linux-kernel - Re: [PATCH v2 2/2] slab: __GFP_ZERO is incompatible with a constructor

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:   Wed, 11 Apr 2018 16:56:52 -0700
From:   Matthew Wilcox <willy@...radead.org>
To:     Christopher Lameter <cl@...ux.com>
Cc:     linux-mm@...ck.org, Matthew Wilcox <mawilcox@...rosoft.com>,
        Pekka Enberg <penberg@...nel.org>,
        David Rientjes <rientjes@...gle.com>,
        Joonsoo Kim <iamjoonsoo.kim@....com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        linux-kernel@...r.kernel.org, Jan Kara <jack@...e.cz>,
        Jeff Layton <jlayton@...hat.com>,
        Mel Gorman <mgorman@...hsingularity.net>
Subject: Re: [PATCH v2 2/2] slab: __GFP_ZERO is incompatible with a
 constructor

On Wed, Apr 11, 2018 at 04:11:17PM -0500, Christopher Lameter wrote:
> On Wed, 11 Apr 2018, Matthew Wilcox wrote:
> > > Please put this in a code path that is enabled by specifying
> > >
> > > slub_debug
> > >
> > > on the kernel command line.
> >
> > I don't understand.  First, I had:
> >
> > 	if (unlikely(gfpflags & __GFP_ZERO) && object && !WARN_ON_ONCE(s->ctor))
> >
> > and you didn't like that because it was putting checking into a (semi)fast
> > path.  Now you want me to add a check for slub_debug somewhere?  I dont
> > see an existing one I can leverage that will hit on every allocation.
> > Perhaps I'm missing something.
> 
> The WARN_ON is only enabled when you configure and build the kernel with
> debugging enabled (CONFIG_VM_DEBUG). That is a compile time debugging
> feature like supported by SLAB.

Yes.  I want to have an option to check *every single* allocation.

> "slub_debug" enables kmem_cache->flags & SLAB_DEBUG and that forces all
> fastpath processing to be disabled. Thus you can check reliably in the
> slow path only for the GFP_ZERO problem.
> 
> Add the check to the other debug stuff already there. F.e. in
> alloc_debug_processing() or after
> 
> if (kmem_cache_debug(s) ...
> 
> in ____slab_alloc()

I don't see how that works ... can you explain a little more?

I see ___slab_alloc() is called from __slab_alloc().  And I see
slab_alloc_node does this:

        object = c->freelist;
        page = c->page;
        if (unlikely(!object || !node_match(page, node))) {
                object = __slab_alloc(s, gfpflags, node, addr, c);
                stat(s, ALLOC_SLOWPATH);

But I don't see how slub_debug leads to c->freelist always being NULL.
It looks like it gets repopulated from page->freelist in ___slab_alloc()
at the load_freelist label.