[systemd-devel] [PATCH] timedated: add LocalTimeUSec via dbus

[Kay Sievers]

On Tue, Mar 24, 2015 at 2:52 AM, Zbigniew Jędrzejewski-Szmek
<zbyszek at in.waw.pl> wrote:
> On Mon, Mar 23, 2015 at 04:24:38AM -0700, Shawn Landden wrote:
>> ---
>>  src/timedate/timedated.c | 24 ++++++++++++++++++++++++
>>  1 file changed, 24 insertions(+)
>>
>> diff --git a/src/timedate/timedated.c b/src/timedate/timedated.c
>> index ca771d5..f83b99c 100644
>> --- a/src/timedate/timedated.c
>> +++ b/src/timedate/timedated.c
>> @@ -38,6 +38,7 @@
>>  #include "bus-common-errors.h"
>>  #include "event-util.h"
>>  #include "selinux-util.h"
>> +#include "time-util.h"
>>
>>  #define NULL_ADJTIME_UTC "0.0 0 0\n0\nUTC\n"
>>  #define NULL_ADJTIME_LOCAL "0.0 0 0\n0\nLOCAL\n"
>> @@ -361,6 +362,28 @@ static int property_get_time(
>>          return sd_bus_message_append(reply, "t", now(CLOCK_REALTIME));
>>  }
>>
>> +/* gmtime(LocalTimeUSec % USEC_PER_SEC) is useful as long as you ignore GNU
>> + * extensions of tm_gmtoff and tm_zone, and their ourgrowth in strftime(3)
>> + */
>> +static int property_get_local_time(
>> +                sd_bus *bus,
>> +                const char *path,
>> +                const char *interface,
>> +                const char *property,
>> +                sd_bus_message *reply,
>> +                void *userdata,
>> +                sd_bus_error *error)
>> +{
>> +        struct tm tm;
>> +        struct timespec ts;
>> +        usec_t t;
>> +
>> +        t = now(CLOCK_REALTIME);
>> +        (void)timespec_store(&ts, t);
>> +        assert_se(localtime_r(&dummy, &ts.tv_sec));
>> +        return sd_bus_message_append(reply, "t", t + (tm->tm_gmtoff * USEC_PER_SEC));
>> +}
>> +
>>  static int property_get_ntp_sync(
>>                  sd_bus *bus,
>>                  const char *path,
>> @@ -671,6 +694,7 @@ static const sd_bus_vtable timedate_vtable[] = {
>>          SD_BUS_PROPERTY("NTP", "b", bus_property_get_bool, offsetof(Context, use_ntp), SD_BUS_VTABLE_PROPERTY_EMITS_CHANGE),
>>          SD_BUS_PROPERTY("NTPSynchronized", "b", property_get_ntp_sync, 0, 0),
>>          SD_BUS_PROPERTY("TimeUSec", "t", property_get_time, 0, 0),
>> +        SD_BUS_PROPERTY("LocalTimeUSec", "t", property_get_local_time, 0, 0),
>>          SD_BUS_PROPERTY("RTCTimeUSec", "t", property_get_rtc_time, 0, 0),
>>          SD_BUS_METHOD("SetTime", "xbb", NULL, method_set_time, SD_BUS_VTABLE_UNPRIVILEGED),
>>          SD_BUS_METHOD("SetTimezone", "sb", NULL, method_set_timezone, SD_BUS_VTABLE_UNPRIVILEGED)
>
> This patch seems the most reasonable of the proposed solutions.
> It answers the question "What does the remote host think local time is?"
> unambigously.
>
> The other patch proposed by Shawn mixes the remote UTC time with local
> information about timezones, which doesn't seem as clean.
>
> Kay said:
>> Transmitting several absolute time stamps sounds wrong.
> The second timestamp is what the user sees, so in some way it is a
> very significant thing. It is the outcome of the remote knowledge
> about the time and its config and timezone database. We aren't making
> any promises about the accuracy of that, but just trying to accurately
> convey that information. Seems fine to me to transmit this bit of
> information.

No, it is not fine, and we are not going to do that.

We do not transmit several timestamps, the local time on the remote
machine is UTC + its configured location. Tools who represent that
need to convert it into their presentation and not systemd.

Kay