Skip to content

LinkProven v2.5.0

Choose a tag to compare

@mhmd2520 mhmd2520 released this 21 Nov 19:22
· 1 commit to main since this release

Both ends of a Linux agent connection now prove the agent token (ADR 0003).
Only set-ups that use the agent token are affected: without a token the
control channel is exactly as before, byte for byte. Reports, sidecars
(schema 13) and the cloud are unchanged.

Security

  • The Linux agent proves the token to this laptop. Before, only the
    laptop proved it, so anything answering on the server's address could run
    the whole sweep and send back made-up Server-side results (loss, NIC
    counters, the server's own details) that went into the report. Now the
    agent proves the token first, and LinkProven stops with its own message
    before a case runs if the proof is wrong or missing. The proof also covers
    the laptop's opening message, so nothing on the way can quietly switch
    off the Server-side evidence a sweep asks for.
  • Every message after the token check is signed. Each control message
    carries a code made from the token and this connection only. A message
    that was changed, replayed or reordered on the way ends the sweep at once,
    and both ends say it was tampered with (never "connection lost"), so no
    one on the test network can change a result in transit. Results are not
    encrypted.
  • A recorded sign-in to the agent can't be replayed. The agent used to
    remember used codes only until it restarted, and a fake server could
    collect a fresh code from a laptop and replay it straight away. Each
    connection now has to answer a fresh challenge from the agent, and the
    agent keeps no list to lose.
  • The web portal sends a Content-Security-Policy. It names everything
    the portal may load: its own files, Flutter's renderer and fonts from
    Google's CDN, and the LinkProven cloud. A script injected from anywhere
    else would not run. It ran in report-only mode on the live portal first;
    a signed-in session logged no violation, so it is now enforced.

Changed

  • Update the Linux agent together with LinkProven. Once a token is in
    use, the agent and every laptop that uses it must all be on 2.5.0 or
    later: an older laptop and a 2.5.0 agent refuse each other. There is no
    fallback to the old check, because a fallback would let the old attacks
    back in. After updating, run the Linux server set-up (Setup →
    Server is Linux…) again to update the agent. A laptop with a token saved
    also refuses a Server that can't prove it, including a Windows LinkProven
    Server at the same address: clear the token in that dialog to use one. Both
    cases say so in plain words when you connect.
  • The phone app's Server code follows the same rules, ready for when it
    offers a token.

Fixed

  • The Windows Server never listens on every network. A Server address
    of 0.0.0.0 (any network) is refused with a message instead of opening
    the control port to the whole LAN, as the Linux agent already did.
  • Error banners and the Multi-Client PDF never treat text as markup. An
    iperf3 error message containing < could stop a Multi-Client PDF from
    saving; the app-wide warning banner now always shows plain text.
  • A malformed number from the Server no longer loses the sweep. A
    number too long or too large to read in a Server's reply (for example
    1e999 packets) used to stop the sweep and throw away every finished
    case. Now that reply is treated as a bad message and the cases already
    measured are kept.