Creating packages

See also → Package Policy Guide, which contains a wealth of extra technical information not covered here.

One of the things that we've attempted to do with OpenWrt's template system is make it incredibly easy to port software to OpenWrt. If you look at a typical package directory in OpenWrt you'll find three things:

  • package/Makefile
  • package/patches
  • package/files

The patches directory is optional and typically contains bug fixes or optimizations to reduce the size of the executable. The files directory is optional. It typically includes default config or init files.

The package Makefile is the important item because it provides the steps actually needed to download and compile the package.

Looking at one of the package makefiles, you'd hardly recognize it as a makefile. Through what can only be described as blatant disregard and abuse of the traditional make format, the Makefile has been transformed into an object oriented template which simplifies the entire ordeal.

Here, for example, is package/bridge/Makefile:

include $(TOPDIR)/
include $(INCLUDE_DIR)/
define Package/bridge
  TITLE:=Ethernet bridging configuration utility
  #DESCRIPTION:=This variable is obsolete. use the Package/name/description define instead!
define Package/bridge/description
 Ethernet bridging configuration utility
 Manage ethernet bridging; a way to connect networks together to
 form a larger network.
define Build/Configure
  $(call Build/Configure/Default,--with-linux-headers=$(LINUX_DIR))
define Package/bridge/install
        $(INSTALL_DIR) $(1)/usr/sbin
        $(INSTALL_BIN) $(PKG_BUILD_DIR)/brctl/brctl $(1)/usr/sbin/
$(eval $(call BuildPackage,bridge))

As you can see, there's not much work to be done; everything is hidden in other makefiles and abstracted to the point where you only need to specify a few variables.

  • PKG_NAME - The name of the package, as seen via menuconfig and ipkg. Avoid using underscores in the package name, to avoid build failures--for example, the underscore separates name from version information, and may confuse the build system in hard-to-spot places.
  • PKG_VERSION - The upstream version number that we're downloading
  • PKG_RELEASE - The version of this package Makefile. Should be initially set to 1, and reset to 1 whenever the PKG_VERSION is changed. Increment it when PKG_VERSION stays the same, but when there are functional changes to the installed artifacts.
  • PKG_LICENSE - The license(s) the package is available under, SPDX form.
  • PKG_LICENSE_FILES- file containing the license text
  • PKG_BUILD_DIR - Where to compile the package
  • PKG_SOURCE - The filename of the original sources
  • PKG_SOURCE_URL - Where to download the sources from (directory)
  • PKG_HASH - A checksum to validate the download. It can be either a MD5 or SHA256 checksum, but SHA256 should be used, see scripts/
  • PKG_CAT - How to decompress the sources (zcat, bzcat, unzip)
  • PKG_URL. - Upstream project homepage
  • PKG_BUILD_DEPENDS - Packages that need to be built before this package. Use this option if you need to make sure that your package has access to includes and/or libraries of another package at build time. Specify the directory name (i.e. openssl) rather than the binary package name (i.e. libopenssl). This build variable only establishes the build time dependency. Use DEPENDS to establish the runtime dependencies. This variable uses the same syntax as DEPENDS below.
  • PKG_CONFIG_DEPENDS - specifies which config options influence the build configuration and should trigger a rerun of Build/Configure on change
  • PKG_INSTALL - Setting it to “1” will call the package's original “make install” with prefix set to PKG_INSTALL_DIR
  • PKG_INSTALL_DIR - Where “make install” copies the compiled files
  • PKG_FIXUP - See below
  • PKG_CVE_IGNORE - Variable for defining CVEs that don't apply to this version of the package due to features not enabled, or affecting other platforms (e.g. Windows issues or features that are not used and so not relevant)
  • PKG_CVE_FIXED - Variable for defining CVEs that are patches in the current version, but aren't properly marked as fixed at in the current version

Optional support for fetching sources from a VCS (git, bzr, svn, etc), see Use source repository below for more information:

