Fix for bz14333 -- race between atexit() and exit()

  Carlos,

Thanks for the review.

On Thu, Sep 14, 2017 at 8:07 AM, Carlos O'Donell <carlos@redhat.com> wrote:

> Suggest:
>
> /* See concurrency notes in stdlib/exit.h where this lock is defined.  */

Nit: the lock is defined here, but is declared in exit.h

>> diff --git a/stdlib/test-at_quick_exit-race.c b/stdlib/test-at_quick_exit-race.c
>
> Needs a comment explaining what this specific test is looking for and
> what are the expected results.

Rather than repeating such comment in every test-*exit*-race.c, I
added a note to look in test-atexit-race-common.c.

I also added a test for on_exit/exit -- I missed on_exit in previous iteration.

Thanks,

2017-09-18  Paul Pluzhnikov  <ppluzhnikov@google.com>
            Ricky Zhou  <rickyz@google.com>
            Anoop V Chakkalakkal  <anoop.vijayan@in.ibm.com>

        [BZ #14333]
        * stdlib/cxa_atexit.c (__internal_atexit): Use __exit_funcs_lock.
        Remove atomics.
        (__new_exitfn): Fail registration when we finished at_exit processing.
        * stdlib/cxa_finalize.c (__cxa_finalize): Likewise.
        * stdlib/on_exit.c (__on_exit): Likewise.
        * stdlib/exit.c (__exit_funcs_done): New variable.
        (__run_exit_handlers): Use __exit_funcs_lock.
        * stdlib/exit.h (__exit_funcs_lock, __exit_funcs_done): New
        declarations.
        * stdlib/Makefile (test-atexit-race, test-quick_at_exit-race)
        (test-cxa_atexit-race, test-on_exit-race): New tests.
        * stdlib/test-atexit-race-common.c: New file.
        * stdlib/test-atexit-race.c: New file.
        * stdlib/test-at_quick_exit-race.c: New file.
        * stdlib/test-cxa_atexit-race.c: New file.
        * stdlib/test-on_exit-race.c: New file.
  

On 09/18/2017 10:45 AM, Paul Pluzhnikov wrote:
> Carlos,
> 
> Thanks for the review.
> 
> On Thu, Sep 14, 2017 at 8:07 AM, Carlos O'Donell <carlos@redhat.com> wrote:
> 
>> Suggest:
>>
>> /* See concurrency notes in stdlib/exit.h where this lock is defined.  */
> Nit: the lock is defined here, but is declared in exit.h

OK.

>>> diff --git a/stdlib/test-at_quick_exit-race.c b/stdlib/test-at_quick_exit-race.c
>> Needs a comment explaining what this specific test is looking for and
>> what are the expected results.
> Rather than repeating such comment in every test-*exit*-race.c, I
> added a note to look in test-atexit-race-common.c.
> 
> I also added a test for on_exit/exit -- I missed on_exit in previous iteration.

I noticed something I didn't see on my other review.

Please see below.

