[PATCH] Use conn->session_info->security_token in posix_acls.c to make sysvolreset faster (was: Re: [PATCH] improve performance for samba-tool ntacl sysvolreset)

[Uri Simchoni]

On 07/10/2018 08:10 AM, Andrew Bartlett wrote:
> On Tue, 2018-07-10 at 07:49 +0300, Uri Simchoni wrote:
>> Hi,
<snip>
> 
>> Beside that I'm curious - it seems like the function we're optimizing
>> (uid_entry_in_group()) gets called in one of the following case:
>> 1. If the SD somehow doesn't translate into a POSIX ACL with a USER_OBJ
>> 2. To emulate deny ACE
> 
> Something like that.  I understand it is to fold any group permissions
> into the user permission because of the mismatch between NT and POSIX
> semantics. 
> 
>> Which one of the two gets called in the sysvolreset? (and if it's 1.,
>> why do we get an ACL without a USER_OBJ when we do a "reset" operation
>> which should bring things to the detault state)
> 
> It gets called a lot, I find this code very difficult to follow but
> seems to be needed for every group even if it isn't a DENY or such.
> 

OK I've dug a little deeper and can see that the ACL we're setting is
"O:LAG:BAD:P(A;OICI;0x001f01ff;;;BA)(A;OICI;0x001200a9;;;SO)(A;OICI;0x001f01ff;;;SY)(A;OICI;0x001200a9;;;AU)"
, which means that the owner (LA - basically an alias) doesn't have a
direct ACE. Being an alias, perhaps we should simply add a rule that if
the ACL has a BA ace (BUILTIN\Administrators), then we can use it for an
LA owner and construct the USER_OBJ ACE from that.

> Thanks for the thoughtful feedback.  I was a little worried this might
> not attract enough attention so I added a scarier subject :-)
> 

That certainly got my attention :)
> Andrew Bartlett
>