simple-kdump

About

This is a very simple kdump setup for Archlinux.

It has one and one function only, save the vmcore to /var/crash/ of the root fs.

There are only two dependency, systemd and makedumpfile.

HOW TO USE

Install from AUR

• Modify simple-kdump.conf Note the kernel doesn't need to match the kernel you want to debug. So choose your favorite/smallest/whatever kernel and its initramfs.

  Some kernel configs must be enabled, fortunately linux and linux-lts all have the options enabled for x86_64 Archlinux.

  For ArchlinuxARM linux-aarch64 kernels, CONFIG_CRASH_DUMP=y and CONFIG_PROC_VMCORE=y options are needed, and the PR is already submitted.

  And just use the regular boot option without crashkernel= option.

• Add crashkernel= option to boot entry and reboot Normally I'd recommend 512M as I have hit cases where 256M is not large enough to decompress the huge 50M kdump specific initramfs generated by Fedora's dracut.

• Start and enable simple-kdump-setup.service

• Test if the setup works By sync; echo c > /proc/sysrq-trigger with root privilege.

  The system should reboot into an emergency shell, with makedumpfile running at the background:

[  OK  ] Started Emergency Shell.
[  OK  ] Reached target Emergency Mode.
[  OK  ] Started Collect the vmcore for simple-kdump.

There will be a prompt for the emergency shell login, but you can/should ignore that.

After the vmcore is collected, the system will reboot back to your regular setup. And /var/crash/ will contain the last collected vmcore file.

IMPLEMENTATION

Unlike other distros, Archlinux doesn't have an easy to use way to setup kdump to collect vmcore from crashed kernels. There are projects like kdumpst backed by Valve to improve the situation.

But unfortunately I hate GRUB2 and have too many custom kernels that GRUB2 can never properly detect.

The freedom to choose different kernels and bootloaders is exactly why I'm using Archlinux, and also why it's so hard to implement kdump.

So I just come up with the super simple setup.

The core of the simplicity comes from letting user to choose whatever kernel/initramfs combination as the kexec kernel/initramfs, not reusing the current kernel/initramfs, as long as the combination can boot and have needed options enabled.

This avoids two complex workload:

• Detect the booting kernel/initramfs/cmdline
• Rebuild a dedicated kdump initramfs

Instead use systemd.unit= option to let the kexec environment to boot into a special emergency target which also collects the vmcore and reboot.

But this also means there is a limitation.

If the crash caused serious filesystem corruption that your local fs can not be mounted, the setup will not work at all.

Fortunately even during my filesystem development work, I haven't really hit such case.