> Thanks,
> 
> 2017-09-18  Paul Pluzhnikov  <ppluzhnikov@google.com>
>             Ricky Zhou  <rickyz@google.com>
>             Anoop V Chakkalakkal  <anoop.vijayan@in.ibm.com>
> 
>         [BZ #14333]
>         * stdlib/cxa_atexit.c (__internal_atexit): Use __exit_funcs_lock.
>         Remove atomics.
>         (__new_exitfn): Fail registration when we finished at_exit processing.
>         * stdlib/cxa_finalize.c (__cxa_finalize): Likewise.
>         * stdlib/on_exit.c (__on_exit): Likewise.
>         * stdlib/exit.c (__exit_funcs_done): New variable.
>         (__run_exit_handlers): Use __exit_funcs_lock.
>         * stdlib/exit.h (__exit_funcs_lock, __exit_funcs_done): New
>         declarations.
>         * stdlib/Makefile (test-atexit-race, test-quick_at_exit-race)
>         (test-cxa_atexit-race, test-on_exit-race): New tests.
>         * stdlib/test-atexit-race-common.c: New file.
>         * stdlib/test-atexit-race.c: New file.
>         * stdlib/test-at_quick_exit-race.c: New file.
>         * stdlib/test-cxa_atexit-race.c: New file.
>         * stdlib/test-on_exit-race.c: New file.
> 
> 
> 
> -- Paul Pluzhnikov
> 
> 
> glibc-bz14333-20170918.txt
> 
> 
> 2017-09-18  Paul Pluzhnikov  <ppluzhnikov@google.com>
>             Ricky Zhou  <rickyz@google.com>
>             Anoop V Chakkalakkal  <anoop.vijayan@in.ibm.com>
> 
> 	[BZ #14333]
>  	* stdlib/cxa_atexit.c (__internal_atexit): Use __exit_funcs_lock.
>  	Remove atomics.
>  	(__new_exitfn): Fail registration when we finished at_exit processing.
>  	* stdlib/cxa_finalize.c (__cxa_finalize): Likewise.
>  	* stdlib/on_exit.c (__on_exit): Likewise.
>  	* stdlib/exit.c (__exit_funcs_done): New variable.
>  	(__run_exit_handlers): Use __exit_funcs_lock.
>  	* stdlib/exit.h (__exit_funcs_lock, __exit_funcs_done): New
>  	declarations.
>  	* stdlib/Makefile (test-atexit-race, test-quick_at_exit-race)
>  	(test-cxa_atexit-race, test-on_exit-race): New tests.
>  	* stdlib/test-atexit-race-common.c: New file.
>  	* stdlib/test-atexit-race.c: New file.
>  	* stdlib/test-at_quick_exit-race.c: New file.
>  	* stdlib/test-cxa_atexit-race.c: New file.
>  	* stdlib/test-on_exit-race.c: New file.
> 
> 
> 
> diff --git a/stdlib/Makefile b/stdlib/Makefile
> index 2da39e067c..2fb08342e0 100644
> --- a/stdlib/Makefile
> +++ b/stdlib/Makefile
> @@ -81,7 +81,9 @@ tests		:= tst-strtol tst-strtod testmb testrand testsort testdiv   \
>  		   tst-quick_exit tst-thread-quick_exit tst-width	    \
>  		   tst-width-stdint tst-strfrom tst-strfrom-locale	    \
>  		   tst-getrandom tst-atexit tst-at_quick_exit 		    \
> -		   tst-cxa_atexit tst-on_exit
> +		   tst-cxa_atexit tst-on_exit test-atexit-race 		    \
> +		   test-at_quick_exit-race test-cxa_atexit-race             \
> +		   test-on_exit-race
>  
>  tests-internal	:= tst-strtod1i tst-strtod3 tst-strtod4 tst-strtod5i \
>  		   tst-tls-atexit tst-tls-atexit-nodelete
> @@ -91,6 +93,11 @@ ifeq ($(build-hardcoded-path-in-tests),yes)
>  tests += tst-empty-env
>  endif
>  
> +LDLIBS-test-atexit-race = $(shared-thread-library)
> +LDLIBS-test-at_quick_exit-race = $(shared-thread-library)
> +LDLIBS-test-cxa_atexit-race = $(shared-thread-library)
> +LDLIBS-test-on_exit-race = $(shared-thread-library)
> +
>  ifeq ($(have-cxx-thread_local),yes)
>  CFLAGS-tst-quick_exit.o = -std=c++11
>  LDLIBS-tst-quick_exit = -lstdc++
> diff --git a/stdlib/cxa_atexit.c b/stdlib/cxa_atexit.c
> index ce5d9f22b4..beb31691d5 100644
> --- a/stdlib/cxa_atexit.c
> +++ b/stdlib/cxa_atexit.c
> @@ -21,21 +21,29 @@
>  
