Re: Regression in next with ext4 oops
From: Theodore Ts'o
Date: Tue Oct 04 2016 - 15:08:20 EST
On Tue, Oct 04, 2016 at 03:59:15PM +0100, Al Viro wrote:
> Jan is wrong - we do have per-struct-file serialization for getdents()
> et.al. It might be a race between getdents() on *different* struct
> file for the same directory, but ->private_data is not a problem.
So the rb_insert_color() OOPS seems to indicate that the rbtree has
gotten corrupted, and it hangs off of private_data.
And I've just checked the ext4.git tree and I can't see any changes in
the dev branch (which I was just about to push to Linus) that would
impact the readdir path for non-encrypted directories. So if it's a
race on ->private_data I'm not sure what else might be going wrong.
Hmm... Tony, what version was the last good version where it wouldn't
reproduce for you? v4.7? Does it reproduce for you on v4.8?
Also, would you mind seeing if you can reproduce it on the dev
branch of the ext4 git tree?