Home » General » Support » *SOLVED* VEs with different subnets
| *SOLVED* VEs with different subnets [message #14542] |
Sun, 01 July 2007 23:25  |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
Hello,
Since my local tests with OpenVZ were working perfectly, I've installed yesterday a few boxes in production running OpenVZ in order to deploy my services and to replace my current Xen architecture.
So far so good.... up to the network part... and I'm seriously about to hang myself, I've been stuck for like 7 hours straight.... so any kind of help would be GREATLY appreciated 
Here's the deal :
I want to reproduce a similar network topology I had with Xen which was to put all the HN (dom0) inside a private subnet 10.1.0.0/16, outside of Internet, with a gateway 10.1.1.1 providing internet access to them, to save a few IPs and to isolate the HNs from public access.
And to give to the VEs (domU), IPs from a public routable subnet provided by my ISP, and the VEs to directly connect to the gateway of the ISP (their router) as their default route.
With Xen it wasn't a problem, I just had to create an interface for the domU, to set it up like a normal box directly connected to the Internet with its public IP and the ISP gateway, and that's all....
From what I've read through the documentation and the wiki, it seems it is not possible in that way with OpenVZ... at least not with the venet device....
Am I right ?
I was quite disappointed, until I found the veth device in the wiki.... which seems to be the real thing for me and to act the way that I want
[I don't care about the possible security issue, since I am not a reseller dealing with evil customers, and I'm running in a fully trusted environment]
But still, it's been a few hours that I've tried to make the thing work and I can't :/
I've followed the examples here
http://wiki.openvz.org/Virtual_Ethernet_device
So far, I have the veth101.0 device on the HN for my VE with ID 101, I also have the eth0 on the VE.
I've put all the sysctl calls explained on the wiki, the route add on both the HN and the VE, etc... but still... all I can do is :
ping the VE from the HN with it's Internet IP.
ping the HN and its subnet
and nothing more  (no one can ping the VE from the outside except the HN, and I don't have Internet from the VE)
Is what I want to do impossible with the current way OpenVZ is built ?
Has someone succeeded ? If yes, is the wiki complete ? And someone could share its configuration if it has a similar setup running ?
Thanks a lot in advance !
Ugo
[Updated on: Wed, 11 July 2007 09:53] by Moderator Report message to a moderator |
|
|
|
|
|
|
|
| Re: VEs with different subnets [message #14551 is a reply to message #14548] |
Mon, 02 July 2007 08:31  |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
| dev wrote on Mon, 02 July 2007 03:29 |
have you read the following article aboute multiple devices/GW and routed venet:
http://wiki.openvz.org/Source_based_routing
?
The typical configuration of secured OVZ looks the following way:
HN VEs
----------------------------------- ---------
eth0 (assigned local IP address)
eth1 (assigned global IP address) VEs are routed
and set as default default GW through eth1, since
it's default GW.
the only difference is that eth1 still has global IP assigned to HN. But this is a plus on the other hand, since HN often needs an access to internet for example for upgrades.
In this case you simple need to put "eth1" to /etc/vz/conf variable:
17 # The name of the device whose ip address will be used as source ip for VE.
18 # By default automatically assigned.
19 #VE_ROUTE_SRC_DEV="eth1"
|
Hello,
Thanks for you answer.
Yes, I did read the wiki "Source based routing", but it implies to have a new device set up for each new route that I want to give access to, to my VEs, right ?
So it means to assign a global IP to each HN ?
As I said, I would like to avoid "wasting" global IPs on the HNs, because I only have 64 IP and about 20 HNs... so it would mean only 44 IPs left for the VEs, which is not a lot in my case.
That's why I've built the explained configuration with Xen... well and of course the HN (dom0) have Internet access from their private subnet, because 10.1.1.1 acts as a NAT Gateway to them, so I can do upgrades, etc... with the HNs.
Ugo
|
|
|
|
|
|
|
|
|
|
| Re: VEs with different subnets [message #14558 is a reply to message #14554] |
Mon, 02 July 2007 11:32 |
khorenko
Messages: 533
Registered: January 2006
Location: Moscow, Russia
|
Senior Member |
|
|
Hello, Ugo.
just a small question - did you try to use venet and just configure node as simple as you did on your Xen node?
i mean just set a private IP on a Hardware Node (eth0) and set a public IP for a VE? (vzctl set $VEID --ipadd $PUBLIC_IP --save)
i just checked such a simple configuration and it seems to me - it works fine.
