Re: [PATCH 3/5] blktrace: refcount the request_queue during ioctl
From: Luis Chamberlain
Date: Thu Apr 16 2020 - 01:29:19 EST
On Wed, Apr 15, 2020 at 08:43:32PM -0700, Bart Van Assche wrote:
> On 2020-04-15 18:12, Luis Chamberlain wrote:
> > On Wed, Apr 15, 2020 at 07:18:22AM -0700, Bart Van Assche wrote:
> >> blk_get_queue() prevents concurrent freeing of struct request_queue but
> >> does not prevent concurrent blk_cleanup_queue() calls.
> >
> > Wouldn't concurrent blk_cleanup_queue() calls be a bug? If so should
> > I make it clear that it would be or simply prevent it?
>
> I think calling blk_cleanup_queue() while the queue refcount > 0 is well
> established behavior. At least the SCSI core triggers that behavior
> since a very long time. I prefer not to change that behavior.
I see. An alternative is to simply check if we already are cleaning up
and if so abort early on the blk_cleanup_queue(). That would allow
re-entrant calls, and just be a no-op to the additional calls. Or is
the re-entrant, two attemps to really do all the work
blk_cleanup_queue() expected functionality already?
> Regarding patch 3/5: how about dropping that patch? If the queue
> refcount can drop to zero while blk_trace_ioctl() is in progress I think
> that should be fixed in the block_device_operations.open callback
> instead of in blk_trace_ioctl().
I'll take a look, thanks!
Luis