podman-troubleshooting - Man Page

A large number of issues reported against Podman are often found to already be fixed in more current versions of the project.  Before reporting an issue, please verify the version you are running with podman version and compare it to the latest release documented on the top of Podman's  ⟨README.md⟩.

If they differ, please update your version of PODMAN to the latest possible and retry your command before reporting the issue.

$ podman run -v ~/mycontent:/content fedora touch /content/file
touch: cannot touch '/content/file': Permission denied

This is sometimes caused by SELinux, and sometimes by user namespaces.

Labeling systems like SELinux require that proper labels are placed on volume content mounted into a container. Without a label, the security system might prevent the processes running inside the container from using the content. By default, Podman does not change the labels set by the OS.

To change a label in the container context, you can add either of two suffixes :z or :Z to the volume mount. These suffixes tell Podman to relabel file objects on the shared volumes. The z option tells Podman that two containers share the volume content. As a result, Podman labels the content with a shared content label. Shared volume labels allow all containers to read/write content. The Z option tells Podman to label the content with a private unshared label. Only the current container can use a private volume.

$ podman run -v ~/mycontent:/content:Z fedora touch /content/file

Make sure the content is private for the container.  Do not relabel system directories and content. Relabeling system content might cause other confined services on your machine to fail.  For these types of containers we recommend having SELinux separation disabled.  The option --security-opt label=disable will disable SELinux separation for the container.

$ podman run --security-opt label=disable -v ~:/home/user fedora touch /home/user/file

In cases where the container image runs as a specific, non-root user, though, the solution is to fix the user namespace.  This would include container images such as the Jupyter Notebook image (which runs as "jovyan") and the Postgres image (which runs as "postgres").  In either case, use the --userns switch to map user namespaces, most of the time by using the keep-id option.

$ podman run -v "$PWD":/home/jovyan/work --userns=keep-id jupyter/scipy-notebook

When doing a podman pull or podman build command and a "common" image cannot be pulled, it is likely that the /etc/containers/registries.conf file is either not installed or possibly misconfigured.

$ sudo podman build -f Dockerfile
STEP 1: FROM alpine
error building: error creating build container: no such image "alpine" in registry: image not known

or

$ sudo podman pull fedora
error pulling image "fedora": unable to pull fedora: error getting default registries to try: Near line 9 (last key parsed ''): Bare keys cannot contain ':'.

• Verify that the /etc/containers/registries.conf file exists.  If not, verify that the containers-common package is installed.

• Verify that the entries in the unqualified-search-registries list of the /etc/containers/registries.conf file are valid and reachable.

  • i.e. unqualified-search-registries = ["registry.fedoraproject.org", "quay.io", "registry.access.redhat.com"]

When doing a Podman command such as build, commit, pull, or push to a registry, TLS verification is turned on by default.  If encryption is not used with those commands, this error can occur.

$ sudo podman push alpine docker://localhost:5000/myalpine:latest
Getting image source signatures
Get https://localhost:5000/v2/: http: server gave HTTP response to HTTPS client

By default TLS verification is turned on when communicating to registries from Podman.  If the registry does not require encryption the Podman commands such as build, commit, pull and push will fail unless TLS verification is turned off using the --tls-verify option.  NOTE: It is not at all recommended to communicate with a registry and not use TLS verification.

• Turn off TLS verification by passing false to the tls-verify option.
• I.e. podman push --tls-verify=false alpine docker://localhost:5000/myalpine:latest

For a global workaround, users[1] can create the file /etc/containers/registries.conf.d/registry-NAME.conf (replacing NAME with the name of this registry) with the following content (replacing FULLY.QUALIFIED.NAME.OF.REGISTRY with the address of this registry):

[[registry]]
location = "FULLY.QUALIFIED.NAME.OF.REGISTRY"
insecure = true

[1] If you are using a Mac / Windows, you should execute podman machine ssh to login into podman machine before adding the insecure entry to the registry—conf file.

This is an insecure method and should be used cautiously.

When using the ping command from a non-root container, the command may fail because of a lack of privileges.

$ podman run --rm fedora ping -W10 -c1 redhat.com
PING redhat.com (209.132.183.105): 56 data bytes

--- redhat.com ping statistics ---
1 packets transmitted, 0 packets received, 100% packet loss

It is most likely necessary to enable unprivileged pings on the host. Be sure the UID of the user is part of the range in the /proc/sys/net/ipv4/ping_group_range file.

To change its value you can use something like:

# sysctl -w "net.ipv4.ping_group_range=0 2000000"

To make the change persistent, you'll need to add a file in /etc/sysctl.d that contains net.ipv4.ping_group_range=0 $MAX_UID.

When the Dockerfile contains a command like RUN useradd -u 99999000 -g users newuser the build can hang.

If you are using a useradd command within a Dockerfile with a large UID/GID, it will create a large sparse file /var/log/lastlog.  This can cause the build to hang forever.  Go language does not support sparse files correctly, which can lead to some huge files being created in your container image.

If the entry in the Dockerfile looked like: RUN useradd -u 99999000 -g users newuser then add the --no-log-init parameter to change it to: RUN useradd --no-log-init -u 99999000 -g users newuser. This option tells useradd to stop creating the lastlog file.

When rootless Podman attempts to execute a container on a non exec home directory a permission error will be raised.

If you are running Podman or Buildah on a home directory that is mounted noexec, then they will fail with a message like:

$ podman run centos:7
standard_init_linux.go:203: exec user process caused "permission denied"

Since the administrator of the system set up your home directory to be noexec, you will not be allowed to execute containers from storage in your home directory. It is possible to work around this by manually specifying a container storage path that is not on a noexec mount. Simply copy the file /etc/containers/storage.conf to ~/.config/containers/ (creating the directory if necessary). Specify a graphroot directory which is not on a noexec mount point and to which you have read/write privileges.  You will need to modify other fields to writable directories as well.

For example

$ cat ~/.config/containers/storage.conf
[storage]
  driver = "overlay"
  runroot = "/run/user/1000"
  graphroot = "/execdir/myuser/storage"
  [storage.options]
    mount_program = "/bin/fuse-overlayfs"

When running systemd as PID 1 inside of a container on an SELinux separated machine, it needs to write to the cgroup file system.

