aio: protect reqs_available updates from changes in interrupt handlers
authorBenjamin LaHaise <bcrl@kvack.org>
Mon, 14 Jul 2014 16:49:26 +0000 (12:49 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 28 Jul 2014 15:06:03 +0000 (08:06 -0700)
commit8d378bce5487a81db606c4c8b01305cb4b0fcc97
treeead3cbb7594ab2b7b1a656bd67de357b73a1a0ee
parent2ef68c813d46c975f119f4a8d434dee56677f724
aio: protect reqs_available updates from changes in interrupt handlers

commit 263782c1c95bbddbb022dc092fd89a36bb8d5577 upstream.

As of commit f8567a3845ac05bb28f3c1b478ef752762bd39ef it is now possible to
have put_reqs_available() called from irq context.  While put_reqs_available()
is per cpu, it did not protect itself from interrupts on the same CPU.  This
lead to aio_complete() corrupting the available io requests count when run
under a heavy O_DIRECT workloads as reported by Robert Elliott.  Fix this by
disabling irq updates around the per cpu batch updates of reqs_available.

Many thanks to Robert and folks for testing and tracking this down.

Reported-by: Robert Elliot <Elliott@hp.com>
Tested-by: Robert Elliot <Elliott@hp.com>
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Cc: Jens Axboe <axboe@kernel.dk>, Christoph Hellwig <hch@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/aio.c