Implement performance monitor

Cargo.lock

@@ -3197,13 +3197,14 @@ checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
 
 [[package]]
 name = "hello-agent"
-version = "0.1.3"
+version = "0.1.7"
 dependencies = [
  "anyhow",
  "env_logger 0.10.2",
  "hbb_common",
  "log",
  "rustdesk",
+ "serde 1.0.228",
  "serde_json 1.0.118",
  "tokio",
  "winapi",

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "hello-agent"
-version = "0.1.3"
+version = "0.1.7"
 edition = "2021"
 rust-version = "1.75"
 description = "Headless RustDesk-protocol-compatible support agent for Windows"
@@ -24,7 +24,7 @@ path = "src/main.rs"
 librustdesk = { package = "rustdesk", path = "vendor/rustdesk", default-features = false, features = ["use_dasp", "hwcodec", "vram"] }
 hbb_common  = { path = "vendor/rustdesk/libs/hbb_common" }
 
-tokio       = { version = "1", features = ["rt-multi-thread", "macros", "sync", "time", "io-util"] }
+tokio       = { version = "1", features = ["rt-multi-thread", "macros", "sync", "time", "io-util", "process"] }
 log         = "0.4"
 env_logger  = "0.10"
 anyhow      = "1"
@@ -34,6 +34,12 @@ anyhow      = "1"
 # the inventory module's `use serde_json` doesn't depend on internal
 # implementation details of hbb_common.
 serde_json  = "1"
+# `perf_events.rs` derives Deserialize on the PowerShell row schema.
+# hbb_common re-exports `serde_derive` and `serde_json` but NOT `serde`
+# itself — and `#[derive(Deserialize)]` expands to a path that references
+# the `serde` crate root, so we depend on it explicitly with the `derive`
+# feature.
+serde       = { version = "1", features = ["derive"] }
 
 [target.'cfg(target_os = "windows")'.dependencies]
 windows-service = "0.6"

README.md

@@ -60,6 +60,18 @@ hello-agent.exe --server                       # user session, SYSTEM token
         │       └─ stamps `agent_name` / `agent_version` / `inventory`
         │          into each /api/sysinfo payload (re-uploads when the
         │          inventory collector below transitions empty → ready)
+        │       └─ signs every request with the device's Ed25519 sk
+        │          (same key rendezvous registers via RegisterPk).
+        │          The server's first valid sig flips that peer to
+        │          `managed=1` and unsigned posts get 401 from then on.
+        │          Spec: rustdesk-server/docs/AGENT-API-AUTH.md
+        ├── exec::run_loop                            (background thread)
+        │       └─ subscribes to sync.rs's EXEC_SENDER broadcast; for
+        │          each queued PowerShell command runs `powershell.exe
+        │          -NoProfile -NonInteractive -Command -`, captures
+        │          stdout+stderr with 1 MiB cap & 5-min timeout, POSTs
+        │          signed result to /api/agent/exec-result. Idle unless
+        │          an admin dispatches via the dashboard.
         ├── inventory::collect_inventory             (background thread)
         │       └─ PowerShell + WMI + wlanapi + ipify → `INVENTORY` global
         │          consumed by hbbs_http::sync above; one-shot, no retry
@@ -115,6 +127,15 @@ inventory — keep it in sync when adding new patches.
      `--cm` process can plug a `MessageBoxW`-based `InvokeUiCM` into
      upstream's connection-manager IPC loop and inherit file-transfer,
      chat, and clipboard handling rather than re-implementing them.
+   * `mod hbbs_http` → `pub mod hbbs_http` so hello-agent's
+     `unattended_password::try_report` and `exec::run_loop` can reach
+     `librustdesk::hbbs_http::sign::build_signed_headers` and
+     `librustdesk::hbbs_http::sync::exec_signal_receiver`. Without
+     this the in-crate code can't sign / can't subscribe to the
+     server's queued PowerShell commands, and the build fails with
+     `E0603: module 'hbbs_http' is private`. Tightly coupled to the
+     **Signed agent API** and **Remote PowerShell exec** divergences
+     below.
 2. **Build shape** — [`vendor/rustdesk/Cargo.toml`](vendor/rustdesk/Cargo.toml):
    `[lib] crate-type` reduced from `["cdylib", "staticlib", "rlib"]` to
    `["rlib"]`. We statically link the rlib into hello-agent.exe; the
@@ -148,6 +169,71 @@ inventory — keep it in sync when adding new patches.
    [`src/main.rs`](vendor/rustdesk/src/main.rs) (`.author(...)`).
    Cosmetic, but they show through in the Windows EXE metadata and
    in-app error dialogs.
+6. **Signed agent API** — every `POST /api/heartbeat`,
+   `POST /api/sysinfo`, and `POST /api/unattended-password` carries
+   two extra headers (`X-RD-Device-Id`,
+   `X-RD-Signature: v1.<ts>.<base64-ed25519-sig>`) so the server can
+   bind the request to the device's existing rendezvous keypair
+   instead of trusting the `id` + `uuid` body fields. Without this
+   patch, anyone who knows a peer's id and uuid can inject inventory,
+   heartbeats, and unattended-access passwords for it. Three patch
+   sites in the vendor tree (plus one in the hello-agent crate):
+   * New file
+     [`src/hbbs_http/sign.rs`](vendor/rustdesk/src/hbbs_http/sign.rs) —
+     the signer (`build_signed_headers`, `path_from_url`). Reads
+     `Config::get_key_pair()` and `Config::get_id()`; uses the
+     re-exported `hbb_common::sodiumoxide`.
+   * [`src/hbbs_http.rs`](vendor/rustdesk/src/hbbs_http.rs) — adds
+     `pub mod sign;` next to the existing module declarations.
+   * [`src/common.rs`](vendor/rustdesk/src/common.rs) — the
+     `post_request_` and `parse_simple_header` header-string parsers
+     now accept a `\n`-separated list of `Name: Value` lines so we
+     can pass both signing headers in one call. Old single-pair
+     callers parse identically — there's no newline to split on.
+   * [`src/hbbs_http/sync.rs`](vendor/rustdesk/src/hbbs_http/sync.rs)
+     call sites (the sysinfo POST around the sysinfo-version
+     comparison block, and the heartbeat POST a few dozen lines
+     later) — both build a signed-headers string via
+     `crate::hbbs_http::sign::build_signed_headers("POST",
+     &path_from_url(&url), body.as_bytes()).unwrap_or_default()`
+     and pass it to `post_request` instead of `""`.
+
+   And in the hello-agent crate proper (not the vendor tree, no
+   re-sync concern):
+   * [`src/unattended_password.rs`](src/unattended_password.rs) —
+     `try_report` also signs its `POST /api/unattended-password`
+     via `librustdesk::hbbs_http::sign::build_signed_headers`.
+
+   Matching server side: see rustdesk-server's
+   [`docs/AGENT-API-AUTH.md`](https://github.com/cstudio-ch/rustdesk-server/blob/pro-features/docs/AGENT-API-AUTH.md)
+   for the wire format and verification flow.
+7. **Remote PowerShell exec** — the dashboard can queue a PowerShell
+   script for a managed peer; the agent runs it as its service account
+   and POSTs the result back. Gated server-side on admin role +
+   `peer.managed=1` + strategy `enable-remote-exec=Y`. Vendor-tree
+   patches:
+   * [`src/hbbs_http/sync.rs`](vendor/rustdesk/src/hbbs_http/sync.rs) —
+     new `EXEC_SENDER` broadcast channel, new `ExecRequest` type, new
+     `pub fn exec_signal_receiver()` helper, and the heartbeat-reply
+     parser drains the `exec: [...]` field into the channel. Vanilla
+     rustdesk simply has no subscriber — the channel send errors out
+     with NoReceivers and the requests are dropped silently.
+
+   In the hello-agent crate:
+   * [`src/exec.rs`](src/exec.rs) — the PowerShell runner. Subscribes
+     to the broadcast channel above, spawns
+     `powershell.exe -NoProfile -NonInteractive -ExecutionPolicy
+     Bypass -Command -`, writes the script to stdin, captures
+     stdout+stderr with 1 MiB cap and a 5-minute wall-clock timeout,
+     signs and POSTs the result to `/api/agent/exec-result`. Started
+     from `run_server()` in [`src/main.rs`](src/main.rs) (must live in
+     the `--server` process to share the broadcast channel with
+     sync.rs).
+   * [`Cargo.toml`](Cargo.toml) — adds `process` to tokio's feature
+     list for `tokio::process::Command`.
+
+   Server-side spec: see [`docs/AGENT-API-AUTH.md`](https://github.com/cstudio-ch/rustdesk-server/blob/pro-features/docs/AGENT-API-AUTH.md)
+   §*Remote PowerShell exec*.
 
