Skip to content

Latest commit

 

History

History
1999 lines (1627 loc) · 91.2 KB

CHANGELOG.md

File metadata and controls

1999 lines (1627 loc) · 91.2 KB
Raw

SingularityCE Changelog

Change Since Last Release

Requirements

  • Requires a minimum of Go 1.21 to build. Go 1.20 is end-of-life.

Bug Fixes

  • Set default PATH in container run in OCI-Mode when image does not set PATH.
  • Fix storage of credentials for docker.io to behave the same as for index.docker.io.
  • Improve documentation for remote list command.

4.1.2 [2024-03-05]

Bug Fixes

  • Set OCI runtime-spec annotations that are required by the documented image-spec conversion process.
  • In --oci mode always set inner ID map based on host user, not USER in OCI container. Fixes incorrect permissions for files owned by USER in the container.
  • Provide warning / info message for OCI image-spec features (volumes, exposed ports) that are not supported by singularity.
  • Honor WORKDIR by default for OCI images in --oci mode, as required by OCI image-spec.
  • Restore previous --writable behaviour when running a container image from SIF/SquashFS in user namepace mode. The image will be extracted to a temporary sandbox, which is writable at runtime. Note that any changes are not made to the original image.
  • Fix target: no such file or directory error in native mode when extracting layers from certain OCI images that manipulate hard links across layers.
  • Fix extraction of OCI layers when run in a root mapped user namespace (e.g.. unshare -r).
  • Use user namespace for wrapping of unsquashfs when singularity is run with --userns / -u flag. Fixes temporary sandbox extraction of images in non-root mapped user namespace (e.g. unshare -c).

4.1.1 [2024-02-01]

Security Related Fixes

  • Update github.com/moby/buildkit dependency, used for --oci Dockerfile builds, addressing the following upstream CVEs:
    • CVE-2024-23650 Possible panic when incorrect parameters sent from frontend
    • CVE-2024-23651 Possible race condition with accessing subpaths from cache mounts.
    • CVE-2024-23652 Possible host system access from mount stub cleaner.
    • CVE-2024-23653 Interactive containers API does not validate entitlements check.

Note also that in OCI-Mode, SingularityCE may call out to runc versions vulnerable to CVE-2024-21626. runc is not bundled with SingularityCE, and should be updated via your Linux distribution's package manager, or manually.

Bug Fixes

4.1.0 [2024-01-25]

Changed defaults / behaviours

  • --oci mode containers and native mode instances can now be successfully started as a non-root user on cgroups v2 systems when both:

    • The system configuration / environment does not provide the correct information necessary to communicate with systemd via dbus.
    • Resource limits (e.g. --cpus) have not been requested.

    The container / instance will be started in the current cgroup, and information about the configuration issue displayed to the user as warnings.

  • In native mode, SIF/SquashFS container images will now be mounted with squashfuse when kernel mounts are disabled in singularity.conf, or cannot be used (non-setuid / user namespace workflow). If the FUSE mount fails, Singularity will fall back to extracting the container to a temporary sandbox in order to run it.

  • In native mode, bare extfs container images will now be mounted with fuse2fs when kernel mounts are disabled in singularity.conf, or cannot be used (non-setuid / user namespace workflow).

New Features & Functionality

  • The registry login and registry logout commands now support a --authfile <path> flag, which causes the OCI credentials to be written to / removed from a custom file located at <path> instead of the default location ($HOME/.singularity/docker-config.json). The commands pull, push, run, exec, shell, and instance start can now also be passed a --authfile <path> option, to read OCI registry credentials from this custom file.
  • A new --keep-layers flag, for the pull and run/shell/exec/instance start commands, allows individual layers to be preserved when an OCI-SIF image is created from an OCI source. Multi layer OCI-SIF images can be run with SingularityCE 4.1 and later.
  • Singularity will now build OCI-SIF images from Dockerfiles, if the --oci flag is used with the build command. Provide a Dockerfile as the final argument to build, instead of a Singularity definition (.def) file. Supports --build-arg / --build-arg-file options, --arch for cross-architecture builds, --authfile and other authentication options, and more. See the user guide for more information.
  • Docker-style SCIF containers (https://sci-f.github.io/tutorial-preview-install) are now supported. If the entrypoint of an OCI container is the scif executable, then the run / exec / shell commands in --oci mode can be given the --app <appname> flag, and will automatically invoke the relevant SCIF command.
  • A new --tmp-sandbox flag has been added to the run / shell / exec / instance start commands. This will force Singularity to extract a container to a temporary sandbox before running it, when it would otherwise perform a kernel or FUSE mount.

Bug Fixes

  • Added missing tmp sandbox directive to singularity.conf template.

Deprecated Functionality

  • The experimental --sif-fuse flag, and sif fuse directive in singularity.conf are deprecated. The flag and directive were used to enable experimental mounting of SIF/SquashFS container images with FUSE in prior versions of Singularity. From 4.1, FUSE mounts are used automatically when kernel mounts are disabled / not available.

4.0.3 [2024-01-11]

Bug Fixes

  • Use kernel overlayfs instead of fuse-overlayfs when running as root user, regardless of unprivileged kernel overlay support.
  • Execute correct %appstart script when using instance start with --app.

4.0.2 [2023-11-16]

Changed defaults / behaviours

  • Added libnvidia-nvvm to nvliblist.conf. Newer NVIDIA Drivers (known with

    = 525.85.05) require this lib to compile OpenCL programs against NVIDIA GPUs, i.e. libnvidia-opencl depends on libnvidia-nvvm.

Bug Fixes

  • Support parentheses in test / [ commands in container startup scripts, via dependency update of mvdan.cc/sh.
  • Fix incorrect client timeout during remote build context upload.
  • When user requests a bind of /dev:/dev or /dev/xxx:/dev/xxx in OCI-mode, ensure that it is bind mounted with appropriate flags so that it is usable in the container.

4.0.1 [2023-10-13]

New Features & Functionality

  • Added the upcoming NVIDIA driver library libnvidia-gpucomp.so to the list of libraries to add to NVIDIA GPU-enabled containers.

Bug Fixes

  • Don't bind /var/tmp on top of /tmp in the container, where /var/tmp resolves to same location as /tmp.
  • Fix problem where credentials locally stored with registry login command were not usable in some execution flows. Run registry login again with latest version to ensure credentials are stored correctly.
  • Don't fail in a yum bootstrap on systems where the _db_backend rpm macros is not defined (EL <8).

4.0.0 [2023-09-19]

OCI-mode

Singularity 4 introduces OCI-mode as a fully supported feature. It is enabled by using the --oci flag with the run / shell / exec / pull commands, or by setting oci mode = yes in singularity.conf.

In OCI-mode:

  • Container images from OCI sources will be pull-ed to an OCI-SIF file. An OCI-SIF file encapsulates the OCI image configuration and squashed filesystem using an OCI, rather than Singularity specific, structure.
  • The run / shell / exec commands use a low-level OCI runtime (crun/runc) for container execution.
  • Default operation is compatible with other OCI tools, similar to using --compat in Singularity's non-OCI native mode.
  • OCI-modes support running existing Singularity non-OCI-SIF images, and can be made to imitate native mode default behavior by using the --no-compat flag.

OCI-mode changes from 3.11 to 4.0 include:

  • run / shell / exec in OCI-mode now includes support for the following existing CLI flags:
    • --add-caps
    • --drop-caps
    • --keep-privs
    • --no-privs
    • --overlay from directories, bare squashfs and extfs images.
    • --workdir
    • --scratch
    • --no-home
    • --no-mount (dev cannot be disabled in OCI mode)
    • --no-umask (with --no-compat)
    • --writable-tmpfs (with --no-compat)
  • Added --device flag to "action" commands (run/exec/shell) when run in OCI mode (--oci). Currently supports passing one or more (comma-separated) fully-qualified CDI device names, and those devices will then be made available inside the container.
  • Added --cdi-dirs flag to override the default search locations for CDI json files, allowing, for example, users who don't have root access on their host machine to nevertheless create CDI mappings (into containers run with --fakeroot, for example).
  • A container run as root, or with --fakeroot, has OCI default effective/permitted capabilities.
  • An --env-file is evaluated with respect to the host environment, to match native mode behaviour.
  • If the kernel does not support unprivileged overlays, OCI-mode will attempt to use fuse-overlayfs and fusermount for overlay mounting and unmounting.
  • Support for thee SINGULARITY_CONTAINLIBS env var, to specify libraries to bind into /.singularity.d/libs/ in the container.
  • Support for running OCI-SIF images directly from docker://, http://, https:// and oras:// URIs.
  • A new --no-compat flag can be used with OCI-mode to mirror singularity's historic native mode behavior on a variety of settings, instead of setting them the way other OCI runtimes typically do:
    • $HOME, /tmp, /var/tmp are bind mounted from the host.
    • The full /dev is bind mounted from the host, unless mount dev = minimal in singularity.conf (requires crun, not applied with runc).
    • bind path entries in singularity.conf are mounted into the container.
    • The current working directory is mounted into the container, and is the entry point into the container.
    • The container is read-only unless --writable-tmpfs is also used.
    • The host umask is propagated into the container, unless --no-umask is also used.
    • When a native (non-OCI-SIF) image is run in OCI-mode, environment variables will be shell evaluated on container startup.
  • The pull command now accepts a new flag --oci for OCI image sources. This will create an OCI-SIF image rather than convert to Singularity's native container format.
  • OCI-SIF containers can be pushed/pulled to/from OCI registries as single file artifacts using oras:// URIs.
  • OCI-SIF containers can be pushed/pulled to/from registries as OCI images, with a single squashfs layer, using docker:// URIs.
  • A new oci mode directive in singularity.conf can be set to true to enable OCI-mode by default. It can be negated with a new --no-oci command line flag.

