Rework handling of late HW TX timestamps. Instead of suspending reading
from client-only sockets that have HW TX timestamping enabled, save the
whole response if it is valid and a HW TX timestamp was received for the
source before. When the timestamp is received, or the configurable
timeout is reached, process the saved response again, but skip the
authentication test as the NTS code allows only one response per
request. Only one valid response per source can be saved. If a second
valid response is received while waiting for the timestamp, process both
responses immediately in the order they were received.
The main advantage of this approach is that it works on all sockets, i.e.
even in the symmetric mode and with NTP-over-PTP, and the kernel does
not need to buffer invalid responses.
Previously, in the calculation of the next transmission time
corresponding to the current polling interval, the reference point was
the current time in the client mode (i.e. the time when the response is
processed) and the last transmission time in the symmetric mode.
Rework the code to use the last transmission in both modes and make it
independent from the time when the response is processed to avoid extra
delays due to waiting for HW TX timestamps.
Update the serverstats response to use the new 64-bit integers.
Don't define a new value for the response as it already had an
incompatible change since the latest release (new fields added for
timestamp counters).
Add a structure for 64-bit integers without requiring 64-bit alignment
to be usable in CMD_Reply without struct packing.
Add utility functions for conversion to/from network order. Avoid using
be64toh() and htobe64() as they don't seem to be available on all
supported systems.
Count served timestamps in all combinations of RX/TX and
daemon/kernel/hardware. Repurpose CLG_LogAuthNtpRequest() to update all
NTP-specific stats in one call per accepted request and response.
After 5f4cbaab7e ("ntp: optimize detection of clients using
interleaved mode") the local TX timestamp is saved for all requests
indicating interleaved mode even when no previous RX timestamp is found.
Specify maxpoll for HW timestamping (default minpoll + 1) to track the
PHC well even when there is little NTP traffic on the interface. After
each PHC reading schedule a timeout according to the maxpoll. Polling
between minpoll and maxpoll is still triggered by HW timestamps.
Wait for the first HW timestamp before adding the timeout to avoid
polling PHCs on interfaces that are enabled in the configuration but
not used for NTP. Add a new scheduling class to separate polling of
different PHCs to avoid too long intervals between processing I/O
events.
In some cases even the new timeout of 1 millisecond is not sufficient to
get all HW TX timestamps. Add a new directive to allow users to
specify longer timeouts.
With some hardware it takes milliseconds to get the HW TX timestamp.
Rework the code to handle multiple suspended client-only sockets at the
same time in order to allow longer timeouts, which may overlap for
different sources. Instead of waiting for the first read event simply
suspend the socket and create timeout when the HW TX timestamp is
requested.
The Linux kernel (as of 6.2) has a shared queue of external timestamps
for all descriptors of the same PHC. If multiple refclocks using the
same PHC and the same or different channels were specified, some
refclocks didn't receive any or most of their timestamps, depending on
the rate and timing of the events (with the previous commit avoiding
blocking reads).
Track extpps-enabled refclocks in an array. Add PHC index to the PHC
instance. When a timestamp is read from the descriptor, provide it to
all refclocks that have the same PHC index and a channel matching the
event.
Make sure the timestamp is different from the previous one in case the
kernel will be improved to duplicate the timestamps for different
descriptors.
Reported-by: Matt Corallo <ntp-lists@mattcorallo.com>
The kernel has a common queue for all readers of a PHC device. With
multiple PHC refclocks using the same device some reads blocked. PHC
devices don't seem to support non-blocking reads. Use poll() to check if
a timestamp is available before reading from the descriptor.
Count missing samples for the median filter when
NAU_PrepareRequestAuth() is failing.
Fixes: 4234732b08 ("ntp: rework filter option to count missing samples")
Don't adjust the NTP polling interval and decrement the burst count when
NAU_PrepareRequestAuth() fails (e.g. no NTS-KE response received yet,
network being down, or the server refusing connections), same as if an
NTP request could not be sent. Rely on the rate limiting implemented in
the NTS code.
When chronyd configured with an NTS source not specified as offline and
resolvable without network was started before the network was up, it was
using an unnecessarily long NTS-KE retry interval, same as if the server
was refusing the connections.
When the network is down, the connect() call made from NKC_Start() on
the non-blocking TCP socket should fail with a different error than
EINPROGRESS and cause NKC_Start() to return with failure. Add a constant
2-second retry interval (matching default iburst) for this case.
When NKC_Start() fails (e.g. due to unreachable network), don't wait for
the next poll to destroy the client and another poll to create and start
it again.
In a non-tty session with chronyc it is not possible to detect the
end of the response without relying on timeouts, or separate responses
to a repeated command if using the -c option.
Add -e option to end each response with a line containing a single dot.
The sample time used in calculation of the last_meas_ago (LastRx) value
in the sources report is aligned to the second to minimize the leak
of the NTP receive timestamp, which could be useful in some attacks.
There is no need to do that with reference clocks, which are often used
with very short polling intervals and an extra second in the LastRx
value can be misinterpreted as a missed sample.
Log a warning message for each detected falseticker, but only once
between changes in the selection of the best source. Don't print all
sources when no majority is reached as that case has its own warning
message.
After dropping root privileges, log a warning message if chronyd
doesn't have read access or has (unnecessary) write access to the
files containing symmetric and server NTS keys.
Log a warning message if the file specified by the keyfile or
ntsserverkey directive is world-readable or writable, which is likely
an insecure misconfiguration. There is no check of directories
containing the file.
Split the new SOCK conditional using __GLIBC_PREREQ macro (which has
arguments) to fix compilation when it is not defined.
Fix also debug message using sizeof(time_t) in case it's enabled on
64-bit systems.
Reported-by: Bryan Christianson <bryan@whatroute.net>
Fixes: badaa83c31 ("refclock: convert mismatched timeval in SOCK messages")
On 32-bit glibc-based (>=2.34) systems, allow the SOCK client to send
messages with timevals using the other time_t size than chrony. If the
length of the received message corresponds to the other size, convert
the timeval and move the rest of the message before its processing.
This is needed for compatibility with the current development version of
gpsd, which forces 64-bit time_t on these systems, while chrony needs to
be compiled with the same time_t as gnutls.
The NTP SHM refclock protocol has the following properties:
- the memory segments have a predictable key (first segment 0x4e545030)
- it's expected to work in any order of starting chronyd and the program
providing samples to chronyd, i.e. both the consumer and producer need
to be able to create the segment
- the producer and consumer generally don't know under which user is
the other side running (e.g. gpsd can create the segment as root and
also as nobody after it drops root privileges)
- there is no authentication of data provided via SHM
- there is no way to restart the protocol
This makes it difficult for chronyd to ensure it is receiving
measurements from the process that the admin expects it to and not some
other process that managed to create the segment before it was started.
It's up to the admin to configure the system so that chronyd or the
producer is started before untrusted applications or users can create
the segment, or at least verify at some point later that the segment was
created with the expected owner and permissions.
There doesn't seem to be a backward-compatible fix of the protocol. Even
if one side could detect the segment had a wrong owner or permissions,
it wouldn't be able to tell the other side to reattach after recreating
the segment with the expected owner and permissions, if it still had the
permissions to do that.
The protocol would need to specify which side is responsible for
creating the segment and the start order would need to strictly follow
that.
As gpsd (likely the most common refclock source for chronyd) now
supports in the latest version SOCK even for message-based timing,
update the man page and FAQ to deprecate SHM in favor of SOCK.
