feat(vdisplay): harden keep-alive reconnect — same-client preempt, quit-skips-linger, configurable idle

On-glass testing (Test 2, KWin .116) surfaced that a reconnect within the QUIC idle-timeout
window (~8s) lands on a fresh SECOND display instead of reusing the kept one: the old session
was still Active (not yet Lingering), so the registry's keep-alive reuse (which only matches
Lingering) skipped it and the old session kept streaming to nobody. Three fixes:

#3 Same-client reconnect preempt (the real fix): admission::preempt_same_identity() lists a
   reconnecting client's OWN still-live session(s) (same cert fingerprint); serve_session signals
   their stop + waits the release grace BEFORE acquiring, so the zombie tears down → its display
   lingers → the reconnect REUSES it instead of making a second. Implements the "preempts
   downstream" the admission docs already promised. Independent of the mode_conflict policy; the
   pure core (same_identity_stops) is unit-tested.

#2 Deliberate quit skips linger: a client that deliberately disconnects closes the QUIC connection
   with QUIT_CLOSE_CODE (0x51, shared in core::quic); the host reads the ApplicationClosed reason
   and tears the display down immediately (registry release() gained force_immediate →
   Linger::Immediate; multi-session-safe via the pure lifecycle machine), while a bare disconnect
   still lingers for reconnect. Threaded via a session quit flag → the DisplayLease.
   NativeClient::disconnect_quit() + punktfunk-probe --quit drive it; GameStream (Quit App /
   h_cancel) is a documented follow-up.

#1 Configurable disconnect-detection latency: the QUIC control-connection idle timeout
   (stream_transport, 8s default) is host-tunable via --idle-timeout-ms / PUNKTFUNK_IDLE_TIMEOUT_MS,
   clamped >=1s with a keep-alive that scales to it so a live session never false-closes. Default
   unchanged (8s stays load-bearing for the Windows IDD-push reconnect flow).

Workspace check + 63 core / 215 host / 47 vdisplay tests green; clippy clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-07-05 16:41:06 +00:00
parent c1acfe8b85
commit b53710da1a
9 changed files with 272 additions and 38 deletions
@@ -115,6 +115,31 @@ pub fn admit(req_identity: Option<[u8; 32]>) -> Admission {
decide(effective_conflict(), req_identity, &table().lock().unwrap())
}
/// Pure core of [`preempt_same_identity`]: the stop flags of live sessions owned by the SAME client
/// as `req_identity` (its own zombies). Testable over a slice (the public fn locks the global table).
fn same_identity_stops(
req_identity: Option<[u8; 32]>,
live: &[LiveSession],
) -> Vec<Arc<AtomicBool>> {
live.iter()
.filter(|s| same_client(s.identity, req_identity))
.map(|s| Arc::clone(&s.stop))
.collect()
}
/// Preempt this reconnecting client's OWN still-live session(s). A client has at most one live
/// session, so a new connection from an already-registered identity is a **reconnect** — the old
/// session is a zombie whose QUIC idle timer hasn't fired yet (an unwanted disconnect is only
/// declared dead after `max_idle_timeout`, ~seconds later). Return its stop flag(s) so the caller
/// signals them and waits the release grace: the zombie tears its display down, which (keep-alive on)
/// lingers, and THIS reconnect **reuses** that kept display instead of landing on a fresh SECOND one
/// (the "thrown onto a second display while the old one keeps streaming" bug). Anonymous (`None`)
/// never matches — same limitation as `steal`/`reject`. Call this BEFORE [`admit`] and before this
/// session registers itself, so it only ever signals a *prior* session's flag, never its own.
pub fn preempt_same_identity(req_identity: Option<[u8; 32]>) -> Vec<Arc<AtomicBool>> {
same_identity_stops(req_identity, &table().lock().unwrap())
}
/// Register a now-admitted, live session; the returned guard removes it on drop (session end). Call
/// AFTER [`admit`] (so a session never conflicts with itself) and once the mode + stop flag are known.
pub fn register(
@@ -225,6 +250,20 @@ mod tests {
));
}
#[test]
fn same_identity_stops_targets_own_zombie_only() {
let live = [
sess(Some(1), (2560, 1440, 60)), // this client's prior (zombie) session
sess(Some(2), (1920, 1080, 60)), // a different client
];
// Reconnecting as client 1 → its own zombie's stop is returned (to preempt), not client 2's.
assert_eq!(same_identity_stops(fp(1), &live).len(), 1);
// A client with no prior session (fp 3) has nothing of its own to preempt.
assert_eq!(same_identity_stops(fp(3), &live).len(), 0);
// Anonymous never matches — we can't prove it's the same client.
assert_eq!(same_identity_stops(None, &live).len(), 0);
}
#[test]
fn join_targets_the_oldest_other_session() {
let live = [
+43 -14
View File
@@ -81,16 +81,23 @@ fn topology_str() -> String {
///
/// Windows delegates to the [`manager`](super::manager) via `vd.create` (unchanged); Linux uses the
/// pool below; other platforms pass through.
/// `quit` is the session's deliberate-quit flag: when the session ends with it set (the client closed
/// with the quit application code — a user "stop", not a network drop), the display is torn down
/// **immediately**, skipping the keep-alive linger. A bare disconnect leaves it `false` → normal linger.
pub fn acquire(
vd: &mut Box<dyn super::VirtualDisplay>,
mode: super::Mode,
quit: std::sync::Arc<std::sync::atomic::AtomicBool>,
) -> Result<super::VirtualOutput> {
#[cfg(target_os = "linux")]
{
linux::acquire(vd, mode)
linux::acquire(vd, mode, quit)
}
#[cfg(not(target_os = "linux"))]
{
// Windows leases in the manager (its own linger); the deliberate-quit skip is not wired
// through there yet, so the flag is accepted but unused off Linux.
let _ = quit;
vd.create(mode)
}
}
@@ -163,8 +170,8 @@ pub fn release(slot: Option<u64>) -> usize {
#[cfg(target_os = "linux")]
mod linux {
use std::sync::atomic::{AtomicU64, Ordering};
use std::sync::{Mutex, Once, OnceLock};
use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
use std::sync::{Arc, Mutex, Once, OnceLock};
use std::time::{Duration, Instant};
use anyhow::Result;
@@ -304,16 +311,21 @@ mod linux {
node_id: u32,
preferred_mode: Option<(u32, u32, u32)>,
gen: u64,
quit: Arc<AtomicBool>,
) -> VirtualOutput {
VirtualOutput {
node_id,
remote_fd: None,
preferred_mode,
keepalive: Box::new(DisplayLease { gen }),
keepalive: Box::new(DisplayLease { gen, quit }),
}
}
pub(super) fn acquire(vd: &mut Box<dyn VirtualDisplay>, mode: Mode) -> Result<VirtualOutput> {
pub(super) fn acquire(
vd: &mut Box<dyn VirtualDisplay>,
mode: Mode,
quit: Arc<AtomicBool>,
) -> Result<VirtualOutput> {
ensure_timer();
let backend = vd.name();
let r = reg();
@@ -343,7 +355,7 @@ mod linux {
e.life.acquire();
let gen = r.gen.fetch_add(1, Ordering::Relaxed);
e.gen = gen;
let out = output_for(e.node_id, e.preferred_mode, gen);
let out = output_for(e.node_id, e.preferred_mode, gen, quit);
tracing::info!(
backend,
node_id = e.node_id,
@@ -443,15 +455,21 @@ mod linux {
if (position.x, position.y) != (0, 0) {
vd.apply_position(position.x, position.y);
}
Ok(output_for(node_id, preferred_mode, gen))
Ok(output_for(node_id, preferred_mode, gen, quit))
}
/// The [`DisplayLease`] `Drop` path: release the session's hold on the pooled display. The
/// lifecycle machine decides linger / pin / teardown; a torn-down entry's keepalive drops *after*
/// the lock is released.
fn release(gen: u64) {
fn release(gen: u64, force_immediate: bool) {
let Some(r) = REG.get() else { return };
let linger = linger();
// A deliberate quit (the client closed with the quit code — a user "stop") tears the display
// down NOW, overriding the keep-alive linger; a bare disconnect honors the policy.
let linger = if force_immediate {
Linger::Immediate
} else {
linger()
};
let (torn_down, restore) = {
let mut es = r.entries.lock().unwrap();
let Some(idx) = es.iter().position(|e| e.gen == gen) else {
@@ -489,10 +507,17 @@ mod linux {
restore();
}
if let Some(e) = torn_down {
tracing::info!(
backend = e.backend,
"virtual display torn down (keep-alive off / released)"
);
if force_immediate {
tracing::info!(
backend = e.backend,
"virtual display torn down (deliberate quit — keep-alive skipped)"
);
} else {
tracing::info!(
backend = e.backend,
"virtual display torn down (keep-alive off / released)"
);
}
drop(e); // outside the lock — the keepalive Drop may block
}
}
@@ -683,11 +708,15 @@ mod linux {
/// registry hold; a stale lease (its entry was reused + re-stamped, or torn down) is a no-op.
struct DisplayLease {
gen: u64,
/// The session's deliberate-quit flag: set when the client closes with the quit application
/// code (a user "stop", not a network drop), so this lease's `Drop` tears the display down
/// immediately instead of lingering. `false` on a bare disconnect → normal keep-alive.
quit: Arc<AtomicBool>,
}
impl Drop for DisplayLease {
fn drop(&mut self) {
release(self.gen);
release(self.gen, self.quit.load(Ordering::SeqCst));
}
}