Re: [PATCH] tmpfs symlink corrupts mempolicy

[Andrea Arcangeli <andrea@novell.com>]

On Tue, Nov 16, 2004 at 12:26:05AM +0000, Hugh Dickins wrote:
> On Mon, 15 Nov 2004, Andrea Arcangeli wrote:
> > 
> > this patch is completely broken, delete_inode isn't going to be called
> > when the inode is being shrunk. delete_inode is only good for truncate,
> 
> Do you mean "when the inode cache is being shrunk"?
> By "truncate" there you mean "unlink"?

yes.

> > this patch will tend to work until the vm shrink the dcache, then it'll
> > crash, sorry.
> 
> Sorry, you can see I'm having some trouble understanding your reply.
> 
> I think you're forgetting that tmpfs inodes live nowhere but in memory:
> if shrinking the inode cache were to remove tmpfs inodes, it would be a
> considerably more temporary fs than could ever be useful.  There's an
> extra dget that keeps dentry and inode safe from pruning.

I was wrong about it crashing, but it will not crash only for shmfs,
if you want to export that to MAP_SHARED on other fs, you won't be
allowed anymore to depending on this shmfs speciaility of delete_inode
being recalled only during unlink. I don't see how depending on this
subtle detail to free the mpol could ever be an improvement instead of
doing in destroy_inode where it belongs to.