Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <bc4616002932b25973533c39c07f48ea57afa3dc.camel@redhat.com>
Subject: Re: [PATCH v2] net: sched: fix memory leak in tcindex_set_parms
From:   Paolo Abeni <pabeni@redhat.com>
To:     Jakub Kicinski <kuba@kernel.org>,
        Hawkins Jiawei <yin31149@gmail.com>
Cc:     Jamal Hadi Salim <jhs@mojatatu.com>,
        Cong Wang <xiyou.wangcong@gmail.com>,
        Jiri Pirko <jiri@resnulli.us>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>, 18801353760@163.com,
        syzbot+232ebdbd36706c965ebf@syzkaller.appspotmail.com,
        syzkaller-bugs@googlegroups.com,
        Cong Wang <cong.wang@bytedance.com>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Tue, 15 Nov 2022 19:57:10 +0100
In-Reply-To: <20221115090237.5d5988bb@kernel.org>
References: <20221113170507.8205-1-yin31149@gmail.com>
         <20221115090237.5d5988bb@kernel.org>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.42.4 (3.42.4-2.fc35) 
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Tue, 2022-11-15 at 09:02 -0800, Jakub Kicinski wrote:
> On Mon, 14 Nov 2022 01:05:08 +0800 Hawkins Jiawei wrote:
> 
> > @@ -479,6 +480,7 @@ tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
> >  	}
> >  
> >  	if (old_r && old_r != r) {
> > +		old_e = old_r->exts;
> >  		err = tcindex_filter_result_init(old_r, cp, net);
> >  		if (err < 0) {
> >  			kfree(f);
> > @@ -510,6 +512,12 @@ tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
> >  		tcf_exts_destroy(&new_filter_result.exts);
> >  	}
> >  
> > +	/* Note: old_e should be destroyed after the RCU grace period,
> > +	 * to avoid possible use-after-free by concurrent readers.
> > +	 */
> > +	synchronize_rcu();
> > +	tcf_exts_destroy(&old_e);
> 
> I don't think this dance is required, @cp is a copy of the original
> data, and the original (@p) is destroyed in a safe manner below.

This code confuses me more than a bit, and I don't follow ?!? it looks
like that at this point:

* the data path could access 'old_r->exts' contents via 'p' just before
the previous 'tcindex_filter_result_init(old_r, cp, net);' but still
potentially within the same RCU grace period

* 'tcindex_filter_result_init(old_r, cp, net);' has 'unlinked' the old
exts from 'p'  so that will not be freed by later
tcindex_partial_destroy_work() 

Overall it looks to me that we need some somewhat wait for the RCU
grace period, 

Somewhat side question: it looks like that the 'perfect hashing' usage
is the root cause of the issue addressed here, and very likely is
afflicted by other problems, e.g. the data curruption in 'err =
tcindex_filter_result_init(old_r, cp, net);'.

AFAICS 'perfect hashing' usage is a sort of optimization that the user-
space may trigger with some combination of the tcindex arguments. I'm
wondering if we could drop all perfect hashing related code?

Paolo