fix(auth): a member's clock running ahead of the register's is refused too
The full reading's L10: issueSessionAfter refused a member clock more than five seconds behind the register's, but one running ahead without bound, so sessions made on the member in the gap before a reset's moment outlived the register's notice. Now a clock that far ahead is refused alike (PasswordRefusal.passwordChanged: the reset is done, the device signs in again), with a warning to set both clocks by NTP. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
This commit is contained in:
parent
0119aacbc7
commit
8e6625855a
1 changed files with 73 additions and 6 deletions
|
|
@ -73,9 +73,10 @@ Future<AuthSuccess> issueSession(
|
|||
/// password or of the account): those here end first, as the register's
|
||||
/// notice to this server will end them, and the new one is made only once
|
||||
/// this server's clock is past that moment, so the notice, whenever it
|
||||
/// arrives, leaves it. A clock more than [maxClockSkew] behind is refused
|
||||
/// ([PasswordRefusal.passwordChanged]): the reset is done, and the device
|
||||
/// signs in again with the new password.
|
||||
/// arrives, leaves it. A clock more than [maxClockSkew] behind, or ahead
|
||||
/// (the full reading's L10: sessions made here in the gap would outlive the
|
||||
/// notice), is refused ([PasswordRefusal.passwordChanged]): the reset is
|
||||
/// done, and the device signs in again with the new password.
|
||||
Future<AuthSuccess> issueSessionAfter(
|
||||
Session session,
|
||||
UuidValue account, {
|
||||
|
|
@ -94,6 +95,18 @@ Future<AuthSuccess> issueSessionAfter(
|
|||
// The reset is done: the device signs in again, with the new password.
|
||||
throw PasswordRefusedException(reason: PasswordRefusal.passwordChanged);
|
||||
}
|
||||
// Ahead as far, sessions made here in the gap before the register's
|
||||
// moment would outlive its notice (the full reading's L10): refused alike.
|
||||
// (The register answered moments ago: its moment can't be further back
|
||||
// than that, but for a clock running ahead.)
|
||||
if (-wait > maxClockSkew) {
|
||||
session.log(
|
||||
"This server's clock is ${(-wait).inSeconds} s ahead of the "
|
||||
"register's: set both by NTP.",
|
||||
level: LogLevel.warning,
|
||||
);
|
||||
throw PasswordRefusedException(reason: PasswordRefusal.passwordChanged);
|
||||
}
|
||||
if (!wait.isNegative) {
|
||||
await Future<void>.delayed(wait + const Duration(milliseconds: 1));
|
||||
}
|
||||
|
|
@ -102,9 +115,9 @@ Future<AuthSuccess> issueSessionAfter(
|
|||
|
||||
/// How far behind the register's clock a member's may run for
|
||||
/// [issueSessionAfter] to wait for it: further, the request would hang as
|
||||
/// long, so it is refused. Clocks set by NTP differ by milliseconds; a
|
||||
/// member's clock running ahead of the register's by up to this much
|
||||
/// leaves sessions made in that gap, which the ADR accepts.
|
||||
/// long, so it is refused; as far ahead, alike. Clocks set by NTP differ by
|
||||
/// milliseconds; a member's clock running ahead of the register's by up to
|
||||
/// this much leaves sessions made in that gap, which the ADR accepts.
|
||||
const maxClockSkew = Duration(seconds: 5);
|
||||
|
||||
/// What [person] becomes here once a way in is theirs, as linking makes them
|
||||
|
|
|
|||
Loading…
Reference in a new issue