Hi Daniel, First of all, awfully sorry for replying so late. Unfortunately your reply had gone to the Spam drawer... Also, I'm answering from Gmail's webmail which IIRC only allows for 'quote original post below'. So please forgive me for not following the proper netiquette of 'quote original post above'.> Is net.ipv4.ip_forward set to 0 ? > >I assume you're asking if this is setup on the host and not on the VM's. I've checked the host and it is configured like this: $ sysctl net.ipv4.ip_forward net.ipv4.ip_forward = 1 Should I change it to =0 ? It wouldn't make sense to me if I'd change it to =0 because that way IP Forwarding would be disabled and not enabled as I think enabled is what is needed in this particular case. Also, for this to work fully, your router switch needs to know where to> send traffic for the 10.2.2.1/24 network. IOW, you need to configure > static routes on your router switch for 10.2.2.1 >I see. So, let me get this straight. What you're saying is that in order for WAN/LAN traffic to reach my VM's I need to set up static IP routes on my router - which is the default gateway for the network - to let it know how to route the packets to the VM's, which needs to be setup to go through the KVM host. Right? But, for testing purposes (trying to reach the VM's from the KVM host) I don't need those static routes, right? Because right now I'd be ok if I could reach the VM's from the KVM host and right now I can't. I've covered a similar setup to yours here:> > > https://www.berrange.com/posts/2009/12/13/routed-subnets-without-nat-for-libvirt-managed-virtual-machines-in-fedora/ > >I've looked at your article and I can see that it resembles my setup. But I think I've got that covered. This is what the xml file for the virtual network 'routed' looks like: <network> <name>routed</name> <uuid>970a25f7-29b6-4a6b-b890-f593eae4fc15</uuid> <forward dev="wlo1" mode="route"> <interface dev="wlo1"/> </forward> <bridge name="virbr2" stp="on" delay="0"/> <mac address="52:54:00:bf:35:42"/> <domain name="routed"/> <ip address="10.2.2.1" netmask="255.255.255.0"> <dhcp> <range start="10.2.2.11" end="10.11.22.254"/> </dhcp> </ip> </network> I've got this 'routed' network in use for the virtual nic of my VM's. Weird this is, when I run a net-list command, it comes back empty... $ virsh net-list Name State Autostart Persistent ---------------------------------------- $ This doesn't look ok, right? I mean, running a net-list command it should list all the networks I have set up on this server. But it's coming back empty as if there weren't any networks. I've double checked on Virtual Machine Manager and both the 'default' and 'routed' networks are active and are set up to 'auto start on boot'. Could this somehow be related to my problem? I guess not because the 'default' network is the NAT network that I've used before and it worked and still works when I set the VM's to use it. I'm puzzled with this stuff. Hope you or someone else can put me back on the right track. Thanks in advance. Cheers, Rui Correia On Mon, Jul 20, 2020 at 10:32 AM Daniel P. Berrange <dan@berrange.com> wrote:> On Sun, Jul 19, 2020 at 11:54:06AM +0100, Rui Correia wrote: > > Greetings folks. > > I've setup libvirtd on my manjaro linux laptop. > > Got a couple of VM's running (Win10 and Debian10) through NAT without any > > issues. > > > > This is what the current network diagram looks like and it works fine: > > > > +-----------------------------------+ > > | +---------------------+ | > > | | +----------+ | | > > | | |Win 10 VM | | | > > | | |10.1.1.10 | | | > > | | +----------+ | | > > | Laptop | | | > > | Manjaro | +-------------+ | | > > | 10.0.0.10 | |Debian 10 VM | | | > > +-------->+ | |10.1.1.11 | | | > > | | | +-------------+ | | > > | | |NAT | | > > | | |10.1.1.0/24 | | > > | | +---------------------+ | > > +------------+ | +-----------------------------------+ > > |router | | > > |switch +---+ > > |10.0.0.0/24 | | +---------+ > > +------------+ | |Desktop | > > +-------->+Manjaro | > > |10.0.0.11| > > +---------+ > > > > But now I need the debian machine to be accessible from another host on > the > > lan 10.0.0.0/24 which of course is outside the host. > > That network diagram would look like this: > > > > +-----------------------------------------+ > > | +------------------+ | > > | | +----------+ | | > > | | |Win 10 VM | | | > > | | |10.1.1.10 | | | > > | | NAT +----------+ | | > > | | 10.1.1.0/24 | | > > | +------------------+ | > > | Laptop | > > +-------->+ Manjaro +------------------------+ | > > | | 10.0.0.10 | +-------------+ | | > > | | | |Debian 10 VM | | | > > | | | |10.2.2.10 | | | > > | | | Routed +-------------+ | | > > +------------+ | | | 10.2.2.0/24 | | > > |router | | | +------------------------+ | > > |switch +---+ +-----------------------------------------+ > > |10.0.0.0/24 | | > > +------------+ | > > | > > | +---------+ > > | |Desktop | > > +-------->+Manjaro | > > |10.0.0.11| > > +---------+ > > > > > > So, I've setup a 'routed network' for the Debian 10 VM but it's not > working > > as I would expect. > > The host can ping the Debian VM and the Debian VM can ping the host but > the > > Debian VM cannot ping the router 10.0.0.1 or any ip address on the > internet. > > I've been using Virtual Machine Manager to set everything up. > > And this is how the routed network is configured > > <network connections="1"> > > <name>routed</name> > > <uuid>970a25f7-29b6-4a6b-b890-f593eae4fc15</uuid> > > <forward dev="wlo1" mode="route"> > > <interface dev="wlo1"/> > > </forward> > > <bridge name="virbr2" stp="on" delay="0"/> > > <mac address="52:54:00:bf:35:42"/> > > <domain name="routed"/> > > <ip address="10.2.2.1" netmask="255.255.255.0"> > > <dhcp> > > <range start="10.2.2.2" end="10.2.2.254"/> > > </dhcp> > > </ip> > > </network> > > > > Any idea on what i might be doing wrong? > > Is net.ipv4.ip_forward set to 0 ? > > > Also, for this to work fully, your router switch needs to know where to > send traffic for the 10.2.2.1/24 network. IOW, you need to configure > static routes on your router switch for 10.2.2.1 > > I've covered a similar setup to yours here: > > > https://www.berrange.com/posts/2009/12/13/routed-subnets-without-nat-for-libvirt-managed-virtual-machines-in-fedora/ > > > > Regards, > Daniel > -- > |: https://berrange.com -o- > https://www.flickr.com/photos/dberrange :| > |: https://libvirt.org -o- > https://fstop138.berrange.com :| > |: https://entangle-photo.org -o- > https://www.instagram.com/dberrange :| > >
On Thu, Jul 23, 2020 at 03:34:03PM +0100, Rui Correia wrote:> > Is net.ipv4.ip_forward set to 0 ? > > > > > I assume you're asking if this is setup on the host and not on the VM's. > I've checked the host and it is configured like this: > $ sysctl net.ipv4.ip_forward > net.ipv4.ip_forward = 1This is good.> Should I change it to =0 ? It wouldn't make sense to me if I'd change it to > =0 because that way IP Forwarding would be disabled and not enabled as I > think enabled is what is needed in this particular case.No, don't change it to 0. We need ip_forward enabled as you say.> Also, for this to work fully, your router switch needs to know where to > > send traffic for the 10.2.2.1/24 network. IOW, you need to configure > > static routes on your router switch for 10.2.2.1 > > > > I see. So, let me get this straight. What you're saying is that in order > for WAN/LAN traffic to reach my VM's I need to set up static IP routes on > my router - which is the default gateway for the network - to let it know > how to route the packets to the VM's, which needs to be setup to go through > the KVM host. Right? > > But, for testing purposes (trying to reach the VM's from the KVM host) I > don't need those static routes, right?Correct. The KVM host knows where the 10.2.2.1/24 subnet is - it owns it. The other hosts on your LAN don't know anything about 10.2.2.1/24, so if they try to access VMs on that subnet, traffic will go to the default route, aka your LAN router. It then has to know which KVM host has the 10.2.2.1/24 subnet to send the traffic onwards.> Because right now I'd be ok if I could reach the VM's from the KVM host and > right now I can't.Yep, so that suggests a more fundamental problem with the KVM host config. Since ip_forward is confirmed as set to 1, I feel the most likely problem is something related to the firewall rules. Libvirt will create iptables rules to allow traffic. Tradititionally this would have been sufficient, in iptables all rules are in the single set of global tables. If your OS distro has enabled "nft" to replace iptables though, things become more tricky. In nft world there is no single set of global tables. Any app using nft can define its own top level tables. So while libvirt adds iptables rules to allow traffic, there is the possibility that "nft" rules may none the less deny traffic. In the specific case of distros using "firewalld", libvirt does some further workarounds for this problem. Overall though, I'd be investigating your firewall. It is helpful to add logging rules to your firewall immediately before any REJECT / DROP rules so you can spot packets being dropped. That combined with tcpdump on the TAP devices should help you confirm what is happening to traffic.