Podman
Podman is an alternative to Docker, providing a similar interface. It supports rootless containers and a shim service for docker-compose.
Installation
Podman depends on the netavark package as the default network backend for rootful containers (see podman-network(1)). Netavark depends on aardvark-dns for name resolution among containers in the same network. Support for the alternative network backend (CNI, cni-plugins) is deprecated.
If you want to replace Docker, one can install podman-docker to mimic the docker binary along with man pages.
Unlike Docker, Podman does not require a daemon, but there is one providing an API for services like cockpit via cockpit-podman.
For advanced usage related to building containers see podman-build(1) which is based on Buildah.
Configuration
Configuration files for configuring how containers behave are located at /usr/share/containers/. You must copy necessary files to /etc/containers before editing. To configure the network bridge interface used by Podman, see /etc/cni/net.d/87-podman.conflist.
Registries
By default, no container image registries are configured in Arch Linux [1]. This means unqualified searches like podman search httpd will not work. To make Podman behave like Docker, configure containers-registries.conf(5):
/etc/containers/registries.conf.d/10-unqualified-search-registries.conf
unqualified-search-registries = ["docker.io"]
/etc/containers/registries.conf.d/shortnames.conf.User namespace mode
By default, processes in Podman containers run within the same user namespace as the caller, i.e. containers are not isolated by the user_namespaces(7) feature. This is the behavior of --userns=host, see podman-run(1).
The --userns=auto flag automatically creates a unique user namespace for the container using an empty range of UIDs and GIDs:
- For containers started by root, the
--userns=autoflag requires the user namecontainersto be specified in the/etc/subuidand/etc/subgidfiles with an unused range of IDs. For example:containers:2147483647:2147483648. - For containers started by other users, the user's range from the
/etc/subuidand/etc/subgidfiles will be used. See #Rootless Podman for the necessary configuration.
There are other valid values for the --userns flag, see podman-run(1) for details. The user namespace mode can also be configured in the containers.conf(5) file on a per-system or per-user basis.
Rootless Podman
CONFIG_USER_NS_UNPRIVILEGED) which has some serious security implications, see Security#Sandboxing applications for details.By default, only root is allowed to run containers (or namespaces in kernelspeak). Running rootless Podman improves security as an attacker will not have root privileges over your system, and also allows multiple unprivileged users to run containers on the same machine. See also podman(1) § Rootless mode and the official rootless tutorial (may be outdated).
Enable kernel.unprivileged_userns_clone
First, check the value of kernel.unprivileged_userns_clone by running:
$ sysctl kernel.unprivileged_userns_clone
If it is currently set to 0, enable it by setting 1 via sysctl or a kernel parameter.
Set subuid and subgid
In order for users to run rootless Podman, a subuid(5) and subgid(5) configuration entry must exist for each user that wants to use it. New users created using useradd(8) have these entries by default.
Migration for users created prior to shadow 4.11.1-3
Users created prior to shadow 4.11.1-3 do not have entries in /etc/subuid and /etc/subgid by default. An entry can be created for them using the usermod(8) command or by manually modifying the files.
The following command enables the username user and group to run Podman containers (or other types of containers in that case). It allocates a given range of UIDs and GIDs to the given user and group.
# usermod --add-subuids 100000-165535 --add-subgids 100000-165535 username
The above range for the user username may already be taken by another user as it defines the default range for the first user on the system. If in doubt, first consult the /etc/subuid and /etc/subgid files to find the already reserved ranges.
Workaround for users managed by homed
Homed does not seem to allocate gid and uid entries to its users. To do this manually, run:
# usermod --add-subuids 524288-589823 --add-subgids 524288-589823 username
Or simply edit the following configuration files as root and add these lines
/etc/subuid
username:524288:65536
/etc/subgid
username:524288:65536
This allocates uid and gid range 524288-589823 to the username user. If these ranges are already taken by other users, you need to shift/adjust the ranges accordingly.
You might need to reboot to reflect the changes.
- This is a workaround only, Podman does not seem to support homed officially.
- This is a known issue of systemd-homed.
