From mboxrd@z Thu Jan 1 00:00:00 1970 From: Steven Whitehouse Date: Tue, 05 Dec 2006 14:36:16 +0000 Subject: [Cluster-devel] Re: [GFS2] Don't flush everything on fdatasync [70/70] In-Reply-To: <20061201110927.ec6ee073.akpm@osdl.org> References: <1164889448.3752.449.camel@quoit.chygwyn.com> <20061130230158.174e995c.akpm@osdl.org> <1164970738.3752.508.camel@quoit.chygwyn.com> <20061201110927.ec6ee073.akpm@osdl.org> Message-ID: <1165329376.3752.669.camel@quoit.chygwyn.com> List-Id: To: cluster-devel.redhat.com MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Hi, On Fri, 2006-12-01 at 11:09 -0800, Andrew Morton wrote: > On Fri, 01 Dec 2006 10:58:58 +0000 > Steven Whitehouse wrote: > > > On Thu, 2006-11-30 at 23:01 -0800, Andrew Morton wrote: > > > On Thu, 30 Nov 2006 12:24:08 +0000 > > > Steven Whitehouse wrote: > > > > > > > static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync) > > > > { > > > > - struct gfs2_inode *ip = GFS2_I(dentry->d_inode); > > > > + struct inode *inode = dentry->d_inode; > > > > + int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); > > > > + int ret = 0; > > > > + struct writeback_control wbc = { > > > > + .sync_mode = WB_SYNC_ALL, > > > > + .nr_to_write = 0, > > > > + }; > > > > + > > > > + if (gfs2_is_jdata(GFS2_I(inode))) { > > > > + gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl); > > > > + return 0; > > > > + } > > > > > > > > - gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl); > > > > + if (sync_state != 0) { > > > > + if (!datasync) > > > > + ret = sync_inode(inode, &wbc); > > > > > > filemap_fdatawrite() would be simpler. > > > > I was taking my cue here from ext3 which does something similar. The > > filemap_fdatawrite() is done by the VFS before this is called with a > > filemap_fdatawait() afterwards. This was intended to flush the metadata > > via (eventually) ->write_inode() although I guess I should be calling > > write_inode_now() instead? > > oh I see, you're jsut trying to write the inode itself, not the pages. > > write_inode_now() will write the pages, which you seem to not want to do. > Whatever. The APIs here are a bit awkward. Well its not so much that we want to avoid it, but since the VFS has already done that, it should be pretty much a no-op aside from the waiting which will happen within the call (so that presumably the VFS's wait after the call will be more or less a no-op). I notice that fs/sync.c:file_fsync() uses write_inode_now(), so maybe it is a better choice as it at least means that I don't have to fill in a struct write_back control for a mapping that I don't really want to write again, Steve.