netdev - Re: [PATCH nf-next v4 1/5] net: sched: Micro-optimize egress handling

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:   Mon, 25 Jan 2021 11:39:08 -0800
From:   Jakub Kicinski <kuba@...nel.org>
To:     Lukas Wunner <lukas@...ner.de>
Cc:     Eric Dumazet <edumazet@...gle.com>,
        Pablo Neira Ayuso <pablo@...filter.org>,
        Jozsef Kadlecsik <kadlec@...filter.org>,
        Florian Westphal <fw@...len.de>,
        netfilter-devel@...r.kernel.org, coreteam@...filter.org,
        netdev <netdev@...r.kernel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Alexei Starovoitov <ast@...nel.org>,
        Thomas Graf <tgraf@...g.ch>,
        Laura Garcia Liebana <nevola@...il.com>,
        John Fastabend <john.fastabend@...il.com>,
        Dan Carpenter <dan.carpenter@...cle.com>
Subject: Re: [PATCH nf-next v4 1/5] net: sched: Micro-optimize egress
 handling

On Sun, 24 Jan 2021 11:33:01 +0100 Lukas Wunner wrote:
> On Fri, Jan 22, 2021 at 10:40:05AM +0100, Eric Dumazet wrote:
> > On Fri, Jan 22, 2021 at 9:55 AM Lukas Wunner <lukas@...ner.de> wrote:  
> > > sch_handle_egress() returns either the skb or NULL to signal to its
> > > caller __dev_queue_xmit() whether a packet should continue to be
> > > processed.
> > >
> > > The skb is always non-NULL, otherwise __dev_queue_xmit() would hit a
> > > NULL pointer deref right at its top.
> > >
> > > But the compiler doesn't know that.  So if sch_handle_egress() signals
> > > success by returning the skb, the "if (!skb) goto out;" statement
> > > results in a gratuitous NULL pointer check in the Assembler output.
> > >
> > > Avoid by telling the compiler that __dev_queue_xmit() is never passed a
> > > NULL skb.  
> [...]
> > > we're about to add a netfilter egress hook to __dev_queue_xmit()
> > > and without the micro-optimization, it will result in a performance
> > > degradation which is indeed measurable:  
> [...]
> > > --- a/net/core/dev.c
> > > +++ b/net/core/dev.c
> > > +__attribute__((nonnull(1)))
> > >  static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
> > >  {
> > >         struct net_device *dev = skb->dev;  
> > 
> > It is a bit sad the compilers do not automatically get this knowledge
> > from the very first instruction :
> > 
> >  struct net_device *dev = skb->dev;  
> 
> The compiler (gcc) is capable of doing that, but the feature was disabled by:
> 
>     commit a3ca86aea507904148870946d599e07a340b39bf
>     Author: Eugene Teo <eteo@...hat.com>
>     Date:   Wed Jul 15 14:59:10 2009 +0800
>     
>     Add '-fno-delete-null-pointer-checks' to gcc CFLAGS
> 
> If -fno-delete-null-pointer-checks is dropped from the top-level Makefile
> then the gratuitous NULL pointer checks disappear from the Assembler output,
> obviating the need to litter hot paths with __attribute__((nonnull(1)))
> annotations.
> 
> Taking a closer look at that commit, its rationale appears questionable:
> It says that broken code such as ...
> 
> 	struct agnx_priv *priv = dev->priv;
> 
> 	if (!dev)
> 		return;
> 
> ... would result in the NULL pointer check being optimized away.
> The commit message claims that keeping the NULL pointer check in
> "makes it harder to abuse" the broken code.
> 
> I don't see how that's the case:  If dev is NULL, the NULL pointer
> dereference at the function's top causes termination of the task
> in kernel/exit.c:do_exit().  So the NULL pointer check is never
> reached by the task.  If on the other hand dev is non-NULL,
> the task isn't terminated but then the NULL pointer check is
> unnecessary as well.
> 
> So the point of the commit remains elusive to me.  I could submit
> an RFC patch which drops -fno-delete-null-pointer-checks and see
> if any security folks cry foul.  Thoughts?

I wonder if modern compilers can't simply warn about this particular
case. Not to mention our static checkers..


Dan, do you think the concern from the above-quoted commit is still
valid? Is this something that smatch flags these days? We're apparently
paying a real performance price in networking for tying compiler's hands
with -fno-delete-null-pointer-checks