cloud-hypervisor/docs/api.md
Dean Sheather bb2139a408 vmm/api: Add vm.add-fs route
Currently unimplemented. Once implemented, this API will allow for
creating virtio-fs devices in the VM after it has booted.

Signed-off-by: Dean Sheather <dean@coder.com>
2020-04-20 20:36:26 +02:00

17 KiB

Cloud Hypervisor API

The Cloud Hypervisor API is made of 2 distinct interfaces:

  1. The external API. This is the user facing API. Users and operators can control and manage Cloud Hypervisor through either a REST API or a Command Line Interface (CLI).
  2. The internal API, based on rust's Multi-Producer, Single-Consumer (MPSC) module. This API is used internally by the Cloud Hypervisor threads to communicate between each others.

The goal of this document is to describe the Cloud Hypervisor API as a whole, and to outline how the internal and external APIs are architecturally related.

External API

REST API

The Cloud Hypervisor REST API triggers VM and VMM specific actions, and as such it is designed as a collection of RPC-style, static methods.

The API is OpenAPI 3.0 compliant. Please consult the Cloud Hypervisor API document for more details about the API payloads and responses.

Location and availability

The REST API is available as soon as the Cloud Hypervisor binary is started, through a local UNIX socket. By default, it is located at /run/user/{user ID}/cloud-hypervisor.{Cloud Hypervisor PID}. For example, if you launched Cloud Hypervisor as user ID 1000 and its PID is 123456, the Cloud Hypervisor REST API will be available at /run/user/1000/cloud-hypervisor.123456.

The REST API default URL can be overridden through the Cloud Hypervisor option --api-socket:

$ ./target/debug/cloud-hypervisor --api-socket /tmp/cloud-hypervisor.sock
Cloud Hypervisor Guest
    API server: /tmp/cloud-hypervisor.sock
    vCPUs: 1
    Memory: 512 MB
    Kernel: None
    Kernel cmdline:
    Disk(s): None

Endpoints

The Cloud Hypervisor API exposes the following actions through its endpoints:

Virtual Machine Manager (VMM) Actions

Action Endpoint Request Body Response Body Prerequisites
Check for the REST API availability /vmm.ping N/A /schemas/VmmPingResponse N/A
Shut the VMM down /vmm.shutdown N/A N/A The VMM is running

Virtual Machine (VM) Actions

Action Endpoint Request Body Response Body Prerequisites
Create the VM /vm.create /schemas/VmConfig N/A The VM is not created yet
Delete the VM /vm.delete N/A N/A The VM is created but not booted
Boot the VM /vm.boot N/A N/A The VM is created
Shut the VM down /vm.shutdown N/A N/A The VM is booted
Reboot the VM /vm.reboot N/A N/A The VM is booted
Pause the VM /vm.pause N/A N/A The VM is booted
Resume the VM /vm.resume N/A N/A The VM is paused
Add/remove CPUs to/from the VM /vm.resize /schemas/VmResize N/A The VM is booted
Remove memory from the VM /vm.resize /schemas/VmResize N/A The VM is booted
Dump the VM information /vm.info N/A /schemas/VmInfo The VM is created
Add VFIO PCI device to the VM /vm.add-device /schemas/VmAddDevice N/A The VM is booted
Remove VFIO PCI device from the VM /vm.remove-device /schemas/VmRemoveDevice N/A The VM is booted
Add disk device to the VM /vm.add-disk /schemas/DiskConfig N/A The VM is booted
Add fs device to the VM /vm.add-fs /schemas/FsConfig N/A The VM is booted
Add pmem device to the VM /vm.add-pmem /schemas/PmemConfig N/A The VM is booted
Add network device to the VM /vm.add-net /schemas/NetConfig N/A The VM is booted

REST API Examples

For the following set of examples, we assume Cloud Hypervisor is started with the REST API available at /tmp/cloud-hypervisor.sock:

$ ./target/debug/cloud-hypervisor --api-socket /tmp/cloud-hypervisor.sock
Cloud Hypervisor Guest
    API server: /tmp/cloud-hypervisor.sock
    vCPUs: 1
    Memory: 512 MB
    Kernel: None
    Kernel cmdline:
    Disk(s): None

Create a Virtual Machine

We want to create a virtual machine with the following characteristics:

