From: John E. Malmberg [wb8tyw@qsl.network]
Sent: Sunday, October 31, 2004 9:13 PM
To: Info-VAX@Mvb.Saic.Com
Subject: SAMBA 2.2.12 - (was: Samba 2.2.8 for OpenVMS, new release)

John Reinhardt wrote:
>>
>>
>>For those interested in being more on the bleeding edge, I have now
>>merged in the SAMBA 2.2.12 changes with last source kit that I produced.
>>
>>I also fixed the bug that I accidently introduced, and implemented the
>>fsync() on the 1 byte writes to try to improve the large file issues.
>>
>>The source kits can be found at:
>>
>> http://encompasserve.org/~malmberg/samba/

>> SAMBA_2_2_12-V82-SRC-20041017_BCK.ZIP

That kit has some serious bugs, and has been replaced with:

SAMBA_2_2_12-V82-SRC-20041031.ZIP;1

In the previous 2.2.12 kit and the 2.2.8 kit that I posted on 
Encompasserve, I introduced two major bugs.  One is that it bypassed the 
filename translation code, and as a result also bypassed the stat cache 
for performance.

It also broke the operation of the stat cache.

This new 2.2.12 kit corrects those bugs, and also incorporates all of 
Jean-Yves Collot's (JYC) fixes that have been documented to date.

In addition, it has several other changes:

1. In previous versions of both the JYC and my source kits in the module 
vms_support.c, sections of the item_list structure to the SYS$GETUAI 
call were not initialized.  This could lead to internal data structures 
being corrupted, or crashes with access violations.

2. Attempting to debug at level 5 or higher will result in a infinite 
recursion until the stack is exhausted.

3. strcpy was used in several places to move overlapping sections of a 
string.  This is documented as to cause undefined behavior, and showed 
up as causing a corrupted stack during debug.  Replaced all occurrences 
that I found of this with memmove() instructions.

4. Out of memory conditions from realloc() calls were not being 
detected, resulting in the SMBD process crashing with an access 
violation.  Added messages to write the error condition to the log and 
then exit the SMBD.

5. To prevent memory fragmentation, which can cause premature cases of 
the SMBD process running out of memory, I increased the amount of memory 
chunks allocated for storing a cached list of directory contents.

6. The tests for VMS Version were failing for EFT and VMS internal 
builds.  This was used to determine ODS-5 support.  Changed to assume 
that any EFT or VMS internal builds that are not on VAX support ODS-5.

This has not undergone any significant testing, all I did was verify:

* that the server will transfer some files.
* that new files and directories can be created and renamed.

I have only done a test build on OpenVMS ALPHA 8.2 EFT at this time.  I 
attempted to make this code build on older versions, but as yet I have 
not had time to do any test builds.


If it is found not to build/work on a release before 8.2 EFT, post on 
the SAMBA-VMS mailing list the fix, and I will attempt to incorporate it
into a future source code drop.  My main focus is going to start being 
on Samba 3 and Samba 4 and the build support for them, so I will not 
have time to either prior version or VAX testing.

> Work is also being done on Samba 3.X and 4.X for VMS.

So far it has been in the investigation stage, I do not know of anyone 
that is attempting a build at this time.  If there is anyone doing so, 
please post in the SAMBA-VMS mailing list.

Instructions for either joining the list, or using the gmane newsgroup 
mirror are available at http://www.samba.org.

>   John H. Reinhardt
> 
> (Apologies to John for reproducing his post)

None needed.

It is available to the public from either the archives of the SAMBA-VMS 
mailing list or from the gmane newsgroups that mirror.

-John
wb8tyw@qsl.network
Personal Opinion Only