VM recovery via cloud console
This is not a complete article: This is a Draft, a work in progress that is intended to be published into an article, which may or may not be ready for inclusion in the main wiki. It should not necessarily be considered factual or authoritative.
VM recovery via cloud console
If the VM can't be accessed anymore via ssh or via a local user, the OS can be booted into single user mode or a recovery kernel can be launched, which provides privileged access to the OS image. The only requirement is that he boot manager is accessible and can be modified.
The recovery procedure is not that easy and convenient as you would expect from CentOS, the functionality is the same or at least similar. Most cloud images have the root account locked, so just booting single user won;t help us. However, when a linux based system boots, regardless what falvor it is, the kernel gives up the control into userspace for all userspace related things like running daemons etc. That is done as soon as all the hardware is initilized, then the kenrel runs a single userspace binary, called the init process which always has PID1, in most recent distributions it is either
upstart. Via the bootmanager we are able to mdifythat and tell the kernel to execute a shell instead and manually mount the image filesystem and do our recovery operations. The debian10 image comes with grub2 as well, the menu however looks a little different, but the keys and key cominations we need to use, are all the same. Reboot or boot the system until you see the grub menu then hit
e for edit and remove the serial consoles and add
init=/bin/bash to let the kernel know the new initprocess.
Modify the line after
linux like the below example:
linux /boot/vmlinuz-4.19.0-6-cloud-amd64 root=UUID=d187d85e-8a80-4664-8b5a-dce4d7ceca9e ro biosdevname=0 net.ifnames=0 console=tty0 init=/bin/bash
That will boot the kernel, initializes initrd and executes
/bin/bash as the init process, now we basically landed in memory and are mounted r/o, since the userspace init process is supposed to take care of the root filesystem, the kernel just needs to know where to find it before it hands over the control. To do a useful recovery, the next steps will be remounting the initrd filesystem r/w, mount the OS image disk, chroot into it, set a root password and restart the VM. After a successfull restart, we can login as root. Take note that bash has no
reboot or any powercontrol mechanism, so we have to unmount everything cleanly and stop the VM.
Within our initrd remount the file system r/w:
mount -o remount,rw /
Mount /dev/vda1 (the first primary partition) to /mnt:
mount /dev/vda1 /mnt
We have now the image root filesystem r/w mounted at /mnt, to use tools like
passwd via chroot in there, we need to mount
/dev to gain tty access and
/sys, since we can then also access the network.
mount -o bind /proc /mnt/proc
mount -o bind /sys /mnt/sys
mount -o bind /dev /mnt/dev
Then chroot into
/mnt, which will show an ioctl error for the terminal process group, we can ignore that. Now we can just use
passwd to reset the root password, once done we leave the chroot via
Ctrl+D, unmount out previously mounted mount points and restart the system by using the
Ctrl+Alt+Del submit button on the openstack console page, you can also just stop and start the VM since we unmounted all real filesystems, they are already synced and all buffers flushed to the virtual disk. So we are stricly operating in memory, which is volatile anyway.
After the VM has started, you can now login as user root with the password you have chose. Once completed, remove the root password again, or disable direct root logins via ssh.
Open the console via Horizon and reboot the VM, in the upper right corner the
CtrlAltDel button can be used for that, unless you need to recover a kernel persisten kernel panik. At one point the boot manager shows up, which is currently for all cloud images GRUB or GRUB2. Other would work as well, they will only have a different key sequence to gain access to the append parameters for the kernel. Once the GRUB menu is visiable, hit
e on your keyboard to get into edit mode, you will see something like below.
insmod xfs set root='hd0,msdos1' if [ x$feature_platform_search_hint = xy ]; then search --no-floppy --fs-uuid --set=root --hint='hd0,msdos1' 3ef2b806-efd7-4eef-aaa2-2584909365ff else search --no-floppy --fs-uuid --set=root 3ef2b806-efd7-4eef-aaa2-2584909365ff fi linux16 /boot/vmlinuz-3.10.0-1127.19.1.el7.x86_64 root=UUID=3ef2b806-efd7-4eef-aaa2-2584909365ff ro console=tty0 console=ttyS0,115200n8 crashkernel=auto console=ttyS0,115200 LANG=en_US.UTF-8 initrd16 /boot/initramfs-3.10.0-1127.19.1.el7.x86_64.img
Now navigate to the line which starts with
linux16, here all
console parameters need to be removed, since qemu uses the serial console (ttySX), we would have to go onto the compute node directly and attach it there to a terminal. The easier option is just to leave
console=tty0 in there. If we want to have the filesystem from the image mounted r/w we would have to change the paramter
rw, but that can be done later as well if something needs to investigated r/o is a very good option to leave timestamps intact on inodes. Centos has a parameter to interrupt the boot process in an early stage, which is
rd.break. The linux16 line should then look like the below (the order of the parameters do not matter):
linux16 /boot/vmlinuz-3.10.0-1127.19.1.el7.x86_64 root=UUID=3ef2b806-efd7-4eef-aaa2-2584909365ff ro rd.break console=tty0 crashkernel=auto LANG=en_US.UTF-8
To boot the kernel with the changes hit
/sysroot you will find the ro mounted filesystem from the image, you can chroot into it or modify it directly. To make it rw, it needs to be remounted:
mount -o remount,rw /sysroot.
The steps are very similar compared to the CentOS7 recovery procedure, the option
console needs to initialize a tty instead of a serial console and
rd.break will launch the recovery environment.
root=UUID=c7b1ead0-f176-4f23-b9c7-299eb4a06cef ro console=tty no_timer_check net.ifnames=0 crashkernel=auto