[udev] /dev/disk/by-path missing

[James Page]

Latest cloud image for Ubuntu Saucy; /dev/disk/by-path is missing:

ubuntu@juju-saucy-instance-0:~$ find /dev/disk
/dev/disk
/dev/disk/by-uuid
/dev/disk/by-uuid/80aed723-838e-44b9-90a9-f1b2c04f7c87
/dev/disk/by-label
/dev/disk/by-label/cloudimg-rootfs
/dev/disk/by-label/config-2
ubuntu@juju-saucy-instance-0:~$

I've not seen this in < Saucy and can repeat this reliably using the saucy daily images.

ProblemType: Bug
DistroRelease: Ubuntu 13.10
Package: udev 204-0ubuntu4
ProcVersionSignature: User Name 3.9.0-6.14-generic 3.9.6
Uname: Linux 3.9.0-6-generic x86_64
ApportVersion: 2.10.2-0ubuntu2
Architecture: amd64
CurrentDmesg: [ 6.952094] init: plymouth-stop pre-start process (1125) terminated with status 1
Date: Sat Jun 22 21:32:23 2013
Ec2AMI: ami-0000002a
Ec2AMIManifest: FIXME
Ec2AvailabilityZone: nova
Ec2InstanceType: m1.tiny
Ec2Kernel: aki-00000002
Ec2Ramdisk: ari-00000002
Lsusb:
 Bus 001 Device 002: ID 0627:0001 Adomax Technology Co., Ltd
 Bus 001 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
MachineType: OpenStack Foundation OpenStack Nova
MarkForUpload: True
ProcEnviron:
 TERM=screen
 PATH=(custom, no user)
 XDG_RUNTIME_DIR=<set>
 LANG=en_US.UTF-8
 SHELL=/bin/bash
ProcKernelCmdLine: BOOT_IMAGE=/boot/vmlinuz-3.9.0-6-generic root=LABEL=cloudimg-rootfs ro console=tty1 console=ttyS0
SourcePackage: systemd
UpgradeStatus: No upgrade log present (probably fresh install)
dmi.bios.date: 01/01/2011
dmi.bios.vendor: Bochs
dmi.bios.version: Bochs
dmi.chassis.type: 1
dmi.chassis.vendor: Bochs
dmi.modalias: dmi:bvnBochs:bvrBochs:bd01/01/2011:svnOpenStackFoundation:pnOpenStackNova:pvr2013.1.1:cvnBochs:ct1:cvr:
dmi.product.name: OpenStack Nova
dmi.product.version: 2013.1.1
dmi.sys.vendor: OpenStack Foundation

[Martin Pitt (pitti)]

I bisected this down to http://git.kernel.org/cgit/linux/hotplug/udev.git/commit/?id=481dcf7c8f; reverting it makes the by-path/ reappear, but disabling it for ATA devices (which only have a SCSI driver, but are not physically SCSI) was quite deliberate. Is this causing actual problems, or did you just happen to notice it?

[James Page (james-page)]

Martin

I picked it up because it causes problems with the tooling that ceph uses to prepare disks for use in a ceph cluster.

The tools use the /dev/disk/by-path tree to detect whether specific devices are already in use or not; so it would be nice if it could be restored!

[Martin Pitt (pitti)]

> The tools use the /dev/disk/by-path tree to detect whether specific devices are already in use or not;

The by-path for ATA devices was removed because it was possible for the same device to have multiple different by-path names; what does "in use" mean in that context?

[James Page (james-page)]

Easiest to refer to the code that uses this:

List all devices and partitions:
http://bazaar.launchpad.net/~ubuntu-branches/ubuntu/saucy/ceph/saucy/view/head:/src/ceph-disk#L179

Check whether a given device is a partition or a full disk:
http://bazaar.launchpad.net/~ubuntu-branches/ubuntu/saucy/ceph/saucy/view/head:/src/ceph-disk#L222

[James Page (james-page)]

Raising a task for ceph as well as this is the key impact I'm currently aware of.

[Martin Pitt (pitti)]

While we can technically put /dev/disks/by-path back for ATA devices in Ubuntu, that still won't help upstream ceph on other distros, and I believe this approach is buggy anyway. is_partition() currently looks a bit underdefined/buggy to me, as it's not uncommon to have file systems on unpartitioned block devices (i. e. mounting /dev/sdb directly); this will also completely miss devmapper or other "virtual" devices. To mimic the current behaviou, one could do something like

  return not os.path.exists('/sys/block/%s' % os.path.basename(dev))

as devices will be there (like /sys/block/sdb), and partitions under that (like /sys/block/sdb/sdb2). But again, that's not really what you want there: Judging by how is_partition() is being used in verify_not_in_use(), it should probably be called is_filesystem() and query udev (CLI example: udevadm info --name=/dev/sda2 | grep ID_FS) or blkid.

There are a couple of options how to rewrite these functions, including reading /sys/block/, /proc/partitions, asking blkid, querying udevadm, or querying udisks.

If you want a list of block devices with mountable file systems which are present, and don't want to use any extra libraries, then perhaps the simplest thing is

  udevadm trigger --dry-run --verbose --property-match=ID_FS_USAGE=filesystem

(this doesn't need any root privileges). If you want all block devices, then

  udevadm trigger --dry-run --verbose --subsystem-match=block

will give a list, including device-mapper (and also loopback, etc.).

[James Page (james-page)]

Hi Martin

Thanks for the detailed response on different ways to tackle this issue; I'll review and see what we can do with upstream to resolve this.

[James Page (james-page)]

This has been resolved in the latest upstream releases.