See the admin guide and user guide for full requirements of OCI-mode and usage information.

Changed defaults / behaviours

Packages / Requirements

  • RPM packages now use /var/lib/singularity (rather than /var/singularity) to store local state files.
  • Bash completions are now install to the modern share/bash-completion/completions location, rather than under etc.
  • The --vm and related flags to start singularity inside a VM have been removed. This functionality was related to the retired Singularity Desktop / SyOS projects.
  • Singularity uses squashfuse_ll / squashfuse, which is now built from a git submodule unless --without-squashfuse is specified as an argument to mconfig. When built with --without-squashfuse, squashfuse_ll or squashfuse will be located on PATH. Version 0.2.0 or later is required.

CLI

  • The commands related to OCI/Docker registries that were under remote have been moved to their own, dedicated registry command. Run singularity help registry for more information.
  • The remote list subcommand now outputs only remote endpoints (with keyservers and OCI/Docker registries having been moved to separate commands), and the output has been streamlined.
  • Adding a new remote endpoint using the singularity remote add command will now set the new endpoint as default. This behavior can be suppressed by supplying the --no-default (or -n) flag to remote add.
  • The keyserver-related commands that were under remote have been moved to their own, dedicated keyserver command. Run singularity help keyserver for more information.
  • Improved the clarity of singularity key list output.
  • --cwd is now the preferred form of the flag for setting the container's working directory, though --pwd is still supported for compatibility.

Runtime Behaviour

  • The way --home is handled when running as root (e.g. sudo singularity) or with --fakeroot has changed. Previously, we were only modifying the HOME environment variable in these cases, while leaving the container's /etc/passwd file unchanged (with its homedir field pointing to /root, regardless of the value passed to --home). With this change, both the value of HOME and the contents of /etc/passwd in the container will reflect the value passed to --home.
  • Bind mounts are now performed in the order of their occurrence on the command line, or within the value of the SINGULARITY_BIND environment variable. (Previously, image-mounts were always performed first, regardless of order.)
  • Default OCI config generated with singularity mount no longer sets any inheritable / ambient capabilites, matching other OCI runtimes.
  • singularity oci mount now uses, and requires, squashfuse_ll or squashfuse to mount a SIF image to an OCI bundle. Note that squashfuse_ll is built with singularity unless --without-squashfuse is passed to mconfig.
  • The current working directory is created in the container when it doesn't exist, so that it can be entered. You must now specify --no-mount home,cwd instead of just --no-mount home to avoid mounting from $HOME if you run singularity from inside $HOME.
  • If the path of the current working directory in the container and on the host contain symlinks to different locations, it will not be mounted.

New Features & Functionality

  • Templating support for definition files: users can now define variables in definition files via a matching pair of double curly brackets. Variables of the form {{ variable }} will be replaced by a value defined either by a variable=value entry in the %arguments section of the definition file, or through new build options --build-arg or --build-arg-file.
  • Added --secret flag (shorthand: -s) to key remove subcommand, to allow removal of a private key by fingerprint.
  • Added --private as a synonym for --secret in key list, key export, and key remove subcommands.
  • The remote status command will now print the username, realname, and email of the logged-in user, if available.
  • The cache commands now accept --type oci-sif to list and clean cached OCI-SIF image conversions of OCI sources.
  • The instance start command now accepts an optional --app <name> argument which invokes start script within the %appstart <name> section in the definition file. The instance stop command still only requires the instance name.
  • A new --no-pid flag for singularity run/shell/exec disables the PID namespace inferred by --containall and --compat.
  • A new --platform flag can be used to specify an OS/Architecture[/Variant] when pulling images from OCI or library sources. When pulling from library sources the optional variant is ignored.
  • The --arch flag can now be used to specify a required architecture when pulling images from OCI, as well as library sources.
  • Execution flows that unpack an image into a temporary sandbox dir can now be disabled, by setting "tmp sandbox = no" in singularity.conf or by passing --no-tmp-sandbox to the relevant run / shell / exec command.

Developer / API

  • Support for image driver plugins, deprecated at 3.11, has been removed. Unprivileged kernel overlay is supported without a plugin. In singularity.conf, the image driver directive has been removed, and enable overlay no longer supports the driver option.
  • Changes in pkg/build/types.Definition struct. New .FullRaw field introduced, which always contains the raw data for the entire definition file. Behavior of .Raw field has changed: for multi-stage builds parsed with pkg/build/types/parser.All(), .Raw contains the raw content of a single build stage. Otherwise, it is equal to .FullRaw.
  • The SingularityCE go module is now github.com/sylabs/singularity/v4, reflecting the major version of the application.

Bug Fixes

  • Fix interaction between --workdir when given relative path and --scratch.
  • Set correct $HOME in --oci mode when mount home = no in singularity.conf.
  • Lookup and store user/group information in stage one prior to entering any namespaces to fix issue with winbind not correctly lookup user/group information when using user namespace.
  • Caching of OCI images is now architecture aware. This fixes behaviour where a user's home directory is shared between systems of different architectures.
  • Fix compilation with the mconfig -b option (custom builddir).

3.11.5 [2023-09-15]

Changed defaults / behaviours

  • If commands that expect an image file are given an OCI-SIF image, an error will be generated advising the user that this format is only supported in versions 4.0 and up.

Bug Fixes

  • Improved help text for compile and install subcommands of plugin command. Thanks to tonghuaroot (https://github.com/tonghuaroot) for the suggested improvements.

3.11.4 [2023-06-22]

Changed defaults / behaviours

  • Add xino=on mount option for writable kernel overlay mount points to fix inode numbers consistency after kernel cache flush.

New Features & Functionality

  • The tap CNI plugin, new to github.com/containernetworking/plugins v1.3.0, is now provided.
  • Added remote get-login-password subcommand that allows the user to retrieve a CLI token to interact with the OCI registry of a Singularity Enterprise instance.
  • Added --no-setgroups flag for --fakeroot builds and run/shell/exec. This prevents the setgroups syscall being used on the container process in the fakeroot user namespace. Maintains access from within the user namespace to files on the host that have permissions based on supplementary group membership. Note that supplementary groups are mapped to nobody in the container, and chgrp, newgrp, etc. cannot be used.
  • Added ability to set a custom user config directory (default $HOME/.singularity) via the new SINGULARITY_CONFIGDIR environment variable.

Bug Fixes

  • In --oci mode, do not attempt to use unprivileged overlay on systems that do not support it.
  • Fix dropped "n" characters on some platforms in definition file stored as part of SIF metadata.
  • Pass STDIN to --oci containers correctly, to fix piping input to a container.
  • Fix compilation on 32-bit systems.
  • Fix seccomp filters to allow mknod/mknodat syscalls to create pipe/socket and character devices with device number 0 for fakeroot builds.
  • Fix freeze when copying files between stages in an unprivileged proot build.
  • Fix non-POSIX sh operator in mconfig.
  • Correct internal name for CAP_BLOCK_SUSPEND.

3.11.3 [2023-05-04]

Changed defaults / behaviours

  • --oci mode now provides a writable container by default, using a tmpfs overlay. This improves parity with --compat mode in the native runtime, as --compat enables --writable-tmpfs.

Bug Fixes

  • Ensure the allow kernel squashfs directive in singularity.conf applies to encrypted squashfs filesystems in a SIF.

3.11.2 [2023-04-27]

New Features & Functionality

  • OCI mode now supports --hostname (requires UTS namespace, therefore this flag will infer --uts).
  • OCI mode now supports --scratch (shorthand: -S) to mount a tmpfs scratch directory in the container.
  • Support --pwd in OCI mode.
  • OCI mode now supports --home. Supplying a single location (e.g. --home /myhomedir) will result in a new tmpfs directory being created at the specified location inside the container, and that dir being set as the in-container user's home dir. Supplying two locations separated by a colon (e.g. --home /home/user:/myhomedir) will result in the first location on the host being bind-mounted as the second location in-container, and set as the in-container user's home dir.
  • OCI mode now handles --dns and resolv.conf on par with native mode: the --dns flag can be used to pass a comma-separated list of DNS servers that will be used in the container; if this flag is not used, the container will use the same resolv.conf settings as the host.
  • Added allow kernel squashfs directive to singularity.conf. Defaults to yes. When set to no, Singularity will not mount squashfs filesystems using the kernel squashfs driver.
  • Added allow kernel extfs directive to singularity.conf. Defaults to yes. When set to no, Singularity will not mount extfs filesystems using the kernel extfs driver.

Bug Fixes

  • Require runc in RPM packages built on SLES, not crun, because crun is part of the Package Hub community repository that may not be enabled. SingularityCE will still prefer crun if it has been installed.
  • Use /dev/loop-control for loop device creation, to avoid issues with recent kernel patch where max_loop is not set.
  • Always request inner userns in --oci mode without --fakeroot, so that inner id mapping is applied correctly.
  • Use correct target uid/gid for inner id mappings in --oci mode.
  • Avoid runc cgroup creation error when using --oci from a root-owned cgroup (e.g. ssh login session scope).
  • Pass host's TERM environment variable to container in OCI mode. Can be overridden by setting SINGULARITYENV_TERM on host.
  • Honour config passwd and config group directives from singularity.conf in --oci mode.
  • Honour mount proc / mount sys / mount tmp / mount home directives from singularity.conf in --oci mode.
  • Corrected singularity.conf comment, to refer to correct file as source of default capabilities when root default capabilities = file.

