Re: [Cluster-devel] [PATCH] gfs2: Fix loop in gfs2_rbm_find (2)

----- Original Message -----

----- Original Message -----

...

The fix from commit 2d29f6b96d8f introduced an unexpected performance regression when allocating blocks. Fix by rewriting the overly complicated wrap-around logic in gfs2_rbm_find in a more reasonable way. Discovered and verified with iozone.

Fixes: 2d29f6b96d8f ("gfs2: Fix loop in gfs2_rbm_find") Cc: stable@vger.kernel.org # v3.13+ Signed-off-by: Andreas Gruenbacher agruenba@redhat.com

(snip)

With this new patch, it will just go back next_bitmap, but there are no more bitmaps. It sets wrap, but since we're at the end, it returns -ENOSPC, when, in fact, there is a free block, no?

On closer inspection, the patch looks okay to me. I still want to test it before we push it.

Bob Peterson