On Wed, 2018-04-04 at 19:26 -0600, Jens Axboe wrote: > Leaving the whole trace here, but I'm having a hard time making sense of it. > It complains about a user-after-free in the inflight iteration, which is only > working on the queue, request, and on-stack mi data. None of these would be > freed. The below trace on allocation and free indicates a bio, but that isn't > used in the inflight path at all. Is it possible that kasan gets confused here? > Not sure what to make of it so far. Hello Jens, In the many block layer tests I ran with KASAN enabled I have never seen anything like this nor have I seen anything that made me wonder about the reliability of KASAN. Maybe some code outside the block layer core corrupted a request queue data structure and triggered this weird report? Bart.
