[email protected]
2022-02-15

This directory contains a bunch of scripts and working directories to build a
Linux guest kernel and a tinycore (TC) image as an initrd.

The scripts expect to be run from this directory and require sudo for mucking
with the initrd.  Sorry.

Typical usage: run full_guest_build.sh $FULL_PATH once, then as you muck with
things, run rebuild_cpio_and_linux.sh $FULL_PATH.  If you aren't changing the
kernel, run SKIP_KERNEL=1 ./rebuild_cpio_and_linux.sh.


Scripts:
--------------------
- full_guest_build.sh: produces vmlinux and initrd from scratch.  Calls the
  other scripts.

  Run with: ./full_guest_build.sh $FULL_PATH_TO_LINUX_REPO
  - copies kernel.config to $FULL_PATH_TO_LINUX_REPO/.config
  - setup_tinycore.sh (creates tc_root/)
  - copies stuff into tc_root/
  - rebuild_cpio_and_linux.sh

  Customizations:
  - CUSTOM_BINARIES: programs from your machine to add to tc_root/usr/local/bin
  - CUSTOM_REMOVALS: files to remove from tc_root.  e.g. libstdc++.
  - ssh keys (looks for my key named "db_rsa", provide your own keys)


- rebuild_cpio_and_linux.sh: build linux (vmlinux & bzimage) and the initrd,
  using the contents of tc_root.  If you manually changed tc_root (edited files,
  added programs) or changed the kernel (including modules), run this.

  Run with: ./rebuild_cpio_and_linux.sh $FULL_PATH_TO_LINUX_REPO
  - builds the kernel modules
  - cpios tc_root; puts it in this directory.
  - builds the kernel; puts it in the linux repo.

  You can skip the kernel build if you want with SKIP_KERNEL=1.


- setup_tinycore.sh: downloads TC, extracts to tc_core, and installs PACKAGES.
  Edit the PACKAGES variable to add any tinycore packages you want, e.g.
  openssh.  The script should find and include any dependencies.  If you change
  any of the Magic Symlink Overrides, e.g. rcS-link, you need to rerun this.

  Run with: ./setup_tinycore.sh
  - Optionally overrides rcS and tc-sys.sh.
  - Misc console setup
  - Downloads and extracts PACKAGES and dependencies to tc_root
  - SSH server setup, including hokey server private keys


Various directories:
--------------------
- kernel_mods: storage for kernel mods created during the kernel build.
- progs: compiled helper programs to run in the guest.
- tc_root: TC's root filesystem.  Extracted and edited by the scripts
- tczs: storage for downloaded TC programs, e.g. openssh and openssl
- vm-apps: Akaros scripts to run an app as a guest VM, e.g. an ext4 9p server.


Magic Symlink Overrides:
--------------------
You may noticed symlinks such as "rcS-link -> xxx".  If these symlinks point to
a file, it will override or install the file into the TC image.  These are quick
ways to change up what your guest will do at boottime without editing tc_root/
directly.

- rcS-link: replace the low-level /etc/init.d/rcS script.  This short-circuits
  most of TC's initialization.  The default rcS mounts /proc, /, fstab, and runs
  tc-config.  tc-config does the bulk of TC's init, such as modprobing, cmdline
  parsing, udev stuff, etc.  Ultimately, this calls /opt/bootlocal.sh and
  tc-sys.sh.  For an example of how to override rcS, check out
  rcS-example-apps-and-info.  Note that "rcS-link -> xxx" is a dead link, so we
  will not automatically override rcS.

- tc-sys.sh: script run during the guest's boot, called from TC's
  /opt/bootlocal.sh.  I use this for setting up networking.  If you override
  rcS, you'll need to call this manually from your own rcS.


Misc helpers:
--------------------
- cat_cpio_gz.sh: Dumps the contents of an initramfs to easily see what takes up
  the most space.

- embed_payload.sh: Embeds a payload with a shell script and makes it
  executable.  Used for Akaros's vm-apps.


Example Commands to run the Virtual Machine:
--------------------
- Akaros:
  	vmrunkernel -k tinycore_cmdline -n vnet_opts -N 8 -i initramfs.cpio.gz vmlinux

	tinycore_cmdline:
		earlyprintk=akaros	# optional
		console=hvc0
		mitigations=off
		nozswap			# for tinycore faster boots
	vnet_opts:			# all optional
		snoop
		nat_timeout = 30
		map_diagnostics
		port:tcp:23:22

	With this setup, you can ssh to localhost:23 and it'll port-foward to the guest.

- Tun/Tap:
	A bunch of VMMs on Linux want a tun/tap setup.  I use tun-up.sh, which
	is mostly this:

	ip link add br-vm type bridge
	ip addr add 10.0.2.2/24 dev br-vm 2>/dev/null || true
	ip addr add fd0:1234:4321::2/64 dev br-vm
	ip link set up dev br-vm
	ip tuntap add tap-vm mode tap
	ip link set up dev tap-vm
	ip link set tap-vm master br-vm

	This creates a virtual network similar to the one that Qemu's usermode
	networking will make: guest will get 10.0.2.15, host gets 10.0.2.2.
	Akaros's VMM NAT does the same thing.

