On Tue, Dec 21, 2021 at 7:25 AM Jens Axboe [off-list ref] wrote:

On 12/21/21 3:44 AM, Eric Dumazet wrote:

quoted

On Tue, Dec 21, 2021 at 1:52 AM syzbot
[off-list ref] wrote:

quoted

syzbot has bisected this issue to:

commit e4b8954074f6d0db01c8c97d338a67f9389c042f
Author: Eric Dumazet [off-list ref]
Date:   Tue Dec 7 01:30:37 2021 +0000

    netlink: add net device refcount tracker to struct ethnl_req_info

Unfortunately this commit will be in the way of many bisections.

Real bug was added in

commit 5fc11eebb4a98df5324a4de369bb5ab7f0007ff7
Author: Christoph Hellwig [off-list ref]
Date:   Thu Dec 9 07:31:29 2021 +0100

    block: open code create_task_io_context in set_task_ioprio

    The flow in set_task_ioprio can be simplified by simply open coding
    create_task_io_context, which removes a refcount roundtrip on the I/O
    context.

    Signed-off-by: Christoph Hellwig [off-list ref]
    Reviewed-by: Jan Kara [off-list ref]
    Link: https://lore.kernel.org/r/20211209063131.18537-10-hch@lst.de (local)
    Signed-off-by: Jens Axboe [off-list ref]

There are only really 5 patches in between the broken commit and the one
that fixes it, and it only affects things trying to set the ioprio with
a dead task. Is this a huge issue? I don't see why this would cause a
lot of bisection headaches.

I was saying that my commit was polluting syzbot bisection, this is a
distraction in this report.
(Or if you prefer, please ignore syzbot bisection)

linux-next has still this bug in set_task_ioprio()

--
Jens Axboe