| Commit message (Collapse) | Author | Age | Files |
|---|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We use a dedicated, non-routable, IPv4 subnet for IPSec. Furthermore
the subnet is nullrouted in the absence of xfrm lookup (i.e., when there
is no matching IPSec Security Association) to avoid data leaks.
Each host is associated with an IP in that subnet (thus only reachble
within that subnet, either by the host itself or by its IPSec peers).
The peers authenticate each other using RSA public key authentication.
Kernel traps are used to ensure that connections are only established
when traffic is detected between the peers; after 30m of inactivity
(this value needs to be less than the rekeying period) the connection is
brought down and a kernel trap is installed.
|
| |
|
|
|
|
|
|
|
|
| |
Ideally we we should also increase the Diffie-Hellman group size from
2048-bit to 3072-bit, as per ENISA 2014 report.
https://www.enisa.europa.eu/publications/algorithms-key-size-and-parameters-report-2014
But we postpone that for now until we are reasonably certain that older
client won't be left out.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
| |
Interhost communications are protected by stunnel4. The graphs are only
visible on the master itself, and content is generated by Fast CGI.
|
| |
|
|
|
| |
Using client-side data signing/encryption and wrapping inter-host
communication into stunnel.
|
| |
|
|
|
| |
XXX: this is a workaround the CAcert root CAs not being present in
Jessie. In stretch, we would merely install the 'ca-cacert' package.
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
It should be installed on the dom0 instead.
|
| |
|
|
| |
S.M.A.R.T makes little sense for virtual HDDs.
|
| | |
|
| | |
|
| |
|
|
|
| |
For some reason giraff doesn't like IPSec. App-level TLS sessions are
less efficient, but thanks to ansible it still scales well.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
Instead, generate a server certificate for each host (on the machine
itself). Then fetch all these certs locally, and copy them over to each
IPSec peer. That requires more certs to be stored on each machines (n
vs 2), but it can be done automatically, and is easier to deploy.
Note: When adding a new machine to the inventory, one needs to run the
playbook on that machine (to generate the cert and fetch it locally)
first, then on all other machines.
|
| |
|
|
|
|
| |
To avoid low-entropy conditions, see
http://www.issihosts.com/haveged/
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
We use a "master" NTP server, which synchronizes against stratum 1
servers (hence is a stratum 2 itself); all other clients synchronize to
this master server through IPSec.
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
We use a dedicated instance for each role: MDA, MTA out, MX, etc.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
|
|
To run the playbook:
cd ./ansible
ansible-playbook -i vms site.yml
|
