Skip to content

RTSP handshake failed with error 60 while using Moonlight on Tailscale + Mobile Data #669

Open

Description

@parksoonometry

Describe the bug
When using Moonlight in combination with Tailscale over Mobile Data (5G), there is a "Connection Failed" error with the description being

"RTSP handshake failed with error 60

Check your firewall and port forwarding rules for port(s):
TCP 48010
UDP 48000
UDP 48010"

This only appears to be an issue on mobile data + Tailscale only. Using Tailscale and Wi-Fi appears to work fine on the exact same setup.

Steps to reproduce

  • Register Tailscale to iOS device
  • Enable Tailscale VPN in iOS settings
  • Register Tailscale IP to Moonlight
  • Connect to computer using Tailscale IP

Screenshots

Image

Affected games
N/A

Other Moonlight clients

  • Tailscale + Moonlight works fine with Windows and Steam Deck.

Moonlight settings (please complete the following information)

  • Have any settings been adjusted from defaults? None
  • If so, which settings have been changed? None
  • Does the problem still occur after reverting settings back to default?
    N/A

Device details (please complete the following information)

  • iOS/tvOS version: iOS 18.3.1
  • Device model: iPhone 13 Pro Max

Server PC details (please complete the following information)

  • OS: Windows 10 19045.5487
  • Apollo Version: v0.3.1
  • Nvidia GPU driver: 572.42
  • Antivirus and firewall software: Windows Defender and Windows Firewall

Additional context
Current Mobile Phone provider is T-Mobile

Activity

  1. cerool commented on Apr 27, 2025

    @cerool

    Me and a couple of my friends are having the same vein of issue, though I am the only one without tailscale and my carrier myself is Cricket who go through AT&T, we all have later iPhones on IOS 18.4 iirc so a bit different but linked. For fun I decided to attempt to use apollo for the desktop client instead of sunshine and got the same error as well, starting to thing apple broke something in a update push? idk

  2. SullieLore commented on May 6, 2025

    @SullieLore

    Same issue.

  3. poizuun-exe commented on Aug 2, 2025

    @poizuun-exe

    Same issue...still. any update on this from anyone with success?

  4. treyf711 commented on Dec 28, 2025

    @treyf711

    The only way I’ve found to allow moonlight when on a cellular connection is by using an exit node on the same network as my computer running sunshine.

  5. NuclearJoe2020 commented on Feb 2, 2026

    @NuclearJoe2020

    The exit node thing works. I am just using the PC running sunshine as an exit node for my iPhone.

  6. gabe4278 commented on Jul 11, 2026

    @gabe4278

    Same issue as of WireGuard, I opened a port over the internet from a Raspberry Pi, then WireGuard mobile to connect via cellular, using Moonlight and the RTSP handshake failed.

    Unless you got Internet forwarding enabled from network configuration, this may seem to work.

  7. harrison001 commented on Aug 15, 2026

    @harrison001

    Root cause below — but the important part is that this is already fixed upstream, and the iOS client has never picked the fix up.

    What's happening

    On an IPv6-only cellular network (NAT64/DNS64), getaddrinfo() on an IPv4 literal
    returns a synthesised IPv6 address. If that IPv4 belongs to a VPN (Tailscale's
    100.64.0.0/10, WireGuard's 10.0.0.0/8, …), the synthesised address is not inside the
    VPN's routes — so those packets leave the tunnel, hit the carrier's NAT64 gateway, and go
    looking for a CGNAT address on the public internet. Nothing answers.

    Why one app takes two different paths: the HTTP calls (/serverinfo, /launch) go
    through NSURLSession/CFNetwork, which hands the literal to the stack and lets the VPN
    route it. RTSP goes through moonlight-common-c, which resolves the address itself with
    getaddrinfo(). That is exactly why pairing and the app list work while streaming fails.

    Captured on the phone with rvictl + tcpdump — same capture, same host, same second.
    Addresses are masked but the structure is verbatim: 100.a.b.c is the host's Tailscale
    address, <nat64> is the carrier's NAT64 prefix.

    # RTSP — retransmits into silence, then ETIMEDOUT (60)
    IP6 <phone>.65521 > <nat64>::<hex(100.a.b.c)>.48010: Flags [S], seq 2739160459
    IP6 <phone>.65521 > <nat64>::<hex(100.a.b.c)>.48010: Flags [S], seq 2739160459   # +1s
    IP6 <phone>.65521 > <nat64>::<hex(100.a.b.c)>.48010: Flags [S], seq 2739160459   # +2s
    ... 7 retransmits, no reply ...
    
    # HTTPS to the SAME host in the same capture — plain IPv4, inside the tunnel, works
    IP  <phone>.58018 > 100.a.b.c.47984: Flags [S], ...   -> SYN-ACK, session proceeds
    

    Decoding the RTSP destination makes it obvious: its low four bytes are the host's
    Tailscale IPv4 written in hex, carried inside the carrier's NAT64 prefix. The client
    never dialled the address I configured.

    This is already fixed in moonlight-common-c

    • 0975a86
      (2025-07-15) — "Implement workaround for bad synthesized IPv6 addresses when connecting
      through a v4-only VPN on iOS"
    • 5f22801
      (2025-07-15) — "Fix CGN subnet mask", which is what makes it cover 100.64.0.0/10,
      i.e. Tailscale

    But moonlight-ios has not picked it up

    moonlight-ios master pins the submodule at 48d7f1a (2024-02-25) — 24 commits behind
    that fix
    . The App Store build is older still, so building from source doesn't help either
    unless you bump the submodule yourself.

    Comparing the three clients:

    client pinned moonlight-common-c has the fix
    moonlight-qt e41355e (2026-07-23) yes
    moonlight-android 8af4562 (2024-02-18) no
    moonlight-ios 48d7f1a (2024-02-25) no

    Which is the awkward part: the fix is written for iOS — the commit message says so — and
    the only client carrying it is the one that doesn't run on iOS.

