Re: [Nbd] nbd-client: bug in passing flags?

[Wouter Verhelst <w@...112...>]

Hi Alex,

On Mon, May 16, 2011 at 06:23:33PM +0100, Alex Bligh wrote:
> --On 16 May 2011 17:29:37 +0100 Alex Bligh <alex@...872...> wrote:
> 
> > I think the "<< 16" on line 215 is superfluous. The flags should
> > simply be converted to host order, and assigned to "flags", so
> > they will live in least significant two bytes (where they came
> > from).
> 
> In fact, unless lack of caffeine is getting to me, I don't see
> how these ever get passed to the kernel anyway. The read only
> flag, if it is set right, causes the BLKROSET ioctl to be
> executed.

Currently there are no other flags that are of importance to the kernel,
so indeed there's no way to inform the kernel of whatever it negotiated
during handshake.

> However, there is also a member of struct nbd_device in the
> kernel called flags. As far as I can tell, this is always zero,
> as it is never set to anything else. However, it is tested
> for NBD_READ_ONLY at line 460 of nbd.c in the kernel. Unless
> I'm missing something (quite possible) I don't think this
> check is ever activated.

Can't help you there; I don't do the kernel side of things, and indeed
am utterly unfamiliar with kernel hacking (you mean there's another side
to the syscall? whoa!)

> I need to pass the remainder of the flags anyway, so I proposed adding
> a NBD_SET_FLAGS ioctl to pass them through, and not checking for
> errors in the client (as older kernels won't support it).

Yeah, that's been suggested previously[1], and Paul seemed agreeable to
that. It hasn't been implemented yet, though, TTBOMK; I guess Paul is
busy with other things currently.

[1] http://sourceforge.net/mailarchive/forum.php?thread_name=20101229221020.GA32735%40celtic.nixsys.be&forum_name=nbd-general
    Okay, so we called it NBD_SET_OPTS. Whatever.

-- 
The volume of a pizza of thickness a and radius z can be described by
the following formula:

pi zz a