Systemd gets permission denied when attempting to write to the cgroup file system, and AVC messages start to show up in the audit.log file or journal on the system.

Newer versions of Podman (2.0 or greater) support running init based containers with a different SELinux labels, which allow the container process access to the cgroup file system. This feature requires container-selinux-2.132 or newer versions.

Prior to Podman 2.0, the SELinux boolean container_manage_cgroup allows container processes to write to the cgroup file system. Turn on this boolean, on SELinux separated systems, to allow systemd to run properly in the container. Only do this on systems running older versions of Podman.

# setsebool -P container_manage_cgroup true

Rootless Podman requires the newuidmap and newgidmap programs to be installed.

If you are running Podman or Buildah as a rootless user, you get an error complaining about a missing newuidmap executable.

$ podman run -ti fedora sh
command required for rootless mode with multiple IDs: exec: "newuidmap": executable file not found in $PATH

Install a version of shadow-utils that includes these executables.  Note that for RHEL and CentOS 7, at least the 7.7 release must be installed for support to be available.

Rootless Podman requires the user running it to have a range of UIDs listed in /etc/subuid and /etc/subgid.

A user, either via --user or through the default configured for the image, is not mapped inside the namespace.

$ podman run --rm -ti --user 1000000 alpine echo hi
Error: container create failed: container_linux.go:344: starting container process caused "setup user: invalid argument"

Update the /etc/subuid and /etc/subgid with fields for users that look like:

$ cat /etc/subuid
johndoe:100000:65536
test:165536:65536

The format of this file is USERNAME:UID:RANGE

• username as listed in /etc/passwd or getpwent.
• The initial uid allocated for the user.
• The size of the range of UIDs allocated for the user.

This means johndoe is allocated UIDs 100000-165535 as well as his standard UID in the /etc/passwd file.

You should ensure that each user has a unique range of UIDs, because overlapping UIDs, would potentially allow one user to attack another user. In addition, make sure that the range of UIDs you allocate can cover all UIDs that the container requires. For example, if the container has a user with UID 10000, ensure you have at least 10000 subuids, and if the container needs to be run as a user with UID 1000000, ensure you have at least 1000000 subuids.

You could also use the usermod program to assign UIDs to a user.

If you update either the /etc/subuid or /etc/subgid file, you need to stop all running containers and kill the pause process.  This is done automatically by the system migrate command, which can also be used to stop all the containers and kill the pause process.

# usermod --add-subuids 200000-201000 --add-subgids 200000-201000 johndoe
# grep johndoe /etc/subuid /etc/subgid
/etc/subuid:johndoe:200000:1001
/etc/subgid:johndoe:200000:1001

When I change the graphroot storage location in storage.conf, the next time I run Podman, I get an error like:

# podman run -p 5000:5000 -it centos bash

bash: error while loading shared libraries: /lib64/libc.so.6: cannot apply additional memory protection after relocation: Permission denied

For example, the admin sets up a spare disk to be mounted at /src/containers, and points storage.conf at this directory.

SELinux blocks containers from using arbitrary locations for overlay storage. These directories need to be labeled with the same labels as if the content was under /var/lib/containers/storage.

Tell SELinux about the new containers storage by setting up an equivalence record. This tells SELinux to label content under the new path, as if it was stored under /var/lib/containers/storage.

# semanage fcontext -a -e /var/lib/containers /srv/containers
# restorecon -R -v /srv/containers

The semanage command above tells SELinux to set up the default labeling of /srv/containers to match /var/lib/containers.  The restorecon command tells SELinux to apply the labels to the actual content.

Now all new content created in these directories will automatically be created with the correct label.

Pulling an anonymous image that doesn't require authentication can result in an invalid username/password error.

If you pull an anonymous image, one that should not require credentials, you can receive an invalid username/password error if you have credentials established in the authentication file for the target container registry that are no longer valid.

$ podman run -it --rm docker://docker.io/library/alpine:latest ls
Trying to pull docker://docker.io/library/alpine:latest...ERRO[0000] Error pulling image ref //alpine:latest: Error determining manifest MIME type for docker://alpine:latest: unable to retrieve auth token: invalid username/password
Failed
Error: unable to pull docker://docker.io/library/alpine:latest: unable to pull image: Error determining manifest MIME type for docker://alpine:latest: unable to retrieve auth token: invalid username/password

This can happen if the authentication file is modified 'by hand' or if the credentials are established locally and then the password is updated later in the container registry.

Depending upon which container tool was used to establish the credentials, use podman logout or docker logout to remove the credentials from the authentication file.

Running Podman in a container and forwarding some, but not all, of the required host directories can cause inconsistent container behavior.

After creating a container with Podman's storage directories mounted in from the host and running Podman inside a container, all containers show their state as "configured" or "created", even if they were running or stopped.

When running Podman inside a container, it is recommended to mount at a minimum /var/lib/containers/storage/ as a volume. Typically, you will not mount in the host version of the directory, but if you wish to share containers with the host, you can do so. If you do mount in the host's /var/lib/containers/storage, however, you must also mount in the host's /run/libpod and /run/containers/storage directories. Not doing this will cause Podman in the container to detect that temporary files have been cleared, leading it to assume a system restart has taken place. This can cause Podman to reset container states and lose track of running containers.

For running containers on the host from inside a container, we also recommend the Podman remote client ⟨docs/tutorials/remote_client.md⟩, which only requires a single socket to be mounted into the container.

NFS enforces file creation on different UIDs on the server side and does not understand user namespace, which rootless Podman requires. When a container root process like YUM attempts to create a file owned by a different UID, NFS Server denies the creation. NFS is also a problem for the file locks when the storage is on it.  Other distributed file systems (for example: Lustre, Spectrum Scale, the General Parallel File System (GPFS)) are also not supported when running in rootless mode as these file systems do not understand user namespace.

