GNU bug report logs - #41143
[PATCH 1/2] Add 'lvm-device-mapping'

Previous Next

Package: guix-patches;

Reported by: tsmish <tsymsh <at> gmail.com>

Date: Sat, 9 May 2020 01:13:01 UTC

Severity: normal

Tags: patch

Done: Ludovic Courtès <ludo <at> gnu.org>

Bug is archived. No further changes may be made.

Full log


View this message in rfc822 format

From: Mikhail Tsykalov <tsymsh <at> gmail.com>
To: 41143 <at> debbugs.gnu.org
Subject: [bug#41143] Some clarification
Date: Fri, 15 May 2020 01:53:26 +0300
Hi,

In LVM one volume group usually haves more than one logical volume, so 
target need to be a list. Just setting target to list in config doesn't 
work because that field is actually used for checking when (and whether 
at all) the device needs to be mounted. Because we still need to support 
using a string as target, we need to handle two situations: target is a 
string and target is list of strings. Patch converts plain string target 
to single element list and proceeds with assuming a list. grep by 
mapped-device-target revealed two uses in gnu/system.scm and one in 
gnu/services/base.scm. First two checked if any filesystems used a 
device, so I replaced plain string comparison with comparison with every 
element of list and returning true if any of them matched. Last one used 
target for building service name for shepherd, so I replaced plain 
target with concatenation of every target that mapped device has.

Moving on to second patch. Basically to mount LVM you need to activate 
volume group(s) with command "vgchange -ay". It worked fine with 
non-root mounts, but when it got moved in initrd it crashed. Quick look 
around found that while /dev/dm-X devices were being created, 
/dev/mapper/ nodes were not. Looking at man pages (and internet) "vgscan 
--mknodes" looked like what I wanted (there is also dmsetup mknodes, but 
basically first just calls it). After adding it, boot with root on LVM 
started working, so I left it. Also I check that targets actually exist, 
so you get crash in a bit more appropriate place. This check should 
probably be moved to somewhere else, but I haven't found better solution.

Looking at 
https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/logical_volume_manager_administration/udev_device_manager 
10-dm.rules looks like what creates /dev/mapper nodes in normal systems, 
so probably more proper solution will be to add udevd to initrd, but I 
don't know how to run daemons in initrd properly. dracut also creates 
file /etc/lvm/lvm.conf, which doesn't seem to be very important, but I 
probably should look closer to it. Also both dracut and LVM add 
"--ignorelockingfailure" flag to vgchange, which I forgot and will 
change in future patches.

There seems to be close to no resources about mounting LVM in initrd, so 
I'm adding code from other initrd generators:

Dracut: 
https://github.com/dracutdevs/dracut/blob/master/modules.d/90lvm/lvm_scan.sh

LVM: 
https://github.com/lvmteam/lvm2/blob/master/scripts/lvm2create_initrd/lvm2create_initrd





This bug report was last modified 4 years and 176 days ago.

Previous Next


GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.