samba - May 2020 - Win7 clients problem after upgrading samba file server to 4.12 on Arch

Hi Volker, Jermey, and the Community,

To be clear: I didn't analyze your code, I don't know structures of
Samba,
definitely I am not a person to judge and blame anybody :)
I am sure that there were reasons, you wanted to clarify some issues around
controlling share mode flags. And it's rather everything OK with your code.

Even more: most probably your change revealed a bug in Win7, that has been
fixed by Microsoft in later Win editions.
Because I have no problems when accessing my Samba shares from Win10
computer, android devices or other linux systems.

My recommendation is as follows: after you double check your code, please
make it user-configurable - I mean a configuration parameter dedicated to
people using Win7-like versions in their networks, the parameter which by
default uses your enhancements, but may be switched off to bypass your code
in such special cases like mine.

I will be more than happy to help you in testing any change you decide to
apply.

Kind regards,
Pawel.


sob., 9 maj 2020 o 07:15 Andrew Bartlett <abartlet at samba.org>
napisa?(a):
> G'Day Volker and Jermey,
>
> Can you take a look at Pavlos's analysis and see if you can work out
> why he has been having trouble with Samba since this commit?
>
> Pavlos,
>
> That is an incredible effort, and so very, very much appreciated!
>
> Thanks,
>
> Andrew Bartlett
>
> On Sat, 2020-05-09 at 06:08 +0200, pavlos wrote:
> > Hi All,
> >
> > It's already 6am at my side. After the whole night of bisecting,
> compiling and testing I think I have nailed it:
> >
> > > fd61e550acec6e9924a71712eef739b58c7d68e8 is the first bad commit
> > > commit fd61e550acec6e9924a71712eef739b58c7d68e8
> > > Author: Volker Lendecke <vl at samba.org>
> > > Date:   Mon Sep 16 16:16:40 2019 -0700
> > >
> > >     smbd: Don't always walk the share mode array in
open_mode_check()
> > >
> > >     share_mode_data->flags contains the "most
restrictive" share mode
> of
> > >     the whole array. This is maintained lazily: Whenever
> set_share_mode()
> > >     is called, d->flags is updated if the new share mode got
more
> > >     restrictive. It is not updated when a file is closed, as this
would
> > >     mean we would have to walk the whole array, making sure that
the
> > >     closed handle was indeed the only most restrictive one.
Instead, we
> > >     walk the share mode array only when a conflict happens: Then
we
> need
> > >     to know "the truth" and recalculate it by walking
the share mode
> > >     array.
> > >
> > >     Signed-off-by: Volker Lendecke <vl at samba.org>
> > >     Reviewed-by: Jeremy Allison <jra at samba.org>
> > >
> > >     Autobuild-User(master): Jeremy Allison <jra at
samba.org>
> > >     Autobuild-Date(master): Wed Sep 18 00:07:13 UTC 2019 on
> sn-devel-184
> > >
> > >  source3/smbd/open.c | 217
> ++++++++++++++++++++++++++++++++++++++++++++++++----
> > >  1 file changed, 200 insertions(+), 17 deletions(-)
> >
> > Why do I think so?
> > Because:
> > I did a reset my current Samba master - my last clone was about 10
hours
> ago.
> > It is version 4.13.0pre1-GIT-e907f002a7f
> > Then I have carefuly edited the file 'source3/smbd/open.c' and
withdrawn
> all changes
> > Saved, compiled, installed, started services, tested different
scenarios
> and it works PROPERLY.
> > For the first time in any version greater that 4.11
> > Yes, it's magic :)
> >
> > Please, be so kind and check - everybody who suffers this bug, and, of
> course, the author of this change - and let me know whether my night was
> worth spending in front of my computer.
> >
> > Looking forward to hearing from you.
> > Pawel.
> >
> > ..
> >
> >
> > ?r., 6 maj 2020 o 00:06 pavlos <pavlos.audiofreak at gmail.com>
napisa?(a):
> > > Andrew, please forget about mu dumb question.
> > > Nice tutorial here: https://flaviocopes.com/git-bisect/
> > > I know how to proceed and what is the 'git clean...' for.
> > >
> > > Will come back :)
> > > Pawel
> > >
> > > wt., 5 maj 2020 o 23:43 pavlos <pavlos.audiofreak at
gmail.com>
> napisa?(a):
> > > > 'git checkout' in your first sentence- please be
more detailed.
> > > > And please remember where I start - I am just after 'git
clone' and
> build of 4.13.x
> > > >
> > > > Pawel
> > > >
> > > > wt., 5 maj 2020 o 23:28 Andrew Bartlett <abartlet at
samba.org>
> napisa?(a):
> > > > > On Tue, 2020-05-05 at 22:32 +0200, pavlos wrote:
> > > > > > Hi Andrew et al,
> > > > > >
> > > > > > Now is the time to come back to my leisure...
> > > > > >
> > > > > > Preparing the environment to the 'git
bisect' surgery:
> > > > > > Got a clone of the last VM with 4.12.2 installed
from
> > > > > >
https://download.samba.org/pub/samba/samba-latest.tar.gz
> > > > > > Stopped smb/nmb services and uninstalled 'sudo
make uninstall'
> within
> > > > > > the directory 4.12.2.
> > > > > > Got the current master with 'git clone
> > > > > > https://gitlab.com/samba-team/samba/' -
it's version is
> 4.13.0pre1.
> > > > > > From inside the directory: './confugure',
'make' and 'sudo make
> > > > > > install'. No issues.
> > > > > > Started smb/nmb services. No issues, as all
