netdev - Re: [PATCH v1 net 1/2] raw: Fix NULL deref in raw_get

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:   Tue, 4 Apr 2023 09:23:15 +0200
From:   Eric Dumazet <edumazet@...gle.com>
To:     Jason Xing <kerneljasonxing@...il.com>
Cc:     Kuniyuki Iwashima <kuniyu@...zon.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        Kuniyuki Iwashima <kuni1840@...il.com>, netdev@...r.kernel.org,
        syzbot <syzkaller@...glegroups.com>,
        "Dae R . Jeong" <threeearcat@...il.com>
Subject: Re: [PATCH v1 net 1/2] raw: Fix NULL deref in raw_get_next().

On Tue, Apr 4, 2023 at 8:56 AM Jason Xing <kerneljasonxing@...il.com> wrote:
>
> On Tue, Apr 4, 2023 at 12:07 PM Eric Dumazet <edumazet@...gle.com> wrote:
> >
> > On Tue, Apr 4, 2023 at 4:46 AM Jason Xing <kerneljasonxing@...il.com> wrote:
> > >
> > > I would like to ask two questions which make me confused:
> > > 1) Why would we use spin_lock to protect the socket in a raw hashtable
> > > for reader's safety under the rcu protection? Normally, if we use the
> > > RCU protection, we only make sure that we need to destroy the socket
> > > by calling call_rcu() which would prevent the READER of the socket
> > > from getting a NULL pointer.
> >
> > Yes, but then we can not sleep or yield the cpu.
>
> Indeed. We also cannot sleep/yield under the protection of the spin
> lock. And I checked the caller in fs/seq_file.c and noticed that we
> have no chance to sleep/yield between ->start and ->stop.
>

You missed my point.
The spinlock can trivially be replaced by a mutex, now the fast path
has been RCU converted.
This would allow raw_get_idx()/raw_get_first() to use cond_resched(),
if some hosts try to use 10,000 raw sockets :/
Is it a real problem to solve right now ?  I do not think so.

> So I wonder why we couldn't use RCU directly like the patch[1] you
> proposed before and choose deliberately to switch to spin lock? Spin
> lock for the whole hashinfo to protect the reader side is heavy, and
> RCU outperforms spin lock in this case, I think.

spinlock is just fine enough, most hosts have less than 10 raw sockets,
because raw sockets make things _much_ slower.

RCU 'just because' does not make sense, it would suggest that RAW sockets
scale, while they do not.