  • PKG_SOURCE_PROTO - the protocol to use for fetching the sources (git, svn, etc).
  • PKG_SOURCE_URL - source repository to fetch from. The URL scheme must be consistent with PKG_SOURCE_PROTO (e.g. git://), but most VCS accept http:// or https:// URLs nowadays.
  • PKG_SOURCE_VERSION - must be specified, the commit hash or SVN revision to check out.
  • PKG_SOURCE_DATE - a date like 2017-12-25, will be used in the name of generated tarballs.
  • PKG_MIRROR_HASH - SHA256 checksum of the tarball generated from the source repository checkout (previously named PKG_MIRROR_MD5SUM). See below for details.
  • PKG_SOURCE_SUBDIR - where the temporary source checkout should be stored, defaults to $(PKG_NAME)-$(PKG_VERSION)

The PKG_* variables define where to download the package from; @SF is a special keyword for downloading packages from sourceforge. The md5sum is used to verify the package was downloaded correctly and PKG_BUILD_DIR defines where to find the package after the sources are uncompressed into $(BUILD_DIR). PKG_INSTALL_DIR defines where the files will be copied after calling “make install” (set with the PKG_INSTALL variable), and after that you can package them in the install section.

At the bottom of the file is where the real magic happens, “BuildPackage” is a macro setup by the earlier include statements. BuildPackage only takes one argument directly -- the name of the package to be built, in this case “bridge”. All other information is taken from the define blocks. This is a way of providing a level of verbosity, it's inherently clear what the DESCRIPTION variable in Package/bridge is, which wouldn't be the case if we passed this information directly as the Nth argument to BuildPackage.

Avoid reuse of PKG_NAME in call, define and eval lines for consistency and readability. Write the full name instead.

There is support for a range of sanity checks like mismatched checksums.

To check your package:

make package/nlbwmon/download
make package/nlbwmon/check V=s

To automatically attempt to fix the Makefile:

make package/nlbwmon/check V=s FIXUP=1

Note: despite the similar name, this has nothing to do with the PKG_FIXUP variable presented below.

Some packages that use autotools end up needing fixes to work around autotools using host tools instead of the build environment tools. OpenWrt defines some PKG_FIXUP rules to help work around this.


Any variations of this you see in the wild are simply aliases for these.


This fixup performs

  • autoreconf -f -i
  • touch required but maybe missing files
  • ensures that openwrt-libtool is linked
  • suppresses autopoint/gettext

If the shipped automake recipes are broken beyond repair, then simply find instances of libtool, detect their version and apply OpenWrt fix patches to it.


This fixup suppresses version mismatch errors in automake's gettext support.


Packages that are using Autotools should work with simply “PKG_FIXUP:=autoreconf”. However there might be issues with required versions.

:!: Instead of patching ./configure, one should fix the file from which ./configure is generated in autotools: (or, for very old packages). Another important file is from which Makefiles (with configure output) are generated.

OpenWrt Buildroot supports many different ways to download external source code.

Most packages use a packed .tar.gz, .tar.bz2, .tar.xz or similar source code file.

PKG_SOURCE_PROTO supports download from various repositories to integrate development versions:


Besides the source repository PKG_SOURCE_URL, you also need to specify which exact version you are building using PKG_SOURCE_VERSION e.g. a commit hash for git, or a revision number for svn. The PKG_SOURCE_VERSION can be a git tag and specified like PKG_SOURCE_VERSION:=v$(PKG_VERSION).

Buildroot will first clone the source repository, and then generate a tarball from the source repository, with a name like dl/odhcpd-2017-08-16-94e65ee0.tar.xz.

You should also define PKG_MIRROR_HASH with the SHA256 checksum of this generated tarball. This way, users building OpenWrt will directly download the generated tarball from a buildbot and verify its checksum, thus avoiding a clone of the source repository.

:!: The tarballs generated from svn checkouts are not reproducible, so you should avoid defining PKG_MIRROR_HASH when building from svn!

To generate PKG_MIRROR_HASH automatically, use the following (replace package/odhcpd by your package):

# First add "PKG_MIRROR_HASH:=skip" to the package Makefile and/or "HASH:=skip", if required.
make package/odhcpd/download V=s
make package/odhcpd/check FIXUP=1 V=s

Complete git example:


History: PKG_MIRROR_MD5SUM was introduced in 2011 and renamed to ''PKG_MIRROR_HASH'' in 2016.

