[Android] - Strange things happen with a custom configuration and data connection only

[d3cim]

What is affected by this bug?

My Android rooted devices (no matter the Android version)

When does this occur?

• The device boot up with dnscrypt-proxy installed on it.
• routes are configured with 2 relays at least.
• skip_incompatible = true.
• Wi-fi connection disabled.
• Data connection only.

How do we replicate the issue?

On a Android rooted device with Magisk installed:

1. Disable wi-fi connection.
2. Enable data connection.
3. Download my latest module release.
4. Magisk > Modules > Install from storage > dnscrypt-proxy-android-v2.1.1-1.zip.
5. Reboot.
6. Unlock the device (PIN and/or lock screen).
7. See the issue happen: it take almost 10 minutes to initialize the connection and make the X sign under LTE icon disappear. (sometimes after 10 minutes the connection isn't initialized yet).

Expected behavior (i.e. solution)

The device should connect immediatly, just like under wi-fi only network.

1. I found useful (but again in random ways) toggling on/off the airplane mode.
2. I found useful toggling on the wi-fi connection. At this point the dnscrypt-proxy initialize the connection almost immediatly and then I can use data connection as well if I switch back to it.
3. skip_incompatible = false (not considered due to the lose of the routes correct functionality)

Other Comments

This is the dnscrypt-proxy.toml used.

Take logs was very difficult due to the fact that the issue happen randomly.
At the actual state, with no-wifi connection, is an high risk use dnscrypt-proxy with this configuration.

log_level = 0

dnscrypt-proxy-wifi-only.log

dnscrypt-proxy-data-only.log
dnscrypt-proxy-data-only.jpg

[Gedsh]

Have you tried force_tcp = true ?

[d3cim]

I did about 7 reboots without the issue using force_tcp = true, the initial connection is really fast, I will keep the configuration and do further tests.

As far as the issue happens randomly, and happen that dnscrypt-proxy can manage the initial connection with UDP queries as well, do you think is there anything that can be done to fix the issue without forcing TCP connections?

[Gedsh]

do you think is there anything that can be done to fix the issue without forcing TCP connections?

I don't know Golang but there is a func _dnsExchange

dnscrypt-proxy/dnscrypt-proxy/dnsutils.go

Line 383 in ef1c70e

func _dnsExchange(proxy *Proxy, proto string, query *dns.Msg, serverAddress string, relay *DNSCryptRelay, paddedLen int) DNSExchangeResponse {

It checks if the network supports the fragmented queries required for the relay to work. Only @jedisct1 can know if this algorithm is not working correctly or if LTE networks simply do not support fragmented queries over udp.