This is the prototype ivshmem-server implementation.
We are breaking from the general philosophy of ivshmem-server inter-vm communication. In this prototype there is no concept of inter-vm communication; guests can only talk to daemons running on host.
Cuttlefish requires the following packages to be installed on your system:
Users running cuttlefish must be a member of a relevant group enabling them to use virsh tool, eg. libvirtd.
Group is created automatically when installing libvirt-bin package.
Users may need to log out after their membership has been updated; optionally you can use newgrp to switch currently active group to libvirtd.
sudo usermod -a -G libvirtd $(whoami)
Once configured, users should be able to execute
$ virsh -c qemu:///system net-list --all Name State Autostart Persistent ---------------------------------------------------------- [...]
You will need to update your configuration /etc/libvirt/qemu.conf to disable dynamic permission management for image files. Uncomment and modify relevant config line:
dynamic_ownership = 1 user = "libvirt-qemu" group = "kvm" # Apparmor would stop us from creating files in /tmp. # TODO(ender): find out a better way to manage these permissions. security_driver = "none"
and restart libvirt-bin service:
sudo service libvirt-bin restart
Create a directory to host your files. This directory will need to be accessible not only by you, but by libvirt, too, and libvirt will likely update ownership of your files. My recommendation is to use either /srv/cf or /run/cf folder.
mkdir /srv/cf
sudo chown -R libvirt-qemu:root /srv/cf
sudo setfacl -m u:${USER}:rwx /srv/cf
sudo chmod 0770 /srv/cf
If you‘ve done the above right, you should be able to create files there, even if you’re not working on behalf of libvirt-qemu user or root group.
You will need to copy (or link) the following files from your build directory:
This artifact needs to be built and copied from this repo:
gce_ramdisk.img
build: bazel build //guest:gce_ramdisk, copy or link file from workspace root's bazel-bin folder.
These files need to be manually created:
data.img and
cache.img
truncate -s 10G data.img mkfs.ext4 data.img truncate -s 2G cache.img mkfs.ext4 cache.img
After you‘re done linking/copying/creating files, set posix acls on these files so that you don’t lose access to them:
setfacl -m u:${USER}:rw /srv/cf/*
Done.
permission denied errorslibvirt is not executing virtual machines on behalf of the calling user. Instead, it calls its own privileged process to configure VM on user‘s behalf. If you’re seeing permission denied errors chances are that the QEmu does not have access to relevant files OR folders.
To work with this problem, it's best to copy (not link!) all files QEmu would need to a separate folder (placed eg. under /tmp or /run), and give that folder proper permissions.
➜ ls -l /srv/cf total 1569216 drwxr-x--- 2 libvirt-qemu eng 180 Jun 28 14:27 . drwxr-xr-x 45 root root 2080 Jun 28 14:27 .. -rwxr-x--- 1 root root 2147483648 Jun 28 14:27 cache.img -rwxr-x--- 1 root root 10737418240 Jun 28 14:27 data.img -rwxr-x--- 1 root root 825340 Jun 28 14:27 gce_ramdisk.img -rwxr-x--- 1 root root 6065728 Jun 28 14:27 kernel -rwxr-x--- 1 root root 2083099 Jun 28 14:27 ramdisk.img -rwxr-x--- 1 root root 3221225472 Jun 28 14:27 system.img
Note: the /run/cf folder's owner is libvirt-qemu. This allows QEmu to access images - and me to poke in the folder.
Now don't worry about the root ownership. Libvirt manages permissions dynamically.