It is also possible to have the source code in the package/<packagename> directory. Often a ./src/ subdirectory is used.

Examples: px5g , px5g-standalone

Bundled source code does not need overriding.

You can download additional data from external sources.

define Download/usb_ids
$(eval $(call Download,usb_ids))

and unpack it or integrate it into the build process

define Build/Prepare
        $(CP) $(DL_DIR)/$(USB_IDS_FILE) $(PKG_BUILD_DIR)/usb.ids

You can modify UNPACK_CMD or call/modify PKG_UNPACK manually in your Build/Prepare section.

UNPACK_CMD=ar -p "$(DL_DIR)/$(PKG_SOURCE)" data.tar.xz | xzcat | tar -C $(1) -xf -
define Build/Prepare
#       we have to download additional stuff before patching
        (cd $(PKG_BUILD_DIR) && ./contrib/download_prerequisites)
Examples: px5g, px5g-standalone, usbutils, debootstrap, gcc, 

matches the argument passed to buildroot, this describes the package the menuconfig and ipkg entries. Within Package/ you can define the following variables:

  • SECTION - The type of package (currently unused)
  • CATEGORY - Which menu it appears in menuconfig
  • TITLE - A short description of the package
  • DESCRIPTION - (deprecated) A long description of the package
  • URL - Where to find the original software
  • MAINTAINER - (required for new packages) Who to contact concerning the package
  • DEPENDS - (optional) Which packages must be built/installed before this package. See below for the syntax
  • EXTRA_DEPENDS - (optional) Runtime dependencies, don't get built, only added to package control file
  • PROVIDES - (optional) allow to define a virtual package that might be provided by multiple real-packages
  • PKGARCH - (optional) Set this to “all” to produce a package with “Architecture: all” (See below)
  • USERID - (optional) a username:groupname pair to create at package installation time.
PKGARCH (optional)

By default, packages are built for the target architecture, and the ipk files generated are tagged that way. This is normally correct, for any compiled code, but if a package only contains scripts or resources, marking it with PKGARCH:=all will make a single ipk file that can be installed on any target architecture. (It will still be compiled into bin/packages/arch/, however.)

Package/conffiles (optional)

A list of config files installed by this package, one file per line. The file list section should not be indented: no leading tabs or spaces in the section.


A free text description of the package

Build/Prepare (optional)

A set of commands to unpack and patch the sources. You may safely leave this undefined.

Build/Configure (optional)

You can leave this undefined if the source doesn't use configure or has a normal config script, otherwise you can put your own commands here or use “$(call Build/Configure/Default,)” as above to pass in additional arguments for a standard configure script.

Build/Compile (optional)

How to compile the source; in most cases you should leave this undefined, because then the default is used, which calls make. If you want to pass special arguments to make, use e.g. “$(call Build/Compile/Default,FOO=bar)”

Build/Install (optional)

How to install the compiled source. The default is to call “make install”. Again, to pass special arguments or targets, use “$(call Build/Install/Default,install install-foo)”. Note that you need put all the needed make arguments here. If you just need to add something to the “install” argument, don't forget the “install” itself.

Build/InstallDev (optional)

For things needed to compile packages against it (static libs, header files), but that are of no use on the target device.

Build/Clean (optional)

For things needed to be wiped out during cleanup procedure.


A set of commands to copy files into the ipkg which is represented by the $(1) directory. As source you can use relative paths which will install from the unpacked and compiled source, or $(PKG_INSTALL_DIR) which is where the files in the Build/Install step above end up.


The actual text of the script which is to be executed before installation. Don't forget to include the #!/bin/sh. If you need to abort installation, have the script return false.


The actual text of the script which is to be executed after installation. Don't forget to include the #!/bin/sh. Alternatively you can also use an uci-default script which will be executed automatically on runtime installations by opkg or embedding into an image.


The actual text of the script which is to be executed before removal. Don't forget to include the #!/bin/sh. If you need to abort removal, have the script return false.


The actual text of the script which is to be executed after removal. Don't forget to include the #!/bin/sh.