>  #include <libc-lock.h>
>  #include "exit.h"
> -#include <atomic.h>
>  #include <sysdep.h>
>  
>  #undef __cxa_atexit
>  
> +/* See concurrency notes in stdlib/exit.h where this lock is declared.  */
> +__libc_lock_define_initialized (, __exit_funcs_lock)
> +
>  
>  int
>  attribute_hidden
>  __internal_atexit (void (*func) (void *), void *arg, void *d,
>  		   struct exit_function_list **listp)
>  {
> -  struct exit_function *new = __new_exitfn (listp);
> +  struct exit_function *new;
> +
> +  __libc_lock_lock (__exit_funcs_lock);
> +  new = __new_exitfn (listp);
>  
>    if (new == NULL)
> -    return -1;
> +    {
> +      __libc_lock_unlock (__exit_funcs_lock);
> +      return -1;
> +    }
>  
>  #ifdef PTR_MANGLE
>    PTR_MANGLE (func);
> @@ -43,8 +51,8 @@ __internal_atexit (void (*func) (void *), void *arg, void *d,
>    new->func.cxa.fn = (void (*) (void *, int)) func;
>    new->func.cxa.arg = arg;
>    new->func.cxa.dso_handle = d;
> -  atomic_write_barrier ();
>    new->flavor = ef_cxa;
> +  __libc_lock_unlock (__exit_funcs_lock);
>    return 0;
>  }
>  
> @@ -60,14 +68,11 @@ __cxa_atexit (void (*func) (void *), void *arg, void *d)
>  libc_hidden_def (__cxa_atexit)
>  
>  
> -/* We change global data, so we need locking.  */
> -__libc_lock_define_initialized (static, lock)
> -
> -
>  static struct exit_function_list initial;
>  struct exit_function_list *__exit_funcs = &initial;
>  uint64_t __new_exitfn_called;
>  
> +/* Must be called with __exit_funcs_lock held.  */
>  struct exit_function *
>  __new_exitfn (struct exit_function_list **listp)
>  {
> @@ -76,7 +81,10 @@ __new_exitfn (struct exit_function_list **listp)
>    struct exit_function *r = NULL;
>    size_t i = 0;
>  
> -  __libc_lock_lock (lock);
> +  if (__exit_funcs_done)
> +    /* Exit code is finished processing all registered exit functions,
> +       therefore we fail this registration.  */
> +    return NULL;
>  
>    for (l = *listp; l != NULL; p = l, l = l->next)
>      {
> @@ -127,7 +135,5 @@ __new_exitfn (struct exit_function_list **listp)
>        ++__new_exitfn_called;
>      }
>  
> -  __libc_lock_unlock (lock);
> -
>    return r;
>  }
> diff --git a/stdlib/cxa_finalize.c b/stdlib/cxa_finalize.c
> index aa0a70cb58..213acafe12 100644
> --- a/stdlib/cxa_finalize.c
> +++ b/stdlib/cxa_finalize.c
> @@ -17,7 +17,6 @@
>  
>  #include <assert.h>
>  #include <stdlib.h>
> -#include <atomic.h>
>  #include "exit.h"
>  #include <fork.h>
>  #include <sysdep.h>
> @@ -31,36 +30,36 @@ __cxa_finalize (void *d)
>  {
>    struct exit_function_list *funcs;
>  
> +  __libc_lock_lock (__exit_funcs_lock);
> +
>   restart:
>    for (funcs = __exit_funcs; funcs; funcs = funcs->next)
>      {
>        struct exit_function *f;
>  
>        for (f = &funcs->fns[funcs->idx - 1]; f >= &funcs->fns[0]; --f)
> -	{
> -	  void (*cxafn) (void *arg, int status);
> -	  void *cxaarg;
> +	if ((d == NULL || d == f->func.cxa.dso_handle) && f->flavor == ef_cxa)
> +	  {
> +	    const uint64_t check = __new_exitfn_called;
> +	    void (*cxafn) (void *arg, int status) = f->func.cxa.fn;
> +	    void *cxaarg = f->func.cxa.arg;
>  
> -	  if ((d == NULL || d == f->func.cxa.dso_handle)
> -	      /* We don't want to run this cleanup more than once.  */
> -	      && (cxafn = f->func.cxa.fn,
> -		  cxaarg = f->func.cxa.arg,
> -		  ! catomic_compare_and_exchange_bool_acq (&f->flavor, ef_free,
> -							   ef_cxa)))
> -	    {
> -	      uint64_t check = __new_exitfn_called;
> +	    /* We don't want to run this cleanup more than once.  */

