jpeach at samba.org
Sat Dec 22 03:44:32 GMT 2007
On 19/12/2007, at 11:14 PM, Volker Lendecke wrote:
> Attached find my current streams patch. It is far from
> finished, but in particular the VFS streaminfo call has an a
> bit different prototype. Can we somehow coordinate on that?
I think this is a really nice patch. I have a few comments, but I
think is a good improvement for streams support.
General streams support
- We should only claim support for FILE_NAMED_STREAMS in the
filesystem capabilities if the underlying filesystem supports it.
There should be a patch in my streams tree that copies FsCapabilities
from the FsCapabilities to the connection_struct at treeconnect time.
- I was a bit confused by build_stream_path(). Should the first
talloc_sprintf have a ':' between the base path and the stream name?
I'm not sure why you need the stream names if the VFS_STAT succeeds.
- Can you please ensure that stream name is canonicalised with a
leading ":"? The darwin_streams module needs this to distinguish
between xattrs that are xattrs that are streams.
Streams VFS module
- This looks a lot like fork depot <http://www.linux.org.au/conf/2006/abstract1f24.html?id=384
- At least SMB_VFS_OP_STREAMINFO should be opaque.
- You need to make sure that the attributes of the streams files are
the same as the attributes of the base file. At least the same
permissions and ACLs, etc. Timestamps you can fudge by taking them
from the base file.
- I'm not sure that this module should be called just "streams", maybe
"streams_depot". I think that there are so many different trade-offs
with streams implementations that it's difficult to imply that any one
is the canonical one.
James Peach | jpeach at samba.org
More information about the samba-technical