$ podman build .
ERRO[0014] Error while applying layer: ApplyLayer exit status 1 stdout:  stderr: open /root/.bash_logout: permission denied
error creating build container: Error committing the finished image: error adding layer with blob "sha256:a02a4930cb5d36f3290eb84f4bfa30668ef2e9fe3a1fb73ec015fc58b9958b17": ApplyLayer exit status 1 stdout:  stderr: open /root/.bash_logout: permission denied

Choose one of the following:
 * Set up containers/storage in a different directory, not on an NFS share.
   * Create a directory on a local file system.
   * Edit ~/.config/containers/containers.conf and point the volume_path option to that local directory. (Copy /usr/share/containers/containers.conf if ~/.config/containers/containers.conf does not exist)
 * Otherwise just run Podman as root, via sudo podman

The Overlay file system (OverlayFS) requires the ability to call the mknod command when creating whiteout files when extracting an image.  However, a rootless user does not have the privileges to use mknod in this capacity.

$ podman build --storage-driver overlay .
STEP 1: FROM docker.io/ubuntu:xenial
Getting image source signatures
Copying blob edf72af6d627 done
Copying blob 3e4f86211d23 done
Copying blob 8d3eac894db4 done
Copying blob f7277927d38a done
Copying config 5e13f8dd4c done
Writing manifest to image destination
Storing signatures
Error: creating build container: Error committing the finished image: error adding layer with blob "sha256:8d3eac894db4dc4154377ad28643dfe6625ff0e54bcfa63e0d04921f1a8ef7f8": Error processing tar file(exit status 1): operation not permitted
$ podman build .
ERRO[0014] Error while applying layer: ApplyLayer exit status 1 stdout:  stderr: open /root/.bash_logout: permission denied
error creating build container: Error committing the finished image: error adding layer with blob "sha256:a02a4930cb5d36f3290eb84f4bfa30668ef2e9fe3a1fb73ec015fc58b9958b17": ApplyLayer exit status 1 stdout:  stderr: open /root/.bash_logout: permission denied

Choose one of the following:
 * Complete the build operation as a privileged user.
 * Install and configure fuse-overlayfs.
   * Install the fuse-overlayfs package for your Linux Distribution.
   * Add mount_program = "/usr/bin/fuse-overlayfs" under [storage.options] in your ~/.config/containers/storage.conf file.

The systemd version shipped in RHEL 7 and CentOS 7 doesn't have support for cgroup v2.  Support for cgroup v2 requires version 230 of systemd or newer, which was never shipped or supported on RHEL 7 or CentOS 7.

# podman run --name test -d registry.access.redhat.com/rhel7-init:latest && sleep 10 && podman exec test systemctl status
c8567461948439bce72fad3076a91ececfb7b14d469bfa5fbc32c6403185beff
Failed to get D-Bus connection: Operation not permitted
Error: non zero exit code: 1: OCI runtime error

You'll need to either:

• configure the host to use cgroup v1. On Fedora you can do:

  # dnf install -y grubby
  # grubby --update-kernel=ALL --args=”systemd.unified_cgroup_hierarchy=0"
  # reboot

• update the image to use an updated version of systemd.

You need to set lingering mode through loginctl to prevent user processes to be killed once the user session completed.

Once the user logs out all the containers exit.

# loginctl enable-linger $UID

The Kernel Lockdown patches deny eBPF programs when Secure Boot is enabled in the BIOS. Matthew Garrett's post describes the relationship between Lockdown and Secure Boot and Jan-Philip Gehrcke's connects this with eBPF. RH bug 1768125 contains some additional details.

Attempts to run podman result in

#### Solution

One workaround is to disable Secure Boot in your BIOS.

### 19) error creating libpod runtime: there might not be enough IDs available in the namespace

Unable to pull images

#### Symptom

```console
$ podman unshare cat /proc/self/uid_map
	 0       1000          1

$ podman system migrate

Original command now returns

$ podman unshare cat /proc/self/uid_map
	 0       1000          1
	 1     100000      65536

Reference subuid and subgid man pages for more detail.

As a non-root user you have group access rights to a device or files that you want to pass into a rootless container with --device=... or --volume=...

Any access inside the container is rejected with "Permission denied".

The runtime uses setgroups(2) hence the process loses all additional groups the non-root user has. Use the --group-add keep-groups flag to pass the user's supplementary group access into the container. Currently only available with the crun OCI runtime.

When running a container with a command like podman run --detach httpd as a rootless user, the container is closed upon logout and is not kept running.

When logging out of a rootless user session, all containers that were started in detached mode are stopped and are not kept running.  As the root user, these same containers would survive the logout and continue running.

When systemd notes that a session that started a Podman container has exited, it will also stop any containers that have been associated with it.  To avoid this, use the following command before logging out: loginctl enable-linger. To later revert the linger functionality, use loginctl disable-linger.

LOGINCTL(1), SYSTEMD(1)

Podman defaults to ctrl-p,ctrl-q to detach from a running containers. The bash and zsh shells default to ctrl-p for the displaying of the previous command.  This causes issues when running a shell inside of a container.

With the default detach key combo ctrl-p,ctrl-q, shell history navigation (tested in bash and zsh) using ctrl-p to access the previous command will not display this previous command, or anything else.  Conmon is waiting for an additional character to see if the user wants to detach from the container. Adding additional characters to the command will cause it to be displayed along with the additional character. If the user types ctrl-p a second time the shell display the 2nd to last command.

The solution to this is to change the default detach_keys. For example in order to change the defaults to ctrl-q,ctrl-q use the --detach-keys option.

$ podman run -ti --detach-keys ctrl-q,ctrl-q fedora sh

To make this change the default for all containers, users can modify the containers.conf file. This can be done simply in your home directory, but adding the following lines to users containers.conf

$ cat >> ~/.config/containers/containers.conf << _eof
[engine]
detach_keys="ctrl-q,ctrl-q"
_eof

In order to effect root running containers and all users, modify the system wide defaults in /etc/containers/containers.conf.

A container with ports that have been published with the --publish or -p option can not be run within a pod.

$ podman pod create --name srcview -p 127.0.0.1:3434:3434 -p 127.0.0.1:7080:7080 -p 127.0.0.1:3370:3370                        4b2f4611fa2cbd60b3899b936368c2b3f4f0f68bc8e6593416e0ab8ecb0a3f1d

$ podman run --pod srcview --name src-expose -p 3434:3434 -v "${PWD}:/var/opt/localrepo":Z,ro sourcegraph/src-expose:latest serve /var/opt/localrepo
Error: cannot set port bindings on an existing container network namespace

This is a known limitation.  If a container will be run within a pod, it is not necessary to publish the port for the containers in the pod. The port must only be published by the pod itself.  Pod network stacks act like the network stack on the host - you have a variety of containers in the pod, and programs in the container, all sharing a single interface and IP address, and associated ports. If one container binds to a port, no other container can use that port within the pod while it is in use. Containers in the pod can also communicate over localhost by having one container bind to localhost in the pod, and another connect to that port.

In the example from the symptom section, dropping the -p 3434:3434 would allow the podman run command to complete, and the container as part of the pod would still have access to that port.  For example:

$ podman run --pod srcview --name src-expose -v "${PWD}:/var/opt/localrepo":Z,ro sourcegraph/src-expose:latest serve /var/opt/localrepo

Some container images require that the fuse kernel module is loaded in the kernel before they will run with the fuse filesystem in play.

When trying to run the container images found at quay.io/podman, quay.io/containers registry.access.redhat.com/ubi8 or other locations, an error will sometimes be returned:

ERRO error unmounting /var/lib/containers/storage/overlay/30c058cdadc888177361dd14a7ed7edab441c58525b341df321f07bc11440e68/merged: invalid argument
error mounting container "1ae176ca72b3da7c70af31db7434bcf6f94b07dbc0328bc7e4e8fc9579d0dc2e": error mounting build container "1ae176ca72b3da7c70af31db7434bcf6f94b07dbc0328bc7e4e8fc9579d0dc2e": error creating overlay mount to /var/lib/containers/storage/overlay/30c058cdadc888177361dd14a7ed7edab441c58525b341df321f07bc11440e68/merged: using mount program /usr/bin/fuse-overlayfs: fuse: device not found, try 'modprobe fuse' first
fuse-overlayfs: cannot mount: No such device
: exit status 1
ERRO exit status 1

If you encounter a fuse: device not found error when running the container image, it is likely that the fuse kernel module has not been loaded on your host system.  Use the command modprobe fuse to load the module and then run the container image afterwards.  To enable this automatically at boot time, you can add a configuration file to /etc/modules.load.d.  See man modules-load.d for more details.

An error such as "OCI runtime error" on a read-only filesystem or the error "{image} is not an absolute path or is a symlink" are often times indicators for this issue.  For more details, review this issue ⟨https://github.com/containers/podman/issues/5895⟩.

Rootless Podman requires certain files to exist in a file system in order to run. Podman will create /etc/resolv.conf, /etc/hosts and other file descriptors on the rootfs in order to mount volumes on them.

Run the container once in read/write mode, Podman will generate all of the FDs on the rootfs, and from that point forward you can run with a read-only rootfs.

$ podman run --rm --rootfs /path/to/rootfs true

The command above will create all the missing directories needed to run the container.

After that, it can be used in read-only mode, by multiple containers at the same time:

$ podman run --read-only --rootfs /path/to/rootfs ....

Another option is to use an Overlay Rootfs Mount:

$ podman run --rootfs /path/to/rootfs:O ....

Modifications to the mount point are destroyed when the container finishes executing, similar to a tmpfs mount point being unmounted.

On some systemd-based systems, non-root users do not have resource limit delegation permissions. This causes setting resource limits to fail.

Running a container with a resource limit options will fail with an error similar to the following:

--cpus, --cpu-period, --cpu-quota, --cpu-shares:

Error: OCI runtime error: crun: the requested cgroup controller `cpu` is not available

--cpuset-cpus, --cpuset-mems:

Error: OCI runtime error: crun: the requested cgroup controller `cpuset` is not available

This means that resource limit delegation is not enabled for the current user.

You can verify whether resource limit delegation is enabled by running the following command:

$ cat "/sys/fs/cgroup/user.slice/user-$(id -u).slice/user@$(id -u).service/cgroup.controllers"

Example output might be:

memory pids

In the above example, cpu and cpuset are not listed, which means the current user does not have permission to set CPU or CPUSET limits.

If you want to enable CPU or CPUSET limit delegation for all users, you can create the file /etc/systemd/system/user@.service.d/delegate.conf with the contents:

[Service]
Delegate=memory pids cpu cpuset

After logging out and logging back in, you should have permission to set CPU and CPUSET limits.

This can happen when running a container from an image for another architecture than the one you are running on.

For example, if a remote repository only has, and thus send you, a linux/arm64 OS/ARCH but you run on linux/amd64 (as happened in https://github.com/openMF/community-app/issues/3323 due to https://github.com/timbru31/docker-ruby-node/issues/564).

In some situations where the client is not on the same machine as where the podman daemon is running the client key could be using a cipher not supported by the host. This indicates an issue with one's SSH config. Until remedied using podman over ssh with a pre-shared key will be impossible.

The accepted ciphers per /etc/crypto-policies/back-ends/openssh.config are not one that was used to create the public/private key pair that was transferred over to the host for ssh authentication.

You can confirm this is the case by attempting to connect to the host via podman-remote info from the client and simultaneously on the host running journalctl -f and watching for the error userauth_pubkey: key type ssh-rsa not in PubkeyAcceptedAlgorithms [preauth].

Create a new key using a supported algorithm e.g. ecdsa:

$ ssh-keygen -t ecdsa -f ~/.ssh/podman

Then copy the new id over:

$ ssh-copy-id -i ~/.ssh/podman.pub user@host

And then re-add the connection (removing the old one if necessary):

$ podman-remote system connection add myuser --identity ~/.ssh/podman ssh://user@host/run/user/1000/podman/podman.sock

And now this should work:

$ podman-remote info

A failure is encountered when trying to use networking on a rootless container in Podman v2.2.1 through v3.0.1 on RHEL.  This error does not occur on other Linux distributions.

A rootless container is created using a CNI network, but the podman run command returns an error that an image must be built.

In order to use a CNI network in a rootless container on RHEL, an Infra container image for CNI-in-slirp4netns must be created.  The instructions for building the Infra container image can be found for v2.2.1 here, and for v3.0.1 here.

Container network can't be reached after firewall-cmd --reload and systemctl restart firewalld Running podman network reload will fix it but it has to be done manually.

The firewall rules created by podman are lost when the firewall is reloaded.

@ranjithrajaram has created a systemd-hook to fix this issue

1) For "firewall-cmd --reload", create a systemd unit file with the following

[Unit]
Description=firewalld reload hook - run a hook script on firewalld reload
Wants=dbus.service
After=dbus.service

[Service]
Type=simple
ExecStart=/bin/bash -c '/bin/busctl monitor --system --match "interface=org.fedoraproject.FirewallD1,member=Reloaded" --match "interface=org.fedoraproject.FirewallD1,member=PropertiesChanged" | while read -r line ; do podman network reload --all ; done'

[Install]
WantedBy=default.target

2) For "systemctl restart firewalld", create a systemd unit file with the following

[Unit]
Description=podman network reload
Wants=firewalld.service
After=firewalld.service
PartOf=firewalld.service

[Service]
Type=simple
RemainAfterExit=yes
ExecStart=/usr/bin/podman network reload --all

[Install]
WantedBy=default.target

However, If you use busctl monitor then you can't get machine-readable output on RHEL 8. Since it doesn't have busctl -j as mentioned here by @yrro.

For RHEL 8, you can use the following one-liner bash script.

[Unit]
Description=Redo podman NAT rules after firewalld starts or reloads
Wants=dbus.service
After=dbus.service
Requires=firewalld.service

[Service]
Type=simple
ExecStart=/bin/bash -c "dbus-monitor --profile --system 'type=signal,sender=org.freedesktop.DBus,path=/org/freedesktop/DBus,interface=org.freedesktop.DBus,member=NameAcquired,arg0=org.fedoraproject.FirewallD1' 'type=signal,path=/org/fedoraproject/FirewallD1,interface=org.fedoraproject.FirewallD1,member=Reloaded' | sed -u '/^#/d' | while read -r type timestamp serial sender destination path interface member _junk; do if [[ $type = '#'* ]]; then continue; elif [[ $interface = org.freedesktop.DBus && $member = NameAcquired ]]; then echo 'firewalld started'; podman network reload --all; elif [[ $interface = org.fedoraproject.FirewallD1 && $member = Reloaded ]]; then echo 'firewalld reloaded'; podman network reload --all; fi; done"
Restart=Always

[Install]
WantedBy=default.target

busctl-monitor is almost usable in RHEL 8, except that it always outputs two bogus events when it starts up, one of which is (in its only machine-readable format) indistinguishable from the NameOwnerChanged that you get when firewalld starts up. This means you would get an extra podman network reload --all when this unit starts.

Apart from this, you can use the following systemd service with the python3 code.

[Unit]
Description=Redo podman NAT rules after firewalld starts or reloads
Wants=dbus.service
Requires=firewalld.service
After=dbus.service

[Service]
Type=simple
ExecStart=/usr/bin/python  /path/to/python/code/podman-redo-nat.py
Restart=always

[Install]
WantedBy=default.target

The code reloads podman network twice when you use systemctl restart firewalld.

import dbus
from gi.repository import GLib
from dbus.mainloop.glib import DBusGMainLoop
import subprocess
import sys

# I'm a bit confused on the return values in the code
# Not sure if they are needed.

def reload_podman_network():
    try:
        subprocess.run(["podman","network","reload","--all"],timeout=90)
        # I'm not sure about this part
        sys.stdout.write("podman network reload done\n")
        sys.stdout.flush()
    except subprocess.TimeoutExpired as t:
        sys.stderr.write(f"Podman reload failed due to Timeout {t}")
    except subprocess.CalledProcessError as e:
        sys.stderr.write(f"Podman reload failed due to {e}")
    except Exception as e:
        sys.stderr.write(f"Podman reload failed with an Unhandled Exception {e}")

    return False

def signal_handler(*args, **kwargs):
    if kwargs.get('member') == "Reloaded":
        reload_podman_network()
    elif kwargs.get('member') == "NameOwnerChanged":
        reload_podman_network()
    else:
        return None
    return None

def signal_listener():
    try:
        DBusGMainLoop(set_as_default=True)# Define the loop.
        loop = GLib.MainLoop()
        system_bus = dbus.SystemBus()
        # Listens to systemctl restart firewalld with a filter added, will cause podman network to be reloaded twice
        system_bus.add_signal_receiver(signal_handler,dbus_interface='org.freedesktop.DBus',arg0='org.fedoraproject.FirewallD1',member_keyword='member')
        # Listens to firewall-cmd --reload
        system_bus.add_signal_receiver(signal_handler,dbus_interface='org.fedoraproject.FirewallD1',signal_name='Reloaded',member_keyword='member')
        loop.run()
    except KeyboardInterrupt:
        loop.quit()
        sys.exit(0)
    except Exception as e:
        loop.quit()
        sys.stderr.write(f"Error occurred {e}")
        sys.exit(1)

if __name__ == "__main__":
    signal_listener()

A failure is encountered when performing podman run with a warning XDG_RUNTIME_DIR is pointing to a path which is not writable. Most likely podman will fail.

A rootless container is being invoked with cgroup configuration as cgroupv2 for user with missing or invalid systemd session.

Example cases

# su user1 -c 'podman images'
ERRO[0000] XDG_RUNTIME_DIR directory "/run/user/0" is not owned by the current user

# su - user1 -c 'podman images'
Error: creating tmpdir: mkdir /run/user/1000: permission denied

Podman expects a valid login session for the rootless+cgroupv2 use-case. Podman execution is expected to fail if the login session is not present. In most cases, podman will figure out a solution on its own but if XDG_RUNTIME_DIR is pointing to a path that is not writable execution will most likely fail. Typical scenarios of such cases are seen when users are trying to use Podman with su - <user> -c '<podman-command>', or sudo -l and badly configured systemd session.

Alternatives:

• Execute Podman via systemd-run that will first start a systemd login session:

    $ sudo systemd-run --machine=username@ --quiet --user --collect --pipe --wait podman run --rm docker.io/library/alpine echo hello

• Start an interactive shell in a systemd login session with the command machinectl shell <username>@ and then run Podman

    $ sudo -i
    # machinectl shell username@
    Connected to the local host. Press ^] three times within 1s to exit session.
    $ podman run --rm docker.io/library/alpine echo hello

• Start a new systemd login session by logging in with ssh i.e. ssh <username>@localhost and then run Podman.
• Before invoking Podman command create a valid login session for your rootless user using loginctl enable-linger <username>

After deleting a VM on macOS, the initialization of subsequent VMs fails.

After deleting a client VM on macOS via podman machine stop && podman machine rm, attempting to podman machine init a new client VM leads to an error with the 127.0.0.1:7777 port already bound.

You will need to remove the hanging gv-proxy process bound to the port in question. For example, if the port mentioned in the error message is 127.0.0.1:7777, you can use the command kill -9 $(lsof -i:7777) in order to identify and remove the hanging process which prevents you from starting a new VM on that default port.

The sshd process running inside the container fails with the error "Error writing /proc/self/loginuid".

If the /proc/self/loginuid file is already initialized then the CAP_AUDIT_CONTROL capability is required to override it.

This happens when running Podman from a user session since the /proc/self/loginuid file is already initialized.  The solution is to run Podman from a system service, either using the Podman service, and then using podman -remote to start the container or simply by running something like systemd-run podman run ....  In this case the container will only need CAP_AUDIT_WRITE.

After running a container with rootless Podman, the non-root user sees a numerical UID and GID instead of a username and groupname.

When listing file permissions with ls -l on the host in a directory that was passed as --volume /some/dir to podman run, the UID and GID are displayed rather than the corresponding username and groupname. The UID and GID numbers displayed are from the user's subordinate UID and GID ranges on the host system.

An example

$ mkdir dir1
$ chmod 777 dir1
$ podman run --rm -v ./dir1:/dir1:Z   
             --user 2003:2003   
             docker.io/library/ubuntu bash -c "touch /dir1/a; chmod 600 /dir1/a"
$ ls -l dir1/a
-rw-------. 1 102002 102002 0 Jan 19 19:35 dir1/a
$ less dir1/a
less: dir1/a: Permission denied

If you want to read, chown, or remove such a file, enter a user namespace. Instead of running commands such as less dir1/a or rm dir1/a, you need to prepend the command-line with podman unshare, i.e., podman unshare less dir1/a or podman unshare rm dir1/a. To change the ownership of the file dir1/a to your regular user's UID and GID, run podman unshare chown 0:0 dir1/a. A file having the ownership 0:0 in the user namespace is owned by the regular user on the host. To use Bash features, such as variable expansion and globbing, you need to wrap the command with bash -c, e.g. podman unshare bash -c 'ls $HOME/dir1/a*'.

Would it have been possible to run Podman in another way so that your regular user would have become the owner of the file? Yes, you can use the option --userns keep-id:uid=$uid,gid=$gid to change how UIDs and GIDs are mapped between the container and the host. Let's try it out.

In the example above ls -l shows the UID 102002 and GID 102002. Set shell variables

$ uid_from_ls=102002
$ gid_from_ls=102002

Set shell variables to the lowest subordinate UID and GID

$ lowest_subuid=$(podman info --format "{{ (index .Host.IDMappings.UIDMap 1).HostID }}")
$ lowest_subgid=$(podman info --format "{{ (index .Host.IDMappings.GIDMap 1).HostID }}")

Compute the UID and GID inside the container that map to the owner of the created file on the host.

$ uid=$(( $uid_from_ls - $lowest_subuid + 1))
$ gid=$(( $gid_from_ls - $lowest_subgid + 1))

(In the computation it was assumed that there is only one subuid range and one subgid range)

$ echo $uid
2003
$ echo $gid
2003

The computation shows that the UID is 2003 and the GID is 2003 inside the container. This comes as no surprise as this is what was specified before with --user=2003:2003, but the same computation could be used whenever a username is specified or the --user option is not used.

Run the container again but now with UIDs and GIDs mapped

$ mkdir dir1
$ chmod 777 dir1
$ podman run --rm
  -v ./dir1:/dir1:Z   
  --user $uid:$gid   
  --userns keep-id:uid=$uid,gid=$gid   
     docker.io/library/ubuntu bash -c "touch /dir1/a; chmod 600 /dir1/a"
$ id -u
tester
$ id -g
tester
$ ls -l dir1/a
-rw-------. 1 tester tester 0 Jan 19 20:31 dir1/a
$

In this example the --user option specified a rootless user in the container. As the rootless user could also have been specified in the container image, e.g.

$ podman image inspect --format "user: {{.User}}" IMAGE
user: hpc

the same problem could also occur even without specifying --user.

Another variant of the same problem could occur when using --user=root:root (the default), but where the root user creates non-root owned files in some way (e.g by creating them themselves, or switching the effective UID to a rootless user and then creates files).

See also the troubleshooting tip:

⟨#39-podman-run-fails-with-error-unrecognized-namespace-mode-keep-iduid1000gid1000-passed⟩

As a non-root user you have access rights to devices, files and directories that you want to pass into a rootless container with --device=..., --volume=... or --mount=...

Podman by default maps a non-root user inside a container to one of the user's subordinate UIDs and subordinate GIDs on the host. When the container user tries to access a file, a "Permission denied" error could occur because the container user does not have the permissions of the regular user of the host.