 ## Build
 

src/exec.rs

@@ -0,0 +1,259 @@
+//! PowerShell remote-exec worker.
+//!
+//! Subscribes to `librustdesk::hbbs_http::sync::exec_signal_receiver()` — a
+//! broadcast channel that the vendored sync loop populates whenever the
+//! server returns an `exec` field in a heartbeat reply (see
+//! rustdesk-server/docs/AGENT-API-AUTH.md). For each `ExecRequest` we:
+//!
+//!   1. Spawn `powershell.exe -NoProfile -NonInteractive -ExecutionPolicy
+//!      Bypass -Command -` and write the script to stdin.
+//!   2. Concurrently drain stdout and stderr into 1 MiB-capped buffers.
+//!   3. Apply a wall-clock timeout (default 5 min); kill on expiry.
+//!   4. POST the result to `/api/agent/exec-result` with the same Ed25519
+//!      signature the heartbeat / sysinfo posts use.
+//!
+//! The whole thing only makes sense on Windows (the agent's target OS),
+//! so the module body is `#[cfg(windows)]` and other platforms get a
+//! no-op `start()` to keep the call site in `service.rs` portable.
+
+#[cfg(windows)]
+mod windows_impl {
+    use anyhow::{anyhow, Result};
+    use hbb_common::config::Config;
+    use librustdesk::hbbs_http::sync::ExecRequest;
+    use std::process::Stdio;
+    use std::sync::{Arc, Mutex};
+    use std::time::Duration;
+    use tokio::io::{AsyncReadExt, AsyncWriteExt};
+
+    pub fn start() {
+        std::thread::spawn(|| {
+            let rt = match tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+            {
+                Ok(rt) => rt,
+                Err(e) => {
+                    log::warn!("exec worker: build runtime: {e}");
+                    return;
+                }
+            };
+            rt.block_on(run_loop());
+        });
+    }
+
+    async fn run_loop() {
+        // The vendored sync layer creates the broadcast channel lazily on
+        // first `subscribe()`. Calling here also primes it for the parser.
+        let mut rx = librustdesk::hbbs_http::sync::exec_signal_receiver();
+        log::info!("exec worker: subscribed to heartbeat exec channel");
+        loop {
+            match rx.recv().await {
+                Ok(req) => {
+                    log::info!(
+                        "exec worker: received cmd_id={} script_len={} max_secs={} max_bytes={}",
+                        req.cmd_id,
+                        req.script.len(),
+                        req.max_secs,
+                        req.max_bytes
+                    );
+                    let outcome = run_one(&req).await;
+                    if let Err(e) = report(&req, &outcome).await {
+                        log::warn!(
+                            "exec worker: report failed for cmd_id={}: {e:#}",
+                            req.cmd_id
+                        );
+                    }
+                }
+                Err(tokio::sync::broadcast::error::RecvError::Lagged(n)) => {
+                    log::warn!("exec worker: lagged, dropped {n} exec requests");
+                }
+                Err(tokio::sync::broadcast::error::RecvError::Closed) => {
+                    log::warn!("exec worker: channel closed, exiting");
+                    return;
+                }
+            }
+        }
+    }
+
+    struct Outcome {
+        exit_code: i64,
+        stdout: String,
+        stderr: String,
+        timed_out: bool,
+        truncated: bool,
+    }
+
+    async fn run_one(req: &ExecRequest) -> Outcome {
+        // Defensive lower bound — a misconfigured server shouldn't be able to
+        // send max_secs=0 and have us skip the wait.
+        let timeout = Duration::from_secs(req.max_secs.max(1));
+        let max_bytes = req.max_bytes.max(1024) as usize;
+
+        // `-Command -` makes PowerShell read the script body from stdin,
+        // which avoids quoting / length issues that plague `-Command "…"`
+        // for multi-line scripts. `-NoProfile` skips both the
+        // machine-wide and user-wide profile loads — those would change
+        // behaviour depending on which AD-managed PowerShell profile the
+        // service account inherited. `-NonInteractive` makes prompts fail
+        // instead of hanging the run.
+        let mut child = match tokio::process::Command::new("powershell.exe")
+            .args([
+                "-NoProfile",
+                "-NonInteractive",
+                "-ExecutionPolicy",
+                "Bypass",
+                "-Command",
+                "-",
+            ])
+            .stdin(Stdio::piped())
+            .stdout(Stdio::piped())
+            .stderr(Stdio::piped())
+            .spawn()
+        {
+            Ok(c) => c,
+            Err(e) => {
+                return Outcome {
+                    exit_code: -1,
+                    stdout: String::new(),
+                    stderr: format!("spawn failed: {e}"),
+                    timed_out: false,
+                    truncated: false,
+                };
+            }
+        };
+
+        if let Some(mut stdin) = child.stdin.take() {
+            let _ = stdin.write_all(req.script.as_bytes()).await;
+            let _ = stdin.shutdown().await;
+        }
+
+        let stdout = child.stdout.take().expect("piped stdout was requested");
+        let stderr = child.stderr.take().expect("piped stderr was requested");
+
+        // Concurrent capped readers. Each task accumulates up to
+        // `max_bytes` bytes, then drains and discards the rest so the
+        // pipe doesn't block the child writer.
+        let stdout_buf: Arc<Mutex<(Vec<u8>, bool)>> = Arc::new(Mutex::new((Vec::new(), false)));
+        let stderr_buf: Arc<Mutex<(Vec<u8>, bool)>> = Arc::new(Mutex::new((Vec::new(), false)));
+        let so = tokio::spawn(read_capped(stdout, stdout_buf.clone(), max_bytes));
+        let se = tokio::spawn(read_capped(stderr, stderr_buf.clone(), max_bytes));
+
+        let wait_result = tokio::time::timeout(timeout, child.wait()).await;
+        let (exit_code, timed_out) = match wait_result {
+            Ok(Ok(s)) => (s.code().unwrap_or(-1) as i64, false),
+            Ok(Err(_)) => (-1, false),
+            Err(_) => {
+                // Timed out: kill, then wait the killed child so it
+                // reaps cleanly (and so the read tasks finish via EOF).
+                let _ = child.kill().await;
+                let _ = child.wait().await;
+                (-1, true)
+            }
+        };
+        let _ = so.await;
+        let _ = se.await;
+
+        let (out_bytes, out_trunc) = {
+            let g = stdout_buf.lock().unwrap();
+            (g.0.clone(), g.1)
+        };
+        let (err_bytes, err_trunc) = {
+            let g = stderr_buf.lock().unwrap();
+            (g.0.clone(), g.1)
+        };
+        Outcome {
+            exit_code,
+            // PowerShell on a current Windows defaults to UTF-8 when
+            // OutputEncoding is set, but the agent service inherits the
+            // legacy code page on older boxes. `from_utf8_lossy`
+            // guarantees we always have a UTF-8 string to ship; the
+            // operator sees a U+FFFD when raw bytes weren't UTF-8.
+            stdout: String::from_utf8_lossy(&out_bytes).into_owned(),
+            stderr: String::from_utf8_lossy(&err_bytes).into_owned(),
+            timed_out,
+            truncated: out_trunc || err_trunc,
+        }
+    }
+
+    async fn read_capped<R: AsyncReadExt + Unpin>(
+        mut reader: R,
+        buf: Arc<Mutex<(Vec<u8>, bool)>>,
+        cap: usize,
+    ) {
+        let mut chunk = [0u8; 8192];
+        loop {
+            match reader.read(&mut chunk).await {
+                Ok(0) => return,
+                Ok(n) => {
+                    let mut g = buf.lock().unwrap();
+                    if g.0.len() < cap {
+                        let room = cap - g.0.len();
+                        if n <= room {
+                            g.0.extend_from_slice(&chunk[..n]);
+                        } else {
+                            g.0.extend_from_slice(&chunk[..room]);
+                            g.1 = true; // truncated; keep draining
+                        }
+                    }
+                    // else: already truncated, drop this chunk on the floor.
+                }
+                Err(_) => return,
+            }
+        }
+    }
+
+    async fn report(req: &ExecRequest, out: &Outcome) -> Result<()> {
+        let api = librustdesk::common::get_api_server(
+            Config::get_option("api-server"),
+            Config::get_option("custom-rendezvous-server"),
+        );
+        if api.is_empty() {
+            return Err(anyhow!("no api-server configured"));
+        }
+        let url = format!("{api}/api/agent/exec-result");
+        let id = Config::get_id();
+        let uuid = librustdesk::common::encode64(hbb_common::get_uuid());
+        let body = hbb_common::serde_json::json!({
+            "id": id,
+            "uuid": uuid,
+            "cmd_id": req.cmd_id,
+            "exit_code": out.exit_code,
+            "stdout": out.stdout,
+            "stderr": out.stderr,
+            "timed_out": out.timed_out,
+            "truncated": out.truncated,
+        })
+        .to_string();
+
+        let headers = librustdesk::hbbs_http::sign::build_signed_headers(
+            "POST",
+            "/api/agent/exec-result",
+            body.as_bytes(),
+        )
+        .unwrap_or_default();
+        if headers.is_empty() {
+            // Server rejects unsigned exec-result posts unconditionally
+            // (see api/agent_exec.rs); bail loudly so the operator can
+            // see the agent isn't ready to sign yet.
+            return Err(anyhow!("no signing keypair available"));
+        }
+
+        let resp = librustdesk::common::post_request(url, body, &headers)
+            .await
+            .map_err(|e| anyhow!("post: {e}"))?;
+        if resp.trim() == "OK" {
+            Ok(())
+        } else {
+            Err(anyhow!("unexpected response: {}", resp.trim()))
+        }
+    }
+}
+
+#[cfg(windows)]
+pub use windows_impl::start;
+
+#[cfg(not(windows))]
+pub fn start() {
+    log::info!("exec worker: skipped (non-Windows build)");
+}

