VirtualBox SF Driver Ejected From The Linux 5.4 Kernel

[chilinux]

I don't understand why the Oracle proprietary file system code ever needs to be accepted into the mainline kernel. Microsoft accomplished the same goal of shared folders using the 9P filesystem driver already in the Linux kernel. 9P is used for this also by QEMU, KVM and Xen. VirtualBox already has a virtio transport implementation as an option for networking. They just need to extend their support for virtio to include 9P over virtio. Looking over the 3,200 lines of code that makes up vboxfs, this looks to just duplicate a subset of the functionality of 9P and contributes nothing new other than using Oracle proprietary method instead of an established open standard. The VirtualBox people are just being lazy in not supporting 9P as the shared folder protocol and have a history of also being lazy in maintaining the code they have gotten mainlined into the kernel in the past. Adding more duplicate functionality that will likely go stale and poorly maintained doesn't seem worth it.

Here is Oracle's 3200+ vboxfs patch:

https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git/commit/?h=staging-linus&id=df4028658f9dd7f82fd190e9db5bd82c05f94625

Here is VirtualBox's documentation indicating the virtio transport is already an option for networking:

https://www.virtualbox.org/manual/ch06.html

Here is the documentation for the existing 9P file system driver:

https://www.kernel.org/doc/Documentation/filesystems/9p.txt

Here is a Microsoft blog post about doing shared folders via 9P:

https://devblogs.microsoft.com/commandline/whats-new-for-wsl-in-windows-10-version-1903/

Here is the documentation for 9P with QEMU:

https://wiki.qemu.org/Documentation/9psetup

Here it is for KVM:

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

Here it is for Xen:

http://xenbits.xenproject.org/docs/4.11-testing/misc/9pfs.html

Anyone have a good reason why the kernel developers shouldn't tell Oracle to either use 9P or just use DKMS to add their proprietary file system driver? Is it likely that any other third party will ever make use of vboxsf instead of just using 9P in the future? Or will the vboxsf always be just a resource for one virtualization app from one company and never really be of generalized use beyond that? How can even MIcrosoft get this right and Oracle still claim they need their own method?

Seriously! Just use the well vetted and already available 9P client driver already in the kernel like everyone else!

[Terrablit]

Hm, so is this driver even compatible with the userspace part of the stack?

Hans de Goede has Oracle's blessing for rewriting all these drivers. He actually got them to commit to not changing the public interface in the short term so that the work wouldn't be wasted later on. So it's as compatible as the originals that Oracle ships are - probably more. He's rewritten a lot of the guest VM stack drivers already, with the resulting efforts being far cleaner and smaller. It was actually pretty exciting, but it started before Oracle started squeezing people for VirtualBox licenses. I don't know that Red Hat would allocate resources for it had that been the case when he started. Virtualbox being a honeytrap should drop marketshare significantly.

Who pushed this code to the kernel so late? Greg KH pull yet another stunt? Did a longtime Redhat Linux developer try to pull a fast one?

I wish whoever pushed this code to Linus would realize that Linux kernel development is not the place for: "It compiles this time! Ok, send it to Linus to merge it!"

Most likely it was a miscommunication on expectations. Hans de Goede does a lot of good kernel work, but I don't think he's frequent in the file system side of things. And each subsystem has its own quirks. There are precedents for new drivers that don't affect anything else to go directly into staging. However, it's only in very limited scenarios, and, as Linus points out, file system drivers aren't one of them.

This went into staging in rc7. It turns out that was a mistake, and apparently it wasn't even supposed to go there at all, but be introduced as a regular filesystem.

So he was probably used to new drivers being eligible to go directly to staging, and didn't realize that the fs subsystem is stricter. It's highly unlikely that this was malicious or outright lazy. It's probably from miscommunications and confusion.

[starshipeleven]

Hm, so is this driver even compatible with the userspace part of the stack?

Afaik yes. The developer said there is an "informal" agreement between him and the Virtualbox developers to try to keep the interface stable

[starshipeleven]

it started before Oracle started squeezing people for VirtualBox licenses

Oracle always trolled companies for Virtualbox businnes licenses.

[caligula]

Afaik yes. The developer said there is an "informal" agreement between him and the Virtualbox developers to try to keep the interface stable

Ok good to know. At the moment, some distros ship with two, somewhat incompatible versions of Virtualbox (that is, 5 and 6).

[Redi44]

So many fucking SJWs in the FOSS development these days...