Probably should find a linux networking specific community for this one…
I have a strange issue that feels very familiar, like I’ve fixed it before, but I can’t remember how.
I try to rtsp to security cam:
ffplay rtsp://user:[email protected]:554/h264Preview_01_main
And I get a no route:
Connection to tcp://192.168.19.137:554?timeout=0 failed: No route to host
rtsp://user:[email protected]:554/h264Preview_01_main: No route to host
Strange, I’m in the same subnet 192.168.19.129/24, and it worked a few days ago.
Check ping:
ping 192.168.19.137
PING 192.168.19.137 (192.168.19.137) 56(84) bytes of data.
64 bytes from 192.168.19.137: icmp_seq=1 ttl=64 time=5.69 ms
Of course… So I run the command again;
ffplay rtsp://user:[email protected]:554/h264Preview_01_main
And now it works.
I could bandaid by crontabbing a ping every hour or something, but I would really like to know why I’m getting a ‘no route’ until I ping.
My routing table is pretty basic:
default via 192.168.19.1 dev enp4s0 proto dhcp src 192.168.19.129 metric 100
default via 192.168.19.1 dev enp4s0 proto dhcp src 192.168.19.129 metric 1002
172.17.0.0/16 dev docker0 proto kernel scope link src 172.17.0.1 linkdown
172.18.0.0/16 dev br-68c1e0344e27 proto kernel scope link src 172.18.0.1 linkdown
192.168.19.0/24 dev enp4s0 proto dhcp scope link src 192.168.19.129 metric 1002
192.168.19.1 dev enp4s0 proto dhcp scope link src 192.168.19.129 metric 1024
And I don’t think I have any rules in firewall for LAN.
Any ideas?
Are you sure the camera is not going into some kind of low power sleep mode and then has a wake-on-LAN functionality that only responds to icmp/magic packets? The number of embedded devices that have aggressive power saving measures built in is kinda stupid tbh.
Can you reliably rtsp from a different machine to rule out the routing table?
I can try test that.
The camera is always accessible via the android app, but that’s not quite the same.
The android app, if proprietary and manufacturer specific, could very well be sending its own magic wake-on-lan packet to check if the camera is alive.
I unfortunately don’t know enough about nitty-gritty networking stuff to help with the actual routing though, refer to the other commenters for that.
This is what I thought it sounded like too. I’ve seen similar behavior with wake on lan before.
I want to say ARP. Can I say ARP?
This is the answer.
Do you have an IP address conflict?
I would check this first too, seems a bit like it. Check your arp table for anything nasty.
I would put my money on that too.
Why do you have two default routes?
The last line in your table also doesn’t look right.
Im rusty. So don’t trust this.
It looks like you have a routing issue with your default route. The fact that a ping gets the IP to start working, tells me that you need a broadcast packet on the local network, that broadcast excites the other computer to send a message out, that updates the IP to Mac table, which then makes the machine routable because now there is a direct ethernet pathway.
So I think the magic here is the initial broadcast ping is doing.
Ideally this isn’t necessary, ethernet should be sending out a broadcast packet for the Mac to IP table anyway for your TCP traffic. I don’t know why that’s not happening. I would do an TCP dump in both scenarios, and see if the broadcast is going out.
My intuition is that there’s something fucky going on with your default route, that ping is not being affected by. I bet you don’t send out a broadcast address discovery packet in the TCP scenario but you do in the ping scenario
But any IP packet should trigger an arp “who has?”.
Yes. It should. Hence the fucky mystery. Good to check with a network dump to rule it out.
When this happened to me, the broadcast would trigger the ARP update; the camera would respond ever so slightly slower and since it was the last device to claim it was at the IP the ARP table would be updated with it. It would work until the conflicting device would send a packet which would update the ARP table again, back to the original device. The services I expected to respond would no longer receive the packets, they’d go to the wrong machine and it of course wouldn’t respond to requests for services it was not running.
That’s how you end up in this situation of “works for a bit then stops responding”
Either this or a firewall issue.
Dunno how helpful this is but when I’ve had this problem in the past it was an IP conflict. Are you setting static IPs or are you using DHCP reserves for everything?
Check your ARP table before and after the ping? Might also be interesting to tcpdump (for the device’s address or 0.0.0.0) while doing it.
This is some Schrodinger’s network address shit. The act of observing the network address modifies it so it can be accessed.
Is there an IPSec tunel in the mix? Often, IPSec Phase2 goes down when idle, while Phase1 stays up. Upon traffic, Phase2 is brought up again, with a delay.
I usually work around this with a crontab on one of the remote servers that sends a single ping packet every minute to a local server, and pipe any output to /dev/null.
I can guess at some things but let me first start with what I think is happening:
You have a gateway set. Your device sends a broadcast arp message asking 'who has ip ’ and the device with that ip is supposed to send back ‘me with this mac address!’.
That device is either sending it so slowly that your machine says ‘I can’t go past the gateway, the gateway isn’t responding’ which in your error message is no route to host.
Assuming that you have no custom manual network route in play.
So things that can cause that are usually link layer and layer two issues and sometimes duplicate IPs. Two devices with the gateway ip.
You should watch your mac address table and arp table (arp a) and watch if the router gateway disappears or changes Mac addresses.
I had an issue where my docker networks started overlapping my network addresses and I had some issues to connect to the machine etc. so perhaps check to make sure your docker addresses don’t overlap your home network range somehow. Also keep the subnet mask in mind as the range might be bigger than what the first few numbers indicate.
