[Btrfs-devel] btrfs and git-reflog
tytso at MIT.EDU
Fri Jan 25 10:47:34 PST 2008
On Fri, Jan 25, 2008 at 07:07:58PM +0100, Jan Engelhardt wrote:
> There was a debate about whether telldir/seekdir should continue to be
> supported at the kernel level (http://lkml.org/lkml/2007/4/7/107)
That's not a good summary of the thread.
I encourage folks to re-read the old LKML discussion, though.
Basically, the problem with readdir off_t collisions are (1)
telldir/seekdir, and (2) NFS.
It's not really possible to emulate telldir/seekdir in userspace,
unless you want glibc to suck in the *entire* directory (which in the
case of a squid cache directory might potentially be megabtes or
gigbytes of storage) all in one go.
Nuking telldir() and seekdir() support entirely would be politically
sensitive, but given that it's needed for NFS server support, and
that's probably harder to make go away than old programs that used
telldir()/seekdir(), no one has ever really been interested in pushing
forward with a deprecation proposal for telldir()/seekdir() at a POSIX
or SuS standards meeting.
Don't get me wrong, as a filesystem developer I would love being able
to make that problem go away. Unfortunately, we're cursed with the
legacy software and legacy network filesystems, and the latter is
still very much in use today. Life would be a lot easier if NFS got
flushed and replaced with OpenAFS, for example --- but that's not
going to happen any time soon, at least not in this reality. :-)
More information about the Btrfs-devel