Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Organization: Red Hat UK Ltd. Registered Address: Red Hat UK Ltd, Amberley
        Place, 107-111 Peascod Street, Windsor, Berkshire, SI4 1TE, United
        Kingdom.
        Registered in England and Wales under Company Registration No. 3798903
From:   David Howells <dhowells@redhat.com>
In-Reply-To: <CACT4Y+YajDb5QpSziTazoyn587JXwXet2w7Jkqkj9v31HZtJxw@mail.gmail.com>
References: <CACT4Y+YajDb5QpSziTazoyn587JXwXet2w7Jkqkj9v31HZtJxw@mail.gmail.com> <000000000000e7c6d205fa10a3cd@google.com>
To:     Dmitry Vyukov <dvyukov@google.com>
Cc:     dhowells@redhat.com,
        syzbot <syzbot+ebc945fdb4acd72cba78@syzkaller.appspotmail.com>,
        davem@davemloft.net, edumazet@google.com, kuba@kernel.org,
        linux-afs@lists.infradead.org, linux-fsdevel@vger.kernel.org,
        linux-kernel@vger.kernel.org, marc.dionne@auristor.com,
        netdev@vger.kernel.org, pabeni@redhat.com,
        syzkaller-bugs@googlegroups.com
Subject: Re: [syzbot] [afs?] [net?] KCSAN: data-race in rxrpc_send_data / rxrpc_set_call_completion
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <457893.1682352358.1@warthog.procyon.org.uk>
Content-Transfer-Encoding: quoted-printable
Date:   Mon, 24 Apr 2023 17:05:58 +0100
Message-ID: <457894.1682352358@warthog.procyon.org.uk>
Precedence: bulk

Dmitry Vyukov <dvyukov@google.com> wrote:

> If I am reading this correctly, rxrpc_send_data() can read wrong
> call->completion and state and incorrectly exit with an error if
> rxrpc_wait_to_be_connected() exists early right after observing error
> set here:
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/=
net/rxrpc/sendmsg.c?id=3D148341f0a2f53b5e8808d093333d85170586a15d#n58
> The code seems to assume that at that point all writes done by
> rxrpc_set_call_completion() are already finished, but it's not
> necessarily the case.

I'm not sure it matters.  call->error can only be set by the I/O thread an=
d
only if a call fails - in which case the call state will be set shortly
thereafter - plus a couple of places where we fail to set the call up, in
which case we're under the call's user_mutex or didn't even manage to full=
y
allocate it.

That said, I probably should check the call state first.  I might also wan=
t to
ignore any signal if the call did manage to get connected, lest I leave it
dangling - a problem might come if userspace issues a single sendmsg() to =
set
up the call and supply data to be transmitted.  I need to have a ponder on
that one.

David