I just updated mdadm, and noticed the message recommending I read /usr/share/doc/mdadm/README.upgrading-2.5.3.gz:
Upgrading to mdadm 2.5.3 from previous versions=============================================== With version 2.5.3, the Debian mdadm package now requires a mdadm.conf file. This file should be in /etc/mdadm/mdadm.conf, although /etc/mdadm.conf will also be read if the former is not present, but it is less preferred. The reason the configuration file is now required is because it was impossible to come up with robust heuristics to replace the previously used mdrun, while addressing its primary deficiency of not honouring the super-minor field and randomly assembling devices (see #354705). Also see /usr/share/doc/mdadm/README.mdrun Systems without a mdadm.conf file~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If your system does not have an mdadm.conf file, the package creates one for you during the installation phase. Furthermore, a configuration file is created during the updating of the initramfs (initial ramdisk), if no file is found at the time. In both cases, the running system is used as a reference, meaning that the file is generated automatically in such a way as to reflect the running system. This should work in most cases. I strongly suggest, however, that you inspect the generated file (/etc/mdadm/mdadm.conf) before you reboot. If you find any anomaly or mistakes, correct them and proceed with the next step ("Systems with a mdadm.conf file"). Systems with a mdadm.conf file~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If an existing configuration file is found, it is *ignored* until you checked it and gave mdadm permission to use it. Even though this is a nuisance to some users, it is a necessary measure: previous versions of mdadm did not necessarily use the information in this file, even if it existed; thus there is no guarantee that the file properly describes the system's configuration. Therefore, you are required to inspect /etc/mdadm/mdadm.conf or/etc/mdadm.conf (whichever one is present, the first gets priority if both are present) and ensure that all arrays are properly identified. Here are a number of recommended checks: - Verify that all arrays referenced by /etc/fstab, /etc/crypttab, your LVM metadata, and whatever other subsystem uses MD arrays (RAIDs) on your machine have a corresponding line in the configuration file. Make sure to verify that your bootloader refers to the proper device name, in case your root filesystem is on an MD array. In particular, verify that the device node name is exactly the same; /dev/md6 is *not* identical to /dev/md/6. Partitionable arrays are a slight exception: if /dev/md_d0p3 is referenced, you need an entry for /dev/md_d0 in the configuration file. - Compare your file with the output of /usr/share/mdadm/mkconf . In particular, make sure that the UUID matches for each array, whenever a UUID is specified. Also compare the values of super-minor, name, and devices. Only one match identifier (UUID, super-minor, name, devices) is needed for each array, but if multiple identifiers are specified, all must match. See mdadm.conf(5). Identifying arrays by UUID is the preferred method. Once you have verified that the configuration file is accurate, you need to let mdadm know, and update the initial ramdisk. This is accomplished with the following two commands: rm -f /var/lib/mdadm/CONF-UNCHECKED update-initramfs -u -k all Depending on your setup, mdadm should print an appropriate informational message. Please make sure that it is in accordance with what you would expect. What to do if my system does not boot anymore?~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First thing: don't panic! If your system does not boot anymore as a result of an mdadm upgrade, it is *very* unlikely that you lost any data. Having said that, it is important that you exercise great care and understand what you are doing. If you do not know how to revive a system that does not boot because it fails to assemble MD arrays (RAIDs), consider asking someone who does for help. If you are using initramfs (which is the default since Debian "etch"), it will probably be easy to revive the system. If you are experiencing problems with another initial ramdisk alternative, you will have to resort to using a rescue disk, such as the debian-installer CD/DVD, or a live system, such as Knoppix. With initramfs, specify 'break=mount' as a kernel boot command line option, and commence the boot process. Eventually, you will be dumped into a shell. From here, you can execute ./scripts/local-top/mdadm and inspect the output. If you want to see what exactly the script is doing, run sh -x ./scripts/local-top/mdadm instead. Your primary goal is the assembly of the array(s) needed to bring up your root filesystem. If the above script fails to do that, take note of its output (for a later bug report), and then proceed to simply assemble the array by hand. For instance, if your root array is on /dev/md1, which is made up of/dev/sd[abc]1, just run: /sbin/mdadm -A --auto=yes /dev/md1 --run /dev/sd[abc]1 /sbin/mdadm -Q /dev/md1 # to verify If your root filesystem is on multiple arrays (e.g. on LVM), repeat for each constituent array. When you are done, hit ctrl-d and watch the system boot. When it is back up, carefully inspect your mdadm.conf file and ensure that INITRDSTART is properly set in /etc/default/mdadm; set it to 'all' to be safe. In case you are not using initramfs, you need to find out what's going wrong during the boot process. Both yaird and initrd use a single mdadm --assemble call to bring up the device holding the root filesystem. If that command fails, you may need to recreate the image and make sure it gets the right parameters. If you rely on the kernel to assemble the arrays (MD support built-in, not modular, and partitions of type 0xfd) and the process fails, the superblock information is probably not accurate. In such a case, try to assemble the arrays from a rescue disc or live system just as you would expect the kernel to do it. Afterwards, verify the superblock information (mdadm --examine). If you continue to experience problems, please file a bug report with all the relevant information. The reportbug tool will automatically include the most relevant data in the report. -- martin f. krafft
Mon, 23 Sep 2006 23:49:10 +0100$Id: README.initramfs-transition 89 2006-08-08 09:33:05Z madduck $
Ugh. I'll have to take care of this on Monday.