3.11.1 [2023-03-14]

New Features & Functionality

  • Add setopt definition file header for the yum bootstrap agent. The setopt value is passed to yum / dnf using the --setopt flag. This permits setting e.g. install_weak_deps=False to bootstrap recent versions of Fedora, where systemd (a weak dependency) cannot install correctly in the container. See examples/Fedora for an example defintion file.
  • Warn user that a yum bootstrap of an older distro may fail if the host rpm _db_backend is not bdb.

Bug Fixes

  • Fix implied --writable-tmpfs with --nvccli, to avoid r/o filesytem error.
  • Avoid incorrect error when requesting fakeroot network.
  • Pass computed LD_LIBRARY_PATH to wrapped unsquashfs. Fixes issues where unsquashfs on host uses libraries in non-default paths.
  • Show correct memory limit in instance stats when a limit is set.
  • Ensure consistent binding of libraries under --nv/--rocm when duplicate <library>.so[.version] files are listed by ldconfig -p.
  • Fix systemd cgroup manager error when running a container as a non-root user with --oci, on systems with cgroups v1 and runc.
  • Fix joining cgroup of instance started as root, with cgroups v1, non-default cgroupfs manager, and no device rules.

Changed defaults / behaviours

  • Show standard output of yum bootstrap if log level is verbose or higher.

3.11.0 [2023-02-10]

Changed defaults / behaviours

  • Image driver plugins, implementing the RegisterImageDriver callback, are deprecated and will be removed in 4.0. Support for the example plugin, permitting Ubuntu unprivileged overlay functionality, has been replaced with direct support for kernel unprivileged overlay.
  • When the kernel supports unprivileged overlay mounts in a user namespace, the container will be constructed using an overlay instead of underlay layout.
  • crun will be used as the low-level OCI runtime, when available, rather than runc. If crun is not available, runc will be used.
  • sessiondir maxsize in singularity.conf now defaults to 64 MiB for new installations. This is an increase from 16 MiB in prior versions.
  • Instances are started in a cgroup, by default, when run as root or when unified cgroups v2 with systemd as manager is configured. This allows singularity instance stats to be supported by default when possible.

New features / functionalities

Image Building

  • Support for a custom hashbang in the %test section of a Singularity recipe (akin to the runscript and start sections).
  • Non-root users can now build from a definition file, on systems that do not support --fakeroot. This requires the statically built proot command (https://proot-me.github.io/) to be available on the user PATH. These builds:
    • Do not support arch / debootstrap / yum / zypper bootstraps. Use localimage, library, oras, or one of the docker/oci sources.
    • Do not support %pre and %setup sections.
    • Run the %post sections of a build in the container as an emulated root user.
    • Run the %test section of a build as the non-root user, like singularity test.
    • Are subject to any restrictions imposed in singularity.conf.
    • Incur a performance penalty due to proot's ptrace based interception of syscalls.
    • May fail if the %post script requires privileged operations that proot cannot emulate.

Instances

  • Instances started by a non-root user can use --apply-cgroups to apply resource limits. Requires cgroups v2, and delegation configured via systemd.
  • A new instance stats command displays basic resource usage statistics for a specified instance, running within a cgroup.

Mounts & Overlays

  • --writable-tmpfs is now available when running unprivileged, or explicitly requesting a user namespace, on systems with a kernel that supports unprivileged overlay mounts in a user namespace.
  • The --no-mount flag now accepts the value bind-paths to disable mounting of all bind path entries in singularity.conf.
  • Persistent overlays (--overlay) from a directory are now available when running unprivileged, or explicitly requesting a user namespace, on systems with a kernel that supports unprivileged overlay mounts in a user namespace.
  • Add --sparse flag to overlay create command to allow generation of a sparse ext3 overlay image.

OCI / Docker Compatibility

  • Support for DOCKER_HOST parsing when using docker-daemon://
  • DOCKER_USERNAME and DOCKER_PASSWORD supported without SINGULARITY_ prefix.
  • A new --oci flag for run/exec/shell enables the experimental OCI runtime mode. This mode:
    • Runs OCI container images from an OCI bundle, using runc or crun.
    • Supports docker://, docker-archive:, docker-daemon:, oci:, oci-archive: image sources.
    • Does not support running Singularity SIF, SquashFS, or EXT3 images.
    • Provides an environment similar to Singularity's native runtime, running with --compat.
    • Supports the following options / flags. Other options are not yet supported:
      • --fakeroot for effective root in the container. Requires subuid/subgid mappings.
      • Bind mounts via --bind or --mount. No image mounts.
      • Additional namespaces requests with --net, --uts, --user.
      • Container environment variables via --env, --env-file, and SINGULARITYENV_ host env vars.
      • --rocm to bind ROCm GPU libraries and devices into the container.
      • --nv to bind Nvidia driver / basic CUDA libraries and devices into the container.
      • --apply-cgroups, and the --cpu*, --blkio*, --memory*, --pids-limit flags to apply resource limits.
  • Instance name is available inside an instance via the new SINGULARITY_INSTANCE environment variable.

Signing & Verification

  • The sign command now supports signing with non-PGP key material by specifying the path to a private key via the --key flag.
  • The verify command now supports verification with non-PGP key material by specifying the path to a public key via the --key flag.
  • The verify command now supports verification with X.509 certificates by specifying the path to a certificate via the --certificate flag. By default, the system root certificate pool is used as trust anchors unless overridden via the --certificate-roots flag. A pool of intermediate certificates that are not trust anchors, but can be used to form a certificate chain can also be specified via the --certificate-intermediates flag.
  • Support for online verification checks of x509 certificates using OCSP protocol. (introduced flag: verify --ocsp-verify)

Other

  • Add new Linux capabilities: CAP_PERFMON, CAP_BPF, CAP_CHECKPOINT_RESTORE.
  • A new --reproducible flag for ./mconfig will configure Singularity so that its binaries do not contain non-reproducible paths. This disables plugin functionality.

Bug Fixes

  • In --rocm mode, the whole of /dev/dri is now bound into the container when --contain is in use. This makes /dev/dri/render devices available, required for later ROCm versions.
  • Overlay is blocked on the panfs filesystem, allowing sandbox directories to be run from panfs without error.
  • Avoid UID / GID readonly var warnings with --env-file.

Development / Testing

  • Significant reduction in the use of network image sources in the e2e tests.
  • Improved parallelization and use of image caches in the e2e tests.
  • The e2e-test makefile target now accepts an argument E2E_GROUPS to only run specified groups of end to end tests. E.g. make -C builddir e2e-test E2E_GROUPS=VERSION,HELP will run end to end tests in the VERSION and HELP groups only.
  • The e2e-test makefile target now accepts an argument E2E_TESTS which is a regular expression specifying the names of (top level) end to end tests that should be run. E.g. make -C builddir e2e-test E2E_TESTS=^semantic will only run end to end tests with a name that begins with semantic. These E2E_ variables offer an alternative to the -run flag, which may be easier to use given the structure of e2e tests.

3.10.5 [2022-01-17]

Security Related Fixes

  • CVE-2022-23538: The github.com/sylabs/scs-library-client dependency included in SingularityCE >=3.10.0, <3.10.5 may leak user credentials to a third-party service via HTTP redirect. This issue is limited to library:// access to specific Singularity Enterprise 1.x or 3rd party library configurations, which implement a concurrent multi-part download flow. Access to Singularity Enterprise 2.x, or Singularity Container Services (cloud.sylabs.io), does not trigger the vulnerable flow. See the linked advisory for full details.

3.10.4 [2022-11-10]

Bug Fixes

  • Ensure make dist doesn't include conmon binary or intermediate files.
  • Do not hang on pull from http(s) source that doesn't provide a content-length.
  • Avoid hang on fakeroot cleanup under high load seen on some distributions / kernels.

3.10.3 [2022-10-06]

Security Related Fixes

  • CVE-2022-39237: The github.com/sylabs/sif/v2 dependency included in SingularityCE <=3.10.3 does not verify that the hash algorithm(s) used are cryptographically secure when verifying digital signatures. This release updates to sif v2.8.1 which corrects this issue. See the linked advisory for references and a workaround.

Bug Fixes

  • Ensure bootstrap_history directory is populated with previous definition files, present in source containers used in a build.

3.10.2 [2022-07-25]

New features / functionalities

  • Added EL9 package builds to CI for GitHub releases.

Bug Fixes

  • Ensure no empty if branch is present in generated OCI image runscripts. Would prevent execution of container by other tools that are not using mvdan.cc/sh.

3.10.1 [2022-07-18]

New features / functionalities

  • Debug output can now be enabled by setting the SINGULARITY_DEBUG env var.
  • Debug output is now shown for nested singularity calls, in wrapped unsquashfs image extraction, and build stages.

Bug Fixes

  • Fix test code that implied %test -c <shell> was supported - it is not.
  • Fix compilation on mipsel.

