From Fedora Project Wiki

Revision as of 16:37, 7 August 2012 by Spot (talk | contribs) (→‎Systemd)

RPM scriptlet recipes

Rpm spec files have several sections which allow packages to run code on installation and removal. These scriptlets are mostly used to update the running system with information from the package. This page offers a quick overview of the RPM scriptlets and a number of common recipes for scriptlets in packages. For a more complete treatment of scriptlets, please see the Maximum RPM book .

Contents:

Default Shell

In Fedora, you can assume that the default shell (/bin/sh) is bash. Thus, all scriptlets can safely assume that if they are running in shell code, they are running within bash.

Syntax

The basic syntax is similar to the %build, %install, and other sections of the rpm spec file. The scripts support a special flag, -p which allows the scriptlet to invoke a single program directly rather than having to spawn a shell to invoke the programs. (ie: %post -p /sbin/ldconfig)

The scriptlets also take an argument, passed into them by the controlling rpmbuild process. This argument, accessed via $1 is the number of packages of this name which will be left on the system when the action completes, except for %pretrans and %posttrans which are always run with $1 as 0 (%pretrans and %posttrans are available in rpm 4.4 and later). So for the common case of install, upgrade, and uninstall we have:

install upgrade uninstall
%pretrans $1 == 0 $1 == 0 (N/A)
%pre $1 == 1 $1 == 2 (N/A)
%post $1 == 1 $1 == 2 (N/A)
%preun (N/A) $1 == 1 $1 == 0
%postun (N/A) $1 == 1 $1 == 0
%posttrans $1 == 0 $1 == 0 (N/A)

Note that these values will vary if there are multiple versions of the same package installed (This mostly occurs with parallel installable packages such as the kernel and multilib packages. However, it can also occur when errors prevent a package upgrade from completing.) So it is a good idea to use this construct:

%pre
if [ $1 -gt 1 ] ; then
fi

...for %pre and %post scripts rather than checking that it equals 2.

Except in some really exceptional cases (if any), we want all scriptlets to exit with the zero exit status. Because rpm in its default configuration does not at the moment execute shell scriptlets with the -e argument to the shell, excluding explicit exit calls (frowned upon with a non-zero argument!), the exit status of the last command in a scriptlet determines its exit status. Most commands in the snippets in this document have a "|| :" appended to them, which is a generic trick to force the zero exit status for those commands whether they worked or not. Usually the most important bit is to apply this to the last command executed in a scriptlet, or to add a separate command such as plain ":" or "exit 0" as the last one in a scriptlet. Note that depending on the case, other error checking/prevention measures may be more appropriate, as well as running some commands only if we saw a previous command in the scriptlet which is a must prerequisite to succeed.

Non-zero exit codes from scriptlets break installs/upgrades/erases so that no further actions will be taken for that package in a transaction (see scriptlet ordering below), which may for example prevent an old version of a package from being erased on upgrades, leaving behind duplicate rpmdb entries and possibly stale, unowned files on the filesystem. There are some cases where letting the transaction to proceed when some things in scriptlets failed may result in partially broken setup. It is however often limited to that package only whereas letting a transaction to proceed with some packages dropped out on the fly is more likely to result in broader system wide problems.

Scriptlet Ordering

The scriptlets in %pre and %post are respectively run before and after a package is installed. The scriptlets %preun and %postun are run before and after a package is uninstalled. The scriptlets %pretrans and %posttrans are run at start and end of a transaction. On upgrade, the scripts are run in the following order:

  1. %pretrans of new package
  2. %pre of new package
  3. (package install)
  4. %post of new package
  5. %triggerin of other packages (set off by installing new package)
  6. %triggerin of new package (if any are true)
  7. %triggerun of old package (if it's set off by uninstalling the old package)
  8. %triggerun of other packages (set off by uninstalling old package)
  9. %preun of old package
  10. (removal of old package)
  11. %postun of old package
  12. %triggerpostun of old package (if it's set off by uninstalling the old package)
  13. %triggerpostun of other packages (if they're setu off by uninstalling the old package)
  14. %posttrans of new package

Writing scriptlets

Some tips for writing good scriptlets

Saving state between scriptlets

Sometimes a scriptlet needs to save some state from an earlier running scriptlet in order to use it at a later running scriptlet. This is especially common when trying to optimize the scriptlets. Examples:

  • If a %posttrans needs to de-register some piece of information when upgrading but the file that has that information is removed when the old package is removed the scriptlets need to save that file during %pre or %post so that the script in %posttrans can access it.
  • If we only want the program in %posttrans to do its work once per-transaction, we may need to write out a flag file so that the %posttrans knows whether to perform an action.

