The cap_free() already does that, plus it has several sanity checks:

int cap_free(void *data_p)
{
    if ( !data_p )
        return 0;

    if ( good_cap_t(data_p) ) {
        data_p = -1 + (__u32 *) data_p;
        memset(data_p, 0, sizeof(__u32) + sizeof(struct _cap_struct));
        free(data_p);
        data_p = NULL;
        return 0;
    }

    if ( good_cap_string(data_p) ) {
        int length = strlen(data_p) + sizeof(__u32);
        data_p = -1 + (__u32 *) data_p;
        memset(data_p, 0, length);
        free(data_p);
        data_p = NULL;
        return 0;
    }

    _cap_debug("don't recognize what we're supposed to liberate");
    errno = EINVAL;
    return -1;
}

  This issue should not normally happen.

  I found it by accident while debugging the issue
that ended up with filling
https://bugzilla.redhat.com/show_bug.cgi?id=1410154
[crash due to not finding libnss_wins.so.2 dependencies in getaddrinfo]
and uses ping in one of the steps.

  I did some tests with ping from rpbmbuild/BUILD/...
and the problem still exists; it will read 4 bytes of
released memory due to calling cap_free again on a
released pointer.

first call it will be a good_cap_t(), so, it will
change the pointer to 4 bytes behind, where it
"hides" the state information, *and* release the
pointer, so, some new memory allocation can get
that same pointer at some point, and write to it;
the second call, will fail check of good_cap_t,
but only if the memory was not reallocated and
not overwritten, as it is a dangling pointer.

This looks like a bug in libcap to me. Do you have a reproducer?

  This is how I found the issue:

$ rpmbuild -ba SPECS/iputils.spec 
[...]
$ sudo valgrind BUILD/iputils-s20160308/ping redhat.com
==13693== Invalid read of size 4
==13693==    at 0x4E34779: cap_free (cap_alloc.c:119)
==13693==    by 0x405CF6: modify_capability (ping_common.c:166)
==13693==    by 0x402AAA: enable_capability_raw (ping.h:271)
==13693==    by 0x402AAA: main (ping.c:449)
==13693==  Address 0x65526e0 is 0 bytes inside a block of size 36 free'd
==13693==    at 0x4C28CDD: free (vg_replace_malloc.c:530)
==13693==    by 0x4E347FF: cap_free (cap_alloc.c:122)
==13693==    by 0x405D3C: modify_capability (ping_common.c:161)
==13693==    by 0x402AAA: enable_capability_raw (ping.h:271)
==13693==    by 0x402AAA: main (ping.c:449)
==13693==  Block was alloc'd at
==13693==    at 0x4C27BE3: malloc (vg_replace_malloc.c:299)
==13693==    by 0x4E3481F: cap_init (cap_alloc.c:19)
==13693==    by 0x4E34915: cap_get_proc (cap_proc.c:16)
==13693==    by 0x405CB0: modify_capability (ping_common.c:138)
==13693==    by 0x402AAA: enable_capability_raw (ping.h:271)
==13693==    by 0x402AAA: main (ping.c:449)
==13693== 
==13693== Invalid read of size 4
==13693==    at 0x4E34779: cap_free (cap_alloc.c:119)
==13693==    by 0x405CF6: modify_capability (ping_common.c:166)
==13693==    by 0x402AF0: disable_capability_raw (ping.h:272)
==13693==    by 0x402AF0: main (ping.c:454)
==13693==  Address 0x6552750 is 0 bytes inside a block of size 36 free'd
==13693==    at 0x4C28CDD: free (vg_replace_malloc.c:530)
==13693==    by 0x4E347FF: cap_free (cap_alloc.c:122)
==13693==    by 0x405D3C: modify_capability (ping_common.c:161)
==13693==    by 0x402AF0: disable_capability_raw (ping.h:272)
==13693==    by 0x402AF0: main (ping.c:454)
==13693==  Block was alloc'd at
==13693==    at 0x4C27BE3: malloc (vg_replace_malloc.c:299)
==13693==    by 0x4E3481F: cap_init (cap_alloc.c:19)
==13693==    by 0x4E34915: cap_get_proc (cap_proc.c:16)
==13693==    by 0x405CB0: modify_capability (ping_common.c:138)
==13693==    by 0x402AF0: disable_capability_raw (ping.h:272)
==13693==    by 0x402AF0: main (ping.c:454)
==13693== 
PING redhat.com (10.4.164.55) 56(84) bytes of data.
64 bytes from redirect-redhat-com.edge.prod.ext.phx2.redhat.com (10.4.164.55): icmp_seq=1 ttl=248 time=180 ms

--- redhat.com ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 1ms
rtt min/avg/max/mdev = 180.740/180.740/180.740/0.000 ms

  I did the above mostly because I did want to run the full
process under valgrind, and as well check loading of shared
objects in the getaddrinfo call.

  It would be a far more serious issue if this were handled
after libraries were loaded, as then memory would very likely
have been released.

  It is not a bug in libcap, because the second call to cap_free
is passing, again, a pointer it had already released, and the
suggested patch just sets the pointer to NULL to remember about
it, and prevent the second cap_free call.

https://github.com/iputils/iputils/pull/71

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHBA-2017:1987