TCP KeepAlive: Handle USBIP Client interrupts about usbip-win

[yelfarri]

@Oxalin the keepalive (heartbeat) was completed I see the keepalive message when using Wireshark. Isn't keepAlive a part of the TCPIP or from the application? Since we are getting the keepAlive messages does that mean that the configuration of time and interval in application side is wrong?

[Oxalin]

It can be set system wide and per application / socket. From my quick overlook at usbipd, the application is not setting any customized Keep-Alive options, thus relying on the system's default values.

[yelfarri]

I understand but I m talking about usbip client side. I Found in code that usbip has a keepalive function call in "usbip_network.c" line 206. If I remember correctly this call does not set time and intervals. It might be using system default values.
Is "usbip_network.c" shared between usbipd server and usbip client?

[Oxalin]

Yes, indeed. setsockopt() could be called to set the IPPROTO_TCP socket options (as described here https://docs.microsoft.com/en-us/windows/desktop/WinSock/ipproto-tcp-socket-options).However, many options are only available under Windows 10 (1709 and up), even though there are commonly available under Linux...

Thus, I would be tempted to add a heartbeat (keepalive) at the application level. This way, we will be able to support a decent heartbeat in any Windows version. That's why the usbip_net_set_keepalive() doesn't set any IPPROTO_TCP socket options: it was supposed to be added under usbip_net_tcp_connect(). See line 259.

[cezuni]

@Oxalin : I agree that application-level heartbeat can help to detect abortive disconnection more quickly. But it's impossible without usbip protocol modification. I think that it's a too big task to resolve at usbip-win project level.

[cezanne]

Socket-specific KeepAlive is implemented by be88d56. Application-level heartbeat might be not useful unless a linux implementation follows.