Re: oops when calling dm_set_disp()

[Damian Hazen <dhazen@xxxxxxx>]

Damian Hazen wrote:
> Hi -
>
> I'm using XFS/DMAPI with a 2.6.10 kernel pulled from the CVS repository 
> a couple of days ago.  I'm regularly getting an oops when calling 
> dm_set_disp() after receiving a mount event but before responding.  This 
> has worked well for us in the past.  Any insights into what's changed?
> Thanks,
> -Damian
Hi -

It looks like the vfs superblock structure is not being filled in
completely until after the mount event is responded to.  If 
dm_set_disp() is called before responding, it causes an oops in 
dm_handle_to_ip() when it attempts to get the dentry here:

        /* Now that the mutex is released, wait until we have access to the
           inode.
        */

        sb = fsrp->fr_sb;
        if (fidp->fid_len == 0) {       /* filesystem handle */
=>                ip = sb->s_root->d_inode;
                igrab(ip);


If you wait and call dm_set_disp() after responding, there's still a 
race with linvfs_fill_super() to get s_root set.  Would it be possible 
to move up setting s_root to somewhere before dm_send_mount_event() is 
called?  It's nice to be able to set disposition and event lists before 
turning the mount event loose to prevent anyone sneaking into the 
filesystem before we're ready.
Thanks,
-Damian