Re: [RFC PATCH] fs: Plumb case sensitivity bits into statx

[Chuck Lever <cel@kernel.org>]

On 9/25/25 9:20 PM, Christoph Hellwig wrote:
> On Thu, Sep 25, 2025 at 11:11:40AM -0400, Chuck Lever wrote:
>> +	if (request_mask & STATX_CASE_INFO) {
>> +		stat->result_mask |= STATX_CASE_INFO;
>> +		/* STATX_CASE_PRESERVING is cleared */
>> +		stat->case_info = statx_case_ascii;
> 
> FAT is using code pages specified on the command line for it's case
> insensitivity handling, which coverse much more than ASCISS.

I noticed that a mount option controls whether the filename encoding is
UTF-8. Clearly this will need more logic to set the correct bit.


>> +/* Values stored in the low-order byte of .case_info */
>> +enum {
>> +	statx_case_sensitive = 0,
>> +	statx_case_ascii,
>> +	statx_case_utf8,
>> +	statx_case_utf16,
>> +};
> 
> What are these supposed to mean?

For the moment, the meaning is unclear because I simply wanted to
demonstrate that more than one behavior can be reported. Someone with
greater expertise than mine can help refine the specific semantics.


> ASCII, utf8 and utf16 are all
> encodings and not case folding algorithms.  While the folding is obvious
> for ASCII, it is not for unicode and there are all kinds of different
> variants.

Fair enough... this is the right spot to report those variants. Or we
can decide that is inconsequential or impossible and simply reduce this
to a single "filename case is {in}sensitive" bit.


> Also I don't know of any file systems using utf16 encoding
> and even if it did, it would interact with the VFS and nfsd using
> utf8.  Note that the 16-bit ucs-2 encoding used by windows file systems
> is a different things than unicode encodings like utf16.

Sure, UTF16 can be dropped or replaced.


-- 
Chuck Lever