Re: [PATCH] security: move LSM xattrnames to xattr.h

[Steve Grubb <sgrubb@redhat.com>]

On Tuesday, October 12, 2010 09:40:34 am Mimi Zohar wrote:
> On Tue, 2010-10-12 at 09:19 -0400, Steve Grubb wrote:
> > On Tuesday, October 12, 2010 09:06:09 am Mimi Zohar wrote:
> > > On Tue, 2010-10-12 at 14:14 +0300, Ozan Çağlayan wrote:
> > > > Cuma 02 Temmuz 2010 günü (saat 03:16:01) James Morris şunları yazmıştı:
> > > > > On Thu, 1 Jul 2010, Mimi Zohar wrote:
> > > > > > Make the security extended attributes names global. Updated to
> > > > > > move the remaining Smack xattrs.
> > > > > > 
> > > > > > Signed-off-by: Mimi Zohar <zohar@us.ibm.com>
> > > > > > Acked-by: Serge Hallyn <serue@us.ibm.com>
> > > > 
> > > > This drops
> > > > 
> > > > #define XATTR_CAPS_SUFFIX "capability"
> > > > #define XATTR_NAME_CAPS XATTR_SECURITY_PREFIX XATTR_CAPS_SUFFIX
> > > > 
> > > > definitions from capability.h and puts them in xattr.h's #ifdef
> > > > __KERNEL__ section making them invisible to userspace like libcap-ng
> > > > causing build failures.
> > > > 
> > > > Am I wrong?
> > > 
> > > You're correct.  It's the same reason that cap-ng.c has to define
> > > 'security'.
> > > 
> > > #ifdef VFS_CAP_U32
> > > 
> > >  #include <attr/xattr.h>
> > >  #define XATTR_SECURITY_PREFIX "security."
> > > 
> > > Am cc'ing Steve.
> > 
> > So does this mean I need to provide more definitions for libcap-ng to
> > work with future kernels or are you asking my opinion? My opinion is
> > that if user space needs it to work correctly, please let it be
> > available so I don't have to make my own define which may be inaccurate
> > one day.
> > 
> > Thanks,
> > -Steve
> 
> Before making any changes to the kernel xattr.h, I want to understand
> the reason for two xattr.h files, one in /usr/include/linux/ and the
> other in /usr/include/xattr/.  /usr/include/linux/xattr.h contains those
> elements not defined as __kernel__, while /usr/include/xattr/xattr.h
> contains that and other definitions.  Will changing the kernel xattr.h
> version change both?
> 
> As long as we're making this change, should 'security' also be defined
> outside of the __kernel__ definitions?

I guess no one fixed this before 2.6.36 was finalized. Removing the define has broke user 
space compilation for anything that works on file based capabilities. I can define it 
myself, but if the kernel folks ever change the string, then we have more than just a 
compile problem, we have runtime problems because I can no longer use the correct 
string.

So, what was the gain for breaking user space?

-Steve