Re: [PATCH] don't do pointless NULL checks and casts before kfree() in security/

[Jesper Juhl <juhl-lkml@dif.dk>]

On Tue, 22 Mar 2005, David Howells wrote:

> 
> Jesper Juhl <juhl-lkml@dif.dk> wrote:
> 
> > --- linux-2.6.11-mm4-orig/security/keys/key.c	2005-03-16 15:45:42.000000000 +0100
> > +++ linux-2.6.11-mm4/security/keys/key.c	2005-03-20 12:40:19.000000000 +0100
> > ...
> > -	if (candidate)
> > -		kfree(candidate);
> > +	kfree(candidate);
> 
> Looks okay to me. It's probably less efficient though, but more space
> efficient.
> 
>From looking at the code gcc generates it looks to me like the bennefits
of the smaller code should outweigh the overhead of a few more function
calls - especially if the branch is usually taken (which I must admit I 
can't really tell if it will be). If the branch is rarely taken you are 
probably right.


> > --- linux-2.6.11-mm4-orig/security/keys/user_defined.c	2005-03-16 15:45:42.000000000 +0100
> > +++ linux-2.6.11-mm4/security/keys/user_defined.c	2005-03-20 12:41:54.000000000 +0100
> > @@ -182,9 +182,7 @@ static int user_match(const struct key *
> >   */
> >  static void user_destroy(struct key *key)
> >  {
> > -	struct user_key_payload *upayload = key->payload.data;
> > -
> > -	kfree(upayload);
> > +	kfree(key->payload.data);
> 
> There's a patch in Andrew Morton's tree that changes this to make use of RCU,
> so I'd prefer you didn't do this just yet.
> 
Huh? I just checked 2.6.12-rc1-mm1, and the user_destroy function still 
looks as above...  But no problem, I'll just send Andrew the bits in 
security/selinux/ for now and wait a bit with the rest. 

Thank you for your comments.


-- 
Jesper