[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230727164820.48c9e685@kernel.org>
Date: Thu, 27 Jul 2023 16:48:20 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Sergey Shtylyov <s.shtylyov@....ru>
Cc: Paolo Abeni <pabeni@...hat.com>, Zheng Wang <zyytlz.wz@....com>,
<lee@...nel.org>, <linyunsheng@...wei.com>, <davem@...emloft.net>,
<edumazet@...gle.com>, <richardcochran@...il.com>,
<p.zabel@...gutronix.de>, <geert+renesas@...der.be>,
<magnus.damm@...il.com>, <yoshihiro.shimoda.uh@...esas.com>,
<biju.das.jz@...renesas.com>, <wsa+renesas@...g-engineering.com>,
<netdev@...r.kernel.org>, <linux-renesas-soc@...r.kernel.org>,
<linux-kernel@...r.kernel.org>, <hackerzheng666@...il.com>,
<1395428693sheep@...il.com>, <alex000young@...il.com>
Subject: Re: [PATCH v4] net: ravb: Fix possible UAF bug in ravb_remove
On Thu, 27 Jul 2023 21:48:41 +0300 Sergey Shtylyov wrote:
> >> Still racy, the carrier can come back up after canceling the work.
> >
> > I must admit I don't see how/when this driver sets the carrier on ?!?
>
> The phylib code does it for this MAC driver, see the call tree of
> phy_link_change(), on e.g. https://elixir.bootlin.com/linux/v6.5-rc3/source/...
>
> >> But whatever, this is a non-issue in the first place.
> >
> > Do you mean the UaF can't happen? I think that is real.
>
> Looks possible to me, at least now... and anyway, shouldn't we clean up
> after ourselves if we call schedule_work()?However my current impression is
> that cancel_work_sync() should be called from ravb_close(), after calling
> phy_{stop|disconnect}()...
>
> >> The fact that ravb_tx_timeout_work doesn't take any locks seems much
> >> more suspicious.
> >
> > Indeed! But that should be a different patch, right?
>
> Yes.
>
> > Waiting a little more for feedback from renesas.
>
> Renesas historically hasn't shown much interest to reviewing the sh_eth/ravb
> driver patches, so I took that task upon myself. I also happen to be a nominal
> author of this driver... :-)
Simplest fix I can think of is to take a reference on the netdev before
scheduling the work, and then check if it's still registered in the work
itself. Wrap the timeout work in rtnl_lock() to avoid any races there.
Powered by blists - more mailing lists