Kernel 525 Changed Behavior

[Jeremy Katz]

Glossing over a few implementation details...

On Tue, 2004-08-24 at 14:51 -0700, Steve G wrote:
> >udev was using a dynamic ramfs based dev and thus had to set the
> >contexts.  Which then failed because it was ramfs which doesn't have an
> >xattr handler.  See fedora-selinux-list for more gory details.
> 
> I saw that. But there's 3 issues here that are very important that were not
> discussed in that thread:
> 
> 1) It tried to modifify my system and did not log the fact that it was about to
> do that.

It's trying to modify a ramdisk.  And this is going to happen everytime
udev creates a device node (which already gets created).  Part of udev
creating the device node is setting the SELinux xattr (as otherwise,
things get screwed up with only the default context)

> 2) It failed and did not log that fact either.

The failing is a bug because tmpfs/ramfs don't support xattrs, usually
not the case.  And actually, with the reintegration of the SELinux stuff
into udev proper (in progress, see fedora-selinux-list), logging will be
better on failure.

> 3) It scanned my system and tried to "fix it" automatically. Usually for fs
> problems it drops to nash and lets you (or the admin) do the fixing manually.

There's no scanning involved here.  The device node doesn't exist before
udev creates it.  The "filesystem" for /dev is created on boot each time
in ram.

Cheers,

Jeremy