On Sun, Jun 19, 2011 at 11:56:46PM +0800, Christoph Hellwig wrote:
> On Fri, Jun 10, 2011 at 11:21:49AM +0800, Wu Fengguang wrote:
> > > Can you check if that helps with your livelock test case? (And yes, I
> > > should go and try it myself. Will do as soon as I'll get a bit more
> > > time).
> >
> > My pleasure. It's easier to compare results on the same test bed and
> > I've made it very convenient to test patches :)
>
> Here's one that should be even better, although in it's current form it
> break a subtile corner case of sync semantics, so it'll need more work
> if I go down that way:
The sync(1)s are much faster now, in most invocations it's in the same
level as ext4 :)
root@fat /home/wfg# ./sync-livelock.sh
sync time: 3
sync time: 5
sync time: 5
sync time: 6
sync time: 15
sync time: 5
sync time: 5
sync time: 5
sync time: 4
sync time: 10
sync time: 4
sync time: 4
Thanks,
Fengguang
> Index: xfs/fs/xfs/linux-2.6/xfs_sync.c
> ===================================================================
> --- xfs.orig/fs/xfs/linux-2.6/xfs_sync.c 2011-06-17 14:16:18.442399481 +0200
> +++ xfs/fs/xfs/linux-2.6/xfs_sync.c 2011-06-18 17:55:44.864025123 +0200
> @@ -359,14 +359,16 @@ xfs_quiesce_data(
> {
> int error, error2 = 0;
>
> - /* push non-blocking */
> - xfs_sync_data(mp, 0);
> xfs_qm_sync(mp, SYNC_TRYLOCK);
> -
> - /* push and block till complete */
> - xfs_sync_data(mp, SYNC_WAIT);
> xfs_qm_sync(mp, SYNC_WAIT);
>
> + /* flush all pending size updates and unwritten extent conversions */
> + flush_workqueue(xfsconvertd_workqueue);
> + flush_workqueue(xfsdatad_workqueue);
> +
> + /* force out the newly dirtied log buffers */
> + xfs_log_force(mp, XFS_LOG_SYNC);
> +
> /* write superblock and hoover up shutdown errors */
> error = xfs_sync_fsdata(mp);
>
> Index: xfs/fs/xfs/linux-2.6/xfs_super.c
> ===================================================================
> --- xfs.orig/fs/xfs/linux-2.6/xfs_super.c 2011-06-18 17:51:05.660705925 +0200
> +++ xfs/fs/xfs/linux-2.6/xfs_super.c 2011-06-18 17:52:50.107367305 +0200
> @@ -929,45 +929,12 @@ xfs_fs_write_inode(
> * ->sync_fs call do that for thus, which reduces the number
> * of synchronous log foces dramatically.
> */
> - xfs_ioend_wait(ip);
> xfs_ilock(ip, XFS_ILOCK_SHARED);
> - if (ip->i_update_core) {
> + if (ip->i_update_core)
> error = xfs_log_inode(ip);
> - if (error)
> - goto out_unlock;
> - }
> - } else {
> - /*
> - * We make this non-blocking if the inode is contended, return
> - * EAGAIN to indicate to the caller that they did not succeed.
> - * This prevents the flush path from blocking on inodes inside
> - * another operation right now, they get caught later by
> - * xfs_sync.
> - */
> - if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
> - goto out;
> -
> - if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip))
> - goto out_unlock;
> -
> - /*
> - * Now we have the flush lock and the inode is not pinned, we
> - * can check if the inode is really clean as we know that
> - * there are no pending transaction completions, it is not
> - * waiting on the delayed write queue and there is no IO in
> - * progress.
> - */
> - if (xfs_inode_clean(ip)) {
> - xfs_ifunlock(ip);
> - error = 0;
> - goto out_unlock;
> - }
> - error = xfs_iflush(ip, SYNC_TRYLOCK);
> + xfs_iunlock(ip, XFS_ILOCK_SHARED);
> }
>
> - out_unlock:
> - xfs_iunlock(ip, XFS_ILOCK_SHARED);
> - out:
> /*
> * if we failed to write out the inode then mark
> * it dirty again so we'll try again later.
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
