|
| From: | Antonios Motakis |
| Subject: | Re: [Qemu-devel] [PATCH v9 00/20] Vhost and vhost-net support for userspace based backends |
| Date: | Tue, 4 Mar 2014 19:33:06 +0100 |
Il 04/03/2014 19:22, Antonios Motakis ha scritto:Hi,
In this patch series we would like to introduce our approach for putting a
virtio-net backend in an external userspace process. Our eventual target is to
run the network backend in the Snabbswitch ethernet switch, while receiving
traffic from a guest inside QEMU/KVM which runs an unmodified virtio-net
implementation.
For this, we are working into extending vhost to allow equivalent functionality
for userspace. Vhost already passes control of the data plane of virtio-net to
the host kernel; we want to realize a similar model, but for userspace.
In this patch series the concept of a vhost-backend is introduced.
We define two vhost backend types - vhost-kernel and vhost-user. The former is
the interface to the current kernel module implementation. Its control plane is
ioctl based. The data plane is realized by the kernel directly accessing the
QEMU allocated, guest memory.
In the new vhost-user backend, the control plane is based on communication
between QEMU and another userspace process using a unix domain socket. This
allows to implement a virtio backend for a guest running in QEMU, inside the
other userspace process. For this communication we use a chardev with a Unix
domain socket backend. Vhost-user is client/server agnostic regarding the
chardev, however it does not support the 'nowait' and 'telnet' options.
A reconnection in 'server' mode is supported, but the backend's exposed virtio
features need to be compatible with the first connected slave.
We change -mem-path to QemuOpts and add prealloc and share as properties
to it. HugeTLBFS is required for this option to work.
The data path is realized by directly accessing the vrings and the buffer data
off the guest's memory.
The current user of vhost-user is only vhost-net. We add new netdev backend
that is intended to initialize vhost-net with vhost-user backend.
Example usage:
qemu -m 1024 -mem-path /hugetlbfs,share=on \
-chardev socket,id=chr0,path=/path/to/
socket \
-netdev type=vhost-user,id=net0,chardev=chr0 \
-device virtio-net-pci,netdev=net0
On non-MSIX guests the vhost feture can be forced using a special option:
...
-netdev type=vhost-user,id=net0,chardev=chr0,vhostforce
...
In order to use ioeventfds, kvm should be enabled.
This code can be pulled from address@hidden:virtualopensystems/qemu.git vhost-user-v9
A simple functional test is available in tests/vhost-user-test.c
A reference vhost-user slave for testing is also available from address@hidden:virtualopensystems/vapp.git
did you see the series I posted today? It would be great if you tested vhost-user on top of it. The replacement for the above -mem-path incantation should be the following:
-object memory-file,id=mem,path=/hugetlbfs,share=on \
-numa node,memdev=mem
Do the ideas behind vhost-user work if you have multiple hugetlbfs files, one per node?
Paolo
Changes from v8:
- Removed prealloc property from the -mem-path refactoring
- Added and use new function - kvm_eventfds_enabled
- Add virtio_queue_get_avail_idx used in vhost_virtqueue_stop to
get a sane value in case of VHOST_GET_VRING_BASE failure
- vhost user uses kvm_eventfds_enabled to check whether the ioeventfd
capability of KVM is available
- Added flag VHOST_USER_VRING_NOFD_MASK to be set when KICK, CALL or ERR file
descriptor is invalid or ioeventfd is not available
Changes from v7:
- Slave reconnection when using chardev in server mode
- qtest vhost-user-test added
- New qemu_chr_fe_get_msgfds for reading multiple fds from the chardev
- Mandatory features in vhost_dev, used on reconnect to verify for conflicts
- Add vhostforce parameter to -netdev vhost-user (for non-MSIX guests)
- Extend libqemustub.a to support qemu-char.c
Changes from v6:
- Remove the 'unlink' property of '-mem-path'
- Extend qemu-char: blocking read, send fds, monitor for connection close
- Vhost-user uses chardev as a backend
- Poll and reconnect removed (no VHOST_USER_ECHO).
- Disconnect is deteced by the chardev (G_IO_HUP event)
- vhost-backend.c split to vhost-user.c
Changes from v5:
- Split -mem-path unlink option to a separate patch
- Fds are passed only in the ancillary data
- Stricter message size checks on receive/send
- Netdev vhost-user now includes path and poll_time options
- The connection probing interval is configurable
Changes from v4:
- Use error_report for errors
- VhostUserMsg has new field `size` indicating the following payload length.
