Re: Error build -STABLE ports on bento

From: Joe Marcus Clarke <marcus_at_marcuscom.com>
Date: Fri, 16 Jan 2004 03:48:10 -0500
On Fri, 2004-01-16 at 03:43, Bruce Evans wrote:
> On Fri, 16 Jan 2004, Joe Marcus Clarke wrote:
> 
> > On Fri, 2004-01-16 at 02:44, Joe Marcus Clarke wrote:
> > > I'm trying to get the latest 4-exp build done, and I'm seeing a strange
> > > problem.  The -STABLE chroots (running 490101) built under FreeBSD
> > > -CURRENT (501114) refuse to copy 0 length files.  Whenever cp encounters
> > > a 0 length file, it fails with EINVAL.  For example:
> > >
> > > cp: /tmp/a/ports/sysutils/usermin/work/usermin-1.051/install-type:
> > > Invalid argument
> > >
> > > This is not happening with the same -STABLE version built and run under
> > > FreeBSD -CURRENT 501108.  Non-zero length files are copied just fine.
> > > Is this something I should expect to be fixed if the cluster is upgraded
> > > to today's -CURRENT?  Is there a simple workaround that can be used in
> > > lieu of upgrading at this time?  Thanks.
> >
> > Hmmm...sorry to follow up to myself, but I just tested the same -STABLE
> > under a -CURRENT 502101 machine, and the behavior is the same.  cp
> > returns EINVAL when trying to copy zero-length files.
> 
> munmap(2) was recently perfectly broken to POSIX spec so that it doesn't
> work on zero-length files.  cp and probably other utilities in RELENG_4
> and older versions of FreeBSD are missing the fix for this, so they
> can't work right under -current.
> 
> % RCS file: /home/ncvs/src/sys/vm/vm_mmap.c,v
> % Working file: vm_mmap.c
> % head: 1.177
> % ...
> % ----------------------------
> % revision 1.171
> % date: 2003/11/10 01:37:40;  author: alc;  state: Exp;  lines: +8 -6
> %  - The Open Group Base Specifications Issue 6 specifies that an munmap(2)
> %    must return EINVAL if size is zero.  Submitted by: tegge
> %  - In order to avoid a race condition in multithreaded applications, the
> %    check and removal operations by munmap(2) must be in the same critical
> %    section.  To accomodate this, vm_map_check_protection() is modified to
> %    require its caller to obtain at least a read lock on the map.
> % ----------------------------
> 
> % RCS file: /home/ncvs/src/bin/cp/utils.c,v
> % Working file: utils.c
> % head: 1.42
> % ...
> % ----------------------------
> % revision 1.42
> % date: 2003/11/13 05:26:55;  author: alc;  state: Exp;  lines: +2 -1
> % Don't mmap(2) and munmap(2) zero-length files.
> %
> % Submitted by:	Wiktor Niesiobedzki <bsd_at_w.evip.pl>
> % ----------------------------
> 
> Note that only munmap(2) is specified to be (or implemented to be) broken.
> You can mmap() zero-length files, but you can't munmap() them.

Thanks for the pointer!  Are there plans to MFC this?  At least I can
hack our local tree in the meantime.

Joe

> 
> Bruce
-- 
PGP Key : http://www.marcuscom.com/pgp.asc

Received on Thu Jan 15 2004 - 23:48:33 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:38 UTC