Re: New SELinux toolchain build system

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Joshua Brindle wrote:
Vikram Ambrose wrote:
Joshua Brindle wrote:
Vikram Ambrose wrote:
Attached to this email is a tarball snapshot of a new build system
for the SELinux toolchain that I have been working on for the last
couple of months.

This autoconf/automake build system has many advantages over the
current Makefiles.

1) It is capable of correctly cross compiling all libraries and
 tools, including the python wrappers.
2) It allows the distro maintainer to set sysconfdir, prefix,
 pamdir, startupdir etc..
3) It employs no "hacks", it is to-the-book, clean, GNU Automake,
 Autoconf code, formatted to ~80 char line length.
4) It lets RPM spec and .deb control take advantage of the autoconf
 infrastructure already built into these package formats, allowing
 distro maintainers to quickly and effortlessly build distro
 packages.
5) It is completely self dependent, ie. one can build the complete
 toolchain now resolving dependencies within the build, without
 needing to link against a system wide library.

Due to an incompatibility with libtool, related to the current
naming convention of the python wrappers and classes, a patch is
supplied to rename:
 *) _selinux.so to libpyselinux.so
 *) _semanage.so to libpysemanage.so
 *) _audit2why.so to libpyaudit2why.so

Similarily, the update also renames:
 *) selinux.py to __init__.py
 *) semanage.py to __init__.py
and installs these plugins into their own sub site-packages folder.

For Example: libselinux/src/Makefile.am
line 35 # When the upstream tree is patched this should be removed
line 36 # and selinux.py must be renamed to __init__.py
line 37 __init__.py : selinux.py
line 38        cp selinux.py $@

A similar measure is taken in libsemanage/src/Makefile.am for
semanage.py.

On a similar note, the code currently uses a macro called SHARED,
which is currently being wrapped with a forced -include libtool_compat.h
using -DPIC (a libtool defined macro). This too can be removed if the
code
can be reformatted.

This build system has been thoroughly tested to function correctly.
Compiling natively for localhost, as well as cross compiling for
the following platforms:
 *) PPC32
 *) PPC64
 *) ARM (Versatile family)
 *) Common PC X86_64

It has also been tested on mainstream Linux distributions such
as Ubuntu 8.04 and Fedora 9.

This new build system is a replacement for the existing build system.
They cannot be used concurrently. I leave no guarantee on backward
compatibility after applying this update.

Recently a Ruby wrapper was added to the SELinux trunk, this build
system does not build it. Though with the current infrastructure in
place, it is a very simple procedure for anyone familiar with ruby
to include it.

This contribution comes out of the integration efforts WindRiver has
recently taken to add a SELinux feature to WindRiver Linux. It comes
with no obligation or indemnity from WindRiver or myself. Distributed
as not copyrighted, public domain software, in accordance with
libselinux.

I ask if this update can be merged into the main SELinux trunk.
I will help coordinate the merge if necessary.

I've just started looking at this. We need to be sure that it covers
current make targets like make swigify which generate the c wrappers
from swig interface files (we run this before checkins). Also make
test needs to work.

Also there is no top level ./configure so this won't let you build the
entire repo. I'd like something (even if it was just the makefile that
ran ./configure in each dir.

I also get a failure:
[root@misterfreeze libsepol]# autoreconf -iv
autoreconf: Entering directory `.'
autoreconf: configure.ac: not using Gettext
autoreconf: running: aclocal autoreconf: configure.ac: tracing
autoreconf: configure.ac: not using Libtool
autoreconf: running: /usr/bin/autoconf
autoreconf: running: /usr/bin/autoheader
autoreconf: running: automake --add-missing --copy --no-force
configure.ac:17: installing `./compile'
configure.ac:6: installing `./install-sh'
configure.ac:6: installing `./missing'
include/Makefile.am:2: whitespace following trailing backslash
include/Makefile.am:5: whitespace following trailing backslash
include/Makefile.am:8: whitespace following trailing backslash
include/Makefile.am:9: whitespace following trailing backslash
include/Makefile.am:11: whitespace following trailing backslash
include/Makefile.am:14: whitespace following trailing backslash
include/Makefile.am:20: whitespace following trailing backslash
src/Makefile.am:10: Libtool library used but `LIBTOOL' is undefined
src/Makefile.am:10:   The usual way to define `LIBTOOL' is to add
`AC_PROG_LIBTOOL'
src/Makefile.am:10:   to `configure.ac' and run `aclocal' and
`autoconf' again.
src/Makefile.am:10:   If `AC_PROG_LIBTOOL' is in `configure.ac', make
sure
src/Makefile.am:10:   its definition is in aclocal's search path.
src/Makefile.am: installing `./depcomp'
Makefile.am: installing `./INSTALL'
Makefile.am: required file `./NEWS' not found
Makefile.am: required file `./README' not found
Makefile.am: required file `./AUTHORS' not found
autoreconf: automake failed with exit status: 1


Formatting error caused by the sanitation process. An updated tarball is
attached to this email, with the following changes from the previous:

