In addition to numerous new and upgraded packages, this release has the following highlights: Support is planned until the end of April 2021, handing over to 21.03. GNOME desktop environment was upgraded to 3.36, see its release notes. maxx package removed along with services.xserver.desktopManager.maxx module. Please migrate to cdesktopenv and services.xserver.desktopManager.cde module. We now distribute a GNOME ISO. PHP now defaults to PHP 7.4, updated from 7.3. PHP 7.2 is no longer supported due to upstream not supporting this version for the entire lifecycle of the 20.09 release. Python 3 now defaults to Python 3.8 instead of 3.7. Two new options, authorizedKeysCommand and authorizedKeysCommandUser, have been added to the openssh module. If you have AuthorizedKeysCommand in your services.openssh.extraConfig you should make use of these new options instead. There is a new module for Podman(virtualisation.podman), a drop-in replacement for the Docker command line. The new virtualisation.containers module manages configuration shared by the CRI-O and Podman modules. Declarative Docker containers are renamed from docker-containers to virtualisation.oci-containers.containers. This is to make it possible to use podman instead of docker. MariaDB has been updated to 10.4, MariaDB Galera to 26.4. Before you upgrade, it would be best to take a backup of your database. For MariaDB Galera Cluster, see Upgrading from MariaDB 10.3 to MariaDB 10.4 with Galera Cluster instead. Before doing the upgrade read Incompatible Changes Between 10.3 and 10.4. After the upgrade you will need to run mysql_upgrade. MariaDB 10.4 introduces a number of changes to the authentication process, intended to make things easier and more intuitive. See Authentication from MariaDB 10.4. unix_socket auth plugin does not use a password, and uses the connecting user's UID instead. When a new MariaDB data directory is initialized, two MariaDB users are created and can be used with new unix_socket auth plugin, as well as traditional mysql_native_password plugin: root@localhost and mysql@localhost. To actually use the traditional mysql_native_password plugin method, one must run the following: services.mysql.initialScript = pkgs.writeText "mariadb-init.sql" '' ALTER USER root@localhost IDENTIFIED VIA mysql_native_password USING PASSWORD("verysecret"); ''; When MariaDB data directory is just upgraded (not initialized), the users are not created or modified. MySQL server is now started with additional systemd sandbox/hardening options for better security. The PrivateTmp, ProtectHome, and ProtectSystem options may be problematic when MySQL is attempting to read from or write to your filesystem anywhere outside of its own state directory, for example when calling LOAD DATA INFILE or SELECT * INTO OUTFILE. In this scenario a variant of the following may be required: - allow MySQL to read from /home and /tmp directories when using LOAD DATA INFILE systemd.services.mysql.serviceConfig.ProtectHome = lib.mkForce "read-only"; - allow MySQL to write to custom folder /var/data when using SELECT * INTO OUTFILE, assuming the mysql user has write access to /var/data systemd.services.mysql.serviceConfig.ReadWritePaths = [ "/var/data" ]; The MySQL service no longer runs its systemd service startup script as root anymore. A dedicated non root super user account is required for operation. This means users with an existing MySQL or MariaDB database server are required to run the following SQL statements as a super admin user before upgrading: CREATE USER IF NOT EXISTS 'mysql'@'localhost' identified with unix_socket; GRANT ALL PRIVILEGES ON *.* TO 'mysql'@'localhost' WITH GRANT OPTION; If you use MySQL instead of MariaDB please replace unix_socket with auth_socket. If you have changed the value of from the default of mysql to a different user please change 'mysql'@'localhost' to the corresponding user instead. The new option documentation.man.generateCaches has been added to automatically generate the man-db caches, which are needed by utilities like whatis and apropos. The caches are generated during the build of the NixOS configuration: since this can be expensive when a large number of packages are installed, the feature is disabled by default. services.postfix.sslCACert was replaced by services.postfix.tlsTrustedAuthorities which now defaults to system certificate authorities. Subordinate GID and UID mappings are now set up automatically for all normal users. This will make container tools like Podman work as non-root users out of the box. The various documented workarounds to use steam have been converted to a module. programs.steam.enable enables steam, controller support and the workarounds. Support for built-in LCDs in various pieces