[PATCH] keys: Allow disabling read permissions for key possessor

[Eric Biggers]

On Mon, Mar 22, 2021 at 12:57:26PM +0300, Andrey Ryabinin wrote:
> keyctl_read_key() has a strange code which allows possessor to read
> key's payload regardless of READ permission status:
> 
> $ keyctl add user test test @u
> 196773443
> $ keyctl print 196773443
> test
> $ keyctl describe 196773443
> 196773443: alswrv-----v------------  1000  1000 user: test
> $ keyctl rdescribe 196773443
> user;1000;1000;3f010000;test
> $ keyctl setperm 196773443 0x3d010000
> $ keyctl describe 196773443
> 196773443: alsw-v-----v------------  1000  1000 user: test
> $ keyctl  print 196773443
> test
> 
> The last keyctl print should fail with -EACCESS instead of success.
> Fix this by removing weird possessor checks.
> 
> Signed-off-by: Andrey Ryabinin <arbn at yandex-team.ru>
> ---
> 
>  - This was noticed by code review. It seems like a bug to me,
>  but if I'm wrong and current behavior is correct, I think we need
>  at least better comment here.
>    
> 
>  security/keys/keyctl.c | 15 +--------------
>  1 file changed, 1 insertion(+), 14 deletions(-)
> 
> diff --git a/security/keys/keyctl.c b/security/keys/keyctl.c
> index 96a92a645216d..2ec021c7adc12 100644
> --- a/security/keys/keyctl.c
> +++ b/security/keys/keyctl.c
> @@ -845,22 +845,9 @@ long keyctl_read_key(key_serial_t keyid, char __user *buffer, size_t buflen)
>  
>  	/* see if we can read it directly */
>  	ret = key_permission(key_ref, KEY_NEED_READ);
> -	if (ret == 0)
> -		goto can_read_key;
> -	if (ret != -EACCES)
> +	if (ret != 0)
>  		goto key_put_out;
>  
> -	/* we can't; see if it's searchable from this process's keyrings
> -	 * - we automatically take account of the fact that it may be
> -	 *   dangling off an instantiation key
> -	 */
> -	if (!is_key_possessed(key_ref)) {
> -		ret = -EACCES;
> -		goto key_put_out;
> -	}
> -

This might be intentional, given the comment above the function:

 * The key must either grant the caller Read permission, or it must grant the
 * caller Search permission when searched for from the process keyrings.

The 'is_key_possessed()' check is implementing the second part, right?

Maybe check if this shows up in the documentation and tests too.

- Eric