I think the dhcp client on the R7800 has a bug in it that's causing the router to stop all traffic through the wan for 5-10 seconds when renewing a lease.
To try and track it down I've been looking at the code of several dhcp clients I could find online (the one in the Netgear GPL release, the busybox version of udhcp, the isc one, dhcpcd, maybe even some I've forgotten about) and trying to see differences in their behavior and what would cause the symptom I've been seeing.
Right now my best guess is that the arpping implementation in Netgear's hacked up fork of udhcp has a bug in it where, upon renewing its lease on the ip it sends out a ping to see if that ip is already in use by anyone and then gets a false positive so it rejects the lease and starts the whole process over again.
I also have a suspicion that the false positive is caused by the router getting back a response to its ping that says "yes someone is using that IP already" and the one using it is the router itself. The reason I suspect that is that several of the other clients have code in there to check for that but the Netgear implementation does not. There's also a "fixme" note in the Netgear code that looks like it's reminding themselves to implement the check in question.
To test and further diagnose the issue, I've been trying to build the Voxel firmware (and also the GPL fw, and the latest build of OpenWRT) but haven't had any success getting it working. I think the tools and toolchain are pretty finicky. I was hoping I could at least cross compile my own version of the client to debug with but I can't without the prerequisites. I was following the OpenWRT documentation to try to get my environment right.
Has anyone else successfully built any of these firmwares? Do you have any tips?
To try and track it down I've been looking at the code of several dhcp clients I could find online (the one in the Netgear GPL release, the busybox version of udhcp, the isc one, dhcpcd, maybe even some I've forgotten about) and trying to see differences in their behavior and what would cause the symptom I've been seeing.
Right now my best guess is that the arpping implementation in Netgear's hacked up fork of udhcp has a bug in it where, upon renewing its lease on the ip it sends out a ping to see if that ip is already in use by anyone and then gets a false positive so it rejects the lease and starts the whole process over again.
I also have a suspicion that the false positive is caused by the router getting back a response to its ping that says "yes someone is using that IP already" and the one using it is the router itself. The reason I suspect that is that several of the other clients have code in there to check for that but the Netgear implementation does not. There's also a "fixme" note in the Netgear code that looks like it's reminding themselves to implement the check in question.
To test and further diagnose the issue, I've been trying to build the Voxel firmware (and also the GPL fw, and the latest build of OpenWRT) but haven't had any success getting it working. I think the tools and toolchain are pretty finicky. I was hoping I could at least cross compile my own version of the client to debug with but I can't without the prerequisites. I was following the OpenWRT documentation to try to get my environment right.
Has anyone else successfully built any of these firmwares? Do you have any tips?
