'call to get_tasks_recursive failed' errors from su

Status changed to 'Confirmed' because the bug affects multiple users.

I'm actually getting the same on xenial.

I see the same on 14.04, even without "-" on the su.
The test also lacked the -c for the command, and "-" would need to be at the end or specified as "-l".
Also we might execute something that doesn't add output, to easen debugging (/bin/true)

Anyway - that just refines the simple test, the issue is there as reported.

Simplified test:
su testuser -c "/bin/true"

The only reference I found to get_tasks_recursive seems far away in cgmanager (http://changelogs.ubuntu.com/changelogs/pool/main/c/cgmanager/cgmanager_0.32-4ubuntu1.1/changelog and https://linuxcontainers.org/it/cgmanager/manpages/man1/cgm.1.html)

Yet I found that my Xenial Test system didn't show the issue, while Michael before mentioned it did for him.
I realized that this Xenial System had no cgmanager installed.

So I installed full lxd (which is what usually brings cgmanager to your system) and e voila I was able to reproduce the issue on Xenial as well now.

The output is slightly different (more and repeated on Xenial)
Trusty
call to get_tasks_recursive failed: Invalid arguments received in reply

Xenial
call to remove-on-empty (freezer:0) failed: invalid request
call to get_tasks_recursive failed: Invalid arguments received in reply
call to get_tasks_recursive failed: Invalid arguments received in reply

I removed lxd and all its dependencies again and only installed cgmanager on the Xenial system. This pulls in libcgmanager0 and libnih-dbus1 as well, but nothing else.
With those installed the issue does not yet appear, so it seems to need more of these packages to actually trigger.
I tried to separate the lxd dependencies and found that libpam-cgm, likely to hook from pam (which is what su uses) into cgmanager.

cgmanager is a service you can reach via dbus for cgroup management.
It is used in this case via libpam-cgm to "Create cgroups for user login sessions"

I stopped the service and reran it in verbose mode
sudo /sbin/cgmanager -m name=systemd -v

Then in another console I tried to trigger the issue again with
su testuser -c "/bin/true"

On the triggering task I still get:
su -m testuser -c "/bin/true"
call to remove-on-empty (freezer:1) failed: invalid request
call to get_tasks_recursive failed: Invalid arguments received in reply

On the verbose log I get:
Connection from private client
ListControllers: Client fd is: 6 (pid=21503, uid=0, gid=1000)
MovePid: Client fd is: 6 (pid=21503, uid=0, gid=1000)
21503 moved to freezer:/ by 21503's request
Create: Client fd is: 6 (pid=21503, uid=0, gid=1000)
Created /run/cgmanager/fs/freezer/user/paelzer for 21503 (0:1000)
cgmanager_create: returning 0; existed is 1
MovePid: Client fd is: 6 (pid=21503, uid=0, gid=1000)
21503 moved to freezer:user/paelzer by 21503's request
Create: Client fd is: 6 (pid=21503, uid=0, gid=1000)
Created /run/cgmanager/fs/freezer/user/paelzer/0 for 21503 (0:1000)
cgmanager_create: returning 0; existed is 1
Create: Client fd is: 6 (pid=21503, uid=0, gid=1000)
Created /run/cgmanager/fs/freezer/user/paelzer/1 for 21503 (0:1000)
cgmanager_create: returning 0; existed is -1
Chown: Client fd is: 6 (pid=21503, uid=0, gid=1000)
RemoveOnEmpty: Client fd is: 6 (pid=21503, uid=0, gid=1000)
cgmanager: remove-on-empty request for pre-mounted controller
MovePid: Client fd is: 6 (pid=21503, uid=0, gid=1000)
21503 moved to freezer:1 by 21503's request
Disconnected from private client
Connection from private client
ListControllers: Client fd is: 6 (pid=21503, uid=0, gid=1000)
MovePid: Client fd is: 6 (pid=21503, uid=0, gid=1000)
21503 moved to freezer:/ by 21503's request
ListChildren: Client fd is: 6 (pid=21503, uid=0, gid=1000)
GetTasksRecursive: Client fd is: 6 (pid=21503, uid=0, gid=1000)
GetTasksRecursive: Client fd is: 6 (pid=21503, uid=0, gid=1000)
Remove: Client fd is: 6 (pid=21503, uid=0, gid=1000)
Removed /run/cgmanager/fs/freezer/user/paelzer/1 for 21503 (0:1000)
GetTasksRecursive: Client fd is: 6 (pid=21503, uid=0, gid=1000)
Disconnected from private client

I can use cgm to do some getrecursive calls without issue, like this:
cgm gettasksrecursive freezer

With that I think my quick debugging has reached its end, but lets assign to the proper components now that we know.

@tkedwards,

Could you please show your cgmanager version?

dpkg -l | grep cgmanager

Confirmed it happens for me with the lxd-git-master ppa.

I'll remove all those error messages printed to stderr in pam/cgmanager.c. If cgmanager
were going to last I would take the time to pass the errors back up to log them to syslog,
but it's just not worth it.

This bug was fixed in the package cgmanager - 0.39-2ubuntu5

---------------
cgmanager (0.39-2ubuntu5) xenial; urgency=medium

  * Update previous patch to remove all the (non-initial) errors printed to
    stderr. (LP: #1534090)

 -- Serge Hallyn <email address hidden> Mon, 18 Jan 2016 10:50:00 -0800

Thanks! Would it also be possible to fix in trusty and wily?

Trusty doesn't have libpam-cgm. Do you mean in ppas?

Yeah, sorry meant in the ppa. I use it on both trusty and wily: deb http://ppa.launchpad.net/ubuntu-l
xc/lxd-stable/ubuntu wily main

Yup that'll get the update.

Thanks!

Status changed to 'Confirmed' because the bug affects multiple users.

Actually libpam-cgm is in trusty-backports now. Can we have the fix backported for this package too?

Ye,s we should SRU to that, thanks.