3.10.0 [2022-05-17]

Changed defaults / behaviours

  • master branch of GitHub repository has been renamed to main.
  • oci mount sets Process.Terminal: true when creating an OCI config.json, so that oci run provides expected interactive behavior by default.
  • Default hostname for oci mount containers is now singularity instead of mrsdalloway.
  • systemd is now supported and used as the default cgroups manager. Set systemd cgroups = no in singularity.conf to manage cgroups directly via the cgroupfs.
  • The singularity oci command group now uses runc to manage containers.
  • The singularity oci commands use conmon which is built from a git submodule, unless --without-conmon is specified as an argument to mconfig, in which case Singularity will search PATH for conmon. Version >=2.0.24 of conmon is required.
  • The singularity oci flags --sync-socket, --empty-process, and --timeout have been removed.
  • Don't prompt for y/n to overwrite an existing file when build is called from a non-interactive environment. Fail with an error.
  • Plugins must be compiled from inside the SingularityCE source directory, and will use the main SingularityCE go.mod file. Required for Go 1.18 support.
  • seccomp support is not disabled automatically in the absence of seccomp headers at build time. Run mconfig using --without-seccomp and --without-conmon to disable seccomp support and building of conmon (which requires seccomp headers).
  • SingularityCE now requires squashfs-tools >=4.3, which is satisfied by current EL / Ubuntu / Debian and other distributions.
  • Added --no-eval to the list of flags set by the OCI/Docker --compat mode (see below).

New features / functionalities

  • Updated seccomp support allows use of seccomp profiles that set an error return code with errnoRet and defaultErrnoRet. Previously EPERM was hard coded. The example etc/seccomp-profiles/default.json has been updated.
  • Native cgroups v2 resource limits can be specified using the [unified] key in a cgroups toml file applied via --apply-cgroups.
  • The --no-mount flag & SINGULARITY_NO_MOUNT env var can now be used to disable a bind path entry from singularity.conf by specifying the absolute path to the destination of the bind.
  • Non-root users can now use --apply-cgroups with run/shell/exec to limit container resource usage on a system using cgroups v2 and the systemd cgroups manager.
  • Added --cpu*, --blkio*, --memory*, --pids-limit flags to apply cgroups resource limits to a container directly.
  • Allow experimental direct mount of SIF images with squashfuse in user-namespace / no-setuid mode.
  • New action flag --no-eval which:
    • Prevents shell evaluation of SINGULARITYENV_ / --env / --env-file environment variables as they are injected in the container, to match OCI behavior. Applies to all containers.
    • Prevents shell evaluation of the values of CMD / ENTRYPOINT and command line arguments for containers run or built directly from an OCI/Docker source. Applies to newly built containers only, use singularity inspect to check version that container was built with.
  • Add support for %files section in remote builds, when a compatible remote is used.

Bug Fixes

  • Allow newgidmap / newuidmap that use capabilities instead of setuid root.
  • Corrected key search output for results from some servers, and keys with multiple names.
  • Pass through a literal \n in host environment variables to container.
  • Address 401 error pulling from private library:// projects.
  • Correctly launch CleanupHost process only when needed in --sif-fuse flow.
  • Add specific error for unreadable image / overlay file.
  • Ensure cgroups device limits are default allow per past behavior.
  • Improve error message when remote build server does not support the %files section.
  • Fix non-root instance join with unprivileged systemd managed cgroups, when join is from outside a user-owned cgroup.

v3.9.9 [2022-04-22]

Bug Fixes

  • Use HEAD request when checking digest of remote OCI image sources, with GET as a fall-back. Greatly reduces Singularity's impact on Docker Hub API limits.

New features / functionalities

  • Add package build for Ubuntu 22.04 LTS.

v3.9.8 [2022-04-07]

Bug fixes

  • Do not truncate environment variables with commas.
  • Fix error when pushing to host-less library:// URIs.

v3.9.7 [2022-03-23]

Bug fixes

  • Support nvidia-container-cli v1.8.0 and above, via fix to capability set.
  • Avoid cleanup panic when invalid file specified for --apply-cgroups.

v3.9.6 [2022-03-10]

New features / functionalities

  • SingularityCE now supports the riscv64 architecture.

Bug fixes

  • Correct library bindings for unsquashfs containment. Fixes errors where resolved library filename does not match library filename in binary (e.g. EL8, POWER9 with glibc-hwcaps).

v3.9.5 [2022-02-04]

Changed defaults / behaviours

  • make install now installs man pages. A separate make man is not required.

Bug fixes

  • GitHub .deb packages correctly include man pages.
  • Update dependency to correctly unset variables in container startup environment processing. Fixes regression in v3.9.2.
  • Remove subshell overhead when processing large environments on container startup.

v3.9.4 [2022-01-19]

Bug fixes

  • Address timeout in library pull single stream download.

v3.9.3 [2022-01-11]

Bug fixes

  • Ensure MIGs are visible with --nvccli in non-contained mode, to match the legacy GPU binding behaviour.
  • Avoid fd leak in loop device transient error path.

v3.9.2 [2021-12-10]

Bug fixes

  • Ensure gengodep in build uses vendor dir when present.
  • Fix source of a script on PATH and scoping of environment variables in definition files (via dependency update).
  • Ensure a local build does not fail unnecessarily if a keyserver config cannot be retrieved from the remote endpoint.
  • Correct documentation for sign command r.e. source of key index.
  • Restructure loop device discovery to address an issue where a transient EBUSY error could lead to failure under Arvados. Also greedily try for a working loop device, rather than perform delayed retries on encountering EAGAIN, since we hold an exclusive lock which can block other processes.

v3.9.1 [2021-11-22]

This is a security release for SingularityCE 3.9, addressing a security issue in SingularityCE's dependencies.

Security Related Fixes

  • CVE-2021-41190 / GHSA-77vh-xpmg-72qh: OCI specifications allow ambiguous documents that contain both "manifests" and "layers" fields. Interpretation depends on the presence / value of a Content-Type header. SingularityCE dependencies handling the retrieval of OCI images have been updated to versions that reject ambiguous documents.

v3.9.0 [2021-11-16]

This is the first release of SingularityCE 3.9, the Community Edition of the Singularity container runtime hosted at https://github.com/sylabs/singularity.

Changed defaults / behaviours

  • Building SingularityCE 3.9.0 requires go >=1.16. We now aim to support the two most recent stable versions of Go. This corresponds to the Go Release Maintenance Policy and Security Policy, ensuring critical bug fixes and security patches are available for all supported language versions.
  • LABELs from Docker/OCI images are now inherited. This fixes a longstanding regression from Singularity 2.x. Note that you will now need to use --force in a build to override a label that already exists in the source Docker/OCI container.
  • The source paths for %files lines in a definition file are no longer interpreted by a shell. This means that environment variable substitution is not performed. Previously, environment variables were substituted for source paths, but not destination paths, leading to unexpected copy behaviour. Globbing for source files will now follow the Go filepath.Match pattern syntax.
  • Removed --nonet flag, which was intended to disable networking for in-VM execution, but has no effect.
  • --nohttps flag has been deprecated in favour of --no-https. The old flag is still accepted, but will display a deprecation warning.
  • Paths for cryptsetup, go, ldconfig, mksquashfs, nvidia-container-cli, unsquashfs are now found at build time by mconfig and written into singularity.conf. The path to these executables can be overridden by changing the value in singularity.conf.
  • When calling ldconfig to find GPU libraries, singularity will not fall back to /sbin/ldconfig if the configured ldconfig errors. If installing in a Guix/Nix on environment on top of a standard host distribution you must set ldconfig path = /sbin/ldconfig to use the host distribution ldconfig to find GPU libraries.
  • --nv will not call nvidia-container-cli to find host libraries, unless the new experimental GPU setup flow that employs nvidia-container-cli for all GPU related operations is enabled (see below).
  • If a container is run with --nvcli and --contain, only GPU devices specified via the NVIDIA_VISIBLE_DEVICES environment variable will be exposed within the container. Use NVIDIA_VISIBLE_DEVICES=all to access all GPUs inside a container run with --nvccli.
  • Example log-plugin rewritten as a CLI callback that can log all commands executed, instead of only container execution, and has access to command arguments.
  • The bundled reference CNI plugins are updated to v1.0.1. The flannel plugin is no longer included, as it is maintained as a separate plugin at: https://github.com/flannel-io/cni-plugin. If you use the flannel CNI plugin you should install it from this repository.
  • Instances are no longer created with an IPC namespace by default. An IPC namespace can be specified with the -i|--ipc flag.
  • The behaviour of the allow container directives in singularity.conf has been modified, to support more intuitive limitations on the usage of SIF and non-SIF container images. If you use these directives, you may need to make changes to singularity.conf to preserve behaviour.
    • A new allow container sif directive permits or denies usage of unencrypted SIF images, irrespective of the filesystem(s) inside the SIF.
    • The allow container encrypted directive permits or denies usage of SIF images with an encrypted root filesystem.
    • The allow container squashfs/extfs directives in singularity.conf permit or deny usage of bare SquashFS and EXT image files only.
    • The effect of the allow container dir directive is unchanged.

