[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZUq-GrWMvbfhX74a@hog>
Date: Tue, 7 Nov 2023 23:45:46 +0100
From: Sabrina Dubroca <sd@...asysnail.net>
To: Jakub Kicinski <kuba@...nel.org>
Cc: "Dae R. Jeong" <threeearcat@...il.com>, borisp@...dia.com,
john.fastabend@...il.com, davem@...emloft.net, edumazet@...gle.com,
pabeni@...hat.com, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org, ywchoi@...ys.kaist.ac.kr
Subject: Re: Missing a write memory barrier in tls_init()
2023-11-06, 14:36:59 -0800, Jakub Kicinski wrote:
> On Thu, 2 Nov 2023 16:11:29 +0900 Dae R. Jeong wrote:
> > In addition, I believe the {tls_setsockopt, tls_getsockopt}
> > implementation is fine because of the address dependency. I think
> > load-load reordering is prohibited in this case so we don't need a
> > read barrier.
>
> Sounds plausible, could you send a patch?
>
> The smb_wmb() would be better placed in tls_init(), IMHO.
Wouldn't it be enough to just move the rcu_assign_pointer after ctx is
fully initialized, ie just before update_sk_prot? also clearer wrt
RCU.
(and maybe get rid of tls_ctx_create and move all that into tls_init,
it's not much and we don't even set ctx->{tx,rx}_conf in there)
--
Sabrina
Powered by blists - more mailing lists