To address these issues scriptlets that run earlier need to write out information that is used in %posttrans. We recommend using a subdirectory of %{_localstatedir}/lib/rpm-state/ for that. For instance, the eclipse plugin scripts touch a file in %{_localstatedir}/lib/rpm-state/eclipse/ when they're installed. The %posttrans runs a script that checks if that file exists. If it does, it performs its action and then deletes the file. That way the script only performs its action once per transaction.

Note.png
We're asking that the %{_localstatedir}/lib/rpm-state/ directory be owned by either the rpm or filesystem packages in Fedora. Until that happens, a package that is providing this functionality needs to own %{_localstatedir}/lib/rpm-state/ as well as their subdirectory of it.

Macros

Complex scriptlets may be placed in rpm macros. This has two benefits:

  1. The standard package authors only have to remember the macros, not the complex stuff that it does
  2. The macros' implementations may change without having to update the package

When writing the macros, the FPC will still want to review the macros (and perhaps include the implementation of the macros in the guideline to show packagers what's happening behind the scenes).

One principle that the FPC follows is that macros must not contain the start of scriptlet tags (for instance, %pre). This also means that a single macro must not be defined to do things in both %pre and %post. Instead, write one macro that performs the actions in %pre and a separate macro that performs the actions in %post. This principle makes it so that all spec files can use your macros in the same manner even if they already have a %pre or %post defined.

Snippets

Shared libraries

Installing shared libraries requires running /sbin/ldconfig to update the dynamic linker's cache files. These can be invoked like:

%post
/sbin/ldconfig
%postun
/sbin/ldconfig

It is also common to invoke these with the '-p' option as they are often the only program invoked in a scriptlet:

%post -p /sbin/ldconfig
%postun -p /sbin/ldconfig

If applicable, the latter way is recommended because doing so will automatically add appropriate dependencies on /sbin/ldconfig to the package (and FWIW, will prevent unnecessarily launching a shell process in the scriptlets).

Users and groups

These are discussed on a separate page

Services

Initscripts Conventions

Full guidelines for SysV-style initscripts can be found here: Packaging/SysVInitScript
Scriptlet specifics can be found here: Packaging/SysVInitScript#InitscriptScriptlets

GConf

GConf is a configuration scheme currently used by the GNOME desktop. Programs which use it setup default values in a [NAME] .schemas file which is installed under %{_sysconfdir}/gconf/schemas/[NAME] .schemas. These defaults are then registered with the gconf daemon which monitors the configuration values and alerts applications when values the applications are interested in change. The schema files also provide documentation about what each value in the configuration system means (which gets displayed when you browse the database in the gconf-editor program).

For packaging purposes, we have to disable schema installation during build, and also register the values in the [NAME] .schemas file with the gconf daemon on installation and unregister them on removal. Due to the ordering of the scriptlets, this is a four step process.

Disabling the GConf installation during the package creation can be done like so:

%install
export GCONF_DISABLE_MAKEFILE_SCHEMA_INSTALL=1
make install DESTDIR=$RPM_BUILD_ROOT
...

The GCONF_DISABLE_MAKEFILE_SCHEMA_INSTALL environment variable suppresses the installation of the schema during the building of the package. An alternative for some packages is to pass a configure flag:

%build
%configure --disable-schemas
...

Unfortunately, this configure switch only works if the upstream packager has adapted their Makefile.am to handle it. If the Makefile.am is not configured, this switch won't do anything and you'll need to use the environment variable instead.

Here's the second part:

BuildRequires: GConf2
Requires(pre): GConf2
Requires(post): GConf2
Requires(preun): GConf2
...
%pre
%gconf_schema_prepare schema1 schema2
%gconf_schema_obsolete schema3

In this section we uninstall old schemas during upgrade using one of two macros.

%gconf_schema_prepare is used for any current GConf schemas. It takes care of uninstalling previous versions of schemas that this package currently installs. It takes a space separated list of schema names without path or suffix that the package installs. Note that behind the scenes, this macro works with the %post scriptlet to only process GConf schemas if changes have occurred.

%gconf_schema_obsolete is used for schemas that this package previously provided but no longer does. It will deregister the old schema if it is present on the system. Nothing will happen if the old schema is not present. This macro takes a space separated list of schemas to uninstall. One example of using this might be if the package changed names. If the old schema was named foo.schemas and the new schema is named foobar.schemas you'd use:

%gconf_schema_prepare foobar
%gconf_schema_obsolete foo

The next section does the processing of the newly installed schemas:

%post
%gconf_schema_upgrade schema1 schema2

%gconf_schema_upgrade takes a space separated list of schemas that the package currently installs just like %gconf_schema_prepare. Behind the scenes, it does the actual work of registering the new version of the schema and deregistering the old version.

