On Saturday, November 12, 2016 9:56:40 PM CET Doug Graham wrote: > As far as I can tell, we run into this problem only when the inode number > of the hardlinked file exceeds 2^31. This is happening on a tmpfs > filesystem of ~48GB where some of the machines are using very large inode > numbers. It still looks like 2.12 does not have this problem so I guess we > need to upgrade.
There's not planned update to 2.12 in RHEL6, but there's backport request which (if everything goes fine) should be in RHEL 6.9: https://bugzilla.redhat.com/show_bug.cgi?id=1155814 Should you have more RHEL related questions please contact Red Hat support channels. Pavel > > On Sat, Nov 12, 2016 at 6:11 PM, Doug Graham <edou...@gmail.com> wrote: > > > Hi, > > > > Our builds are done on a pool of RHEL 6.6 x86_64 machines on which cpio > > 2.10-12 is installed. The builds build a Linux root filesystem like so: > > > > find . | fakeroot cpio -H new -o | xz ... > > > > This works most of the time, but when run on specific machines in the > > pool, all hardlinked regular files within the directory being archived are > > zero length in the archive. The files are still hardlinked but the contents > > are gone. > > > > An md5sum on the /bin/cpio binaries shows that they are different, but > > this must be because of prelinking and address randomization. > > > > I suspect a bug in cpio, but can't rule out a bug in fakeroot or something > > else. I did try building cpio 2.12 from source and using that, and that > > does seem to have made the problem go away, but without knowing what the > > root cause was, I can't really be sure it won't come back. > > > > Any idea what might be causing this? > > > > Thanks, > > Doug > > > > >
