New subject: [Bug 33656] NamedPipe datagrams need to be _really_ datagrams

25 May 2013


      http://bugs.winehq.org/show_bug.cgi?id=33656
Bug #: 33656
           Summary: NamedPipe datagrams need to be _really_ datagrams
           Product: Wine
           Version: unspecified
          Platform: Other
        OS/Version: other
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
        AssignedTo: wine-bugs@winehq.org
        ReportedBy: joe_rocks@hotmail.com
                CC: adam.r.martinson@gmail.com, adys.wh@gmail.com,
                    asb@asbradbury.org, berillions@gmail.com,
                    dank@kegel.com, davilando@gmail.com,
                    djelinski1@gmail.com, ernst@planetcoms.co.za,
                    euug3576+winebugs@gmail.com, focht@gmx.net,
                    grinnz@gmail.com, jjmckenzie51@gmail.com,
                    jonnybarnes@gmail.com, knaprigt@gmail.com,
                    kyle.devir@gmx.com, lkcl@lkcl.net,
                    l_bratch@yahoo.co.uk, m.b.lankhorst@gmail.com,
                    mooroon2@mail.ru, napoliste@gmail.com,
                    rmlipman@gmail.com, saulius2@gmail.com,
                    surfing86@live.it, techtonik@gmail.com,
                    trilavia@gmail.com, update.microsoft@mail.ru,
                    valentyn.pavliuchenko@gmail.com, vegarh@gmail.com
    Classification: Unclassified
+++ This bug was initially created as a clone of Bug #17195 +++
ok, after a little bit of investigation, i think i understand the pipes code
enough to be able to say what's going on, and i'm seeing something like this:
process 1:
recvpipe = CreateNamedPipe("\pipe\fred")
ReadFile(recvpipe, buffer, &length);
printf(length) ===> 43
ReadPipe(recvpipe, buffer, &length); /* no data */
process 2:
sendpipe = CreateNamedPipe("\pipe\fred")
length = 9;
WriteFile(sendpipe, buffer, &length);
length = 34;
WriteFile(sendpipe, buffer, &length);
what's happening is that the data being sent down the pipes isn't being done as
datagrams.  the implementation is using a stream-based fd.
the solution is: you _must_ implement a protocol on top of the pipes which
sends the length (as a 32-bit int, whatever) which is read off the fd, followed
by the data stream of _exactly_ that length.
_must_.  there's no two ways about this.
the protocol of Pipes is unfortunately a combination of both datagrams and
streams.  datagrams because the lengths of data sent are absolute and
inviolate.  streams because the data order and reliability are _also_ absolute
and inviolate.
you can't use datagrams (because they're unreliable).  you can't expect all
unixen to support datagrams on top of unix sockets (if that's what's being
used).
so - you have to send the length, as part of the implementation of the
pipe-data-send.
sending a length will also solve the issue of trying to send zero-length pipe
datagrams.
as a first implementation, you _might_ be able to get away with assuming that
when someone asks for some data, they _will_ provide a buffer big enough.
... actually... i don't see any ERR_MORE_DATA error codes in NtReadFile, so
that would explain... this is going to get messy :)
-- 
Configure bugmail: http://bugs.winehq.org/userprefs.cgi?tab=email
Do not reply to this email, post in Bugzilla using the
above URL to reply.
------- You are receiving this mail because: -------
You are watching all bug changes.

[Bug 33656] New: NamedPipe datagrams need to be _really_ datagrams