-
Notifications
You must be signed in to change notification settings - Fork 45
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Duplicate /dev mounts #242
Comments
Hrmm... Looks like there's a few issues .... if we keep The @bensallen ... You're the last I see that made changes here. What are your thoughts on this? |
Can we revive this? I'm happy to throw together a PR if there's some consensus on how. |
I have some ideas ... there's an if block in We also have: I theory this file just needs to have the |
If we want to maintain the old behavior of supporting kernels without devtmpfs, as we do currently, can we just |
Also agreed that https://github.com/warewulf/warewulf3/blob/development/provision/initramfs/init#L49 should move up above the dmesg setting console level to 1. |
The init shouldn't be dependent on them already being mounted should it? From a boot initramfs?
That should work. We're actually mounting it twice... once in |
Coreutils' switch_root appear to move /dev, /proc, /sys, /run if they're mounted. We might as well too as busybox's switch_root doesn't do this.
Right this was done to follow the original legacy mdev + cp method of populating /dev. We don't support any kernel's that old at this point (RHEL6 support is dropped for example). Thus as Lowell suggest we could just simplify this whole thing, drop 70-devtree completely. Do the three Any case you can think of where someone might be futzing with $NEWROOT/dev via postscript? As this change would force said futzing to using /dev instead. |
It seems like a pretty odd move to futz with $NEWROOT/dev. I can't think of any realistic use-cases for it with modern kernels. Maybe just call that "unsupported"? And yeah, completely agree that it needs to move up. I should have caught that the first time around. |
Yea agreed, so I think we should:
|
I say if we're cleaning house here, then go a head and remove it and just mark
No... but it wouldn't surprise me for someone to want to use |
PR #255 attempts to implement this as discussed. |
Closing with #255 merged. |
When a node is vnfs provisioned, it ends up with two entries for the
/dev
mount in/proc/mounts
(etc). After digging a bit, this happens becauseprovision/initramfs/capabilities/provision-vnfs/70-devtree
mounts dev in$NEWROOT/dev
, but the one in the initramfs remains.For almost everything, having a duplicate
/dev
mount doesn't hurt anything. I ran into an issue specifically with runninglibvirt
on a warewulf provisioned node, where the duplicate/dev
entry confuseslibvirt
(specifically, because of a not-so-robust routine used inlibvirt
, it fails to mount/dev/pts
in the domain namespace).Here's one way this could be managed:
provision/initramfs/capabilities/provision-vnfs/70-devtree
may no longer be needed. It has the ability to populate a/dev
for linux that doesn't supportdevtmpfs
, butdevtmpfs
has been in the kernel for the last decade. Maybe just get rid of it.initramfs/init
handles mounting/proc
,/sys
, and/dev
initially, it makes sense to do the following:I've tested this, and it resolved this issue for a centos7 vnfs image.
Happy to create a PR if this seems reasonable.
The text was updated successfully, but these errors were encountered: