netdev - Re: [PATCH net v2] tcp: fix mishandling when the sack compression is deferred

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] [day] [month] [year] [list]

Message-ID: <CAL+tcoD0hXzymHyGJm2Rfk1hnVieFiAP5SY_WqdwE++APkskFA@mail.gmail.com>
Date: Wed, 31 May 2023 13:31:55 +0800
From: Jason Xing <kerneljasonxing@...il.com>
To: Eric Dumazet <edumazet@...gle.com>
Cc: fuyuanli <fuyuanli@...iglobal.com>, "David S. Miller" <davem@...emloft.net>, 
	Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>, 
	Neal Cardwell <ncardwell@...gle.com>, ycheng <ycheng@...gle.com>, toke <toke@...e.dk>, 
	netdev@...r.kernel.org, Weiping Zhang <zhangweiping@...iglobal.com>, 
	Tio Zhang <tiozhang@...iglobal.com>
Subject: Re: [PATCH net v2] tcp: fix mishandling when the sack compression is deferred

On Wed, May 31, 2023 at 12:13 PM Eric Dumazet <edumazet@...gle.com> wrote:
>
> On Wed, May 31, 2023 at 5:50 AM Jason Xing <kerneljasonxing@...il.com> wrote:
> >
> > On Tue, May 30, 2023 at 10:51 PM Eric Dumazet <edumazet@...gle.com> wrote:
> > >
> > > On Tue, May 30, 2023 at 4:32 PM Jason Xing <kerneljasonxing@...il.com> wrote:
> > > >
> > > > I'm confused. You said in the previous email:
> > > > "As a bonus, no need to send one patch for net, and another in net-next,
> > > > trying to 'fix' issues that should have been fixed cleanly in a single patch."
> > > >
> > > > So we added "introducing ICSK_ACK_TIMER flag for sack compression" to
> > > > fix them on top of the patch you suggested.
> > > >
> > > > I can remove the Suggested-by label. For now, I do care about your
> > > > opinion on the current patch.
> > > >
> > > > Well...should I give up introducing that flag and then leave that
> > > > 'issue' behind? :S
> > >
> > > Please let the fix go alone.
> > >
> > > Then I will look at your patch, but honestly I fail to see the _reason_ for it.
> > >
> > > In case you missed it, tcp_event_ack_sent() calls
> > > inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
> >
> > Hello Eric,
> >
> > Sorry, I didn't explain that 'issue' well last night. Let me try it once more:
> >
> > In the tcp_event_ack_sent(), since we're going to transmit data with
> > ack header, we should cancel those timers which could start before to
> > avoid sending useless/redundant acks. Right?
> >
> > But what if the timer, say, icsk_delack_timer, was triggered before
> > and had to postpone it in the release cb phrase because currently
> > socket (in the tcp sending process) has owned its @owned
> > field(sk->sk_lock.owned == 1).
> >
> > We could avoid sending extra useless ack by removing the
> > ICSK_ACK_TIMER flag to stop sending an ack in
> > tcp_delack_timer_handler().
> >
> > In the current logic, see in the tcp_event_ack_sent():
> > 1) hrtimer_try_to_cancel(&tp->compressed_ack_timer)
> > 2) sk_stop_timer(sk, &icsk->icsk_delack_timer)
> > Those two statements can prevent the timers from sending a useless ack
> > but cannot prevent sending a useless ack in the deferred process.
> >
> > Does it make any sense? Like I said, it's not a bug, but more like an
> > improvement.
>
> Your patch adds a bug. An skb allocation can fail, and ACK would not be sent.
>
> Timer handlers are not canceled in TCP stack.
> We do not call sk_stop_timer() because include/net/inet_connection_sock.h says
>
> /* Cancel timers, when they are not required. */
> #undef INET_CSK_CLEAR_TIMERS
>
> So claiming the following is nonsense:
>
> <quote>
>  2) sk_stop_timer(sk, &icsk->icsk_delack_timer)
> Those two statements can prevent ...
> </quote>
>

[...]
> We do not send extra ACK, because icsk->icsk_ack.pending (or icsk->icsk_pending)
> is cleared in inet_csk_clear_xmit_timer()

You're right about this. Thanks for your kind explanation. I missed
this key point again :(

It is a really happy ending cause I only need to focus on the v3
patch. Please help us review that patch if you're available (no rush)
:)

Thanks,
Jason

>
> This clearing is happening already at strategic places.
>
> When tcp_delack_timer_handler() is finally run (when owning socket lock),
> it will return early if icsk->icsk_ack.pending was already cleared.
>
> hrtimer_try_to_cancel(&tp->compressed_ack_timer) has to be called because
> we rely on the hrtimer status (hrtimer_is_queued()) in __tcp_ack_snd_check()