BINDIR/SBINDIR/... variables in Makefiles

BINDIR/SBINDIR/... variables in Makefiles

[Nicolas Iooss]

Hello,

While updating packages for SELinux userspace 2.7-rc2 release, I
noticed there are still some inconsistencies in the Makefile variables
related to bin and sbin directories. A quick search (with find . -name
Makefile -exec cat {} + |grep '^[A-Z_]*BINDIR' |sort -u) shows:

BINDIR=$(PREFIX)/bin
BINDIR ?= $(PREFIX)/bin
BINDIR ?= $(PREFIX)/sbin
SBINDIR ?= $(DESTDIR)/sbin
SBINDIR ?= $(PREFIX)/sbin
USRSBINDIR ?= $(PREFIX)/sbin

In the coming days I will not have time to write and test patches in
order to make these variables more consistent. I am therefore sending
this message in case someone would want to fix these inconsistencies
before the 2.7 release.

By the way, I noticed a few things in the RC that may be worth
mentioning in the final release announcement:
* libsemanage no longer needs ustr.
* libselinux/utils Makefile now uses USRSBINDIR instead of USRBINDIR
(since commit 055d14a99a84 ("libselinux/utils: do not create an empty
/sbin directory")).
* Some packages (libselinux, checkpolicy, selinux-python,
semodule-utils and mcstrans) require LIBSEPOLA to the set to the
absolute path to libsepol.a when building with DESTDIR set.

Cheers,
Nicolas

Re: BINDIR/SBINDIR/... variables in Makefiles

[Stephen Smalley]

On Sat, 2017-06-17 at 15:55 +0200, Nicolas Iooss wrote:
> Hello,
> 
> While updating packages for SELinux userspace 2.7-rc2 release, I
> noticed there are still some inconsistencies in the Makefile
> variables
> related to bin and sbin directories. A quick search (with find .
> -name
> Makefile -exec cat {} + |grep '^[A-Z_]*BINDIR' |sort -u) shows:
> 
> BINDIR=$(PREFIX)/bin
> BINDIR ?= $(PREFIX)/bin
> BINDIR ?= $(PREFIX)/sbin
> SBINDIR ?= $(DESTDIR)/sbin
> SBINDIR ?= $(PREFIX)/sbin
> USRSBINDIR ?= $(PREFIX)/sbin
> 
> In the coming days I will not have time to write and test patches in
> order to make these variables more consistent. I am therefore sending
> this message in case someone would want to fix these inconsistencies
> before the 2.7 release.

I guess the question is do we need to still support distributions that
do not enable usrmerge?

If so, then the difference is meaningful, and changing all SBINDIR
instances to be either $(PREFIX)/sbin or $(DESTDIR)/sbin could create
breakage.  Originally I was thinking we should convert them all to
$(PREFIX)/sbin (and kill the load_policy symlink) and just drop
USRSBINDIR entirely, since the majority of cases are already using that
definition and in the usrmerge case it will correspond to the real
path.  This would however break fixfiles (unless we also update it to
specify /usr/sbin/{setfiles,restorecon}) and the make relabel targets
in the selinux Makefiles (unless we update them all to specify
/usr/sbin/restorecon) on non-usrmerge distributions.  More importantly,
it would prevent running fixfiles, setfiles, and restorecon before /usr
is mounted on non-usrmerge distributions, which could be problematic
for e.g. autorelabel, invocation of restorecon from init scripts, etc.

Converting them all to $(DESTDIR)/sbin doesn't seem beneficial, as that
would end up installing far more utilities into /sbin that aren't
actually needed before /usr is mounted.

So, if non-usrmerge still matters, we probably can't make them
consistent.  If it doesn't matter, then I'd go with the first option
above.