- Using Docker seems to work (adding the user to the
dockergroup, but it has its own security implications).
Propagate changes to subuid and subgid
Rootless Podman uses a pause process to keep the unprivileged namespaces alive. This prevents any change to the /etc/subuid and /etc/subgid files from being propagated to the rootless containers while the pause process is running. For these changes to be propagated it is necessary to run:
$ podman system migrate
After this, the user/group specified in the above files is able to start and run Podman containers.
Enable native rootless overlays
Previously, it was necessary to use the fuse-overlayfs package for FUSE overlay mounts in a rootless environment. However, modern versions of Podman and Linux kernel support native rootless overlays, which yields better performance.
--userns auto where different UID/GID mappings could potentially be used on each invocation. See the Podman performance guide for details.To migrate from fuse-overlayfs, run the following command (it will unfortunately delete all pulled images):
$ podman system reset
Also make sure that Podman uses the overlay driver and that the mount_program parameter is not defined in containers-storage.conf(5). Follow the instructions in Docker#Enable native overlay diff engine.
To verify that native rootless overlays are enabled, run
$ podman info | grep -i overlay
It should show graphDriverName: overlay and Native Overlay Diff: "true".
Networking
Podman depends on passt, which provides pasta as the default rootless network backend.
An alternative rootless network backend is slirp4netns, which was the default up to Podman 5.
A major difference between the two is outlined in Podman 5.0 breaking changes in detail:
- Pasta by default performs no Network Address Translation (NAT) and copies the ip addresses from your main interface into the container namespace.
The consequences of this change are explained in upstream's Shortcomings of Rootless Podman:
- Since pasta copies the IP address of the main interface, connections to that IP from containers do not work. This means that unless you have more than one interface, inter-container connections cannot be made without explicitly passing a pasta network configuration, either in
containers.confor at runtime.
An example to mimic slirp4netns behavior is given in the "Podman 5.0 breaking changes" blog post:
containers.conf
[network] pasta_options = ["-a", "10.0.2.0", "-n", "24", "-g", "10.0.2.2", "--dns-forward", "10.0.2.3"]
Also, the default rootless networking tool can be selected in containers.conf under the [network] section with default_rootless_network_cmd, which can be set to pasta or slirp4netns. So, if you run into bugs, you can always revert to slirp4netns like so (provided it is installed):
containers.conf
[network] default_rootless_network_cmd = "slirp4netns"
Storage
The configuration for how and where container images and instances are stored takes place in /etc/containers/storage.conf.
$XDG_CONFIG_HOME/containers/storage.conf on a per-user basis.The default overlay driver is well tested and supports reflink copies [3] on filesystems that support it (Btrfs, XFS, ZFS...) [4].
For more information on the available alternatives and other configuration options, see containers-storage.conf(5) § STORAGE_TABLE.
Foreign architectures
Podman is able to run images built for different CPU architecture than the host using the Wikipedia:binfmt_misc system.
To enable it, install qemu-user-static and qemu-user-static-binfmt.
systemd comes with the systemd-binfmt.service service which should enable new rules.
Verify that binfmt rules have been added:
$ ls /proc/sys/fs/binfmt_misc
DOSWin qemu-cris qemu-ppc qemu-sh4eb status qemu-aarch64 qemu-m68k qemu-ppc64 qemu-sparc qemu-alpha qemu-microblaze qemu-riscv64 qemu-sparc32plus qemu-arm qemu-mips qemu-s390x qemu-sparc64 qemu-armeb qemu-mipsel qemu-sh4 register
Podman should now be able to run foreign architecture images. Most commands use the foreign architecture when --arch option is passed.
Example:
# podman run --arch arm64 'docker.io/alpine:latest' arch
aarch64
Docker Compose
Podman has a compose subcommand which is a thin wrapper around a compose provider, either docker-compose or podman-compose. If both are installed, docker-compose takes precedence. You can override this using the PODMAN_COMPOSE_PROVIDER environment variable.
If you want to use docker-compose, you will need to enable the podman.socket user unit and set docker socket environment variable for that user:
$ export DOCKER_HOST=unix://$XDG_RUNTIME_DIR/podman/podman.sock
This is not required when using podman-compose as it will use podman directly.
