On Thu, Apr 1, 2021 at 6:03 AM heming.zhao@xxxxxxxx
<heming.zhao@xxxxxxxx> wrote:
>
> On 4/1/21 2:17 PM, Christoph Hellwig wrote:
> > On Thu, Apr 01, 2021 at 09:34:56AM +0800, Zhao Heming wrote:
> >> commit d3374825ce57 ("md: make devices disappear when they are no longer
> >> needed.") introduced protection between mddev creating & removing. The
> >> md_open shouldn't create mddev when all_mddevs list doesn't contain
> >> mddev. With currently code logic, there will be very easy to trigger
> >> soft lockup in non-preempt env.
> >
> > As mention below, please don't make this even more of a mess than it
> > needs to. We can just pick the two patches doing this from the series
> > I sent:
> >
>
> Hi,
>
> I already got your meaning on previously email.
> I sent v2 patch for Song's review comment. My patch is bugfix, it may need
> to back port into branch maintenance.
>
> Your attachment patch files is partly my patch.
> The left part is in md_open (response [PATCH 01/15] md: remove the code to flush
> an old instance in md_open)
> I still think you directly use bdev->bd_disk->private_data as mddev pointer is not safe.
>
Hi Christoph and Heming,
Trying to understand the whole picture here. Please let me know if I
misunderstood anything.
IIUC, the primary goal of Christoph's set is to get rid of the
ERESTARTSYS hack from md,
and eventually move bd_mutext. 02/15 to 07/15 of this set cleans up
code in md.c, and they
are not very important for the rest of the set (is this correct?).
Heming, you mentioned "the solution may simply return -EBUSY (instead
of -ENODEV) to
fail the open path". Could you please show the code? Maybe that would
be enough to unblock
the second half of Christoph's set (08/15 to 15/15)?
Once this part is resolved, the bug fix (this patch thread) and the
clean up (Christoph's 02 - 07)
should be easier.
Would this work?
Thanks,
Song
