From mboxrd@z Thu Jan 1 00:00:00 1970 From: Fabio Massimo Di Nitto Date: Thu Mar 29 02:04:50 2007 Subject: [Ocfs2-devel] [PATCH 0/3] Properly protect the extent map In-Reply-To: <20070329005041.518030000@oracle.com> References: <20070329005041.518030000@oracle.com> Message-ID: <460B7B00.30004@ubuntu.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ocfs2-devel@oss.oracle.com Joel Becker wrote: > The extent map code was seeing an error trip when inserting records. > There is a sanity check ensuring that the new record is within the known > size of the file - as known by the extent map. The problem was that the > "known" size was smaller than expected, returning an error to the caller. > > This is a lot easier to see in local mounts. It has only been rarely seen > in clusters. > > We have three "independant" problems. > > 1) We don't take ip_alloc_sem when touching directory allocations. > 2) We don't lock em_clusters properly. > 3) Local mounts race to ocfs2_extent_map_trunc() in ocfs2_meta_lock_update(). > > The patches that follow are against mainline. They have been tested > against 1.2 with local mounts. > > Patches (1) and (2) will be ported to the ocfs2 1.2 branch directly, > and all three fixes will be pushed to mainline. > > Joel add a Signed-off-by: Fabio M. Di Nitto on all 3 of them. They fix the problem as i reported it, but it looks like they introduce a performance hit. Somebody other than me should test it too. Fabio -- I'm going to make him an offer he can't refuse.