On Sat, 2021-05-15 at 08:02 +0100, Christoph Hellwig wrote: > On Fri, May 14, 2021 at 03:46:57PM +0000, Trond Myklebust wrote: > > Why leave the commit_metadata() call under the lock? If you're > > concerned about latency, then it makes more sense to call > > fh_unlock() > > before flushing those metadata updates to disk. > > Also I'm not sure why the extra inode reference is needed. What > speaks > against just moving the dput out of the locked section? Isn't the inode reference taken just in order to ensure that the call to iput_final() (and in particular the call to truncate_inode_pages_final()) is performed outside the lock? The dput() is presumably usually not particularly expensive, since the dentry is just a completely ordinary negative dentry at this point. -- Trond Myklebust Linux NFS client maintainer, Hammerspace trond.myklebust@xxxxxxxxxxxxxxx
