virtio-net have two usage of hashes: one is RSS and another is hash
reporting. Conventionally the hash calculation was done by the VMM.
However, computing the hash after the queue was chosen defeats the
purpose of RSS.

Another approach is to use eBPF steering program. This approach has
another downside: it cannot report the calculated hash due to the
restrictive nature of eBPF.

Introduce the code to compute hashes to the kernel in order to overcome
thse challenges.

An alternative solution is to extend the eBPF steering program so that it
will be able to report to the userspace, but it is based on context
rewrites, which is in feature freeze. We can adopt kfuncs, but they will
not be UAPIs. We opt to ioctl to align with other relevant UAPIs (KVM
and vhost_net).

QEMU patched to use this new feature is available at:
https://github.com/daynix/qemu/tree/akihikodaki/rss2

The QEMU patches will soon be submitted to the upstream as RFC too.

This work will be presented at LPC 2024:
https://lpc.events/event/18/contributions/1963/

V1 -> V2:
  Changed to introduce a new BPF program type.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
---
Changes in v3:
- Reverted back to add ioctl.
- Split patch "tun: Introduce virtio-net hashing feature" into
  "tun: Introduce virtio-net hash reporting feature" and
  "tun: Introduce virtio-net RSS".
- Changed to reuse hash values computed for automq instead of performing
  RSS hashing when hash reporting is requested but RSS is not.
- Extracted relevant data from struct tun_struct to keep it minimal.
- Added kernel-doc.
- Changed to allow calling TUNGETVNETHASHCAP before TUNSETIFF.
- Initialized num_buffers with 1.
- Added a test case for unclassified packets.
- Fixed error handling in tests.
- Changed tests to verify that the queue index will not overflow.
- Rebased.
- Link to v2: https://lore.kernel.org/r/20231015141644.260646-1-akihiko.odaki@daynix.com

---
Akihiko Odaki (9):
      skbuff: Introduce SKB_EXT_TUN_VNET_HASH
      virtio_net: Add functions for hashing
      net: flow_dissector: Export flow_keys_dissector_symmetric
      tap: Pad virtio header with zero
      tun: Pad virtio header with zero
      tun: Introduce virtio-net hash reporting feature
      tun: Introduce virtio-net RSS
      selftest: tun: Add tests for virtio-net hashing
      vhost/net: Support VIRTIO_NET_F_HASH_REPORT

 Documentation/networking/tuntap.rst  |   7 +
 drivers/net/Kconfig                  |   1 +
 drivers/net/tap.c                    |   2 +-
 drivers/net/tun.c                    | 255 ++++++++++++--
 drivers/vhost/net.c                  |  16 +-
 include/linux/skbuff.h               |  10 +
 include/linux/virtio_net.h           | 198 +++++++++++
 include/net/flow_dissector.h         |   1 +
 include/uapi/linux/if_tun.h          |  71 ++++
 net/core/flow_dissector.c            |   3 +-
 net/core/skbuff.c                    |   3 +
 tools/testing/selftests/net/Makefile |   2 +-
 tools/testing/selftests/net/tun.c    | 666 ++++++++++++++++++++++++++++++++++-
 13 files changed, 1195 insertions(+), 40 deletions(-)
---
base-commit: 46a0057a5853cbdb58211c19e89ba7777dc6fd50
change-id: 20240403-rss-e737d89efa77

Best regards,
-- 
Akihiko Odaki <akihiko.odaki@daynix.com>

On Sun, 15 Sep 2024 10:17:39 +0900
Akihiko Odaki <akihiko.odaki@daynix.com> wrote:

> virtio-net have two usage of hashes: one is RSS and another is hash
> reporting. Conventionally the hash calculation was done by the VMM.
> However, computing the hash after the queue was chosen defeats the
> purpose of RSS.
> 
> Another approach is to use eBPF steering program. This approach has
> another downside: it cannot report the calculated hash due to the
> restrictive nature of eBPF.
> 
> Introduce the code to compute hashes to the kernel in order to overcome
> thse challenges.
> 
> An alternative solution is to extend the eBPF steering program so that it
> will be able to report to the userspace, but it is based on context
> rewrites, which is in feature freeze. We can adopt kfuncs, but they will
> not be UAPIs. We opt to ioctl to align with other relevant UAPIs (KVM
> and vhost_net).

This will be useful for DPDK. But there still are cases where custom
flow rules are needed. I.e the RSS happens after other TC rules.
It would be a good if skbedit supported RSS as an option.

On 2024/09/15 21:48, Stephen Hemminger wrote:
> On Sun, 15 Sep 2024 10:17:39 +0900
> Akihiko Odaki <akihiko.odaki@daynix.com> wrote:
> 
>> virtio-net have two usage of hashes: one is RSS and another is hash
>> reporting. Conventionally the hash calculation was done by the VMM.
>> However, computing the hash after the queue was chosen defeats the
>> purpose of RSS.
>>
>> Another approach is to use eBPF steering program. This approach has
>> another downside: it cannot report the calculated hash due to the
>> restrictive nature of eBPF.
>>
>> Introduce the code to compute hashes to the kernel in order to overcome
>> thse challenges.
>>
>> An alternative solution is to extend the eBPF steering program so that it
>> will be able to report to the userspace, but it is based on context
>> rewrites, which is in feature freeze. We can adopt kfuncs, but they will
>> not be UAPIs. We opt to ioctl to align with other relevant UAPIs (KVM
>> and vhost_net).
> 
> This will be useful for DPDK. But there still are cases where custom
> flow rules are needed. I.e the RSS happens after other TC rules.
> It would be a good if skbedit supported RSS as an option.

Hi,

It is nice to hear about a use case other than QEMU or virtualization. I 
implemented RSS as tuntap ioctl because:
- It is easier to configure for the user of tuntap (e.g., QEMU)
- It implements hash reporting, which is specific to tuntap.

You can still add skbedit if you want to override RSS for some packets 
with filter. Please tell me if it is not sufficient for your use case.

Regards,
Akihiko Odaki