Provide docker-based development environments to support 100-500 concurrent workspaces on a single server. The idea is for developers to edit code and perform light compilation or testing within the workspace.
- Serious testing is the duty of a continuous integration server.
- Serious compilation is the duty of a build server.
Please add any issues you find with this software to the upstream wormhole.
Architectural considerations
OVAL vulnerability scan
Test harness
User instructions
Admin instructions
Contributing
License
source: docs/uml.md
Notes:
-
User data lives in a data container and persists across upgrades of the app container.
-
sshd is restrictive:
- ssh host key persists across upgrades of the app container.
-
Weak firewall allows:
- inbound from Internet to wormhole ephemeral TCP ports
- outbound from wormhole to Internet on all ports
-
Strong firewall allows:
- inbound from wormhole IP to internal TCP 22
- no outbound connections other than return traffic for inbound connections
-
Must be capable of running on CoreOS. Therefore no outside dependencies.
-
Internal infrastructure should use appropriate access control mechanisms based on risk evaluation of the wormhole.
The Red Hat Security Response Team provides OVAL definitions for all vulnerabilities (identified by CVE name) that affect RHEL or Fedora. This enables users to perform a vulnerability scan and diagnose whether the system is vulnerable.
The Dockerfile in this repo adds a script to download the latest
OVAL definitions from Red Hat and perform a vulnerability scan
against the image. If the image has one or more known vulnerabilies,
the script exits non-zero, and the docker build
fails.
Implications:
-
We must resolve all known vulnerabilities in order to successfully build an image.
-
The scan is time-dependent as of image build, so we should rebuild the image when Red Hat updates the OVAL feed.
-
The vulnerability scan is distinct from the SCAP secure configuration scan described by the test harness.
It is possible to scan an existing image:
docker run --rm -t jumanjiman/wormhole /usr/sbin/oval-vulnerability-scan.sh
The exact output of the vulnerability scan varies according to the latest Red Hat OVAL feed, but it looks similar to this snapshot from August 2014:
-snip copious checks-
RHSA-2014:1051: flash-plugin security update (Critical)
oval-com.redhat.rhsa-def-20141051
CVE-2014-0538
CVE-2014-0540
CVE-2014-0541
CVE-2014-0542
CVE-2014-0543
CVE-2014-0544
CVE-2014-0545
pass
RHSA-2014:1052: openssl security update (Moderate)
oval-com.redhat.rhsa-def-20141052
CVE-2014-3505
CVE-2014-3506
CVE-2014-3507
CVE-2014-3508
CVE-2014-3509
CVE-2014-3510
CVE-2014-3511
pass
RHSA-2014:1053: openssl security update (Moderate)
oval-com.redhat.rhsa-def-20141053
CVE-2014-0221
CVE-2014-3505
CVE-2014-3506
CVE-2014-3508
CVE-2014-3510
pass
vulnerability scan exit status 0
TODO: Implement some sort of CD system to poll the OVAL feed and rebuild the image on any update. https://github.com/jumanjiman/docker-gocd may be a candidate for the solution.
