Re: [Buildroot] lvm2 seems to require systemd

[Andreas Ziegler <br015@umbiko.net>]

Hi Frank,

On 2024-07-21 19:37, Frank Wunderlich wrote:
> Hi
> 
>> Gesendet: Sonntag, 21. Juli 2024 um 19:58 Uhr
>> Von: "Andreas Ziegler" <br015@umbiko.net>
>> An: "Frank Wunderlich" <frank-w@public-files.de>
>> Cc: "buildroot" <buildroot@buildroot.org>
>> Betreff: Re: Aw: Re:  Re: lvm2 seems to require systemd
>> 
>> Hi Frank,
>> 
>> On 2024-07-21 15:50, Frank Wunderlich wrote:
>> > Hi Andreas
>> >
>> > noticed some errors when calling the vgchange command
>> >
>> > # /usr/sbin/lvm vgchange -aay --autoactivation event vg-nvme
>> >   1 logical volume(s) in volume group "vg-nvme" now active
>> > [ 8023.796866] udevd[1614]: failed to execute '/usr/sbin/blkid'
>> > '/usr/sbin/blkid -o udev -p /dev/dm-0': No such file or directory
>> > [ 8023.808977] udevd[1610]: conflicting device node
>> > '/dev/mapper/vg--nvme-var' found, link to '/dev/dm-0' will not be
>> > created
>> > # lvscan
>> >   ACTIVE            '/dev/vg-nvme/var' [10.00 GiB] inherit
>> 
>> Your execution of vgchange triggers additional udev messages, that are
>> not handled correctly. Try to find out what is missing and add it to
>> your image; blkid, for example, is part of package util-linux.
> 
> i have already enabled util_linux
> Symbol: BR2_PACKAGE_UTIL_LINUX [=y]
> BR2_PACKAGE_HOST_UTIL_LINUX [=y]
> BR2_PACKAGE_UTIL_LINUX_LIBBLKID [=y]
> 
> so strange that blkid is not there, but it is in /sbin
> 
> # which blkid
> /sbin/blkid
> 
> maybe a symlink needs to be created?
> 

Sorry, my mistake: blkid is provided by busybox and busybox installs it 
in /sbin. Since this is a Buildroot feature (unless you activate 
BR2_ROOTFS_MERGED_USER), I would modify the udev rule.

>> The double-dash in vg--nvme-var looks strange: is this intentional?
> 
> as you can see above i used only one dash in vg-nvme, not sure why 
> another one was inserted here
> 
> # pvscan
>   PV /dev/nvme0n1p5   VG vg-nvme   lvm2 [<1.79 TiB / <1.78 TiB free]
>   Total: 1 [<1.79 TiB] / in use: 1 [<1.79 TiB] / in no VG: 0 [0   ]
> 
>> > but i'm able to mount it
>> >
>> > # mkdir /mnt/var
>> > # mount /dev/mapper/vg--nvme-var /mnt/var/
>> > [ 8514.687803] EXT4-fs (dm-0): mounted filesystem
>> > be8b31d6-96e2-43a1-8416-a058972118b3 r/w with ordered data mode. Quota
>> > mode: disabled.
>> > # ls /mnt/var
>> > cache       lib         log         lost+found  run         tmp
>> > #
>> >
>> > how can i test the script? currently i'm in the initrd, so changing
>> > anything will be lost on reboot, so i want to test the script before
>> > putting it into the initrd.
>> 
>> I usually mount the target's file system via ssh://root@hostname in a
>> file browser; that allows me to copy to and from the target machine.
>> Script and configuration changes are made directly on the target 
>> system,
>> and copied back to the development machine when everything runs
>> satisfactorily.
>> 
>> > as far as i see the original call via udev contains some
>> > systemd-specific params and environment-vars i do not know currently
>> 
>> They are probably safe to ignore; when in doubt, consult the systemd
>> documentation. Evironment variables are extensively used within the 
>> udev
>> context; I doubt that the called lvm2 tool makes use of them.
> 
> they seem to be systemd options which are not stripped by the script 
> yet which
> will result in invalid call (imho only first param is shifted away)
> 

RUN+="/usr/bin/systemd-run \
   --no-block \
   --property DefaultDependencies=no \
   --unit lvm-activate-$env{LVM_VG_NAME_COMPLETE} \
   /usr/sbin/lvm \
   vgchange \
   -aay \
   -autoactivation \
   event $env{LVM_VG_NAME_COMPLETE}"

If you put a 'shift 5' at the beginning of the script, it should get rid 
of the systemd parameters. Alternatively, you could modify the udev 
rule.

