So I've stumbled across this weird networking issue while using my Android phone (a rooted Droid 4). What I was seeing was this:
I had gone to our club hangout and had used the wifi there. After coming back home again, and using my wifi at home, my phone wouldn't connect to my server, neither via SSH nor via IMAPS nor any other protocol. Fearing something had caused my server to go offline, I tried to reach it in other ways (i.e. by connecting through other hosts).
I then discovered something weird.
- my server was online
- I could connect to it via some other remote host that took a different path into the server network
- oh wait, I could connect to it via a different host that was in the same network as my server? (Meaning the network as such was reachable)
- oh wait, other computers inside my wifi can connect to the server? Uh .. what?
- oh wait, when I disconnect from my wifi and use the 3G connection instead, I can connect to the server
So, next, I looked at the shell on my phone again and tried to see if maybe the routing was broken, but
$ ip route show
gave a completely normal output (i.e. there was only the default route to my wifi router plus the local netmask). Next, I tried to see whether there was a pattern to to the IP addresses I could reach and the ones I couldn't. There was. It was only the IP address of my server that was unreachable. I tried connecting to the IP addresses of some VMs that run on that server (i.e. behind the same bridging interface) and they were reachable.
Just for laughs, I used another IP address from the same server network that was unused at that time and added that one to the interface, and my server was instantly reachable via that address. So it really was that one, specific, IP address that was unreachable.
At which point a colleague of mine (I had thrown my story out to some people I know to aid in brainstorming) said "Have you looked at the arp cache?"
So I did.
$ arp fnanp.in-ulm.n-addr.arpa (188.8.131.52) at <incomplete> on wlan0
So. A full two days earlier, my phone had been connected to a wifi in which said server can be reached without going through a router and therefore had added the servers address to its arp cache. If I read the arp(7) manpage correctly, the lifetime of an entry in the arp cache should be around the 30 seconds mark, but at this point it was already 48 hours that it had been there. Another way to see that the phone thought the server should be in the same network:
$ ip neigh show 184.108.40.206 dev wlan0 FAILED
$ ip route show cache | grep 220.127.116.11
showed it as a local address that was offline. At this point I was pondering whether I should flush the cache and just be done with it or whether I wanted to find out more about this. But then my parents came over to visit and we went out to have dinner. When we came back and my phone reconnected to my wifi, the issue was gone.
Well, I'll be back in the wifi that caused the whole problem tomorrow evening. I'm curious whether it'll happen again.