- If you have enabled buildkit in docker, the integration will not work. You need to disable buildkit by setting the
DOCKER_BUILDKIT=0environment variable. - podman-compose has compatibility issues, e.g. passing of environment variables does not match behaviour of docker-compose.
NVIDIA GPUs
NVIDIA Container Toolkit provides container runtime for NVIDIA GPUs. Install the nvidia-container-toolkit package. It contains a pacman hook that generates the CDI specification for your GPU and saves it in /etc/cdi/nvidia.yaml.
Test the setup:
$ podman run --rm --gpus all archlinux nvidia-smi -L
--userns nomap and --userns auto podman run parameters. [5]
Containers with restart policy
To automatically start containers with a restart policy, enable podman-restart.service.
Quadlet
Quadlet allows to manage Podman containers with systemd.
For rootless Podman, place Quadlet files under one of following directories:
-
$XDG_CONFIG_HOME/containers/systemd/or~/.config/containers/systemd/ -
/etc/containers/systemd/users/UIDfor the user matchingUID -
/etc/containers/systemd/users/for all users
For Podman with root permissions, the directory is /etc/containers/systemd/.
Podman will read Quadlet files with extensions .container, .volume, .network, .kube, .image, and .pod. A corresponding .service file will be generated using systemd.generator(7). The Quadlet files are read during boot or manually by running a daemon-reload.
Quadlet files can also be generated from Podman commands using podletAUR.
For example, here is a command that will run Syncthing container from LinuxServer.io:
$ podman run \
--rm \
--replace \
--label io.containers.autoupdate=registry \
--name syncthing \
--hostname=syncthing \
--uidmap 1000:0:1 \
--uidmap 0:1:1000 \
--uidmap 1001:1001:64536 \
--env PUID=1000 \
--env PGID=1000 \
--env TZ=Etc/UTC \
--publish 127.0.0.1:8384:8384/tcp \
--publish 22000:22000/tcp \
--volume /path/to/syncthing/config:/config \
--volume /path/to/data1:/data1 \
lscr.io/linuxserver/syncthing:latest
To manage it as a systemd service, create the following Quadlet file:
~/.config/containers/systemd/syncthing-lsio.container
[Unit] Description=Syncthing container # Specify the dependencies Wants=network-online.target After=network-online.target nss-lookup.target # If other container depends on this one, use syncthing-lsio.service not syncthing-lsio.container [Container] ContainerName=syncthing Image=lscr.io/linuxserver/syncthing:latest # Enable auto-update container AutoUpdate=registry Volume=/path/to/syncthing/config:/config Volume=/path/to/data1:/data1 HostName=syncthing PublishPort=127.0.0.1:8384:8384/tcp PublishPort=22000:22000/tcp Environment=PUID=1000 Environment=PGID=1000 Environment=TZ=Etc/UTC # UID mapping is needed to run linuxserver.io container as rootless podman. # This will map UID=1000 inside the container to intermediate UID=0. # For rootless podman intermediate UID=0 will be mapped to the UID of current user. UIDMap=1000:0:1 UIDMap=0:1:1000 UIDMap=1001:1001:64536 [Service] Restart=on-failure # Extend Timeout to allow time to pull the image TimeoutStartSec=300 # The [Install] section allows enabling the generated service. [Install] WantedBy=default.target
We can validate the Quadlet file via
$ /usr/lib/podman/quadlet -dryrun -user
Then, reload and start/enable syncthing-lsio.service.
Valid options for the Container section are listed under podman-systemd.unit(5) § Container units [Container]. PodmanArgs= can be used to add other Podman arguments that do not have corresponding file options.
See podman-systemd.unit(5) § EXAMPLES for more examples including Pod, Volume, Network and Image units.
Images
/etc/containers/registries.conf at unqualified-search-registries in the defined order. The following images will always contain the prefix, to allow for configurations without docker.io in the configuration.Arch Linux
The following command pulls the Arch Linux x86_64 image from Docker Hub.
# podman pull docker.io/archlinux
See the Docker Hub page for a full list of available tags, including versions with and without build tools.
