Re: CVS commit: src/sys/ufs/ffs

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Wed, Nov 12, 2008 at 08:25:34PM +0200, Antti Kantee wrote:
> On Wed Nov 12 2008 at 19:08:37 +0100, Joerg Sonnenberger wrote:
> > On Wed, Nov 12, 2008 at 07:38:22PM +0200, Antti Kantee wrote:
> > > > The PR is about the optional reallocation fcntl?
> > > 
> > > The what?
> > 
> > The ancient FFS code had a sysctl doreallocblks. That allowed on a
> > per-inode base to force reallocation of the blocks. I think this was
> > meant to be used for some online defragmentation. The actual code for it
> > was what I ultimately removed.
> 
> Try to decide between sysctl and fcntl ;)

The sysctl controlled whether the fcntl allowed anything.

> As far as I understood it, the point was to reallocate blocks while a
> file wasn't yet flushed to disk.  This allowed better locality for files
> written in increments.  Since vfs_cluster, VOP_REALLOCBLKS() and with your
> commit even the backend code gone, the PR could use a little updating.
> You struck the final blow, so it's your job ;)

I won't promise anything. I agree with the problem and I might work on
it at some point, but not sure when. Basically, fixing the time of
allocation == time of initial write issue is very difficult with the
current FFS code. It won't be done as long as soft updates are in it, at
least.

Joerg