Re: [PATCH] Fix possible deadlock in stdio locking code

[Siddhesh Poyarekar <spoyarek at redhat dot com>]

On Fri, Nov 11, 2011 at 03:09:40AM -0600, Jonathan Nieder wrote:
> Siddhesh Poyarekar wrote:
> 
> > 	while(1) {
> > 		int child = fork();
> >
> > 		if (child == 0)
> > 			exit(0);
> 
> POSIX explains (under fork()):
> 
> 	If a multi-threaded process calls fork( ), the new process
> 	shall contain a replica of the calling thread and its entire
> 	address space, possibly including the states of mutexes and
> 	other resources. Consequently, to avoid errors, the child
> 	process may only execute async-signal-safe operations until
> 	such time as one of the exec functions is called.
> 
> exit() is not async-signal-safe, though _exit() is.  I don't think
> that glibc is the right tool to diagnose or work around application
> problems like this.  (Maybe helgrind could be tweaked to make the
> diagnosis.)

Thanks, the problem is not due to this though. I believe Andreas has
already fixed it; I was wrong (and careless) when I thought that it
could not have been fixed inside _IO_flush_all_lockp:


commit 7583a88d1c7170caad26966bcea8bfc2c92093ba
Author: Andreas Schwab <schwab@redhat.com>
Date:   Mon Nov 7 15:07:31 2011 +0100

    Fix locking in _IO_flush_all_lockp

diff --git a/libio/genops.c b/libio/genops.c
index 5d21c42..bb40c34 100644
--- a/libio/genops.c
+++ b/libio/genops.c
@@ -826,7 +826,7 @@ _IO_flush_all_lockp (int do_lock)
   int last_stamp;
 
 #ifdef _IO_MTSAFE_IO
-  _IO_cleanup_region_start_noarg (flush_cleanup);
+  __libc_cleanup_region_start (do_lock, flush_cleanup, 0);
   if (do_lock)
     _IO_lock_lock (list_all_lock);
 #endif
@@ -866,7 +866,7 @@ _IO_flush_all_lockp (int do_lock)
 #ifdef _IO_MTSAFE_IO
   if (do_lock)
     _IO_lock_unlock (list_all_lock);
-  _IO_cleanup_region_end (0);
+  __libc_cleanup_region_end (0);
 #endif
 
   return result;

--
Siddhesh