From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from bombadil.infradead.org ([198.137.202.9]:53435 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753520AbcJUQuL (ORCPT ); Fri, 21 Oct 2016 12:50:11 -0400 From: Christoph Hellwig Subject: remove the XFS iolock, V2 Date: Fri, 21 Oct 2016 18:50:04 +0200 Message-Id: <1477068606-21936-1-git-send-email-hch@lst.de> Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: linux-xfs@vger.kernel.org Cc: peterz@infradead.org Hi all, this series remove the per-inode iolock in XFS which has historically been used to serialize I/O. But now that the VFS i_mutex has been replaced with a rw_semaphore called i_rwsem there is no need to keep the XFS lock - by relying on the VFS i_rwsem we unify the locking constraints between difference file systems, reduce the XFS inode size and reduce the amount of atomic instruction we do with every read or write call. Note that Dave asked to kill the XFS mrlock first the last time this series was posted - while that seems desirable it's not easily possible due to the way XFS may offload allocator work to a workqueues, and how lockdep doesn't recognize that. As removing mrlocks entirely would be just a cleanup without an affect on locking hierchies or inode size reduction (except for debug builds) I would prefer to get this series to remove the iolock in first, and then see what we can do about the two remaining mrlocks later.