Field `flags` now has version and reply bits. The structure is packed.
- Send data is of variable length (`size` field in message)
- Receive in 2 steps, header and payload
- Add new message type VHOST_USER_ECHO, to check connection status
Changes from v3:
- Convert -mem-path to QemuOpts with prealloc, share and unlink properties
- Set 1 sec timeout when read/write to the unix domain socket
- Fix file descriptor leak
Changes from v2:
- Reconnect when the backend disappears
Changes from v1:
- Implementation of vhost-user netdev backend
- Code improvements
Antonios Motakis (20):
Convert -mem-path to QemuOpts and add share property
Add kvm_eventfds_enabled function
Add chardev API qemu_chr_fe_read_all
Add chardev API qemu_chr_fe_set_msgfds
Add chardev API qemu_chr_fe_get_msgfds
Add G_IO_HUP handler for socket chardev
vhost_net should call the poll callback only when it is set
Refactor virtio-net to use generic get_vhost_net
Add new virtio API virtio_queue_get_avail_idx
Gracefully handle ioctl failure in vhost_virtqueue_stop
vhost_net_init will use VhostNetOptions to get all its arguments
Add vhost_ops to vhost_dev struct and replace all relevant ioctls
Add mandatory_features to vhost_dev
Add vhost-backend and VhostBackendType
Add vhost-user as a vhost backend.
Add new vhost-user netdev backend
Add the vhost-user netdev backend to the command line
Add vhost-user protocol documentation
libqemustub: add stubs to be able to use qemu-char.c
Add qtest for vhost-user
docs/specs/vhost-user.txt | 261 ++++++++++++++++++++++++++++
exec.c | 21 ++-
hmp-commands.hx | 4 +-
hw/net/vhost_net.c | 141 ++++++++++-----
hw/net/virtio-net.c | 29 +---
hw/scsi/vhost-scsi.c | 20 ++-
hw/virtio/Makefile.objs | 2 +-
hw/virtio/vhost-backend.c | 71 ++++++++
hw/virtio/vhost-user.c | 356 ++++++++++++++++++++++++++++++++++++++
hw/virtio/vhost.c | 58 +++----
hw/virtio/virtio.c | 5 +
include/exec/cpu-all.h | 1 -
include/hw/virtio/vhost-backend.h | 38 ++++
include/hw/virtio/vhost.h | 9 +-
include/hw/virtio/virtio.h | 1 +
include/net/vhost-user.h | 17 ++
include/net/vhost_net.h | 13 +-
include/sysemu/char.h | 43 ++++-
include/sysemu/kvm.h | 11 ++
kvm-all.c | 4 +
kvm-stub.c | 1 +
net/Makefile.objs | 2 +-
net/clients.h | 3 +
net/hub.c | 1 +
net/net.c | 25 +--
net/tap.c | 18 +-
net/vhost-user.c | 273 +++++++++++++++++++++++++++++
qapi-schema.json | 19 +-
qemu-char.c | 272 +++++++++++++++++++++++++----
qemu-options.hx | 25 ++-
stubs/Makefile.objs | 8 +
stubs/bdrv-commit-all.c | 7 +
stubs/chr-msmouse.c | 7 +
stubs/get-next-serial.c | 3 +
stubs/is-daemonized.c | 7 +
stubs/machine-init-done.c | 6 +
stubs/monitor-init.c | 6 +
stubs/notify-event.c | 6 +
stubs/vc-init.c | 7 +
tests/Makefile | 4 +
tests/vhost-user-test.c | 309 +++++++++++++++++++++++++++++++++
vl.c | 23 ++-
42 files changed, 1979 insertions(+), 158 deletions(-)
create mode 100644 docs/specs/vhost-user.txt
create mode 100644 hw/virtio/vhost-backend.c
create mode 100644 hw/virtio/vhost-user.c
create mode 100644 include/hw/virtio/vhost-backend.h
create mode 100644 include/net/vhost-user.h
create mode 100644 net/vhost-user.c
create mode 100644 stubs/bdrv-commit-all.c
create mode 100644 stubs/chr-msmouse.c
create mode 100644 stubs/get-next-serial.c
create mode 100644 stubs/is-daemonized.c
create mode 100644 stubs/machine-init-done.c
create mode 100644 stubs/monitor-init.c
create mode 100644 stubs/notify-event.c
create mode 100644 stubs/vc-init.c
create mode 100644 tests/vhost-user-test.c
| [Prev in Thread] | Current Thread | [Next in Thread] |