Could you please check it? If it won't work, could you please post
- ip a l
- ip r l
- vzctl exec $VEID ip a l
- vzctl exec $VEID ip r l
- the command that didn't work?
If your problem is solved - please, report it!
It's even more important than reporting the problem itself...
|
|
|
|
|
|
| Re: VEs with different subnets [message #14560 is a reply to message #14559] |
Mon, 02 July 2007 12:19 |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
titan:/var/lib/vz/template/cache# vzctl create 101 --ostemplate fedora-core-6-i686-default --config vps.basic
Creating VE private area (fedora-core-6-i686-default)
Performing postcreate actions
VE private area was created
titan:/var/lib/vz/template/cache# vzctl set 101 --onboot yes --save
Saved parameters for VE 101
titan:/var/lib/vz/template/cache# vzctl set 101 --hostname test --save
Saved parameters for VE 101
titan:/var/lib/vz/template/cache# vzctl set 101 --ipadd 87.98.196.135 --save
Saved parameters for VE 101
titan:/var/lib/vz/template/cache# vzctl set 101 --nameserver 194.2.0.50 --save
that's how I've created the VZ (just in case).
87.98.196.135, being the public IP address I want to try to assign to the VE.
Here are the parameters that would do the trick if the VE had a "classical" ethernet device :
address 87.98.196.135
netmask 255.255.255.192
network 87.98.196.128
broadcast 87.98.196.191
gateway 87.98.196.129
Here's the output of the commands you've requested. (titan is the HN, test the VE)
titan:~# ip a l
2: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop
link/ether 00:00:00:00:00:00 brd ff:ff:ff:ff:ff:ff
4: lo: <LOOPBACK,UP,10000> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
6: eth0: <BROADCAST,MULTICAST,UP,10000> mtu 1500 qdisc pfifo_fast qlen 1000
link/ether 00:50:70:26:53:9d brd ff:ff:ff:ff:ff:ff
inet 10.1.1.15/16 brd 10.1.255.255 scope global eth0
inet6 fe80::250:70ff:fe26:539d/64 scope link
valid_lft forever preferred_lft forever
8: dummy0: <BROADCAST,NOARP> mtu 1500 qdisc noop
link/ether 12:62:b8:e1:94:93 brd ff:ff:ff:ff:ff:ff
10: teql0: <NOARP> mtu 1500 qdisc noop qlen 100
link/void
12: tunl0: <NOARP> mtu 1480 qdisc noop
link/ipip 0.0.0.0 brd 0.0.0.0
14: gre0: <NOARP> mtu 1476 qdisc noop
link/gre 0.0.0.0 brd 0.0.0.0
16: sit0: <NOARP> mtu 1480 qdisc noop
link/sit 0.0.0.0 brd 0.0.0.0
18: ip6tnl0: <NOARP> mtu 1460 qdisc noop
link/tunnel6 00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00 brd 00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00
1: venet0: <BROADCAST,POINTOPOINT,NOARP,UP,10000> mtu 1500 qdisc noqueue
link/void
titan:~#
titan:~# ip r l
87.98.196.135 dev venet0 scope link
10.1.0.0/16 dev eth0 proto kernel scope link src 10.1.1.15
default via 10.1.1.1 dev eth0
titan:~#
(10.1.1.1 being my gateway providing NAT Internet access for the HNs, and not the ISP one)
titan:~# vzctl exec 101 ip a l
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
3: venet0: <BROADCAST,POINTOPOINT,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue
link/void
inet 127.0.0.1/32 scope host venet0
inet 87.98.196.135/32 brd 87.98.196.135 scope global venet0:0
titan:~#
titan:~# vzctl exec 101 ip r l
192.0.2.0/24 dev venet0 scope host
169.254.0.0/16 dev venet0 scope link
default via 192.0.2.1 dev venet0
titan:~#
(I really don't get the 192.0.2.0/24 that I've just seen from the route....)
I can ping from the HN and the VE can ping the HN
titan:~# ping 87.98.196.135
PING 87.98.196.135 (87.98.196.135) 56(84) bytes of data.
