Documentation/QMP

QEMU Machine Protocol

The QEMU Machine Protocol (QMP) is a JSON-based protocol which allows applications to control a QEMU instance.

Features:

• Lightweight, text-based, easy to parse data format
• Asynchronous messages support (events)
• Capabilities negotiation
• API/ABI stability guarantees

Please, also check the README file.

Examples

The first example explains some important details about QMP. The others are simpler and run on top of the first one.

In all examples 'C' stands for 'Client' and 'S' stands for 'Server'.

Capabilities Negotiation

When a new QMP connection is established, QMP sends its greeting message and enters capabilities negotiation mode. In this mode, only the qmp_capabilities command works. To exit capabilities negotiation mode and enter command mode, the qmp_capabilities command must be issued:

 S: {
        "QMP": {
            "version": {
                "qemu": {
                    "micro": 0,
                    "minor": 6,
                    "major": 1
                },
                "package": ""
            },
            "capabilities": [
            ]
        }
    }
C: { "execute": "qmp_capabilities" }
S: { "return": {}}

The { "return": {} } response is QMP's success response. An error response will contain the "error" keyword instead of "return".

Eject a medium

C: { "execute": "eject", "arguments": { "device": "ide1-cd0" } }
S: { "return": {}}

Query VM status

C: { "execute": "query-status" }
S: {
       "return": {
           "status": "running",
           "singlestep": false,
           "running": true
       }
   }

Asynchronous message

S: { "event": "BLOCK_IO_ERROR",
     "data": { "device": "ide0-hd1",
               "operation": "write",
               "action": "stop" },
     "timestamp": { "seconds": 1265044230, "microseconds": 450486 } }

Development

Main developer in charge is Luiz Capitulino. All QMP-related discussions happen on the qemu-devel mailing list.

Luiz's QMP queue can be found at:

http://repo.or.cz/w/qemu/qmp-unstable.git/shortlog/refs/heads/queue/qmp

IMPORTANT: This branch is constantly rebased!

TODO

In order of importance.

Drop the qmp-commands.hx file

The qmp-commands.hx file is basically a command table. Today, with the advent of the QAPI, the qapi-schema.json file serves the same purpose. We should drop qmp-commands.hx in favor of qapi-schema.json.

This is more or less what it takes to do that:

1. Convert do_device_add() and do_qmp_capabilities() to propagate errors (ie, fill an Error ** object instead of doing qerror_report() calls)
2. Do something about qmp_netdev_add(), qmp_qom_get() and qmp_qom_set() (drop them or add QAPI wrappers?)
3. Change QMP code in monitor.c to use the QAPI to do command parameter validation and command dispatching (just like qemu-ga does)
4. Move QMP examples from qmp-commands.hx to qapi-schema.json
5. Drop qmp-commands.hx

Add full introspection support

This is about adding a command which returns all QMP types, commands (input and output parameters) and events.

Amos Kong is working on this:

http://marc.info/?l=qemu-devel&m=137059998004419&w=2

Add events support to the QAPI

Today events are exclusively part of the text protocol. This means that they are open coded, QEMU subsystems can't make use of them and they don't appear in qapi-schema.json.

We would solve all those problems by adding event support to the QAPI.

QMP testing

There should be a QMP test-suite. Either in QEMU itself or externally, like in kvm-autotest (which seems to be the best option).

HMP command conversion to the QAPI

Our long term plan is to make HMP, the human monitor, independent of QEMU. This means that HMP code shouldn't call QEMU code other than the monitor.

To achieve this we have to split existing HMP-only commands (eg. do_wav_capture() or do_ioport_read()) into a QMP part and an HMP part. Just like new QMP commands.

Testing

This section describes the following ways of testing QMP:

• By hand (cumbersome, only worth it if you're chasing a specific bug)
• qmp-shell script (automates part of the job)
• Libvirt integration (assumes familiarity with libvirt)
• libvirt-TCK
• kvm-autotest

Please, note that it's very recommended to read the README and spec files, as some of the testing procedures may require knowledge about the protocol format.

By hand

1. Start QMP on a TCP socket, so that telnet can be used

# qemu [...] -qmp tcp:localhost:4444,server

2. Run telnet

$ telnet localhost 4444

3. You should see QMP's greeting banner

{"QMP": {"version": {"qemu": {"micro": 50, "minor": 13, "major": 0}, "package": ""}, "capabilities": []}}

4. Issue the qmp_capabilities command, so that QMP enters command mode

{ "execute": "qmp_capabilities" }

5. You can now issue commands. For example, to get a list of QMP supported commands, issue query-commands

{ "execute": "query-commands" }

NOTE: all "info" commands are available under QMP as "query-", for example "info vnc" is "query-vnc"

There's an optimization to this procedure in case you plan to use it often (eg. QMP development):

1. Install programs socat and rlwrap. If you're running Fedora, you can do

# yum install socat rlwrap

2. Add the following sections to your QEMU config file (or create a qemu-qmp.conf one):

[chardev "qmp"]
  backend = "socket"
  path = "path to the QMP unix socket"
  server = "on"
  wait = "off"
[mon "qmp"]
  mode = "control"
  chardev = "qmp"

3. Run QEMU

# qemu [...] -readconfig qemu-qmp.conf

4. Run rlwrap

# rlwrap -C qmp socat STDIO UNIX:path-to-the-QMP-unix-socket

You can now issue commands, rlwrap will give you readline support (including persistent history).

qmp-shell script

This script is available under the QMP directory in QEMU's source-tree. It automates a bit the testing work, as it can construct commands objects for you.

1. Start QMP on a unix socket

# qemu [...] -qmp unix:./qmp-sock,server

2. Run the script

# qmp-shell ./qmp-sock

3. You should get the following prompt

(QEMU)

4. You can now issue commands. For example, let's add a new device

(QEMU) device_add driver=e1000 id=net1

Libvirt

Libvirt already got QMP support, but it's currently disabled. This test procedure explains how to enable it, so that you can have libvirt running on top of QMP.

1. Install yajl-devel (If you're running Fedora 12 or above just do 'yum install yajl-devel')

2. From a fresh checkout of libvirt master branch run the following:

./autogen.sh --system --enable-compile-warnings=error

NOTE (1): The '--system' flag is a shortcut for compiling with the --prefix and other directories matching a system RPM build.

NOTE (2): Make sure the final summary of autogen.sh tells you that it found the yajl library

3. To enable QMP support, edit src/qemu/qemu_conf.c and find:

#if 0
   if (version >= 13000)
       flags |= QEMUD_CMD_FLAG_MONITOR_JSON;
#endif

Change to '#if 1', and change the version to 12000 so it detects your GIT build of QEMU

4. Run 'make' to build. There is no need to 'make install' anything especially since that would overwrite your RPM based install

5. As root simply stop the current daemon & start the one you built

/etc/init.d/libvirtd stop
$HOME/your/git/checkout/of/libvirt/daemon/libvirtd

6. As root you can use the newly built virsh too

cd $HOME/your/git/checkout/of/libvirt/src
./virsh <BLAH>

Libvirt-TCK

TODO: describe libvirt's testing tool.

kvm-autotest

TODO: describe kvm-autotest QMP support.

Other information

• Luiz's QMP talk on KVM Forum 2010 can be found here
• Old QMP page can be accessed here