[PATCHES] vfs_shadow_copy2: more tests and fix of a bug with SMB1

[Jeremy Allison]

On Wed, Aug 24, 2016 at 09:08:48PM +0300, Uri Simchoni wrote:
> Hi,
> 
> The attached patch set (some of which I already sent yesterday) adds
> some more tests to shadow_copy2 module and then fixes a bug that was
> discovered by one of those tests, namely that snapshot folders cannot be
> listed using SMB1.
> 
> The fix may be controversial. The cause of the bug is that when listing
> files with SMB1, call_trans2findfirst() is given a path of
> path\to\dir\*, as this is what goes over the wire, and the
> call_trans2findfirst() passes this path (after minimum SMB->unix
> conversion) to some VFS functions (lstat, connectpath, realpath - via
> check_reduced_name()). With the trailing "*". The connectpath VFS
> function of shadow_copy2 overrides default behavior and fails with this
> trailing "*".
> 
> The obvious fix seems to be to strip the "*" before going to the VFS,
> because the VFS layer doesn't expect those things.

Actually that's not true. call_trans2findfirst() passes the given
name ultimately to filename_convert_internal() which takes the
flags:

UCF_COND_ALLOW_WCARD_LCOMP
UCF_ALWAYS_ALLOW_WCARD_LCOMP

which explicitly allow a wildcard as the last component.

The default vfs copes with these things fine, so I'm
guessing this is something broken in shadow_copy2, which
is where you're doing the fix. So I'd argue you're fixing
the right place.

Haven't reviewed your code yet, but just wanted to correct
the misapprehension above.