Re: [PATCH 1/6] livepatch v5: avoid race when checking for state of the patch

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Tue, Dec 09, 2014 at 07:05:02PM +0100, Petr Mladek wrote:
> klp_patch_enable() and klp_patch_disable() should check the current state
> of the patch under the klp_lock. Otherwise, it might detect that the operation
> is valid but the situation might change before it takes the lock.

Hi Petr,

Thanks for the patches.

I don't think this patch is necessary.  klp_is_enabled() doesn't check
the state of the patch.  It checks the initialization state of the core
module (klp_root_kobj), which can only be set in klp_init().  It's not
protected by the lock, so I don't see the point of this patch.

> 
> Signed-off-by: Petr Mladek <pmladek@suse.cz>
> ---
>  kernel/livepatch/core.c | 16 ++++++++++------
>  1 file changed, 10 insertions(+), 6 deletions(-)
> 
> diff --git a/kernel/livepatch/core.c b/kernel/livepatch/core.c
> index d6d0f50e81f8..b848069e44cc 100644
> --- a/kernel/livepatch/core.c
> +++ b/kernel/livepatch/core.c
> @@ -425,11 +425,13 @@ int klp_disable_patch(struct klp_patch *patch)
>  {
>  	int ret;
>  
> -	if (!klp_is_enabled())
> -		return -ENODEV;
> -
>  	mutex_lock(&klp_mutex);
>  
> +	if (!klp_is_enabled()) {
> +		ret = -ENODEV;
> +		goto err;
> +	}
> +
>  	if (!klp_patch_is_registered(patch)) {
>  		ret = -EINVAL;
>  		goto err;
> @@ -489,11 +491,13 @@ int klp_enable_patch(struct klp_patch *patch)
>  {
>  	int ret;
>  
> -	if (!klp_is_enabled())
> -		return -ENODEV;
> -
>  	mutex_lock(&klp_mutex);
>  
> +	if (!klp_is_enabled()) {
> +		ret = -ENODEV;
> +		goto err;
> +	}
> +
>  	if (!klp_patch_is_registered(patch)) {
>  		ret = -EINVAL;
>  		goto err;
> -- 
> 1.8.5.2
> 

-- 
Josh