The reason that some of the defines are prefixed by “Package/” and others are simply “Build” is because of the possibility of generating multiple packages from a single source. OpenWrt works under the assumption of one source per package Makefile, but you can split that source into as many packages as desired. Since you only need to compile the sources once, there's one global set of “Build” defines, but you can add as many “Package/” defines as you want by adding extra calls to BuildPackage -- see the dropbear package for an example.

Some software has no Makefile directly at the root of the tarball. For instance, it is common to have a src/ directory with all source files and a Makefile. The problem is that OpenWRT's build system will try to run make in PKG_BUILD_DIR; this will fail if there is no Makefile there. To solve this problem, use the MAKE_PATH variable, for instance:


This path is relative to PKG_BUILD_DIR and defaults to .. Alternatively, you can override Build/Compile (see above), though this is more work.

Various types of dependencies can be specified, which require a bit of explanation for their differences. More documentation is available at Using Dependencies

+<foo> Package will depend on package <foo> and will select it when selected.
<foo> Package will depend on package <foo> and will be invisible until <foo> is selected.
@FOO Package depends on the config symbol CONFIG_FOO and will be invisible unless CONFIG_FOO is set. This usually used for depending on certain Linux versions or targets, e.g. @TARGET_foo will make a package only available for target foo. You can also use boolean expressions for complex dependencies, e.g. @(!TARGET_foo&&!TARGET_bar) will make the package unavailable for foo and bar.
+FOO:<bar> Package will depend on <bar> if CONFIG_FOO is set, and will select <bar> when it is selected itself. The typical use case would be if there are compile time options for this package toggling features that depend on external libraries. :!: Note that the + replaces the @. :!: There is limited support for boolean operators here compared to the @ type above. Negation ! is only supported to negate the whole condition. Parentheses are ignored, so use them only for readability. Like C, && has a higher precedence than ||. So +(YYY||FOO&&BAR):package will select package if CONFIG_YYY is set or if both CONFIG_FOO and CONFIG_BAR are set.
@FOO:<bar> Package will depend on <bar> if CONFIG_FOO is set, and will be invisible until <bar> is selected when CONFIG_FOO is set.

Some typical config symbols for (conditional) dependencies are:

TARGET_<foo> Target <foo> is selected
TARGET_<foo>_<bar> If the target <foo> has subtargets, subtarget <foo> is selected. If not, profile <foo> is selected. This is in addition to TARGET_<foo>
TARGET_<foo>_<bar>_<baz> Target <foo> with subtarget <bar> and profile <baz> is selected.
LINUX_3_X Linux version used is 3.x.*
LINUX_2_6_X Linux version used is 2.6.x.* (:1: only used for backfire and earlier)
LINUX_2_4 Linux version is 2.4 (:!: only used in backfire and earlier, and only for target brcm-2.4)
USE_UCLIBC, USE_GLIBC, USE_EGLIBC To (not) depend on a certain libc.
BROKEN Package doesn't build or work, and should only be visible if “Show broken targets/packages” is selected. Prevents the package from failing builds by accidentally selecting it.
IPV6 IPv6 support in packages is selected.

Note that the syntax above applies to the DEPENDS field only.

PKG_BUILD_DEPENDS does not use + or @, but otherwise uses the same syntax. You can write FOO:bar to build bar if CONFIG_FOO is defined (if you want to build bar if package FOO is selected, use PACKAGE_FOO:bar). You may use !, || and && as explained above. PKG_BUILD_DEPENDS uses the name from the PKG_NAME, not the individual packages. For example, if you want to have openssl to be a build-dependency, you would write PKG_BUILD_DEPENDS:=openssl, whereas if your package depends and selects the openssl library, you'd have DEPENDS:=+libopenssl. Notice that there's no installable package named openssl: the library is libopenssl, the utility is openssl-util, but their PKG_NAME is openssl. If you need a host-built package, append /host to the PKG_NAME, e.g. PKG_BUILD_DEPENDS:=openssl/host. A package listed in PKG_BUILD_DEPENDS will be built even if it is not selected in make menuconfig.