configuration was
> adapted
> > > > > > to different directories layout and missing
packages installed
> during
> > > > > > previous attempt.
> > > > > > # smbstatus --version
> > > > > > Version 4.13.0pre1-GIT-f80c97cb8da
> > > > > > Stubborn problems with multiple Win7 clients do
persist,
> > > > > > unfortunately.
> > > > > >
> > > > > > Now we have to move backwards till the last commit
when it was
> > > > > > working properly.
> > > > > > We already know that 4.12.x is hopeless and we
have to find last
> > > > > > working commit within 4.11.x branch.
> > > > > > For instance, the 4.11.3 - the last provided by
Arch packagers -
> was
> > > > > > working.
> > > > > > But there were many other patches published by
samba, the
> current one
> > > > > > is 4.11.9.
> > > > > >
> > > > > > Would it be possible to perform 'git
bisect' between different
> > > > > > branches?
> > > > >
> > > > > Yes, it can do that, in the git checkout.  Because of
the different
> > > > > branches it will do an extra at the point they diverge.
> > > > >
> > > > > > Please give me some instructions as I am new in
this school :)
> > > > >
> > > > > git bisect good samba-4.11.2
> > > > > (say yes to starting the bisect)
> > > > > git bisect bad samba-4.12.2
> > > > >
> > > > > Then it will move to various points it needs to test,
counting
> down the
> > > > > bisect revisions.
> > > > >
> > > > > Between each one, do a 'git clean -x -f -d
&& ./configure && make
> -j &&
> > > > > sudo make install'
> > > > >
> > > > > Finally, first install ccache if you can, it will be
much faster.
> > > > >
> > > > > Andrew Bartlett
> > > > >
> > > > > > Regards, Pawel.
> > > > > >
> > > > > > wt., 5 maj 2020 o 11:50 Andrew Bartlett
<abartlet at samba.org>
> > > > > > napisa?(a):
> > > > > > > On Tue, 2020-05-05 at 11:08 +0200, pavlos
wrote:
> > > > > > > > My plans are as follows, step by step:
> > > > > > > > - clone the VM, clean and build the
newest from gitlab, I
> can see
> > > > > > > it
> > > > > > > > is 4.13.0pre1
> > > > > > > > - clone the VM. clean and build the
latest supported 4.11.8
> > > > > > > > - create a new fresh VM with another
distro, not Arch or
> Arch-
> > > > > > > based,
> > > > > > > > build the official 4.12.2 - to
check/confirm that it is not
> > > > > > > distro-
> > > > > > > > related
> > > > > > >
> > > > > > > Thanks.  Even if it is distro-related, the
bisect between the
> two
> > > > > > > versions will be helpful.
> > > > > > >
> > > > > > > Andrew Bartlett
> > > > > > >
> > > > > > > > I will start playing in couple of hours,
now I am at work...
> > > > > > > >
> > > > > > > > Pawel.
> > > > > > > >
> > > > > > > > wt., 5 maj 2020 o 03:33 Andrew Bartlett
<abartlet at samba.org>
> > > > > > > > napisa?(a):
> > > > > > > > > On Tue, 2020-05-05 at 03:05 +0200,
pavlos wrote:
> > > > > > > > > > @Andrew
> > > > > > > > > > Unfortunately I didn't
make notes on exact time of errors
> > > > > > > when
> > > > > > > > > doing
> > > > > > > > > > my tests...
> > > > > > > > > >
> > > > > > > > > > But, I've spend last
couple of hours on building samba
> from
> > > > > > > > > master:
> > > > > > > > > > Have made a clone of the
virtual machine, with new name
> and
> > > > > > > > > network
> > > > > > > > > > interface mac address, and a
new dynamic IP assigned by
> my
> > > > > > > > > router.
> > > > > > > > > > Uninstalled Arch packages:
samba, smbclient and
> cifs-tools
> > > > > > > > > (depending
> > > > > > > > > > on smbclient)
> > > > > > > > > > As 'ala' downloaded
> > > > > > > > > >
https://download.samba.org/pub/samba/samba-latest.tar.gz
> and
> > > > > > > > > unpacked
> > > > > > > > > > it, obtaining
'samba-4.12.2' directory
> > > > > > > > >
> > > > > > > > > Thanks.  This gives you Samba
4.12.2.  However our
> development
> > > > > > > > > branch,
> > > > > > > > > master is newer.  Could you try
that?
> > > > > > > > >
> > > > > > > > > git clone
https://gitlab.com/samba-team/samba/
> > > > > > > > >
> > > > > > > > > That can be downloaded from
> > > > > > > > >
> > > > > > >
> > > > >
> https://gitlab.com/samba-team/samba/-/archive/master/samba-master.tar.bz2
> > > > > > > > >
> > > > > > > > > but if you use git, then you can
really help us with the
> > > > > > > bisect.
> > > > > > > > >
> > > > > > > > > But the rest of your steps look
good.
> > > > > > > > >
> > > > > > > > > The next step, assuming master
still fails, is to do a
> bisect.
> > > > > > > > > http://webchick.net/node/99
> > > > > > > > >
> > > > > > > > > You may need to run 'git clean
-x -f -d' between steps.
> > > > > > > > >
> > > > > > > > > This will be slow, particularly as
you can't automate it
> (need
> > > > > > > test
> > > > > > > > > with the client), but it is a very
good way to find what
> the
> > > > > > > issue
> > > > > > > > > is.
> > > > > > > > >
> > > > > > > > > Thanks,
> > > > > > > > >
> > > > > > > > > Andrew Bartlett
> > > > > > > > >
> --
> Andrew Bartlett                       http://samba.org/~abartlet/
> Authentication Developer, Samba Team  http://samba.org
> Samba Developer, Catalyst IT
> http://catalyst.net.nz/services/samba
>
>
>