See also README.md.
Alpine Linux
Alpine Linux is a popular choice for small container images, especially for software compiled as static binaries. The following command pulls the latest Alpine Linux image from Docker Hub:
# podman pull docker.io/alpine
Alpine Linux uses the musl libc implementation instead of the glibc libc implementation used by most Linux distributions. Because Arch Linux uses glibc, there are a number of functional differences between an Arch Linux host and an Alpine Linux container that can impact the performance and correctness of software. A list of these differences is documented in https://wiki.musl-libc.org/functional-differences-from-glibc.html.
Note that dynamically linked software built on Arch Linux (or any other system using glibc) may have bugs and performance problems when run on Alpine Linux (or any other system using a different libc). See [6], [7] and [8] for examples.
CentOS
The following command pulls the latest CentOS image from Docker Hub:
# podman pull docker.io/centos
See the Docker Hub page for a full list of available tags for each CentOS release.
Debian
The following command pulls the latest Debian image from Docker Hub:
# podman pull docker.io/debian
See the Docker Hub page for a full list of available tags, including both standard and slim versions for each Debian release.
Troubleshooting
Add pause to process
WARN[0000] Failed to add pause process to systemd sandbox cgroup: Process org.freedesktop.systemd1 exited with status 1
Can be solved using: https://github.com/containers/crun/issues/704
# echo +cpu +cpuset +io +memory +pids > /sys/fs/cgroup/cgroup.subtree_control
Containers terminate on shell logout
After logging out from machine, Podman containers are stopped for some users. To prevent that, enable lingering for users running containers.
You can also create user systemd unit as described in podman-auto-update(1) § EXAMPLES.
Error on commit in rootless mode
Error committing the finished image: error adding layer with blob "sha256:02823fca9b5444c196f1f406aa235213254af9909fca270f462e32793e2260d8": Error processing tar file(exit status 1) permitted operation
Check that the storage driver is overlay in the storage configuration.
Error when creating a container with bridge network in rootless mode
If you are using AppArmor you might end up with problems when creating container using a bridge network with the dnsname plugin enabled:
$ podman network create foo
/home/user/.config/cni/net.d/foo.conflist
$ podman run --rm -it --network=foo docker.io/library/alpine:latest ip addr
Error: command rootless-cni-infra [alloc 89398a9315256cb1938075c377275d29c2b6ebdd75a96b5c26051a89541eb928 foo festive_hofstadter ] in container 1f4344bbd1087c892a18bacc35f4fdafbb61106c146952426488bc940a751efe failed with status 1, stdout="", stderr="exit status 3\n"
This can be solved by adding the following lines to /etc/apparmor.d/local/usr.sbin.dnsmasq:
owner /run/user/[0-9]*/containers/cni/dnsname/*/dnsmasq.conf r, owner /run/user/[0-9]*/containers/cni/dnsname/*/addnhosts r, owner /run/user/[0-9]*/containers/cni/dnsname/*/pidfile rw,
And then reloading the AppArmor profile:
# apparmor_parser -R /etc/apparmor.d/usr.sbin.dnsmasq # apparmor_parser /etc/apparmor.d/usr.sbin.dnsmasq
No image found
By default, the registry list is not populated as the files in the package come from upstream. This means that by default, trying to pull any image without specifying the registry will result in an error similar to the following:
Error: short-name "archlinux" did not resolve to an alias and no unqualified-search registries are defined in "/etc/containers/registries.conf"
A starting configuration could be the following:
/etc/containers/registries.conf.d/00-unqualified-search-registries.conf
unqualified-search-registries = ["docker.io"]
/etc/containers/registries.conf.d/01-registries.conf
[[registry]] location = "docker.io"
This is equivalent to the default docker configuration.
A less convenient alternative, but having a higher compatibility with systems without configured shortnames, use the full registry path in the Containerfile or Dockerfile.
