com>
Fixes: ab9d8d69 ("xfs_admin: support adding features to V5 filesystems")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
---
 db/xfs_admin.sh |    9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/db/xfs_admin.sh b/db/xfs_admin.sh
index 916050cb..409975b2 100755
--- a/db/xfs_admin.sh
+++ b/db/xfs_admin.sh
@@ -8,7 +8,7 @@ status=0
 DB_OPTS=""
 REPAIR_OPTS=""
 REPAIR_DEV_OPTS=""
-DB_LOG_OPTS=""
+LOG_OPTS=""
 USAGE="Usage: xfs_admin [-efjlpuV] [-c 0|1] [-L label] [-O v5_feature] [-r rtdev] [-U uuid] device [logdev]"
 
 while getopts "c:efjlL:O:pr:uU:V" c
@@ -40,19 +40,18 @@ case $# in
 	1|2)
 		# Pick up the log device, if present
 		if [ -n "$2" ]; then
-			DB_OPTS=$DB_OPTS" -l '$2'"
-			REPAIR_DEV_OPTS=$REPAIR_DEV_OPTS" -l '$2'"
+			LOG_OPTS=" -l '$2'"
 		fi
 
 		if [ -n "$DB_OPTS" ]
 		then
-			eval xfs_db -x -p xfs_admin $DB_OPTS "$1"
+			eval xfs_db -x -p xfs_admin $LOG_OPTS $DB_OPTS "$1"
 			status=$?
 		fi
 		if [ -n "$REPAIR_OPTS" ]
 		then
 			echo "Running xfs_repair to upgrade filesystem."
-			eval xfs_repair $REPAIR_DEV_OPTS $REPAIR_OPTS "$1"
+			eval xfs_repair $LOG_OPTS $REPAIR_DEV_OPTS $REPAIR_OPTS "$1"
 			status=`expr $? + $status`
 		fi
 		;;

Re: [PATCH] xfs_admin: pick up log arguments correctly

[Eric Sandeen]

On 3/23/21 9:10 PM, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@kernel.org>
> 
> In commit ab9d8d69, we added support to xfs_admin to pass an external
> log to xfs_db and xfs_repair.  Unfortunately, we didn't do this
> correctly -- by appending the log arguments to DB_OPTS, we now guarantee
> an invocation of xfs_db when we don't have any work for it to do.
> 
> Brian Foster noticed that this results in xfs/764 hanging fstests
> because xfs_db (when not compiled with libeditline) will wait for input
> on stdin.  I didn't notice because my build includes libeditline and my
> test runner script does silly things with pipes such that xfs_db would
> exit immediately.
> 
> Reported-by: Brian Foster <bfoster@redhat.com>
> Fixes: ab9d8d69 ("xfs_admin: support adding features to V5 filesystems")
> Signed-off-by: Darrick J. Wong <djwong@kernel.org>

This seems fine.  While astute bashophiles will have no problem with this,
at some point it might be nice to add some comments above DB_OPTS and
REPAIR_OPTS that point out hey, if you set these, you WILL be invoking the
tool.

I also chafe a little at accumulating some device options in REPAIR_DEV_OPTS
and others in LOG_OPTS; why not REPAIR_DEV_OPTS and DB_DEV_OPTS for some
consistency?

But, this does seem to solve the problem, and in the
Spirit of Lets Not Navel-Gaze And Just Keep Fixing Things(tm),

Reviewed-by: Eric Sandeen <sandeen@redhat.com>

> ---
>  db/xfs_admin.sh |    9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/db/xfs_admin.sh b/db/xfs_admin.sh
> index 916050cb..409975b2 100755
> --- a/db/xfs_admin.sh
> +++ b/db/xfs_admin.sh
> @@ -8,7 +8,7 @@ status=0
>  DB_OPTS=""
>  REPAIR_OPTS=""
>  REPAIR_DEV_OPTS=""
> -DB_LOG_OPTS=""
> +LOG_OPTS=""
>  USAGE="Usage: xfs_admin [-efjlpuV] [-c 0|1] [-L label] [-O v5_feature] [-r rtdev] [-U uuid] device [logdev]"
>  
>  while getopts "c:efjlL:O:pr:uU:V" c
> @@ -40,19 +40,18 @@ case $# in
>  	1|2)
>  		# Pick up the log device, if present
>  		if [ -n "$2" ]; then
> -			DB_OPTS=$DB_OPTS" -l '$2'"
> -			REPAIR_DEV_OPTS=$REPAIR_DEV_OPTS" -l '$2'"
> +			LOG_OPTS=" -l '$2'"
>  		fi
>  
>  		if [ -n "$DB_OPTS" ]
>  		then
> -			eval xfs_db -x -p xfs_admin $DB_OPTS "$1"
> +			eval xfs_db -x -p xfs_admin $LOG_OPTS $DB_OPTS "$1"
>  			status=$?
>  		fi
>  		if [ -n "$REPAIR_OPTS" ]
>  		then
>  			echo "Running xfs_repair to upgrade filesystem."
> -			eval xfs_repair $REPAIR_DEV_OPTS $REPAIR_OPTS "$1"
> +			eval xfs_repair $LOG_OPTS $REPAIR_DEV_OPTS $REPAIR_OPTS "$1"
>  			status=`expr $? + $status`
>  		fi
>  		;;
> 

Re: [PATCH] xfs_admin: pick up log arguments correctly

[Darrick J. Wong]

On Wed, Mar 24, 2021 at 03:47:54PM -0500, Eric Sandeen wrote:
> On 3/23/21 9:10 PM, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > In commit ab9d8d69, we added support to xfs_admin to pass an external
> > log to xfs_db and xfs_repair.  Unfortunately, we didn't do this
> > correctly -- by appending the log arguments to DB_OPTS, we now guarantee
> > an invocation of xfs_db when we don't have any work for it to do.
> > 
> > Brian Foster noticed that this results in xfs/764 hanging fstests
> > because xfs_db (when not compiled with libeditline) will wait for input
> > on stdin.  I didn't notice because my build includes libeditline and my
> > test runner script does silly things with pipes such that xfs_db would
> > exit immediately.
> > 
> > Reported-by: Brian Foster <bfoster@redhat.com>
> > Fixes: ab9d8d69 ("xfs_admin: support adding features to V5 filesystems")
> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> 
> This seems fine.  While astute bashophiles will have no problem with this,
> at some point it might be nice to add some comments above DB_OPTS and
> REPAIR_OPTS that point out hey, if you set these, you WILL be invoking the
> tool.

That "should" be obvious from reading the source code, but this
maintainer obviously didn't notice that, and bash is horrible so I don't
blame anyone else for missing things.

> I also chafe a little at accumulating some device options in REPAIR_DEV_OPTS
> and others in LOG_OPTS; why not REPAIR_DEV_OPTS and DB_DEV_OPTS for some
> consistency?

Why not dump each arg into a bash array so that we don't have to deal
with all this shell space-escaping crap? :P

> But, this does seem to solve the problem, and in the
> Spirit of Lets Not Navel-Gaze And Just Keep Fixing Things(tm),

Ok, thanks.  Enjoy your vacation! :)

--D

> Reviewed-by: Eric Sandeen <sandeen@redhat.com>
> 
> > ---
> >  db/xfs_admin.sh |    9 ++++-----
> >  1 file changed, 4 insertions(+), 5 deletions(-)
> > 
> > diff --git a/db/xfs_admin.sh b/db/xfs_admin.sh
> > index 916050cb..409975b2 100755
> > --- a/db/xfs_admin.sh
> > +++ b/db/xfs_admin.sh
> > @@ -8,7 +8,7 @@ status=0
> >  DB_OPTS=""
> >  REPAIR_OPTS=""
> >  REPAIR_DEV_OPTS=""
> > -DB_LOG_OPTS=""
> > +LOG_OPTS=""
> >  USAGE="Usage: xfs_admin [-efjlpuV] [-c 0|1] [-L label] [-O v5_feature] [-r rtdev] [-U uuid] device [logdev]"
> >  
> >  while getopts "c:efjlL:O:pr:uU:V" c
> > @@ -40,19 +40,18 @@ case $# in
> >  	1|2)
> >  		# Pick up the log device, if present
> >  		if [ -n "$2" ]; then
> > -			DB_OPTS=$DB_OPTS" -l '$2'"
> > -			REPAIR_DEV_OPTS=$REPAIR_DEV_OPTS" -l '$2'"
> > +			LOG_OPTS=" -l '$2'"
> >  		fi
> >  
> >  		if [ -n "$DB_OPTS" ]
> >  		then
> > -			eval xfs_db -x -p xfs_admin $DB_OPTS "$1"
> > +			eval xfs_db -x -p xfs_admin $LOG_OPTS $DB_OPTS "$1"
> >  			status=$?
> >  		fi
> >  		if [ -n "$REPAIR_OPTS" ]
> >  		then
> >  			echo "Running xfs_repair to upgrade filesystem."
> > -			eval xfs_repair $REPAIR_DEV_OPTS $REPAIR_OPTS "$1"
> > +			eval xfs_repair $LOG_OPTS $REPAIR_DEV_OPTS $REPAIR_OPTS "$1"
> >  			status=`expr $? + $status`
> >  		fi
> >  		;;
> >