On 05/02/2011 01:09 PM, Dave Reisner wrote:
On Mon, May 02, 2011 at 12:28:41PM -0300, Gerardo Exequiel Pozzi wrote:
On 05/02/2011 11:23 AM, Thomas Bächler wrote:
Am 02.05.2011 16:00, schrieb Gerardo Exequiel Pozzi:
N0: device-mapper/LVM udev rules must be parsed for initramfs creation
to add OPTIONS+="db_persist"
Patch looks good now.
Great!
I am unsure and confused why we would need this db_persist option. In
the past, we killed udev and the db completely and everything worked out
fine. What kind of state would need to be persistent? Is upstream adding
this option?
We've never killed the DB on leaving early userspace because the DB
previously lived in /dev/.udev.
Yes, see above.
We did a mount --move of /dev to
/new_root/dev and continued on our merry way. Nothing different with the
DB in /run/udev. We transfer that tmpfs over to the root and everything
is preserved.
Yes.
Yes, that is what I not understand, If I am not wrong this was
before devtmpfs times.
Not sure where devtmpfs comes into play. This is all created by udevd in
userland regardless of the filesystem /dev is mounted on.
In old times (about one year ago), "/proc" "/sys" was discarded from
initramfs [#1]. Also no separate "/dev" exist at initramfs [#2] stage,
so all "udev DB" was discarded at all.
[#1]
http://projects.archlinux.org/mkinitcpio.git/commit/init?id=64c0a72e9fc0a83d4313f37e09b2fa5390f2d28d
[#2]
http://projects.archlinux.org/mkinitcpio.git/commit/init?id=3bb6c683094490e88883b20b5c6cffc17df165e7
Maybe this is needed in case if rules on initramfs differs from real
root. Dave can you ask to Kay on IRC about this? (I am going to my
work now) Thanks.
Seems to me like lvm2 is just a pain in the ass for udev all around. In
systemd, it's the only reason that a call to udevadm settle is required
during late bootup. Everything else can be handled as hotplug events as
they come up.
Snippets from Kay...
11:52 kay » falconindy: only if you use --db-cleanup, and yes lvm
plays dirty tricks with udev and needs the db preserved
11:54 kay » falconindy: they need to catch up with 2011 a bit though
and stop scanning /dev, or have crazy regexes in their
config files
11:58 kay » falconindy: in the past we used to rm -rf /dev/.udev
in the real root
11:59 falconindy » which i assume is a more of a sledgehammer approach
to what the cleanup does
12:00 kay » falconindy: not that different, the db_persist just
adds a sticky bit to the db file
So LVM is just a pain in the butt all around. State can be wiped for all
other devices and the rebuilt DB in later userspace doesn't contain
failed/unhandled events from the initramfs.
dave
Thanks.
--
Gerardo Exequiel Pozzi
\cos^2\alpha + \sin^2\alpha = 1
