Wolfram Schlich created CLOUDSTACK-585:
------------------------------------------
Summary: DHCP entry provisioning is broken in the KVM agent
Key: CLOUDSTACK-585
URL: https://issues.apache.org/jira/browse/CLOUDSTACK-585
Project: CloudStack
Issue Type: Bug
Security Level: Public (Anyone can view this level - this is the default.)
Components: Hypervisor Controller, KVM
Environment: CloudStack 3.0.2 KVM agent running on Fedora 14
Reporter: Wolfram Schlich
When adding an instance to a routerVM DHCP configuration, it seems that the KVM
agent calls /usr/lib64/cloud/agent/scripts/network/domr/dhcp_entry.sh with
wrongly constructed command line arguments, making the script fail to add
correct entries (specifically default router, DNS servers and static routes)
for that instance to the routerVM's /etc/dhcphosts.txt + /etc/dhcpopts.txt.
Especially adding a specific default gateway fails, so the routerVM will always
announce itself as the default router, because the correct entry in
/etc/dhcpopts specifying the gateway of the instance's default network as the
gateway is missing. This is especially nasty for non-default/additional
networks of an instance, messing up the default routing.
Examples:
Management server log entry:
2012-11-29 14:36:37,764 DEBUG [resource.virtualnetwork.VirtualRoutingResource]
(agentRequest-Handler-3:null) Executing:
/usr/lib64/cloud/agent/./scripts/network/domr/dhcp_entry.sh -r 169.254.0.122 -v
172.31.2.233 -m 06:b5:88:00:02:30 -n vmname -d 172.31.2.1 -N 172.31.2.201
<END-OF-LINE>
Notice the double spaces before -d and -N (and the extra space at the EOL).
After patching /usr/lib64/cloud/agent/scripts/network/domr/dhcp_entry.sh to do
meaningful logging, it's clear that the script does not get called with "-d",
but " -d" instead (same for -N), so with an extra space before the dash. Thus,
getopts fails to parse/recognize these two arguments correctly and passed empty
values for $dfltrt and $dns to the /root/edithosts.sh being called on the
routerVM.
It's also clear that the CloudStack KVM Java agent calls it with the wrongly
constructed command line, because if a shell would interpret this command line,
it would just ignore the extra spaces itself.
I've not been able to dig it down, but I somehow suspect that one of
./utils/src/com/cloud/utils/script/Script.java:protected String
buildCommandLine(String[] command) { }
./utils/src/com/cloud/utils/script/Script.java:public String
execute(OutputInterpreter interpreter) { }
might mess up building the command line of the command that had been built by
./core/src/com/cloud/agent/resource/virtualnetwork/VirtualRoutingResource.java:protected
synchronized Answer execute (final DhcpEntryCommand cmd) { }
before.
I've not tried 4.0.0 so far, thus I cannot say whether it might be affected or
not.
As a workaround, I've patched dhcp_entry.sh to re-evaluate the positional
parameters using 'set -- ${@}' (will attach a patch, also one for logging).
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
