[PATCH] smbd: Fix snapshot query on shares with DFS enabled

[Christof Schmitt]

On Wed, Aug 17, 2016 at 08:52:31AM -0700, Jeremy Allison wrote:
> On Wed, Aug 17, 2016 at 10:12:45AM +0300, Uri Simchoni wrote:
> > > 
> > I've played with it some (haven't carefully reviewed it). Adding a patch
> > that runs shadow-copy2 tests using both SMB3 and NT1, and a test-patch
> > that puts the @GMT at the end of the path.
> > 
> > The SMB3 variant fails without your snapshot listing code, but passes
> > with it, no matter where the @GMT is, so I don't understand why you said
> > that path/@GMT-xxx fails (tested both path\@GMT and path/@GMT).
> 
> I was running against Windows, that's why :-). Windows needs TWrp.
> 
> > IMHO smbclient should place a TWrp in its allinfo query - that's what's
> > going to work against any SMB2 server. Perhaps we can also add smbclient
> > functionality to access previous versions (somewhat like cd - smbclient
> > keeps the dir locally and prepends it to paths. In the same way the TWrp
> > can be added to any open).
> 
> I've been thinking about that. I don't want the client code to arbitrarily
> scan SMB2 opens for @GMT-YYYY... paths and convert them into TWrp tokens,
> so I was thinking of a modal switch (yeah sucks but I can't think of any
> other way).
> 
> Like we have cli_set_backup_intent(), I think we need cli_set_previous_version(cli, time_t t).
> 
> If that is set, the SMB2 client code then scans for and removes
> any @GMT-YYYY... part of the path that contains the time_t t
> and sends a TWtp token containing t.
> 
> I'm trying to keep compatibility with the SMB1 path-based selection,
> as well as the SMB2 TWtp based selection, which is rather hard. The
> source3/client/client.c code can set a previous version time using
> cli_set_previous_version() as well as append the @GMT-YYYY.. to the
> path to cover both cases.
> 
> Setting cli_set_previous_version(cli, 0) removes any special casing
> in the SMB2 client SMB2create code.
> 
> 
> Thoughts ?

If cli_set_previous_version(cli, time_t t) already registers the
timestamp that should be used for opens, then the cli code could create
the twrp context from that for SMB2, or in the SMB1 case insert the @GMT
token in the path. That way the protocol differences between SMB1 and
SMB2 would be hidden from the application using the cli.

> > For the time being, I think we should apply your shadow_copy_data()
> > patches and not my patch that puts the @GMT at the end. That would make
> > "allinfo" properly list previous versions against all (reasonably
> > recent) versions of samba in SMB2 mode, at the expense of providing a
> > not-quite-realistic test for vfs_shadow_copy2.
> 
> I need to fix the source3/modules/vfs_snapper.c code first :-).
> 
> I'll send a patch to review soon !

Should we push the fixes you have for shadowcopy2, or do you want to wait
until everything is complete, including snapper and tests?

Christof