We have just changed the way locking works, and the above comment
worries me, particularly for test coverage.

Under what conditions can this function be called more than once?

The C++ runtime is responsible for calling __cxa_finalize to call
the destructors for C++ functions.

The Itanium C++ ABI specifically says things like:
~~~
Multiple calls to __cxa_finalize shall not result in calling 
termination function entries multiple times; the implementation 
may either remove entries or mark them finished.
~~~

In theory perhaps if one thread is calling dlclose() while another
calls exit() we might have a case where the dlclose() has released
the list lock to call a function, then another thread calls exit()
and we might run the same function twice.

Could we amend the comment here to be more descriptive then?

/* We don't want to run this cleanup more than once. The Itanium
   C++ ABI requires that multiple calls to __cxa_finalize not
   result in calling termination functions more than once. One
   potential scenario where that could happen is with a concurrent
   dlclose and exit, where the running dlclose must at some point
   release the list lock, an exiting thread may acquire it, and
   without setting flavor to ef_free, might re-run this destructor
   which could result in undefined behaviour.  Therefore we must
   set flavor to ef_free to avoid calling this destructor again.
   Technically there is a race condition in this example, the thread
   calling dlclose may not have enough time to complete the execution
   of the recently called function before the other thread completes
   the exit processing and terminates the process.  */

Is it a bug that the thread calling dlclose may be the only thread
running this particular function while the other thread is running
to exit?

T1-> dlclose
T1-> library destructors call __cxa_finalize
T1-> picks function foo off the list, marks flavor ef_free
T1-> unlocks list, starts executing foo.
T2-> exit
T2-> starts executing all destructors, skips foo marked ef_free
T2-> proceeds to terminate the process

Is T1's call to foo incomplete?

... I remember having a discussion about exit() having to be
delayed indefinitely waiting for something, but I can't find our
libc-alpha conversation about this. It would seem like this would be
another case where exit() would have to wait for other in-process
termination functions?

> +	    f->flavor = ef_free;
>  
>  #ifdef PTR_DEMANGLE
> -	      PTR_DEMANGLE (cxafn);
> +	    PTR_DEMANGLE (cxafn);
>  #endif
> -	      cxafn (cxaarg, 0);
> +	    /* Unlock the list while we call foreign function.  */

Unlock the list while we call foreign functions.

or

Unlock the list while we call a foreign function.

> +	    __libc_lock_unlock (__exit_funcs_lock);
> +	    cxafn (cxaarg, 0);
> +	    __libc_lock_lock (__exit_funcs_lock);
>  
> -	      /* It is possible that that last exit function registered
> -		 more exit functions.  Start the loop over.  */
> -	      if (__glibc_unlikely (check != __new_exitfn_called))
> -		goto restart;
> -	    }
> -	}
> +	    /* It is possible that that last exit function registered
> +	       more exit functions.  Start the loop over.  */
> +	    if (__glibc_unlikely (check != __new_exitfn_called))
> +	      goto restart;
> +	  }
>      }
>  
>    /* Also remove the quick_exit handlers, but do not call them.  */
> @@ -79,4 +78,5 @@ __cxa_finalize (void *d)
>    if (d != NULL)
>      UNREGISTER_ATFORK (d);
>  #endif
> +  __libc_lock_unlock (__exit_funcs_lock);
>  }
> diff --git a/stdlib/exit.c b/stdlib/exit.c
> index c0b6d666c7..b18e252235 100644
> --- a/stdlib/exit.c
> +++ b/stdlib/exit.c
> @@ -19,11 +19,16 @@
>  #include <stdlib.h>
>  #include <unistd.h>
>  #include <sysdep.h>
> +#include <libc-lock.h>
>  #include "exit.h"
>  
>  #include "set-hooks.h"
>  DEFINE_HOOK (__libc_atexit, (void))
>  
> +/* Initialize the flag that indicates exit function processing
> +   is complete. See concurrency notes in stdlib/exit.h where
> +   __exit_funcs_lock is declared.  */
> +bool __exit_funcs_done = false;
>  
>  /* Call all functions registered with `atexit' and `on_exit',
>     in the reverse of the order in which they were registered
> @@ -44,14 +49,32 @@ __run_exit_handlers (int status, struct exit_function_list **listp,
>       the functions registered with `atexit' and `on_exit'. We call
>       everyone on the list and use the status value in the last
>       exit (). */
> -  while (*listp != NULL)
> +  while (true)
>      {
> -      struct exit_function_list *cur = *listp;
> +      struct exit_function_list *cur;
> +
> +      __libc_lock_lock (__exit_funcs_lock);
> +
> +    restart:
> +      cur = *listp;
> +
> +      if (cur == NULL)
> +	{
> +	  /* Exit processing complete.  We will not allow any more
> +	     atexit/on_exit registrations.  */
> +	  __exit_funcs_done = true;
> +	  __libc_lock_unlock (__exit_funcs_lock);
> +	  break;
> +	}
>  
>        while (cur->idx > 0)
>  	{
>  	  const struct exit_function *const f =
>  	    &cur->fns[--cur->idx];
> +	  const uint64_t new_exitfn_called = __new_exitfn_called;
> +
> +	  /* Unlock the list while we call foreign function.  */
> +	  __libc_lock_unlock (__exit_funcs_lock);
>  	  switch (f->flavor)
>  	    {
>  	      void (*atfct) (void);
> @@ -83,6 +106,13 @@ __run_exit_handlers (int status, struct exit_function_list **listp,
>  	      cxafct (f->func.cxa.arg, status);
>  	      break;
>  	    }
> +	  /* Re-lock again before looking at global state.  */
> +	  __libc_lock_lock (__exit_funcs_lock);
> +
> +	  if (__glibc_unlikely (new_exitfn_called != __new_exitfn_called))
> +	    /* The last exit function, or another thread, has registered
> +	       more exit functions.  Start the loop over.  */
> +	    goto restart;
>  	}
>  
>        *listp = cur->next;
> @@ -90,6 +120,8 @@ __run_exit_handlers (int status, struct exit_function_list **listp,
>  	/* Don't free the last element in the chain, this is the statically
>  	   allocate element.  */
>  	free (cur);
> +
> +      __libc_lock_unlock (__exit_funcs_lock);
>      }
>  
>    if (run_list_atexit)
> diff --git a/stdlib/exit.h b/stdlib/exit.h
> index 7f2e679246..dbf9f2d01f 100644
> --- a/stdlib/exit.h
> +++ b/stdlib/exit.h
> @@ -20,6 +20,7 @@
>  
>  #include <stdbool.h>
>  #include <stdint.h>
> +#include <libc-lock.h>
>  
>  enum
>  {
> @@ -57,11 +58,27 @@ struct exit_function_list
>      size_t idx;
>      struct exit_function fns[32];
>    };
> +
>  extern struct exit_function_list *__exit_funcs attribute_hidden;
>  extern struct exit_function_list *__quick_exit_funcs attribute_hidden;
> +extern uint64_t __new_exitfn_called attribute_hidden;
> +
> +/* True once all registered atexit/at_quick_exit/onexit handlers have been
> +   called */
> +extern bool __exit_funcs_done attribute_hidden;
> +
> +/* This lock protects __exit_funcs, __quick_exit_funcs, __exit_funcs_done
> +   and __new_exitfn_called globals against simultaneous access from
> +   atexit/on_exit/at_quick_exit in multiple threads, and also from
> +   simultaneous access while another thread is in the middle of calling
> +   exit handlers.  See BZ#14333.  Note: for lists, the entire list, and
> +   each associated entry in the list, is protected for all access by this
> +   lock.  */
> +__libc_lock_define (extern, __exit_funcs_lock);
> +
>  
>  extern struct exit_function *__new_exitfn (struct exit_function_list **listp);
> -extern uint64_t __new_exitfn_called attribute_hidden;
> +
>  
>  extern void __run_exit_handlers (int status,
>  				 struct exit_function_list **listp,
> diff --git a/stdlib/on_exit.c b/stdlib/on_exit.c
> index 83845e76d8..f4ede2b1a7 100644
> --- a/stdlib/on_exit.c
> +++ b/stdlib/on_exit.c
> @@ -17,25 +17,30 @@
>  
>  #include <stdlib.h>
>  #include "exit.h"
> -#include <atomic.h>
>  #include <sysdep.h>
>  
>  /* Register a function to be called by exit.  */
>  int
>  __on_exit (void (*func) (int status, void *arg), void *arg)
>  {
> -  struct exit_function *new = __new_exitfn (&__exit_funcs);
> +  struct exit_function *new;
> +
> +   __libc_lock_lock (__exit_funcs_lock);
> +  new = __new_exitfn (&__exit_funcs);
>  
>    if (new == NULL)
> -    return -1;
> +    {
> +      __libc_lock_unlock (__exit_funcs_lock);
> +      return -1;
> +    }
>  
>  #ifdef PTR_MANGLE
>    PTR_MANGLE (func);
>  #endif
>    new->func.on.fn = func;
>    new->func.on.arg = arg;
> -  atomic_write_barrier ();
>    new->flavor = ef_on;
> +  __libc_lock_unlock (__exit_funcs_lock);
>    return 0;
>  }
>  weak_alias (__on_exit, on_exit)
> diff --git a/stdlib/test-at_quick_exit-race.c b/stdlib/test-at_quick_exit-race.c
> new file mode 100644
> index 0000000000..f93fb852a2
> --- /dev/null
> +++ b/stdlib/test-at_quick_exit-race.c
> @@ -0,0 +1,32 @@
> +/* Bug 14333: a test for at_quick_exit/quick_exit race.
> +
> +   Copyright (C) 2017 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +/* This file must be run from within a directory called "stdlib".  */
> +
> +/* See stdlib/test-atexit-race-common.c for details on this test.  */
> +
> +#define CALL_ATEXIT at_quick_exit (&no_op)
> +#define CALL_EXIT quick_exit (0)
> +
> +static void
> +no_op (void)
> +{
> +}
> +
> +#include <stdlib/test-atexit-race-common.c>
> diff --git a/stdlib/test-atexit-race-common.c b/stdlib/test-atexit-race-common.c
> new file mode 100644
> index 0000000000..365c9d9c5a
> --- /dev/null
> +++ b/stdlib/test-atexit-race-common.c
> @@ -0,0 +1,70 @@
> +/* Bug 14333: Support file for atexit/exit, etc. race tests.
> +
> +   Copyright (C) 2017 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +/* This file must be run from within a directory called "stdlib".  */
> +
> +/* The atexit/exit, at_quick_exit/quick_exit, __cxa_atexit/exit, etc.
> +   exhibited data race while accessing destructor function list (Bug 14333).
> +
> +   This test spawns large number of threads, which all race to register
> +   large number of destructors.
> +
> +   Before the fix, running this test resulted in a SIGSEGV.
> +   After the fix, we expect clean process termination.  */
> +
> +#if !defined(CALL_EXIT) || !defined(CALL_ATEXIT)
> +#error Must define CALL_EXIT and CALL_ATEXIT before using this file.
> +#endif
> +
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <support/xthread.h>
> +
> +const size_t kNumThreads = 1024;
> +const size_t kNumHandlers = 1024;
> +
> +static void *
> +threadfunc (void *unused)
> +{
> +  size_t i;
> +  for (i = 0; i < kNumHandlers; ++i) {
> +    CALL_ATEXIT;
> +  }
> +  return NULL;
> +}
> +
> +static int
> +do_test (void)
> +{
> +  size_t i;
> +  pthread_attr_t attr;
> +
> +  xpthread_attr_init (&attr);
> +  xpthread_attr_setdetachstate (&attr, 1);
> +
> +  for (i = 0; i < kNumThreads; ++i) {
> +    xpthread_create (&attr, threadfunc, NULL);
> +  }
> +  xpthread_attr_destroy (&attr);
> +
> +  CALL_EXIT;
> +}
> +
> +#define TEST_FUNCTION do_test
> +#include <support/test-driver.c>
> diff --git a/stdlib/test-atexit-race.c b/stdlib/test-atexit-race.c
> new file mode 100644
> index 0000000000..a4df532ce5
> --- /dev/null
> +++ b/stdlib/test-atexit-race.c
> @@ -0,0 +1,32 @@
> +/* Bug 14333: a test for atexit/exit race.
> +
> +   Copyright (C) 2017 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +/* This file must be run from within a directory called "stdlib".  */
> +
> +/* See stdlib/test-atexit-race-common.c for details on this test.  */
> +
> +#define CALL_ATEXIT atexit (&no_op)
> +#define CALL_EXIT exit (0)
> +
> +static void
> +no_op (void)
> +{
> +}
> +
> +#include <stdlib/test-atexit-race-common.c>
> diff --git a/stdlib/test-cxa_atexit-race.c b/stdlib/test-cxa_atexit-race.c
> new file mode 100644
> index 0000000000..670f67538e
> --- /dev/null
> +++ b/stdlib/test-cxa_atexit-race.c
> @@ -0,0 +1,36 @@
> +/* Bug 14333: a test for __cxa_atexit/exit race.
> +
> +   Copyright (C) 2017 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +/* This file must be run from within a directory called "stdlib".  */
> +
> +/* See stdlib/test-atexit-race-common.c for details on this test.  */
> +
> +#include <stdio.h>
> +
> +#define CALL_ATEXIT __cxa_atexit (&no_op, NULL, NULL)
> +#define CALL_EXIT exit (0)
> +
> +int __cxa_atexit (void (*func) (void *), void *arg, void *d);
> +
> +static void
> +no_op (void *ignored)
> +{
> +}
> +
> +#include <stdlib/test-atexit-race-common.c>
> diff --git a/stdlib/test-on_exit-race.c b/stdlib/test-on_exit-race.c
> new file mode 100644
> index 0000000000..fce0fa7492
> --- /dev/null
> +++ b/stdlib/test-on_exit-race.c
> @@ -0,0 +1,32 @@
> +/* Bug 14333: a test for on_exit/exit race.
> +
> +   Copyright (C) 2017 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +/* This file must be run from within a directory called "stdlib".  */
> +
> +/* See stdlib/test-atexit-race-common.c for details on this test.  */
> +
> +#define CALL_ATEXIT on_exit (&no_op, (void *) 0)
> +#define CALL_EXIT exit (0)
> +
> +static void
> +no_op (int exit_code, void *ignored)
> +{
> +}
> +
> +#include <stdlib/test-atexit-race-common.c>
  