64 bytes from 87.98.196.135: icmp_seq=1 ttl=64 time=0.063 ms
64 bytes from 87.98.196.135: icmp_seq=2 ttl=64 time=0.017 ms
[root@test /]# ping 10.1.1.15
PING 10.1.1.15 (10.1.1.15) 56(84) bytes of data.
64 bytes from 10.1.1.15: icmp_seq=1 ttl=64 time=0.061 ms
64 bytes from 10.1.1.15: icmp_seq=2 ttl=64 time=0.017 ms
But nothing else, but it's quite logical since I can't set the gateway.... and I don't know how to do it with the venet device....
Just in case, here's the version of the vzctl program
titan:~# vzctl
vzctl version 3.0.16-5dso1
Copyright (C) 2000-2007 SWsoft.
This program may be distributed under the terms of the GNU GPL License.
Thanks a lot.
Ugo
|
|
|
|
| Re: VEs with different subnets [message #14577 is a reply to message #14560] |
Mon, 02 July 2007 16:08 |
khorenko
Messages: 533
Registered: January 2006
Location: Moscow, Russia
|
Senior Member |
|
|
Thank you for the logs.
yes, my suggestion implied that 10.1.1.1 will route the packets from VEs also as it does for Hardware Nodes.
Well, if you want to set up another gateway for VEs then just try source base routing as dev already suggested.
# /sbin/ip rule add from 87.98.196.135 table 10
# /sbin/ip route add default dev eth0 via 87.98.196.129 table 10
This should do the trick i suppose. (you won't be forced to do this manually each time, when you get the commands consequence required to get it work, you can create a script with these commands).
If it won't work again, please, let me know, we'll think about the reasons. Or just provide an access to the node, it will take much less time to undestand and eliminate the problem.
| ugo123 wrote on Mon, 02 July 2007 16:19 |
titan:~# vzctl exec 101 ip r l
192.0.2.0/24 dev venet0 scope host
169.254.0.0/16 dev venet0 scope link
default via 192.0.2.1 dev venet0
titan:~#
(I really don't get the 192.0.2.0/24 that I've just seen from the route....)
|
This IP 192.0.2.1 is fake one, it's not used, venet is just a point-to-point conenction, so all the packets got to the venet0 inside a VE appeared on the venet0 interface on a Hardware Node and are routed further according to the rules on Hardware Node.
If your problem is solved - please, report it!
It's even more important than reporting the problem itself...
|
|
|
|
| Re: VEs with different subnets [message #14578 is a reply to message #14577] |
Mon, 02 July 2007 17:17 |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
| finist wrote on Mon, 02 July 2007 12:08 |
Thank you for the logs.
yes, my suggestion implied that 10.1.1.1 will route the packets from VEs also as it does for Hardware Nodes.
Well, if you want to set up another gateway for VEs then just try source base routing as dev already suggested.
# /sbin/ip rule add from 87.98.196.135 table 10
# /sbin/ip route add default dev eth0 via 87.98.196.129 table 10
This should do the trick i suppose. (you won't be forced to do this manually each time, when you get the commands consequence required to get it work, you can create a script with these commands).
If it won't work again, please, let me know, we'll think about the reasons. Or just provide an access to the node, it will take much less time to undestand and eliminate the problem.
| ugo123 wrote on Mon, 02 July 2007 16:19 |
titan:~# vzctl exec 101 ip r l
192.0.2.0/24 dev venet0 scope host
169.254.0.0/16 dev venet0 scope link
default via 192.0.2.1 dev venet0
titan:~#
(I really don't get the 192.0.2.0/24 that I've just seen from the route....)
|
This IP 192.0.2.1 is fake one, it's not used, venet is just a point-to-point conenction, so all the packets got to the venet0 inside a VE appeared on the venet0 interface on a Hardware Node and are routed further according to the rules on Hardware Node.
|
Ok, thank you both again for you support (at least with Xen I sure had the network, but not the support and the people behind  ) hahaha
Yes I don't want to use 10.1.1.1 as the gateway for the VE....
Because it would be a Single Point Of Failure for the whole network.... whereas the gateway of my ISP is an huge backbone and is fully redondant in many ways...
But it's perfectly acceptable for providing Internet access to the HN (just for the upgrades)
I'm 99% sure that I've tried what you're suggesting, and that it blocked the network of the HN where I've tried that. (as I told dev, I've already read and tried the Source Based Routing)
But I will try again and tell you if it works or not with your EXACT same lines.... but I actually can't test it right now because I've locked the network of one of the HN by wrongfully trying to add the veth device and the eth device to a same bridge (it locked the network too).