    A submodule bump would close this issue.

    Confirmed: the fix works, it just needs to reach the client

    I cherry-picked 0975a86 and 5f22801 onto the currently pinned 48d7f1a — both apply
    cleanly, they only touch PlatformSockets.c, and the helper they need comes with them —
    then built and installed that on my phone alongside the App Store build.

    Same phone, same cellular connection, same minutes, same host entry (the IPv4 Tailscale
    literal, not the IPv6 workaround), Wi-Fi off. Both builds installed side by side:

    App Store 9.0.2 patched build
    Streaming fails, as in this issue works
    RTSP reaching the host never arrives 7 × RTSP/1.0 200 OK, then CLIENT CONNECTED

    The host-side log shows the signature this issue is about: HTTP requests keep arriving
    throughout, and exactly one RTSP handshake and one session ever appear — the one from the
    patched build. (Sunshine cannot label which client sent which HTTP request, since
    moonlight-common-c sends a fixed uniqueid, so the A/B itself is my observation and the
    log is corroboration.)

    The phone really was on cellular: its Tailscale peer endpoint at the time was a global IPv6
    address on the carrier's network, not a LAN address.

    So the fix itself is good. The problem is purely that it has never shipped — and note that
    cherry-picking is what made this testable at all, since a straight submodule bump does not
    build against the current iOS project.

    Why the exit-node workaround works

    @treyf711 / @NuclearJoe2020 — with an exit node, all traffic goes into the tunnel, so
    the synthesised address no longer escapes it. That's consistent with the capture, and it's
    a good sanity check on the diagnosis. @gabe4278's WireGuard case is the same shape:
    10.0.0.0/8 is a private range, so it gets synthesised too.

    Workaround if you can't use an exit node

    Give the client an address that cannot be synthesised — the host's IPv6 literal:

    1. Sunshine: address_family = both (it listens on IPv4 only by default)
    2. Moonlight: delete every existing host entry, then add only the IPv6 literal,
      e.g. [fd7a:xxxx:xxxx::xxxx:xxxx]

    Both halves are required. Adding the IPv6 entry alongside the old IPv4 one still failed,
    and the capture showed HTTP still going to the IPv4 address.

    Note on error 60

    It's ambiguous and it misled me for a while. ETIMEDOUT is produced both by the
    receive-timeout path and by *error = LastSocketError() after a failed connect, since a
    TCP connect timeout on Apple platforms is ETIMEDOUT. It reads like "connected but got
    no reply"; here it was actually "never connected".


    Happy to test a branch with the submodule bumped, on the same cellular setup — I can
    reproduce this on demand.

  8. MythicHH commented on Aug 20, 2026

    @MythicHH

    It'd be nice if the iOS client could be updated.

  9. harrison001 commented on Sep 11, 2026

    @harrison001

    Follow-up to my analysis above, since a few people have asked for the client to be updated.

    I bumped moonlight-common-c in my fork — it now carries 0975a86 and 5f22801, the two commits that fix this — and that build is on the App Store, free, if you need something that works on cellular today: https://apps.apple.com/app/id6802870894

    To be clear about what that is and is not. The fix is upstream's, written by the Moonlight developers for exactly this case; all I did was ship a client with it. A submodule bump here still closes this issue properly, and I would rather that happened than have people go to a fork for it. The offer in my earlier comment stands — I can reproduce this on demand on cellular and will test a branch with the bump.

    One practical note for anyone stuck right now: the exit-node workaround @treyf711 and @NuclearJoe2020 described still works and costs nothing, and the IPv6-literal workaround at the end of my earlier comment works without an exit node.

  10. jnobbs commented on Sep 12, 2026

    @jnobbs

    Follow-up to my analysis above, since a few people have asked for the client to be updated.

    I bumped moonlight-common-c in my fork — it now carries 0975a86 and 5f22801, the two commits that fix this — and that build is on the App Store, free, if you need something that works on cellular today: https://apps.apple.com/app/id6802870894

    To be clear about what that is and is not. The fix is upstream's, written by the Moonlight developers for exactly this case; all I did was ship a client with it. A submodule bump here still closes this issue properly, and I would rather that happened than have people go to a fork for it. The offer in my earlier comment stands — I can reproduce this on demand on cellular and will test a branch with the bump.

    One practical note for anyone stuck right now: the exit-node workaround @treyf711 and @NuclearJoe2020 described still works and costs nothing, and the IPv6-literal workaround at the end of my earlier comment works without an exit node.

    Worked immediately. Thankyou so much!! I’ve been running around in circles trying to work out what was happening before I found your post.

    Your intellect is gargantuan honestly, impressive!!!

  11. CharlesMcquade commented on Oct 26, 2026

    @CharlesMcquade

    Ran into this exact thing recently and wanted to add my experience.

    Got one of the new iPhone Duos and switched to T-Mobile. Moonlight over Tailscale worked fine on WiFi but every cellular attempt failed with "RTSP handshake failed with error 60". Spent way too long blaming my own setup (firewall rules, Tailscale ACLs) before finding this thread. T-Mobile's network is IPv6-only, and the NAT64 translation is what kills the connection.

    The fix is already on master via the moonlight-common-c update, but the App Store build (9.0.2) is from 2024 so none of that is actually shipped.

    In the meantime I sideloaded master plus PR #729 (the UIScene / Duo resize fix) built with Xcode, and everything works. Duo layout is fixed, cellular streaming is back to normal.

    Would love to see an App Store update with both. Would save the next Duo on T-Mobile person from this exact wall, and me from maintaining my own sideloaded copy. Thanks for Moonlight!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions