svn commit: r194040 - stable/7/lib/libc/gen

[John Baldwin]

On Friday 12 June 2009 9:34:18 am Jilles Tjoelker wrote:
> On Fri, Jun 12, 2009 at 08:51:32AM -0400, John Baldwin wrote:
> > On Friday 12 June 2009 6:13:08 am Dmitry Morozovsky wrote:
> > > On Thu, 11 Jun 2009, John Baldwin wrote:
> > > 
> > > JB> Author: jhb
> > > JB> Date: Thu Jun 11 21:37:23 2009
> > > JB> New Revision: 194040
> > > JB> URL: http://svn.freebsd.org/changeset/base/194040
> > > JB> 
> > > JB> Log:
> > > JB>   If the running kernel has support for shm_open() and shm_unlink() as
> > > JB>   system calls (i.e. 8.0+), then invoke the system calls instead of using
> > > JB>   open/fcntl/unlink.  This is a direct commit instead of an MFC since libc
> > > JB>   in 8.0 always uses the system calls directly.
> > > 
> > > So it could help running RELENG_7 jails under 8 host? 
> 
> > In that case it probably doesn't matter either way.  The goal is that if you
> > have an 8.x binary and a 7.x binary that are both using shm_open() with a
> > path to open a SHM region for IPC, they will both access the same object.
> 
> > If you have all 8.x binaries or all < 8.x binaries in the group that shares
> > access to a named SHM region then that case would work regardless.
> 
> where "< 8.x binaries" refers to those using 7.x's libc.so.7.
> Normally, a 7.x binary running on 8.x will link to 8.x's libc.so.7 and
> therefore use the new SHM objects even without this change.

True.  It is more relevant to libc.so.[46] which I plan to update next week
when the MFC after timer expires, and that is the true target audience I have
in mind (many people still use 4.x and 6.x binaries on newer kernels).

-- 
John Baldwin