Re: [PATCH 08/32] aio: replace kiocb_set_cancel_fn with a cancel_kiocb file operation

[Al Viro]

On Tue, May 15, 2018 at 09:48:09PM +0200, Christoph Hellwig wrote:

>  	case -EIOCBQUEUED:
> +		if (req->ki_filp->f_op->cancel_kiocb) {
> +			struct aio_kiocb *iocb =
> +				container_of(req, struct aio_kiocb, rw);
> +			struct kioctx *ctx = iocb->ki_ctx;
> +			unsigned long flags;
> +
> +			spin_lock_irqsave(&ctx->ctx_lock, flags);
> +			list_add_tail(&iocb->ki_list, &ctx->active_reqs);

Use after free - that list insertion used to be done by drivers and doing
so before any ->ki_complete() calls might've happened used to be their
responsibility.  Now you've taken that to the point after ->read_iter()
(or ->write_iter()) return, so there's no way in hell to guarantee it's
not been completed (and freed) by that point.

Incidentally, none of the callers gives a damn about the difference between
0 and -EIOCBQUEUED now, so aio_rw_ret() might as well had been made void...