> > I've covered a similar setup to yours here: > > > > > > https://www.berrange.com/posts/2009/12/13/routed-subnets-without-nat-for-libvirt-managed-virtual-machines-in-fedora/ > > > > > I've looked at your article and I can see that it resembles my setup. But > I think I've got that covered. > This is what the xml file for the virtual network 'routed' looks like: > > <network> > <name>routed</name> > <uuid>970a25f7-29b6-4a6b-b890-f593eae4fc15</uuid> > <forward dev="wlo1" mode="route"> > <interface dev="wlo1"/> > </forward> > <bridge name="virbr2" stp="on" delay="0"/> > <mac address="52:54:00:bf:35:42"/> > <domain name="routed"/> > <ip address="10.2.2.1" netmask="255.255.255.0"> > <dhcp> > <range start="10.2.2.11" end="10.11.22.254"/> > </dhcp> > </ip> > </network> > > I've got this 'routed' network in use for the virtual nic of my VM's. > Weird this is, when I run a net-list command, it comes back empty... > > $ virsh net-list > Name State Autostart Persistent > ---------------------------------------- > > $ > > This doesn't look ok, right? I mean, running a net-list command it should > list all the networks I have set up on this server. But it's coming back > empty as if there weren't any networks. I've double checked on Virtual > Machine Manager and both the 'default' and 'routed' networks are active and > are set up to 'auto start on boot'. > Could this somehow be related to my problem? I guess not because the > 'default' network is the NAT network that I've used before and it worked > and still works when I set the VM's to use it.Often missed is that there are multiple instances of libvirtd. One global instance that runs as root for privileged set ups, and then one per user instance that runs unprivileged. If you run "virsh" as non-root, you'll be querying the per-user instance. virt-manager uses the privileged instance by default. Try 'virsh -c qemu:///system netlist' instead, or simply run virsh as root. Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
On Thu, Jul 23, 2020 at 3:54 PM Daniel P. Berrangé <berrange@redhat.com> wrote:> No, don't change it to 0. We need ip_forward enabled as you say. >That's what I thought. I'm leaving it as it is.> Correct. The KVM host knows where the 10.2.2.1/24 subnet is - it owns > it. The other hosts on your LAN don't know anything about 10.2.2.1/24, > so if they try to access VMs on that subnet, traffic will go to the > default route, aka your LAN router. It then has to know which KVM host > has the 10.2.2.1/24 subnet to send the traffic onwards. >Yep, again that's what I thought. For now I'll be leaving it as it is because right now I just need the host to be able to communicate with the VM's. Yep, so that suggests a more fundamental problem with the KVM host> config. > > Since ip_forward is confirmed as set to 1, I feel the most likely problem > is something related to the firewall rules. > > Libvirt will create iptables rules to allow traffic. Tradititionally > this would have been sufficient, in iptables all rules are in the single > set of global tables. > > If your OS distro has enabled "nft" to replace iptables though, things > become more tricky. In nft world there is no single set of global tables. > Any app using nft can define its own top level tables. > > So while libvirt adds iptables rules to allow traffic, there is the > possibility that "nft" rules may none the less deny traffic. > > In the specific case of distros using "firewalld", libvirt does some > further workarounds for this problem. > > Overall though, I'd be investigating your firewall. >Okay, I think I've understood but how can I tell if my distro has 'nft' enabled? Guess I'll ask down at their IRC channel and see if someone can tell me. Otherwise I think I'm fried because I googled it and I came out empty handed.> It is helpful to add logging rules to your firewall immediately before > any REJECT / DROP rules so you can spot packets being dropped. That > combined with tcpdump on the TAP devices should help you confirm > what is happening to traffic. >I don't have the faintest idea on how to set up logging rules, or worst, how to get a tcpdump on my TAP devices and analyse the dump. I'd use wireshark but I wouldn't know what I'd be doing to analyse the dump with it. Often missed is that there are multiple instances of libvirtd. One global> instance that runs as root for privileged set ups, and then one per > user instance that runs unprivileged. > > If you run "virsh" as non-root, you'll be querying the per-user instance. > > virt-manager uses the privileged instance by default. > > Try 'virsh -c qemu:///system netlist' instead, or simply run > virsh as root. >Got it. Here's the output with sudo: $ sudo virsh net-list [sudo] password for ******: Name State Autostart Persistent -------------------------------------------- default active yes yes routed active yes yes $ $ virsh -c qemu:///system net-list Name State Autostart Persistent -------------------------------------------- default active yes yes routed active yes yes $ This means both network profiles are created, loaded, active and set up for autostart. Thanks for the headsup. I'll ask the Manjaro guys about the nft. Hopefully they'll know if nft is installed and running. Cheers, Rui Correia