EXTRA_DEPENDS does not accept the conditional or select dependency syntax. However, unlike DEPENDS and PKG_BUILD_DEPENDS, it is generated at package-build-time, so you may use Makefile functions to add them conditionally. For example, to get the equivalent of DEPENDS:=@SOMESYMBOL:foo +PACKAGE_somepkg:bar, you can write EXTRA_DEPENDS:=$(if $(CONFIG_SOMESYMBOL),foo) $(if $(CONFIG_PACKAGE_somepkg),bar). :!: Note that neither foo or bar will be selected in menuconfig, or guaranteed to be built before your package, even if selected somewhere else. For this reason, it is seldom used in packages.

EXTRA_DEPENDS is more often used to depend on specific versions, by adding the desired version specification in parentheses, using >=,>,<,<=,=. Make sure you add the PKG_RELEASE number if you're using '=', such as EXTRA_DEPENDS:=foo (=2.0.0-1): where foo's PKG_VERSION is 2.0.0, and PKG_RELEASE is 1.


        --disable-native-affinity \
        --disable-unicode \
        ac_cv_file__proc_stat=yes \
        ac_cv_file__proc_meminfo=yes \
        ac_cv_func_malloc_0_nonnull=yes \

To set variables (autoconfig internal ones or CPPFLAGS,CFLAGS, CXXFLAGS, LDFLAGS for example) or configure arguments. Setting configure arguments is common. Setting VARS is needed when the autoconf source script does not work well on cross compilation or finding libraries.

:!: The article packages.flags contains more information and examples about overriding and setting these.

In order to build your package, you may require some extra build tools or libraries that are not already in the standard OpenWrt toolchain. These must be built as part of the OpenWrt build process ie. you cannot rely on them being installed via the build system's package manager (or manually by the user or administrator). This is so that OpenWrt can be built reliably and repeatably on a wide variety of machines. These are referred to as host tools because they run on (or are compiled for) the host, not the target.

If your package requires host tools in order to be built for the target machine, these should go in PKG_BUILD_DEPENDS and will end with /host. For example, the json-glib package requires the Meson build system to generate build files, as well as Glib2 on the host, so it has:

PKG_BUILD_DEPENDS:=glib2/host meson/host

A package might itself provide host tools, and building or using those might require other host tools to be built first. These other tools go in HOST_BUILD_DEPENDS. For example, the host tool that the Meson package provides requires another build tool, Ninja, so it has this line:


The makefile for a package that provides host tools will:

  • Include $(INCLUDE_DIR)/ You can look at this makefile for the details of the host tool build process.
  • Have sections similar to the Build/... sections of other packages, but they will start with Host/. For example, Host/Configure, Host/Compile and Host/Install are common.
  • Call $(eval $(call HostBuild)) at the end.

Some examples of packages that provide host tools (and their makefiles):

  • Samba 4 - note that the Samba 4 target package actually depends on the Samba 4 host package provided in the same makefile
  • Go (golang) - this is a much more complex example, and useful if you're thinking of adding support for a new language in OpenWrt.

If you want to build only the host tool to test or check a compilation error for host compilation, then you could also build only the host tool with the following command.

  • Compile:
    make ./package/<package_name>/host/compile
  • Clean:
    make ./package/<package_name>/host/clean
  • Update:
    make ./package/<package_name>/host/update

The make arguments QUILT=1 and V=s are also valid.


If you want to patch the host and target tool separately, then you have to add HOST_PATCH_DIR:=./<directory>. For example, add HOST_PATCH_DIR:=./patches-host to the Makefile so the host tool has its own patch directory. The target tool will still use the standard patch directory ./patches in its package directory.


All variables in your pre/post install/removal scripts should have double ($$) instead of a single ($) string characters. This will inform “make” to not interpret the value as a variable, but rather just ignore the string and replace the double $$ by a single $ -- More Info

After you've created your package Makefile, the new package will automatically show in the menu the next time you run “make menuconfig” and if selected will be built automatically the next time “make” is run.

DESCRIPTION is obsolete, use Package/PKG_NAME/description.

If you would like to configure your package installation/compilation in the menuconfig you can do the following: Add MENU:=1 to your package definition like this:

define Package/mjpg-streamer
  DEPENDS:=@!LINUX_2_4 +libpthread-stubs +jpeg

Create a config key in the Makefile:

define Package/mjpg-streamer/config
	source "$(SOURCE)/"

Create a file directory where the Makefile is located with the content like this:

	# Mjpg-streamer configuration
	menu "Configuration"
		depends on PACKAGE_mjpg-streamer
		bool "Autostart enabled"
		default n
		menu "Input plugins"
			depends on PACKAGE_mjpg-streamer
				bool "File input plugin"
					You can stream pictures from jpg files on the filesystem
				default n
				bool "UVC input plugin"
					You can stream pictures from an Universal Video Class compatible webcamera
				default y
				int "Maximum FPS"
				default 15
				int "Picture height"
				default 640
				int "Picture width"
				default 480
				string "Device"
				default /dev/video0
				bool "GSPCA input plugin"
					You can stream pictures from a gspca supported webcamera Note this module is deprecated, use the UVVC plugin instead
				default n
		# ......

Above, you can see examples for various types of config parameters. Finally, you can check your configuration parameters in your Makefile in the following way (note that you can reference the parameter's value with its name prefixed with CONFIG_):

    $(CP) $(PKG_BUILD_DIR)/ $(1)/usr/lib

If you are still working on the application, itself, at the same time as you are working on the packaging, it can be very useful to have OpenWrt build your work in progress code, rather than a specific version+md5sum combination checked out of revision control, or downloaded from your final “release” location. There are a few ways of doing this.

This is an option in menuconfig. See “Advanced configuration options (for developers)” → “Enable package source tree override”

This allows you to point to a local git tree. (And only git) Say your package is defined in my_cool_feed/awesome_app.

ln -s /path/to/local/awesome_app_tree/.git feeds/my_cool_feed/awesome_app/git-src
make package/awesome_app/{clean,compile} V=s

Benefits of this approach are that you don't need any special infrastructure in your package makefiles, they stay completely as they would be for a final build. The downside is that it only builds whatever is currently committed in HEAD of your local tree. (This could be a private testing branch, but everything you want to include in the package must be committed: uncommitted local changes will not be included in the build.) This will also use a separate directory for building and checking out the code. So, any built objects in your local git tree (for example, a build targeting a different architecture) will be left alone, but whichever branch is checked out in your tree determines where HEAD is.

As part of deprecating (below), a method to point directly to local source was introduced.

make package/awesome_app/clean V=s
make package/awesome_app/prepare USE_SOURCE_DIR=~/src/awesome_src V=s
make package/awesome_app/clean V=s

(V=s is optional above)

This doesn't require any config change to enable rules, doesn't require that you have a local git tree, and doesn't require any files to be committed.

At least at present, however, this has the following problems:

  • make clean doesn't clean the source link directory, but still seems to be removing a link
  • make prepare needs to be run every time
  • make package/awesome_app/{clean,compile} USE_SOURCE_DIR=~blah doesn't work
  • Seems to have bad interactions with leaving USE_SOURCE_DIR set for other (dependent?) packages.

See for the original discussion of this new feature

:!: Since OpenWrt 19.07.5, USE_SOURCE_DIR only works with packages that have a valid PKG_MIRROR_HASH. For packages under development, hash checking can be disabled by setting PKG_MIRROR_HASH=skip. This allows using USE_SOURCE_DIR for packages pulling sources from a developer branch (e.g. having PKG_SOURCE_VERSION:=branchname) without updating PKG_MIRROR_HASH every time.

:!: Don't use this anymore!

Support for this style of local source building was removed. This style required a permanent modification to your package makefile, and then entering a path via menuconfig to where the source was found. It was fairly easy to use, and didn't care whether your local source was in git or svn or visual source safe even, but it had the major downside that the “clean” target simply didn't work (as it simply removed a symlink for cleaning).

If you build a current OpenWrt tree, with packages that still attempt to use this style of local building, you will receive errors like so: ERROR: please fix package/feeds/feed_name/application_name/Makefile - see logs/package/feeds/feed_name/application_name/dump.txt for details

If you need/want to keep using this style, where it's available, make sure you include without failing if it was missing:

-include $(INCLUDE_DIR)/

A kernel module is an installable program which extends the behavior of the linux kernel. A kernel module gets loaded after the kernel itself, E.G. using insmod.

Many kernel programs are included in the Linux source distribution; typically the kernel build may be configured to, for each program,

  • compile it into the kernel as a built-in,
  • compile it as a loadable kernel module, or
  • ignore it.

See FIX:Customizingthekerneloptions customizing the kernel options for including it in the kernel.

To include one of these programs as a loadable module, select the corresponding kernel option in the OpenWrt configuration (see Build Configuration). If your favorite kernel module does not appear in the OpenWrt configuration menus, you must add a stanza to one of the files in the package/kernel/linux/modules directory. Here is an example extracted from .../modules/

define KernelPackage/loop
  TITLE:=Loopback device support
  AUTOLOAD:=$(call AutoLoad,30,loop)
define KernelPackage/loop/description
 Kernel module for loopback device support
$(eval $(call KernelPackage,loop))

Changes to the *.mk files are not automatically picked up by the build system. To force re-reading the metadata, either touch the kernel package Makefile using touch package/kernel/linux/Makefile (on older revisions touch package/kernel/Makefile) or to delete the tmp/ directory of the buildroot.

You can also add kernel modules which are not part of the linux source distribution. In this case, a kernel module appears in the package/ directory, just as any other package does. The package/Makefile uses KernelPackage/xxx definitions in place of Package/xxx.

For example, here is package/madwifi/Makefile:

# Copyright (C) 2006
# This is free software, licensed under the GNU General Public License v2.
# See /LICENSE for more information.
# $Id$
include $(TOPDIR)/
include $(INCLUDE_DIR)/
include $(INCLUDE_DIR)/
ifeq ($(ARCH),mips)
ifeq ($(ARCH),mipsel)
ifeq ($(ARCH),i386)
ifeq ($(ARCH),armeb)
ifeq ($(ARCH),powerpc)
ifneq ($(CONFIG_LINUX_2_4_AR531X),)
ifneq ($(CONFIG_LINUX_2_6_ARUBA),)
  BUS:=PCI AHB	# no suitable HAL for AHB yet.
ifeq ($(findstring AHB,$(BUS)),AHB)
ifeq ($(findstring PCI,$(BUS)),PCI)
	wlan \
	wlan_scan_ap \
	wlan_scan_sta \
	ath_hal \
	ath_rate_$(RATE_CONTROL) \
	wlan_acl \
	wlan_ccmp \
	wlan_tkip \
	wlan_wep \
ifeq ($(findstring AHB,$(BUS)),AHB)
ifeq ($(findstring PCI,$(BUS)),PCI)
define KernelPackage/madwifi
  SUBMENU:=Wireless Drivers
  DEFAULT:=y if LINUX_2_6_BRCM |  LINUX_2_6_ARUBA |  LINUX_2_4_AR531X |  LINUX_2_6_XSCALE, m if ALL
  TITLE:=Driver for Atheros wireless chipsets
	This package contains a driver for Atheros 802.11a/b/g chipsets.
  FILES:= \
		$(PKG_BUILD_DIR)/ath/ath_hal.$(LINUX_KMOD_SUFFIX) \
  AUTOLOAD:=$(call AutoLoad,50,$(MADWIFI_AUTOLOAD))
		LDOPTS=" " \
		ATH_RATE="ath_rate/$(RATE_CONTROL)" \
ifeq ($(findstring AHB,$(BUS)),AHB)
  define Build/Compile/ahb
ifeq ($(findstring PCI,$(BUS)),PCI)
  define Build/Compile/pci
define Build/Compile
	$(call Build/Compile/ahb)
	$(call Build/Compile/pci)
define Build/InstallDev
	$(INSTALL_DIR) $(STAGING_DIR)/usr/include/madwifi
	$(CP) $(PKG_BUILD_DIR)/include $(STAGING_DIR)/usr/include/madwifi/
	$(INSTALL_DIR) $(STAGING_DIR)/usr/include/madwifi/net80211
	$(CP) $(PKG_BUILD_DIR)/net80211/*.h $(STAGING_DIR)/usr/include/madwifi/net80211/
define KernelPackage/madwifi/install
	$(INSTALL_DIR) $(1)/etc/init.d
	$(INSTALL_DIR) $(1)/lib/modules/$(LINUX_VERSION)
	$(INSTALL_DIR) $(1)/usr/sbin
	$(INSTALL_BIN) ./files/madwifi.init $(1)/etc/init.d/madwifi
	$(CP) $(PKG_BUILD_DIR)/tools/{madwifi_multi,80211debug,80211stats,athchans,athctrl,athdebug,athkey,athstats,wlanconfig} $(1)/usr/sbin/
$(eval $(call KernelPackage,madwifi))

The use of MODPARAMS

If a module require some special param to be passed on Autoload, it's available MODPARAMS using the following syntax:

MODPARAMS.module_ko:=example_param1=example_value example_param2=example_value2

For example here is package/kernel/cryptodev-linux/Makefile:

define KernelPackage/cryptodev
	SUBMENU:=Cryptographic API modules
	TITLE:=Driver for cryptographic acceleration
	DEPENDS:=+kmod-crypto-authenc +kmod-crypto-hash
	AUTOLOAD:=$(call AutoLoad,50,cryptodev)

Make a Kernel Module required for boot

Some modules may be required for the correct operation of the device. One example would be an ethernet driver required for the correct operation of the switch on the device.

To flag a Kernel Module this way it's needed to append 1 to AUTOLOAD at the end.

This cause the module file to get placed in /etc/modules-boot.d/ instead of /etc/modules.d/, modules-boot.d is processed by procd init before launching preinit and correctly works both in a normal boot and in a failsafe boot. All of this is with the assumption that the module is installed in the firmware and not with OPKG on a loaded system as it needs to be present before /overlay is mounted. (OPKG installed module are present only in after /overlay is mounted)

For example here is phy-realtek in package/kernel/linux/modules/

define KernelPackage/phy-realtek
   TITLE:=Realtek Ethernet PHY driver
   AUTOLOAD:=$(call AutoLoad,18,realtek,1)

INSTALL_DIR, INSTALL_BIN, INSTALL_DATA are used for creating a directory, copying an executable, or copying a data file. +x is set on the target file for INSTALL_BIN, independent of its mode on the host.

From the big document:


A set of commands to copy files out of the compiled source and into the ipkg which is represented by the $(1) directory. Note that there are currently 4 defined install macros:

install -d -m0755
install -m0755
install -m0644
install -m0600

If you want to install a service (something that should start/stop at boot time, that has a /etc/init.d/blah script), read the Init Scripts section of the Technical Reference and the Procd init scripts section of the Developer's Guide. A key point is to make sure that the init.d script can be run on the host. At image build time, all init.d scripts found are run on the host, looking for the START=20/STOP=99 lines. This is what installs the symlinks in /etc/rc.d.

Packages have default postinst/prerm scripts that will run /etc/init.d/foo enable (creating the symlinks) or /etc/init.d/foo disable (removing the symlinks) when they are installed/removed by opkg.

Very basic example of a suitable init.d script. Please note that the newer style version does not work properly with interpreted executables (i.e. scripts). That is because start-stop-daemon is used by service_stop() in a way that it makes it confuse the script name with the interpreter name.

:!: procd style init is used in some init.d scripts since this commit. See procd-init-scripts for more details on that.

#!/bin/sh /etc/rc.common
# "new(er)" style init script
# Look at /lib/functions/ on a running system for explanations of what other SERVICE_
# options you can use, and when you might want them.
start() {
        service_start /usr/bin/$APP
stop() {
        service_stop /usr/bin/$APP
#!/bin/sh /etc/rc.common
# NOTE - this is an old style init script #
start() {
        start-stop-daemon -S -x $APP -p $PID_FILE -m -b
stop() {
        start-stop-daemon -K -n $APP -p $PID_FILE -s TERM
        rm -rf $PID_FILE

See Configuration in scripts for details on how to access UCI configuration information from an init.d script, for instance to set command-line parameters or to generate a config file for your service.

Packages are maintained in a separate repository to reduce maintenance overhead. The general guidelines for OpenWrt still apply, but see the README in the packages repository for latest information.

See the original announcement of this change.

This website uses cookies. By using the website, you agree with storing cookies on your computer. Also you acknowledge that you have read and understand our Privacy Policy. If you do not agree leave the website.More information about cookies
  • Last modified: 2024/05/22 20:59
  • by bam