New features / functionalities

  • --writable-tmpfs can be used with singularity build to run the %test section of the build with a ephemeral tmpfs overlay, permitting tests that write to the container filesystem.
  • The --compat flag for actions is a new short-hand to enable a number of options that increase OCI/Docker compatibility. Infers --containall, --no-init, --no-umask, --writable-tmpfs. Does not use user, uts, or network namespaces as these may not be supported on many installations.
  • remote add --insecure may be used to configure endpoints that are only accessible via http.
  • The experimental --nvccli flag will use nvidia-container-cli to setup the container for Nvidia GPU operation. SingularityCE will not bind GPU libraries itself. Environment variables that are used with Nvidia's docker-nvidia runtime to configure GPU visibility / driver capabilities & requirements are parsed by the --nvccli flag from the environment of the calling user. By default, the compute and utility GPU capabilities are configured. The use nvidia-container-cli option in singularity.conf can be set to yes to always use nvidia-container-cli when supported. Note that in a setuid install, nvidia-container-cli will be run as root with required ambient capabilities. --nvccli is not currently supported in the hybrid fakeroot (setuid install + --fakeroot) workflow. Please see documentation for more details.
  • The --apply-cgroups flag can be used to apply cgroups resource and device restrictions on a system using the v2 unified cgroups hierarchy. The resource restrictions must still be specified in the v1 / OCI format, which will be translated into v2 cgroups resource restrictions, and eBPF device restrictions.
  • A new --mount flag and SINGULARITY_MOUNT environment variable can be used to specify bind mounts in type=bind,source=<src>,destination=<dst>[,options...] format. This improves CLI compatibility with other runtimes, and allows binding paths containing : and , characters (using CSV style escaping).
  • Perform concurrent multi-part downloads for library:// URIs. Uses 3 concurrent downloads by default, and is configurable in singularity.conf or via environment variables.

Bug fixes

  • The oci commands will operate on systems that use the v2 unified cgroups hierarchy.
  • Ensure invalid values passed to config global --set cannot lead to an empty configuration file being written.
  • An invalid remote build source (bootstrap) will be identified before attempting to submit the build.
  • --no-https now applies to connections made to library services specified in library://<hostname>/... URIs.

v3.8.4 [2021-10-28]

Bug fixes

  • Update oras-go dependency to address push failures to some registry configurations.
  • Implement context cancellation when a signal is received in several CLI commands.

v3.8.3 [2021-09-01]

Bug fixes

  • Fix regression when files sourced from %environment contain \ escaped shell builtins (fixes issue with source of conda profile.d script).

Additional changes include dependency updates for the SIF module (to v2.0.0), and migration to maintained versions of other modules. There is no change to functionality, on-disk SIF format etc.

v3.8.2 [2021-08-19]

Bug fixes

  • singularity delete will use the correct library service when the hostname is specified in the library:// URI.
  • singularity build will use the correct library service when the hostname is specified in the library:// URI / definition file.
  • Fix download of default pacman.conf in arch bootstrap.
  • Call debootstrap with correct Debian arch when it is not identical to the value of runtime.GOARCH. E.g. ppc64el -> ppc64le.
  • When destination is omitted in %files entry in definition file, ensure globbed files are copied to correct resolved path.
  • Return an error if --tokenfile used for remote login to an OCI registry, as this is not supported.
  • Ensure repeated remote login to same URI does not create duplicate entries in ~/.singularity/remote.yaml.
  • Avoid panic when mountinfo line has a blank field.
  • Properly escape single quotes in Docker CMD / ENTRYPOINT translation.
  • Use host uid when choosing unsquashfs flags, to avoid selinux xattr errors with --fakeroot on non-EL/Fedora distributions with recent squashfs-tools.

v3.8.1 [2021-07-20]

Bug Fixes

  • Allow escaped \$ in a SINGULARITYENV_ var to set a literal $ in a container env var.
  • Handle absolute symlinks correctly in multi-stage build %copy from blocks.
  • Fix incorrect reference in sandbox restrictive permissions warning.

v3.8.0 [2021-05-26]

This is the first release of SingularityCE 3.8.0, the Community Edition of the Singularity container runtime hosted at https://github.com/sylabs/singularity.

Changed defaults / behaviours

  • The package name for this release is now singularity-ce. This name is used for the source tarball, output of an rpmbuild, and displayed in --version information.
  • The name of the top level directory in the source tarball from make dist now includes the version string.

New features / functionalities

  • A new overlay command allows creation and addition of writable overlays.
  • Administrators can allow named users/groups to use specific CNI network configurations. Managed by directives in singularity.conf.
  • The build command now honors --nv, --rocm, and --bind flags, permitting builds that require GPU access or files bound in from the host.
  • A library service hostname can be specified as the first component of a library:// URL.
  • Singularity is now relocatable for unprivileged installations only.

Bug Fixes

  • Respect http proxy server environment variables in key operations.
  • When pushing SIF images to oras:// endpoints, work around Harbor & GitLab failure to accept the SifConfigMediaType.
  • Avoid a setfsuid compilation warning on some gcc versions.
  • Fix a crash when silent/quiet log levels used on pulls from shub:// and http(s):// URIs.
  • Wait for dm device to appear when mounting an encrypted container rootfs.

Testing / Development

Testing changes are not generally itemized. However, developers and contributors should note that this release has modified the behavior of make test for ease of use:

  • make test runs limited unit and integration tests that will not require docker hub credentials.
  • make testall runs the full unit/integration/e2e test suite that requires docker credentials to be set with E2E_DOCKER_USERNAME and E2E_DOCKER_PASSWORD environment variables.
  • Fix privilege handling issue with tests on Go >=1.16.

v3.7.4 - [2021-05-26]

Singularity 3.7.4 is the most recent stable release of Singularity prior to Sylabs' fork from https://github.com/hpcng/singularity

The 3.7.4 release is identical to https://github.com/hpcng/singularity/releases/tag/v3.7.4 and is provided for convenience to users arriving from outdated links.

Security Related Fixes

  • CVE-2021-32635: Due to incorrect use of a default URL, singularity action commands (run/shell/exec) specifying a container using a library:// URI will always attempt to retrieve the container from the default remote endpoint (cloud.sylabs.io) rather than the configured remote endpoint. An attacker may be able to push a malicious container to the default remote endpoint with a URI that is identical to the URI used by a victim with a non-default remote endpoint, thus executing the malicious container.

v3.7.3 - [2021-04-06]

Security Related Fixes

  • CVE-2021-29136: A dependency used by Singularity to extract docker/OCI image layers can be tricked into modifying host files by creating a malicious layer that has a symlink with the name "." (or "/"), when running as root. This vulnerability affects a singularity build or singularity pull as root, from a docker or OCI source.

v3.7.2 - [2021-03-09]

Bug Fixes

  • Fix progress bar display when source image size is unknown.
  • Fix a memory usage / leak issue when building from an existing image file.
  • Fix to allow use of --library flag to point push/pull at default cloud library when another remote is in use.
  • Address false positive loop test errors, and an e2e test registry setup issue.

v3.7.1 - [2021-01-12]

Bug Fixes

  • Accommodate /sys/fs/selinux mount changes on kernel 5.9+.
  • Fix loop devices file descriptor leak when shared loop devices is enabled.
  • Use MaxLoopDevices variable from config file in all appropriate locations.
  • Use -buildmode=default (non pie) on ppc64le to prevent crashes when using plugins.
  • Remove spurious warning in parseTokenSection()
  • e2e test fixes for new kernels, new unsquashfs version.
  • Show correct web URI for detached builds against alternate remotes.

New features / functionalities

  • The singularity binary is now relocatable when built without setuid support

v3.7.0 - [2020-11-24]

New features / functionalities

  • Allow configuration of global custom keyservers, separate from remote endpoints.
  • Add a new global keyring, for public keys only (used for ECL).
  • The remote login command now supports authentication to Docker/OCI registries and custom keyservers.
  • New --exclusive option for remote use allows admin to lock usage to a specific remote.
  • A new Fingerprints: header in definition files will check that a SIF source image can be verified, and is signed with keys matching all specified fingerprints.
  • Labels can be set dynamically from a build's %post section by setting them in the SINGULARITY_LABELS environment variable.
  • New build-arch label is automatically set to the architecture of the host during a container build.
  • New -D/--description flag for singularity push sets description for a library container image.
  • singularity remote status shows validity of authentication token if set.
  • singularity push reports quota usage and URL on successful push to a library server that supports this.
  • A new --no-mount flag for actions allows a user to disable proc/sys/dev/devpts/home/tmp/hostfs/cwd mounts, even if they are enabled in singularity.conf.

Changed defaults / behaviours

  • When actions (run/shell/exec...) are used without --fakeroot the umask from the calling environment will be propagated into the container, so that files are created with expected permissions. Use the new --no-umask flag to return to the previous behaviour of setting a default 0022 umask.
  • Container metadata, environment, scripts are recorded in a descriptor in builds to SIF files, and inspect will use this if present.
  • The --nv flag for NVIDIA GPU support will not resolve libraries reported by nvidia-container-cli via the ld cache. Will instead respect absolute paths to libraries reported by the tool, and bind all versioned symlinks to them.
  • General re-work of the remote login flow, adds prompts and token verification before replacing an existing authentication token.
  • The Execution Control List (ECL) now verifies container fingerprints using the new global keyring. Previously all users would need relevant keys in their own keyring.
  • The SIF layer mediatype for ORAS has been changed to application/vnd.sylabs.sif.layer.v1.sif reflecting the published opencontainers/artifacts value.
  • SINGULARITY_BIND has been restored as an environment variable set within a running container. It now reflects all user binds requested by the -B/--bind flag, as well as via SINGULARITY_BIND[PATHS].
  • singularity search now correctly searches for container images matching the host architecture by default. A new --arch flag allows searching for other architectures. A new results format gives more detail about container image results, while users and collections are no longer returned.

