[PATCH v2] security: Restore passing final prot to ima_file_mmap()

[Roberto Sassu]

On Fri, 2023-01-06 at 16:14 -0500, Paul Moore wrote:
> On Wed, Dec 21, 2022 at 9:10 AM Roberto Sassu
> <roberto.sassu at huaweicloud.com> wrote:
> > From: Roberto Sassu <roberto.sassu at huawei.com>
> > 
> > Commit 98de59bfe4b2f ("take calculation of final prot in
> > security_mmap_file() into a helper") moved the code to update prot with the
> > actual protection flags to be granted to the requestor by the kernel to a
> > helper called mmap_prot(). However, the patch didn't update the argument
> > passed to ima_file_mmap(), making it receive the requested prot instead of
> > the final computed prot.
> > 
> > A possible consequence is that files mmapped as executable might not be
> > measured/appraised if PROT_EXEC is not requested but subsequently added in
> > the final prot.
> > 
> > Replace prot with mmap_prot(file, prot) as the second argument of
> > ima_file_mmap() to restore the original behavior.
> > 
> > Cc: stable at vger.kernel.org
> > Fixes: 98de59bfe4b2 ("take calculation of final prot in security_mmap_file() into a helper")
> > Signed-off-by: Roberto Sassu <roberto.sassu at huawei.com>
> > ---
> >  security/security.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/security/security.c b/security/security.c
> > index d1571900a8c7..0d2359d588a1 100644
> > --- a/security/security.c
> > +++ b/security/security.c
> > @@ -1666,7 +1666,7 @@ int security_mmap_file(struct file *file, unsigned long prot,
> >                                         mmap_prot(file, prot), flags);
> >         if (ret)
> >                 return ret;
> > -       return ima_file_mmap(file, prot);
> > +       return ima_file_mmap(file, mmap_prot(file, prot));
> >  }
> 
> This seems like a reasonable fix, although as the original commit is
> ~10 years old at this point I am a little concerned about the impact
> this might have on IMA.  Mimi, what do you think?
> 
> Beyond that, my only other comment would be to only call mmap_prot()
> once and cache the results in a local variable.  You could also fix up
> some of the ugly indentation crimes in security_mmap_file() while you
> are at it, e.g. something like this:

Hi Paul

thanks for the comments. With the patch set to move IMA and EVM to the
LSM infrastructure we will be back to calling mmap_prot() only once,
but I guess we could do anyway, as the patch (if accepted) would be
likely backported to stable kernels.

Thanks

Roberto

> diff --git a/security/security.c b/security/security.c
> index d1571900a8c7..2f9cad9ecac8 100644
> --- a/security/security.c
> +++ b/security/security.c
> @@ -1662,11 +1662,12 @@ int security_mmap_file(struct file *file, unsigned long
> prot,
>                        unsigned long flags)
> {
>        int ret;
> -       ret = call_int_hook(mmap_file, 0, file, prot,
> -                                       mmap_prot(file, prot), flags);
> +       unsigned long prot_adj = mmap_prot(file, prot);
> +
> +       ret = call_int_hook(mmap_file, 0, file, prot, prot_adj, flags);
>        if (ret)
>                return ret;
> -       return ima_file_mmap(file, prot);
> +       return ima_file_mmap(file, prot_adj);
> }
> 
> --
> paul-moore.com