	It also adds an IPv6 address, in case your host kernel is IPv6-only.
	The guest will use fd0:1234:4321::15/64

- Qemu:
	Note that qemu takes the bzImage, but not vmlinux.  There might be some
	kernel config issue, but my qemu with the vmlinux from these scripts
	will complain with "qemu: linux kernel too old to load a ram disk".

	If you have a tun/tap setup, with a tap named "tap-vm":

	qemu-system-x86_64 -s -enable-kvm -cpu host -smp 8 -m 1024 -nographic \
	-kernel arch/x86/boot/bzImage -initrd initramfs.cpio.gz \
	-append "console=ttyS0 nozswap" \
	-device virtio-net-pci,netdev=vmnet,mac=00:01:02:03:04:0b \
	-netdev tap,id=vmnet,ifname=tap-vm,script=no,downscript=no

	You can also use the old user-mode networking (forwarding host 5555 to
	guest 22).  Replace the -device and -netdev with:
	-net nic,model=e1000 -net user,hostfwd=tcp::5555-:22

	There are plenty of other options.

	Either way, the guest will have IP 10.0.2.15, gateway 10.0.2.2 (the
	host).  For usermode networking, that comes from DHCP.  For the tun/tap,
	that comes from the magic MAC address 00:01:02:03:04:0b.  The tinycore VM
	has an init script that looks for that address and statically configures
	networking.

	You can add other linux command line args, such as console=hvc or
	"console=ttyS0 earlyprintk=serial".  hvc (virtio-cons) didn't get the
	bootspew that ttyS0 gets, which helps with debugging, so YMMV.

	To stop the guest, try "Ctl-A, X".  "Ctl-A, C" gives you the qemu
	monitor.

	The biggest thing with qemu for me is that it jacks up your terminal,
	such that the console text won't wrap anymore.  Try:

	tput smam

	I put that in all of my qemu scripts and even have a bash alias for it.

- Cloud Hypervisor:

	./target/release/cloud-hypervisor --console off --serial tty \
	--cpus boot=8 --memory size=1024M \
	--kernel vmlinux --initramfs initramfs.cpio.gz \
	--cmdline "console=ttyS0 nozswap" \
	--net tap=tap-vm,mac=00:01:02:03:04:0b,ip=,mask=

	Same as with qemu, hvc0 works, but is a little on the quiet side.
	nozswap is for tinycore.  The MAC address tells the guest to set itself
	up as 10.0.2.15.

	To stop the guest, there's some involved incantation to tell
	cloud-hypervisor to shutdown, but I just:

	killall cloud-hypervisor

- SSHing to the guest under various environments:

	If the guest is listening on host port 5555 (from my example qemu
	usermode networking), set up your ~/.ssh/config like so:  The
	IdentityFile is the one you specified in full_guest_build.sh.

	Host qemu
		Hostname 127.0.0.1
		User root
		Port 5555
		IdentitiesOnly yes
		IdentityFile ~/.ssh/db_rsa

	This ought to work for a tap:

	Host qemu-tap
		Hostname 10.0.2.15
		User root
		Port 22
		IdentitiesOnly yes
		IdentityFile ~/.ssh/db_rsa

	But if your workstation has an "ssh-unfriendly" policy, you might have
	issues.

	Instead, you can use port forwarding.  Set up a tunnel from your
	workstation's 5555 to the VM guest's 22 on the tun/tap network of your
	workstation.

	ssh -L 5555:10.0.2.15:22 127.0.0.1

	Now you can ssh to the guest just like with qemu's usermode net port
	forwarding. (ssh qemu, from above).

	Similarly, if you're running the VM on *another* host that you have ssh
	access to, but you can't ssh to any port other than 22, you can forward
	from your workstation's 2345 to the host's port where the guest is
	listening.

	ssh -L 2345:127.0.0.1:23 $HOST

	I'll use that to ssh from my Linux workstation to an Akaros $HOST
	(though it could also be Linux).  The tunnel endpoint is $HOST,
	connecting to it's localhost:23, which I'll often port-forward to the
	guest (e.g. in Akaros's vnet_opts above).

	Host linux-guest
		Hostname 127.0.0.1
		User root
		Port 2345
		IdentitiesOnly yes
		IdentityFile ~/.ssh/db_rsa

	If you don't want to mess around with ssh-config, you can do:

	ssh -o IdentitiesOnly=yes -o IdentityFile=db_rsa root@fd0:1234:4321::15

	or even make a bash alias:

	alias ssh-guest-v6="ssh -o IdentitiesOnly=yes -o IdentityFile=db_rsa root@fd0:1234:4321::15"