Re: [PATCH] devtmpfs: move NULLing the thread pointer before unregistering fs

[Greg Kroah-Hartman]

On Fri, Dec 02, 2022 at 02:45:01PM +0200, Alexander Atanasov wrote:
> In commit
> 31c779f293b3 ("devtmpfs: fix the dangling pointer of global devtmpfsd thread")
> a dangling pointer on an error condition was fixed. But the fix
> left the dangling pointer during unregister_filesystem and printk calls.

And how could it be used there?

> Improve the fix to clear the pointer before unregistration to close
> the window  where the dangling pointer can be potentially used.

Again, how can that happen?  And you have an extra ' ' in that line :(

> Make it clear the pointer at only one place in the function.
> 
> Signed-off-by: Alexander Atanasov <alexander.atanasov@xxxxxxxxxxxxx>
> ---
>  drivers/base/devtmpfs.c | 8 +++-----
>  1 file changed, 3 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/base/devtmpfs.c b/drivers/base/devtmpfs.c
> index e4bffeabf344..773e66ef5642 100644
> --- a/drivers/base/devtmpfs.c
> +++ b/drivers/base/devtmpfs.c
> @@ -472,17 +472,15 @@ int __init devtmpfs_init(void)
>  	}
>  
>  	thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
> -	if (!IS_ERR(thread)) {
> +	if (!IS_ERR(thread))
>  		wait_for_completion(&setup_done);
> -	} else {
> +	else
>  		err = PTR_ERR(thread);
> -		thread = NULL;
> -	}
>  
>  	if (err) {
> +		thread = NULL;
>  		printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
>  		unregister_filesystem(&dev_fs_type);
> -		thread = NULL;
>  		return err;
>  	}

This all feels wrong and way too complex to have to clean up from a call
to kthread_run().  Are you sure this is the correct way to do this?

And how was this "issue" found?  How does the call to kthread_run() ever
fail for you?

thanks,

greg k-h