On 10/1/20 2:38 PM, Linus Torvalds wrote:
On Thu, Oct 1, 2020 at 10:41 AM Josef Bacik <josef@xxxxxxxxxxxxxx> wrote:
Obviously not ideal, but I figured the simpler fix was better for stable, and
then we could work out something better.
I think the attached is the proper fix, and it's not really any more
complicated.
The patch is bigger, but it's pretty obvious: get rid of the
non-specific "pipe_wait()", and replace them with specific versions
that wait for a particular thing.
NOTE! Entirely untested. It seems to build fine for me, and it _looks_
obvious, but I haven't actually rebooted to see if it works at all. I
don't think I have any real splice-heavy test cases.
Mind trying this out on the load that showed problems?
I wrote a simple reproducer, reproduced the problem on an unpatched kernel in
like 20 minutes. I'm loading this up, I'll call it fixed in say a couple of
hours? Thanks,
Josef
