On Tue 07 Nov 2017 05:19:41 PM CET, Anton Nefedov wrote: > BlockBackend gets deleted by another job's stream_complete(), deferred > to the main loop, so the fact that the job is put to sleep by > bdrv_drain_all_begin() doesn't really stop it from execution.
I was debugging this a bit, and the block_job_defer_to_main_loop() call happens _after_ all jobs have been paused, so I think that when the BDS is drained then stream_run() finishes the last iteration without checking if it's paused. Without your patch (i.e. with a smaller STREAM_BUFFER_SIZE) then I assume that the function would have to continue looping and block_job_sleep_ns() would make the job coroutine yield, effectively pausing the job and preventing the crash. I can fix the crash by adding block_job_pause_point(&s->common) at the end of stream_run() (where the 'out' label is). I'm thinking that perhaps we should add the pause point directly to block_job_defer_to_main_loop(), to prevent any block job from running the exit function when it's paused. Somehow I had the impression that we discussed this already in the past (?) because I remember thinking about this very scenario. Berto