Re: [PATCH] Fix vmtruncate race and distributed filesystem race

[Andrew Morton]

Dave McCracken <[e-mail]> wrote:
>
> 
> --On Thursday, June 12, 2003 14:00:14 -0700 Andrew Morton <[e-mail]>
> wrote:
> 
> > And this does require that ->nopage be entered with page_table_lock held,
> > and that it drop it.
> 
> I think that's a worse layer violation than referencing inode in
> do_no_page.  We shouldn't require that the filesystem layer mess with the
> page_table_lock.

Well it is not "worse".  Futzing with i_sem in do_no_page() is pretty gross.
You could add vm_ops->prevalidate() or something if it worries you.

btw, it should be synchronising with
file->f_dentry->d_inode->i_mapping->host->i_sem, not
file->f_dentry->d_inode->i_sem.  do_truncate() also seems to be taking the
(potentially) wrong semaphore.