Re: [PATCH v4] virtio/vsock: replace virtio_vsock_pkt with sk_buff

[Stefano Garzarella]

This is a net-next material, please remember to use net-next tag:
https://www.kernel.org/doc/html/v6.0/process/maintainer-netdev.html#netdev-faq

On Wed, Nov 23, 2022 at 10:07:49PM -0800, Bobby Eshleman wrote:
> This commit changes virtio/vsock to use sk_buff instead of
> virtio_vsock_pkt. Beyond better conforming to other net code, using
> sk_buff allows vsock to use sk_buff-dependent features in the future
> (such as sockmap) and improves throughput.
>
> This patch introduces the following performance changes:
>
> Tool/Config: uperf w/ 64 threads, SOCK_STREAM
> Test Runs: 5, mean of results
> Before: commit 95ec6bce2a0b ("Merge branch 'net-ipa-more-endpoints'")
>
> Test: 64KB, g2h
> Before: 21.63 Gb/s
> After: 25.59 Gb/s (+18%)
>
> Test: 16B, g2h
> Before: 11.86 Mb/s
> After: 17.41 Mb/s (+46%)
>
> Test: 64KB, h2g
> Before: 2.15 Gb/s
> After: 3.6 Gb/s (+67%)
>
> Test: 16B, h2g
> Before: 14.38 Mb/s
> After: 18.43 Mb/s (+28%)
>
> Signed-off-by: Bobby Eshleman <bobby.eshleman@xxxxxxxxxxxxx>
> ---

The patch LGTM. I run several tests (iperf3, vsock_test,
vsock_diag_test, vhost-user-vsock, tcpdump) and IMO we are okay.

I found the following problems that I would like to report:

- vhost-user-vsock [1] is failing, but it is not an issue of this patch,
  but a spec violation in the rust-vmm/vm-virtio/virtio-vsock crate as I
  reported here [2]. We will fix it there, this patch is fine, indeed
  trying a guest with the new layout (1 descriptor for both header and
  data) with vhost-vsock in Linux 6.0, everything works perfectly.

- the new "SOCK_SEQPACKET msg bounds" [3] reworked by Arseniy fails
  intermittently with this patch.

  Using the tests currently in the kernel tree everything is fine, so
  I don't understand if it's a problem in the new test or in this
  patch. I've looked at the code again and don't seem to see any
  criticisms.

  @Arseniy @Bobby can you take a look?

  I'll try to take a closer look too, and before I give my R-b I'd like
  to make sure it's a problem in the test and not in this patch.

  This is what I have (some times, not always) with both host and guest
  with this patch and the series of [3] applied:

  host$ ./vsock_test --control-host=192.168.133.3 --control-port=12345 \
                     --mode=client --peer-cid=4
  Control socket connected to 192.168.133.3:12345.
  0 - SOCK_STREAM connection reset...ok
  1 - SOCK_STREAM bind only...ok
  2 - SOCK_STREAM client close...ok
  3 - SOCK_STREAM server close...ok
  4 - SOCK_STREAM multiple connections...ok
  5 - SOCK_STREAM MSG_PEEK...ok
  6 - SOCK_SEQPACKET msg bounds...ok
  7 - SOCK_SEQPACKET MSG_TRUNC flag...recv: Connection reset by peer

  guest$ ./vsock_test --control-port=12345 --mode=server --peer-cid=2
  Control socket listening on 0.0.0.0:12345
  Control socket connection accepted...
  0 - SOCK_STREAM connection reset...ok
  1 - SOCK_STREAM bind only...ok
  2 - SOCK_STREAM client close...ok
  3 - SOCK_STREAM server close...ok
  4 - SOCK_STREAM multiple connections...ok
  5 - SOCK_STREAM MSG_PEEK...ok
  6 - SOCK_SEQPACKET msg bounds...Message bounds broken

Thanks,
Stefano

[1] https://github.com/rust-vmm/vhost-device/tree/main/crates/vsock
[2] https://github.com/rust-vmm/vm-virtio/issues/204
[3] https://lore.kernel.org/lkml/c991dffd-1dbc-e1d1-b682-a3c71f6ce51c@xxxxxxxxxxxxxx/