Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Sun, 1 Sep 2019 06:17:58 -0400
From:   "Michael S. Tsirkin" <mst@redhat.com>
To:     Stefano Garzarella <sgarzare@redhat.com>
Cc:     netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        Stefan Hajnoczi <stefanha@redhat.com>,
        "David S. Miller" <davem@davemloft.net>,
        virtualization@lists.linux-foundation.org,
        Jason Wang <jasowang@redhat.com>, kvm@vger.kernel.org
Subject: Re: [PATCH v4 1/5] vsock/virtio: limit the memory used per-socket
Message-ID: <20190901061707-mutt-send-email-mst@kernel.org>
References: <20190729114302-mutt-send-email-mst@kernel.org>
 <20190729161903.yhaj5rfcvleexkhc@steredhat>
 <20190729165056.r32uzj6om3o6vfvp@steredhat>
 <20190729143622-mutt-send-email-mst@kernel.org>
 <20190730093539.dcksure3vrykir3g@steredhat>
 <20190730163807-mutt-send-email-mst@kernel.org>
 <20190801104754.lb3ju5xjfmnxioii@steredhat>
 <20190801091106-mutt-send-email-mst@kernel.org>
 <20190801133616.sik5drn6ecesukbb@steredhat>
 <20190901025815-mutt-send-email-mst@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190901025815-mutt-send-email-mst@kernel.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, Sep 01, 2019 at 04:26:19AM -0400, Michael S. Tsirkin wrote:
> On Thu, Aug 01, 2019 at 03:36:16PM +0200, Stefano Garzarella wrote:
> > On Thu, Aug 01, 2019 at 09:21:15AM -0400, Michael S. Tsirkin wrote:
> > > On Thu, Aug 01, 2019 at 12:47:54PM +0200, Stefano Garzarella wrote:
> > > > On Tue, Jul 30, 2019 at 04:42:25PM -0400, Michael S. Tsirkin wrote:
> > > > > On Tue, Jul 30, 2019 at 11:35:39AM +0200, Stefano Garzarella wrote:
> > > > 
> > > > (...)
> > > > 
> > > > > > 
> > > > > > The problem here is the compatibility. Before this series virtio-vsock
> > > > > > and vhost-vsock modules had the RX buffer size hard-coded
> > > > > > (VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE = 4K). So, if we send a buffer smaller
> > > > > > of 4K, there might be issues.
> > > > > 
> > > > > Shouldn't be if they are following the spec. If not let's fix
> > > > > the broken parts.
> > > > > 
> > > > > > 
> > > > > > Maybe it is the time to add add 'features' to virtio-vsock device.
> > > > > > 
> > > > > > Thanks,
> > > > > > Stefano
> > > > > 
> > > > > Why would a remote care about buffer sizes?
> > > > > 
> > > > > Let's first see what the issues are. If they exist
> > > > > we can either fix the bugs, or code the bug as a feature in spec.
> > > > > 
> > > > 
> > > > The vhost_transport '.stream_enqueue' callback
> > > > [virtio_transport_stream_enqueue()] calls the virtio_transport_send_pkt_info(),
> > > > passing the user message. This function allocates a new packet, copying
> > > > the user message, but (before this series) it limits the packet size to
> > > > the VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE (4K):
> > > > 
> > > > static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
> > > > 					  struct virtio_vsock_pkt_info *info)
> > > > {
> > > >  ...
> > > > 	/* we can send less than pkt_len bytes */
> > > > 	if (pkt_len > VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE)
> > > > 		pkt_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
> > > > 
> > > > 	/* virtio_transport_get_credit might return less than pkt_len credit */
> > > > 	pkt_len = virtio_transport_get_credit(vvs, pkt_len);
> > > > 
> > > > 	/* Do not send zero length OP_RW pkt */
> > > > 	if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
> > > > 		return pkt_len;
> > > >  ...
> > > > }
> > > > 
> > > > then it queues the packet for the TX worker calling .send_pkt()
> > > > [vhost_transport_send_pkt() in the vhost_transport case]
> > > > 
> > > > The main function executed by the TX worker is
> > > > vhost_transport_do_send_pkt() that picks up a buffer from the virtqueue
> > > > and it tries to copy the packet (up to 4K) on it.  If the buffer
> > > > allocated from the guest will be smaller then 4K, I think here it will
> > > > be discarded with an error:
> > > > 
> > 
> > I'm adding more lines to explain better.
> > 
> > > > static void
> > > > vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
> > > > 				struct vhost_virtqueue *vq)
> > > > {
> > 		...
> > 
> > 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
> > 					 &out, &in, NULL, NULL);
> > 
> > 		...
> > 
> > 		len = iov_length(&vq->iov[out], in);
> > 		iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len);
> > 
> > 		nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
> > 		if (nbytes != sizeof(pkt->hdr)) {
> > 			virtio_transport_free_pkt(pkt);
> > 			vq_err(vq, "Faulted on copying pkt hdr\n");
> > 			break;
> > 		}
> > 
> > > >  ...
> > > > 		nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter);
> > > 
> > > isn't pck len the actual length though?
> > > 
> > 
> > It is the length of the packet that we are copying in the guest RX
> > buffers pointed by the iov_iter. The guest allocates an iovec with 2
> > buffers, one for the header and one for the payload (4KB).
> 
> BTW at the moment that forces another kmalloc within virtio core. Maybe
> vsock needs a flag to skip allocation in this case.  Worth benchmarking.
> See virtqueue_use_indirect which just does total_sg > 1.
> 
> > 
> > > > 		if (nbytes != pkt->len) {
> > > > 			virtio_transport_free_pkt(pkt);
> > > > 			vq_err(vq, "Faulted on copying pkt buf\n");
> > > > 			break;
> > > > 		}
> > > >  ...
> > > > }
> > > > 
> > > > 
> > > > This series changes this behavior since now we will split the packet in
> > > > vhost_transport_do_send_pkt() depending on the buffer found in the
> > > > virtqueue.
> > > > 
> > > > We didn't change the buffer size in this series, so we still backward
> > > > compatible, but if we will use buffers smaller than 4K, we should
> > > > encounter the error described above.
> 
> So that's an implementation bug then? It made an assumption
> of a 4K sized buffer? Or even PAGE_SIZE sized buffer?

Assuming we miss nothing and buffers < 4K are broken,
I think we need to add this to the spec, possibly with
a feature bit to relax the requirement that all buffers
are at least 4k in size.

> 
> > > > 
> > > > How do you suggest we proceed if we want to change the buffer size?
> > > > Maybe adding a feature to "support any buffer size"?
> > > > 
> > > > Thanks,
> > > > Stefano
> > > 
> > > 
> > 
> > --