of Logitech hardware (keyboards and USB speakers). hardware.logitech.lcd.enable enables support for all hardware supported by the g15daemon project. Zabbix now defaults to 5.0, updated from 4.4. Please carefully read through the upgrade guide and apply any changes required. Be sure to take special note of the section on enabling extended range of numeric (float) values as you will need to apply this database migration manually. If you are using Zabbix Server with a MySQL or MariaDB database you should note that using a character set of utf8 and a collate of utf8_bin has become mandatory with this release. See the upstream issue for further discussion. Before upgrading you should check the character set and collation used by your database and ensure they are correct: SELECT default_character_set_name, default_collation_name FROM information_schema.schemata WHERE schema_name = 'zabbix'; If these values are not correct you should take a backup of your database and convert the character set and collation as required. Here is an example of how to do so, taken from the Zabbix forums: ALTER DATABASE `zabbix` DEFAULT CHARACTER SET utf8 COLLATE utf8_bin; -- the following will produce a list of SQL commands you should subsequently execute SELECT CONCAT("ALTER TABLE ", TABLE_NAME," CONVERT TO CHARACTER SET utf8 COLLATE utf8_bin;") AS ExecuteTheString FROM information_schema.`COLUMNS` WHERE table_schema = "zabbix" AND COLLATION_NAME = "utf8_general_ci"; The NixOS module system now supports freeform modules as a mix between types.attrsOf and types.submodule. These allow you to explicitly declare a subset of options while still permitting definitions without an associated option. See for how to use them. The GRUB module gained support for basic password protection, which allows to restrict non-default entries in the boot menu to one or more users. The users and passwords are defined via the option boot.loader.grub.users. Note: Password support is only avaiable in GRUB version 2. Following its deprecation in 20.03, the Perl NixOS test driver has been removed. All remaining tests have been ported to the Python test framework. Code outside nixpkgs using make-test.nix or testing.nix needs to be ported to make-test-python.nix and testing-python.nix respectively.

The following new services were added since the last release: There is a new module that provides doas, a lighter alternative to sudo with many of the same features. Hercules CI Agent is a specialized build agent for projects built with Nix. See the options and setup.

When upgrading from a previous release, please be aware of the following incompatible changes: buildGoModule now internally creates a vendor directory in the source tree for downloaded modules instead of using go's module proxy protocol. This storage format is simpler and therefore less likely to break with future versions of go. As a result buildGoModule switched from modSha256 to the vendorSha256 attribute to pin fetched version data. Grafana is now built without support for phantomjs by default. Phantomjs support has been deprecated in Grafana and the phantomjs project is currently unmaintained. It can still be enabled by providing phantomJsSupport = true to the package instantiation: { services.grafana.package = pkgs.grafana.overrideAttrs (oldAttrs: rec { phantomJsSupport = false; }); } The supybot module now uses /var/lib/supybot as its default stateDir path if stateVersion is 20.09 or higher. It also enables a number of systemd sandboxing options which may possibly interfere with some plugins. If this is the case you can disable the options through attributes in systemd.services.supybot.serviceConfig. The security.duosec.skey option, which stored a secret in the nix store, has been replaced by a new security.duosec.secretKeyFile option for better security. security.duosec.ikey has been renamed to security.duosec.integrationKey. vmware has been removed from the services.x11.videoDrivers defaults. For VMWare guests set virtualisation.vmware.guest.enable to true which will include the appropriate drivers. The initrd SSH support now uses OpenSSH rather than Dropbear to allow the use of Ed25519 keys and other OpenSSH-specific functionality. Host keys must now be in the OpenSSH format, and at least one pre-generated key must be specified. If you used the boot.initrd.network.ssh.host*Key options, you'll get an error explaining how to convert your host keys and migrate to the new boot.initrd.network.ssh.hostKeys option. Otherwise, if you don't have any host keys set, you'll need to generate some; see the hostKeys option documentation for instructions. Since this release there's an easy way to customize your PHP install to get a much smaller base PHP with only wanted extensions enabled. See the following snippet installing a smaller PHP with the extensions imagick, opcache, pdo and pdo_mysql loaded: environment.systemPackages = [ (pkgs.php.withExtensions ({ all, ... }: with all; [ imagick opcache pdo pdo_mysql ]) ) ]; The default php attribute hasn't lost any extensions. The opcache extension has been added. All upstream PHP extensions are available under ]]>. All PHP config flags have been removed for the following reasons: The updated php attribute is now easily customizable to your liking by using php.withExtensions or php.buildEnv instead of writing config files or changing configure flags. The remaining configuration flags can now be set directly on the php attribute. For example, instead of php.override { config.php.embed = true; config.php.apxs2 = false; } you should now write php.override { embedSupport = true; apxs2Support = false; } The ACME module has been overhauled for simplicity and maintainability. Cert generation now implicitly uses the acme user, and the security.acme.certs._name_.user option has been removed. Instead, certificate access from other services is now managed through group permissions. The module no longer runs lego twice under certain conditions, and will correctly renew certificates if their configuration is changed. Services which reload nginx and httpd after certificate renewal are now properly configured too so you no longer have to do this manually if you are using HTTPS enabled virtual hosts. A mechanism for regenerating certs on demand has also been added and documented. Gollum received a major update to version 5.x and you may have to change some links in your wiki when migrating from gollum 4.x. More information can be found here. Deluge 2.x was added and is used as default for new NixOS installations where stateVersion is >= 20.09. If you are upgrading from a previous NixOS version, you can set service.deluge.package = pkgs.deluge-2_x to upgrade to Deluge 2.x and migrate the state to the new format. Be aware that backwards state migrations are not supported by Deluge. Add option services.nginx.enableSandbox to starting Nginx web server with additional sandbox/hardening options. By default, write access to services.nginx.stateDir is allowed. To allow writing to other folders, use systemd.services.nginx.serviceConfig.ReadWritePaths systemd.services.nginx.serviceConfig.ReadWritePaths = [ "/var/www" ]; The NixOS options nesting.clone and nesting.children have been deleted, and replaced with named configurations. Replace a nesting.clone entry with: { specialisation.example-sub-configuration = { configuration = { ... }; }; Replace a nesting.children entry with: { specialisation.example-sub-configuration = { inheritParentConfig = false; configuration = { ... }; }; To switch to a specialised configuration at runtime you need to run: # sudo /run/current-system/specialisation/example-sub-configuration/bin/switch-to-configuration test Before you would have used: # sudo /run/current-system/fine-tune/child-1/bin/switch-to-configuration test The Nginx log directory has been moved to /var/log/nginx, the cache directory to /var/cache/nginx. The option services.nginx.stateDir has been removed. The httpd web server previously started its main process as root privileged, then ran worker processes as a less privileged identity user. This was changed to start all of httpd as a less privileged user (defined by and ). As a consequence, all files that are needed for httpd to run (included configuration fragments, SSL certificates and keys, etc.) must now be readable by this less privileged user/group. The default value for has been changed from prefork to event. Along with this change the default value for services.httpd.virtualHosts.<name>.http2 has been set to true. The systemd-networkd option systemd.network.networks.<name>.dhcp.CriticalConnection has been removed following upstream systemd's deprecation of the same. It is recommended to use systemd.network.networks.<name>.networkConfig.KeepConfiguration instead. See systemd.network 5 for details. The systemd-networkd option systemd.network.networks._name_.dhcpConfig has been renamed to following upstream systemd's documentation change. See systemd.network 5 for details. In the picom module, several options that accepted floating point numbers encoded as strings (for example ) have been changed to the (relatively) new native float type. To migrate your configuration simply remove the quotes around the numbers. When using buildBazelPackage from Nixpkgs, flat hash mode is now used for dependencies instead of recursive. This is to better allow using hashed mirrors where needed. As a result, these hashes will have changed. The rkt module has been removed, it was archived by upstream. The Bazaar VCS is unmaintained and, as consequence of the Python 2 EOL, the packages bazaar and bazaarTools were removed. Breezy, the backward compatible fork of Bazaar (see the announcement), was packaged as breezy and can be used instead. Regarding Nixpkgs, fetchbzr, nix-prefetch-bzr and Bazaar support in Hydra will continue to work through Breezy. In addition to the hostname, the fully qualified domain name (FQDN), which consists of ${cfg.hostName} and ${cfg.domain} is now added to /etc/hosts, to allow local FQDN resolution, as used by the hostname --fqdn command and other applications that try to determine the FQDN. These new entries take precedence over entries from the DNS which could cause regressions in some very specific setups. Additionally the hostname is now resolved to 127.0.0.2 instead of 127.0.1.1 to be consistent with what nss-myhostname (from systemd) returns. The old behaviour can e.g. be restored by using networking.hosts = lib.mkForce { "127.0.1.1" = [ config.networking.hostName ]; };. The hostname (networking.hostName) must now be a valid DNS label (see RFC 1035) and as such must not contain the domain part. This means that the hostname must start with a letter, end with a letter or digit, and have as interior characters only letters, digits, and hyphen. The maximum length is 63 characters. Additionally it is recommended to only use lower-case characters. The GRUB specific option boot.loader.grub.extraInitrd has been replaced with the generic option boot.initrd.secrets. This option creates a secondary initrd from the specified files, rather than using a manually created initrd file. Due to an existing bug with boot.loader.grub.extraInitrd, it is not possible to directly boot an older generation that used that option. It is still possible to rollback to that generation if the required initrd file has not been deleted. The DNSChain package and NixOS module have been removed from Nixpkgs as the software is unmaintained and can't be built. For more information see issue #89205. In the resilio module, has been changed to listen to [::1] instead of 0.0.0.0. Users of OpenAFS 1.6 must upgrade their services to OpenAFS 1.8! In this release, the OpenAFS package version 1.6.24 is marked broken but can be used during transition to OpenAFS 1.8.x. Use the options services.openafsClient.packages.module, services.openafsClient.packages.programs and services.openafsServer.package to select a different OpenAFS package. OpenAFS 1.6 will be removed in the next release. The package openafs and the service options will then silently point to the OpenAFS 1.8 release. See also the OpenAFS Administrator Guide for instructions. Beware of the following when updating servers: The storage format of the server key has changed and the key must be converted before running the new release. When updating multiple database servers, turn off the database servers from the highest IP down to the lowest with resting periods in between. Start up in reverse order. Do not concurrently run database servers working with different OpenAFS releases! Update servers first, then clients. Radicale's default package has changed from 2.x to 3.x. An upgrade checklist can be found here. You can use the newer version in the NixOS service by setting the package to radicale3, which is done automatically if stateVersion is 20.09 or higher. udpt experienced a complete rewrite from C++ to rust. The configuration format changed from ini to toml. The new configuration documentation can be found at the official website and example configuration is packaged in ${udpt}/share/udpt/udpt.toml. We now have a unified option interface to be used for every display-manager in NixOS. The bitcoind module has changed to multi-instance, using submodules. Therefore, it is now mandatory to name each instance. To use this new multi-instance config with an existing bitcoind data directory and user, you have to adjust the original config, e.g.: services.bitcoind = { enable = true; extraConfig = "..."; ... }; To something similar: services.bitcoind.mainnet = { enable = true; dataDir = "/var/lib/bitcoind"; user = "bitcoin"; extraConfig = "..."; ... }; The key settings are: dataDir - to continue using the same data directory. user - to continue using the same user so that bitcoind maintains access to its files. Graylog introduced a change in the LDAP server certificate validation behaviour for version 3.3.3 which might break existing setups. When updating Graylog from a version before 3.3.3 make sure to check the Graylog release info for information on how to avoid the issue. The dokuwiki module has changed to multi-instance, using submodules. Therefore, it is now mandatory to name each instance. Moreover, forcing SSL by default has been dropped, so