Re: [PATCH] mdadm: Deprecate the obsolete --auto-detect option

["Masalkhi, Abd Alrhman" <abd.masalkhi@gatech.edu>]

Hi,

On Wed, Oct 28, 2026 at 15:00 +0100, Mariusz Tkaczyk wrote:
> On Wed, 28 Oct 2026 12:52:30 +0000
> "Masalkhi, Abd Alrhman" <abd.masalkhi@gatech.edu> wrote:
>
>> Hi Mariusz,
>> 
>> On Wed, Oct 28, 2026 at 10:03 +0100, Mariusz Tkaczyk wrote:
>> > On Thu, 15 Oct 2026 13:03:59 +0000
>> > "Masalkhi, Abd Alrhman" <abd.masalkhi@gatech.edu> wrote:
>> >  
>> >> From: Abd-Alrhman Masalkhi <abd.masalkhi@gatech.edu>
>> >> 
>> >> --auto-detect opened md0 (9:0) and issued the RAID_AUTORUN ioctl to
>> >> ask the kernel to start the arrays it had auto-detected. The kernel
>> >> removed this ioctl in commit d82fa81c2364 ("md: replace the
>> >> RAID_AUTORUN ioctl with a direct function call"). Before that, it
>> >> worked only when md was built into the kernel. Since then,
>> >> md_ioctl()  
>> >
>> > Redhat still builds-in md into kernel so there is chance that
>> > someone is using it.
>> >
>> >  
>> >> has rejected it with ENOTTY, making --auto-detect ineffective on
>> >> current kernels.  
>> >
>> >
>> > Hello,
>> > Do you mind to submit github PRs for mdadm patches?
>> > https://github.com/md-raid-utilities/mdadm
>> >  
>> 
>> Sure, I will submit the next version as a GitHub pull request.
>> 
>> One question, should I also send the RFC patches for updating mdadm to
>> use the new UAPI to the mailing list, or is submitting GitHub PRs
>> sufficient?
>
>
> In this particular case there is nothing interesting we are going to do
> with kernel or any problem that needs to be coordinated with kernel.
> Just take new UAPI and use it once is available so I don't think kernel
> list messaging is needed.
>
> I forgot to say, the problem you are challenging is very legacy. With
> proposed change, we will not need to open the device and generate
> unnecessary events (and problems) anymore! I'm glad to see someone
> working on it.
>

Yes, that is the end goal. I plan to do it in two phases. the first
one is to  but keep the array open as today, except for stop and
read-only and i will open the pull request today or tomorrow.

And the second is to stop opening the array entirely when the control
node is available. It is the harder part.  Much of mdadm is built around
the array fd, it is passed through most functions and used for sysfs
lookups and doing it separately keeps the first series reviewable.

> Mariusz
>

-- 
Best Regards,
Abd-Alrhman