On Mon, Dec 17, 2012 at 06:28:40PM -0500, Chris Mason wrote:

On Mon, Dec 17, 2012 at 04:09:07PM -0700, Zach Brown wrote:

quoted

1) The fundamental fix is to re-use deleted entry positions.  Do we add
another cache to index unlinked positions?  Do we add an unreliable
best-effort walk of the tree looking for holes in the key space?  At the
very least test index_cnt in unlink to get the basically useless
index_cnt--? :)

The index is dense enough that we can search for free spots without too
much pain.  But, more below.

OK.  Want me to take a stab at it?  Is there a similar use somewhere I
should work from?

quoted

2) Regardless of that, we have to deal with existing entry items with
giant keys.  If for no other reason than big jerks making corrupt images
and leaving them on usb keys in Josef's driveway.  Should we drop the
silly INT_MAX setting for 64bit callers and return -EOVERFLOW for 32bit
callers?  (That'd be gross, but not unheard of.  ext4 has grown htree
behaviour that depends on compat detection: see its is_32bit_api()
callers.)

I can make up some fixes but I'd love to hear strong opinions first, if
anyone's got 'em :).

If we go past the 32 bit number we can use the hash offsets in readdir,
and just flag the directory as hashme-in-readdir

Hmm.  That sounds painful given either hash collisions or reconnecting
nfs clients with previous f_pos values in use.

With a dencent entry key allocator it sounds a lot cleaner to me to just
admit that 32bit apps can't see more dirents than their f_pos can
represent.  It's already based on the number of entries rather than the
bytes of entries so it'd be pretty hard to exhaust.

Am I .. not right? :)

- z