The last section is for unregistering schemas when a package is removed:

%preun
%gconf_schema_remove schema1 schema2

When a package is upgraded rpm invokes the %pre scriptlet to register and deregister the schemas. When a package is uninstalled, the %preun scriptlet is used. %gconf_schema_remove takes the list of schemas that this package currently provides and removes them for us.

Rebuilds for changes to macros

When macros change, packages that make use of them have to be rebuilt to pick up the changes. This repoquery command can be used to find the schema including packages to rebuild:

repoquery --whatprovides "/etc/gconf/schemas/*" |sort |uniq |wc -l

EPEL Notes

EPEL does not have macros.gconf2, so please follow the instructions found here: Packaging:EPEL#GConf

GSettings Schema

GSettings is the configuration system used by the GNOME 3 desktop. It replaces the older GConf system, which was used in GNOME 2. GSettings has pluggable backends, the 'native' one for GNOME is using DConf to store settings. The GSettings API and utilities are part of the glib2 package.

Programs which use GSettings install schema information including default values in the directory %{_datadir}/glib-2.0/schemas. Schema files are xml files with the extension .gschema.xml. At runtime, GSettings uses the schemas in a compiled binary (but arch-neutral) form, which is created by running the glib-compile-schemas utility. /usr/bin/glib-compile-schemas must be run whenever the set of installed schemas changes.

%postun
if [ $1 -eq 0 ] ; then
    /usr/bin/glib-compile-schemas %{_datadir}/glib-2.0/schemas &> /dev/null || :
fi

%posttrans
    /usr/bin/glib-compile-schemas %{_datadir}/glib-2.0/schemas &> /dev/null || :

gdk-pixbuf loaders

gdk-pixbuf is a library that is part of the gdk-pixbuf2 package. It is for loading images in various formats in GNOME. gdk-pixbuf can be extended by implementing loaders for image formats in loadable modules. These loadable modules have to be installed in %{_libdir}/gdk-pixbuf-2.0/2.10.0/loaders. To avoid opening all modules in that directory unnecessarily, gdk-pixbuf maintains a cache with information about the available modules in the text file %{_libdir}/gdk-pixbuf-2.0/2.10.0/loaders.cache. This cache file needs to be updated when the set of installed modules changes, by calling the /usr/bin/gdk-pixbuf-query-loaders binary. Multilib considerations force us to install the binary in -32 and -64 variants.

The scriptlets to maintain the cache file are:

%postun
    /usr/bin/gdk-pixbuf-query-loaders-%{__isa_bits} --update-cache &> /dev/null || :

%post
if [ $1 -eq 1 ] ; then
    # For upgrades, the cache will be regenerated by the new package's %postun
    /usr/bin/gdk-pixbuf-query-loaders-%{__isa_bits} --update-cache &> /dev/null || :
fi

Note the use of %{__isa_bits}, which is an rpm macro that expands to either 32 or 64, depending on the architecture of the package.

GTK+ modules

The GTK+ toolkit (in the gtk3 package) can be extended by loadable modules which can provide theme engines, input methods, print backends or other functionality. These modules have to be installed in subdirectories of %{_libdir}/gtk-3.0 or %{_libdir}/gtk-3.0/3.0.0. For the input methods, GTK+ maintains a cache in the text file %{_libdir}/gtk-3.0/3.0.0/immodules.cache. This cache file needs to be updated when the set of installed input methods changes, by calling the gtk-query-immodules-3.0 binary. Multilib considerations force us to install the binary in -32 and -64 variants.

The scriptlets to maintain the cache file are:

%postun
/usr/bin/gtk-query-immodules-3.0-%{__isa_bits} --update-cache &> /dev/null || :

%post
if [ $1 -eq 1 ] ; then
    # For upgrades, the cache will be regenerated by the new package's %postun
    /usr/bin/gtk-query-immodules-3.0-%{__isa_bits} --update-cache &> /dev/null || :
fi

The 3.0 in the binary name is there because gtk2 has its own utility for the same purpose, called gtk-query-immodules-2.0. Note the use of %{__isa_bits}, which is an rpm macro that expands to either 32 or 64, depending on the architecture of the package.

GIO modules

GIO is a library that is part of the glib2 package. It is a low-level part of the GNOME stack. GIO can be extended by implementing extension points in loadable modules. These loadable modules have to be installed in %{_libdir}/gio/modules. To avoid opening all modules in that directory unnecessarily, GIO maintains a cache with information about the available modules in the text file giomodule.cache in the same directory. This cache file needs to be updated when the set of installed modules changes, by calling the gio-querymodules binary. Multilib considerations force us to install the binary in -32 and -64 variants.

The scriptlets to maintain the cache file are:

%postun
/usr/bin/gio-querymodules-%{__isa_bits} %{_libdir}/gio/modules &> /dev/null || :

%post
# We run this after every install or upgrade because of a cornercase
# when installing the second architecture of a multilib package 
/usr/bin/gio-querymodules-%{__isa_bits} %{_libdir}/gio/modules || :

Note the use of %{__isa_bits}, which is an rpm macro that expands to either 32 or 64, depending on the architecture of the package.

Texinfo

The GNU project and many other programs use the texinfo file format for much of its documentation. These info files are usually located in /usr/share/info/. When installing or removing a package, install-info from the info package takes care of adding the newly installed files to the main info index and removing them again on deinstallation.

Requires(post): info
Requires(preun): info
...
%post
/sbin/install-info %{_infodir}/%{name}.info %{_infodir}/dir || :

%preun
if [ $1 = 0 ] ; then
  /sbin/install-info --delete %{_infodir}/%{name}.info %{_infodir}/dir || :
fi

These two scriptlets tell install-info to add entries for the info pages to the main index file on installation and remove them at erase time. The "|| :" in this case prevents failures that would typically affect systems that have been configured not to install any %doc files, or have read-only mounted, %_netsharedpath /usr/share.

Scrollkeeper

In all current Fedora, rarian has replaced scrollkeeper. There is no scriptlet needed for rarian. For instructions on what to do in EPEL releases, see Packaging:EPEL#Scrollkeeper

desktop-database

Use this when a desktop entry has a 'MimeType key.

%post
/usr/bin/update-desktop-database &> /dev/null || :

%postun
/usr/bin/update-desktop-database &> /dev/null || :

Note: For FC5+, this scriptlet follows the same convention as mimeinfo files and gtk-icon-cache. Namely, the spec file should not Require desktop-file-utils for this. For older releases, one should

Requires(post): desktop-file-utils
Requires(postun): desktop-file-utils

(See http://bugzilla.redhat.com/180898 and http://bugzilla.redhat.com/180899)

mimeinfo

Use this when a package drops an XML file in %{_datadir}/mime/packages.

%post
/usr/bin/update-mime-database %{_datadir}/mime &> /dev/null || :

%postun
/usr/bin/update-mime-database %{_datadir}/mime &> /dev/null || :

Note that similarly to the gtk-update-icon-cache code, these scriptlets should be run only if the user has update-mime-info installed and without a specific Requires: shared-mime-info. If shared-mime-info is not installed, update-mime-database won't be run when this package is installed. This does not matter because it will be run when the shared-mime-info package is installed.

Icon Cache

If an application installs icons into one of the subdirectories in %{_datadir}/icons/ (such as hicolor in the following examples), icon caches must be updated so that the installed icons show up in menus right after package installation. This consists of updating the timestamp of the top-level icon directory where the icons were installed, and running gtk-update-icon-cache. 'touch'ing the top-level dir is done so that environments compatible with the Icon theme specification can refresh their caches, and gtk-update-icon-cache which is additionally required for GNOME also does its work based on the dir timestamp.

Note that no dependencies should be added for this. If gtk-update-icon-cache is not available, there's nothing that would be needing the cache update, ditto if "touch" is not available, there's nothing that would benefit from icon cache updates installed yet either. Not adding the dependency on gtk-update-icon-cache (ie. gtk2 >= 2.6.0) or "touch" makes it easier to use the package (or the same specfile) on systems where it's not available nor needed, such as older distro versions or (very) trimmed down installations, and generally results in less entries in specfiles, rpmdb, and repo metadatas.

%post
/bin/touch --no-create %{_datadir}/icons/hicolor &>/dev/null || :

%postun
if [ $1 -eq 0 ] ; then
    /bin/touch --no-create %{_datadir}/icons/hicolor &>/dev/null
    /usr/bin/gtk-update-icon-cache -f %{_datadir}/icons/hicolor &>/dev/null || :
fi

%posttrans
/usr/bin/gtk-update-icon-cache -f %{_datadir}/icons/hicolor &>/dev/null || :

Systemd

Packages containing systemd unit files need to use scriptlets to ensure proper handling of those services. Services can either be enabled or disabled by default. To determine which case your specific service falls into, please refer to FESCo's policy here: Starting_services_by_default. On upgrade, a package may only restart a service if it is running; it may not start it if it is off. Also, the service may not enable itself if it is currently disabled.