nginx.forceSSL and nginx.enableACME are no longer set to true. To continue using your service with the original SSL settings, you have to adjust the original config, e.g.: services.dokuwiki = { enable = true; ... }; To something similar: services.dokuwiki."mywiki" = { enable = true; nginx = { forceSSL = true; enableACME = true; }; ... }; The base package has also been upgraded to the 2020-07-29 "Hogfather" release. Plugins might be incompatible or require upgrading. The option is now set to "/var/lib/postgresql/${cfg.package.psqlSchema}" regardless of your . Users with an existing postgresql install that have a of 17.03 or below should double check what the value of their option is (/var/db/postgresql) and then explicitly set this value to maintain compatibility: services.postgresql.dataDir = "/var/db/postgresql"; The postgresql module now expects there to be a database super user account called postgres regardless of your . Users with an existing postgresql install that have a of 17.03 or below should run the following SQL statements as a database super admin user before upgrading: CREATE ROLE postgres LOGIN SUPERUSER; The USBGuard module now removes options and instead hardcodes values for IPCAccessControlFiles, ruleFiles, and auditFilePath. Audit logs can be found in the journal. The NixOS module system now evaluates option definitions more strictly, allowing it to detect a larger set of problems. As a result, what previously evaluated may not do so anymore. See the PR that changed this for more info. For NixOS configuration options, the type loaOf, after its initial deprecation in release 20.03, has been removed. In NixOS and Nixpkgs options using this type have been converted to attrsOf. For more information on this change have look at these links: issue #1800, PR #63103. config.systemd.services.${name}.path now returns a list of paths instead of a colon-separated string.

SD images are now compressed by default using zstd. The compression for ISO images has also been changed to zstd, but ISO images are still not compressed by default. services.journald.rateLimitBurst was updated from 1000 to 10000 to follow the new upstream systemd default. The notmuch package move its emacs-related binaries and emacs lisp files to a separate output. They're not part of the default out output anymore - if you relied on the notmuch-emacs-mua binary or the emacs lisp files, access them via the notmuch.emacs output. The default output of buildGoPackage is now $out instead of $bin. buildGoModule doCheck now defaults to true. Packages built using buildRustPackage now use release mode for the checkPhase by default. Please note that Rust packages utilizing a custom build/install procedure (e.g. by using a Makefile) or test suites that rely on the structure of the target/ directory may break due to those assumptions. For further information, please read the Rust section in the Nixpkgs manual. The cc- and binutils-wrapper's "infix salt" and _BUILD_ and _TARGET_ user infixes have been replaced with with a "suffix salt" and suffixes and _FOR_BUILD and _FOR_TARGET. This matches the autotools convention for env vars which standard for these things, making interfacing with other tools easier. Additional Git documentation (HTML and text files) is now available via the git-doc package. Default algorithm for ZRAM swap was changed to zstd. The scripted networking system now uses .link files in /etc/systemd/network to configure mac address and link MTU, instead of the sometimes buggy network-link-* units, which have been removed. Bringing the interface up has been moved to the beginning of the network-addresses-* unit. Note this doesn't require systemd-networkd - it's udev that parses .link files. Extra care needs to be taken in the presence of legacy udev rules to rename interfaces, as MAC Address and MTU defined in these options can only match on the original link name. In such cases, you most likely want to create a 10-*.link file through and set both name and MAC Address / MTU there. Grafana received a major update to version 7.x. A plugin is now needed for image rendering support, and plugins must now be signed by default. More information can be found in the Grafana documentation. The hardware.u2f module, which was installing udev rules was removed, as udev gained native support to handle FIDO security tokens. The services.transmission module was enhanced with the new options: , , and . transmission-daemon is now started with additional systemd sandbox/hardening options for better security. Please report any use case where this is not working well. In particular, the RootDirectory option newly set forbids uploading or downloading a torrent outside of the default directory configured at settings.download-dir. If