linux-kernel - Re: [Linux-v4.6-rc1] ext4: WARNING: CPU: 2 PID: 2692 at kernel/locking/lockdep.c:2017 __lock

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 30 Mar 2016 11:50:06 +0200
From:	Peter Zijlstra <peterz@...radead.org>
To:	Ingo Molnar <mingo@...nel.org>
Cc:	Alfredo Alvarez Fernandez <alfredoalvarezfernandez@...il.com>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Sedat Dilek <sedat.dilek@...il.com>,
	Theodore Ts'o <tytso@....edu>,
	linux-fsdevel <linux-fsdevel@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>
Subject: Re: [Linux-v4.6-rc1] ext4: WARNING: CPU: 2 PID: 2692 at
 kernel/locking/lockdep.c:2017 __lock_acquire+0x180e/0x2260

On Wed, Mar 30, 2016 at 11:36:59AM +0200, Peter Zijlstra wrote:
> On Tue, Mar 29, 2016 at 10:47:02AM +0200, Ingo Molnar wrote:
> 
> > > You are right; this is lockdep running into a hash collision; which is a new 
> > > DEBUG_LOCKDEP test. See 9e4e7554e755 ("locking/lockdep: Detect chain_key 
> > > collisions").
> > 
> > I've Cc:-ed Alfredo Alvarez Fernandez who added that test.
> 
> OK, so while the code in check_no_collision() seems sensible, it relies
> on borken bits.
> 
> The whole chain_hlocks and /proc/lockdep_chains stuff appears to have
> been buggered from the start.
> 
> The below patch should fix this.

Note that unless we had more than 65536 chain_hlocks consumed the patch
would not make a difference.

> Furthermore, our hash function has definite room for improvement.

And no matter how good we make it, a u64 hash is bound to collide at
some point (or of any size really).

Also, we could make them non-fatal, returning true from
lookup_chain_cache() is always correct (_very_ expensive, but correct),
so in case of doubt we could just return true.