src/login_events.rs

@@ -0,0 +1,617 @@
+// User-login tracking.
+//
+// Polls the Windows Terminal Services session table at a low cadence and
+// reports logon / logoff events to the rustdesk-server admin API. Each
+// event carries an explicit unix timestamp — for logons that's the OS's
+// session `ConnectTime` (so the recorded time is when the user actually
+// signed in, not when the agent first noticed them); for logoffs it's
+// the agent's wall clock at the moment the session disappeared.
+//
+// Architecture mirrors `unattended_password`:
+//   * One background thread, its own current-thread Tokio runtime, no
+//     entanglement with the SCM supervisor's poll loop.
+//   * In-memory queue with retry-with-backoff on transport / ID_NOT_FOUND
+//     errors (the agent's first POST routinely races rendezvous
+//     registration). Events that never land are eventually dropped to
+//     bound memory — see DROP_AFTER.
+//   * Server-side dedup via UNIQUE INDEX
+//     (peer_id, kind, session_id, at, username) lets the agent re-emit
+//     the same `logon@ConnectTime` event on every service restart without
+//     piling up duplicate rows; agent-side state stays in memory only.
+//
+// Known limits (v1):
+//   * Lock / unlock are not reported — they don't change the
+//     WTSEnumerateSessions output we diff on.
+//   * Logoffs that happen while the service is down are not detected.
+//     The next service start sees "no session" and has nothing to diff
+//     against; this leaves a logon without a paired logoff in the UI.
+//     Tradeoff vs. persisting a snapshot to disk; revisit if operators
+//     ask for it.
+
+use anyhow::{anyhow, Result};
+use std::collections::HashMap;
+use std::sync::Mutex;
+use std::time::Duration;
+
+/// How often to poll the WTS session table. A user can't meaningfully
+/// log in / out faster than this, and the OS keeps the table cheap to
+/// enumerate (it's a kernel-side struct, not a registry scan).
+const POLL_INTERVAL: Duration = Duration::from_secs(5);
+
+/// How often to attempt a flush of the pending queue. Decoupled from the
+/// poll interval so a transient server outage doesn't slow down session
+/// observation; we keep observing locally and just back off the network
+/// retry.
+const FLUSH_INTERVAL_BASE: Duration = Duration::from_secs(5);
+const FLUSH_INTERVAL_MAX: Duration = Duration::from_secs(60);
+
+/// Drop events from the queue after this many failed delivery attempts.
+/// At backoff cap = 60s, this is ~6 hours of trying — enough to ride out
+/// a long server-side outage, short enough that a permanently-misconfigured
+/// agent doesn't blow up memory.
+const DROP_AFTER: u32 = 360;
+
+/// Cap per request — must match the server's MAX_EVENTS_PER_POST. Server
+/// rejects anything larger with a 400 so we'd retry forever if we exceeded
+/// it.
+const MAX_EVENTS_PER_POST: usize = 256;
+
+/// One observed session snapshot. Equality by `(session_id, connect_time)`
+/// — Windows can recycle session IDs across logins, so we use the OS-
+/// reported `ConnectTime` as the disambiguator. Two snapshots compare
+/// equal iff they describe the *same* logical login.
+#[derive(Clone, Debug)]
+struct Session {
+    session_id: u32,
+    /// FILETIME → unix epoch seconds. 0 if the OS returned an unparseable
+    /// value (very old Windows builds); we still report `logon` but the
+    /// server-side dedup degrades to "first observation wins".
+    connect_unix: i64,
+    username: String,
+    domain: String,
+    /// "console" | "rdp" | "" (unknown).
+    session_kind: String,
+}
+
+#[derive(Clone, Debug)]
+struct PendingEvent {
+    at: i64,
+    kind: &'static str,
+    username: String,
+    domain: String,
+    session_id: u32,
+    session_kind: String,
+    /// Number of times we've tried to flush this event. Used to drop
+    /// rows that have been retrying since forever.
+    attempts: u32,
+}
+
+#[cfg(target_os = "windows")]
+static QUEUE: Mutex<Vec<PendingEvent>> = Mutex::new(Vec::new());
+
+/// Kick off the background tracker. Returns immediately. Safe to call
+/// multiple times (subsequent calls are no-ops) — gated by an
+/// `AtomicBool`.
+pub fn start() {
+    #[cfg(not(target_os = "windows"))]
+    {
+        // Login tracking is Windows-only; on other platforms the call is
+        // a no-op so cross-platform builds (CI lint runs on Linux) link.
+    }
+    #[cfg(target_os = "windows")]
+    {
+        use std::sync::atomic::{AtomicBool, Ordering};
+        static STARTED: AtomicBool = AtomicBool::new(false);
+        if STARTED.swap(true, Ordering::SeqCst) {
+            return;
+        }
+        std::thread::spawn(move || {
+            let rt = match tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+            {
+                Ok(rt) => rt,
+                Err(e) => {
+                    log::warn!("login-events: build runtime: {e}");
+                    return;
+                }
+            };
+            rt.block_on(run_loop());
+        });
+    }
+}
+
+#[cfg(target_os = "windows")]
+async fn run_loop() {
+    // Diff snapshot. Keyed by session_id; value carries connect_unix so we
+    // can detect "session id reused for a new login" as well as "session
+    // disappeared".
+    let mut prev: HashMap<u32, Session> = HashMap::new();
+    let mut first_poll = true;
+    let mut flush_backoff = FLUSH_INTERVAL_BASE;
+
+    loop {
+        match enumerate_active_user_sessions() {
+            Ok(snapshot) => {
+                let now_unix = hbb_common::chrono::Utc::now().timestamp();
+                let curr: HashMap<u32, Session> =
+                    snapshot.into_iter().map(|s| (s.session_id, s)).collect();
+                let events =
+                    diff_into_events(&prev, &curr, first_poll, now_unix);
+                if !events.is_empty() {
+                    let mut q = QUEUE.lock().unwrap();
+                    q.extend(events);
+                }
+                prev = curr;
+                first_poll = false;
+            }
+            Err(e) => {
+                log::warn!("login-events: enumerate failed: {e:#}");
+            }
+        }
+
+        // Try to flush. If the queue is empty this is cheap (no network).
+        // The flush also handles its own retry / drop semantics — we just
+        // adjust our local cadence based on whether it succeeded.
+        match flush_once().await {
+            FlushOutcome::Idle | FlushOutcome::AllSent => {
+                flush_backoff = FLUSH_INTERVAL_BASE;
+            }
+            FlushOutcome::Failed => {
+                flush_backoff = (flush_backoff * 2).min(FLUSH_INTERVAL_MAX);
+            }
+        }
+
+        // Poll cadence is constant; the network backoff doesn't slow down
+        // observation — that would risk missing logoffs while the server
+        // is down. We just delay the retry of pending events.
+        tokio::time::sleep(POLL_INTERVAL.min(flush_backoff)).await;
+    }
+}
+
+#[cfg(target_os = "windows")]
+fn diff_into_events(
+    prev: &HashMap<u32, Session>,
+    curr: &HashMap<u32, Session>,
+    first_poll: bool,
+    now_unix: i64,
+) -> Vec<PendingEvent> {
+    let mut out = Vec::new();
+
+    // New / changed sessions → logon.
+    for (sid, sess) in curr {
+        let changed = match prev.get(sid) {
