AW: accept() failed message on kde 1.1.2 ??

["Ralf Habacker" <Ralf dot Habacker at saght dot tessag dot com>]

> Hi!
>
> Monday, 18 June, 2001 Robert Collins robert.collins@itdomain.com.au wrote:
>
> >> BTW, Ralf, are those sockets that fail in you test work in blocking or
> >> non-blocking mode?
>
Whats the default case in cygwin. There seems to be no code for setting
NONBLOCK or something else.

BTW: I have compiled a kdecore-dll (the dll which contains the kde socket
stuff) with and without stderr debug messages relating all socket stuff.
On running the debug version there is no error, running the other produces
the error.
For me this seems highly to be a timing problem. In this case accept() fail
with error code 113.
I try to find out more details.

> RC> Egor, I've been trying to grok the code. Can you tell me if I've got
> RC> this right?
>
> RC> on bind() you create a cookie and write it to the file. You
> also create
> RC> a event with a name related to the cookie.
> RC> on connect() you read the file and attempt to open the previously
> RC> created event. A failure to open is used to indicate failure.
>
> not quite. both peers (one that's accept()ing and one that's
> connect()ing)) open theirs own events, and wait until they're
> signalled. both also open their respective peer's event and signal it.
> see http://sources.redhat.com/ml/cygwin-apps/2001-06/msg00063.html for
> details.
>
> RC> I ask because I've been playing around with this on win2k
> sp1, ntsec on.
> RC> With XFree86 I get a "inode has changed" error on the
> /tmp/.X11-unix/foo
> RC> socket. Removing the .X11-unix directory fixes that. I suspect it's
> RC> related to problem Ralf is describing.
>
> do this problem persists if you run everything's with
> 'strace -b 65000 -f -o strace.log your_program.exe'
> ?
>
> if so, what does 'grep "unauthorized" strace.log'
> prints after that?
>
> RC> Anyway back to the AF_UNIX sockets - two questions:
> RC> 1) what stops someone using the NT namespace to see the
> available events
> RC> with a name format like the secure socket event names?
>
> you can't just "browse" all existing events, unless you're admin. to
> open event you have to know (or guess) its name.
>
> RC> 2) Why not allow the accept() to happen at the win32 level,
> but read and
> RC> write the cookie between cygwin, before letting the user program know
> RC> that cygwin is involved? It seems to me that that would be a somewhat
> RC> easier solution.
>
> i'm not sure i understand what you mean here.
>
> Egor.            mailto:deo@logos-m.ru ICQ 5165414 FidoNet 2:5020/496.19
>
>