• Any access inside the container is rejected with "Permission denied" for files, directories or devices passed in to the container with --device=..,--volume=.. or --mount=.., e.g.

  $ mkdir dir1
  $ chmod 700 dir1
  $ podman run --rm -v ./dir1:/dir1:Z   
               --user 2003:2003   
               docker.io/library/ubuntu ls /dir1
  ls: cannot open directory '/dir1': Permission denied

We follow essentially the same solution as in the previous troubleshooting tip:

⟨#34-container-creates-a-file-that-is-not-owned-by-the-users-regular-uid⟩

but for this problem the container UID and GID can't be as easily computed by mere addition and subtraction.

In other words, it might be more challenging to find out the UID and the GID inside the container that we want to map to the regular user on the host.

If the --user option is used together with a numerical UID and GID to specify a rootless user, we already know the answer.

If the --user option is used together with a username and groupname, we could look up the UID and GID in the file /etc/passwd of the container.

If the container user is not set via --user but instead from the container image, we could inspect the container image

$ podman image inspect --format "user: {{.User}}" IMAGE
user: hpc

and then look it up in /etc/passwd of the container.

If the problem occurs in a container that is started to run as root but later switches to an effictive UID of a rootless user, it might be less straightforward to find out the UID and the GID. Reading the Containerfile, Dockerfile or the /etc/passwd could give a clue.

To run the container with the rootless container UID and GID mapped to the user's regular UID and GID on the host follow these steps:

Set the uid and gid shell variables in a Bash shell to the UID and GID of the user that will be running inside the container, e.g.

$ uid=2003
$ gid=2003

and run

$ mkdir dir1
$ echo hello > dir1/file.txt
$ chmod 700 dir1/file.txt
$ podman run --rm   
  -v ./dir1:/dir1:Z   
  --user $uid:$gid   
  --userns keep-id:uid=$uid,gid=$gid   
  docker.io/library/alpine cat /dir1/file.txt
hello

See also the troubleshooting tip:

⟨#39-podman-run-fails-with-error-unrecognized-namespace-mode-keep-iduid1000gid1000-passed⟩

When an image in an additional store is used, it is not locked thus it can be deleted even if there are containers using it.

WARN[0000] Can't stat lower layer "/var/lib/containers/storage/overlay/l/7HS76F2P5N73FDUKUQAOJA3WI5" because it does not exist. Going through storage to recreate the missing symlinks.

It is the user responsibility to make sure images in an additional store are not deleted while being used by containers in another store.

After upgrading Podman to a newer version an issue with the earlier version of Podman still presents itself while using podman-remote.

While running podman remote commands with the most updated Podman, issues that were fixed in a prior version of Podman can arise either on the Podman client side or the Podman server side.

When upgrading Podman to a particular version for the required fixes, users often make the mistake of only upgrading the Podman client. However, suppose a setup uses podman-remote or uses a client that communicates with the Podman server on a remote machine via the REST API. In that case, it is required to upgrade both the Podman client and the Podman server running on the remote machine. Both the Podman client and server must be upgraded to the same version.

Example: If a particular bug was fixed in v4.1.0 then the Podman client must have version v4.1.0 as well the Podman server must have version v4.1.0.

When using the --tty (-t) flag, unexpected carriage returns are outputted on the terminal.

The container program prints a newline (\n) but the terminal outputs a carriage return and a newline (\r\n).

$ podman run --rm -t fedora echo abc | od -c
0000000   a   b   c  \r  \n
0000005

When run directly on the host, the result is as expected.

$ echo abc | od -c
0000000   a   b   c  \n
0000004

Extra carriage returns can also shift the prompt to the right.

$ podman run --rm -t fedora sh -c "echo 1; echo 2; echo 3" | cat -A
1^M$
    2^M$
        3^M$
            $

Run Podman without the --tty (-t) flag.

$ podman run --rm fedora echo abc | od -c
0000000   a   b   c  \n
0000004

The --tty (-t) flag should only be used when the program requires user interaction in the termainal, for instance expecting the user to type an answer to a question.

Where does the extra carriage return \r come from?

The extra \r is not outputted by Podman but by the terminal. In fact, a reconfiguration of the terminal can make the extra \r go away.

$ podman run --rm -t fedora /bin/sh -c "stty -onlcr && echo abc" | od -c
0000000   a   b   c  \n
0000004

Podman 4.3.0 introduced the options uid and gid that can be given to --userns keep-id which are not recognized by older versions of Podman.

When using a Podman version older than 4.3.0, the options uid and gid are not recognized, and an "unrecognized namespace mode" error is raised.

$ uid=1000
$ gid=1000
$ podman run --rm
  --user $uid:$gid   
  --userns keep-id:uid=$uid,gid=$gid   
     docker.io/library/ubuntu /bin/cat /proc/self/uid_map
Error: unrecognized namespace mode keep-id:uid=1000,gid=1000 passed
$

Use --uidmap and --gidmap options to describe the same UID and GID mapping.

Run

$ uid=1000
$ gid=1000
$ subuidSize=$(( $(podman info --format "{{ range   
   .Host.IDMappings.UIDMap }}+{{.Size }}{{end }}" ) - 1 ))
$ subgidSize=$(( $(podman info --format "{{ range   
   .Host.IDMappings.GIDMap }}+{{.Size }}{{end }}" ) - 1 ))
$ podman run --rm   
  --user $uid:$gid   
  --uidmap 0:1:$uid   
  --uidmap $uid:0:1   
  --uidmap $(($uid+1)):$(($uid+1)):$(($subuidSize-$uid))   
  --gidmap 0:1:$gid   
  --gidmap $gid:0:1   
  --gidmap $(($gid+1)):$(($gid+1)):$(($subgidSize-$gid))   
     docker.io/library/ubuntu /bin/cat /proc/self/uid_map
         0          1       1000
      1000          0          1
      1001       1001      64536

which uses the same UID and GID mapping as when specifying --userns keep-id:uid=$uid,gid=$gid with Podman 4.3.0 (or greater)

$ uid=1000
$ gid=1000
$ podman run --rm   
  --user $uid:$gid   
  --userns keep-id:uid=$uid,gid=$gid   
     docker.io/library/ubuntu /bin/cat /proc/self/uid_map
         0          1       1000
      1000          0          1
      1001       1001      64536

Replace /bin/cat /proc/self/uid_map with /bin/cat /proc/self/gid_map to show the GID mapping.

When trying to do a Podman command that pulls an image from local storage or a remote repository, an error is raised saying "image not known" or "error locating pulled image".  Even though the image had been verified before the Podman command was invoked.

After verifying that an image is in place either locally or on a remote repository, a Podman command referencing that image will fail in a manner like:

# ls Containerfile
FROM registry.access.redhat.com/ubi8-minimal:latest
MAINTAINER Podman Community
USER root

# podman build .
STEP 1/2: FROM registry.access.redhat.com/ubi8-minimal
Trying to pull registry.access.redhat.com/ubi8-minimal:latest...
Getting image source signatures
Checking if image destination supports signatures
Copying blob a6577091999b done
Copying config abb1ba1bce done
Writing manifest to image destination
Storing signatures
Error: error creating build container: error locating pulled image "registry.access.redhat.com/ubi8-minimal:latest" name in containers storage: registry.access.redhat.com/ubi8-minimal:latest: image not known

