Differences between revisions 26 and 27
Revision 26 as of 2011-02-23 15:20:42
Size: 4758
Editor: ?maximilianattems
Revision 27 as of 2011-07-04 17:49:02
Size: 4780
Comment: add to CategoryBootProcess
Deletions are marked like this. Additions are marked like this.
Line 120: Line 120:
CategoryKernel CategoryKernel | CategoryBootProcess

Translation(s): none

initramfs-tools is used in Debian to build the initramfs that contains the early userspace and the needed kernel modules to be able to mount any possible root.


Boot with


or such to shorten the time of waiting for the root device to come up. Also it adds time to scsi device to settle before calling mdadm or lvm thus excluding potential races. Alternatively you may have success with


To fix wrong fs recognition work around boot with


Rescue shell (also known as initramfs shell)

Read man initramfs-tools to learn about the break=something kernel parameter (where valid arguments for something are: top, modules, premount, mount, mountroot, bottom, init), which starts a debug shell. You can try, for example, break=premount. You can edit /boot/grub/grub.cfg to add this to the end of the kernel line, or you can do it interactively from the grub boot menu: "e" to edit, and "b" to boot once you've edited the kernel line.

Ok, now you landed in the initramfs debug shell. This is a surprisingly complete environment provided by "busybox". You can use vi, for example, but not everything is fully the same as bash. For example, sleep only accepts integer arguments. You may find more complete commands in /bin (for example, /bin/sleep does accept decimal arguments, but you will need to use /bin/sleep in your scripts).

The following steps should help to narrow down the problem.

  • Check root device existence
     ls -l /dev/[hs]da*
    If above does not exist checkout for eventual driver conflict or other information
  • Check root boot argument
    • It needs to be the same than above root device
     cat /proc/cmdline
  • Check correct fs recognition
     # for releases >= squeeze:
     blkid -o value -s TYPE /dev/sda1 
     # for releases <= lenny:
     /lib/udev/vol_id -t /dev/sda1
     # compare with:
     fstype /dev/sda1
    • If the output of aboves command differ, your partition was not cleanly formated. For example mkfs.ext2 does not clear all sectors when formating a partition.
  • Check if the ata/ide/scsi driver is loaded
     cat /proc/modules
  • saving output from an initramfs script into a logfile you can review later
    • Add "debug" bootparam and find the logfile in /dev/.initramfs/initramfs.debug. (You'll need initramfs-tools 0.92m at least for that feature. Also see further down in this document on how to save other data, or save data across reboots.

You can exit the shell with exit or ^D and the boot should resume.

initramfs content

It can happen that due to a bug in the initramfs creation a newer initramfs lacks important libraries or utilities. Easiest way is to check contents with lsinitramfs

  •  lsinitramfs /boot/initrd.img-2.6.32-5-686 | less

Old way is to compare the file list of working contra non-working initramfs.

  •  cd $TMPDIR
     mkdir initramfs{,-old}
     cd initramfs
     gunzip -c -9 /boot/initrd.img-2.6.32-5-686 \
      | cpio -i -d -H newc --no-absolute-filenames
     find > ../initramfs.content
     cd ../initramfs-old
     gunzip -c -9 /boot/initrd.img-2.6.32-5-686.bak \
      | cpio -i -d -H newc --no-absolute-filenames
     find > ../initramfs-old.content
     cd ..
     diff -u initramfs-old.content initramfs.content
    If the intiramfs creation of both those images went well there should be no diff. If you are comparing initramfs for different kernels you should see a difference in the modules aka other dir and newer drivers in /lib/modules/$(uname -r)

Saving debug information

If you specify debug as a kernel option during boot, initramfs-tools (>= 0.92m) writes shell script traces to /dev/.initramfs/initramfs.debug, which will be available after the system has booted.

You can write other data to files in that directory to be able to access them later, when the system has booted. Since /dev is a tmpfs however, they won't persist across a reboot.

If you need to get debug data off the system and it can mount the root filesystem, you can do something like

mount -o remount,rw /root
dmesg > /root/root/initramfs.dmesg
mount -o remount,ro /root

If that does not work, use a USB stick: plug it in, observe the kernel output, mkdir /mnt, mount the device node there, and write to the flash drive. Remember to umount before pulling out the stick.

CategoryKernel | CategoryBootProcess