On 24/08/2020 23:35, Pavel Begunkov wrote:
> do_exit() first drops current->mm and then runs task_work, from where
> io_sq_thread_acquire_mm() would try to set mm for a user dying process.
This is a backport of [1] + [2] for 5.8. Let's wait to see if
Roman Gershman can test it.
[1] 8eb06d7e8dd85 ("io_uring: fix missing ->mm on exit")
[2] cbcf72148da4a ("io_uring: return locked and pinned page accounting")
>
> [ 208.004249] WARNING: CPU: 2 PID: 1854 at
> kernel/kthread.c:1238 kthread_use_mm+0x244/0x270
> [ 208.004287] kthread_use_mm+0x244/0x270
> [ 208.004288] io_sq_thread_acquire_mm.part.0+0x54/0x80
> [ 208.004290] io_async_task_func+0x258/0x2ac
> [ 208.004291] task_work_run+0xc8/0x210
> [ 208.004294] do_exit+0x1b8/0x430
> [ 208.004295] do_group_exit+0x44/0xac
> [ 208.004296] get_signal+0x164/0x69c
> [ 208.004298] do_signal+0x94/0x1d0
> [ 208.004299] do_notify_resume+0x18c/0x340
> [ 208.004300] work_pending+0x8/0x3d4
>
> Reported-by: Roman Gershman <>
> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
> ---
> fs/io_uring.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 493e5047e67c..a8b3a608c553 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -4313,7 +4313,8 @@ static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx,
> struct io_kiocb *req)
> {
> if (io_op_defs[req->opcode].needs_mm && !current->mm) {
> - if (unlikely(!mmget_not_zero(ctx->sqo_mm)))
> + if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL) ||
> + !mmget_not_zero(ctx->sqo_mm)))
> return -EFAULT;
> kthread_use_mm(ctx->sqo_mm);
> }
>
--
Pavel Begunkov
