On Thu, 09 Dec 2010 14:17:52 -0500 Trond Myklebust <Trond.Myklebust@xxxxxxxxxx> wrote: > On Thu, 2010-12-09 at 14:01 -0500, Jeff Layton wrote: > > On Tue, 07 Dec 2010 23:19:32 -0500 > > Trond Myklebust <Trond.Myklebust@xxxxxxxxxx> wrote: > > > > > On Tue, 2010-12-07 at 22:45 -0500, Trond Myklebust wrote: > > > > On Tue, 2010-12-07 at 22:17 -0500, Trond Myklebust wrote: > > > > > OK... I think I see why the hang: > > > > > > > > > > I believe that it is basically due to nfs_clear_page_tag_locked() making > > > > > the assumption that if req->wb_page != NULL, then the corresponding > > > > > nfsi->nfs_page_tree lock tag needs to be cleared. > > > > > > > > > > Maybe we can do that differently by just setting a flag to indicate > > > > > whether or not this request is mapped into the radix tree... > > > > > > > > The following patch is completely untested, but should do the trick.... > > > > > > ...and it appears to work correctly for me. > > > > > > > Hi Trond, > > > > I've gotten some preliminary results with the patch that I backported > > to RHEL5: > > > > -----------------------[snip]------------------ > > > > FYI, this version has been running now for about 18 hours on > > the same [host] where I originally observed this problem. A full run takes 48 > > hours, but this is well beyond the normal point of failure so I thought > > I'd share preliminary results. > > > > -----------------------[snip]------------------ > > > > So, it looks good so far. If you like, you can add my. I'll let you > > know once the run is done. > > > > Reviewed-by: Jeff Layton <jlayton@xxxxxxxxxx> > > -- > > To unsubscribe from this list: send the line "unsubscribe linux-nfs" in > > the body of a message to majordomo@xxxxxxxxxxxxxxx > > More majordomo info at http://vger.kernel.org/majordomo-info.html > > Thanks! > > BTW: the reason why the PG_MAPPER is not critical in the RHEL-5 back > port is because the BKL keeps you safe w.r.t. races between the tests in > nfs_flush_incompatible() and the nfs_clear_request call in > nfs_inode_remove_request(). > > Cheers > Trond > RHEL5 also doesn't have nfs_set/clear_page_tag_locked, so that explains why we didn't see the hang there. I still backported the PG_MAPPED change anyway as I think it's cleaner than testing for a NULL wb_page (which rhel5 does in nfs_clear_page_writeback). -- Jeff Layton <jlayton@xxxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
