Bug#859156: multipath-tools: boot ends in rescue mode - boot partition marked as systemd_ready=0

[Alban Browaeys]

Le samedi 01 avril 2017 à 14:24 +0530, Ritesh Raj Sarraf a écrit :
> On Fri, 2017-03-31 at 16:29 +0200, Alban Browaeys wrote:
> > The bug is that all partition are tagged SYSTEMD_READY=0 which
> > prevents
> > systemd from mounting the bare ones. That is the dm, bcache and
> > raid ones are
> > fine. 
> > But I cannot move /boot/efi to lvm thus I cannot fix boot out of
> > modifying
> > multipath configuration.
> > 
> > This since the introduction of 60-multipath.rules the 25th of
> > match.
> > Mind this might be a local issue as I do not know if
> > /etc/multipath/wwids is
> > supposed to list all disks as it was here. Emptying it fixes boot.
> > 
> > If I remove the multipath udev rules that tags all my partitions as
> > SYSTEMD_READY=0 systemd manage to boot.
> 
> With the current information that you have provided, I am not able to
> determine
> the bug. I tried in my local multipath setup and everything is
> working fine.
> 
> If you can help give me clean steps to reproduce in a plain multiapth
> setup,
> that'll help.
> 

I guess the issue is a mix of this old behaviour : 
https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/6/html-single/DM_Multipath/
In older releases of Red Hat Enterprise Linux, multipath always tried
to create a multipath device for every path that was not explicitly
blacklisted. In Red Hat Enterprise Linux 6, however, if the
find_multipaths configuration parameter is set to yes, then multipath
will create a device only if one of three conditions are met 
(...)

[ Note that find_multipaths is set to no here.]

and the 60-multipath.rules in March 2017 which ends up tagging all the
partitions with SYSTEMD_RUN=O. 
Which is fine if all of them are  dm ones (as they are later tagged
back to SYSTEMD_RUN=1 , but not so fine for "bare metal partitions"
that are left so.

To reproduce I add my disk wwid (here <n> is the disk serial) to
/etc/multipath/wwids 
/WDC_WD10SPCX-08S8TT0_WD-<n>/
and with sda7 a "bare metal" partition on this disk (here /boot):
sudo sh -c "env - SEQNUM="1" ACTION="add" $(udevadm info  -x /dev/sda7|sed -n '/^E:/ {   s at E: \(.*\)=\(.*\)@\1="\2"@p  }' | tr '\n' ' ' )  /sbin/multipath -u /dev/sda7; ";echo $?
I get:
/dev/sda7 is a valid multipath device path
0
thus this "non multipath, non dm" partition will be tagged not ready to systemd.
(this also mean this requires a non dm partition).

If /etc/mulitpath/wwids is missing or left empty, obviously the bug is not reproducible.
I need guidance as to what creates or was creating this file at one point in time.
I tried removing it, then flusing, no luck so far.
Otherwise I would say let s close this bug :)


Thanks
Alban

PS: multipath was never setup on this box. The package was installed only (this is a laptop).
This install dates back to at least 0.5.0+git1.656f8865-8 (from apt history).