Re: deadlocks caused by ext3/reiser dirty_inode calls during do_mmap_pgoff

[Andrew Morton <akpm@osdl.org>]

Chris Mason <mason@suse.com> wrote:
>
> On Mon, 2004-06-21 at 20:13, Andrew Morton wrote:
> 
> > It's super-improbable because we fault the source page in by hand in
> > generic_file_aio_write_nolock() via fault_in_pages_readable().  Of course,
> > that prefaulting isn't 100% reliable either, because the VM can come in and
> > steal the page (or at least unmap its pte) before we get to doing the copy.
> > 
> > I think we can fix both problems by changing filemap_copy_from_user() and
> > filemap_copy_from_user_iovec() to not fall back to kmap() - just fail the
> > copy in some way if the atomic copy failed.  Then, in
> > generic_file_aio_write_nolock(), do a zero-length ->commit_write(),
> > put_page(), then go back and retry the whole thing, starting with
> > fault_in_pages_readable().
> 
> Oh, just realized this will break our data=ordered setup.  prepare_write
> is going to do things like allocating blocks in the file etc etc.  If we
> close the transaction via commit_write(0 size) and crash, we'll leak.

OK, then just zero the pagecache page between `from' and `to' and call
->commit_write() with unmodified `from' and `to'?

> We might need to do prefault + pin on the user pages.

That's always been the correct way to fix these problems, but it involves a
pagetable walk, which requires page_table_lock.  Not a popular thing to be
doing on the write() fastpath.