SEO

usr.bin/ssh/monitor_fdpass.c has two functions to pass file descriptors
between processes. These look similar to the functions in Stevens
UNIX Network Programming, Vol. 1, 2nd Ed. I have a question about
the size of the data and msg_controllen. monitor_fdpass.c uses:

mm_send_fd(int socket, int fd)
{
char tmp[CMSG_SPACE(sizeof(int))];

msg.msg_controllen = CMSG_LEN(sizeof(int));

mm_receive_fd(int socket)
{
char tmp[CMSG_SPACE(sizeof(int))];

msg.msg_controllen = sizeof(tmp);

while Stevens (Section 14.7) uses:

union {
struct cmsghdr cm;
char control[CMSG_SPACE(sizeof(int))];
} control_un;
msg.msg_controllen = sizeof(control_un.control);

in both cases. Moreover, today Wietse Venema sent out a patch claiming
that it is required for OpenBSD/Sparc64 (for send_fd and receive_fd):

* Adapted from: W. Richard Stevens, UNIX Network Programming, Volume 1,
+ * Second edition. Except that we use CMSG_LEN instead of CMSG_SPACE, for
+ * portability to LP64 environments.

union {
struct cmsghdr just_for_alignment;
- char control[CMSG_SPACE(sizeof(newfd))];
+ char control[CMSG_LEN(sizeof(newfd))];
} control_un;

- msg.msg_controllen = sizeof(control_un.control);
+ msg.msg_controllen = CMSG_LEN(sizeof(newfd)); /* Fix 200506 */

Unfortunately I don't have access to an OpenBSD/SPARC64 machine,
so I can't test this claim.

Question: what is the "correct" way to specify the sizes of the
buffers and msg_controllen (for the latter: why not simply use
sizeof(buffer)? ok, this would fail is someone changes the buffer
to char* and allocates it).