Kernel+LibC: Add adjtime(2)

Most systems (Linux, OpenBSD) adjust 0.5 ms per second, or 0.5 us per
1 ms tick. That is, the clock is sped up or slowed down by at most
0.05%.  This means adjusting the clock by 1 s takes 2000 s, and the
clock an be adjusted by at most 1.8 s per hour.

FreeBSD adjusts 5 ms per second if the remaining time adjustment is
>= 1 s (0.5%) , else it adjusts by 0.5 ms as well. This allows adjusting
by (almost) 18 s per hour.

Since Serenity OS can lose more than 22 s per hour (#3429), this
picks an adjustment rate up to 1% for now. This allows us to
adjust up to 36s per hour, which should be sufficient to adjust
the clock fast enough to keep up with how much time the clock
currently loses. Once we have a fancier NTP implementation that can
adjust tick rate in addition to offset, we can think about reducing
this.

adjtime is a bit old-school and most current POSIX-y OSs instead
implement adjtimex/ntp_adjtime, but a) we have to start somewhere
b) ntp_adjtime() is a fairly gnarly API. OpenBSD's adjfreq looks
like it might provide similar functionality with a nicer API. But
before worrying about all this, it's probably a good idea to get
to a place where the kernel APIs are (barely) good enough so that
we can write an ntp service, and once we have that we should write
a way to automatically evaluate how well it keeps the time adjusted,
and only then should we add improvements ot the adjustment mechanism.
This commit is contained in:
Nico Weber 2020-11-05 16:00:51 -05:00 committed by Andreas Kling
parent 28abfd6290
commit 323e727a4c
Notes: sideshowbarker 2024-07-19 01:27:53 +09:00
8 changed files with 94 additions and 2 deletions

View file

@ -0,0 +1,33 @@
## Name
adjtime - gradually adjust system clock
## Synopsis
```**c++
#include <sys/time.h>
int adjtime(const struct timeval* delta, struct timeval* old_delta);
```
## Description
`adjtime()` gradually increments the system time by `delta`, if it is non-null.
Serenity OS slows down or speeds up the system clock by at most 1%, so adjusting the time by N seconds takes 100 * n seconds to complete.
Calling `settimeofday()` or `clock_settime()` cancels in-progress time adjustments done by `adjtime`.
If `delta` is not null, `adjtime` can only called by the superuser.
If `old_delta` is not null, it returns the currently remaining time adjustment. Querying the remaining time adjustment does not need special permissions.
## Pledge
In pledged programs, the `settime` promise is required when `delta` is not null.
## Errors
* `EFAULT`: `delta` and/or `old_delta` are not null and not in readable memory.
* `EINVAL`: `delta` is not null and has a `tv_nsec` field that's less than 0 or larger or equal to `10^6`. Negative deltas should have a negative `tv_sec` field but a `tv_nsec` that's larger or equal zero. For example, a delta of -0.5 s is represented by `{-1, 500'000}`.
* `EPERM`: `delta` is not null but geteuid() is not 0.

View file

@ -193,7 +193,8 @@ namespace Kernel {
S(recvfd) \
S(sysconf) \
S(set_process_name) \
S(disown)
S(disown) \
S(adjtime)
namespace Syscall {

View file

@ -241,6 +241,7 @@ public:
int sys$fchdir(int fd);
int sys$sleep(unsigned seconds);
int sys$usleep(useconds_t usec);
int sys$adjtime(Userspace<const timeval*>, Userspace<timeval*>);
int sys$gettimeofday(Userspace<timeval*>);
int sys$clock_gettime(clockid_t, Userspace<timespec*>);
int sys$clock_settime(clockid_t, Userspace<const timespec*>);

View file

@ -24,6 +24,7 @@
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <AK/Time.h>
#include <Kernel/Process.h>
#include <Kernel/Time/TimeManagement.h>
@ -119,6 +120,35 @@ int Process::sys$clock_nanosleep(Userspace<const Syscall::SC_clock_nanosleep_par
}
}
int Process::sys$adjtime(Userspace<const timeval*> user_delta, Userspace<timeval*> user_old_delta)
{
if (user_old_delta) {
timespec old_delta_ts = TimeManagement::the().remaining_epoch_time_adjustment();
timeval old_delta;
timespec_to_timeval(old_delta_ts, old_delta);
if (!copy_to_user(user_old_delta, &old_delta))
return -EFAULT;
}
if (user_delta) {
REQUIRE_PROMISE(settime);
if (!is_superuser())
return -EPERM;
timeval delta;
if (!copy_from_user(&delta, user_delta))
return -EFAULT;
if (delta.tv_usec < 0 || delta.tv_usec >= 1'000'000)
return -EINVAL;
timespec delta_ts;
timeval_to_timespec(delta, delta_ts);
TimeManagement::the().set_remaining_epoch_time_adjustment(delta_ts);
}
return 0;
}
int Process::sys$gettimeofday(Userspace<timeval*> user_tv)
{
REQUIRE_PROMISE(stdio);

