Re: What fields should be used for reporting shared memory?

[Martin Kletzander <mkletzan@redhat.com>]

[-- Attachment #1.1: Type: text/plain, Size: 4444 bytes --]

On Mon, Mar 20, 2017 at 06:21:53PM -0400, Steve Grubb wrote:
>On Monday, March 20, 2017 7:36:14 AM EDT Martin Kletzander wrote:
>> On Thu, Mar 16, 2017 at 09:04:52PM -0400, Steve Grubb wrote:
>> >Hello,
>> >
>> >I apologize for the delay.
>> >
>> >On Tuesday, March 14, 2017 7:42:27 AM EDT Martin Kletzander wrote:
>> >> I am going through the fields in the dictionary and I can't find any
>> >> name to use for the following scenario.
>> >>
>> >> We (libvirt) are running virtual machines and there's a thing nowadays,
>> >> that people like to use, called ivshmem (Inter-VM SHared MEMory).  From
>> >> host's point of view this is just a shared memory region accessed by
>> >> multiple VMs (and possibly to host as well).  The machine maps the
>> >> shared memory given a name (e.g. name "asdf" results in /dev/shm/asdf to
>> >> be mapped) *or* it can communicate with a server over UNIX socket and
>> >> that server handles interrupts and also tells the client which shared
>> >> memory region to map.
>> >
>> >If both of these result in a path, then I think we want to log it as a
>> >resource event.
>>
>> Yes, and they both are resources in its sense. So you are talking
>> particularly about the resrc= field?  Should that also have category and
>> class or anything else set?  Or do you mean we report the path in the
>> resrc= field?
>
>No, I mean we would want the path to the memory in a VIRT_RESOURCE event. :-)
>How about something like this:
>
>type=VIRT_RESOURCE msg=audit(1488441043.591:2977): pid=25464 uid=0
>auid=4294967295 ses=4294967295 subj=system_u:system_r:virtd_t:s0-s0:c0.c1023
>msg='virt=kvm resrc=shmem reason=start vm="rhel7.3" uuid=a7708061-
>faa0-42ce-897a-e92fb75fcf1d size=4194304 path="/dev/shm/my_shmem1" exe="/usr/
>sbin/libvirtd" hostname=? addr=? terminal=? res=success'
>
>Having the full path would make it more normal when reporting on files in use.
>
>
>> >> Talking about information we have; in server-less
>> >> setup it's the shared memory region that is shared, in the server
>> >> scenario it is the socket.  That's information we can output.
>> >
>> >Above you mentioned that the server communicates which region to map. Can
>> >you explain what that means?
>>
>> The server sends a file descriptor to the VM over the socket, details
>> can be found here:
>>
>> http://git.qemu-project.org/?p=qemu.git;a=blob;f=docs/specs/ivshmem-spec.txt
>> #l151
>
>This one is harder. What we really want to know is the size information. But
>libvirtd doesn't have it I assume. That makes it sound like ivshmem should log
>it, but it probably doesn't know the reason, vm name, or uuid or anything
>else. So, this is probably not good.
>
>I guess about all that you can say is that a ivshmem-socket is being assigned.
>I guess let's do it like this:
>
>type=VIRT_RESOURCE msg=audit(1488441043.591:2977): pid=25464 uid=0
>auid=4294967295 ses=4294967295 subj=system_u:system_r:virtd_t:s0-s0:c0.c1023
>msg='virt=kvm resrc=ivshmem-socket reason=start vm="rhel7.3" uuid=a7708061-
>faa0-42ce-897a-e92fb75fcf1d ivshmem="my_shmem2" exe="/usr/sbin/libvirtd"
>hostname=? addr=? terminal=? res=success'
>

Sorry for resurrecting this old thread.  I was under the impression that
I replied to you and no matter how much I look I can't find the reply.

I agree with everything with one exception.  I would change the
ivshmem="my_shmem2" to path="/path/to/ivshmem.socket".  That will also
clearly show what VMs share the shm objects.

If that's OK with you, I can propose the patch for libvirt in a little while.

>-Steve
>
>> >> So my question is, when starting a domain or hot-(un)plugging, what
>> >> naming should we use for this kind of device and what are the things
>> >> that we should describe about it?  Basically, how would you like the
>> >> message to look?
>> >
>> >We need a record recording what is getting assigned to the VM. In the case
>> >of the /dev/shm, you can record that as a path which must be escaped. In
>> >the case of the server, I think we still need to understand what is
>> >happening. Just recording a socket number or path is not terribly useful
>> >in reconstructing the resources given to the VM.
>> >
>> >Audit events have to tell a story. There is a subect, object, action, and
>> >results. It kind of needs to be a sentence. "libvirtd successfully assigned
>> >____ to vm-name."
>> >
>> >-Steve
>> >
>> >> Thanks in advance for any info.
>> >>
>> >> Have a nice day,
>> >> Martin
>
>

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]