Revert Fix zpl_writepage() deadlock
authorRichard Yao <ryao@cs.stonybrook.edu>
Mon, 7 May 2012 18:14:45 +0000 (14:14 -0400)
committerBrian Behlendorf <behlendorf1@llnl.gov>
Mon, 27 Aug 2012 19:01:37 +0000 (12:01 -0700)
commit62c4165a1bbfb7d68f8ebf93d32a6fc8ea4d4e33
tree7b444d4698f60cdcaf184b645eda454cf5ad97ef
parentb876dac776afc8ea2c598eac53b9903de01c6172
Revert Fix zpl_writepage() deadlock

The commit, cfc9a5c88f91f7b4d606fce89505e1f404691ea5, to fix deadlocks
in zpl_writepage() relied on PF_MEMALLOC.   That had the effect of
disabling the direct reclaim path on all allocations originating from
calls to this function, but it failed to address the actual cause of
those deadlocks.  This led to the same deadlocks being observed with
swap on zvols, but not with swap on the loop device, which exercises
this code.

The use of PF_MEMALLOC also had the side effect of permitting
allocations to be made from ZONE_DMA in instances that did not require
it.  This contributes to the possibility of panics caused by depletion
of pages from ZONE_DMA.

As such, we revert this patch in favor of a proper fix for both issues.

Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #726
module/zfs/zpl_file.c