+            None => true,
+            Some(old) => {
+                // Same session id but a different connect_time → the
+                // OS recycled the slot for a fresh login. Emit a
+                // logoff for the old occupant and a logon for the new.
+                if old.connect_unix != sess.connect_unix
+                    || old.username != sess.username
+                {
+                    out.push(PendingEvent {
+                        at: now_unix,
+                        kind: "logoff",
+                        username: old.username.clone(),
+                        domain: old.domain.clone(),
+                        session_id: *sid,
+                        session_kind: old.session_kind.clone(),
+                        attempts: 0,
+                    });
+                    true
+                } else {
+                    false
+                }
+            }
+        };
+        if !changed {
+            continue;
+        }
+        // ConnectTime can be 0 on the login-screen session even when a
+        // user is shown as logged in (rare edge); fall back to `now` so
+        // the row still lands somewhere sensible.
+        let at = if sess.connect_unix > 0 {
+            sess.connect_unix
+        } else {
+            now_unix
+        };
+        out.push(PendingEvent {
+            at,
+            kind: "logon",
+            username: sess.username.clone(),
+            domain: sess.domain.clone(),
+            session_id: *sid,
+            session_kind: sess.session_kind.clone(),
+            attempts: 0,
+        });
+    }
+
+    // Disappeared sessions → logoff. Skipped on the very first poll
+    // because we have no baseline to diff against — see the module-level
+    // "first poll: emit logons-only" note.
+    if !first_poll {
+        for (sid, old) in prev {
+            if !curr.contains_key(sid) {
+                out.push(PendingEvent {
+                    at: now_unix,
+                    kind: "logoff",
+                    username: old.username.clone(),
+                    domain: old.domain.clone(),
+                    session_id: *sid,
+                    session_kind: old.session_kind.clone(),
+                    attempts: 0,
+                });
+            }
+        }
+    }
+
+    out
+}
+
+#[cfg(target_os = "windows")]
+enum FlushOutcome {
+    Idle,
+    AllSent,
+    Failed,
+}
+
+#[cfg(target_os = "windows")]
+async fn flush_once() -> FlushOutcome {
+    // Take a snapshot under the lock so we don't hold it across the
+    // (potentially slow) network call. If the POST succeeds we drop the
+    // matching prefix from the queue; if it fails we put back the
+    // unsent tail with bumped attempt counters.
+    let batch: Vec<PendingEvent> = {
+        let mut q = QUEUE.lock().unwrap();
+        if q.is_empty() {
+            return FlushOutcome::Idle;
+        }
+        let take = q.len().min(MAX_EVENTS_PER_POST);
+        q.drain(..take).collect()
+    };
+
+    let posted = post_batch(&batch).await;
+    match posted {
+        Ok(()) => {
+            // Server accepted (or returned ID_NOT_FOUND, which we treat
+            // as "drop and continue" because no amount of retrying will
+            // make the peer materialize without the rendezvous loop in
+            // --server, which the unattended_password reporter already
+            // hammers on; once that succeeds the next batch lands).
+            FlushOutcome::AllSent
+        }
+        Err(e) => {
+            log::warn!(
+                "login-events: flush of {} event(s) failed: {e:#}",
+                batch.len(),
+            );
+            // Put the batch back with bumped attempt counters, modulo
+            // events that have hit the drop threshold. Prepend so that
+            // any events pushed by the poll loop between the drain and
+            // here stay after the (older) failed batch.
+            let mut requeued: Vec<PendingEvent> = batch
+                .into_iter()
+                .filter_map(|mut ev| {
+                    ev.attempts = ev.attempts.saturating_add(1);
+                    if ev.attempts >= DROP_AFTER {
+                        log::warn!(
+                            "login-events: dropping event after {} attempts: \
+                             kind={} session={} user={}",
+                            ev.attempts, ev.kind, ev.session_id, ev.username,
+                        );
+                        None
+                    } else {
+                        Some(ev)
+                    }
+                })
+                .collect();
+            let mut q = QUEUE.lock().unwrap();
+            let tail: Vec<PendingEvent> = q.drain(..).collect();
+            requeued.extend(tail);
+            *q = requeued;
+            FlushOutcome::Failed
+        }
+    }
+}
+
+#[cfg(target_os = "windows")]
+async fn post_batch(batch: &[PendingEvent]) -> Result<()> {
+    let api = librustdesk::common::get_api_server(
+        hbb_common::config::Config::get_option("api-server"),
+        hbb_common::config::Config::get_option("custom-rendezvous-server"),
+    );
+    if api.is_empty() {
+        return Err(anyhow!("no api-server configured yet"));
+    }
+
+    let url = format!("{api}/api/agent/login-event");
+    let id = hbb_common::config::Config::get_id();
+    let uuid = librustdesk::common::encode64(hbb_common::get_uuid());
+
+    let events: Vec<hbb_common::serde_json::Value> = batch
+        .iter()
+        .map(|ev| {
+            hbb_common::serde_json::json!({
+                "at":           ev.at,
+                "kind":         ev.kind,
+                "username":     ev.username,
+                "domain":       ev.domain,
+                "session_id":   ev.session_id,
+                "session_kind": ev.session_kind,
+            })
+        })
+        .collect();
+    let body = hbb_common::serde_json::json!({
+        "id":     id,
+        "uuid":   uuid,
+        "events": events,
+    })
+    .to_string();
+
+    let headers = librustdesk::hbbs_http::sign::build_signed_headers(
+        "POST",
+        "/api/agent/login-event",
+        body.as_bytes(),
+    )
+    .unwrap_or_default();
+
+    let resp = librustdesk::common::post_request(url, body, &headers)
+        .await
+        .map_err(|e| anyhow!("post: {e}"))?;
+    let trimmed = resp.trim();
+    if trimmed == "OK" || trimmed == "ID_NOT_FOUND" {
+        // ID_NOT_FOUND is "peer not registered yet" — happens on the
+        // first few flushes after a fresh install, before the
+        // rendezvous loop in --server has created the peer row. The
+        // unattended_password reporter races this same window; once
+        // either of them succeeds the peer row exists and subsequent
+        // posts land. We treat it as success here so the agent doesn't
+        // pile up unbounded retries — if rendezvous never registers,
+        // the heartbeat path is also broken and the operator has
+        // bigger problems than missing login events.
+        Ok(())
+    } else {
+        Err(anyhow!("unexpected response: {trimmed}"))
+    }
+}
+
+// ─────────────────────────── Win32 session enumeration ────────────────────
+//
+// Same shape as service.rs's find_active_user_session — we declare just
