TCP connections from host to guest drop on host network reconfiguration events. -> fixed in SVN

[jglogan]

I have started seeing this behavior after upgrading to 4.3.4. It is 100% repeatable on my system, and has a pretty significant impact on productivity.

The forum article I posted is ​https://forums.virtualbox.org/viewtopic.php?f=8&t=58909.

Reproduction steps:

1.) Start guest VM
2.) ssh into guest (localhost:2322 forwarded to guest:22 in my config)
3.) Perform any of the following operations:
  a.) disable, then enable WiFi (issue occurs on enable, not disable)
  b.) disable VPN on host (happens with IPSecuritas and NetExtender VPNs on my system)
  c.) enable VPN on host

Expected behavior: no impact on ssh connection from host to guest

Actual behavior in 4.2.x: same as expected

Actual behavior in 4.3.4: ssh connection terminated, 'Connection to localhost closed by remote host.'

Config:

Host - MacOS 10.8.5, Guest - Ubuntu 13.04, 3.8.0-34-generic x86_64 Guest network - Paravirtualized network (NAT, natdnshostresolver enabled so that I can resolve VPN hostnames, but I have tested with it disabled and the same issue occurs) WiFi physical network connection

[jglogan]

VBox.log, network events at end of log file.

[vasily Levchenko]

Could you please collect pcap file of your ssh session?

[jglogan]

Guest-side capture, failure at 06:11:04.579

[jglogan]

Host-side capture, fails at 06:16:29.183.

[jglogan]

Guest-side capture.

[jglogan]

I used Wireshark on the host side and tcpdump on the guest side; please let me know if that's not okay.

host.pcapng and guest.2.pcap are captures of the same failure event from the host and guest side, respectively. guest.pcap is a capture of a separate failure, I just forgot to check the "replace attachement with same filename" when I uploaded guest.2.pcap.

Thanks! John

[jglogan]

Just curious, could this be a side effect of this bugfix for 4.3.4?

https://www.alldomusa.eu.org/ticket/12225

[jglogan]

Another datapoint...I downgraded to 4.3.2 and I don't see the problem anymore. My guess is that the change for bug 12225 is causing this.

[vasily Levchenko]

Ah, this isn't a bug it's the feature :) the idea behind this is to enforce guest's dhcp client to request new lease, in case you've changed network or location. Perhaps we should think about flexibility there.

[jglogan]

With a NAT, though, shouldn't I be isolated from what happens on the host network?

Or is the DHCP renewal done mainly to handle changes to DNS servers?

If that's the case, then is it necessary to force new DHCP requests if I'm using one of the host DNS proxy options (e.g. natdnshostresolver)?

jgl

[vasily Levchenko]

Replying to jglogan:

With a NAT, though, shouldn't I be isolated from what happens on the host network?

Or is the DHCP renewal done mainly to handle changes to DNS servers?

right

If that's the case, then is it necessary to force new DHCP requests if I'm using one of the host DNS proxy options (e.g. natdnshostresolver)?

for DNS proxy it isn't require to enforce guest renew dhcp lease. Will check whether we can do anything on this direction.

jgl

[jglogan]

Thanks. It'd be mighty helpful. Right now I lose everything if I start/stop my VPN, or sleep/wake my laptop.

[CrashRoX]

I'm also experiencing this problem. Not sure if its machine specific but I've tried on OS X mountain lion and mavericks. Any VMs that I'm sshed into lose their connection when the computer sleeps. The problem started when I upgraded to VBox 4.3.6, not sure what version I was on before.

[CrashRoX]

I'm also experiencing this problem. Not sure if its machine specific but I've tried on OS X mountain lion and mavericks. Any VMs that I'm sshed into lose their connection when the computer sleeps. The problem started when I upgraded to VBox 4.3.6, not sure what version I was on before.

[remillet]

I also started having this problem after updating to VirtualBox 4.3.6 on OS X Mavericks 10.9.1. Very frustrating.

[lukeasrodgers]

Also having this problem with VirtualBox 4.3.6r91406 on OSX 10.9.1.

If anyone has suggestions for a temporary workaround until a bugfix is landed, that would be most appreciated.

[jglogan]

I've resigned myself to running 4.3.2 for now. Fortunately this has worked well enough for me.

[franky22]

Having trouble with this as well. Makes using virtualbox for development nearly worthless.

[vasily Levchenko]

Could you please verify that build-osx fixes issue for you?

[jglogan]

It works properly for three of these test cases described above, thanks!

Any reason why I should not stay with this build on my dev system until 4.3.8 releases? Or should I downgrade back to 4.3.2 for now?

[vasily Levchenko]

Replying to jglogan:

It works properly for three of these test cases described above, thanks!

Thank for testing and update.

Any reason why I should not stay with this build on my dev system until 4.3.8 releases? Or should I downgrade back to 4.3.2 for now?

There are no such reasons, it's hotfix based on stable branch.

[lukeasrodgers]

The above link to build-osx is broken. I was able to download it for one machine, which fixed the problem, but am still running 4.3.6 on other machines which are still subject to this bug.

[yvess]

I also want to download this test build. Would be nice if the link would be get fixed by someone

[Frank Mehnert]

Here is another test build.

[emptysquare]

Recent test build fixes this issue for me--an ssh session to the virtual machine persists after the host sleeps and wakes.

[toobulkeh]

I'm also having this issue. Can't wait to see the release pushed!

[Frank Mehnert]

Fix is part of VBox 4.3.8.

[mattdw]

I'm still seeing exactly this bug in 4.3.8. I've checked that I'm updated; kextstat shows only v4.3.8 extensions loaded, I've upgraded both virtual box and the vbox tools in my VM, and turning the wifi on and off (my primary interface is by ethernet cable) or sleeping my computer will reliably disconnect all ssh sessions into my VM.

[jglogan]

From the fix description in the release notes: "NAT: transparent handling of host sleep/resume and network configuration changes if the dnsproxy is enabled or if the hostresolver is used"

Are you using DNS proxy or the host DNS resolution options? If not, you may not get transparent handling on sleep/resume.

It's fixed for me; I use host DNS resolution on my laptop VMs.

[mattdw]

Hey jglogan, thanks. Where would I find those settings? I haven't set anything up explicitly. (However, I didn't have this problem in older versions of VirtualBox either.

[mattdw]

Ah, found it:

VBoxManage modifyvm "VM name" --natdnshostresolver1 on

And thanks, that appears to have solved the problem for me. A little disappointed that it was this involved to (IMO) restore previous default behaviour, but never mind. Thanks for your help, jglogan.