Re: [PATCH v2 0/10] fs: Introduce FALLOC_FL_INSERT_RANGE for fallocate

[Theodore Ts'o <tytso@xxxxxxx>]

On Mon, Jun 02, 2014 at 03:06:13PM +0200, LukÃÅ Czerner wrote:
> > > So what will happen when there is not enough space when "inserting a
> > > range" ? And how should user proceed from there ?
> > If insert range fails with an ENOSPC error, user could use collapse
> > range on the same range to remove the hole.
> > And after freeing more space, he can again try inserting range.
> > Ofcourse, this type of guidance should be properly documented in
> > manpage. When updating fallocate(2) manpage, I will keep  in mind to
> > describe ENOSPC handling.
> 
> Why collapse ? The hole is already there right ? Why not just use
> fallocate to allocate the space for the hole. And that's my point
> actually. Why not do it this way in the first place, because this is
> really counterintuitive.

It's worse than that.  It's possible that the reason why you got the
ENOSPC warning was because the operation to move the extents down
required allocating a block, and it was *that* block allocation which
failed.  So it's not deterministic whether or not the file's extent
mappings were modified after a ENOSPC error, and so it's not clear
whether or not a collapse_range function will undo the range that had
been inserted --- or whether it ends up deleting existing data blocks.

In generally, you really want system calls to have all-or-nothing
effects, where if the system call returns an error, the state of the
file has not been changed.  And for that reason, I agree with LukÃÅ
that it is really a good idea to decouple moving the blocks down, and
allocating space --- and to make sure that if there is any failure
while inserting the range, the state of the file is not modified at all.

Cheers,

                                        - Ted