[PATCH linux-cr urgent] lsm: fix unlabeled restore case.

[Oren Laadan]

This is already in v20-rc1.

Oren

Serge E. Hallyn wrote:
> If an object is being restored which was checkpointed without an
> LSM label, then we should return 0 BEFORE we try to ckpt_obj_fetch
> a -1 objref (and obviously return error).
> 
> This makes the msg cr_tests ipc testcase fail if, for instance,
> CONFIG_SECURITY=y but no LSM is active.
> 
> IMO this definately should be in the tree that gets sent out...
> 
> Signed-off-by: Serge E. Hallyn <serue at us.ibm.com>
> ---
>  security/security.c |    8 ++++----
>  1 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/security/security.c b/security/security.c
> index 2b147cf..4b3f932 100644
> --- a/security/security.c
> +++ b/security/security.c
> @@ -1528,14 +1528,14 @@ int security_restore_obj(struct ckpt_ctx *ctx, void *v, int sectype,
>  		if (sectype != CKPT_SECURITY_MSG_MSG)
>  			return 0;
>  
> -	l = ckpt_obj_fetch(ctx, secref, CKPT_OBJ_SECURITY);
> -	if (IS_ERR(l))
> -		return PTR_ERR(l);
> -
>  	/* return if checkpointed label was "Not Applicable" */
>  	if (secref == SECURITY_CTX_NONE)
>  		return 0;
>  
> +	l = ckpt_obj_fetch(ctx, secref, CKPT_OBJ_SECURITY);
> +	if (IS_ERR(l))
> +		return PTR_ERR(l);
> +
>  	/* Ask the LSM to apply a void*security to the object
>  	 * based on the checkpointed context string */
>  	switch (sectype) {