milsivich

First time poster, hopefully this format works. Diagnostics package attached (Tailscale diag plugin, generated while the failure was active, hopefully, if I fucked up the order and captured a healthy diag then it's easy to reproduce, I just have to restart the server again and it immediately bricks the NFS functions on the server). Summary: With "Unraid services listen on Tailscale IP" = Yes, the NFS daemon fails during Unraid's normal boot-time reload_services restarts. The Tailscale IPv4 and IPv6 addresses get added to the rpc.nfsd bind list, and registration fails. With the plugin disabled, the same boot-time restarts occur and succeed. This makes NFS shares unreachable after every reboot. Environment: - Unraid 7.3.4, then again Unraid 7.4.0 - Tailscale plugin 2026.10.22, then again 2026.10.28.0147 - "Unraid services listen on Tailscale IP" = Yes - NFS shares exported, clients on the same LAN - Tailscale node assigned both IPv4 (100.x) and IPv6 (fd7a:...) addresses Reproduction: 1. Enable Tailscale; "Unraid services listen on Tailscale IP" = Yes; Apply. 2. Reboot. 3. NFS shares unreachable; port 2049 not listening. 4. Removing tailscale1 from Include Listening Interfaces and restarting rc.rpc + rc.nfsd restores NFS immediately. 5. Reproduced on two separate boots, including across an Unraid update (Oct 27 and Nov 13, 2026). The include-list entry is re-added on every reboot. Evidence (I originally recorded evidence on Oct 27 and then got distracted, reproduced it again on Nov 13 when the Unraid update hit, but I'll post the older data first): Oct 27 21:59:49 clean start, 13 seconds later the reload_services restart is poisoned: rc.nfsd: /usr/sbin/rpc.nfsd -u -H 192.168.8.141 -H 127.0.0.1 -s 8 rc.nfsd: NFS server daemon... Started. reload_services: execute queued job 4 rc.nfsd: Restarting NFS server daemon... rc.nfsd: /usr/sbin/rpc.nfsd -u -H 192.168.8.141 -H 127.0.0.1 -H 100.102.129.5 -H fd7a:115c:a1e0::2b46:8106 -s 8 rpc.nfsd: writing fd to kernel failed: errno 111 (Connection refused) rpc.nfsd: unable to set any sockets for nfsd kernel: svc: failed to register nfsdv3 RPC service (errno 111). kernel: svc: failed to register nfsdv3 RPC service (errno 97). errno 97 is EAFNOSUPPORT — the Tailscale IPv6 address appears to be the trigger. On the Nov 13 boot the failure was severe enough that rpcbind (port 111) was also down; /etc/rc.d/rc.rpc restart + rc.nfsd restart after clearing the include list recovered the server. Additional observations: - rc.nfsd restarts twice during normal boot (after array import, after compose autostart). This happens regardless of the plugin — it is stock reload_services behavior. - tailscale-watcher.php starts ~4 minutes into boot, i.e. AFTER the failing restarts, so the watcher is not the initial trigger (though it does contribute to later churn; killing it stopped nginx restarts on an earlier, related failure). - No INCLUDE_INTERFACE line was found in /boot/config/network.cfg via grep, yet the include-list behavior is present — the interface appears to be injected dynamically. Workarounds found: a) "Unraid services listen on Tailscale IP" = No — fixes NFS, but breaks remote access to all services over the Tailscale IP (the main reason the plugin is used here). b) After each reboot: remove tailscale1 from Include Listening Interfaces, rc.rpc restart, rc.nfsd restart, kill tailscale-watcher.php. Request: Investigate how/when the plugin adds tailscale1 to the include list, and whether the Tailscale IPv6 address can be excluded from the NFS bind list (or from the -H arguments passed to rpc.nfsd) while other services continue to listen on the Tailscale IP. Happy to test builds or provide more logs. Note: I pruned the system-diagnostics a bit (ident.cfg, superdat, folders.txt removed); full package available privately on request. tailscale-diag-20261113-153529prune.zip