On Thu, Aug 03, 2017 at 06:26:47PM -0400, trondd wrote:
> Stuart Henderson <[email protected]> wrote:
>
> > Ah, perhaps the change to disk behaviour wasn't reflected in calculations
> > then..
>
> I got it figured out.
>
> In the checkfs function, the 'eval $(stat...)' command stores a list of disk
> devices and creates a variable named for each device to store the size of the
> files in the patch to be installed there. It uses :+ to accumulate the sizes
> of
> multiple files. But since ksh creates variales as global by default, these
> are
> not cleared between patches and :+ takes an existing value if the variable
> already
> exists. So the value stored in these variables will continue to accumulate.
>
> A simple solution seems to be to mark these variables local. Ksh will clean
> them
> up between calls to checkfs.
Very good find.
Just applied your patch on current, thanks.
> Tim.
>
>
> Index: syspatch.sh
> ===================================================================
> RCS file: /cvs/src/usr.sbin/syspatch/syspatch.sh,v
> retrieving revision 1.120
> diff -u -p -r1.120 syspatch.sh
> --- syspatch.sh 2 Aug 2017 05:58:29 -0000 1.120
> +++ syspatch.sh 3 Aug 2017 21:59:13 -0000
> @@ -91,7 +91,7 @@ checkfs()
> # - nonexistent files (i.e. syspatch is installing new files)
> # - broken interpolation due to bogus devices like remote filesystems
> eval $(cd / &&
> - stat -qf "_dev=\"\${_dev} %Sd\" %Sd=\"\${%Sd:+\${%Sd}\+}%Uz\"" \
> + stat -qf "_dev=\"\${_dev} %Sd\"; local
> %Sd=\"\${%Sd:+\${%Sd}\+}%Uz\"" \
> ${_files}) 2>/dev/null || _ret=$?
> set -e
> [[ ${_ret} == 127 ]] && sp_err "Remote filesystem, aborting"
>
--
Antoine