So now I need to go to the datacenter in order to reboot all of my hard failures
I will tell you ASAP (in about 2h) if it works or not with the Source Based Routing.
Thanks again.
Ugo
|
|
|
|
| Re: VEs with different subnets [message #14584 is a reply to message #14542] |
Mon, 02 July 2007 20:13 |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
Ok, now I remember why source routing wasn't working :
mercury:~# /sbin/ip rule add from 87.98.196.137 table 10
mercury:~# /sbin/ip route add default dev eth0 via 87.98.196.129 table 10
RTNETLINK answers: Network is unreachable
mercury:~#
mercury:~# ping 87.98.196.129
PING 87.98.196.129 (87.98.196.129) 56(84) bytes of data.
64 bytes from 87.98.196.129: icmp_seq=1 ttl=254 time=0.845 ms
64 bytes from 87.98.196.129: icmp_seq=2 ttl=254 time=0.654 ms
64 bytes from 87.98.196.129: icmp_seq=3 ttl=254 time=0.724 ms
--- 87.98.196.129 ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 1998ms
rtt min/avg/max/mdev = 0.654/0.741/0.845/0.078 ms
mercury:~#
mercury:~# vzctl exec 101 ip a l
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
3: venet0: <BROADCAST,POINTOPOINT,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue
link/void
inet 127.0.0.1/32 scope host venet0
inet 87.98.196.137/32 brd 87.98.196.137 scope global venet0:0
mercury:~#
(yes I've done this on another HN and another IP for the VE, but this is EXACTLY the same network, configuration, etc...)
|
|
|
|
| Re: VEs with different subnets [message #14585 is a reply to message #14542] |
Mon, 02 July 2007 21:19 |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
And I have to remember that I only have one interface and my goal is actually not to waste a public IP on the HN since they don't need one at all, they just need an internet access for the upgrade, and that's where my NAT gateway 10.1.1.1 takes place to save some IPs.
After reading the route manual, that's probably why I've got the "RTNETLINK answers: Network is unreachable" message because the program doesn't see any DIRECT interface that knows how to reach the ISP's real gateway.
Actually, after thinking about it.... I don't see how this is do-able at all with "classical routing", because even if we succeed, the ISP will surely filters the "martians" source IP (10.1.x.x) from its routing table, so the packets will be null-routed anyway... and the thing will be pointless in that case
So that's why I think that I really need the veth to get the thing working...because from what I see the veth looks like a lot like the things I was able to do with Xen networking part.... but still I can't get it work (at least by by doing the things explained on the wiki)
Thanks,
Ugo
[Updated on: Mon, 02 July 2007 21:20] Report message to a moderator |
|
|
|
| Re: VEs with different subnets [message #14592 is a reply to message #14584] |
Tue, 03 July 2007 07:20 |
khorenko
Messages: 533
Registered: January 2006
Location: Moscow, Russia
|
Senior Member |
|
|
| ugo123 wrote on Tue, 03 July 2007 00:13 |
Ok, now I remember why source routing wasn't working :
mercury:~# /sbin/ip rule add from 87.98.196.137 table 10
mercury:~# /sbin/ip route add default dev eth0 via 87.98.196.129 table 10
RTNETLINK answers: Network is unreachable
mercury:~#
|
This is simply means that Hardware Node does not know how to get to 87.98.196.129. Could you please try again with:
# ip route add 87.98.196.128/26 dev eth0
# ip rule add from 87.98.196.137 table 10
# ip route add default dev eth0 via 87.98.196.129 table 10
This should eliminate the "Network is unreachable" error.
If your problem is solved - please, report it!