Bug Fixes

  • Support larger definition files, environments etc. by passing engine configuration in the environment vs. via socket buffer.
  • Ensure docker-daemon: and other source operations respect SINGULARITY_TMPDIR for all temporary files.
  • Support double quoted filenames in the %files section of build definitions.
  • Correct cache list sizes to show KiB with powers of 1024, matching du etc.
  • Don't fail on enable fusemount=no when no fuse mounts are needed.
  • Pull OCI images to the correct requested location when the cache is disabled.
  • Ensure Singularity> prompt is set when container has no environment script, or singularity is called through a wrapper script.
  • Avoid build failures in yum/dnf operations against the 'setup' package on RHEL/CentOS/Fedora by ensuring staged /etc/ files do not match distro default content.
  • Failed binds to /etc/hosts and /etc/localtime in a container run with --contain are no longer fatal errors.
  • Don't initialize the cache for actions where it is not required.
  • Increase embedded shell interpreter timeout, to allow slow-running environment scripts to complete.
  • Correct buffer handling for key import to allow import from STDIN.
  • Reset environment to avoid LD_LIBRARY_PATH issues when resolving dependencies for the unsquashfs sandbox.
  • Fall back to /sbin/ldconfig if ldconfig on PATH fails while resolving GPU libraries. Fixes problems on systems using Nix / Guix.
  • Address issues caused by error code changes in unsquashfs version 4.4.
  • Ensure /dev/kfd is bound into container for ROCm when --rocm is used with --contain.
  • Tolerate comments on %files sections in build definition files.
  • Fix a loop device file descriptor leak.

Known Issues

  • A change in Linux kernel 5.9 causes --fakeroot builds to fail with a /sys/fs/selinux remount error. This will be addressed in Singularity v3.7.1.

v3.6.4 - [2020-10-13]

Security related fixes

Singularity 3.6.4 addresses the following security issue.

  • CVE-2020-15229: Due to insecure handling of path traversal and the lack of path sanitization within unsquashfs (a distribution provided utility used by Singularity), it is possible to overwrite/create files on the host filesystem during the extraction of a crafted squashfs filesystem. Affects unprivileged execution of SIF / SquashFS images, and image builds from SIF / SquashFS images.

Bug Fixes

  • Update scs-library-client to support library:// backends using an 3rd party S3 object store that does not strictly conform to v4 signature spec.

v3.6.3 - [2020-09-15]

Security related fixes

Singularity 3.6.3 addresses the following security issues.

  • CVE-2020-25039: When a Singularity action command (run, shell, exec) is run with the fakeroot or user namespace option, Singularity will extract a container image to a temporary sandbox directory. Due to insecure permissions on the temporary directory it is possible for any user with access to the system to read the contents of the image. Additionally, if the image contains a world-writable file or directory, it is possible for a user to inject arbitrary content into the running container.

  • CVE-2020-25040: When a Singularity command that results in a container build operation is executed, it is possible for a user with access to the system to read the contents of the image during the build. Additionally, if the image contains a world-writable file or directory, it is possible for a user to inject arbitrary content into the running build, which in certain circumstances may enable arbitrary code execution during the build and/or when the built container is run.

Change defaults / behaviours

  • The value for maximum number of loop devices in the config file is now used everywhere instead of redefining this value

Bug Fixes

  • Add CAP_MKNOD in capability bounding set of RPC to fix issue with cryptsetup when decrypting image from within a docker container.
  • Fix decryption issue when using both IPC and PID namespaces.
  • Fix unsupported builtins panic from shell interpreter and add umask support for definition file scripts.
  • Do not load keyring in prepare_linux if ECL not enabled.
  • Ensure sandbox option overrides remote build destination.

v3.6.2 - [2020-08-25]

New features / functionalities

  • Add --force option to singularity delete for non-interactive workflows.

Change defaults / behaviours

  • Default to current architecture for singularity delete.

Bug Fixes

  • Respect current remote for singularity delete command.
  • Allow rw as a (noop) bind option.
  • Fix capability handling regression in overlay mount.
  • Fix LD_LIBRARY_PATH environment override regression with --nv/--rocm.
  • Fix environment variable duplication within singularity engine.
  • Use -user-xattrs for unsquashfs to avoid error with rootless extraction using unsquashfs 3.4 (Ubuntu 20.04).
  • Correct --no-home message for 3.6 CWD behavior.
  • Don't fail if parent of cache dir not accessible.
  • Fix tests for Go 1.15 Ctty handling.
  • Fix additional issues with test images on ARM64.
  • Fix FUSE e2e tests to use container ssh_config.

v3.6.1 - [2020-07-21]

New features / functionalities

  • Support compilation with FORTIFY_SOURCE=2 and build in pie mode with fstack-protector enabled (#5433).

Bug Fixes

  • Provide advisory message r.e. need for upper and work to exist in overlay images.
  • Use squashfs mem and processor limits in squashfs gzip check.
  • Ensure build destination path is not an empty string - do not overwrite CWD.
  • Don't unset PATH when interpreting legacy /environment files.

v3.6.0 - [2020-07-14]

Security related fixes

Singularity 3.6.0 introduces a new signature format for SIF images, and changes to the signing / verification code to address:

  • CVE-2020-13845 In Singularity 3.x versions below 3.6.0, issues allow the ECL to be bypassed by a malicious user.
  • CVE-2020-13846 In Singularity 3.5 the --all / -a option to singularity verify returns success even when some objects in a SIF container are not signed, or cannot be verified.
  • CVE-2020-13847 In Singularity 3.x versions below 3.6.0, Singularity's sign and verify commands do not sign metadata found in the global header or data object descriptors of a SIF file, allowing an attacker to cause unexpected behavior. A signed container may verify successfully, even when it has been modified in ways that could be exploited to cause malicious behavior.

Please see the published security advisories at https://github.com/hpcng/singularity/security/advisories for full detail of these security issues.

Note that the new signature format is necessarily incompatible with Singularity < 3.6.0 - e.g. Singularity 3.5.3 cannot verify containers signed by 3.6.0.

We thank Tru Huynh for a report that led to the review of, and changes to, the signature implementation.

New features / functionalities

  • Singularity now supports the execution of minimal Docker/OCI containers that do not contain /bin/sh, e.g. docker://hello-world.
  • A new cache structure is used that is concurrency safe on a filesystem that supports atomic rename. If you downgrade to Singularity 3.5 or older after using 3.6 you will need to run singularity cache clean.
  • A plugin system rework adds new hook points that will allow the development of plugins that modify behavior of the runtime. An image driver concept is introduced for plugins to support new ways of handling image and overlay mounts. Plugins built for <=3.5 are not compatible with 3.6.
  • The --bind flag can now bind directories from a SIF or ext3 image into a container.
  • The --fusemount feature to mount filesystems to a container via FUSE drivers is now a supported feature (previously an experimental hidden flag). This permits users to mount e.g. sshfs and cvmfs filesystems to the container at runtime.
  • A new -c/--config flag allows an alternative singularity.conf to be specified by the root user, or all users in an unprivileged installation.
  • A new --env flag allows container environment variables to be set via the Singularity command line.
  • A new --env-file flag allows container environment variables to be set from a specified file.
  • A new --days flag for cache clean allows removal of items older than a specified number of days. Replaces the --name flag which is not generally useful as the cache entries are stored by hash, not a friendly name.
  • A new '--legacy-insecure' flag to verify allows verification of SIF signatures in the old, insecure format.
  • A new '-l / --logs' flag for instance list that shows the paths to instance STDERR / STDOUT log files.
  • The --json output of instance list now include paths to STDERR / STDOUT log files.

Changed defaults / behaviours

  • New signature format (see security fixes above).
  • Environment variables prefixed with SINGULARITYENV_ always take precedence over variables without SINGULARITYENV_ prefix.
  • The %post build section inherits environment variables from the base image.
  • %files from ... will now follow symlinks for sources that are directly specified, or directly resolved from a glob pattern. It will not follow symlinks found through directory traversal. This mirrors Docker multi-stage COPY behaviour.
  • Restored the CWD mount behaviour of v2, implying that CWD path is not recreated inside container and any symlinks in the CWD path are not resolved anymore to determine the destination path inside container.
  • The %test build section is executed the same manner as singularity test image.
  • --fusemount with the container: default directive will foreground the FUSE process. Use container-daemon: for previous behavior.
  • Fixed spacing of singularity instance list to be dynamically changing based off of input lengths instead of fixed number of spaces to account for long instance names.

Deprecated / removed commands

  • Removed --name flag for cache clean; replaced with --days.
  • Deprecate -a / --all option to sign/verify as new signature behavior makes this the default.

Bug Fixes

  • Don't try to mount $HOME when it is / (e.g. nobody user).
  • Process %appinstall sections in order when building from a definition file.
  • Ensure SINGULARITY_CONTAINER, SINGULARITY_ENVIRONMENT and the custom shell prompt are set inside a container.
  • Honor insecure registry settings from /etc/containers/registries.conf.
  • Fix http_proxy env var handling in yum bootstrap builds.
  • Disable log colorization when output location is not a terminal.
  • Check encryption keys are usable before beginning an encrypted build.
  • Allow app names with non-alphanumeric characters.
  • Use the base metapackage for arch bootstrap builds - arch no longer has a base group.
  • Ensure library client messages are logged with --debug.
  • Do not mount $HOME with --fakeroot --contain.
  • Fall back to underlay automatically when using a sandbox on GPFS.
  • Fix Ctrl-Z handling - propagation of signal.

v3.5.3 - [2020-02-18]

Changed defaults / behaviours

The following minor behaviour changes have been made in 3.5.3 to allow correct operation on CRAY CLE6, and correct an issue with multi-stage image builds that was blocking use by build systems such as Spack:

  • Container action scripts are no longer bound in from etc/actions.d on the host. They are created dynamically and inserted at container startup.
  • %files from ... will no longer follow symlinks when copying between stages in a multi stage build, as symlinks should be copied so that they resolve identically in later stages. Copying %files from the host will still maintain previous behavior of following links.