+// the WTS functions we touch rather than pull in another bindgen. The
+// types are tiny and ABI-stable.
+
+#[cfg(target_os = "windows")]
+#[repr(C)]
+struct WtsSessionInfoW {
+    session_id: u32,
+    win_station_name: *mut u16,
+    state: i32,
+}
+
+#[cfg(target_os = "windows")]
+#[repr(C)]
+struct WtsInfoW {
+    state: i32,
+    session_id: u32,
+    incoming_bytes: u32,
+    outgoing_bytes: u32,
+    incoming_frames: u32,
+    outgoing_frames: u32,
+    incoming_compressed_bytes: u32,
+    outgoing_compressed_bytes: u32,
+    win_station_name: [u16; 32],
+    domain: [u16; 17],
+    user_name: [u16; 21],
+    connect_time: i64,
+    disconnect_time: i64,
+    last_input_time: i64,
+    logon_time: i64,
+    current_time: i64,
+}
+
+#[cfg(target_os = "windows")]
+extern "system" {
+    fn WTSEnumerateSessionsW(
+        h_server: winapi::shared::ntdef::HANDLE,
+        reserved: u32,
+        version: u32,
+        pp_session_info: *mut *mut WtsSessionInfoW,
+        p_count: *mut u32,
+    ) -> i32;
+    fn WTSFreeMemory(p_memory: *mut std::ffi::c_void);
+    fn WTSQuerySessionInformationW(
+        h_server: winapi::shared::ntdef::HANDLE,
+        session_id: u32,
+        info_class: i32,
+        pp_buffer: *mut *mut u16,
+        p_bytes_returned: *mut u32,
+    ) -> i32;
+}
+
+#[cfg(target_os = "windows")]
+const WTS_ACTIVE: i32 = 0;
+#[cfg(target_os = "windows")]
+const WTS_USER_NAME: i32 = 5;
+#[cfg(target_os = "windows")]
+const WTS_DOMAIN_NAME: i32 = 7;
+#[cfg(target_os = "windows")]
+const WTS_CLIENT_PROTOCOL_TYPE: i32 = 16;
+#[cfg(target_os = "windows")]
+const WTS_SESSION_INFO: i32 = 24;
+
+#[cfg(target_os = "windows")]
+fn enumerate_active_user_sessions() -> Result<Vec<Session>> {
+    let mut sessions: *mut WtsSessionInfoW = std::ptr::null_mut();
+    let mut count: u32 = 0;
+    let ok = unsafe {
+        WTSEnumerateSessionsW(
+            std::ptr::null_mut(),
+            0,
+            1,
+            &mut sessions,
+            &mut count,
+        )
+    };
+    if ok == 0 || sessions.is_null() {
+        return Err(anyhow!(
+            "WTSEnumerateSessionsW failed: {}",
+            std::io::Error::last_os_error()
+        ));
+    }
+
+    let mut out = Vec::new();
+    for i in 0..count {
+        let info = unsafe { &*sessions.add(i as usize) };
+        if info.state != WTS_ACTIVE {
+            continue;
+        }
+        let sid = info.session_id;
+
+        // Skip sessions without a logged-in user (login screen, Session 0).
+        let username = match query_wide(sid, WTS_USER_NAME) {
+            Some(s) if !s.is_empty() => s,
+            _ => continue,
+        };
+        let domain = query_wide(sid, WTS_DOMAIN_NAME).unwrap_or_default();
+        let session_kind = match query_protocol_type(sid) {
+            Some(0) => "console".to_string(),
+            Some(2) => "rdp".to_string(),
+            Some(_) | None => String::new(),
+        };
+        let connect_unix = query_connect_time(sid).unwrap_or(0);
+
+        out.push(Session {
+            session_id: sid,
+            connect_unix,
+            username,
+            domain,
+            session_kind,
+        });
+    }
+
+    unsafe { WTSFreeMemory(sessions as *mut std::ffi::c_void) };
+    Ok(out)
+}
+
+/// Pull a WCHAR-string-shaped value (WTSUserName / WTSDomainName / …)
+/// and convert to UTF-8. Returns None on failure or empty result.
+#[cfg(target_os = "windows")]
+fn query_wide(session_id: u32, info_class: i32) -> Option<String> {
+    let mut buf: *mut u16 = std::ptr::null_mut();
+    let mut bytes: u32 = 0;
+    let ok = unsafe {
+        WTSQuerySessionInformationW(
+            std::ptr::null_mut(),
+            session_id,
+            info_class,
+            &mut buf,
+            &mut bytes,
+        )
+    };
+    if ok == 0 || buf.is_null() {
+        return None;
+    }
+    let s = unsafe { wide_to_string(buf, bytes) };
+    unsafe { WTSFreeMemory(buf as *mut std::ffi::c_void) };
+    if s.is_empty() {
+        None
+    } else {
+        Some(s)
+    }
+}
+
+/// WTSClientProtocolType returns a single USHORT (2 bytes). Buffer is
+/// allocated by Windows; we read the 16-bit value and free.
+#[cfg(target_os = "windows")]
+fn query_protocol_type(session_id: u32) -> Option<u16> {
+    let mut buf: *mut u16 = std::ptr::null_mut();
+    let mut bytes: u32 = 0;
+    let ok = unsafe {
+        WTSQuerySessionInformationW(
+            std::ptr::null_mut(),
+            session_id,
+            WTS_CLIENT_PROTOCOL_TYPE,
+            &mut buf,
+            &mut bytes,
+        )
+    };
+    if ok == 0 || buf.is_null() || bytes < 2 {
+        if !buf.is_null() {
+            unsafe { WTSFreeMemory(buf as *mut std::ffi::c_void) };
+        }
+        return None;
+    }
+    let val = unsafe { *buf };
+    unsafe { WTSFreeMemory(buf as *mut std::ffi::c_void) };
+    Some(val)
+}
+
+/// Pull ConnectTime from WTSINFOW (the per-session struct returned by
+/// WTSQuerySessionInformation with class WTSSessionInfo). FILETIME 0
+/// means "OS doesn't know" (rare — happens on the login-screen session
+/// before a user signs in); we surface that as None and the caller
+/// falls back to `now()`.
+#[cfg(target_os = "windows")]
+fn query_connect_time(session_id: u32) -> Option<i64> {
+    let mut buf: *mut u16 = std::ptr::null_mut();
+    let mut bytes: u32 = 0;
+    let ok = unsafe {
+        WTSQuerySessionInformationW(
+            std::ptr::null_mut(),
+            session_id,
+            WTS_SESSION_INFO,
+            &mut buf,
+            &mut bytes,
+        )
+    };
+    if ok == 0 || buf.is_null() || (bytes as usize) < std::mem::size_of::<WtsInfoW>() {
+        if !buf.is_null() {
+            unsafe { WTSFreeMemory(buf as *mut std::ffi::c_void) };
+        }
+        return None;
+    }
+    let info: &WtsInfoW = unsafe { &*(buf as *const WtsInfoW) };
+    let ft = info.connect_time;
+    unsafe { WTSFreeMemory(buf as *mut std::ffi::c_void) };
+    let unix = filetime_to_unix(ft);
+    if unix > 0 {
+        Some(unix)
+    } else {
+        None
+    }
+}
+
+/// FILETIME is 100-nanosecond ticks since 1601-01-01 UTC. Convert to
+/// unix seconds; clamp to >=0 since the table column is signed but a
+/// pre-epoch ConnectTime is nonsense for our purposes.
+#[cfg(target_os = "windows")]
+fn filetime_to_unix(ft: i64) -> i64 {
+    // 11644473600 = seconds between 1601-01-01 and 1970-01-01.
+    const TICKS_PER_SEC: i64 = 10_000_000;
+    const EPOCH_DIFF_SECS: i64 = 11_644_473_600;
+    if ft <= 0 {
+        return 0;
+    }
+    let secs_since_1601 = ft / TICKS_PER_SEC;
+    let unix = secs_since_1601 - EPOCH_DIFF_SECS;
+    unix.max(0)
+}
+
+/// Convert a NUL-terminated UTF-16 buffer to a Rust String. `bytes` is
+/// the byte count Windows returned (NOT the char count); we trust the
+/// trailing NUL but cap on `bytes / 2` so a malformed buffer can't run
+/// us into unallocated memory.
+#[cfg(target_os = "windows")]
+unsafe fn wide_to_string(buf: *const u16, bytes: u32) -> String {
+    if buf.is_null() || bytes == 0 {
+        return String::new();
+    }
+    let max_chars = (bytes as usize) / 2;
+    let mut len = 0usize;
+    while len < max_chars && *buf.add(len) != 0 {
+        len += 1;
+    }
+    let slice = std::slice::from_raw_parts(buf, len);
+    String::from_utf16_lossy(slice)
+}

src/main.rs

@@ -28,6 +28,11 @@ mod inventory;
 #[cfg(target_os = "windows")]
 mod cm_popup;
 #[cfg(target_os = "windows")]
+mod exec;
+mod login_events;
+mod perf;
+mod perf_events;
+#[cfg(target_os = "windows")]
 mod service;
 #[cfg(target_os = "windows")]
 mod unattended_password;
@@ -278,6 +283,15 @@ fn run_server() {
         }
     });
 
+    // Start the PowerShell remote-exec worker. Subscribes to the
+    // broadcast channel in the vendored sync layer; the channel is
+    // shared in-process so the worker MUST run in this --server process
+    // (where sync.rs lives), not the --service supervisor. The worker
+    // is idle until an admin dispatches an exec from the dashboard.
+    // Gated server-side on peer.managed=1 + strategy.enable-remote-exec.
+    #[cfg(target_os = "windows")]
+    exec::start();
+
     // `start_server` is `#[tokio::main]` and runs forever. (is_server=true,
     // no_server=false). It boots the default IPC server, input service,
     // rendezvous mediator, and heartbeat sync.

src/perf.rs

@@ -0,0 +1,311 @@
+// Continuous performance sampling.
+//
+// One sample per minute: overall CPU%, memory used / total, top
+// process by CPU, top process by memory, uptime, process count.
+// Posted in batches to /api/agent/metrics; surfaced on the admin
+// device detail page as a 24 h sparkline plus a "live" snapshot card.
+//
+// Architecture mirrors `login_events`:
+//   * One background thread, its own current-thread Tokio runtime.
+//   * Sysinfo crate (vendored under hbb_common) does the cross-cutting
+//     work — we keep one `System` instance alive across iterations so
+//     its per-process CPU% accounting (which is differential against
+//     the previous refresh) stays accurate.
+//   * In-memory queue with retry-with-backoff and a hard drop cap so
+//     a permanently-misconfigured agent can't balloon memory.
+//   * Server-side `INSERT OR IGNORE` keyed on (peer_id, at) dedups any
+//     retries that get there twice.
+//
+// Process CPU% on Windows is reported by sysinfo as "% of one core",
+// so a busy multi-threaded process can read >100%. We normalise by
+// `cpus().len()` so the snapshot card's "top CPU: chrome.exe 18%"
+// is comparable to the overall CPU%. Memory is in bytes from sysinfo;
+// we convert to MB on the wire.
+
+use anyhow::{anyhow, Result};
+use std::sync::Mutex;
+use std::time::Duration;
+
+/// Sampling cadence. 60 s strikes a balance between resolution (enough
+/// granularity to spot a 2-minute CPU spike) and storage (~1440 rows /
+/// device / day on the server side).
+const SAMPLE_INTERVAL: Duration = Duration::from_secs(60);
+
+/// Flush cadence on the happy path. The reporter tries to flush after
+/// every sample anyway; this is the floor used by the network-error
+/// backoff before it doubles.
+const FLUSH_INTERVAL_BASE: Duration = Duration::from_secs(60);
+const FLUSH_INTERVAL_MAX: Duration = Duration::from_secs(15 * 60);
+
+/// At backoff cap = 15 min, this is ~3 days of trying — enough to ride
+/// out a long server-side outage. Beyond that we drop, on the same
+/// reasoning as `login_events::DROP_AFTER`.
+const DROP_AFTER: u32 = 300;
+
+/// Must match the server's MAX_SAMPLES_PER_POST.
+const MAX_SAMPLES_PER_POST: usize = 512;
+
+#[derive(Clone, Debug, Default)]
+struct PendingSample {
+    at: i64,
+    cpu_pct: f64,
+    mem_used_mb: i64,
+    mem_total_mb: i64,
+    proc_count: i64,
+    uptime_secs: i64,
+    top_cpu_name: String,
+    top_cpu_pct: f64,
+    top_mem_name: String,
+    top_mem_mb: i64,
+    attempts: u32,
+}
+
+#[cfg(target_os = "windows")]
+static QUEUE: Mutex<Vec<PendingSample>> = Mutex::new(Vec::new());
+
+/// Kick off the metrics sampler. Safe to call multiple times — guarded
+/// by an `AtomicBool` so a stray second call is a no-op.
+pub fn start() {
+    #[cfg(not(target_os = "windows"))]
+    {
+        // Cross-platform stub; the implementation only runs on Windows
+        // because that's the only OS hello-agent ships on. Sysinfo is
+        // cross-platform, so a future Linux build can drop the cfg-gate
+        // and reuse this module unchanged.
+    }
+    #[cfg(target_os = "windows")]
+    {
+        use std::sync::atomic::{AtomicBool, Ordering};
+        static STARTED: AtomicBool = AtomicBool::new(false);
+        if STARTED.swap(true, Ordering::SeqCst) {
+            return;
+        }
+        std::thread::spawn(move || {
+            let rt = match tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+            {
+                Ok(rt) => rt,
+                Err(e) => {
+                    log::warn!("perf: build runtime: {e}");
+                    return;
+                }
+            };
+            rt.block_on(run_loop());
+        });
+    }
+}
+
+#[cfg(target_os = "windows")]
+async fn run_loop() {
+    use hbb_common::sysinfo::System;
+
+    // One System instance, refreshed across iterations. Sysinfo's
+    // per-process CPU% is computed against the previous refresh, so a
+    // throwaway `System::new()` each tick would always read 0%.
+    let mut sys = System::new();
+    // Prime CPU + processes once so the first real sample isn't NaN.
+    // The MINIMUM_CPU_UPDATE_INTERVAL sleep is what sysinfo's docs
+    // recommend between back-to-back CPU refreshes; here we have the
+    // full SAMPLE_INTERVAL coming up so we just refresh-and-wait.
+    sys.refresh_cpu();
+    sys.refresh_processes();
+    tokio::time::sleep(SAMPLE_INTERVAL).await;
+
+    let mut flush_backoff = FLUSH_INTERVAL_BASE;
+    loop {
+        if let Some(sample) = collect_sample(&mut sys) {
+            QUEUE.lock().unwrap().push(sample);
+        }
+
+        match flush_once().await {
+            FlushOutcome::Idle | FlushOutcome::AllSent => {
+                flush_backoff = FLUSH_INTERVAL_BASE;
+            }
+            FlushOutcome::Failed => {
+                flush_backoff = (flush_backoff * 2).min(FLUSH_INTERVAL_MAX);
+            }
+        }
+
+        // We always sleep SAMPLE_INTERVAL — the network backoff slows
+        // the *retry* of pending events, not observation. Sampling
+        // continues at full cadence so a transient outage doesn't punch
+        // a hole in the chart on either side of the dropped window.
+        tokio::time::sleep(SAMPLE_INTERVAL.min(flush_backoff)).await;
+    }
+}
+
+#[cfg(target_os = "windows")]
+fn collect_sample(sys: &mut hbb_common::sysinfo::System) -> Option<PendingSample> {
+    sys.refresh_cpu();
+    sys.refresh_memory();
+    sys.refresh_processes();
+
+    let at = hbb_common::chrono::Utc::now().timestamp();
+    let cpu_pct = sys.global_cpu_info().cpu_usage() as f64;
+    // Sysinfo returns bytes on every platform; the server schema stores
+    // MB to keep row sizes small at scale. Divide-then-cast bounds
+    // arithmetic to u64 territory.
+    let mem_total_mb = (sys.total_memory() / 1024 / 1024) as i64;
+    let mem_used_mb = (sys.used_memory() / 1024 / 1024) as i64;
+    let uptime_secs = sys.uptime() as i64;
+
+    // Per-process CPU% from sysinfo is normalised per core: a 4-thread
+    // process pinning 4 cores on a 4-core machine reads 400%. Divide
+    // by core count so the snapshot card's number matches the overall
+    // CPU% reading (both 0-100 of the whole machine).
+    let cpu_count = sys.cpus().len().max(1) as f64;
+
+    let mut top_cpu: Option<(&str, f32)> = None;
+    let mut top_mem: Option<(&str, u64)> = None;
+    let mut proc_count = 0i64;
+    for proc in sys.processes().values() {
+        proc_count += 1;
+        let name = proc.name();
+        // Some kernel-side rows show up with empty names on Windows;
+        // skip them so we don't ever render a top-CPU row with no
+        // label.
+        if name.is_empty() {
+            continue;
+        }
+        let cu = proc.cpu_usage();
+        if cu.is_finite() && cu > top_cpu.map(|(_, v)| v).unwrap_or(0.0) {
+            top_cpu = Some((name, cu));
+        }
+        let mu = proc.memory();
+        if mu > top_mem.map(|(_, v)| v).unwrap_or(0) {
+            top_mem = Some((name, mu));
+        }
+    }
+
+    let (top_cpu_name, top_cpu_pct) = top_cpu
+        .map(|(n, v)| (n.to_string(), (v as f64 / cpu_count).min(100.0)))
+        .unwrap_or_default();
+    let (top_mem_name, top_mem_mb) = top_mem
+        .map(|(n, v)| (n.to_string(), (v / 1024 / 1024) as i64))
+        .unwrap_or_default();
+
+    Some(PendingSample {
+        at,
+        cpu_pct,
+        mem_used_mb,
+        mem_total_mb,
+        proc_count,
+        uptime_secs,
+        top_cpu_name,
+        top_cpu_pct,
+        top_mem_name,
+        top_mem_mb,
+        attempts: 0,
+    })
+}
+
+#[cfg(target_os = "windows")]
+enum FlushOutcome {
+    Idle,
+    AllSent,
+    Failed,
+}
+
+#[cfg(target_os = "windows")]
+async fn flush_once() -> FlushOutcome {
+    let batch: Vec<PendingSample> = {
+        let mut q = QUEUE.lock().unwrap();
+        if q.is_empty() {
+            return FlushOutcome::Idle;
+        }
+        let take = q.len().min(MAX_SAMPLES_PER_POST);
+        q.drain(..take).collect()
+    };
+
+    match post_batch(&batch).await {
+        Ok(()) => FlushOutcome::AllSent,
+        Err(e) => {
+            log::warn!(
+                "perf: flush of {} sample(s) failed: {e:#}",
+                batch.len(),
+            );
+            let mut requeued: Vec<PendingSample> = batch
+                .into_iter()
+                .filter_map(|mut s| {
+                    s.attempts = s.attempts.saturating_add(1);
+                    if s.attempts >= DROP_AFTER {
+                        log::warn!(
+                            "perf: dropping sample after {} attempts: at={}",
+                            s.attempts, s.at,
+                        );
+                        None
+                    } else {
+                        Some(s)
+                    }
+                })
+                .collect();
+            let mut q = QUEUE.lock().unwrap();
+            let tail: Vec<PendingSample> = q.drain(..).collect();
+            requeued.extend(tail);
+            *q = requeued;
+            FlushOutcome::Failed
+        }
+    }
+}
+
+#[cfg(target_os = "windows")]
+async fn post_batch(batch: &[PendingSample]) -> Result<()> {
+    let api = librustdesk::common::get_api_server(
+        hbb_common::config::Config::get_option("api-server"),
+        hbb_common::config::Config::get_option("custom-rendezvous-server"),
+    );
+    if api.is_empty() {
+        return Err(anyhow!("no api-server configured yet"));
+    }
+
+    let url = format!("{api}/api/agent/metrics");
+    let id = hbb_common::config::Config::get_id();
+    let uuid = librustdesk::common::encode64(hbb_common::get_uuid());
+
+    let samples: Vec<hbb_common::serde_json::Value> = batch
+        .iter()
+        .map(|s| {
+            hbb_common::serde_json::json!({
+                "at":           s.at,
+                "cpu_pct":      s.cpu_pct,
+                "mem_used_mb":  s.mem_used_mb,
+                "mem_total_mb": s.mem_total_mb,
+                "proc_count":   s.proc_count,
+                "uptime_secs":  s.uptime_secs,
+                "top_cpu_name": s.top_cpu_name,
+                "top_cpu_pct":  s.top_cpu_pct,
+                "top_mem_name": s.top_mem_name,
+                "top_mem_mb":   s.top_mem_mb,
+            })
+        })
+        .collect();
+    let body = hbb_common::serde_json::json!({
+        "id":      id,
+        "uuid":    uuid,
+        "samples": samples,
+    })
+    .to_string();
+
+    let headers = librustdesk::hbbs_http::sign::build_signed_headers(
+        "POST",
+        "/api/agent/metrics",
+        body.as_bytes(),
+    )
+    .unwrap_or_default();
+
+    let resp = librustdesk::common::post_request(url, body, &headers)
+        .await
+        .map_err(|e| anyhow!("post: {e}"))?;
+    let trimmed = resp.trim();
+    if trimmed == "OK" || trimmed == "ID_NOT_FOUND" {
+        // ID_NOT_FOUND mirrors the unattended_password / login_events
+        // contract: server doesn't know the peer yet (rendezvous race
+        // on first boot). We drop the batch rather than retry forever;
+        // the next sample lands once /api/heartbeat has created the
+        // peer row.
+        Ok(())
+    } else {
+        Err(anyhow!("unexpected response: {trimmed}"))
+    }
+}

