Samba4 shared FS + MacOS issue
George K Colley
gcolley at apple.com
Wed Feb 25 10:10:21 MST 2009
Did some testing here at Connectathon and you are correct. If the
server supports the LARGE_WRITE_CAP then the Apple SMB Client will set
the byte count field zero. We should only do this if the message is
too large to fit in the byte count field. Now with that said not sure
when that change will make it in. Since Windows, Samba 3 and all other
servers don't seem to care I am not sure why Samba 4 should care. I
would suggest that Samba 4 change their code to work with our old
clients, but I will leave that to the Samba 4 team.
Some other notes about Samba 4 and the Apple Client.
1. Samba 4 will return an error if the Apple Client attempts to
delete a stream while the main stream is open. This will cause the
Finder to fail the operation. This is a bug in the Samba 4 code. Samba
3 and Windows Server support doing this operation.
2. Samba 4 will return an error when the client tries to set the write
time using a stream file descriptor. This is perfectly legal, remember
all meta data (except for the EOF and PEOF) is share between streams.
I found some other minor errors with the Apple Client which I need to
fix, but nothing that should be noticeable to the user.
George
On Feb 4, 2009, at 9:28 AM, Jeff Morris wrote:
> Hi George:
>
> In a quick review of the trace, the WriteAndX commands are being
> sent with the bcc field set to 0. (I expected it to be 5. )
>
> I've attached two trace packets, one from an existing trace that I
> had and one that is from the trace that you sent me.
>
> It might not be significant, but it is a difference.
>
> Ciao;
> JLM
>
> WriteAndXmple.pcap (Exemplar):
>
> <WriteAndXmple.pcap>
>
>
> WriteAndXmple2.pcap (from the forwarded trace):
>
> <WriteAndXmple2.pcap>
>
>
>
>
> On Feb 4, 2009, at 8:23 AM, George K Colley wrote:
>
>> <DumpFile.dmp>
>> On Feb 4, 2009, at 1:19 AM, Jeff Morris wrote:
>>
>>> Hi George:
>>>
>>> Was there ever a full packet trace produced for this? The one that
>>> I saw from Xavier did not appear to have captured all of the
>>> packet bytes.
>> This one shows the write errors
>>>
>>>
>>> It sounds like you can reproduce the problem, maybe you can
>>> capture a trace (with all of the packet bytes).
>> I do not have samba 4 running on any system here, so I have not
>> seen this issue. Looks like Xavier can reproduce it at will. I
>> will be a Connectathon wonder if they will have Samba 4 install?
>>>
>>>
>>> Thanks;
>>> JLM
>>>
>>> On Feb 3, 2009, at 2:32 PM, George K Colley wrote:
>>>
>>>> Sorry for the delay. I am going to need someone from the Samba 4
>>>> team to take a look and see if they can figure out what it does
>>>> like. From what I can tell ever thing is ok, the server just
>>>> returns NT_STATUS_UNSUCCESSFUL on my writes. I have no idea why?
>>>>
>>>> George
>>>> On Jan 30, 2009, at 11:37 AM, Xavier Ambrosioni wrote:
>>>>
>>>>> same test with the log set to 10.
>>>>> 172.16.0.4 is the mac client.
>>>>>
>>>>> xavier
>>>>>
>>>>> <samba.log.gz>
>>>>>
>>>>>
>>>>> Le 30 janv. 09 à 20:31, George K Colley a écrit :
>>>>>
>>>>>> Well that wasn't very helpful. I was expecting Samba 4 to log
>>>>>> an error on the write, but they didn't. Lets crank the logging
>>>>>> all the way up, I think 10 is the highest.
>>>>>>
>>>>>> George
>>>>>> On Jan 30, 2009, at 10:59 AM, Xavier Ambrosioni wrote:
>>>>>>
>>>>>>> Hi
>>>>>>>
>>>>>>> in attachment the log file (level 5)
>>>>>>>
>>>>>>> xavier
>>>>>>>
>>>>>>>
>>>>>>> <samba.log.gz>
>>>>>>>
>>>>>>>
>>>>>>> Le 30 janv. 09 à 17:37, George K Colley a écrit :
>>>>>>>
>>>>>>>>
>>>>>>>> On Jan 29, 2009, at 7:43 PM, Xavier Ambrosioni wrote:
>>>>>>>>
>>>>>>>>> Ok, I will send you the samba log file in few hours. Which
>>>>>>>>> log level do you need?
>>>>>>>> Not sure lets try 5 then go from there.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> I found in the code an option called "server signing". I
>>>>>>>>> tried to set it to "off" and "disabled" but the result is
>>>>>>>>> still the same. I don't know if that option is really
>>>>>>>>> implemented yet.
