Re: [PATCH RFC] char: misc: Enforce simple minor space division

[Zijun Hu <zijun_hu@icloud.com>]

On 2025/6/26 20:37, Thadeu Lima de Souza Cascardo wrote:
>> drivers/misc/misc_minor_kunit.c may need to be corrected to reflecting
>> division today.
>>
> Correct, those added tests do not enforce that one cannot register a static
> minor above MISC_DYNAMIC_MINOR.
> 
> However, to some extent [1], it tests that a MISC_DYNAMIC_MINOR allocation
> will not return a number in the range of the static numbers. See
> miscdev_test_dynamic_only_range.
> 
> It also tests that if a dynamic minor is allocated and one tries to
> allocate that same number statically, it will fail. It also tries [2] to
> test the other way around. That is, if one minor was statically allocate in
> the dynamic range, that a dynamic allocation will not return that same
> number.
> 
> Those tests, named miscdev_test_conflict and miscdev_test_conflict_reverse
> were written considering the current implementation, which allows for
> static allocations in the "dynamic range".
> 
> If we are going to change that, you need to change the tests too.
> 

agree.

> I would suggest applying only the last hunk of your patched
> drivers/char/misc.c with a separate commit. Then, when misc_deregister

agree.

> would be called the minor would be restored. However, since statically
> allocating a minor above 255 would still be allowed, it could "restore" it
> wrongly.
> 

now. will disallow statically allocating a minor above 255 since below
disadvantages:

1) if the minor was requested by other dynamic request. requesting a
fixed minor which > 255 will be failed.

2) both fixed minors and dynamic minors interleave int the same space.


> As Greg has observed, if there is no known case in the kernel where minor
> is not set to MISC_DYNAMIC_MINOR by the driver before it calls
> misc_register a second time, then there is nothing to fix here. If there
> is such a case, then the driver must be fixed. It has always been the case,
> even when the ranges were different, that the minor needed to be reset to
> MISC_DYNAMIC_MINOR before calling misc_register after misc_deregister has
> been called.
> 

this patch is to solve such reentry issue.


> As you point out, when misc_register fails, it already restores the minor
> number.
> 
> Let me know if you want to proceed with this change and need help with the
> test case. I may be slow to respond since I am going on vacation.
> 

I will split this RFC patch and modify test cases.

welcome to get your comments at any time. thank you (^^).

> Regards.
> Cascardo.
> 
> 
> [1] To some extent, because due to the large dynamic range, it only tries
> allocating 256 dynamic minors. And only verifies numbers below 16 are not
> returned, because that was the bug that existed before.
> 
> [2] Tries, because due to the large dynamic range, instead of allocating
> all numbers, it just assumes that the allocator would give the first
> number, but it also picks that "first" number by doing a dynamic allocation
> and freeing it.