The general cause for this is a timing issue.  To make Podman commands as efficient as possible, read and write locks are only established for critical sections within the code.  When pulling an image from a repository, a copy of that image is first written to local storage using a write lock.  This lock is released before the image is then acquired/read.  If another process does a harmful command such as podman system prune --all or podman system reset or podman rmi --all, between the time the image is written and before the first process can acquire it, this type of image not known error can arise.

The maintainers of Podman have considered heavier-duty locks to close this timing window. However, the slowdown that all Podman commands would encounter was not considered worth the cost of completely closing this small timing window.

Executing a step in a Dockerfile/Containerfile which mounts secrets using --mount=type=secret fails with "operation not permitted" when running on a host filesystem mounted with nosuid and when using the runc runtime.

A RUN line in the Dockerfile/Containerfile contains a secret mount such as --mount=type=secret,id=MY_USER,target=/etc/dnf/vars/MY_USER. When running podman build the process fails with an error message like:

STEP 3/13: RUN --mount=type=secret,id=MY_USER,target=/etc/dnf/vars/MY_USER     --mount=type=secret,id=MY_USER,target=/etc/dnf/vars/MY_USER     ...: time="2023-06-13T18:04:59+02:00" level=error msg="runc create failed: unable to start container process: error during container init: error mounting \"/var/tmp/buildah2251989386/mnt/buildah-bind-target-11\" to rootfs at \"/etc/dnf/vars/MY_USER\": mount /var/tmp/buildah2251989386/mnt/buildah-bind-target-11:/etc/dnf/vars/MY_USER (via /proc/self/fd/7), flags: 0x1021: operation not permitted"
: exit status 1
ERRO[0002] did not get container create message from subprocess: EOF

• Install crun, e.g. with dnf install crun.
• Use the crun runtime by passing --runtime /usr/bin/crun to podman build.

See also Buildah issue 4228 for a full discussion of the problem.

When using the overlay storage driver to do a nested podman build inside a running container, file I/O operations such as COPY of a large amount of data is very slow or can hang completely.

Using the default overlay storage driver, a COPY, ADD, or an I/O intensive RUN line in a Containerfile that is run inside another container is very slow or hangs completely when running a podman build inside the running parent container.

This could be caused by the child container using fuse-overlayfs for writing to /var/lib/containers/storage. Writes can be slow with fuse-overlayfs. The solution is to use the native overlay filesystem by using a local directory on the host system as a volume to /var/lib/containers/storage like so: podman run --privileged --rm -it -v ./nested_storage:/var/lib/containers/storage parent:latest. Ensure that the base image of parent:latest in this example has no contents in /var/lib/containers/storage in the image itself for this to work. Once using the native volume, the nested container should not fall back to fuse-overlayfs to write files and the nested build will complete much faster.

If you don't have access to the parent run process, such as in a CI environment, then the second option is to change the storage driver to vfs in the parent image by changing changing this line in your storage.conf file: driver = "vfs". You may have to run podman system reset for this to take effect. You know it's changed when podman info |grep graphDriverName outputs graphDriverName: vfs. This method is slower performance than using the volume method above but is significantly faster than fuse-overlayfs

Using --userns=auto when creating new containers does not work as long as any containers exist that were created with --userns=keep-id or --userns=nomap

1. Run with --userns=auto $ podman run --rm -d --userns=auto alpine sleep 3600 The command succeeds.
2. Run with --userns=auto $ podman run --rm -d --userns=auto alpine sleep 3600 The command succeeds.
3. Run with --userns=keep-id $ podman run --rm -d --userns=keep-id alpine sleep 3600 The command succeeds.
4. Run with --userns=auto $ podman run --rm -d --userns=auto alpine sleep 3600 The command fails with the error message Error: creating container storage: not enough unused IDs in user namespace

Any existing containers that were created using --userns=keep-id or --userns=nomap must first be deleted before any new container can be created with --userns=auto

When rootful podman is invoked with --userns=auto, podman needs to pick subranges of subuids and subgids for the user namespace of the container. Rootful podman ensures that the subuid and subgid ranges for such containers do not overlap, but how can rootful podman prevent other tools from accidentally using these IDs?

It's not possible to block other tools that are running as root from using these IDs, but such tools would normally not use subuids and subgids that have already been assigned to a user in /etc/subuid and /etc/subgid.

The username containers on the host has a special function for rootful Podman. Rootful podman uses the subuids and subgids of the user containers when running --userns=auto containers. The user containers has no need for these subuids and subgids because no processes should be started as the user containers. In other words, the user containers is a special user that only exists on the system to reserve subuids and subgids for rootful podman.

containers is the default username but it can be changed by setting the option root-auto-userns-user in the file /etc/containers/storage.conf

Run rootful podman with --userns=auto

sudo podman run --rm --userns=auto alpine echo hello

The command fails with the error message:

ERRO[0000] Cannot find mappings for user "containers": no subuid ranges found for user "containers" in /etc/subuid
Error: creating container storage: not enough unused IDs in user namespace

The files /etc/subuid and /etc/subgid do not have any lines that start with containers:

Add subuid and subgid ranges for the user containers in /etc/subuid and /etc/subgid or provide such ranges with /etc/nsswitch.conf. For details, see subid(5).

The following steps create the user containers and assigns big subuid and subgid ranges to it:

1. Create the user containers sudo useradd --comment "Helper user to reserve subuids and subgids for Podman"  
              --no-create-home  
              --shell /usr/sbin/nologin  
              containers
2. Check the subuid and subgid ranges of the user containers $ grep ^containers: /etc/subuid containers:720896:65536 $ grep ^containers: /etc/subgid containers:720896:65536 By default useradd assigns 65536 subuids and 65536 subgids to a new user. Typically you would like the reserved pool to be bigger than that. The bigger the size, the more containers could be started with sudo podman run --userns=auto ...
3. Edit the line for the user containers in the files  /etc/subuid and /etc/subgid to make the ranges bigger. Ensure that the subuid range of the user containers do not overlap with any other subuid ranges in the files /etc/subuid. Ensure that the subgid range of the user containers do not overlap with any other subgid ranges in the files /etc/subgid.

Test the echo command again

sudo podman run --rm --userns=auto alpine echo hello

The command succeeds and prints hello