When cross-compiling systemd, the introspection XML files fail to be
generated because the systemd host binary is not executable. This patch
works around this by putting the introspection XML data into separate
ELF sections and extracting them from the binary when generating the XML
files.
The extracted XML data is passed through the strings utility in order to
strip the trailing NUL character. A small AWK script is used to prepend
the doctype and add the opening and closing node tags respectively.
Finally, the C preprocessor is used to substitute the correct doctype
information from the D-Bus header files.
Previously Ubuntu was treated as being equivalent to Debian, but the two
distributions require different behaviour in certain places. This commit does
not change the behaviour of systemd on either distro but it creates a
framework for changes to be introduced by later commits.
The following previously meant "Target is Debian or Ubuntu".
* configure option "--with-distro=debian"
* C preprocessor symbol "TARGET_DEBIAN"
* Automake conditional "TARGET_DEBIAN"
After this commit, all of the above are redefined to mean "Target is Debian"
The following are introduced to mean "Target is Ubuntu".
* configure option "--with-distro=ubuntu"
* C preprocessor symbol "TARGET_UBUNTU"
* Automake conditional "TARGET_UBUNTU"
Most code written for Debian will also be applicable to Ubuntu. An extra
Automake conditional "TARGET_DEBIAN_OR_UBUNTU" is introduced to avoid
duplication of code that would otherwise occur.
This commit updates configure.ac, Makefile.am and distro-specific source files
in line with the above definitions.
Reduce number of exported symbols with -fvisibility=hidden by default,
this is safe as we're not generating and loadable library and our
binaries should have no exported symbol other than main(). This alone
reduces around 4kb per binary.
It will also request GCC to emit every function and data variable in
its own section, then request the linker to remove unused
sections. This reduces the size of utility tools
(/lib/systemd/systemd-*) by half or even more (in my system some
binaries went from 84kb to 32kb).
This patch adds a cpp definition HAVE_SYSV_COMPAT that is used to
isolate code dealing with /etc/init.d and /etc/rcN.d for systems where
it does not make sense (one that does not use sysv or one that is fully
systemd native).
The patch tries to be as little intrusive as possible, however in
order to minimize the number of #ifdef'ed regions I've reordered some
code in path-lookup.c:lookup_paths_init() where all code dealing with
sysv is now isolated under running_as == MANAGER_SYSTEM as well.
Moreover, In struct Service, some fields were rearranged to reduce
the number of ifdefs.
Lennart's suggestions were fixed and squashed with the original patch,
that was sent by Gustavo Sverzut Barbieri (barbieri@profusion.mobi).
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 07/28/2010 05:57 AM, Kay Sievers wrote:
> On Wed, Jul 28, 2010 at 11:43, Lennart Poettering
> <lennart@poettering.net> wrote:
>> On Mon, 26.07.10 16:42, Daniel J Walsh (dwalsh@redhat.com) wrote:
>>> tcontext=system_u:object_r:device_t:s0 tclass=chr_file
>>> type=1400 audit(1280174589.476:7): avc: denied { read } for pid=1
>>> comm="systemd" name="autofs" dev=devtmpfs ino=9482
>>> scontext=system_u:system_r:init_t:s0
>>> tcontext=system_u:object_r:device_t:s0 tclass=chr_file
>>> type=1400 audit(1280174589.476:8): avc: denied { read } for pid=1
>>> comm="systemd" name="autofs" dev=devtmpfs ino=9482
>>> scontext=system_u:system_r:init_t:s0
>>> tcontext=system_u:object_r:device_t:s0 tclass=chr_file
>>>
>>> Lennart, we talked about this earlier. I think this is caused by the
>>> modprobe calls to create /dev/autofs. Since udev is not created at the
>>> point that init loads the kernel modules, the devices get created with
>>> the wrong label. Once udev starts the labels get fixed.
>>>
>>> I can allow init_t to read device_t chr_files.
>>
>> Hmm, I think a cleaner fix would be to make systemd relabel this device
>> properly before accessing it? Given that this is only one device this
>> should not be a problem for us to maintain, I think? How would the
>> fixing of the label work? Would we have to spawn restorecon for this, or
>> can we actually do this in C without too much work?
>
> I guess we can just do what udev is doing, and call setfilecon(), with
> a context of an earlier matchpathcon().
>
> Kay
> _______________________________________________
> systemd-devel mailing list
> systemd-devel@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Here is the updated patch with a fix for the labeling of /dev/autofs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.14 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/
iEYEARECAAYFAkxQMyoACgkQrlYvE4MpobNviACfWgxsjW2xzz1qznFex8RVAQHf
gIEAmwRmRcLvGqYtwQaZ3WKIg8wmrwNk
=pC2e
It seems to work on my machine.
/proc/1/fd/20 system_u:system_r:system_dbusd_t:s0
/proc/1/fd/21 system_u:system_r:avahi_t:s0
And the AVC's seem to have dissapeared when a confined app trys to
connect to dbus or avahi.
If you run with this patch and selinux-policy-3.8.8-3.fc14.noarch
You should be able to boot in enforcing mode.