On 09/05/2020 08:38, pavlos via samba wrote:
> Hi Volker, Jermey, and the Community,
>
> To be clear: I didn't analyze your code, I don't know structures of
Samba,
> definitely I am not a person to judge and blame anybody :)
> I am sure that there were reasons, you wanted to clarify some issues around
> controlling share mode flags. And it's rather everything OK with your
code.
>
> Even more: most probably your change revealed a bug in Win7, that has been
> fixed by Microsoft in later Win editions.
> Because I have no problems when accessing my Samba shares from Win10
> computer, android devices or other linux systems.
>
> My recommendation is as follows: after you double check your code, please
> make it user-configurable - I mean a configuration parameter dedicated to
> people using Win7-like versions in their networks, the parameter which by
> default uses your enhancements, but may be switched off to bypass your code
> in such special cases like mine.
>
> I will be more than happy to help you in testing any change you decide to
> apply.
>
> Kind regards,
> Pawel.
I cannot see making it user configurable for Window 7 happening, mainly 
because Windows 7 is EOL and Windows wants you to upgrade. Not that this 
has anything to do with us, but do we really want to fix this, when, by 
your own admission, it is fixed in supported versions of Windows ?

Rowland

On Sat, 2020-05-09 at 09:38 +0200, pavlos wrote:
> Hi Volker, Jermey, and the Community,
> 
> To be clear: I didn't analyze your code, I don't know structures of
Samba, definitely I am not a person to judge and blame anybody :)
> I am sure that there were reasons, you wanted to clarify some issues around
controlling share mode flags. And it's rather everything OK with your code.
G'Day Pavlos,

You appear to have identified a regression.  We take these very
seriously, even if only exposed by older clients.  

You have done us an incredible service in identifying the faulty patch.
It is much easier to fix what things once we know how they were broken.

I notice Ralph has picked this up, which is awesome, and while I can't
speak for his eventual patch, for something like this a smb.conf option
or client OS detection isn't normally required, we are pretty good at
just fixing it to always 'just work'.

But none of this would even be a conversation without your diligent
efforts, and I hope you can get a good sleep after such an exhausting
chase.

You have met the same standard a Samba developer would take, to bisect,
then revert and prove the issue goes away.  Don't doubt yourself!

Thank you.

Andrew Bartlett
-- 
Andrew Bartlett                       http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba

sob., 9 maj 2020 o 10:01 Rowland penny via samba <samba at
lists.samba.org>
napisa?(a):
> I cannot see making it user configurable for Window 7 happening, mainly
> because Windows 7 is EOL and Windows wants you to upgrade. Not that this
> has anything to do with us, but do we really want to fix this, when, by
> your own admission, it is fixed in supported versions of Windows ?
>
> Rowland
>
Thanks, Rowland.
Of course, fixing that "once and done" is always the best. Please
treat my
suggestions of placing a config parameter in smb.conf as a backup solution,
a workaround.

Regarding Windows versions - I have only Win7 32- and 64-bits and Win10
64-bits available at home. Only Win7 - both architectures - are suffering
because of this problem. Win10 as well as other operating systems work OK.
In other posts and forums I have also found some issues around 2008 Server
editions having troubles with Samba recently - they share the same code
with Win7.

I know that if this is a Windows bug - it will never be fixed in Win7 /
Win2008. But there is a lot of hardware with Win7 embedded, with no upgrade
path available. Why should I dispose a good and functioning equipment?
Especially when it is not exposed to the Internet, working in
private networks, where security vulnerability is not an issue at all?...

Regards, Pawel