A possible debugging improvement to talloc ...

[Richard Sharpe]

Hi folks,

When I was chasing down a memory problem recently, I noticed that
there is some spare bytes in the talloc_chunk that could contain
another pointer, and thought this would be interesting:

[rsharpe at sharpe tarballs]$ diff -u talloc-2.1.1/talloc.c
talloc-2.1.1-mod/talloc.c
--- talloc-2.1.1/talloc.c       2016-02-13 13:34:47.266100765 -0800
+++ talloc-2.1.1-mod/talloc.c   2016-02-11 18:00:18.858928133 -0800
@@ -273,6 +273,8 @@
         * from.
         */
        struct talloc_pool_hdr *pool;
+
+        const char *stolen_loc;
 };

 /* 16 byte alignment seems to keep everyone happy */
@@ -850,7 +852,7 @@
        return handle->ptr;
 }

-static void *_talloc_steal_internal(const void *new_ctx, const void *ptr);
+static void *_talloc_steal_internal(const void *new_ctx, const void
*ptr, const char *location);

 static inline void _talloc_free_poolmem(struct talloc_chunk *tc,
                                        const char *location)
@@ -1067,7 +1069,7 @@
    ptr on success, or NULL if it could not be transferred.
    passing NULL as ptr will always return NULL with no side effects.
 */
-static void *_talloc_steal_internal(const void *new_ctx, const void *ptr)
+static void *_talloc_steal_internal(const void *new_ctx, const void
*ptr, const char *location)
 {
        struct talloc_chunk *tc, *new_tc;
        size_t ctx_size = 0;
@@ -1082,6 +1084,11 @@

        tc = talloc_chunk_from_ptr(ptr);

+       /* Stamp the name regardless */
+       if (location) {
+               tc->stolen_loc = location;
+        }
+

There's more ... if it looks useful I will clean it up and post it out.

-- 
Regards,
Richard Sharpe
(何以解憂？唯有杜康。--曹操)