VirtIO-FS: A Proposed Better Approach For Sharing Folders/Files With Guest VMs

[phoronix]

Phoronix: VirtIO-FS: A Proposed Better Approach For Sharing Folders/Files With Guest VMs

Red Hat developers have proposed a new VirtIO-FS component to provide better support for shared folders/files between the host and guest virtual machines...

http://www.phoronix.com/scan.php?page=news_item&px=VirtIO-FS-Folder-Sharing-VM

[tildearrow]

Typo:

Kata containers but coud be used

[varikonniemi]

Very nice! Shared folder support is the one thing i miss from VB under kvm.

[pkunk]

Very nice! Shared folder support is the one thing i miss from VB under kvm.

Why?

https://www.linux-kvm.org/page/9p_virtio

There is already existing solution, just new one should be faster.

[polarathene]

Why?

https://www.linux-kvm.org/page/9p_virtio

There is already existing solution, just new one should be faster.

9p does not work on Windows or macOS guests.

Since there is a shared memory driver for Windows guests that is used with Looking Glass, there's a chance that RedHat will make virtio-fs supported on Windows guests too. I don't know of macOS guests having support for shared memory driver with KVM and RedHat doesn't seem to bother with drivers for that guest so it'd probably be stuck with NFS only.

This is also meant to be much better perf than 9p. Though using shared memory might also mean it is less secure? At least iirc that was one of the drawbacks of using LookingGlass due to the usage of shared memory for performance.

[oiaohm]

9p does not work on Windows or macOS guests.

9p does kind of work on Mac OS with effort.
https://github.com/osxfuse/osxfuse + https://9fans.github.io/plan9port/ Yes the plan9port gives you 9pfuse. This build from source of 9pfuse is absolutely not what OS X users want.

Problem is it not going to be the best performing thing having fuse over head in a virtual machine.

Its not that 9p does not work on MacOS guests is a pure ass to make work. At least you get osxfuse as a built binary.

I don't know of macOS guests having support for shared memory driver with KVM and RedHat doesn't seem to bother with drivers for that guest so it'd probably be stuck with NFS only.

MacOS guests you have a problem. The license of MacOS its only legal to run it on apple hardware. Apple hardware really does not take well to 24/7 running for a build farm so there are not many build farms for open source developers for making and testing MacOS software mostly due these hardware issues.

Since there is a shared memory driver for Windows guests that is used with Looking Glass, there's a chance that RedHat will make virtio-fs supported on Windows guests too.

https://github.com/virtio-win/kvm-guest-drivers-windows

Even if redhat does make a driver for windows don't expect the process to use it is going to be friendly. I like that Microsoft does not provide any support to copies of Windows with the fedora version of the virtio drivers loaded.

There is a way to use fuse on Windows as well to give 9p. Also painful build from source.

This is also meant to be much better perf than 9p. Though using shared memory might also mean it is less secure? At least iirc that was one of the drawbacks of using LookingGlass due to the usage of shared memory for performance.

Really using shared memory does not make it any more insecure than 9p by virtio.

So the Windows and MacOS issues are different but equally nasty why you don't have nice 9p support.

[CuriousTommy]

9p does kind of work on Mac OS with effort.
https://github.com/osxfuse/osxfuse + https://9fans.github.io/plan9port/ Yes the plan9port gives you 9pfuse. This build from source of 9pfuse is absolutely not what OS X users want.

Problem is it not going to be the best performing thing having fuse over head in a virtual machine.

Its not that 9p does not work on MacOS guests is a pure ass to make work. At least you get osxfuse as a built binary.

Has someone actually successfully done this, or is this hypothetical?

MacOS guests you have a problem. The license of MacOS its only legal to run it on apple hardware.

We probably won't ever see a corporation develop QEMU drivers for MacOS, but maybe a Hackintosh developer can develop a driver.

Even if redhat does make a driver for windows don't expect the process to use it is going to be friendly.

I wonder if they would be interested enough to add windows support (I personally hope so).

I like that Microsoft does not provide any support to copies of Windows with the fedora version of the virtio drivers loaded.

Why would they? Wouldn't it be the manufacturer's responsibility to make sure the drivers work?

[oiaohm]

Has someone actually successfully done this, or is this hypothetical?

When virtio-9p was first demo the predecessor to osxfuse was used with the 9pfuse and osxfuse has got patches for 9pfuse in different releases.

Yes windows with windows version of fuse was also in the demonstration. So virtio-9p was support Windows, OS X and Linux.

The windows and OS X sides have not been maintained well since then.

So this is not hypothetical. This is something that has worked at different times and gets to left to bit rot in the middle.

Why would they? Wouldn't it be the manufacturer's responsibility to make sure the drivers work?

Using drivers signed in ways Microsoft does not officially support can be the cause of product activation failures with Windows.

However, systems with cross-signed drivers will not receive Microsoft support.

This line means you load a cross-signed driver and you may find that your copy of windows is no longer activated and will not activate no matter what you do or not receiving windows updates or any other error that happens in the main windows software you will not receive assistance with... This is all Microsoft define of what a test computer is and this is a machine not expected to have general usage so activation or update failures should not be a problem to you right.

Microsoft on drivers is quite a pain these days. In production machines you need to use drivers that have gone though the Microsoft driver publishing process or strange things will happen.

Microsoft makes it quite painful at times to use self signed drivers. This is so vendors of hardware also find this painful.

I was not talking about Microsoft providing support on the drivers. I was talking about Microsoft cutting of support for everything else that is the Windows OS because you have load a driver signed a particular ways.