possible race in end_buffer_io_async()

From: tpepper@vato.org
Date: Fri Sep 21 2001 - 17:46:42 EST


I've got a setup here that is triggering the BUG() in UnlockPage() in
end_buffer_io_async(). Looking at the code in that function we noticed that:

When the put_bh() macro calls got added to fs/buffer.c's
end_buffer_io_async() they look like they got in slightly inconsistently.
If the code branches to still_busy it does:

        put_bh(bh);
        spin_unlock_irqrestore(&page_uptodate_lock, flags);

If the branch is not taken it does:

        spin_unlock_irqrestore(&page_uptodate_lock, flags);
        put_bh(bh);

I don't know which is correct...There's either a possible race if the put_bh()
should be inside the lock (although since it's decrementing bh_count I doubt
it's what would be causing my attempt to unlock an unlocked page) or else
there are a few instructions that can be moved outside the lock.

I've been hunting around on the web and saw some discussion in July
about async I/O races between Andrea and Linus. Are these resolved?
I'm running 2.4.9 on the machine in question (w/ and w/o kgdb) trying
to figure out what's going on. I'm going to try the latest pre patches
this weekend since the 2.4.10-pre's include a big merge against Andrea.

Tim
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sun Sep 23 2001 - 21:00:46 EST