Containerfile
FROM docker.io/archlinux/archlinux
Permission denied: OCI permission denied
$ podman exec openvas_openvas_1 bash
Error: crun: writing file `/sys/fs/cgroup/user.slice/user-1000.slice/user@1000.service/user.slice/libpod-b3e8048a9b91e43c214b4d850ac7132155a684d6502e12e22ceb6f73848d117a.scope/container/cgroup.procs`: Permission denied: OCI permission denied
Can be solved: BBS#253966
$ env DBUS_SESSION_BUS_ADDRESS= podman ... $ env DBUS_SESSION_BUS_ADDRESS= podman-compose ...
Pushing images to Docker Hub: access denied/authentication required
When using podman push to push container images to Docker Hub, the following errors could occur: Requested access to the resource is denied or Authentication required. The following hints can help to fix potential issues:
- Tag the local image:
# podman tag <localImage> docker.io/<dockerHubUsername>/<dockerHubRepository>:<Tag>
- Push the tagged image:
# podman push docker.io/<dockerHubUsername>/<dockerHubRepository>:<Tag> docker://docker.io/<dockerHubUsername>/<dockerHubRepository>:<Tag>
- Login to docker.io, the Docker Hub repository and Docker Hub Registry server:
# podman login -u <DockerHubUsername> -p <DockerHubPassword> registry-1.docker.io # podman login -u <DockerHubUsername> -p <DockerHubPassword> docker.io/<dockerHubUsername>/<dockerHubRepository> # podman login -u <DockerHubUsername> -p <DockerHubPassword> docker.io
- Logout from all registries before the login, e.g.,
# podman logout --all
- Add
<dockerHubUsername>as collaborator in the Docker Hub Collaborators tab of the repository
Buildah/Podman running as rootless expects the bind mount to be shared, check if it is set to private:
$ findmnt -o PROPAGATION /
PROPAGATION private
In this case see mount(8) § Shared_subtree_operations and set temporarily the mount as shared with:
# mount --make-shared /
To set it permanently edit /etc/fstab and add the shared option to the desired mount and reboot. It will result in a entry like:
/etc/fstab
# <device> <dir> <type> <options> <dump> <fsck> UUID=0a3407de-014b-458b-b5c1-848e92a327a3 / ext4 defaults,shared 0 1
Networking issues inside containers
IP networking
Podman containers are by default bridged with the host through their own virtual network interfaces.
For example, inside a container, virtual interface eth0@if6 has IP 10.89.0.3 (IPs might be different on your system!):
container# ip addr
...
2: eth0@if6: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
...
inet 10.89.0.3/24 brd 10.89.0.255 scope global eth0
valid_lft forever preferred_lft forever
On the host, packets from the container exit on the host side from another virtual interface, here named podman1 as if routed via IP 10.89.0.1:
host# ip addr
...
4: podman1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
...
inet 10.89.0.1/24 brd 10.89.0.255 scope global podman1
Despite being virtual IP addresses, packets are still routed through the kernel's packet filtering system and can therefore be blocked by iptables/nftables rules. In particular, default DROP policy in INPUT or FORWARD iptables filter chains and/or running firewalls (ufw, firewalld) can affect containers in some cases. Check your configuration (for example with iptables -L -n -v or nft list ruleset) if you think this may be the case.
After a change in docker-compose.yml, note that created networks (from the networks: section) may not be destroyed when using podman compose down to destroy an environment. Make sure (using podman network ls and podman network rm if necessary) that they are if that is your intention.
DNS and name resolution
Name resolution is handled by subsystems of Podman (for example aardvark-dns), which provide both external DNS (usually through the host's DNS resolver) and container name resolution (e.g. webserver.dns.podman talking to database.dns.podman).
In the example above, containers are configured automatically by Podman via /etc/resolv.conf to ask a DNS resolver running on port 53 on the host-side of the pipe:
container# cat /etc/resolv.conf
search dns.podman nameserver 10.89.0.1
Check that you don't have another DNS resolver running on the host on port 53 (for example Systemd-resolved or Unbound), as it may interfere with Podman name resolution. If that is the case, you can change the port used by Podman on the host to any other available port, and Podman should automatically forward container requests from containers to the correct port on the host:
host# # cat /etc/containers/containers.conf
... dns_bind_port = 20053