nova-specs/specs/queens/implemented/deprecate-file-injection.rst
Matt Riedemann 37c576b9a8 Fix problem description number in deprecate file injection spec
The numbers in this list weren't being rendered properly.

Change-Id: Ifd010ca8efcef4c58b6a5b547e4728c31b91fa2c
2018-07-26 21:51:08 -04:00

9.0 KiB

Deprecate file injection

https://blueprints.launchpad.net/nova/+spec/deprecate-file-injection

Deprecate "personality files", otherwise known as file injection, from the compute REST API. This is something we have talked about for several years since file injection is insecure, not very user friendly, and we have alternatives.

Problem description

There are a few issues with nova's file injection support.

  1. It is not discoverable by the end user

    As noted in1, depending on how the compute host is configured, the user request to inject personality files may be silently ignored. For the libvirt driver, file injection is disabled by default via the [libvirt]/inject_partition configuration option, so the default behavior is to not honor the user request, at least when using libvirt which is by far the most widely deployed compute driver in nova. Note that the [libvirt]/inject_partition configuration option default value has been to disable file injection since the Icehouse release.

  2. It is not secure

    When injecting files, if libguestfs is not available on the compute host, then the VFSLocalFS code is used to inject files via the host. This means malicious images could exploit the host. This was also discussed in the mailing list2. This was originally a hack to workaround an issue in older versions of Ubuntu and is going to be removed regardless of the changes proposed in this spec.3

  3. It is not persisted

    Personality files are not persisted in the database which means they cannot be retrieved via the metadata service API nor are they available during operations like evacuate where the server instance is rebuilt on a different compute host because the source host failed.

  4. There are better alternatives available

    The configuration drive is the standard way for users to inject user data into their server instance. The supplied user_data is persisted with the instance so it is available from the metadata service API and is available when rebuilding the configuration drive on another host. There are three ways to tell nova to create a configuration drive for an instance:

    • The image can have metadata indicating a configuration drive is required.
    • The operator can configure the force_config_drive option in nova.conf.
    • The end user can request a configuration drive via the config_drive parameter in the POST /servers API.

    Note that failure to build a config drive during instance create will result in the build request getting rescheduled to another compute host. The same cannot be said for file injection being disabled on the host.

Use Cases

As a user, I want a predictable way to inject data into my server instance.

As a nova developer, I do not want to maintain legacy code flows for which there are better alternatives.

Proposed change

There are a few changes to the REST API in a new microversion:

  1. Deprecate the personality parameter from the POST /servers create server API and the POST /servers/{server_id}/action rebuild server API.

    Specifying the personality parameter in the request body to either API will result in a 400 Bad Request error response.

  2. We will add support to pass user_data to the rebuild server API as a result of this change. Several people said this would be useful for their users in a related mailing list thread.4

  3. Stop returning maxPersonality and maxPersonalitySize values from the GET /limits API. Since we want to stop accepting personality files when creating or rebuilding a server instance, we should also stop reporting the quota limits on those resources in the new microversion.

  4. Stop accepting and returning injected_files, injected_file_path_bytes, injected_file_content_bytes from the os-quota-sets and os-quota-class-sets APIs.

Note

There are configurable quota limits on injected_files, injected_file_content_bytes and injected_file_path_length. There is no quota on the user_data that is supplied with an instance. The only limitation on user data is the size of the field in the database, which is ~16MiB with MySQL. With the default file injection quota values, the limit is less than 1MiB per request, so we have ample space in the database for storing in user_data what otherwise would have been specified with personality files.

Since personality file injection will still be supported with older microversions, there will be nothing removed from the backend compute code related to file injection (except for the insecure VFSLocalFS code as noted already). Nor will there be any deprecation of related configuration options for file injection. The point of this microversion is really to signal that users should not be using this legacy part of the compute API, and to set a timer on when it could be removed if nova ever starts requiring a higher minimum supported microversion in the distant future.

Alternatives

Persist the supplied personality files, but this is essentially duplicating user_data which is already persisted and made available to the config drive and via the metadata service API and does not solve the security or end user discoverability issues.

Data model impact

None

REST API impact

In a new microversion:

  1. Deprecate the personality parameter from POST /servers create server API and POST /servers/{server_id/action rebuild server API. Specifying that parameter after the microversion will result in a 400 Bad Request error response.

  2. Support passing user_data to the rebuild server action API. The schema would be the same as the POST /servers server create API:

    'user_data': {
        'type': 'string',
        'format': 'base64',
        'maxLength': 65535
    }
  3. Deprecate the maxPersonality and maxPersonalitySize response parameters from the GET /limits API.

  4. Deprecate the injected_files, injected_file_path_bytes, injected_file_content_bytes parameters from the following APIs:

    • GET /os-quota-sets/{tenant_id}
    • PUT /os-quota-sets/{tenant_id}
    • GET /os-quota-sets/{tenant_id}/defaults
    • GET /os-quota-sets/{tenant_id}/detail
    • GET /os-quota-class-sets/{id}
    • PUT /os-quota-class-sets/{id}

Security impact

Removing the VFSLocalFS fallback code will actually be good for security.

Notifications impact

None; personality files were never part of any notifications (thankfully).

Other end user impact

Update python-novaclient CLIs and related python API bindings, specifically:

  • Deprecate the --file option from the nova boot and nova rebuild CLIs and API bindings.
  • Add --user-data to the nova rebuild CLI and API bindings.
  • Deprecate the maxPersonality and maxPersonalitySize fields from the nova limits and nova absolute-limits CLIs and API bindings.
  • Deprecate injected_files, injected_file_content_bytes, and injected_file_path_bytes from the nova quota-show, nova quota-update, nova quota-defaults, nova quota-class-show, and nova quota-class-update CLIs and API bindings.

Performance Impact

None

Other deployer impact

None

Developer impact

None

Implementation

Assignee(s)

Primary assignee:

Matt Riedemann (mriedem) <mriedem.os@gmail.com>

Work Items

  • Add a microversion to make the proposed changes to the server create, server rebuild, limits, os-quota-sets and os-quota-class-sets APIs.
  • Make the related changes in python-novaclient.

Dependencies

None

Testing

  • Unit tests for negative scenarios.
  • Functional API samples tests for the normal API flows with the new microversion.

Documentation Impact

  • The compute API reference will need to be updated for the new microversion impacts.
  • The Manage Compute service quotas doc will need to be updated.

References

More mailing list discussion from Ocata: http://lists.openstack.org/pipermail/openstack-dev/2016-November/107195.html

History

Revisions
Release Name Description
Queens Introduced

  1. http://lists.openstack.org/pipermail/openstack-dev/2016-July/098703.html↩︎

  2. http://lists.openstack.org/pipermail/openstack-dev/2016-November/107233.html↩︎

  3. https://review.openstack.org/#/c/324720/↩︎

  4. http://lists.openstack.org/pipermail/openstack-operators/2017-October/014309.html↩︎