Re: [PATCH 4.19 011/120] tcp: annotate tp->copied_seq lockless reads

From: Eric Dumazet
Date: Tue Mar 16 2021 - 05:50:11 EST


On Tue, Mar 16, 2021 at 10:41 AM Pavel Machek <pavel@xxxxxxx> wrote:
>
> Hi!
>
> > From: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> >
> > From: Eric Dumazet <edumazet@xxxxxxxxxx>
>
> Two From: fields here.
>
> > [ Upstream commit 7db48e983930285b765743ebd665aecf9850582b ]
> >
> > There are few places where we fetch tp->copied_seq while
> > this field can change from IRQ or other cpu.
>
> And there are few such places even after the patch is applied; I
> quoted them below.
>
> Doing addition to variable without locking... is kind of
> interesting. Are you sure it is okay?

We are holding the socket lock here.

The WRITE_ONCE() here is paired with sides doing READ_ONCE() while
socket lock is _not_ held.

We want to make sure compiler won't write into this variable one byte at a time,
or using stupid things.




>
> > @@ -2112,7 +2112,7 @@ int tcp_recvmsg(struct sock *sk, struct
> > if (urg_offset < used) {
> > if (!urg_offset) {
> > if (!sock_flag(sk, SOCK_URGINLINE)) {
> > - ++*seq;
> > + WRITE_ONCE(*seq, *seq + 1);
> > urg_hole++;
> > offset++;
> > used--;
> > @@ -2134,7 +2134,7 @@ int tcp_recvmsg(struct sock *sk, struct
> > }
> > }
> >
> > - *seq += used;
> > + WRITE_ONCE(*seq, *seq + used);
> > copied += used;
> > len -= used;
> >
> > @@ -2163,7 +2163,7 @@ skip_copy:
> >
> > found_fin_ok:
> > /* Process the FIN. */
> > - ++*seq;
> > + WRITE_ONCE(*seq, *seq + 1);
> > if (!(flags & MSG_PEEK))
> > sk_eat_skb(sk, skb);
> > break;
>
> Best regards,
> Pavel
> --
> DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
> HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany