Re: [PATCH v5] Cygwin: pty: Fix race issue between starting and exiting non-cygwin apps

On Tue, 23 Jun 2026 23:24:23 +0900
Takashi Yano wrote:
> On Tue, 23 Jun 2026 21:29:25 +0900
> Takashi Yano wrote:
> > > I am a little concerned if the replacement WFSO is equivalent to the 
> > > looping WFSO being replaced.  I.e., it terminates for the same 
> > > condition(s) with the pty being in correct state.  I can't point to 
> > > something specific though.  Can you reassure me?  Or is this just 
> > > re-establishing code to the way it was before?
> > 
> > The code before the patch intended to leave wait-loop when pcon_start
> > mode is set even though the pipe_sw_mutex was not acquired. With this
> > patch, to_be_read_from_nat_pipe() is not called from master::write()
> > anymore, so the busy-loop is not necessary due to changes below.
> > 
> > @@ -2496,7 +2519,7 @@ fhandler_pty_master::write (const void *ptr, size_t len)
> > @@ -2580,20 +2603,6 @@ fhandler_pty_master::write (const void *ptr, size_t len)
> 
> Ah, I forgot to mention.
> 
> to_be_read_from_nat_pipe() is called in master::write() below, however,
> it is called only when pseudo console is not activated. In this case,
> the slave is never in setup_pseudoconsole(). So to_be_read_from_nat_pipe()
> can acquire pipe_sw_mutex after a short while.

No! I was wrong.
If master::write() is called when setup_pseudoconsole() is called
but the pcon_activated is not set yet, to_be_read_from_nat_pipe()
will be called while the slave holds pipe_sw_mutex.

I'll revert this change, and submit v7 patch.

-- 
Takashi Yano <[email protected]>