[Samba] XP Clients Showing Incorrect Filenames

[TAKAHASHI Motonobu]

From: Andy Theuninck <gohanman at gmail.com>
Date: Fri, 27 May 2011 11:01:36 -0500

> I'm seeing something fairly odd where XP clients show some - but not
> nearly all -filenames as a random jumble of characters.

This is by design. Samba uses hash2 method to generate short
filenames. According to the old document, by default an 8+3 file name
is constructed as

  1st byte       : The first of the filename (if ASCII)
  2nd - 6th bytes: hashed value from the filename
  7th            : ~ 
  8th            : hashed value from the filename

Also the first 3 bytes from extension becomes the 3 byte extension.

> The odd file names all look like "AH6I10~Z". Besides the tilde always
> being 2nd to last, the other characters don't have any consistent
> mapping to the actual file name.

This odd filename is valid for hash2 method. And about the extension
see the BUG#7719.

From: Andy Theuninck <gohanman at gmail.com>
Date: Fri, 27 May 2011 12:39:07 -0500

> On a sidenote, the behavior would be tolerable if it wasn't removing
> the file extension and thus Windows' ability to identify file type. So
> it's more like just 8 shortening; the .3 part is missing.

I've already posted this bug: 
  Long extensions aren't shortened correctly in mangled names  
  https://bugzilla.samba.org/show_bug.cgi?id=7719  

I examined this behavior against at least Samba 3.0.24/3.2.4/3.5.4.

I found other bug in hash2 method that if the extension contains
non-ASCII chars, the extension of short filename is vanished.

Anyway hash2 method generates mostly "randomized" filename so it's
very inconvinient. I hope the short filename is generated with the
same method as Windows and stored in EA.

P.S.
hash method is also completely broken:
  Bug 1521 - SFN process must be based on dos charset, not UCS2 
  and also this method sometimes generates same short filename as 
  described at smb.conf(5).

---
TAKAHASHI Motonobu <monyo at samba.gr.jp>