Accessing raw dd images in a Docker Linux container. I was backing up a Linux server of mine the other day and I wanted to have a full backup (along with regular tar.gz backups) of the main disk mounted on the /dev/sda partition. You can backup your partition using dd with a command such as dd if=/dev/sda dd of=/home/archive/disk.img. The file must be valid YAML using the standard Docker Compose format. The file must use the same 'version' header as the main docker-compose.yml file, currently 2.2. When modifying ports or other lists, you can only add new items, not remove existing ones. $ docker pull samalba/hipache $ docker save samalba/hipache glance image-create -visibility public -container-format=docker -disk-format=raw -name samalba/hipache NOTE: The name you provide to glance must match the name by which the image is known to docker. And until they are zeroed, the raw format can't free up the sectors. If you really need that disk space, you may need to destroy and recreate the file, which would be the equivalent of a reinstall of docker. – BMitchJul 31 '18 at 16:21 Do you think the information Ortomala Lokni posted would cause that to shrink? Version: '2' services: redis-master: image: k8s.gcr.io/redis:e2e ports: - '6379' redis-slave: image: gcr.io/googlesamples/gb-redisslave:v1 ports: - '6379.
August 27th, 2017Somewhat following on from my previous post about running containers in non-root environments I’ve been spending some more time reading up on Capabilities, so thought it would be worth making some notes.
What are Capabilities?
Linux capabilities have been around in the kernel for some time. The idea is to break up the monolithic root privilege that Linux systems have had, so that smaller more specific privileges can be provided where they’re required. This helps reduce the risk that by compromising a single process on a host an attacker is able to fully compromise it.
One point to make note of is, that capabilities are only needed to carry out privileged actions on a host. If your process only needs to carry out actions that an ordinary user could without the use of sudo, su or setuid root binaries, then your process doesn’t need any capabilities assigned to it.
To provide a concrete example, take the ping program which ships with most Linux disitributions. Traditionally this program has been setuid root due to the fact that it needs to send raw network packets and this privilege is not availble to ordinary users. With a capability aware system this can be broken down and only the CAP_NET_RAW privilege can be assigned to the file. This means that an attacker who was able to compromise the ping binary, would only get a small additional level of privilege and not full access to the host, as might have been possible when it was setuid root.
Practical use of capabilities
So how do we actually manipulate capabilites on a Linux system? The most basic way of handing this (without writing custom code) is to use the getcap and setcap binaries which come with the libcap2-bin package on debian derived systems.
If you use getcap on a file which has capabilities, you’ll see something like this
We can see here that the arping file has cap_net_raw with +ep at the end of it, so what does that mean. The e here refers to the effective capability of the file and the p to the permitted capability. Effectively for file capabilities the effective flag is needed where the binary isn’t “capability aware” i.e. it’s not written with capabilities in mind (which is usually the case). For practical purposes if you’re assigning capabilities to files, you’ll use +ep most of the time.
So if you want to assign a capability, for example to apply cap_net_raw to an nmap binary
It’s important to note that you can’t set capabilities on symlinks, it has to be the binary, and also you can’t set capabilities on shell scripts (well unless you have a super-recent kernel)
Some More background - Inheritable and Bounded
If you look at capability sets for files and processes, you’ll run across two additional terms which bear looking at, Inheritable and Bounded.
Inheritable capabilites are capabilites that can be passed from one program to another.
Bounded capabilities are, to quote the Man page for capabilities Torchlight 2 item codes.
The capability bounding set acts as a limiting superset for the capabilities that a thread can add to its inheritable set using capset(2).
So they restrict which capabilities can be inherited by a process.
Back to the practical - Auditing capabilities
This is all well and good, but how do we audit capabilities?
there’s a number of ways of reviewing what capabilities a process or file has got. From a low-level perspective, we can review the contents of /proc/[pid]/status. This will contain some information that looks like this :-
This set was for a user level process (using the command cat /proc/self/status). As you can see the CapPrm and CapEff are both all zero’s indicating that I don’t have any capabilities and assigned.
If I then switch to a root user using sudo bash and run the same command, I get the following
which is quite the difference, here CapPrm and CapEff have a lot more content as I’m a privileged user.
If we try the same in a Docker process using the command docker run alpine:latest cat /proc/self/status we get
which is quite different. In this container we were running as root, so you might have guessed that we’d have the same permissions as we did in the root shell before. However as Docker limits the available default permissions we don’t get as much.
Interpreting capabilities
Of course these long hex strings aren’t exactly the most friendly way of viewing capabilities. Luckily there are ways of making this a bit more readable. if we use capsh (which comes with libcap2-bin on debian derived systems) we can work out what’s meant here.
Running capsh --decode=0000003fffffffff returns
So that shows that our root shell run outside basically had all the capabilities. If we run capsh --decode=00000000a80425fb we can see what Docker provides by default
which corresponds to the list in the source code.
Docker.raw File Size
Dangerous Capabilities
So which of these capabilities are a concern from a security perspective? Well to an extent that’s going to depend on how you’re using them. However there are some good starting points to look at, firstly there’s this post from the grsecurity forum that goes into the risks of allowing various capabilities. You can also look at the default list of capabilities that Docker allows (link above), as the ones they block are things that have been determined as dangerous in the context of containers.
Other Utilities
There are some other utilities which are handy for doing things like auditing capabilities. the libcap-ng-utils package has the very handy filecap and pscap programs which can be used to review capapbilties on all files and all processes on a system by default. There’s also captest which will review capabilities in the context of the current process.
Docker Raw File
Also if you’re running containers and want a nice quick way to assess capabilities amongst other things, you could use Jessie Frazelle’s amicontained
Capabilities and Containers
So what has all this to do with Containers? Well it’s worth noting what was mentioned early in this post which is, if you have a container which will run as a non-root user and which has no setuid or setgid root prgrams in it, you should be good to drop all capabilities. This adds another layer of hardening to the container, which can be helpful in preventing container breakout issues.
If you’re running with root containers, then it’s well worth reviewing the default list of capabilities that is provided by your container runtime an ensuring that you’re happy that these are needed.
Specifically there are ones like CAP_NET_RAW in the default docker set which could be dangerous (see here for more details)
Docker Raw Format Example
Capability Gotcha’s
There are some gotcha’s to be aware of when using capabilities. First up is that, to use file capabilities, the filesystem you’re running from needs have extended attribute (xattr) support. A notable exception here is some versions of aufs that ship with some versions Debian and Ubuntu. This can impact Docker installs, as they’ll use aufs by default.
Another one is that where you’re manipulating files you need to make sure that the tools you’re using understand capabilities. For example when backing up files with tar, you need to use the following switches to make it all work.
In practice for tar you’ll likely want to use --xattrs and --xatttrs-include=security.capability to make backups of files with capabilities.