1) Fixed trailing white space in libsepol/src/Makefile.am
2) Added --enable-swig option to libselinux and libsemanage to regenerate
  c-wrapper
3) Added very very basic top level makefile, proof-of-concept to build
entire
  trunk. make -f Makefile.toolchain PREFIX=/some/where (Requested by
Joshua.B)
4) Minor python related cleanups in configure.ac for libselinux and
  libsemanage


Dan, Manoj, Caleb, you can skip to the bottom of this email

I have a branch with the latest patch, my current observations:
* checkpolicy/test isn't built (this is included on fedora and must be built)

That was done intentionally, I don't use it, so I didn't build it. That doesn't mean no one else can add 3 lines to a new Makefile.am.
* when I generated the configure scripts in a fresh repo apparently it felt the need to relicense libselinux as gplv3 (added a COPYING file)
If you do not provide a COPYING file, autoconf will generate one for you, so I suggest a COPYING, and any other dist file (eg, Changelog, NEWS, AUTHORS etc..) be made by hand and placed in the tree.
* the CFLAGS we previously had have been completely removed, we need those back, particularly the warning ones and they aren't necessarily consistent:
libsepol/src/Makefile:CFLAGS ?= -Werror -Wall -W -Wundef -Wshadow -Wmissing-noreturn -Wmissing-format-attribute
libsemanage/src/Makefile:CFLAGS ?= -Wall -W -Wundef -Wshadow -Wmissing-noreturn -Wmissing-format-attribute -Wno-unused-parameter
checkpolicy/Makefile:CFLAGS ?= -g -Wall -Werror -Wshadow -O2 -pipe -fno-strict-aliasing
libselinux/src/Makefile:CFLAGS ?= -Werror -Wall -W -Wundef -Wshadow -Wmissing-noreturn -Wmissing-format-attribute

Also, we filter out Werror from generated files (parser and lexer files, swig generated files) because they often cause warnings and we can't help it

there is a libblah_la_CFLAGS = in each Makefile.am, you can tack those onto it. But i strongly suggest you only add the necessary, eg -g, -O2 etc.. are not mandatory, and should not be forced on the package builder. autoconf has support for such by adding them on the configure line,
eg ./configure CFLAGS="-Werror -pipe -Wall" --prefix=/blah etc..

* make clean from the top doesn't actually clean anything, it just removes the makefiles

* there is no way to distclean from the top level, eg., make the repository look like it will when we package up a release
The top level makefile was just a proof-of-concept. If you need a top level makefile, then write one. The one i supplied was only intended to show that the tree builds.
* Why doesn't configure automatically find python? the vast majority of selinux installations are going to require the python bits so it should default to on and I'm not sure what the use of the autotools infrastructure is if it can't find the resources we need automatically
This is actually quite a complicated topic. Python is not very good at being used in a cross platform environment. And thus autoconf cannot be blamed for not being able to use python correctly in this manner. Autoconf depends on the "python" interpreter for information on which version and where python is installed. Now the problem with this is that, the build machine must have an identical python to your target machine, which is not possible on eg, x86_32 and ppc_64, as python generates arch dependent C-headers instead of #ifdef'ing the various architectures. There are more intricate problems that I could write a paper on, but i doubt you're interested. (eg Building on Linux for Windows, BSD, Solaris or vice versa, Or broken build systems like that of Redhat/Fedora)

To cut a long story short: No, you cannot build python wrappers outside your target _properly_ without explicitly telling the build system which python to use. To counter this, I have got autoconf to pick up the local python and suggest it to the user, this way autoconf doesn't guess and then get blamed for doing the "wrong thing". The user is the only one to blame if the package doesn't build.
* configure is currently doing alot of tests for unnecessary things, such as a fortran compiler. This is one of the problems I have using autotools, it takes time and adds complexity for no reason
I'll look into it, but a lot of it is done implicitly.
* building an embedded libselinux (--disable-rpm --disable-avc --disable-bool) fails:
	/usr/bin/ld: .libs/libselinux_la-load_policy.o: relocation R_X86_64_PC32 against undefined symbol `security_get_boolean_names_internal' can not be used when making a shared object; recompile with -fPIC

Yup, i'll look at that too.
I think that is probably enough for now..

--

I created an autotools branch on the upstream git repo so all new patches should be based on that instead of a tarball.
Also, I'd like distributions to start testing this branch as they are less likely to be affected by the issues raised above. Distributions will likely be mostly affected by the change of python bindings from _*.so to libpy*.so, and also will need to run configure with --with-python=/usr/include/python2.5 to have a useful build
yes, and --enable-swig too. As the pre-generated .c file will not be patched.


Working hard over the weekend hey method? give me svn write access to your autotools branch, this will speed up the initial ground work. I am moving onto other wild and wonderful things next week and would like to get this working for you guys as soon as possible.

Vikram.

--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to majordomo@xxxxxxxxxxxxx with
the words "unsubscribe selinux" without quotes as the message.

[Index of Archives]     [Selinux Refpolicy]     [Linux SGX]     [Fedora Users]     [Fedora Desktop]     [Yosemite Photos]     [Yosemite Camping]     [Yosemite Campsites]     [KDE Users]     [Gnome Users]

  Powered by Linux