Re: [PATCH] io_uring/io-wq: don't clear PF_IO_WORKER on exit

[Jens Axboe]

On 6/13/23 6:54?PM, Zorro Lang wrote:
> On Mon, Jun 12, 2023 at 12:11:57PM -0600, Jens Axboe wrote:
>> A recent commit gated the core dumping task exit logic on current->flags
>> remaining consistent in terms of PF_{IO,USER}_WORKER at task exit time.
>> This exposed a problem with the io-wq handling of that, which explicitly
>> clears PF_IO_WORKER before calling do_exit().
>>
>> The reasons for this manual clear of PF_IO_WORKER is historical, where
>> io-wq used to potentially trigger a sleep on exit. As the io-wq thread
>> is exiting, it should not participate any further accounting. But these
>> days we don't need to rely on current->flags anymore, so we can safely
>> remove the PF_IO_WORKER clearing.
>>
>> Reported-by: Zorro Lang <zlang@xxxxxxxxxx>
>> Reported-by: Dave Chinner <david@xxxxxxxxxxxxx>
>> Reported-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
>> Link: https://lore.kernel.org/all/ZIZSPyzReZkGBEFy@xxxxxxxxxxxxxxxxxxx/
>> Fixes: f9010dbdce91 ("fork, vhost: Use CLONE_THREAD to fix freezer/ps regression")
>> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
>>
>> ---
> 
> Hi,
> 
> This patch fix the issue I reported. The bug can be reproduced on v6.4-rc6,
> then test passed on v6.4-rc6 with this patch.
> 
> But I found another KASAN bug [1] on aarch64 machine, by running generic/388.
> I hit that 3 times. And hit a panic [2] (once after that kasan bug) on a x86_64
> with pmem device (mount with dax=never), by running geneirc/388 too.

Can you try with this? I suspect the preempt dance isn't really
necessary, but I can't quite convince myself that it isn't. In any case,
I think this should fix it and this was exactly what I was worried about
but apparently not able to easily trigger or prove...


diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c
index fe38eb0cbc82..878ec3feeba9 100644
--- a/io_uring/io-wq.c
+++ b/io_uring/io-wq.c
@@ -220,7 +220,9 @@ static void io_worker_exit(struct io_worker *worker)
 	list_del_rcu(&worker->all_list);
 	raw_spin_unlock(&wq->lock);
 	io_wq_dec_running(worker);
-	worker->flags = 0;
+	preempt_disable();
+	current->worker_private = NULL;
+	preempt_enable();
 
 	kfree_rcu(worker, rcu);
 	io_worker_ref_put(wq);

-- 
Jens Axboe