you really need Transmission to access other directories, you must include those directories into the BindPaths of the service: systemd.services.transmission.serviceConfig.BindPaths = [ "/path/to/alternative/download-dir" ]; Also, connection to the RPC (Remote Procedure Call) of transmission-daemon is now only available on the local network interface by default. Use: services.transmission.settings.rpc-bind-address = "0.0.0.0"; to get the previous behavior of listening on all network interfaces. With this release systemd-networkd (when enabled through ) has it's netlink socket created through a systemd.socket unit. This gives us control over socket buffer sizes and other parameters. For larger setups where networkd has to create a lot of (virtual) devices the default buffer size (currently 128MB) is not enough. On a machine with >100 virtual interfaces (e.g., wireguard tunnels, VLANs, …), that all have to be brought up during system startup, the receive buffer size will spike for a brief period. Eventually some of the message will be dropped since there is not enough (permitted) buffer space available. By having systemd-networkd start with a netlink socket created by systemd we can configure the ReceiveBufferSize= parameter in the socket options (i.e. systemd.sockets.systemd-networkd.socketOptions.ReceiveBufferSize) without recompiling systemd-networkd. Since the actual memory requirements depend on hardware, timing, exact configurations etc. it isn't currently possible to infer a good default from within the NixOS module system. Administrators are advised to monitor the logs of systemd-networkd for rtnl: kernel receive buffer overrun spam and increase the memory limit as they see fit. Note: Increasing the ReceiveBufferSize= doesn't allocate any memory. It just increases the upper bound on the kernel side. The memory allocation depends on the amount of messages that are queued on the kernel side of the netlink socket. Specifying mailboxes in the dovecot2 module as a list is deprecated and will break eval in 21.03. Instead, an attribute-set should be specified where the name should be the key of the attribute. This means that a configuration like this { services.dovecot2.mailboxes = [ { name = "Junk"; auto = "create"; } ]; } should now look like this: { services.dovecot2.mailboxes = { Junk.auto = "create"; }; } netbeans was upgraded to 12.0 and now defaults to OpenJDK 11. This might cause problems if your projects depend on packages that were removed in Java 11. nextcloud has been updated to v19. If you have an existing installation, please make sure that you're on nextcloud18 before upgrading to nextcloud19 since Nextcloud doesn't support upgrades across multiple major versions. The nixos-run-vms script now deletes the previous run machines states on test startup. You can use the --keep-vm-state flag to match the previous behaviour and keep the same VM state between different test runs. The nix.buildMachines option is now type-checked. There are no functional changes, however this may require updating some configurations to use correct types for all attributes. The fontconfig module stopped generating fontconfig 2.10.x config and cache. Fontconfig 2.10.x was removed from Nixpkgs - it hasn't been used in any nixpkgs package anymore. Nginx module nginxModules.fastcgi-cache-purge renamed to official name nginxModules.cache-purge. Nginx module nginxModules.ngx_aws_auth renamed to official name nginxModules.aws-auth. The packages perl, rsync and strace were removed from systemPackages. If you need them, install them again with = with pkgs; [ perl rsync strace ]; in your configuration.nix. The undervolt option no longer needs to apply its settings every 30s. If they still become undone, open an issue and restore the previous behaviour using undervolt.useTimer. Agda has been heavily reworked. agda.mkDerivation has been heavily changed and is now located at agdaPackages.mkDerivation. New top-level packages agda and agda.withPackages have been added, the second of which sets up agda with access to chosen libraries. All agda libraries now live under agdaPackages. Many broken libraries have been removed. See the new documentation for more information. The deepin package set has been removed from nixpkgs. It was a work in progress to package the Deepin Desktop Environment (DDE), including libraries, tools and applications, and it was still missing a service to lauch the desktop environment. It has shown to no longer be a feasible goal due to reasons discussed in issue #94870. The package netease-cloud-music has also been removed, as it depends on libraries from deepin. The opendkim module now uses systemd sandboxing features to limit the exposure of the system towards the opendkim service.