  • 4 vCPUs
  • 1 GB of RAM
  • 1 virtio based networking interface
  • Direct kernel boot from a custom 5.5.0 Linux kernel located at /opt/clh/kernel/vmlinux-virtio-fs-virtio-iommu
  • Using a Clear Linux image as its root filesystem, located at /opt/clh/images/clear-30080-kvm.img
#!/bin/bash

curl --unix-socket /tmp/cloud-hypervisor.sock -i \
     -X PUT 'http://localhost/api/v1/vm.create'  \
     -H 'Accept: application/json'               \
     -H 'Content-Type: application/json'         \
     -d '{
         "cpus":{"boot_vcpus": 4, "max_vcpus": 4},
         "kernel":{"path":"/opt/clh/kernel/vmlinux-virtio-fs-virtio-iommu"},
         "cmdline":{"args":"console=hvc0 reboot=k panic=1 nomodules i8042.noaux i8042.nomux i8042.nopnp i8042.dumbkbd root=/dev/vda3"},
         "disks":[{"path":"/opt/clh/images/clear-30080-kvm.img"}],
         "rng":{"src":"/dev/urandom"},
         "net":[{"ip":"192.168.10.10", "mask":"255.255.255.0", "mac":"12:34:56:78:90:01"}]
         }'

Boot a Virtual Machine

Once the VM is created, we can boot it:

#!/bin/bash

curl --unix-socket /tmp/cloud-hypervisor.sock -i -X PUT 'http://localhost/api/v1/vm.boot'

Dump a Virtual Machine Information

We can fetch information about any VM, as soon as it's created:

#!/bin/bash

curl --unix-socket /tmp/cloud-hypervisor.sock -i \
     -X GET 'http://localhost/api/v1/vm.info' \
     -H 'Accept: application/json'

Reboot a Virtual Machine

We can reboot a VM that's already booted:

#!/bin/bash

curl --unix-socket /tmp/cloud-hypervisor.sock -i -X PUT 'http://localhost/api/v1/vm.reboot'

Shut a Virtual Machine Down

Once booted, we can shut a VM down from the REST API:

#!/bin/bash

curl --unix-socket /tmp/cloud-hypervisor.sock -i -X PUT 'http://localhost/api/v1/vm.shutdown'

Command Line Interface

The Cloud Hypervisor Command Line Interface (CLI) can only be used for launching the Cloud Hypervisor binary, i.e. it can not be used for controlling the VMM or the launched VM once they're up and running.

If you want to inspect the VMM, or control the VM after launching Cloud Hypervisor from the CLI, you must use the REST API.

From the CLI, one can either:

  1. Create and boot a complete virtual machine by using the CLI options to build the VM config. Run cloud-hypervisor --help for a complete list of CLI options. As soon as the cloud-hypervisor binary is launched, the REST API is available for controlling and managing the VM.
  2. Start the REST API server only, by not passing any VM configuration options. The VM can then be asynchronously created and booted by sending HTTP commands to the REST API. Check the REST API examples section for more details.

REST API and CLI Architectural Relationship

The REST API and the CLI both rely on a common, internal API.

The CLI options are parsed by the clap crate and then translated into internal API commands.

The REST API is processed by an HTTP thread using the Firecracker's micro_http crate. As with the CLI, the HTTP requests eventually get translated into internal API commands.

As a summary, the REST API and the CLI are essentially frontends for the internal API:

                                  +------------------+
                        REST API  |                  |
                       +--------->+    micro_http    +--------+
                       |          |                  |        |
                       |          +------------------+        |
                       |                                      |      +------------------------+
                       |                                      |      |                        |
+------------+         |                                      |      |                        |
|            |         |                                      |      | +--------------+       |
|    User    +---------+                                      +------> | Internal API |       |
|            |         |                                      |      | +--------------+       |
+------------+         |                                      |      |                        |
                       |                                      |      |                        |
                       |                                      |      +------------------------+
                       |            +----------+              |                 VMM
                       |     CLI    |          |              |
                       +----------->+   clap   +--------------+
                                    |          |
                                    +----------+


Internal API

The Cloud Hypervisor internal API, as its name suggests, is used internally by the different Cloud Hypervisor threads (VMM, HTTP, control loop, etc) to send commands and responses to each others.

It is based on rust's Multi-Producer, Single-Consumer (MPSC), and the single consumer (a.k.a. the API receiver) is the Cloud Hypervisor control loop.

API producers are the HTTP thread handling the REST API and the main thread that initially parses the CLI.

