Patch to add support for advertising FULLSYNC to Mac OSX Clients

[Uri Simchoni]

On 04/07/2017 09:41 PM, Jeremy Allison via samba-technical wrote:
> On Fri, Apr 07, 2017 at 09:37:02PM +0300, Uri Simchoni wrote:
>> On 04/07/2017 09:04 PM, Jeremy Allison via samba-technical wrote:
>>> Can't we change this to a generic:
>>>
>>> bool VFS_QUERY_FEATURE(fsp, const char *vfs_module_name, const char *feature_name);
>>>
>>> or
>>>
>>> void *VFS_QUERY_FEATURE(fsp, const char *vfs_module_name, const char *feature_name);
>>>
>>> that can sweep aapl-specific flags/structs out of the main interfaces
>>> and back onto the module-specific private data struct where they
>>> belong ?
>>>
>> +1. My thinking of this has been:
>>
>> a. It's Apple-specific, let's hide it
>> b. ...But Macs are important too, let's not hide it (that's where I
>> submitted the file-id patch which adds an Apple-specific flag to the SMB
>> server).
>> c. ...But their version of the protocol is not documented and subject to
>> change without notice, and Microsoft's version currently is documented
>> and supported, so let's keep the main SMB code adhering to the
>> documented protocol version and (try to) hide undocumented dialects.
>>
>> Just my 2c...
> 
> Yeah, each individual decision made sense at the time,
> but I don't like where we ended up if you see what I
> mean.
> 
> Wouldn't a:
> 
> void *VFS_QUERY_FEATURE(fsp, const char *vfs_module_name, const char *feature_name);
> 
> interface allow us to at least clean things up a
> little, and maybe allow us to isolate some of the
> state back where it belongs ?
> 

I think I'm back to b) with this, as:
- There's no way to really hide this special flush behavior, as aapl
mode affects how you parse packets.
- Even if there was a way, the abstraction is wrong - you don't ask your
file system how to interpret packets.

So, with this flush I would have a *global* (smbd_server_connection)
flag that we're running AAPL, and maybe move AAPL parsing from vfs_fruit
to smb layer, make it a first citizen. Making it global serves two things:
1. Keeps files_struct from bloating
2. More correct as AAPL mode is global

For file index, I would *remove* the global aapl_zero_file_id flag, and
hide *that* in the VFS (SMB_VFS_GET_FILE_ID), because it's a good
abstraction with potential uses elsewhere.

So with this, I don't yet see a use case for VFS_QUERY_FEATURE()...

Uri