View file

@ -25,6 +25,7 @@
*/
#include <AK/Singleton.h>
#include <AK/StdLibExtras.h>
#include <AK/Time.h>
#include <Kernel/ACPI/Parser.h>
#include <Kernel/CommandLine.h>
@ -59,6 +60,7 @@ void TimeManagement::set_epoch_time(timespec ts)
{
InterruptDisabler disabler;
m_epoch_time = ts;
m_remaining_epoch_time_adjustment = { 0, 0 };
}
timespec TimeManagement::epoch_time() const
@ -256,7 +258,21 @@ void TimeManagement::increment_time_since_boot(const RegisterState&)
{
ASSERT(!m_time_keeper_timer.is_null());
timespec epoch_tick = { .tv_sec = 0, .tv_nsec = 1'000'000 }; // FIXME: Don't assume that one tick is 1 ms.
// Compute time adjustment for adjtime. Let the clock run up to 1% fast or slow.
// That way, adjtime can adjust up to 36 seconds per hour, without time getting very jumpy.
// Once we have a smarter NTP service that also adjusts the frequency instead of just slewing time, maybe we can lower this.
constexpr long NanosPerTick = 1'000'000; // FIXME: Don't assume that one tick is 1 ms.
constexpr time_t MaxSlewNanos = NanosPerTick / 100;
static_assert(MaxSlewNanos < NanosPerTick);
// Clamp twice, to make sure intermediate fits into a long.
long slew_nanos = clamp(clamp(m_remaining_epoch_time_adjustment.tv_sec, (time_t)-1, (time_t)1) * 1'000'000'000 + m_remaining_epoch_time_adjustment.tv_nsec, -MaxSlewNanos, MaxSlewNanos);
timespec slew_nanos_ts;
timespec_sub({ 0, slew_nanos }, { 0, 0 }, slew_nanos_ts); // Normalize tv_nsec to be positive.
timespec_sub(m_remaining_epoch_time_adjustment, slew_nanos_ts, m_remaining_epoch_time_adjustment);
timespec epoch_tick = { .tv_sec = 0, .tv_nsec = NanosPerTick };
epoch_tick.tv_nsec += slew_nanos; // No need for timespec_add(), guaranteed to be in range.
timespec_add(m_epoch_time, epoch_tick, m_epoch_time);
if (++m_ticks_this_second >= m_time_keeper_timer->ticks_per_second()) {

View file

@ -62,6 +62,9 @@ public:
static timeval now_as_timeval();
timespec remaining_epoch_time_adjustment() const { return m_remaining_epoch_time_adjustment; }
void set_remaining_epoch_time_adjustment(const timespec& adjustment) { m_remaining_epoch_time_adjustment = adjustment; }
private:
bool probe_and_set_legacy_hardware_timers();
bool probe_and_set_non_legacy_hardware_timers();
@ -73,6 +76,7 @@ private:
u32 m_ticks_this_second { 0 };
u32 m_seconds_since_boot { 0 };
timespec m_epoch_time { 0, 0 };
timespec m_remaining_epoch_time_adjustment { 0, 0 };
RefPtr<HardwareTimerBase> m_system_timer;
RefPtr<HardwareTimerBase> m_time_keeper_timer;
};

View file

@ -42,6 +42,7 @@ struct timezone {
int tz_dsttime;
};
int adjtime(const struct timeval* delta, struct timeval* old_delta);
int gettimeofday(struct timeval* __restrict__, void* __restrict__) __attribute__((nonnull(1)));
int settimeofday(struct timeval* __restrict__, void* __restrict__) __attribute__((nonnull(1)));

View file

@ -49,6 +49,12 @@ time_t time(time_t* tloc)
return tv.tv_sec;
}
int adjtime(const struct timeval* delta, struct timeval* old_delta)
{
int rc = syscall(SC_adjtime, delta, old_delta);
__RETURN_WITH_ERRNO(rc, rc, -1);
}
int gettimeofday(struct timeval* __restrict__ tv, void* __restrict__)
{
int rc = syscall(SC_gettimeofday, tv);