On Mon, Sep 18, 2017 at 2:15 PM, Carlos O'Donell <carlos@redhat.com> wrote:

>> -           uint64_t check = __new_exitfn_called;
>> +         /* We don't want to run this cleanup more than once.  */
>
> We have just changed the way locking works, and the above comment
> worries me, particularly for test coverage.
>
> Under what conditions can this function be called more than once?

Presumably the applicaton itself may call __cxa_finalize(NULL) from
multiple threads.

> Could we amend the comment here to be more descriptive then?

Sure.

> Is it a bug that the thread calling dlclose may be the only thread
> running this particular function while the other thread is running
> to exit?
>
> T1-> dlclose
> T1-> library destructors call __cxa_finalize
> T1-> picks function foo off the list, marks flavor ef_free
> T1-> unlocks list, starts executing foo.
> T2-> exit
> T2-> starts executing all destructors, skips foo marked ef_free
> T2-> proceeds to terminate the process
>
> Is T1's call to foo incomplete?

Yes. But an application that calls exit in parallel with running
threads always has the risk that any of its functions  will
"evaporate" mid-sentence.

Also, AFAICT this patch does not change the behavior here: the exact
same incomplete call to foo can happen with current code.

Thanks,
  

On 09/18/2017 05:03 PM, Paul Pluzhnikov wrote:
> On Mon, Sep 18, 2017 at 2:15 PM, Carlos O'Donell <carlos@redhat.com> wrote:
> 
>>> -           uint64_t check = __new_exitfn_called;
>>> +         /* We don't want to run this cleanup more than once.  */
>>
>> We have just changed the way locking works, and the above comment
>> worries me, particularly for test coverage.
>>
>> Under what conditions can this function be called more than once?
> 
> Presumably the applicaton itself may call __cxa_finalize(NULL) from
> multiple threads.
> 
>> Could we amend the comment here to be more descriptive then?
> 
> Sure.

I'll make my concrete suggestion below.

>> Is it a bug that the thread calling dlclose may be the only thread
>> running this particular function while the other thread is running
>> to exit?
>>
>> T1-> dlclose
>> T1-> library destructors call __cxa_finalize
>> T1-> picks function foo off the list, marks flavor ef_free
>> T1-> unlocks list, starts executing foo.
>> T2-> exit
>> T2-> starts executing all destructors, skips foo marked ef_free
>> T2-> proceeds to terminate the process
>>
>> Is T1's call to foo incomplete?
> 
> Yes. But an application that calls exit in parallel with running
> threads always has the risk that any of its functions  will
> "evaporate" mid-sentence.

Yes, and no.

Users expect exit() to run *all* of their registered exit functions.

When dlclose() and exit() interleave, you have the potential for
one function which is currently being run by dlclose() to be unable
to finish, and that's not expected. In fact that function runs
partly in parallel with the next registered function, and that could
be seen as a violation of POSIX requirements that functions run in
LIFO order.

I just had to test this out, so I wrote the following program:

#include <stdio.h>
#include <stdlib.h>
#include <dlfcn.h>
#include <semaphore.h>
#include <pthread.h>

sem_t order;

void *
open_library (char * pathname)
{
  void *dso;
  char *err;
  /* Open the DSO.  */
  dso = dlopen (pathname, RTLD_NOW|RTLD_GLOBAL);
  if (dso == NULL)
    {
      err = dlerror ();
      fprintf (stderr, "%s\n", err);
      exit (1);
    }
  /* Clear any errors.  */
  dlerror ();
  return dso;
}

int
close_library (void *dso)
{
  int ret;
  char *err;
  /* Close the library and look for errors too.  */
  ret = dlclose (dso);
  if (ret != 0)
    {
      err = dlerror ();
      fprintf (stderr, "%s\n", err);
      exit (1);
    }
  return ret;
}

void *
exit_thread (void *arg)
{
  /* Wait for the dlclose to start...  */
  sem_wait (&order);
  /* Then try to run the exit sequence which should call all
     __cxa_atexit registered functions and in parallel with
     the executing dlclose().  */
  exit (0);
}

int
main (void)
{
  void *dso;
  pthread_t thread;
  dso = open_library ("./libhas-dtors.so");
  pthread_create (&thread, NULL, exit_thread, NULL); 
  close_library (dso);
  pthread_join (thread, NULL);
  return 1;
}

