bcache-super-show segfaults reading superblock on dasd

[Ryan Harper]

1. # lsb_release -rd
Description: Ubuntu Disco Dingo (development branch)
Release: 19.04

2. # apt-cache policy bcache-tools
bcache-tools:
  Installed: 1.0.8-3
  Candidate: 1.0.8-3
  Version table:
 *** 1.0.8-3 500
        500 http://us.ports.ubuntu.com/ubuntu-ports disco/main s390x Packages
        100 /var/lib/dpkg/status

3. bcache-super-show /dev/dasda

Should return the superblock information on the device.

4. segfaults.

Reproduce with:

$ ls -al /dev/disk/by-path/ccw-0.0.1520
lrwxrwxrwx 1 root root 11 Mar 22 08:34 /dev/disk/by-path/ccw-0.0.1520 -> ../../dasda
$ make-bcache --writeback --cache /dev/disk/by-path/ccw-0.0.1520
$ bcache-super-show /dev/disk/by-path/ccw-0.0.1520

ProblemType: Bug
DistroRelease: Ubuntu 19.04
Package: bcache-tools 1.0.8-3
ProcVersionSignature: Ubuntu 4.18.0-15.16-generic 4.18.20
Uname: Linux 4.18.0-15-generic s390x
NonfreeKernelModules: zfs zunicode zavl icp zcommon znvpair
ApportVersion: 2.20.10-0ubuntu21
Architecture: s390x
Date: Fri Mar 22 08:38:06 2019
ProcEnviron:
 TERM=xterm-256color
 PATH=(custom, no user)
 LANG=en_US.UTF-8
 SHELL=/bin/bash
SourcePackage: bcache-tools
UpgradeStatus: Upgraded to disco on 2019-02-18 (31 days ago)

[Dimitri John Ledkov (xnox)]

Requesting reverse proxy to ibm for attention

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2019-03-25 04:49 EDT-------
Description will be reversed mirrored from LP

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2019-03-25 05:32 EDT-------
This error is likely due to using the full block device node (dasda) instead of a partition device node (e.g. dasda1) of a CDL formatted DASD device.

Note that the first two tracks (1 track = 48k at 4k block size) of a CDL-formatted DASD device have special properties. Not all data written to these blocks is retained. Instead portions of the data are replaced by an E5 byte pattern. Applications/FS should therefore not use the block device, but only partition devices.

[Andreas Hasenack (ahasenack)]

Can you attach a core dump?

[Dimitri John Ledkov (xnox)]

@Peter
I guess the question is, whether or not, we should be trying to use dasd as a bcache device at all. Or like create a partition table, and use the partition as a bcache device then? In that case should the make-bcache fail on the full dasd device, as shows in the reproducer steps?

If I understand correctly where the request is coming from, MAAS/NewServerInstaller use curtin which supports bcache setup, and that fails with dasd drives at the moment in curtin's implementation / integration test suite.

[Ryan Harper (raharper)]

There are no restrictions on non-dasd devices so we can update curtin to ensure that we don't allow creating bcache devices on dasd without a partition.

I'm not sure whether or not it's worth adding dasd specific checks to bcache-tools to refuse when supplied with a dasd block device.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2019-04-02 04:03 EDT-------
(In reply to comment #12)
> @Peter
> I guess the question is, whether or not, we should be trying to use dasd as
> a bcache device at all.

In my opinion completely preventing the use of DASD for bcache would be overly restrictive.

> Or like create a partition table, and use the
> partition as a bcache device then? In that case should the make-bcache fail
> on the full dasd device, as shows in the reproducer steps?

Creating a partition and using the partition as bcache device is the recommended approach. I don't think that workarounds for DASD particularities would fit into bcache-tools though.

> If I understand correctly where the request is coming from,
> MAAS/NewServerInstaller use curtin which supports bcache setup, and that
> fails with dasd drives at the moment in curtin's implementation /
> integration test suite.

I'm not familiar with curtin, but if that's a layer on top that makes use of bcache-tools, then that may be a good place to catch/prevent the attempt to run bcache on the DASD block device and to suggest using a partition instead.

[Ryan Harper (raharper)]

Marking this invalid since DASDs can't be used directly without a partition table.

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2019-05-29 03:46 EDT-------
IBM bugzilla status -> closed, set to Invalid by Canonical