> However most systems use open(file, O_CREAT|O_WRONLY|O_EXCL, mode) to
> creat a lockfile. And NFS doesn't know about O_EXCL, so this still doesn't
> work. I created a patch for the Linux NFS client code that guarantees
> an atomic creat by first creating a temporary file, and then linking it
> to its destination.
Very elegant. It can't be relied on in general purpose UNIX code, for
portability reasons, but it is a pleasent improvement for Linux specific
code. And even if it isn't perfect, it's a great deal better then it was,
and will help naive code immensely.
My only problem with the patch is that it does not precisely match
traditional NFS linking. The different may be irrelevant, but I'm not
sure. Specifically, you say (pardon the perl code):
$tmp = "$filename.foobar";
$err = nfslink($tmp,$filename);
nfsunlink($tmp);
if($err == 0) {
#success
}
However, the traditional code that I'm familiar with goes more like this:
$tmp = "$filename.foobar";
retry:
nfslink($tmp,$filename);
if(nfsstat($tmp)->{links} == 2) {
nfsunlink($tmp);
#success
} else {
nfsunlink($tmp);
$retries++;
goto retry if $retries<5;
}
Apart from the retrying, the important difference (I think) is the stat
call to verify that the link was actually successful, instead of relying
on the return code.
-- Kenneth Albanowski (kjahds@kjahds.com, CIS: 70705,126)