[gonzaarcr]

Reopen from #886

09-21 13:35:56.361 D gojni: panic in multiTUN.run close of closed channel: goroutine 2 [running]:
runtime/debug.Stack()
/build/.cache/tailscale-go/src/runtime/debug/stack.go:26 +0x64
github.com/tailscale/tailscale-android/libtailscale.(*multiTUN).run.func1()
/build/tailscale-android/libtailscale/multitun.go:106 +0x3c
panic({0x7041e7e660?, 0x7042041950?})
/build/.cache/tailscale-go/src/runtime/panic.go:860 +0x12c
created by github.com/tailscale/tailscale-android/libtailscale.newTUNDevices in goroutine 40
/build/tailscale-android/libtailscale/multitun.go:99 +0x20c

09-21 13:35:56.379 E Go : panic: close of closed channel [recovered, repanicked]
09-21 13:35:56.384 F libc: Fatal signal 6 (SIGABRT), code -6 (SI_TKILL)
in tid 31023 (Thread-21), pid 30757 (m.tailscale.ipn)
09-21 13:35:58.345 I Zygote: Process 30757 exited due to signal 6 (Aborted)

Fixes tailscale/tailscale#21897

[atondwal]

Independent repro of this on a Galaxy Z TriFold (SM-F968B), built from
main (9b62cdc) plus an unrelated network-switch patch. Five peers left the tailnet within one second, which
rebuilt the TUN five times in about 100 ms:

15:09:19.200 wg: Failed to write packets to TUN device: write /dev/tun: input/output error

After that, nothing on the phone could reach the network while the app still
said Connected, until I toggled the VPN about a minute later, which crashed
with the same panic you saw:

15:10:37.704 panic in multiTUN.run close of closed channel

I've been running a build with this PR applied since, with no drops. That
build also includes a separate fix that makes these rebuild bursts rare
(filed as tailscale/tailscale#22129), so I can't credit the stability to this
PR alone. Happy to share the full logcat.