RSpec documents key behaviors and assures no regressions:
contributor friction
there should not be any
jumanjiman/wormhole
should use correct docker API version
image should be available
image properties
should expose ssh port and only ssh port
should have volume /home/user
should have volume /media/state/etc/ssh
admin scripts
given user handle="booga"
everybody knows pubkey
booga knows privkey
`build.sh $handle "$pubkey"` creates wormhole from 2 containers
"booga-data" is a persistent read-write container
should exist
should be stopped
should be created from busybox
should export /home/user volume read-write
should export /media/state/etc/ssh volume read-write
should not mount any volumes
"booga" is a read-only app container
should exist
should be running
should run unprivileged
should be created from jumanjiman/wormhole
should use volumes from booga-data
should have hostname wormhole.example.com
should be limited to 512 MiB RAM
`docker logs` should show sshd running on sshd port
should expose internal sshd port and only sshd port
should map internal sshd port to an outside ephemeral port
arcanist (phabricator client)
`arc` is in user path
`arc version` is functional
uses bash autocompletion
BZ1099206 (slow test)
home directory should exist
go get should work
user convenience
man -k returns results
locate returns the path for issue.net
shell-shock
should resolve CVE-014-6271
should resolve CVE-014-7169
eiffelstudio
has command-line eiffel compiler in path
has estudio in path
ldc D compiler
compiles a D program (PENDING: https://bugzilla.redhat.com/show_bug.cgi?id=1176361)
locale archive
supports C
supports POSIX
supports en_US.utf8
supports en_US.UTF-8
SCAP secure configuration checks (slow test)
should pass all tests
/etc/securetty should be a zero-size file
prohibited packages
should not have at installed
should not have prelink installed
should not have sudo installed
prohibited commands
should not have the at command
should not have the crond command
should not have the crontab command
should not have the /usr/sbin/prelink command
rhc (openshift client)
`rhc` is in user path
`rhc --version` is functional
sshd config
auth
should use privilege separation
should use pam
should allow pubkeyauthentication
should deny passwordauthentication
should deny gssapiauthentication
should deny kerberosauthentication
should deny challengeresponseauthentication
should use strong, perfect forward secrecy for key exchange
tunnels and forwarding
should deny ssh tunnels
should deny TCP forwarding
should deny X11 forwarding
should deny gateway ports
Common Configuration Enumeration (CCE)
CCE-3660-8 Disable remote ssh from accounts with empty passwords
CCE-3845-5 idle timeout interval should be set appropriately
CCE-4325-7 Disable SSH protocol version 1
CCE-4370-3 Disable SSH host-based authentication
CCE-4387-7 Disable root login via SSH
CCE-4431-3 SSH warning banner should be enabled
CCE-4475-0 Disable emulation of rsh command through sshd
CCE-14061-6 "keep alive" msg count should be set appropriately
CCE-14491-5 Use appropriate ciphers for SSH
CCE-14716-5 Users should not be allowed to set env options
obscurity
should hide patch level
users with interactive shells
should only include "root" and "user"
su
"user" cannot su
Finished in 1 minute 6.35 seconds (files took 0.45472 seconds to load)
61 examples, 0 failures
The OpenSCAP secure configuration test shown above uses a tailoring file to adjust the upstream checks. It expands to this inside the container:
Title gpgcheck Enabled In Main Yum Configuration
Rule ensure_gpgcheck_globally_activated
Result pass
Title gpgcheck Enabled For All Yum Package Repositories
Rule ensure_gpgcheck_never_disabled
Result pass
Title Shared Library Files Have Restrictive Permissions
Rule file_permissions_library_dirs
Result pass
Title Shared Library Files Have Root Ownership
Rule file_ownership_library_dirs
Result pass
Title System Executables Have Restrictive Permissions
Rule file_permissions_binary_dirs
Result pass
Title System Executables Have Root Ownership
Rule file_ownership_binary_dirs
Result pass
Title Direct root Logins Not Allowed
Rule no_direct_root_logins
Result notchecked
Title Virtual Console Root Logins Restricted
Rule securetty_root_login_console_only
Result pass
Title Serial Port Root Logins Restricted
Rule restrict_serial_port_logins
Result pass
Title Only Root Has UID 0
Rule no_uidzero_except_root
Result pass
Title Log In to Accounts With Empty Password Impossible
Rule no_empty_passwords
Result pass
Title Password Hashes For Each Account Shadowed
Rule no_hashes_outside_shadow
Result pass
Title netrc Files Do Not Exist
Rule no_netrc_files
Result pass
Title SSH Root Login Disabled
Rule sshd_disable_root_login
Result pass
Title SSH Access via Empty Passwords Disabled
Rule sshd_disable_empty_passwords
Result pass
Title SSH Idle Timeout Interval Used
Rule sshd_set_idle_timeout
Result pass
Title SSH Client Alive Count Used
Rule sshd_set_keepalive
Result pass
New containers begin life with a git-suitable
~/.bashrc
.
This is only the initial bashrc; you can modify it at any time.
Inside the container, your user account is literally named user
.
