On Sun, Aug 15, 2010 at 09:16:32AM +0200, Matthieu Herrb wrote:
> On Sun, Aug 15, 2010 at 2:32 AM, Daniel Ouellet <dan...@presscom.net> wrote:
> > In trying the latest snapshots on sparc64 servers, so far 3 of them all give
> > me the same errors. I can't use the -n flag anymore with rdate as before.
> >
> > It worked on 4.7 and before. Not sure of the exact date it stop working.
> >
> > Any suggestions?
> >
> Probably an aligment problem.
>
> Try this patch:
>
> Index: ntp.c
> ===================================================================
> RCS file: /cvs/src/usr.sbin/rdate/ntp.c,v
> retrieving revision 1.29
> diff -u -r1.29 ntp.c
> --- ntp.c 17 Sep 2006 17:03:56 -0000 1.29
> +++ ntp.c 15 Aug 2010 07:12:11 -0000
> @@ -305,7 +305,8 @@
> int
> read_packet(int fd, struct ntp_data *data, double *off, double *error)
> {
> - u_char receive[NTP_PACKET_MAX];
> + u_int64_t receivebuf[howmany(NTP_PACKET_MAX, sizeof(u_int64_t))];
> + u_char *receive = (u_char *)receivebuf;;
> struct timeval tv;
> double x, y;
> int length, r;
>
>
> --
> Matthieu Herrb
There may be a similar potential issue in write_packet. Here I use
bcopy() rather than aligning the buffer on the stack. Not sure what
the preferred idiom is. This way it will continue to work if
recvck/xmitck ever become u_int128_t. :-).
.... Ken
Index: ntp.c
===================================================================
RCS file: /cvs/src/usr.sbin/rdate/ntp.c,v
retrieving revision 1.29
diff -u -p -r1.29 ntp.c
--- ntp.c 17 Sep 2006 17:03:56 -0000 1.29
+++ ntp.c 15 Aug 2010 11:01:55 -0000
@@ -280,9 +280,11 @@ write_packet(int fd, struct ntp_data *da
* No endian concerns here. Since we're running as a strict
* unicast client, we don't have to worry about anyone else finding
* the transmit field intelligible.
+ *
+ * NOTE: packet is not necessarily aligned, so a (u_int64_t *) would
+ * be bad on aligned architectures!
*/
-
- *(u_int64_t *)(packet + NTP_TRANSMIT) = data->xmitck;
+ bcopy(&data->xmitck, (packet + NTP_TRANSMIT), sizeof(data->xmitck));
data->originate = current_time(JAN_1970);
@@ -423,8 +425,13 @@ unpack_ntp(struct ntp_data *data, u_char
data->transmit = d / NTP_SCALE;
- /* See write_packet for why this isn't an endian problem. */
- data->recvck = *(u_int64_t *)(packet + NTP_ORIGINATE);
+ /*
+ * See write_packet for why this isn't an endian problem.
+ *
+ * NOTE: packet is not necessarily aligned, so a (u_int64_t *) would
+ * be bad on aligned architectures!
+ */
+ bcopy((packet + NTP_ORIGINATE), &data->recvck, sizeof(data->recvck));
}
/*
