Skip to content

User Namespace โ€‹

Learn AWS hacking from zero to hero with htARTE (HackTricks AWS Red Team Expert)!

Other ways to support HackTricks:

Basic Information โ€‹

A user namespace is a Linux kernel feature that provides isolation of user and group ID mappings, allowing each user namespace to have its own set of user and group IDs. This isolation enables processes running in different user namespaces to have different privileges and ownership, even if they share the same user and group IDs numerically.

User namespaces are particularly useful in containerization, where each container should have its own independent set of user and group IDs, allowing for better security and isolation between containers and the host system.

How it works: โ€‹

  1. When a new user namespace is created, it starts with an empty set of user and group ID mappings. This means that any process running in the new user namespace will initially have no privileges outside of the namespace.
  2. ID mappings can be established between the user and group IDs in the new namespace and those in the parent (or host) namespace. This allows processes in the new namespace to have privileges and ownership corresponding to user and group IDs in the parent namespace. However, the ID mappings can be restricted to specific ranges and subsets of IDs, allowing for fine-grained control over the privileges granted to processes in the new namespace.
  3. Within a user namespace, processes can have full root privileges (UID 0) for operations inside the namespace, while still having limited privileges outside the namespace. This allows containers to run with root-like capabilities within their own namespace without having full root privileges on the host system.
  4. Processes can move between namespaces using the setns() system call or create new namespaces using the unshare() or clone() system calls with the CLONE_NEWUSER flag. When a process moves to a new namespace or creates one, it will start using the user and group ID mappings associated with that namespace.

Lab: โ€‹

Create different Namespaces โ€‹

CLI โ€‹

bash
sudo unshare -U [--mount-proc] /bin/bash

By mounting a new instance of the /proc filesystem if you use the param --mount-proc, you ensure that the new mount namespace has an accurate and isolated view of the process information specific to that namespace.

Error: bash: fork: Cannot allocate memory

When unshare is executed without the -f option, an error is encountered due to the way Linux handles new PID (Process ID) namespaces. The key details and the solution are outlined below:

  1. Problem Explanation:

    • The Linux kernel allows a process to create new namespaces using the unshare system call. However, the process that initiates the creation of a new PID namespace (referred to as the "unshare" process) does not enter the new namespace; only its child processes do.
    • Running %unshare -p /bin/bash% starts /bin/bash in the same process as unshare. Consequently, /bin/bash and its child processes are in the original PID namespace.
    • The first child process of /bin/bash in the new namespace becomes PID 1. When this process exits, it triggers the cleanup of the namespace if there are no other processes, as PID 1 has the special role of adopting orphan processes. The Linux kernel will then disable PID allocation in that namespace.
  2. Consequence:

    • The exit of PID 1 in a new namespace leads to the cleaning of the PIDNS_HASH_ADDING flag. This results in the alloc_pid function failing to allocate a new PID when creating a new process, producing the "Cannot allocate memory" error.
  3. Solution:

    • The issue can be resolved by using the -f option with unshare. This option makes unshare fork a new process after creating the new PID namespace.
    • Executing %unshare -fp /bin/bash% ensures that the unshare command itself becomes PID 1 in the new namespace. /bin/bash and its child processes are then safely contained within this new namespace, preventing the premature exit of PID 1 and allowing normal PID allocation.

By ensuring that unshare runs with the -f flag, the new PID namespace is correctly maintained, allowing /bin/bash and its sub-processes to operate without encountering the memory allocation error.

Docker โ€‹

bash
docker run -ti --name ubuntu1 -v /usr:/ubuntu1 ubuntu bash

To use user namespace, Docker daemon needs to be started with --userns-remap=default(In ubuntu 14.04, this can be done by modifying /etc/default/docker and then executing sudo service docker restart)

Check which namespace is your process in โ€‹

bash
ls -l /proc/self/ns/user
lrwxrwxrwx 1 root root 0 Apr  4 20:57 /proc/self/ns/user -> 'user:[4026531837]'

It's possible to check the user map from the docker container with:

bash
cat /proc/self/uid_map 
         0          0 4294967295  --> Root is root in host
         0     231072      65536  --> Root is 231072 userid in host

Or from the host with:

bash
cat /proc/<pid>/uid_map

Find all User namespaces โ€‹

bash
sudo find /proc -maxdepth 3 -type l -name user -exec readlink {} \; 2>/dev/null | sort -u
# Find the processes with an specific namespace
sudo find /proc -maxdepth 3 -type l -name user -exec ls -l  {} \; 2>/dev/null | grep <ns-number>

Enter inside a User namespace โ€‹

bash
nsenter -U TARGET_PID --pid /bin/bash

Also, you can only enter in another process namespace if you are root. And you cannot enter in other namespace without a descriptor pointing to it (like /proc/self/ns/user).

Create new User namespace (with mappings) โ€‹

bash
unshare -U [--map-user=<uid>|<name>] [--map-group=<gid>|<name>] [--map-root-user] [--map-current-user]
bash
# Container
sudo unshare -U /bin/bash
nobody@ip-172-31-28-169:/home/ubuntu$ #Check how the user is nobody

# From the host
ps -ef | grep bash # The user inside the host is still root, not nobody
root       27756   27755  0 21:11 pts/10   00:00:00 /bin/bash

Recovering Capabilities โ€‹

In the case of user namespaces, when a new user namespace is created, the process that enters the namespace is granted a full set of capabilities within that namespace. These capabilities allow the process to perform privileged operations such as mounting filesystems, creating devices, or changing ownership of files, but only within the context of its user namespace.

For example, when you have the CAP_SYS_ADMIN capability within a user namespace, you can perform operations that typically require this capability, like mounting filesystems, but only within the context of your user namespace. Any operations you perform with this capability won't affect the host system or other namespaces.

โš ๏ธ

Therefore, even if getting a new process inside a new User namespace will give you all the capabilities back (CapEff: 000001ffffffffff), you actually can only use the ones related to the namespace (mount for example) but not every one. So, this on its own is not enough to escape from a Docker container.

bash
# There are the syscalls that are filtered after changing User namespace with:
unshare -UmCpf  bash

Probando: 0x067 . . . Error
Probando: 0x070 . . . Error
Probando: 0x074 . . . Error
Probando: 0x09b . . . Error
Probando: 0x0a3 . . . Error
Probando: 0x0a4 . . . Error
Probando: 0x0a7 . . . Error
Probando: 0x0a8 . . . Error
Probando: 0x0aa . . . Error
Probando: 0x0ab . . . Error
Probando: 0x0af . . . Error
Probando: 0x0b0 . . . Error
Probando: 0x0f6 . . . Error
Probando: 0x12c . . . Error
Probando: 0x130 . . . Error
Probando: 0x139 . . . Error
Probando: 0x140 . . . Error
Probando: 0x141 . . . Error
Probando: 0x143 . . . Error

References โ€‹

Learn AWS hacking from zero to hero with htARTE (HackTricks AWS Red Team Expert)!

Other ways to support HackTricks: