Skip to content

Security: rhizomatics/remote_logger

SECURITY.md

Security Policy

Remote Logger follows the Rhizomatics security policy, which covers supported versions, what belongs with Home Assistant itself, and what to expect from a report.

Reporting a vulnerability

Please report it privately, from the Security tab of this repository, with Report a vulnerability. Don't open a public issue, discussion or pull request.

Only the latest release gets security fixes.

What Remote Logger trusts

Remote Logger is a Home Assistant custom integration, so it runs inside Home Assistant with everything Home Assistant can do. It has no login or network service of its own: it only makes outgoing connections, to the log collector it is configured with.

  • Administrators set it up, and are trusted completely, including with the choice of collector, which receives everything that is exported.
  • Other Home Assistant users, and the automations and scripts that call its actions, can add a log record with send_log. Home Assistant lets any user call any action, so they are not trusted with the configuration, the credentials for the collector, or the content of the logs.
  • The content of logs and events is not trusted at all. Log messages, event names and event data often start outside Home Assistant, in a device, an incoming message or another integration.
  • The log collector is trusted with the logs, but not with Home Assistant: nothing it sends back is acted on.

Measures in place

In the integration

  • last_log and flush are for administrators only. last_log gives back log content, which Home Assistant otherwise shows only to administrators.
  • A record added with send_log by a user carries that user's id as user.id, which the caller can't override.
  • last_log masks the value of every configured HTTP header, not only Authorization.
  • The values of event data and state attributes named code, pin, password, secret, token, access_token, refresh_token, api_key, client_secret or entity_picture are replaced with **REDACTED** before export, so alarm and lock codes and camera access tokens stay in Home Assistant.
  • TLS connections verify the server's certificate, for OTLP and for Syslog over TCP. Syslog can also use a private certificate authority and a client certificate (mutual TLS).
  • Syslog over UDP can't be encrypted, and the configuration refuses TLS with UDP rather than quietly sending plain text.
  • A token sent to an OTLP collector over plain HTTP raises a Home Assistant repair issue.
  • HTTP redirects from a collector are not followed, so credentials aren't sent on to another address.
  • Syslog APP-NAME, MSGID and structured data names are reduced to the characters RFC 5424 allows, and a continuation line can't begin with <, so event names, event data and messages can't add fields or records.
  • Records waiting for a collector that can't be reached are capped at 10,000, the oldest dropped first, so an outage can't use up Home Assistant's memory.
  • Error text from a collector is cut to 200 characters before it is logged or shown on a sensor, and a failure to format an event logs the event type, not its data.

In development

  • CodeQL analysis and the OpenSSF Scorecard run on the default branch, and Bandit and zizmor on every change.
  • Dependabot raises dependency and GitHub Action updates, with a cooling-off period for new versions, and pull requests get a dependency review.
  • Every GitHub Action is pinned to a commit. Workflow tokens have no permissions unless a job needs them, and the tests of a pull request run with a read-only token.
  • The integration has no runtime dependencies beyond Home Assistant itself, checked by hassfest.
  • Releases are built and published by GitHub Actions, not from a developer's machine, with a SHA-256 checksum and a build provenance attestation for the zip, which can be checked with gh attestation verify remote_logger.zip --repo rhizomatics/remote_logger.
  • Ruff, mypy and a test suite with over 90% coverage run on every change, with a regression test for each fix.

Known limitations

  • Redaction goes by the name of a key. A secret under another name, or inside a log message or a state value, is exported as it is. Only subscribe to the events you need, see Security.
  • Syslog over UDP, and OTLP over plain HTTP, can be read and altered by anyone on the network path.
  • A syslog receiver can't tell who sent a record: anything that can reach it can send records that look like Home Assistant's, unless it requires client certificates.
  • The collector token and headers are kept as Home Assistant keeps all credentials, in plain text in its .storage folder and in backups.
  • The oldest records are dropped without a count when the cap on waiting records is reached.

There aren't any published security advisories