Re: [Patch 5/8] 2.4.20-pre4/ext3: Fix O_SYNC for non-data-journaled modes.

[Christoph Hellwig <hch@infradead.org>]

On Wed, Aug 28, 2002 at 04:45:20PM +0100, Stephen Tweedie wrote:
> ext3 has its own code which marks buffers dirty, in addition to the setting
> done by the core generic_commit_write code.  However, the core code does
> 
>                         if (!atomic_set_buffer_dirty(bh)) {
>                                 __mark_dirty(bh);
>                                 buffer_insert_inode_queue(bh, inode);
> 
> so if ext3 marks the buffer dirty itself, the core fails to put it on the
> per-inode list of dirty buffers.  Hence, fsync_inode_buffers() misses it.
> 
> The fix is to let ext3 put the buffer on the inode queue manually when
> walking the page's buffer lists in its page write code.

This patch conflicts with the b_inode as bool patch you recently ACKed..