Re: [PATCH v3 12/19] firmware_loader: Use security_post_load_data()

[Mimi Zohar]

On Tue, 2020-07-28 at 12:43 -0700, Kees Cook wrote:
> On Mon, Jul 27, 2020 at 06:57:45AM -0400, Mimi Zohar wrote:
> > On Fri, 2020-07-24 at 14:36 -0700, Kees Cook wrote:
> > > Now that security_post_load_data() is wired up, use it instead
> > > of the NULL file argument style of security_post_read_file(),
> > > and update the security_kernel_load_data() call to indicate that a
> > > security_kernel_post_load_data() call is expected.
> > > 
> > > Wire up the IMA check to match earlier logic. Perhaps a generalized
> > > change to ima_post_load_data() might look something like this:
> > > 
> > >     return process_buffer_measurement(buf, size,
> > >                                       kernel_load_data_id_str(load_id),
> > >                                       read_idmap[load_id] ?: FILE_CHECK,
> > >                                       0, NULL);
> > > 
> > > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > 
> > process_measurement() measures, verifies a file signature -  both
> > signatures stored as an xattr and as an appended buffer signature -
> > and augments audit records with the file hash. (Support for measuring,
> > augmenting audit records, and/or verifying fs-verity signatures has
> > yet to be added.)
> > 
> > As explained in my response to 11/19, the file descriptor provides the
> > file pathname associated with the buffer data.  In addition, IMA
> > policy rules may be defined in terms of other file descriptor info -
> > uid, euid, uuid, etc.
> > 
> > Recently support was added for measuring the kexec boot command line,
> > certificates being loaded onto a keyring, and blacklisted file hashes
> > (limited to appended signatures).  None of these buffers are signed.
> >  process_buffer_measurement() was added for this reason and as a
> > result is limited to just measuring the buffer data.
> > 
> > Whether process_measurement() or process_buffer_measurement() should
> > be modified, needs to be determined.  In either case to support the
> > init_module syscall, would at minimum require the associated file
> > pathname.
> 
> Right -- I don't intend to make changes to the init_module() syscall
> since it's deprecated, so this hook is more of a "fuller LSM coverage
> for old syscalls" addition.
> 
> IMA can happily continue to ignore it, which is what I have here, but I
> thought I'd at least show what it *might* look like. Perhaps BPF LSM is
> a better example.
> 
> Does anything need to change for this patch?

I wasn't aware that init_syscall was deprecated.  From your original comments,
it sounded like you wanted a new LSM for verifying kernel module signatures, as
they're currently supported via init_module().

I was mistaken.  Without a file descriptor, security_post_load_data() will
measure the firmware, as Scott confirmed, but won't be able to verify the
signature, whether he signed it using evmctl or not.

Mimi