New Packages

Note.png
What is a new package?
In this context, a new package is a package which has never included a SysV initscript.

Macroized scriptlets (Fedora 18+)

The systemd package in Fedora 18+ provides a set of helper macros to handle systemd scriptlet operations. These macros are functionally equivalent to the manual scriptlets used in older versions of Fedora, but they also add support for systemd "presets", as documented here: https://fedoraproject.org/wiki/Features/PackagePresets

Requires(post): systemd-units
Requires(preun): systemd-units
Requires(postun): systemd-units

[...]
%post
%{systemd_post} apache-httpd.service

%preun
%{systemd_preun} apache-httpd.service

%postun
%{systemd_postun} 

Please note that if the service is not enabled by default, systemctl will not list your service until it is either manually started (systemctl start foo.service) or manually enabled (systemctl enable foo.service).

If your service should be enabled by default, then use the following %post scriptlet instead of the one shown above:

%post
%{systemd_postun_with_restart} apache-httpd.service

For details on what these macros evaluate to, refer to the following sources:
http://cgit.freedesktop.org/systemd/systemd/tree/src/core/macros.systemd.in
http://www.freedesktop.org/software/systemd/man/daemon.html

Manual scriptlets (Fedora 17 or older)

Requires(post): systemd-units
Requires(preun): systemd-units
Requires(postun): systemd-units

[...]
%post
if [ $1 -eq 1 ] ; then 
    # Initial installation 
    /bin/systemctl daemon-reload >/dev/null 2>&1 || :
fi

%preun
if [ $1 -eq 0 ] ; then
    # Package removal, not upgrade
    /bin/systemctl --no-reload disable apache-httpd.service > /dev/null 2>&1 || :
    /bin/systemctl stop apache-httpd.service > /dev/null 2>&1 || :
fi

%postun
/bin/systemctl daemon-reload >/dev/null 2>&1 || :
if [ $1 -ge 1 ] ; then
    # Package upgrade, not uninstall
    /bin/systemctl try-restart apache-httpd.service >/dev/null 2>&1 || :
fi

Note that /bin/systemctl daemon-reload will automatically detect new systemd unit .service files placed into %{_unitdir}. There is no equivalent to 'chkconfig --add <service>', because it is unnecessary. However, if the service is not enabled by default, systemctl will not list your service until it is either manually started (systemctl start foo.service) or manually enabled (systemctl enable foo.service).

If your service should be enabled by default, then use the following %post scriptlet instead of the one shown above:

%post
if [ $1 -eq 1 ] ; then 
    # Initial installation
    /bin/systemctl enable apache-httpd.service >/dev/null 2>&1 || :
fi

Packages migrating to a systemd unit file from a SysV initscript

Warning.png
Packages are strictly forbidden from migrating to systemd within updates to a Fedora release. The migration is only allowed between Fedora releases.

When updating from a package containing a SysV initscript to a package containing a systemd unit file, we "start-over fresh" with default start and stop policy from the new package, and do not migrate what the user had previously configured. systemd provides a tool (systemd-sysv-convert --apply) to help do this conversion if the user wants after the package is updated.

Packages in this scenario need to include all of the above scriptlets from "New Packages" and also add the additional scriptlets below:

# This is actually needed for the %triggerun script but Requires(triggerun)
# is not valid.  We can use %post because this particular %triggerun script
# should fire just after this package is installed.
Requires(post): systemd-sysv

...

### A sysv => systemd migration contains all of the same scriptlets as a
### systemd package.  These are additional scriptlets

# Note: the NEVR in trigger scripts should all be the version in
# which the package switched to systemd unit files and the comparision
# should be less than.  Using <= the last version with the sysV script won't
# work for several reasons:
# 1) disttag is different between Fedora releases
# 2) An update in an old Fedora release may create a newer NEVR
#    Note that this means an update in an older Fedora release must be NEVR
#    lower than this.  Freezing the version and release of the old package and
#    using a number after the disttag is one way to do this.  Example:
#        httpd-1.0-1%{?dist} => httpd-1.0-1%{?dist}.1

%triggerun -- httpd < 1.0-2
# Save the current service runlevel info
# User must manually run systemd-sysv-convert --apply httpd
# to migrate them to systemd targets
/usr/bin/systemd-sysv-convert --save httpd >/dev/null 2>&1 ||:

# If the package is allowed to autostart:
/bin/systemctl --no-reload enable apache-httpd.service >/dev/null 2>&1 ||:

# Run these because the SysV package being removed won't do them
/sbin/chkconfig --del httpd >/dev/null 2>&1 || :
/bin/systemctl try-restart apache-httpd.service >/dev/null 2>&1 || :