UTC behavior

This page outlines the behavior of the UTC clock on Fuchsia. See the UTC overview for more information on UTC time, and the time overview for information on the other clocks available on Fuchsia.

UTC time is supplied by a kernel clock object that is distributed to components. A process cannot read UTC time unless it is passed a handle to the clock when launched. All components started by Component Manager recieve a handle to the clock.

Timekeeper sets and maintains the UTC clock by synchronizing against either a Real Time Clock (RTC) or an external source generally accessible over the network.

Differences from other operating systems

UTC time on Fuchsia differs from time on most other operating systems in that it enforces a backstop time. The backstop time is set to the time of the latest commit included in a build, and serves as a “well known” time. Fuchsia will never report a time earlier than the backstop time, even if some time source reports an earlier time. This backstop is included to limit attacks where an adversary manipulates time on the device. As an example, such an attack could be used to force the device to accept an expired TLS certificate.

Before UTC time is first synchronized, the only estimate of UTC time available to Fuchsia is the backstop time. Backstop time by itself is not a reliable estimate of time as any amount of time may have passed between the time of the commit and the time the device is booted. To communicate this uncertainty, prior to the first time synchronization the UTC clock does not run and reading the clock returns the fixed backstop time.

In summary, the UTC clock on Fuchsia has two states. The UTC clock starts in a fixed state and ends in the running state.

Properties

The UTC clock always exhibits the following properties:

• Backstop time - the clock is created with a backstop time set to the time of the last commit in the build. The clock will never report a time earlier than the backstop.
• The UTC clock is neither monotonic nor continuous. As UTC time must be synchronized from an external source, Timekeeper may jump the time backwards if it finds that its estimate of UTC time has drifted far ahead of the external source.

In addition, the clock is in either a running or fixed state. The UTC clock is not started on creation and instead starts running when its time is set for the first time. Whether or not the clock is running serves as an indication as to whether or not the clock is synchronized.

Before Timekeeper first synchronizes time, the best estimate of UTC time available on the device is the backstop time. During this period, the clock is not running. Instead, any attempt to read the clock returns the backstop time. In this state, the device has no indication as to how much time has passed since the backstop. Therefore, it is important to note that during this state the UTC time may contain an arbitrarily large error.

After Timekeeper synchronizes the time, it sets the UTC clock. This starts running the clock. From this point on the clock continues to run, but Timekeeper will continue to update the clock by adjusting the clock frequency to run slightly faster or slower, or by jumping the clock to a new time. While accuracy varies between products and time synchronization methods, the clock is generally within a few hundred milliseconds of the actual UTC time once running. Note that even when an RTC, network, or both are available, the UTC clock may never be synchronized as it must be retrieved over fallible protocols from fallible sources.

Observable behaviors

As a result of the properties above, you may observe the following behaviors:

• UTC time may run up to a few hundred parts per million (ppm) faster or slower than monotonic time. This occurs as Timekeeper slews the clock slower or faster to compensate for oscillator errors or to correct small errors.
• UTC time may jump forwards or backwards an unbounded duration. This occurs if Timekeeper needs to correct for a large error. A large jump forward is expected when time is first synchronized. Subsequent jumps forwards and backwards should be very rare and are usually caused by errors in the time source.
• UTC time may not be running. This occurs prior to the first time synchronization. Note UTC time will never run under conditions where time synchronization will never succeed. For example, on a device with neither an RTC nor network access.

Strategies for handling the unsynchronized state

Components launched soon after a device boots, or that need to run before any network is available should expect to encounter situations where the UTC clock is not yet synchronized. In the rare case time synchronization never succeeds, components launched later will also see an unsynchronized clock. Some example strategies are:

• Ignore the unsynchronized state and read UTC time. This strategy is appropriate for cases where the UTC time doesn’t strictly need to be accurate, such as generating timestamps for debug purposes. It is also appropriate for cases where it is known a component will not be run until time is synchronized. The downside of this strategy is that you may see consecutive timestamps that all report the same time.
• Wait for UTC time to synchronize before reading the clock. This strategy is appropriate for cases where UTC time accuracy is critical, such as using UTC time to validate credentials. Note that this is not always the preferred strategy as UTC time may never synchronize.

Checking clock properties

You may check clock properties by first obtaining a handle to the zircon clock object, then passing the handle to the appropriate syscalls. As an example, this is useful if you need to check that time is synchronized before reading the clock. This is particularly important for applications such as TLS certificate validation as some reasonably accurate time is needed to verify expiry dates.

A handle to the UTC clock provided to the runtime is retrievable using the zx_utc_reference_get method provided in zircon/utc.h.

The ZX_CLOCK_STARTED signal is asserted when the clock is running (and therefore synchronized). You may check or wait for the signal using one of:

• zx_object_wait_async
• zx_object_wait_many
• zx_object_wait_one

You may check details such as the clock’s error bound using zx_clock_get_details. For the UTC clock the error bound is defined as half of a 95% confidence interval. In other words, for a randomly selected time on a randomly selected Fuchsia device, there is a ≥95% probability that the true value of UTC is between reported_utc - error_bound and reported_utc + error_bound. Until the system has an estimate of error_bound, the error_bound is set to ZX_CLOCK_UNKNOWN_ERROR. In some cases, if a Fuchsia device is running abnormal workloads or has defective hardware, the true UTC time may fall outside the range defined by error_bound. See the How can clock error be bounded? section for details on how the error bound is calculated. If you require additional details about the UTC clock, see the kernel clock reference for a list of details provided through zx_clock_get_details.

Note that components are provided a read-only handle and are unable to use the provided handle to modify the clock.

For language specific bindings and examples, see language support.