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;
Message-ID: <7cad964394ce47cff28ec7c2f5f1559880e29ae2.camel@redhat.com>
Subject: Re: [PATCH v5] virtio/vsock: replace virtio_vsock_pkt with sk_buff
From:   Paolo Abeni <pabeni@redhat.com>
To:     Bobby Eshleman <bobbyeshleman@gmail.com>
Cc:     Bobby Eshleman <bobby.eshleman@bytedance.com>,
        Cong Wang <cong.wang@bytedance.com>,
        Jiang Wang <jiang.wang@bytedance.com>,
        Krasnov Arseniy <oxffffaa@gmail.com>,
        Stefan Hajnoczi <stefanha@redhat.com>,
        Stefano Garzarella <sgarzare@redhat.com>,
        "Michael S. Tsirkin" <mst@redhat.com>,
        Jason Wang <jasowang@redhat.com>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>, kvm@vger.kernel.org,
        virtualization@lists.linux-foundation.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Wed, 07 Dec 2022 10:33:22 +0100
In-Reply-To: <Y3toiPtBgOcrb8TL@bullseye>
References: <20221202173520.10428-1-bobby.eshleman@bytedance.com>
         <863a58452b4a4c0d63a41b0f78b59d32919067fa.camel@redhat.com>
         <Y3toiPtBgOcrb8TL@bullseye>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.42.4 (3.42.4-2.fc35) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Mon, 2022-11-21 at 12:01 +0000, Bobby Eshleman wrote:
> On Tue, Dec 06, 2022 at 11:20:21AM +0100, Paolo Abeni wrote:
> > Hello,
> > 
> > On Fri, 2022-12-02 at 09:35 -0800, Bobby Eshleman wrote:
> > [...]
> > > diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
> > > index 35d7eedb5e8e..6c0b2d4da3fe 100644
> > > --- a/include/linux/virtio_vsock.h
> > > +++ b/include/linux/virtio_vsock.h
> > > @@ -3,10 +3,129 @@
> > >  #define _LINUX_VIRTIO_VSOCK_H
> > >  
> > >  #include <uapi/linux/virtio_vsock.h>
> > > +#include <linux/bits.h>
> > >  #include <linux/socket.h>
> > >  #include <net/sock.h>
> > >  #include <net/af_vsock.h>
> > >  
> > > +#define VIRTIO_VSOCK_SKB_HEADROOM (sizeof(struct virtio_vsock_hdr))
> > > +
> > > +enum virtio_vsock_skb_flags {
> > > +	VIRTIO_VSOCK_SKB_FLAGS_REPLY		= BIT(0),
> > > +	VIRTIO_VSOCK_SKB_FLAGS_TAP_DELIVERED	= BIT(1),
> > > +};
> > > +
> > > +static inline struct virtio_vsock_hdr *virtio_vsock_hdr(struct sk_buff *skb)
> > > +{
> > > +	return (struct virtio_vsock_hdr *)skb->head;
> > > +}
> > > +
> > > +static inline bool virtio_vsock_skb_reply(struct sk_buff *skb)
> > > +{
> > > +	return skb->_skb_refdst & VIRTIO_VSOCK_SKB_FLAGS_REPLY;
> > > +}
> > 
> > I'm sorry for the late feedback. The above is extremelly risky: if the
> > skb will land later into the networking stack, we could experience the
> > most difficult to track bugs.
> > 
> > You should use the skb control buffer instead (skb->cb), with the
> > additional benefit you could use e.g. bool - the compiler could emit
> > better code to manipulate such fields - and you will not need to clear
> > the field before release nor enqueue.
> > 
> > [...]
> > 
> 
> Hey Paolo, thank you for the review. For my own learning, this would
> happen presumably when the skb is dropped? And I assume we don't see
> this in sockmap because it is always cleared before leaving sockmap's
> hands? I sanity checked this patch with an out-of-tree patch I have that
> uses the networking stack, but I suspect I didn't see issues because my
> test harness didn't induce dropping...

skb->_skb_refdst carries a dst and a flag in the less significative bit
specifying if the dst is refcounted. Passing to the network stack a skb
overloading such bit semanthic is quite alike intentionally corrupting
the kernel memory.

> I originally avoided skb->cb because the reply flag is set at allocation
> and would potentially be clobbered by a pass through the networking
> stack. The reply flag would be used after a pass through the networking
> stack (e.g., during transmission at the device level and when sockets
> close while skbs are still queued for xmit).

I assumed the 'tap_delivered' and 'reply' flag where relevant only
while the skb is owned by the virtio socket. If you need to preserve
such information _after_ delivering the skb to the network stack, that
is quite unfortunate - and skb->cb will not work.

The are a couple of options for adding new metadata inside the skb,
both of them are quite discouraged/need a strong use-case:

- adding new fields in some skb hole
- adding a new skb extension.

Could you please describe the 'reply' and 'tap_delivered' life-cycle
and their interaction with the network stack?


Cheers,

Paolo