It's even more important than reporting the problem itself...
|
|
|
|
|
|
| Re: VEs with different subnets [message #14599 is a reply to message #14597] |
Tue, 03 July 2007 09:26 |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
mercury:~# ip r l
87.98.196.137 dev venet0 scope link src 10.1.1.14
87.98.196.128/26 dev eth0 scope link
10.1.0.0/16 dev eth0 proto kernel scope link src 10.1.1.14
default via 10.1.1.1 dev eth0
mercury:~# ip a l
2: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop
link/ether 00:00:00:00:00:00 brd ff:ff:ff:ff:ff:ff
4: lo: <LOOPBACK,UP,10000> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
6: eth0: <BROADCAST,MULTICAST,UP,10000> mtu 1500 qdisc pfifo_fast qlen 1000
link/ether 00:50:70:26:57:37 brd ff:ff:ff:ff:ff:ff
inet 10.1.1.14/16 brd 10.1.255.255 scope global eth0
inet6 fe80::250:70ff:fe26:5737/64 scope link
valid_lft forever preferred_lft forever
8: dummy0: <BROADCAST,NOARP> mtu 1500 qdisc noop
link/ether e6:1f:aa:72:10:de brd ff:ff:ff:ff:ff:ff
10: teql0: <NOARP> mtu 1500 qdisc noop qlen 100
link/void
12: tunl0: <NOARP> mtu 1480 qdisc noop
link/ipip 0.0.0.0 brd 0.0.0.0
14: gre0: <NOARP> mtu 1476 qdisc noop
link/gre 0.0.0.0 brd 0.0.0.0
16: sit0: <NOARP> mtu 1480 qdisc noop
link/sit 0.0.0.0 brd 0.0.0.0
18: ip6tnl0: <NOARP> mtu 1460 qdisc noop
link/tunnel6 00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00 brd 00:00:00:00:00:00:00:00:00:00:00:00:00:00:00:00
1: venet0: <BROADCAST,POINTOPOINT,NOARP,UP,10000> mtu 1500 qdisc noqueue
link/void
mercury:~#
|
|
|
|
| Re: VEs with different subnets [message #14601 is a reply to message #14599] |
Tue, 03 July 2007 10:28 |
khorenko
Messages: 533
Registered: January 2006
Location: Moscow, Russia
|
Senior Member |
|
|
Is it possible to get an access to the node?
Please, it will be much more simple...
If your problem is solved - please, report it!
It's even more important than reporting the problem itself...
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: VEs with different subnets [message #14743 is a reply to message #14741] |
Mon, 09 July 2007 12:51 |
ugo123
Messages: 22
Registered: June 2007
|
Junior Member |
|
|
| n00b_admin wrote on Mon, 09 July 2007 08:35 |
I know i'm not much of an expert but i'm learning too from your experience...
Why do you need public addresses on the VE's ?
You assign public addresses to the HN's and private to the VE's.. what you need to do that cannot be done because of NAT-ing the VE's ?
You setup the VE's with venet0 and that's all you need to do.
|
Hello,
I need public addresses on the VEs because those VEs are the one who are providing Internet services, and I want one public IP per VE.
I don't want to assign a single public IP to the HN because it's a waste of IP (and being in Europe, we are tight on public IP with the RIPE), it's also useless in my case and it exposes the HN to the Internet (even with a firewall, I don't like the idea).
The HN should be for me the most secured box (because if compromised, everything is going down), and a private subnet is ideal to answer both of my problems : no public IP wasted, impossible to reach from the Internet.
I don't want to do NAT either, because NAT is less than ideal both in terms of performance and configuration, it would be of course okay if I had a single HN and a single IP.
like set the port 25 to my mail_ve, 80 to my web_ve, etc..etc..
But it's not my case and I want a full network capacity on each VE... and to configure each VE the most easy way, like a real box.... and to don't mind any above configuration.... like NATing and so on... so if a HN dies, I can migrate the VE to the HN, launch it again, and it directly works... no configuration involved.
Finally I could have tweaked my main gateway 10.1.1.1 for my internal network to provide a kind of mixed routing to support my case, but it would have meant that the WHOLE network would have relied and transited on a single machine aka a SPoF (Single Point of Failure), whereas my ISP is providing me a nice IP gateway, with full redondancy, heavy reliability, etc..etc....
So I guess for most simple cases or when you can't trust your VE, venet is definitively the way to go....
But when you need to create more complex infrastructure, it has some limitations, it has nothing to do with the way OpenVZ is built, it's just the limitation of Layer 3 IP routing.... you sometimes need to a Level 2 (with MAC addresses and so on) to do more tricky things.
Hope it answers your questions.
Ugo
[Updated on: Mon, 09 July 2007 12:51] Report message to a moderator |
|
|
|
|
|
|
|
Goto Forum:
Current Time: Fri Nov 15 22:12:21 GMT 2024
Total time taken to generate the page: 0.03512 seconds
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
