thread safety of fhandler_base_overlapped

[Takashi Yano]

On Tue, 21 Dec 2021 08:58:13 +0900
Takashi Yano wrote:
> On Mon, 20 Dec 2021 19:02:56 -0400
> David McFarland wrote:
> > I'm seeing a problem where occasionally write() returns EAGAIN on a
> > blocking pipe. I believe this happens when multiple threads are writing
> > concurrently to the pipe, and the overlapped structure is modified by
> > one thread while another one is between calls to WFMO and
> > GetOverlappedResult.
> > 
> > I found this change:
> > 
> > =====
> > commit 2a4b1de773e6159ea8197b6fc3f7e4845479b476
> > Author: Corinna Vinschen <corinna@vinschen.de>
> > Date:   Mon Mar 23 21:06:00 2020 +0100
> > 
> >     Cygwin: serial: use per call OVERLAPPED structs
> > 
> >     Sharing the OVERLAPPED struct and event object in there between
> >     read and select calls in the fhandler might have been a nice
> >     optimization way back when, but it is a dangerous, not thread-safe
> >     approach.  Fix this by creating per-fhandler, per-call OVERLAPPED
> >     structs and event objects on demand.
> > 
> >     Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
> > =====
> > 
> > which seems to fix a similar problem in fhandler_serial.
> > 
> > Is is possible that this needs to be done for fhandler_base_overlapped
> > as well? I'll try using per-call buffers to see if it fixes my problem.
> 
> Pipe code was completely overhauled. Could you please try
> the latest developer snapshot?
> https://cygwin.com/snapshots/

Sorry, the new pipe code is also available in cygwin 3.3.3.
Do you have the problem in the cygwin 3.3.3?

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>