On Thu, Jan 17, 2019 at 4:51 PM Linus Torvalds
[off-list ref] wrote:

On Thu, Jan 17, 2019 at 9:37 AM Matthew Wilcox [off-list ref] wrote:

quoted

Your patch 3/3 just removes the test.  Am I right in thinking that it
doesn't need to be *moved* because the existing test after !PageUptodate
catches it?

That's the _hope_.

That's the simplest patch I can come up with as a potential solution.
But it's possible that there's some nasty performance regression
because somebody really relies on not even triggering read-ahead, and
we might need to do some totally different thing.

Oh, and somebody should probably check that there isn't some simple
way to just avoid that readahead code entirely.

In particular, right now we skip readahead for at least these cases:

        /* no read-ahead */
        if (!ra->ra_pages)
                return;

        if (blk_cgroup_congested())
                return;

and I don't think we need to worry about the cgroup congestion case -
if the attack has to also congest its cgroup with IO, I think they
have bigger problems.

And I think 'ra_pages' can be zero only in the presence of IO errors,
but I might be wrong. It would be good if somebody double-checks that.

               Linus