Kind regards,
Andreas

>> Kind regards,
>> Andreas
>> 
>> >
>> > regards Frank
>> >
>> >
>> >> Gesendet: Sonntag, 21. Juli 2024 um 12:16 Uhr
>> >> Von: "Andreas Ziegler" <br015@umbiko.net>
>> >> An: "Frank Wunderlich" <frank-w@public-files.de>
>> >> Cc: "buildroot" <buildroot@buildroot.org>
>> >> Betreff: Re: Aw: Re: lvm2 seems to require systemd
>> >>
>> >> Hi Frank,
>> >>
>> >> On 2024-07-21 09:54, Frank Wunderlich wrote:
>> >> > regards Frank
>> >> >
>> >> >
>> >> >> Gesendet: Sonntag, 21. Juli 2024 um 11:17 Uhr
>> >> >> Von: "Andreas Ziegler" <br015@umbiko.net>
>> >> >> An: frank-w@public-files.de
>> >> >> Cc: "buildroot" <buildroot@buildroot.org>
>> >> >> Betreff: Re: lvm2 seems to require systemd
>> >> >>
>> >> >> Hi Frank,
>> >> >>
>> >> >> On 2024-07-20 14:34, Frank Wunderlich wrote:
>> >> >> > Hi,
>> >> >> >
>> >> >> > i try to build a buildroot (2024.05) initrd with lvm2 support
>> >> >> >
>> >> >> > so basicly added these options:
>> >> >> >
>> >> >> > #to select BR2_PACKAGE_HAS_UDEV
>> >> >> > BR2_PACKAGE_EUDEV=y
>> >> >> > BR2_ROOTFS_DEVICE_CREATION_DYNAMIC_EUDEV=y
>> >> >> > BR2_PACKAGE_LIBBLOCKDEV=y
>> >> >> > BR2_PACKAGE_LVM2=y
>> >> >> >
>> >> >> > build is fine, but i see this while bootup:
>> >> >> >
>> >> >> > [    3.665010] udevd[1520]: failed to execute '/usr/bin/systemd-run'
>> >> >> > '/usr/bin/systemd-run --no-block --property DefaultDependencies=
>> >> >> > no --unit lvm-activate-vg-nvme /usr/sbin/lvm vgchange -aay
>> >> >> > --autoactivation event vg-nvme': No such file or directory
>> >> >> > done
>> >> >>
>> >> >> lvm2 installs a bunch of udev rules; one of them relies on systemd-run
>> >> >> to execute a command in the background:
>> >> >>
>> >> >> # pvscan will check if this device completes a VG,
>> >> >> # i.e. all PVs in the VG are now present with the
>> >> >> # arrival of this PV.  If so, it prints to stdout:
>> >> >> # LVM_VG_NAME_COMPLETE='foo'
>> >> >> #
>> >> >> # When the VG is complete it can be activated, so
>> >> >> # vgchange -aay <vgname> is run.  It is run via
>> >> >> # systemd since it can take longer to run than
>> >> >> # udev wants to block when processing rules.
>> >> >> # (if there are hundreds of LVs to activate,
>> >> >> # the vgchange can take many seconds.)
>> >> >>
>> >> >> ...
>> >> >>
>> >> >> IMPORT{program}="/usr/sbin/lvm pvscan --cache --listvg --checkcomplete
>> >> >> --vgonline --autoactivation event --udevoutput --journal=output
>> >> >> $env{DEVNAME}"
>> >> >> ENV{LVM_VG_NAME_COMPLETE}=="?*", RUN+="/usr/bin/systemd-run --no-block
>> >> >> --property DefaultDependencies=no --unit
>> >> >> lvm-activate-$env{LVM_VG_NAME_COMPLETE} /usr/sbin/lvm vgchange -aay
>> >> >> --autoactivation event $env{LVM_VG_NAME_COMPLETE}"
>> >> >> GOTO="lvm_end"
>> >> >>
>> >> >> > i see the physical volume, but logical volumes are disabled
>> >> >> >
>> >> >> > # pvscan
>> >> >> >   PV /dev/nvme0n1p5   VG vg-nvme   lvm2 [<1.79 TiB / <1.78 TiB free]
>> >> >> >   Total: 1 [<1.79 TiB] / in use: 1 [<1.79 TiB] / in no VG: 0 [0   ]
>> >> >> > # vgscan
>> >> >> >   Found volume group "vg-nvme" using metadata type lvm2
>> >> >> > # lvscan
>> >> >> >   inactive          '/dev/vg-nvme/var' [10.00 GiB] inherit
>> >> >> >
>> >> >> > and so i cannot mount the logical volume
>> >> >> >
>> >> >> > # mount /dev/vg-nvme/var /mnt
>> >> >> > [  281.006375] /dev/vg-nvme/var: Can't lookup blockdev
>> >> >> > [  281.011279] /dev/vg-nvme/var: Can't lookup blockdev
>> >> >> > [  281.016202] /dev/vg-nvme/var: Can't lookup blockdev
>> >> >> > [  281.021080] /dev/vg-nvme/var: Can't lookup blockdev
>> >> >> > mount: mounting /dev/vg-nvme/var on /mnt failed: No such file or
>> >> >> > directory
>> >> >> >
>> >> >> > any ideas?
>> >> >>
>> >> >> I would try to run vgchange manually to see if the setup works at all,
>> >> >> then create a /usr/bin/systemd-run script that is able to execute
>> >> >> commands in the background. Something like this (untested):
>> >> >
>> >> > i was able to manually initialize the lvm value after adding the
>> >> > systemd without init (running the command without prefix
>> >> > "systemd-run")...
>> >> >
>> >> > basicly with this command:
>> >> >
>> >> > /usr/sbin/lvm vgchange -aay --autoactivation event vg-nvme
>> >> >
>> >> > with init i had compile issues missing libmount, see my followup post
>> >> >
>> >> > now i have to remove systems from my initrd somehow to do further
>> >> > testing with only the eudev. any idea doing this without making all
>> >> > clear (or is this dropped when recreating the cpio file)?
>> >>
>> >> Without running make clean, unwanted files need to be removed
>> >> manually.
>> >> Buildroot creates per-package logs .files-list,
>> >> .files-list-{host,images,staging}, which contain the names of files
>> >> copied to the respective destination. Remove at least those from
>> >> .files-list to revert to a system without the package; preferrably
>> >> also
>> >> those from .files-list-staging.
>> >>
>> >> Kind regards,
>> >> Andreas
>> >>
>> >> >
>> >> >>    #!/bin/sh
>> >> >>    prg=$1
>> >> >>    shift
>> >> >>    (
>> >> >>      $prg $@ < /dev/zero > /tmp/error.log 2>&1
>> >> >>    )&
>> >> >
>> >> >> Kind regards,
>> >> >> Andreas
>> >> >>
>> >> >> > regards Frank
>> >>
>> 
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot