netdev - Re: [PATCH net] net: avoid potential UAF in default

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]

Message-ID: <20241204143423.eewibpbgnuoqyyzn@skbuf>
Date: Wed, 4 Dec 2024 16:34:23 +0200
From: Vladimir Oltean <vladimir.oltean@....com>
To: Eric Dumazet <edumazet@...gle.com>
Cc: "David S . Miller" <davem@...emloft.net>,
	Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>,
	netdev@...r.kernel.org, eric.dumazet@...il.com,
	syzbot+1939f24bdb783e9e43d9@...kaller.appspotmail.com
Subject: Re: [PATCH net] net: avoid potential UAF in default_operstate()

On Wed, Dec 04, 2024 at 02:38:14PM +0100, Eric Dumazet wrote:
> On Wed, Dec 4, 2024 at 1:57 PM Vladimir Oltean <vladimir.oltean@....com> wrote:
> >
> > On Wed, Dec 04, 2024 at 12:46:11PM +0100, Eric Dumazet wrote:
> > > On Wed, Dec 4, 2024 at 12:41â€¯PM Vladimir Oltean <vladimir.oltean@....com> wrote:
> > > >
> > > > I meant: linkwatch runs periodically, via linkwatch_event(). Isn't there
> > > > a chance that linkwatch_event() can run once, immediately after
> > > > __rtnl_unlock() in netdev_run_todo(), while the netdev is in the
> > > > NETREG_UNREGISTERING state? Won't that create problems for __dev_get_by_index()
> > > > too? I guess it depends on when the netns is torn down, which I couldn't find.
> > >
> > > I think lweventlist_lock and dev->link_watch_list are supposed to
> > > synchronize things.
> > >
> > > linkwatch_sync_dev() only calls linkwatch_do_dev() if the device was
> > > atomically unlinked from lweventlist
> >
> > No, I don't mean calls from linkwatch_sync_dev(). I mean other call
> > paths towards linkwatch_do_dev(), like for example linkwatch_fire_event() -
> > carrier down, whatever. Can't these be pending on an unregistering
> > net_device at the time we run __rtnl_unlock() in netdev_run_todo?
> > Otherwise, why would netdev_wait_allrefs_any() have a linkwatch_run_queue()
> > call just later?
> 
> I do not know, this predates git history.
> 
> All these questions seem orthogonal.
> My patch fixes an issue added recently. not something added 10 years ago.
> I suggest we fix proven issues first, step by step.
> If you want to take over and send a series, just say so.
> 
> Thank you.

My understanding is certainly fuzzy, but I am not talking about some
behavior from 10 years ago. If I made default_operstate() require
rtnl_mutex last year, I did so for all call paths, not just for the
direct linkwatch_sync_dev() call that you point out. I agree we can take
them step by step if the UNREGISTERING state also proves problematic
(I don't have enough data now), but I disagree that the problem is
orthogonal.