Bug Fixes

  • Bind additional CUDA 10.2 libs when using the --nv option without nvidia-container-cli.
  • Fix an NVIDIA persistenced socket bind error with --writable.
  • Add detection of ceph to allow workarounds that avoid issues with sandboxes on ceph filesystems.
  • Ensure setgid is inherited during make install.
  • Ensure the root directory of a build has owner write permissions, regardless of the permissions in the bootstrap source.
  • Fix a regression in %post and %test to honor the -c option.
  • Fix an issue running %post when a container doesn't have /etc/resolv.conf or /etc/hosts files.
  • Fix an issue with UID detection on RHEL6 when running instances.
  • Fix a logic error when a sandbox image is in an overlay incompatible location, and both overlay and underlay are disabled globally.
  • Fix an issue causing user namespace to always be used when allow-setuid=no was configured in a setuid installation.
  • Always allow key IDs and fingerprints to be specified with or without a 0x prefix when using singularity keys
  • Fix an issue preventing joining an instance started with --boot.
  • Provide a useful error message if an invalid library:// path is provided.
  • Bring in multi-part upload client functionality that will address large image upload / proxied upload issues with a future update to Sylabs cloud.

In addition, numerous improvements have been made to the test suites, allowing them to pass cleanly on a range of kernel versions and distributions that are not covered by the open-source CI runs.

v3.5.2 - [2019-12-17]

Security related fix

  • 700 permissions are enforced on $HOME/.singularity and SINGULARITY_CACHEDIR directories (CVE-2019-19724). Many thanks to Stuart Barkley for reporting this issue.

Bug Fixes

  • Fixes an issue preventing use of .docker/config for docker registry authentication.

  • Fixes the run-help command in the unprivileged workflow.

  • Fixes a regression in the inspect command to support older image formats.

  • Adds a workaround for an EL6 kernel bug regarding shared bind mounts.

  • Fixes caching of http(s) sources with conflicting filenames.

  • Fixes a fakeroot sandbox build error on certain filesystems, e.g. lustre, GPFS.

  • Fixes a fakeroot build failure to a sandbox in $HOME.

  • Fixes a fakeroot build failure from a bad def file section script location.

  • Fixes container execution errors when CWD is a symlink.

  • Provides a useful warning r.e. possible fakeroot build issues when seccomp support is not available.

  • Fixes an issue where the --disable-cache option was not being honored.

  • Deprecated --groupid flag for sign and verify; replaced with --group-id.

  • Removed useless flag --url for sign.

v3.5.1 - [2019-12-05]

New features / functionalities

A single feature has been added in the bugfix release, with specific functionality:

  • A new option allow container encrypted can be set to no in singularity.conf to prevent execution of encrypted containers.

Bug Fixes

This point release addresses the following issues:

  • Fixes a disk space leak when building from docker-archive.
  • Makes container process SIGABRT return the expected code.
  • Fixes the inspect command in unprivileged workflow.
  • Sets an appropriate default umask during build stages, to avoid issues with very restrictive user umasks.
  • Fixes an issue with build script content being consumed from STDIN.
  • Corrects the behaviour of underlay with non-empty / symlinked CWD and absolute symlink binds targets.
  • Fixes execution of containers when binding BTRFS filesystems.
  • Fixes build / check failures for MIPS & PPC64.
  • Ensures file ownership maintained when building image from sandbox.
  • Fixes a squashfs mount error on kernel 5.4.0 and above.
  • Fixes an underlay fallback problem, which prevented use of sandboxes on lustre filesystems.

v3.5.0 - [2019-11-13]

New features / functionalities

  • New support for AMD GPUs via --rocm option added to bind ROCm devices and libraries into containers.
  • Plugins can now modify Singularity behaviour with two mutators: CLI and Runtime.
  • Introduced the config global command to edit singularity.conf settings from the CLI.
  • Introduced the config fakeroot command to setup subuid and subgid mappings for --fakeroot from the Singularity CLI.

Changed defaults / behaviours

  • Go 1.13 adopted.
  • Vendored modules removed from the Git tree, will be included in release tarballs.
  • Singularity will now fail with an error if a requested bind mount cannot be made.
    • This is beneficial to fail fast in workflows where a task may fail a long way downstream if a bind mount is unavailable.
    • Any unavailable bind mount sources must be removed from singularity.conf.
  • Docker/OCI image extraction now faithfully respects layer permissions.
    • This may lead to sandboxes that cannot be removed without modifying permissions.
    • --fix-perms option added to preserve old behaviour when building sandboxes.
    • Discussion issue for this change at: https://github.com/sylabs/singularity/issues/4671
  • Singularity> prompt is always set when entering shell in a container.
  • The current umask will be honored when building a SIF file.
  • instance exec processes acquire cgroups set on instance start
  • --fakeroot supports uid/subgid ranges >65536
  • singularity version now reports semver compliant version information.

Deprecated / removed commands

  • Deprecated --id flag for sign and verify; replaced with --sif-id.

v3.4.2 - [2019-10-08]

  • This point release addresses the following issues:
    • Sets workable permissions on OCI -> sandbox rootless builds
    • Fallback correctly to user namespace for non setuid installation
    • Correctly handle the starter-suid binary for non-root installs
    • Creates CACHEDIR if it doesn't exist
    • Set apex loglevel for umoci to match singularity loglevel

v3.4.1 - [2019-09-17]

  • This point release addresses the following issues:
    • Fixes an issue where a PID namespace was always being used
    • Fixes compilation on non 64-bit architectures
    • Allows fakeroot builds for zypper, pacstrap, and debootstrap
    • Correctly detects seccomp on OpenSUSE
    • Honors GO_MODFLAGS properly in the mconfig generated makefile
    • Passes the Mac hostname to the VM in MacOS Singularity builds
    • Handles temporary EAGAIN failures when setting up loop devices on recent kernels
    • Fixes excessive memory usage in singularity push

v3.4.0 - [2019-08-30]

New features / functionalities

  • New support for building and running encrypted containers with RSA keys and passphrases
    • --pem-path option added to the build and action commands for RSA based encrypted containers
    • --passphrase option added to build and action commands for passphrase based encrypted containers
    • SINGULARITY_ENCRYPTION_PEM_PATH and SINGULARITY_ENCRYPTION_PASSPHRASE environment variables added to serve same functions as above
    • --encrypt option added to build command to build an encrypted container when environment variables contain a secret
  • New --disable-cache flag prevents caching of downloaded containers
  • Added support for multi-line variables in singularity def-files
  • Added support for 'indexed' def-file variables (like arrays)
  • Added support for SUSE SLE Products
  • Added the def-file variables: product, user, regcode, productpgp, registerurl, modules, otherurl (indexed)
  • Support multiple-architecture tags in the SCS library
  • Added a --dry-run flag to cache clean
  • Added a SINGULARITY_SYPGPDIR environment variable to specify the location of PGP key data
  • Added a --nonet option to the action commands to disable networking when running with the --vm option
  • Added a --long-list flag to the key search command to preserve
  • Added experimental, hidden --fusemount flag to pass a command to mount a libfuse3 based file system within the container

Changed defaults / behaviors

  • Runtime now properly honors SINGULARITY_DISABLE_CACHE environment variable
  • remote add command now automatically attempts to login and a --no-login flag is added to disable this behavior
  • Using the pull command to download an unsigned container no longer produces an error code
  • cache clean command now prompts user before cleaning when run without --force option and is more verbose
  • Shortened the default output of the key search command

Deprecated / removed commands

  • The --allow-unsigned flag to pull has been deprecated and will be removed in the future

v3.3.0 - [2019-06-17]

Changed defaults / behaviors

  • Remote login and status commands will now use the default remote if a remote name is not supplied
  • Added Singularity hub (shub) cache support when using the pull command
  • Clean cache in a safer way by only deleting the cache subdirectories
  • Improvements to the cache clean command

New features / functionalities

  • new oras URI for pushing and pulling SIF files to and from supported OCI registries
  • added the --fakeroot option to build, exec, run, shell, test, and instance start commands to run container in a new user namespace as uid 0
  • added the fakeroot network type for use with the --network option
  • sif command to allow for the inspection and manipulation of SIF files with the following subcommands
    • add Add a data object to a SIF file
    • del Delete a specified object descriptor and data from SIF file
    • dump Extract and output data objects from SIF files
    • header Display SIF global headers
    • info Display detailed information of object descriptors
    • list List object descriptors from SIF files
    • new Create a new empty SIF image file
    • setprim Set primary system partition

v3.2.1 - [2019-05-28]

  • This point release fixes the following bugs:
    • Allows users to join instances with non-suid workflow
    • Removes false warning when seccomp is disabled on the host
    • Fixes an issue in the terminal when piping output to commands
    • Binds NVIDIA persistenced socket when --nv is invoked

v3.2.0 - [2019-05-14]

Security related fix

  • Instance files are now stored in user's home directory for privacy and many checks have been added to ensure that a user can't manipulate files to change starter-suid behavior when instances are joined (many thanks to Matthias Gerstner from the SUSE security team for finding and securely reporting this vulnerability)

New features / functionalities

  • Introduced a new basic framework for creating and managing plugins
  • Added the ability to create containers through multi-stage builds
    • Definitions now require Bootstrap be the first parameter of header
  • Created the concept of a Sylabs Cloud "remote" endpoint and added the ability for users and admins to set them through CLI and conf files
  • Added caching for images from Singularity Hub
  • Made it possible to compile Singularity outside of $GOPATH
  • Added a json partition to SIF files for OCI configuration when building from an OCI source
  • Full integration with Singularity desktop for MacOS code base

