Re: [uml-devel] Re: [uml-user] UNIX socket networking

[BlaisorBlade <bla...@ya...>]

Alle 12:56, domenica 7 marzo 2004, Patrick Kilian ha scritto:
> Hi all,
>
> > You refer to bind in the guest, right? Well, you call the
> > socket->ops->bind, while you should probably at least call sys_bind.
>
> Ok. I'll give it a try.
>
> > Your kernel thread is simply a process running in kernel mode - so
> > instead of calling bind(), you just call sys_bind() skipping some
> > useless code (int 0x80, userspace/kernelspace transition and so on).
> > But you cannot, in general, skip anything else unless you have a
> > reason to do so.
>
> I didn't know what the exact difference between bind()
Note also bind() is likely to be bound to the glibc version, and thus operate 
on the host.
>, sys_bind() and
> sockert->ops->bind() was.
>
> > Probably, the same reasoning applies to a lot of other code - why
> > sock_create instead of sys_socket, why sock_recvmsg  ... and so on.
> > Please, at least change both sys_socket and sys_bind before asking
> > again.
>
> Ok the next version will be updated.
>
> > I *do not* understand how ENOENT is actually returned, but probably
> > something that normally fails for a missing file failed for the lack
> > of coherency.
>
> Yeah that could be the reason.
>
> >> Most of the above code should probably be edited with rm
> >
> > What is that? Indentation cleanup?
>
> rm as in "man 1 rm".
Ah, ok :-) I don't agree anyway.

> >> , but you wanted to have my results so far. Now go and find a brown
> >> paper bag.
> >
> > No, do not - seeing the code is very useful indeed.
>
> If you say so...
> I'll give you sugestions a try and be back in some days with Yet Another
> Version of my code.
I've seen it; does it work on your part? I've got a lot less time now for UML, 
so you must be kind with me.
-- 
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729