Re: [patch 1/9] sys: Fix missing rcu protection for __task_cred() access

From: Thomas Gleixner <tglx@linutronix.de>
To: David Howells <dhowells@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Dipankar Sarma <dipankar@in.ibm.com>, Ingo Molnar <mingo@elte.hu>,
	Peter Zijlstra <peterz@infradead.org>,
	Al Viro <viro@zeniv.linux.org.uk>,
	James Morris <jmorris@namei.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	linux-security-module@vger.kernel.org
Subject: Re: [patch 1/9] sys: Fix missing rcu protection for __task_cred() access
Date: Fri, 11 Dec 2009 14:52:10 +0100 (CET)	[thread overview]
Message-ID: <alpine.LFD.2.00.0912111449030.3089@localhost.localdomain> (raw)
In-Reply-To: <13183.1260539120@redhat.com>

On Fri, 11 Dec 2009, David Howells wrote:

> Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> > > > Or are there updates that are carried out without write-holding
> > > > tasklist_lock that I am missing?
> > > 
> > > Yes, commit_creds() is called lockless.
> > 
> > Right, and that's what the problem is. commit_creds(), which rcu frees
> > the old creds, does not take tasklist lock write lock.
> 
> commit_creds() does not need to hold a write lock, because it is implicitly
> write-locked by only being permitted to run in the thread to which it is
> committing.
> 
> I don't think commit_creds() needs to take the RCU read lock as no-one else
> can alter/delete the creds it is dealing with.

commit_cred() is not required to take anything, but the reader side
needs to take rcu_read_lock() when accessing __task_cred() of another
task as that task could update its own creds right at that point.

The point is that read_lock(tasklist_lock) is not sufficient for the
reader side.

Thanks,

	tglx