New Commands

  • Introduced the plugin command group for creating and managing plugins

    • compile Compile a singularity plugin
    • disable disable an installed singularity plugin
    • enable Enable an installed singularity plugin
    • inspect Inspect a singularity plugin (either an installed one or an image)
    • install Install a singularity plugin
    • list List installed singularity plugins
    • uninstall Uninstall removes the named plugin from the system

  • Introduced the remote command group to support management of Singularity endpoints:

    • add Create a new Sylabs Cloud remote endpoint
    • list List all remote endpoints that are configured
    • login Log into a remote endpoint using an authentication token
    • remove Remove an existing Sylabs Cloud remote endpoint
    • status Check the status of the services at an endpoint
    • use Set a remote endpoint to be used by default

  • Added to the key command group to improve PGP key management:

    • export Export a public or private key into a specific file
    • import Import a local key into the local keyring
    • remove Remove a local public key

  • Added the Stage: <name> keyword to the definition file header and the from <stage name> option/argument pair to the %files section to support multistage builds

Deprecated / removed commands

  • The --token/-t option has been deprecated in favor of the singularity remote command group

Changed defaults / behaviors

  • Ask to confirm password on a newly generated PGP key
  • Prompt to push a key to the KeyStore when generated
  • Refuse to push an unsigned container unless overridden with --allow-unauthenticated/-U option
  • Warn and prompt when pulling an unsigned container without the --allow-unauthenticated/-U option
  • Bootstrap must now be the first field of every header because of parser requirements for multi-stage builds

v3.1.1 - [2019-04-02]

New Commands

  • New hidden buildcfg command to display compile-time parameters
  • Added support for LDFLAGS, CFLAGS, CGO_ variables in build system
  • Added --nocolor flag to Singularity client to disable color in logging

Removed Commands

  • singularity capability <add/drop> --desc has been removed
  • singularity capability list <--all/--group/--user> flags have all been removed

New features / functionalities

  • The --builder flag to the build command implicitly sets --remote
  • Repeated binds no longer cause Singularity to exit and fail, just warn instead
  • Corrected typos and improved docstrings throughout
  • Removed warning when CWD does not exist on the host system
  • Added support to spec file for RPM building on SLES 11

v3.1.0 - [2019-02-22]

New Commands

  • Introduced the oci command group to support a new OCI compliant variant of the Singularity runtime:
    • attach Attach console to a running container process
    • create Create a container from a bundle directory
    • delete Delete container
    • exec Execute a command within container
    • kill Kill a container
    • mount Mount create an OCI bundle from SIF image
    • pause Suspends all processes inside the container
    • resume Resumes all processes previously paused inside the container
    • run Create/start/attach/delete a container from a bundle directory
    • start Start container process
    • state Query state of a container
    • umount Umount delete bundle
    • update Update container cgroups resources
  • Added cache command group to inspect and manage cached files
    • clean Clean your local Singularity cache
    • list List your local Singularity cache

New features / functionalities

  • Can now build CLI on darwin for limited functionality on Mac
  • Added the scratch bootstrap agent to build from anything
  • Reintroduced support for zypper bootstrap agent
  • Added the ability to overwrite a new singularity.conf when building from RPM if desired
  • Fixed several regressions and omissions in SCIF support
  • Added caching for containers pulled/built from the Container Library
  • Changed keys command group to key (retained hidden keys command for backward compatibility)
  • Created an RPMPREFIX variable to allow RPMs to be installed in custom locations
  • Greatly expanded CI unit and end-to-end testing

v3.0.3 - [2019-01-21]

  • Bind paths in singularity.conf are properly parsed and applied at runtime
  • Singularity runtime will properly fail if singularity.conf file is not owned by the root user
  • Several improvements to RPM packaging including using golang from epel, improved support for Fedora, and avoiding overwriting conf file on new RPM install
  • Unprivileged --contain option now properly mounts devpts on older kernels
  • Uppercase proxy environment variables are now rightly respected
  • Add http/https protocols for singularity run/pull commands
  • Update to SIF 1.0.2
  • Add noPrompt parameter to pkg/signing/Verify function to enable silent verification

v3.0.2 - [2019-01-04]

  • Added the --docker-login flag to enable interactive authentication with docker registries
  • Added support for pulling directly from HTTP and HTTPS
  • Made minor improvements to RPM packaging and added basic support for alpine packaging
  • The $SINGULARITY_NOHTTPS,$SINGULARITY_TMPDIR, and $SINGULARITY_DOCKER_USERNAME/$SINGULARITY_DOCKER_PASSWORD environment variables are now correctly respected
  • Pulling from a private shub registry now works as expected
  • Running a container with --network="none" no longer incorrectly fails with an error message
  • Commands now correctly return 1 when incorrectly executed without arguments
  • Progress bars no longer incorrectly display when running with --quiet or --silent
  • Contents of 91-environment.sh file are now displayed if appropriate when running inspect --environment

v3.0.1 - [2018-10-31]

  • Improved RPM packaging procedure via makeit
  • Enhanced general stability of runtime

v3.0.0 - [2018-10-08]

  • Singularity is now written primarily in Go to bring better integration with the existing container ecosystem
  • Added support for new URIs (build & run/exec/shell/start):
    • library:// - Supports the Sylabs.io Cloud Library
    • docker-daemon: - Supports images managed by the locally running docker daemon
    • docker-archive: - Supports archived docker images
    • oci: - Supports oci images
    • oci-archive: - Supports archived oci images
  • Handling of docker & oci URIs/images now utilizes containers/image to parse and convert those image types in a supported way
  • Replaced singularity instance.* command group with singularity instance *
  • The command singularity help now only provides help regarding the usage of the singularity command. To display an image's help message, use singularity run-help <image path> instead

Removed Deprecated Commands

  • Removed deprecated singularity image.* command group
  • Removed deprecated singularity create command
  • Removed deprecated singularity bootstrap command
  • Removed deprecated singularity mount command
  • Removed deprecated singularity check command

New Commands

  • Added singularity run-help <image path> command to output an image's help message
  • Added singularity sign <image path> command to allow a user to cryptographically sign a SIF image
  • Added singularity verify <image path> command to allow a user to verify a SIF image's cryptographic signatures
  • Added singularity keys command to allow the management of OpenPGP key stores
  • Added singularity capability command to allow fine grained control over the capabilities of running containers
  • Added singularity push command to push images to the Sylabs.io Cloud Library

Changed Commands

Action Command Group (run/shell/exec/instance start)

  • Added flags:
    • --add-caps <string>: Run the contained process with the specified capability set (requires root)
    • --allow-setuid: Allows setuid binaries to be mounted into the container (requires root)
    • --apply-cgroups <path>: Apply cgroups configuration from file to contained processes (requires root)
    • --dns <string>: Adds the comma separated list of DNS servers to the containers resolv.conf file
    • --drop-caps <string>: Drop the specified capabilities from the container (requires root)
    • --fakeroot: Run the container in a user namespace as uid=0. Requires a recent kernel to function properly
    • --hostname <string>: Set the hostname of the container
    • --keep-privs: Keep root user privilege inside the container (requires root)
    • --network <string>: Specify a list of comma separated network types (CNI Plugins) to be present inside the container, each with its own dedicated interface in the container
    • --network-args <string>: Specify arguments to pass to CNI network plugins (set by --network)
    • --no-privs: Drop all privileges from root user inside the container (requires root)
    • --security <string>: Configure security features such as SELinux, Apparmor, Seccomp...
    • --writable-tmpfs: Run container with a tmpfs overlay
  • The command singularity instance start now supports the --boot flag to boot the container via /sbin/init
  • Changes to image mounting behavior:
    • All image formats are mounted as read only by default
    • --writable only works on images which can be mounted in read/write [applicable to: sandbox and legacy ext3 images]
    • --writable-tmpfs runs the container with a writable tmpfs-based overlay [applicable to: all image formats]
    • --overlay <string> now specifies a list of ext3/sandbox images which are set as the containers overlay [applicable to: all image formats]

Build Command

  • All images are now built as Singularity Image Format (SIF) images by default
  • When building to a path that already exists, singularity build will now prompt the user if they wish to overwrite the file existing at the specified location
  • The -w|--writable flag has been removed
  • The -F|--force flag now overrides the interactive prompt and will always attempt to overwrite the file existing at the specified location
  • The -u|--update flag has been added to support the workflow of running a definition file on top of an existing container [implies --sandbox, only supports sandbox image types]
  • The singularity build command now supports the following flags for integration with the Sylabs.io Cloud Library:
    • -r|--remote: Build the image remotely on the Sylabs Remote Builder (currently unavailable)
    • -d|--detached: Detach from the stdout of the remote build [requires --remote]
    • --builder <string>: Specifies the URL of the remote builder to access
    • --library <string>: Specifies the URL of the Sylabs.io Cloud Library to push the built image to when the build command destination is in the form library://<reference>
  • The bootstrap keyword in the definition file now supports the following values:
    • library
    • docker-daemon
    • docker-archive
    • oci
    • oci-archive
  • The from keyword in the definition file now correctly parses a docker URI which includes the registry and/or namespace components
  • The registry and namespace keywords in the definition file are no longer supported. Instead, those values may all go into the from keyword
  • Building from a tar archive of a sandbox no longer works