> By the way, I noticed a few things in the RC that may be worth
> mentioning in the final release announcement:
> * libsemanage no longer needs ustr.
> * libselinux/utils Makefile now uses USRSBINDIR instead of USRBINDIR
> (since commit 055d14a99a84 ("libselinux/utils: do not create an empty
> /sbin directory")).
> * Some packages (libselinux, checkpolicy, selinux-python,
> semodule-utils and mcstrans) require LIBSEPOLA to the set to the
> absolute path to libsepol.a when building with DESTDIR set.

Thanks, will add to the release notes.

Re: BINDIR/SBINDIR/... variables in Makefiles

[Jason Zaman]

On Mon, Jun 19, 2017 at 03:50:26PM -0400, Stephen Smalley wrote:
> On Sat, 2017-06-17 at 15:55 +0200, Nicolas Iooss wrote:
> > Hello,
> > 
> > While updating packages for SELinux userspace 2.7-rc2 release, I
> > noticed there are still some inconsistencies in the Makefile
> > variables
> > related to bin and sbin directories. A quick search (with find .
> > -name
> > Makefile -exec cat {} + |grep '^[A-Z_]*BINDIR' |sort -u) shows:
> > 
> > BINDIR=$(PREFIX)/bin
> > BINDIR ?= $(PREFIX)/bin
> > BINDIR ?= $(PREFIX)/sbin
> > SBINDIR ?= $(DESTDIR)/sbin
> > SBINDIR ?= $(PREFIX)/sbin
> > USRSBINDIR ?= $(PREFIX)/sbin
> > 
> > In the coming days I will not have time to write and test patches in
> > order to make these variables more consistent. I am therefore sending
> > this message in case someone would want to fix these inconsistencies
> > before the 2.7 release.
> 
> I guess the question is do we need to still support distributions that
> do not enable usrmerge?

Yes please, Gentoo hasnt done usrmerge officially (yet?). You can of
course do usrmerge on gentoo but its not the default and since we're a
rolling distro even if/when we do switch we'll likely have to support
unmerged for quite a long time since there isnt a complete re-install on
new releases.

Although, there is still inconsistency in the naming which could perhaps
be fixed. one of them up there has BINDIR?=../sbin so should technically
be called SBINDIR i guess.

-- Jason
> 
> If so, then the difference is meaningful, and changing all SBINDIR
> instances to be either $(PREFIX)/sbin or $(DESTDIR)/sbin could create
> breakage.  Originally I was thinking we should convert them all to
> $(PREFIX)/sbin (and kill the load_policy symlink) and just drop
> USRSBINDIR entirely, since the majority of cases are already using that
> definition and in the usrmerge case it will correspond to the real
> path.  This would however break fixfiles (unless we also update it to
> specify /usr/sbin/{setfiles,restorecon}) and the make relabel targets
> in the selinux Makefiles (unless we update them all to specify
> /usr/sbin/restorecon) on non-usrmerge distributions.  More importantly,
> it would prevent running fixfiles, setfiles, and restorecon before /usr
> is mounted on non-usrmerge distributions, which could be problematic
> for e.g. autorelabel, invocation of restorecon from init scripts, etc.
> 
> Converting them all to $(DESTDIR)/sbin doesn't seem beneficial, as that
> would end up installing far more utilities into /sbin that aren't
> actually needed before /usr is mounted.
> 
> So, if non-usrmerge still matters, we probably can't make them
> consistent.  If it doesn't matter, then I'd go with the first option
> above.
> 
> > By the way, I noticed a few things in the RC that may be worth
> > mentioning in the final release announcement:
> > * libsemanage no longer needs ustr.
> > * libselinux/utils Makefile now uses USRSBINDIR instead of USRBINDIR
> > (since commit 055d14a99a84 ("libselinux/utils: do not create an empty
> > /sbin directory")).
> > * Some packages (libselinux, checkpolicy, selinux-python,
> > semodule-utils and mcstrans) require LIBSEPOLA to the set to the
> > absolute path to libsepol.a when building with DESTDIR set.
> 
> Thanks, will add to the release notes.
>