Goals and Design

The internal API is designed for controlling, managing and inspecting a Cloud Hypervisor VMM and its guest. It is a backend for handling external, user visible requests through either the REST API or the CLI interfaces.

The API follows a command-response scheme that closely maps the REST API. Any command must be replied to with a response.

Commands are MPSC based messages and are received and processed by the VMM control loop.

In order for the VMM control loop to respond to any internal API command, it must be able to send a response back to the MPSC sender. For that purpose, all internal API command payload carry the Sender end of an MPSC channel.

The sender of any internal API command is therefore responsible for:

  1. Creating an MPSC response channel.
  2. Passing the Sender end of the response channel as part of the internal API command payload.
  3. Waiting for the internal API command's response on the Receiver end of the response channel.

End to End Example

In order to further understand how the external and internal Cloud Hypervisor APIs work together, let's look at a complete VM creation flow, from the REST API call, to the reply the external user will receive:

  1. A user or operator sends an HTTP request to the Cloud Hypervisor REST API in order to creates a virtual machine:
    shell
    #!/bin/bash
    
     curl --unix-socket /tmp/cloud-hypervisor.sock -i \
     	-X PUT 'http://localhost/api/v1/vm.create'  \
     	-H 'Accept: application/json'               \
     	-H 'Content-Type: application/json'         \
     	-d '{
     		"cpus":{"boot_vcpus": 4, "max_vcpus": 4},
     		"kernel":{"path":"/opt/clh/kernel/vmlinux-virtio-fs-virtio-iommu"},
     		"cmdline":{"args":"console=hvc0 reboot=k panic=1 nomodules i8042.noaux i8042.nomux i8042.nopnp i8042.dumbkbd root=/dev/vda3"},
     		"disks":[{"path":"/opt/clh/images/clear-30080-kvm.img"}],
     		"rng":{"src":"/dev/urandom"},
     		"net":[{"ip":"192.168.10.10", "mask":"255.255.255.0", "mac":"12:34:56:78:90:01"}]
     		}'
    
  2. The Cloud Hypervisor HTTP thread processes the request and de-serializes the HTTP request JSON body into an internal VmConfig structure.
  3. The Cloud Hypervisor HTTP thread creates an MPSC channel for the internal API server to send its response back.
  4. The Cloud Hypervisor HTTP thread prepares an internal API command for creating a virtual machine. The command's payload is made of the de-serialized VmConfig structure and the response channel:
    VmCreate(Arc<Mutex<VmConfig>>, Sender<ApiResponse>)
    
  5. The Cloud Hypervisor HTTP thread sends the internal API command, and waits for the response:
    // Send the VM creation request.
     api_sender
         .send(ApiRequest::VmCreate(config, response_sender))
         .map_err(ApiError::RequestSend)?;
     api_evt.write(1).map_err(ApiError::EventFdWrite)?;
    
     response_receiver.recv().map_err(ApiError::ResponseRecv)??;
    
  6. The Cloud Hypervisor control loop receives the command, as it listens on the internal API MPSC channel:
    // Read from the API receiver channel
    let api_request = api_receiver.recv().map_err(Error::ApiRequestRecv)?;
    
  7. The Cloud Hypervisor control loop matches the received internal API against the VmCreate payload, and extracts both the VmConfig structure and the Sender from the command payload. It stores the VmConfig structure and replies back to the sender ((The HTTP thread):
    match api_request {
        ApiRequest::VmCreate(config, sender) => {
     	   // We only store the passed VM config.
     	   // The VM will be created when being asked to boot it.
     	   let response = if self.vm_config.is_none() {
     		   self.vm_config = Some(config);
     		   Ok(ApiResponsePayload::Empty)
     	   } else {
     		   Err(ApiError::VmAlreadyCreated)
     	   };
    
            sender.send(response).map_err(Error::ApiResponseSend)?;
        }
    
  8. The Cloud Hypervisor HTTP thread receives the internal API command response as the return value from its VmCreate HTTP handler. Depending on the control loop internal API response, it generates the appropriate HTTP response:
    // Call vm_create()
    match vm_create(api_notifier, api_sender, Arc::new(Mutex::new(vm_config)))
        .map_err(HttpError::VmCreate)
    {
        Ok(_) => Response::new(Version::Http11, StatusCode::NoContent),
        Err(e) => error_response(e, StatusCode::InternalServerError),
    }
    
  9. The Cloud Hypervisor HTTP thread sends the formed HTTP response back to the user. This is abstracted by the micro_http crate.