USB storage plumbing for the Linux Kernel Library
an invention by dmdiss
Updated about 1 month ago. 2 hacker ♥️. 1 follower.

This project builds on my previous efforts to plumb the Linux Kernel Library (LKL) into USB storage devices. This time I plan on mostly ignoring Android and will instead focus on lklfuse USB support.

lklfuse uses virtio over MMIO to route block device I/O between the host block layer and LKL guest kernel. xHCI emulation appears to be how proper USB traffic routing between host and guest would normally be implemented, but I'm unfamiliar with (and daunted by) the specifications and protocols involved. Given that both guest and host are Linux, I'm considering a simpler approach using USB/IP over virtio, but would need to investigate further to see whether it's viable.

Goals

  • Explore options for USB traffic routing between LKL guest and Linux hosts
    • Learn about and play with the USB/IP and virtio protocols
    • Read about the kernels usbdevice_fs.h APIs
  • Implement lklfuse USB storage support
  • stretch goal: map a LUKS/dm-crypt device
  • stretch goal: benchmark USB against virtio-block devices

Looking for hackers with the skills:

Nothing? Add some keywords!

This project is part of:

Hack Week 24

Activity

  • about 2 months ago: mwilck liked this project.
  • about 2 months ago: mbrugger liked this project.
  • 3 months ago: dmdiss started this project.
  • 3 months ago: dmdiss originated this project.


    • Comments

    • dmdiss
      about 1 month ago by dmdiss | Reply

      post hack-week report:

      I spent plenty of time on research, given that I wasn't very familiar with
USB or virtio low-level transport and protocol details.
Read original USB/IP paper:
- https://www.usenix.org/legacy/events/usenix05/tech/freenix/hirofuchi/hirofuchi.pdf
- USB Request Block (URB) as underlying USB/IP protocol unit appears to
  suit virtio-mmio transport quite well
  + avoids breakdown into Transfer Descriptor microframes
  + (isochronous) I/O scheduling complications can be initially ignored
Read kernel documentation on virtio, virtqueue details and test examples
- https://www.redhat.com/en/blog/virtqueues-and-virtio-ring-how-data-travels

Played with existing kernel / user-space USB/IP network implementation
- https://github.com/rapido-linux/rapido/commit/9ca2d7a0e27940d59f1eaf7d7e9dc559e77c89ef
- https://github.com/rapido-linux/rapido/commit/c4e0cff00bfdaea03b960311266db5f59f7c5390
- wireshark PDU dissection works well

For Linux Kernel Library (LKL) / lklfuse plumbing I revisited my walkley
project from prior hackweeks, and pushed new WIP branches:
https://gitlab.com/ceph-walkley/walkley/-/tree/wip_hackweek_2024
- rebased atop current LKL master
- added new walkley-usb binary
- added mostly-stubbed virtio-usb host-side driver
- started work on guest kernel side virtio-usb driver, reusing as much
  of usbip/vhci-hcd as possible
  + TCP tx/rx threads spawned on sysfs/vhci/attach can be swapped out for
    corresponding virtio transport logic
  + usbipd user-space component can mostly be avoided, as virtio has
    better options for device discovery
- still quite a bit more work needed to support full USB/IP protocol
  over virtio-mmio
  + QEMU / standalone virtio host implementation would likely also be
    needed, so that mainline inclusion isn't dependent on LKL

    Similar Projects

    This project is one of its kind!