On Wed, 2 Jul 2003, Nathan Scott wrote:
> Looks alot like a sync/iget interaction bug we were seeing recently -
> Steve fixed this a couple of weeks ago, the change was:
>
i was doing a lot of writes (40mb/s, probably as much writes as this
stripe of 2 serial ATA's raid5's can take), and i have big write
caches, i think (logbufs=8,logbsize=32768 as recommended in faq).
i'll try the patch out, thanks.
> --- /usr/tmp/TmpDir.2068614-0/linux/fs/xfs/xfs_inode.c_1.376 Wed Jul 2
> 08:51:27 2003
> +++ /usr/tmp/TmpDir.2068614-0/linux/fs/xfs/xfs_inode.c_1.377 Wed Jul 2
> 08:51:27 2003
> @@ -2629,7 +2629,8 @@
> if (vp) {
> struct inode *inode = LINVFS_GET_IP(vp);
>
> - mark_inode_dirty_sync(inode);
> + if (!(inode->i_state & I_NEW))
> + mark_inode_dirty_sync(inode);
> }
>
> wake_up(&ip->i_ipin_wait);
>
>
> cheers.
>
>
|