That means, with default build options, your prompt inside the container is:
user@wormhole:~$
Connect to your container with info provided by admin:
ssh -i path/to/privkey -p <your port> user@<ip>
Clone this repo, then inspect and optionally modify global.conf
.
When you build a user box for the first time, the build script
copies global.conf
into /etc/wormhole/global.conf
.
This step is optional. The default configuration does not use Duo.
If you want to use Duo Security for multi-factor authentication,
you must edit both global.conf
and login_duo.conf
.
-
When you build your first user box (see below), these two files are copied into
/etc/wormhole/
. -
If you modify
/etc/wormhole/*.conf
, you must restart the user app container(s).
Use the build script with a unique user id (such as github handle) and the user ssh pubkey to create a runtime container:
user=jumanjiman
pubkey="ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEArSm80/8UD/eMolUXU3j6geyNrxthnQxbE4dpwRhXU+F6fbQG+wk9SdWev9NcLLWg9a4zBUSMJUXrrU/8ik3WshSpZpqQary4ZiFFQKgSfYriouchc20S3wwFQZcbOJgH5t5wgGeNaDMzc2GRFhqbuuBiBBF+W5llk0X9CGE1o1iAlyVPAn4UfrJ4//5OXMhYwmU+fO9df3y5Kpn/0SY/lRwWuZeVVIXC+nZcFYXNzPyBVTNEooOXLVXivddtU82jfp65ggTMdLfUafZqia1/smfWQP23lU8F4ySayAOa1lhXXvrGtpxl3lu7vaSvdEg7+F4YaIhxnWZqt769joDraw== pmorgan@github"
./build.sh $user "$pubkey"
A runtime container should be up on a random ssh port:
$ docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
689479673e8e jumanjiman:latest /bin/sh -c /usr/sbin About an hour ago Up About an hour 0.0.0.0:49153->22/tcp jumanjiman
The build script depends on systemd and enables a "wormhole@"
service to start at boot-time for the container you just created and
persists the port as an environment variable in /etc/wormhole/<username>.conf
.
For example: If the username is jumanjiman, you can run
sudo systemctl status wormhole@jumanjiman
to see:
● [email protected] - jumanjiman app container
Loaded: loaded (/etc/systemd/system/[email protected]; enabled)
Active: active (running) since Wed 2014-06-18 14:08:05 UTC; 3h 18min ago
Main PID: 701 (docker)
CGroup: /system.slice/system-wormhole.slice/[email protected]
└─701 /usr/bin/docker run --rm -t -m 512m --volumes-from jumanjiman-data -p 49153:22 -h wormhole.example.com --name jumanjiman...
Jun 18 14:08:05 ip-192-168-254-21 bash[478]: jumanjiman
Jun 18 14:08:05 ip-192-168-254-21 systemd[1]: Started jumanjiman app container.
Jun 18 14:08:09 ip-192-168-254-21 docker[701]: Server listening on 0.0.0.0 port 22.
Jun 18 14:08:09 ip-192-168-254-21 docker[701]: Server listening on :: port 22.
global.conf
limits each app container to 512 MiB memory.
You can override this for a single user in /etc/wormhole/<username>.conf
.
You can view the current limit for a container via the sys filesystem. For example, here is a container that was started with a 1 MiB limit:
$ cat /sys/fs/cgroup/memory/lxc/<hash>/memory.limit_in_bytes
1048576
If a PID inside a container gets killed due to the memory limit,
you can view details in dmesg
output.
Create a local file called jumanjiman-data.tar
with contents of
user data container.
./backup.sh jumanjiman
./restore.sh jumanjiman
Rebuild the jumanjiman/wormhole
image as described above, then...
./upgrade.sh jumanjiman
Backup, then discard both the app and data containers for a user.
./destroy.sh jumanjiman
This image serves as a template for an app container.
You can build the image locally:
script/build
Trigger a rebuild-and-test cycle to get latest updates:
cd wormhole/
date > REBUILD
git add REBUILD
git commit -m 'test build with latest package updates'
# Open pull request.
Fedora kernel on DigitalOcean
has an older LXC implementation that leads to inconsistent builds.
For example, it sometimes builds the base image with bad perms on
/var
and other directories that must be 0755
.