On Tue, Nov 24, 2020 at 08:28:16AM -0800, Hugh Dickins wrote:
> On Tue, 24 Nov 2020, Matthew Wilcox wrote:
> > On Mon, Nov 23, 2020 at 08:07:24PM -0800, Hugh Dickins wrote:
> > >
> > > Then on crashing a second time, realized there's a stronger reason against
> > > that approach. If my testing just occasionally crashes on that check,
> > > when the page is reused for part of a compound page, wouldn't it be much
> > > more common for the page to get reused as an order-0 page before reaching
> > > wake_up_page()? And on rare occasions, might that reused page already be
> > > marked PageWriteback by its new user, and already be waited upon? What
> > > would that look like?
> > >
> > > It would look like BUG_ON(PageWriteback) after wait_on_page_writeback()
> > > in write_cache_pages() (though I have never seen that crash myself).
> >
> > I don't think this is it. write_cache_pages() holds a reference to the
> > page -- indeed, it holds the page lock! So this particular race cannot
> > cause the page to get recycled. I still have no good ideas what this
> > is :-(
>
> It is confusing. I tried to explain that in the final paragraph:
>
> > > Was there a chance of missed wakeups before, since a page freed before
> > > reaching wake_up_page() would have PageWaiters cleared? I think not,
> > > because each waiter does hold a reference on the page: this bug comes
> > > not from real waiters, but from when PageWaiters is a false positive.
>
> but got lost in between the original end_page_writeback() and the patched
> version when writing that last part - false positive PageWaiters are not
> relevant. I'll try rewording that in the simpler version, following.
>
> The BUG_ON(PageWriteback) would occur when the old use of the page, the
> one we do TestClearPageWriteback on, had *no* waiters, so no additional
> page reference beyond the page cache (and whoever racily frees it). The
> reuse of the page definitely has a waiter holding a reference, as you
> point out, and PageWriteback still set; but our belated wake_up_page()
> has woken it to hit the BUG_ON.
I ... think I see. Let me try to write it out:
page is allocated, added to page cache, dirtied, writeback starts,
--- thread A ---
filesystem calls end_page_writeback()
test_clear_page_writeback()
--- context switch to thread B ---
truncate_inode_pages_range() finds the page, it doesn't have writeback set,
we delete it from the page cache. Page gets reallocated, dirtied, writeback
starts again. Then we call write_cache_pages(), see
PageWriteback() set, call wait_on_page_writeback()
--- context switch back to thread A ---
wake_up_page(page, PG_writeback);
... thread B is woken, but because the wakeup was for the old use of
the page, PageWriteback is still set.
Devious.
We could fix this by turning that 'if' into a 'while' in
write_cache_pages(). Just accept that spurious wakeups can happen
and they're harmless. We do need to remove that check of PageWaiters
in wake_up_page() -- as you say, we shouldn't be checking that after
dropping the reference. I had patches to do that ..
https://lore.kernel.org/linux-mm/20200416220130.13343-1-wi...@infradead.org/
specifically:
https://lore.kernel.org/linux-mm/20200416220130.13343-11-wi...@infradead.org/
