Bug#361150: Kernel install process contains a time window that can lead to an unusable system
Package: linux-2.6
Severity: normal
This is about the kernel image installation process (all linux-image-*)
There is a critical time window during install that occurs if
- multiple packages are being installed at once (for exemple during
"apt-get upgrade")
- *AND* the same kernel is being updated
during which time, an interruption of the installation process can lead
to a non-bootable
system (non coherent kernel/modules combination, invalid kernel map for
bootloader, etc..)
The installation procedure first unpacks and replaces the actual running
kernel.
If anything happens during that time several bad things can happen :
- The initrd image (if needed) is not built
- The bootloader is not updated to reflect the disk location map of the
new kernel (lilo ?)
Unfortunatelly, this can happen very easily if one of the package that
is loaded during
an 'apt-get upgrade' happens to fail to install (recent exemple :
manpages-dev and modutils
conflict over a single file). However, this can also occur for other
various reasons (power
outtage, CTRL-C, etc..)
The, hmmm.. solution (?) - would be to actually *install* the new kernel
ONLY when all steps
have been taken to ensure that the system is ready (or about ready) to
boot that kernel
(read : initrd is ready)..
There is probably also a problem with /lib/modules.. I believe the
solution here
is to unload the /lib/modules somewhere else, chroot to create the
initrd, and mv the
right /lib/modules/XXX just before finishing linux-image package
installation setup.
It is possible the previously mentioned solutions are not applicable to
the way dpkg/apt
work. But I am not familiar enough with the debian install procedures to
tell whether this
is the case or not.
It may also be impossible to completelly remove any critical time window
(eg: lilo update needs
everything to be in place first) - but at least it could be reduced to
encompass only a single installation step.
If it is believed this report belongs to another place (apt, dpkg,
etc..) please
don't hesitate to say so - I'll report wherever it is necessary.
Finally - the workaround is usually to boot from a rescue medium (or
another kernel if
feasable) and restart the installation process - until the kernel is
'setup' properly.
-- System Information:
Debian Release: testing/unstable
APT prefers unstable
APT policy: (500, 'unstable')
Architecture: powerpc (ppc64)
Shell: /bin/sh linked to /bin/bash
Kernel: Linux 2.6.16-1-vserver-powerpc64
Reply to: