Skip to content

Latest commit

 

History

History
109 lines (86 loc) · 3.87 KB

metadata.md

File metadata and controls

109 lines (86 loc) · 3.87 KB
Raw

Metadata and Userdata handling

Most providers offer two general mechanisms to provide compute instances with information about the instance that cannot be discovered by any other means. There are usually two types of information, namely metadata and user-data. Metadata is usually set by the provider (e.g. geographical region of the datacentre, name given to the instance, external IP address, tags and other similar information), while userdata is fully custom, hence the name, and it is the information that user may supply to their instances before launch (it is immutable in most providers).

The metadata package handles both metadata and userdata for a number of providers (see below). It abstracts over the provider differences by exposing both metadata and userdata in a directory hierarchy under /run/config. For example, sshd config files from the metadata are placed under /run/config/ssh.

Userdata is assumed to be a single string and the contents will be stored under /run/config/userdata. If userdata is a JSON file, the contents will be further processed, where different keys cause directories to be created and the directories are populated with files. For example, the following userdata file:

{
  "ssh": {
    "entries": {
      "sshd_config": {
        "perm": "0600",
        "content": "PermitRootLogin yes\nPasswordAuthentication no"
      }
    }
  },
  "foo": {
    "entries": {
      "bar": {
        "content": "foobar"
      },
      "baz": {
        "perm": "0600",
        "content": "bar"
      }
    }
  }
}

will generate the following files:

/run/config/ssh/sshd_config
/run/config/foo/bar
/run/config/foo/baz

The JSON file consists of a map from name to an entry object. Each entry object has the following fields:

  • content: if present then the entry is a file. The value is a string containing the desired contents of the file.
  • entries: if present then the entry is a directory. The value is a map from name to entry objects.
  • perm: the permissions to create the file with.

The content and entries fields are mutually exclusive, it is an error to include both, one or the other must be present. The file or directory's name in each case is the same as the key which referred to that entry.

This hierarchy can then be used by individual containers, who can bind mount the config sub-directory into their namespace where it is needed.

Metadata image creation

linuxkit run backends accept two options to pass metadata to the VM in a platform specific manner to be picked up by the pkg/metadata component:

  • -data=STRING will cause the given STRING to be passed to the VM
  • -data-file=PATH will cause the contents of the file at PATH to be passed to the VM

Alternatively linuxkit metadata create meta.iso STRING will produce a correctly formatted ISO image which can be passed to a VM as a CDROM device for consumption by the pkg/metadata component.

Providers

Below is a list of supported providers and notes on what is supported. We will add more over time.

GCP

GCP metadata is reached via a well known URL (http://metadata.google.internal/) and currently we extract the hostname and populate the /run/config/ssh/authorized_keys from metadata. In the future we'll add more complete SSH support.

GCP userdata is extracted from /computeMetadata/v1/instance/attributes/userdata and made available in /run/config/userdata.

AWS

AWS metadata is reached via the following URL (http://169.254.169.254/latest/meta-data/) and currently we extract the hostname and populate the /run/config/ssh/authorized_keys from metadata.

AWS userdata is extracted from http://169.254.169.254/latest/user-data and and made available in /run/config/userdata.

HyperKit

HyperKit does not distinguish metadata and userdata, it's simply refered to as data, which is passed to the VM as a disk image in ISO9660 format.