> Note the file insensitiveness rules differ for
> different filesystems (e.g. MS-DOS applied case-insensitiveness to
> ASCII subset only).

Ok, this kills my approach. I didn't know that.

> Perhaps it would be useful to repeat why exactly you'd like generic 
> 'case-insensitiveness' support for all filesystems.

I personally have the need for one special NFS mount to behave case 
insensitive with the Samba semantics. I.e. I have that filesystem NFS
mounted and write to it, while most other clients have it SMB mounted (mostly
read only). There are some pitfalls where my scripts accidently create
filenames differing in case only - and boom - the SMB clients can't read
neither anymore.

Not knowing the different semantics I searched for the most general solution.

> IMHO Samba can do much better job in translating file names than a kernel
> reasonably can.

Yes, but: creating a file should be a simple creat() call. Only the kernel
can provide this with atomic semantics if case-only-different file names are
not allowed.

> FWIW, I don't expect filename translation to be
> the actual performance bottleneck for Samba.

Sure, it isn't. Samba is a total loss when it comes to performance, mainly
due to it being userland only. See how NFS servers loose when implemented in 
userland. You can easily outperform any NetBSD file server in a SMB 
environment with 1/10 the hardware performance by installing Win2K.

But this is completely off topic now.

Sorry for the waste bandwitht, thanks for all the enlightenment


Martin