diff mbox series

__cxa_thread_atexit_impl: Return -1 on allocation failure

Message ID 20210728062522.3904733-1-siddhesh@sourceware.org
State New
Headers show
Series __cxa_thread_atexit_impl: Return -1 on allocation failure | expand

Commit Message

Siddhesh Poyarekar July 28, 2021, 6:25 a.m. UTC
Return -1 in the unlikely event that allocation fails when trying to
register a TLS destructor.
---
 stdlib/cxa_thread_atexit_impl.c | 3 +++
 1 file changed, 3 insertions(+)

Comments

Florian Weimer July 28, 2021, 7:14 a.m. UTC | #1
* Siddhesh Poyarekar via Libc-alpha:

> Return -1 in the unlikely event that allocation fails when trying to
> register a TLS destructor.

This is bug 18524.  As said on the bug, the interface is broken.  We
really need to rework C++ global destructor support so that these
dynamic memory allocations become unnecessary.

The GCC-generated caller code does not check the return value (and could
do anything with it anyway).  You probable need to add __libc_fatal.

Thanks,
Florian
Siddhesh Poyarekar July 28, 2021, 7:26 a.m. UTC | #2
On 7/28/21 12:44 PM, Florian Weimer wrote:
> * Siddhesh Poyarekar via Libc-alpha:
> 
>> Return -1 in the unlikely event that allocation fails when trying to
>> register a TLS destructor.
> 
> This is bug 18524.  As said on the bug, the interface is broken.  We
> really need to rework C++ global destructor support so that these
> dynamic memory allocations become unnecessary.
> 
> The GCC-generated caller code does not check the return value (and could
> do anything with it anyway).  You probable need to add __libc_fatal.

Hmm, so I started with __libc_fatal and changed it to return -1 when I 
saw that the libstdc++ compat implementation did the same.  Now I see 
that there's no check for return value from the call.  I'll update the 
patch with __libc_fatal and resend.

Thanks,
Siddhesh
Andreas Schwab July 28, 2021, 7:59 a.m. UTC | #3
On Jul 28 2021, Siddhesh Poyarekar via Libc-alpha wrote:

> Return -1 in the unlikely event that allocation fails when trying to
> register a TLS destructor.

Ok.

Andreas.
diff mbox series

Patch

diff --git a/stdlib/cxa_thread_atexit_impl.c b/stdlib/cxa_thread_atexit_impl.c
index 577ed30931..e956f4d804 100644
--- a/stdlib/cxa_thread_atexit_impl.c
+++ b/stdlib/cxa_thread_atexit_impl.c
@@ -72,6 +72,7 @@ 
    is not very different from a case where __call_tls_dtors is called after
    _dl_close_worker on the DSO and hence is an accepted execution.  */
 
+#include <stdio.h>
 #include <stdlib.h>
 #include <ldsodefs.h>
 
@@ -104,6 +105,8 @@  __cxa_thread_atexit_impl (dtor_func func, void *obj, void *dso_symbol)
 
   /* Prepend.  */
   struct dtor_list *new = calloc (1, sizeof (struct dtor_list));
+  if (__glibc_unlikely (new == NULL))
+    return -1;
   new->func = func;
   new->obj = obj;
   new->next = tls_dtor_list;