>>>>>>>>>
>>>>>>>>> Xavier
>>>>>>>>>
>>>>>>>>> Le 29 janv. 09 à 22:57, George K Colley <gcolley at apple.com>
>>>>>>>>> a écrit :
>>>>>>>>>
>>>>>>>>>> So I am not sure what they don't like. We should get the
>>>>>>>>>> samba 4 logs and see if they have any info.
>>>>>>>>>> On Jan 29, 2009, at 11:35 AM, Xavier Ambrosioni wrote:
>>>>>>>>>>
>>>>>>>>>>> Hi George,
>>>>>>>>>>>
>>>>>>>>>>> in attachment the dump file starting before the mount of
>>>>>>>>>>> the drive.
>>>>>>>>>>> Yes there is the same issue even with 1 byte.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Xavier
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> <DumpFile.dmp.gz>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Le 29 janv. 09 à 01:10, George K Colley a écrit :
>>>>>>>>>>>
>>>>>>>>>>>> One other note doesn't seem to be a size issue because
>>>>>>>>>>>> writes of 1 byte have the same issue?
>>>>>>>>>>>> On Jan 28, 2009, at 4:03 PM, George K Colley wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>> So the SMB packet signing issue which was reported with
>>>>>>>>>>>>> Win2k and Win2K8 system was correct in 10.5.3. That
>>>>>>>>>>>>> issue was cause by the client using the wrong buffer
>>>>>>>>>>>>> size on writes.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Now in the trace packet 353 opens a file for read/
>>>>>>>>>>>>> writing, then pack 356 attempts to write 4096 and the
>>>>>>>>>>>>> server returns an error NT_STATUS_UNSUCCESSFUL
>>>>>>>>>>>>> (0xc0000001)?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Now packet signing is on, so it could be related just
>>>>>>>>>>>>> not sure. Would like to retest with packet signing
>>>>>>>>>>>>> turned off, does anyone on the list know how to turn
>>>>>>>>>>>>> signing off with Samba4?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Next can you get me a tcpdump starting before the mount
>>>>>>>>>>>>> so I can see the CAP of the server?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Gerge
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> On Jan 28, 2009, at 11:09 AM, Xavier Ambrosioni wrote:
>>>>>>>>>>>>>
>>>>>>>>>>>>>> Hi,
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> sorry, I did not know that '-w' option produce a dump
>>>>>>>>>>>>>> with more information.
>>>>>>>>>>>>>> You will find that dump in attachment.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> 172.16.0.2 is the Mac
>>>>>>>>>>>>>> 172.16.0.10 is the samba server
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Thank you
>>>>>>>>>>>>>> Xavier
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> <DumpFile.dmp>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Le 28 janv. 09 à 10:22, Gavin Beatty a écrit :
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> 2009/1/27 Xavier Ambrosioni <xavier.ambrosioni at cinema-voiron.fr
>>>>>>>>>>>>>>> >:
>>>>>>>>>>>>>>>> Hi,
>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>> you will find in attachment the dump file. I used the
>>>>>>>>>>>>>>>> command 'tcpdump -i
>>>>>>>>>>>>>>>> en0 -s 0 |grep -v vnc >DumpFile.txt' because I'm
>>>>>>>>>>>>>>>> currently using VNC to
>>>>>>>>>>>>>>>> connect to my mac and I wanted to remove the vnc
>>>>>>>>>>>>>>>> packets from the dump file
>>>>>>>>>>>>>>>> to decrease its size.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Hi Xavier,
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Please follow George's instructions as the text output
>>>>>>>>>>>>>>> of tcpdump is
>>>>>>>>>>>>>>> not a "dump" in the proper sense. The important part
>>>>>>>>>>>>>>> is the -w
>>>>>>>>>>>>>>> ~/Desktop/DumpFile.dmp. This writes a true dump file
>>>>>>>>>>>>>>> which has much
>>>>>>>>>>>>>>> more (extremely useful!) information. Using wireshark
>>>>>>>>>>>>>>> (or ethereal),
>>>>>>>>>>>>>>> it's easy for the samba guys to visually exclude your
>>>>>>>>>>>>>>> VNC traffic when
>>>>>>>>>>>>>>> they view the DumpFile.dmp so there is no need for you
>>>>>>>>>>>>>>> to worry about
>>>>>>>>>>>>>>> it.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> So, if you follow George's instructions exactly and
>>>>>>>>>>>>>>> attach the
>>>>>>>>>>>>>>> DumpFile.dmp, all will be well.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Cheers,
>>>>>>>>>>>>>>> Gavin
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> --
>>>>>>>>>>>>>>> Gavin Beatty
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> SEMPER UBI SUB UBI
>>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>> Jeff Morris
>>> jlm at dessertfirst.com
>>> (510) 517-9157 (Cell)
>>>
>>>
>>>
>>>
>>>
>>
>
> Jeff Morris
> jlm at dessertfirst.com
> (510) 517-9157 (Cell)
>
>
>
>
>
More information about the samba-technical
mailing list