[PATCH] Fix Gluster bug - unable to create file/directory under root of the share due to empty dirpath during get_real_filename()

[Michael Adam]

On 2018-08-22 at 09:27 -0700, Jeremy Allison wrote:
> On Wed, Aug 22, 2018 at 06:09:25PM +0200, Michael Adam wrote:
> > 
> > Conceptually, changing the return from "" to "."
> > and changing all existing callers that check for ""
> > afterwards is semantically not changing much really, is it?
> > 
> > So I don't get the big concern about this change.
> 
> Michael you have *no* idea :-).
> 
> Look into the statcache code (which is *really* old)
> which also gets passed the dirpath to make sure that
> is still working if you make this change.
> 
> Historically, working with the difference between
> "" and "." in pathnames has been the source of
> many, *many* Samba pathname bugs.
> 
> I have (a lot of) historical context that teaches
> me to be wary here. It you want to get really
> scared ask Volker about this stuff also :-).

Yeah, FUD helps convince me. :-)

It's ok. But cleaning up (later) the areas of code that are
so scaring is probably a good thing.

> This is why I ended up re-writing the patch
> to be a very simple fix after *really* careful
> review of the possible issues (and I'm still
> not sure I've gotten them all yet :-).
> 
> When I post my cleanup patch for you to review
> you can be sure I've checked out all the possibilities
> I can think of :-).

Looking forward to seeing it.

> > We might want to change that function so that the
> > callers do not actually have to deal with checking
> > the dirpath at all but encode it in a return code
> > or bool or so... Just rambling, sorry.
> > 
> > But I will buy the benefits of getting a more
> > isolated patch for the specific problem first.
> 
> Thanks. Already pushed mine :-).

Er. Where is the review+? (probably not on the list)
My mail was more of a review- until we sort out
this discussion.

> Your gluster-specific fix, whilst not wrong, isn't general enough.

Hm. You say it's a gluster-specific bug to be fixed.
And you fixed it in one layer above the call to
the gluster specific function. So yours is too
general (I'd say). But mine fixed it at the gluster level
and you say it's not general enough. How can that
be? There's exactly one call between the two places.

Please explain...

Thanks - Michael
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 163 bytes
Desc: not available
URL: <http://lists.samba.org/pipermail/samba-technical/attachments/20180822/5bd6edf3/signature.sig>