bug#50784: coreutils 9.0 sometimes gets spurious failures in chmod -R

[Pádraig Brady]

On 24/09/2021 20:29, Pádraig Brady wrote:
> On 24/09/2021 19:21, Daniel Schepler wrote:
> > After I upgraded my system (locally built) to coreutils 9.0, I noticed
> > that my lightly modified sbuild started getting build failures when it
> > tries to run "chmod -R g+w /<<BUILDDIR>>" (as root) in the chroot.
> > I'm not sure exactly what combination of files triggers the error, but
> > I'm able to reproduce it roughly as follows:
> >
> > cd /tmp
> > tar -x -f .../coreutils-9.0.tar.xz
> > cd coreutils-9.0
> > mkdir obj-x86_64-linux-gnu
> > cd obj-x86_64-linux-gnu
> > ../configure --prefix=/usr
> > make -j8
> > chmod -R u+w /tmp/coreutils-9.0/obj-x86_64-linux-gnu
> > echo $?
> >
> > In the last chmod command, there are no failures reported; and even if
> > I run it under strace, I see no hints of any failed system calls other
> > than one in access("/etc/ld.so.preload", R_OK) at startup -- and
> > certainly no failures in the fchmodat() system calls.  Yet the exit
> > code from the process is 1.
> >
> > I wasn't seeing any such errors under sbuild using coreutils 8.32.
>
> Drats. I think it's determining ignoring symlinks as an error.
> Could you try with the following untested diff:
>
> diff --git a/src/chmod.c b/src/chmod.c
> index 37b04f500..9ab60898d 100644
> --- a/src/chmod.c
> +++ b/src/chmod.c
> @@ -322,6 +322,9 @@ process_file (FTS *fts, FTSENT *ent)
>      if ( ! recurse)
>        fts_set (fts, ent, FTS_SKIP);
>
> +  if (ch.status == CH_NOT_APPLIED)  /* For symlinks.  */
> +    ch.status = CH_NO_CHANGE_REQUESTED;
> +
>      return CH_NO_CHANGE_REQUESTED <= ch.status;
>    }

Tested equivalent patch is attached.

chmod-fix-symlink-exit.patch
Description: Text Data