Re: [PATCH 0/5] ebpf: Added ebpf helper for libvirtd.

[Jason Wang <jasowang@redhat.com>]

在 2021/7/13 下午11:37, Andrew Melnychenko 写道:
> Libvirt usually launches qemu with strict permissions.
> To enable eBPF RSS steering, qemu-ebpf-rss-helper was added.
>
> Added property "ebpf_rss_fds" for "virtio-net" that allows to
> initialize eBPF RSS context with passed program & maps fds.
>
> Added qemu-ebpf-rss-helper - simple helper that loads eBPF
> context and passes fds through unix socket.
> Libvirt should call the helper and pass fds to qemu through
> "ebpf_rss_fds" property.
>
> Added explicit target OS check for libbpf dependency in meson.
> eBPF RSS works only with Linux TAP, so there is no reason to
> build eBPF loader/helper for non-Linux.
>
> Changed Qemu updates eBPF maps to array mmaping. Mmaping allows
> bypassing unprivileged BPF map update. Also, instead of 3 maps
> (config, key and indirection table) there is one map that
> combines everything.
>
> Added helper stamp. To check that helper was build with qemu,
> qemu would check helper symbols that should contain the stamp.
> It was done similar to qemu modules, but checking was performed
> by the helper's ELF parsing.
>
> Overall, libvirt process should not be aware of the "interface"
> of eBPF RSS, it will not be aware of eBPF maps/program "type" and
> their quantity. That's why qemu and the helper should be from
> the same build and be "synchronized". Technically each qemu may
> have its own helper. That's why "query-helper-paths" qmp command
> was added. Qemu should return the path to the helper that suits
> and libvirt should use "that" helper for "that" emulator.
>
> qmp sample:
> C: { "execute": "query-helper-paths" }
> S: { "return": [
>       {
>         "name": "qemu-ebpf-rss-helper",
>         "path": "/usr/local/libexec/qemu-ebpf-rss-helper"
>       }
>      ]
>     }
>
> Changes since v1:
> * Mmap() used instead if bpf_map_update_elem().


Some questions:

1) Is the mmap() part a stable ABI?
2) Is there a change that we may use other kinds of bpf map that can be 
mmaped?
3) What's the advantages of using bpf mmap() over the rx-filter-event?

Thanks


> * Added helper stamp.
>
> Andrew Melnychenko (5):
>    ebpf: Added eBPF initialization by fds and map update.
>    virtio-net: Added property to load eBPF RSS with fds.
>    qmp: Added the helper stamp check.
>    ebpf_rss_helper: Added helper for eBPF RSS.
>    qmp: Added qemu-ebpf-rss-path command.
>
>   ebpf/ebpf_rss-stub.c              |   6 +
>   ebpf/ebpf_rss.c                   | 120 ++++---
>   ebpf/ebpf_rss.h                   |   8 +-
>   ebpf/qemu-ebpf-rss-helper.c       | 130 +++++++
>   ebpf/rss.bpf.skeleton.h           | 557 +++++++++++++++---------------
>   hw/net/virtio-net.c               |  77 ++++-
>   include/hw/virtio/virtio-net.h    |   1 +
>   meson.build                       |  47 ++-
>   monitor/meson.build               |   1 +
>   monitor/qemu-helper-stamp-utils.c | 297 ++++++++++++++++
>   monitor/qemu-helper-stamp-utils.h |  24 ++
>   monitor/qmp-cmds.c                |  32 ++
>   qapi/misc.json                    |  33 ++
>   tools/ebpf/rss.bpf.c              |  67 ++--
>   14 files changed, 990 insertions(+), 410 deletions(-)
>   create mode 100644 ebpf/qemu-ebpf-rss-helper.c
>   create mode 100644 monitor/qemu-helper-stamp-utils.c
>   create mode 100644 monitor/qemu-helper-stamp-utils.h
>