On Mon 09-05-11 09:55:16, Ted Tso wrote: > On Mon, May 09, 2011 at 01:30:52PM +0200, Jan Kara wrote: > > Ah, I see. But then you just reintroduced the bug I was trying to fix. So > > either do_split() has to do the marking of buffer dirty, or we have to do > > it before calllig do_split(), or do_split() has to be changed and not > > release passed buffer (and the two callers have to do it - which they seem > > to do anyway). I don't mind either way but your fix is wrong. > > I think it's OK. We do call ext4_handle_dirty_metadata on frame->bh, > which deals with the original version of bh. And the cases where > do_split() sets bh to NULL is either (a) a journal error, in which > case we will have already aborted the journal, or an I/O error while > reading in the block, so bh won't have gotten modified yet. > > Is there a case that you're worried about that I'm missing? Yes. ext4_append() can return ENOSPC and passed bh will get set to NULL without being marked dirty. Note that we need to call ext4_handle_dirty_metadata() on the passed bh as well specifically in the make_indexed_dir() case because there we move contents of the first block (in frame->bh) to the second block (passed bh) and create indexed tree root in the first block. Then we call do_split() to further split the second block... Honza -- Jan Kara <jack@xxxxxxx> SUSE Labs, CR -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
