samba - Jun 2008 - multiple smb commands (some non AndX) in one packet

Does samba server handle multiple smb commands (not necessarily andx commands)
within one packet?  If a client were to send commands such as 0x2f, 0x32, 0x32
in one packet instead of three separate packets, does samba server
handle it i.e.
respond to each of the three requests?

On Wed, Jun 04, 2008 at 01:36:20PM -0500, Shirish Pargaonkar
wrote:
> Does samba server handle multiple smb commands (not necessarily andx
commands)
> within one packet?  If a client were to send commands such as 0x2f, 0x32,
0x32
> in one packet instead of three separate packets, does samba server
> handle it i.e.
> respond to each of the three requests?
Yes, we should do that. Of course there are earlier
bugs in the implementation but I believe as of our
latest code (3.0.30 and 3.2rc's) we handle this
correctly.

Jeremy.

On Thu, Jun 5, 2008 at 1:34 AM, Volker Lendecke
<Volker.Lendecke@sernet.de> wrote:
> On Wed, Jun 04, 2008 at 03:28:31PM -0500, Shirish Pargaonkar wrote:
>> cifs.ko.
>> You mean 4-byte header FF 53 4D 42?
>> In another recreate, tcp frame has two smb requests,
>> close (0x4) and locking andx (0x24) each with this four byte header.
>
> Wait a second -- XP chokes on something that the Linux
> cifsfs sends??? How can you reproduce this?
>
> Volker
>


Mounted a share and ran this command

fsstress -d /mnt -l 0 -n 1000 -p 10 -r

I also took out the code in connect.c where it sets send and receive
buffer sizes
of a socket to 200K and 140K respectively (thus letting tcp auto-tuning take
care of socket buffer sizes).

        if ((*csocket)->sk->sk_sndbuf < (200 * 1024))
                (*csocket)->sk->sk_sndbuf = 200 * 1024;
        if ((*csocket)->sk->sk_rcvbuf < (140 * 1024))
                (*csocket)->sk->sk_rcvbuf = 140 * 1024;

You will start getting messages like server not responding and No
response to/for cmd

On Thu, Jun 05, 2008 at 07:29:49AM -0500, Shirish Pargaonkar
wrote:
> Mounted a share and ran this command
> 
> fsstress -d /mnt -l 0 -n 1000 -p 10 -r
google points at quite a few sites when asking it for
"fsstress". What particular tool is that?

And, this is definitely a cifsfs bug.

Volker
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url :
http://lists.samba.org/archive/samba/attachments/20080605/8221a0f2/attachment.bin

On Thu, Jun 5, 2008 at 8:58 AM, Volker Lendecke
<Volker.Lendecke@sernet.de> wrote:
> On Thu, Jun 05, 2008 at 07:29:49AM -0500, Shirish Pargaonkar wrote:
>> Mounted a share and ran this command
>>
>> fsstress -d /mnt -l 0 -n 1000 -p 10 -r
>
> google points at quite a few sites when asking it for
> "fsstress". What particular tool is that?
>
> And, this is definitely a cifsfs bug.
>
> Volker
>
I downloaded ltp-full-20080131.tgz few months back and built it and using
network/nfs/nfs_fsstress/fsstress.

I just recreated the problem a while ago against XP server, this time there were
single commands per tcp packet.  So I think the problem is server at some point
just does not respond to cifs client in time (for whatever reasons)
for cifs client to
decide to reconnect declaring server not responding.

I suspect the same for samba server.  In case of samba server, I had turned on
log level to 3 to see whether any messages are logged by samba server as to why
it is not responding to cifs client smb requests but found nothing
significant in the
log file that was generated.