[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20191206110836.GB31416@linux.home>
Date: Fri, 6 Dec 2019 12:08:36 +0100
From: Guillaume Nault <gnault@...hat.com>
To: Eric Dumazet <eric.dumazet@...il.com>
Cc: David Miller <davem@...emloft.net>,
Jakub Kicinski <jakub.kicinski@...ronome.com>,
netdev@...r.kernel.org, Eric Dumazet <edumazet@...gle.com>
Subject: Re: [PATCH net v3 3/3] tcp: Protect accesses to .ts_recent_stamp
with {READ,WRITE}_ONCE()
On Thu, Dec 05, 2019 at 07:50:39PM -0800, Eric Dumazet wrote:
>
>
> On 12/5/19 5:50 PM, Guillaume Nault wrote:
> > Syncookies borrow the ->rx_opt.ts_recent_stamp field to store the
> > timestamp of the last synflood. Protect them with READ_ONCE() and
> > WRITE_ONCE() since reads and writes aren't serialised.
> >
> > Fixes: 264ea103a747 ("tcp: syncookies: extend validity range")
> > Signed-off-by: Guillaume Nault <gnault@...hat.com>
>
> Signed-off-by: Eric Dumazet <edumazet@...gle.com>
>
> To be fair, bug was there before the patch mentioned in the Fixes: tag,
> but we probably do not care enough to backport this to very old kernels.
>
I used this commit because it introduced the conditional in
tcp_synq_overflow(), which I believe made the lockless accesses more
dangerous than they were. But yes, the problem has been there forever.
I have to post a v4 anyway, so I'll change this tag to reference the
first commit in the tree.
Powered by blists - more mailing lists