On Mon, 2009-02-02 at 10:46 +1100, Neil Brown wrote: > On Wednesday January 21, dan.j.williams@xxxxxxxxx wrote: > > Beyond the straightforward fixes the more interesting bits are: > > > > mdmon: support scanning for containers > > This is an attempt to make mdmon more manageable in the initramfs > > environment. Once mdadm has assembled the rootfs we need to switch the > > currently running mdmon instance(s) over to the new mount point. With > > the current code it is awkward to do this in a generic way because a > > script needs to know the names of all the active containers. This tree > > allows a script to do "mdmon /proc/mdstat /newroot" to batch convert all > > mdmon instances to /newroot. This is probably a good time to start > > interfacing with initramfs@xxxxxxxxxxxxxxx to make sure these "initramfs > > helper" changes are relevant, and to see what else is missing. > > We really need a man page for mdmon don't we. Then this sort of text > could be placed there for safely. Yes, a man page is definitely needed. I have no intention of putting mdmon in the initramfs. The only arrays that should be started during initramfs operation are / and possibly /boot. Both of those are started readonly. It isn't until later in the boot process (in rc.sysinit on Fedora/RHEL) that we remount the / device rw. It's my intention to modify our rc.sysinit so that before the root device goes rw, we start the necessary mdmon instances. This should be sufficient to keep the container metadata state consistent with reality and avoids having to add mdmon to the initramfs. However, that means I need to know how to invoke mdmon when mdadm isn't starting it for me. -- Doug Ledford <dledford@xxxxxxxxxx> GPG KeyID: CFBFF194 http://people.redhat.com/dledford Infiniband specific RPMs available at http://people.redhat.com/dledford/Infiniband
Attachment:
signature.asc
Description: This is a digitally signed message part