Re: [PATCH] debugfs: ignore auto and noauto options if given

[Eric Sandeen]

On 6/3/24 9:33 AM, Christian Brauner wrote:
> On Mon, Jun 03, 2024 at 09:17:10AM -0500, Eric Sandeen wrote:
>> On 6/3/24 8:31 AM, Christian Brauner wrote:
>>> On Mon, Jun 03, 2024 at 09:24:50AM +0200, Wolfram Sang wrote:
>>>>
>>>>>>> Does that fix it for you?
>>>>>>
>>>>>> Yes, it does, thank you.
>>>>>>
>>>>>> Reported-by: Wolfram Sang <wsa+renesas@xxxxxxxxxxxxxxxxxxxx>
>>>>>> Tested-by: Wolfram Sang <wsa+renesas@xxxxxxxxxxxxxxxxxxxx>
>>>>>
>>>>> Thanks, applied. Should be fixed by end of the week.
>>>>
>>>> It is in -next but not in rc2. rc3 then?
>>>
>>> Yes, it wasn't ready when I sent the fixes for -rc2 as I just put it in
>>> that day.
>>>
>>
>> See my other reply, are you sure we should make this change? From a
>> "keep the old behavior" POV maybe so, but this looks to me like a
>> bug in busybox, passing fstab hint "options" like "auto" as actual mount
>> options being the root cause of the problem. debugfs isn't uniquely
>> affected by this behavior.
>>
>> I'm not dead set against the change, just wanted to point this out.
> 
> Hm, it seems I forgot your other mail, sorry.

No worries!

> So the issue is that we're breaking existing userspace and it doesn't
> seem like a situation where we can just ignore broken userspace. If
> busybox has been doing that for a long time we might just have to
> accommodate their brokenness. Thoughts?

Yep, I can totally see that POV.

It's just that surely every other strict-parsing filesystem is also
broken in this same way, so coding around the busybox bug only in debugfs
seems a little strange. (Surely we won't change every filesystem to accept
unknown options just for busybox's benefit.)

IOWS: why do we accomodate busybox brokenness only for debugfs, given that
"auto" can be used in fstab for any filesystem?

But in simplest terms - it was, in fact, debugfs that a) changed and
b) got the bug report, so I don't have strong objections to going back
to the old behavior.

-Eric