#include <stdio.h>
#include <stdlib.h>
#include <semaphore.h>
#include <unistd.h>

/* Semaphore defined in executable to ensure we have
   a happens-before between the first function starting
   and exit being called.  */
extern sem_t order;

/* glibc function for registering DSO-specific exit functions.  */
extern int __cxa_atexit (void (*func) (void *), void *arg, void *dso_handle);

/* Hidden compiler handle to this shared object.  */
extern void *__dso_handle __attribute__ ((__weak__));

static void
first (void *start)
{
  sem_post (&order);
  sleep (10);
  printf ("first\n");
}

static void
second (void *start)
{
  printf ("second\n");
}


__attribute__ ((constructor)) static void
constructor (void)
{
  sem_init (&order, 0, 0);
  __cxa_atexit (second, NULL, __dso_handle);
  __cxa_atexit (first, NULL, __dso_handle);
}

gcc -O0 -g3 -shared -fPIC -o libhas-dtors.so has-dtors.c -lpthread
gcc -O0 -g3 -export-dynamic -o tst-dlclose-exit tst-dlclose-exit.c -lpthread -ldl

$ ./tst-dlclose-exit 
second
first
second

Which is just wrong, so it shows the existing implementation is broken.

The thread runs second very quickly, then first runs, then second runs
again.

It doesn't exit right away... and the trick is this:

* exit() must call _dl_fini, which needs the loader lock.
* dlclose() already holds the loader lock.
* therefore exit() blocks on the completion of dlclose().

So we are actually safe to finish running our existing handlers, but as
you see it runs second twice.

Can you run the above test case with your patch?

> Also, AFAICT this patch does not change the behavior here: the exact
> same incomplete call to foo can happen with current code.

OK, concrete suggestion per my previous email:

/* We don't want to run this cleanup more than once. The Itanium
   C++ ABI requires that multiple calls to __cxa_finalize not
   result in calling termination functions more than once. One
   potential scenario where that could happen is with a concurrent
   dlclose and exit, where the running dlclose must at some point
   release the list lock, an exiting thread may acquire it, and
   without setting flavor to ef_free, might re-run this destructor
   which could result in undefined behaviour.  Therefore we must
   set flavor to ef_free to avoid calling this destructor again.
   Note that the concurrent exit must also take the dynamic loader
   lock (for library finalizer processing) and therefore will 
   block while dlclose completes the processing of any in-progress
   exit functions. Lastly, once we release the list lock for the 
   entry marked ef_free, we must not read from that entry again 
   since it may have been reused by the time we take the list lock
   again. Lastly the detection of new registered exit functions is 
   based on a monotonically incrementing counter, and there is an 
   ABA if between the unlock to run the exit function and the 
   re-lock after completion the user registers 2^64 exit functions,
   the implementation will not detect this and continue without
   executing any more functions.
 
   One minor issue remains: A registered exit function that is in
   progress by a call to dlclose() may not completely finish before
   the next registered exit function is run. This may, according to
   some readings of POSIX violate the requirement that functions
   run in effective LIFO order. This should probably be fixed in a
   future implementation to ensure the functions do not run in
   parallel.  */

Before I sign off on this I'd like to see the results of your patch
running the example I provided above.

I expect it to print:
second
first

Thanks for wading through these issues.
  

On Tue, Sep 19, 2017 at 2:32 PM, Carlos O'Donell <carlos@redhat.com> wrote:

> Can you run the above test case with your patch?
...
> I expect it to print:
> second
> first

The bad news: it still prints:

second
first
second

The good news: there is a trivial fix: the loop in stdlib/exit.c must
*also* mark f->flavor as ef_free, and with that fix we get expected
output.

I'll add your test to the set of tests and send updated patch.

Thanks,
  

On 09/19/2017 04:33 PM, Paul Pluzhnikov wrote:
> On Tue, Sep 19, 2017 at 2:32 PM, Carlos O'Donell <carlos@redhat.com> wrote:
> 
>> Can you run the above test case with your patch?
> ...
>> I expect it to print:
>> second
>> first
> 
> The bad news: it still prints:
> 
> second
> first
> second
> 
> The good news: there is a trivial fix: the loop in stdlib/exit.c must
> *also* mark f->flavor as ef_free, and with that fix we get expected
> output.
> 
> I'll add your test to the set of tests and send updated patch.

My test has an implicit timing dependency. We wait 10 seconds to allow
the other thread to make progress against the exit() call, we should
change that to use another semaphore so it proceeds in a tick-tock
fashion without any timing requirement.