src/perf_events.rs

@@ -0,0 +1,507 @@
+// Windows event-log scraper for performance-related events.
+//
+// Pulls fresh entries from three channels Microsoft itself uses to
+// flag "the OS noticed this machine was slow":
+//
+//   * `Microsoft-Windows-Diagnostics-Performance/Operational` — boot,
+//     shutdown, standby, resume degradation (Event IDs 100/200/300/400
+//     family), each carrying total time + the component that caused it.
+//   * `Microsoft-Windows-Resource-Exhaustion-Detector/Operational` —
+//     Event ID 2004 fires when virtual memory hits the wall, with the
+//     top processes by working set.
+//   * `System` — IDs 41 (Kernel-Power unexpected reboot), 6008 (dirty
+//     shutdown), 1001 (BugCheck / BSOD).
+//
+// Cadence is intentionally low (5 min): these events are sparse — a
+// healthy machine may produce zero per day, a sick one a handful.
+//
+// State machine per channel: a numeric RecordId cursor persisted in
+// the agent config (`Config::set_option`). First run with no cursor
+// pulls the trailing 7 days bounded by `MAX_FIRST_RUN_EVENTS`; every
+// subsequent run pulls everything with `EventRecordID > cursor`. The
+// cursor advances on observation, not on successful POST — see the
+// `flush_once` comment for the tradeoff.
+
+#![cfg_attr(not(target_os = "windows"), allow(dead_code))]
+
+use anyhow::{anyhow, Result};
+use serde::Deserialize;
+use std::sync::Mutex;
+use std::time::Duration;
+
+/// How often to scrape. 5 min keeps the UI freshness reasonable
+/// (operators looking at a complaint won't usually see a stale view)
+/// without burning host CPU on PowerShell startup every minute.
+const SCRAPE_INTERVAL: Duration = Duration::from_secs(5 * 60);
+
+const FLUSH_INTERVAL_BASE: Duration = Duration::from_secs(60);
+const FLUSH_INTERVAL_MAX: Duration = Duration::from_secs(15 * 60);
+
+/// First-run lookback. A box that's been alive for months would
+/// otherwise dump thousands of `Diagnostics-Performance` events on a
+/// fresh install — useful in theory, but the UI shows the most recent
+/// 20 and the older entries are mostly noise.
+const MAX_FIRST_RUN_EVENTS: u32 = 100;
+
+/// Per-scrape cap so a misbehaving event log can't blow up an
+/// individual run. Anything beyond this on a single pass is dropped on
+/// the floor and picked up on the next scrape (cursor advances to the
+/// last seen, so we don't oscillate).
+const MAX_PER_SCRAPE: u32 = 200;
+
+/// Must match the server's MAX_EVENTS_PER_POST.
+const MAX_EVENTS_PER_POST: usize = 128;
+
+/// Drop pending events after this many retries — at 15 min cap this
+/// is ~5 days, plenty for any realistic outage window.
+const DROP_AFTER: u32 = 480;
+
+/// One channel config: the WEL log name, the short provider tag stored
+/// server-side (matches the `devices.perf_src_*` i18n keys), and an
+/// optional ID allow-list. `None` means "everything in this channel";
+/// `Some(&[…])` restricts to those event IDs (used for `System` to
+/// avoid pulling boot / service-start chatter).
+struct ChannelCfg {
+    provider: &'static str,
+    log_name: &'static str,
+    event_ids: Option<&'static [u32]>,
+}
+
+const CHANNELS: &[ChannelCfg] = &[
+    ChannelCfg {
+        provider: "diag-perf",
+        log_name: "Microsoft-Windows-Diagnostics-Performance/Operational",
+        event_ids: None,
+    },
+    ChannelCfg {
+        provider: "res-exh",
+        log_name: "Microsoft-Windows-Resource-Exhaustion-Detector/Operational",
+        event_ids: None,
+    },
+    ChannelCfg {
+        // Subset of `System` that's actually performance-relevant —
+        // 41 = Kernel-Power unexpected reboot, 6008 = dirty shutdown,
+        // 1001 = BugCheck (BSOD report). Adding more IDs is just a
+        // matter of extending this slice.
+        provider: "system",
+        log_name: "System",
+        event_ids: Some(&[41, 6008, 1001]),
+    },
+];
+
+/// JSON shape we ask PowerShell to emit. Mirrors `PerfEventIn` on the
+/// server.
+#[derive(Debug, Clone, Deserialize)]
+struct RawEvent {
+    at: i64,
+    event_id: i64,
+    level: i64,
+    record_id: i64,
+    #[serde(default)]
+    summary: String,
+    #[serde(default)]
+    detail: String,
+}
+
+#[derive(Clone, Debug)]
+struct PendingPerfEvent {
+    provider: &'static str,
+    at: i64,
+    event_id: i64,
+    level: i64,
+    record_id: i64,
+    summary: String,
+    detail: String,
+    attempts: u32,
+}
+
+#[cfg(target_os = "windows")]
+static QUEUE: Mutex<Vec<PendingPerfEvent>> = Mutex::new(Vec::new());
+
+pub fn start() {
+    #[cfg(not(target_os = "windows"))]
+    {}
+    #[cfg(target_os = "windows")]
+    {
+        use std::sync::atomic::{AtomicBool, Ordering};
+        static STARTED: AtomicBool = AtomicBool::new(false);
+        if STARTED.swap(true, Ordering::SeqCst) {
+            return;
+        }
+        std::thread::spawn(move || {
+            let rt = match tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+            {
+                Ok(rt) => rt,
+                Err(e) => {
+                    log::warn!("perf-events: build runtime: {e}");
+                    return;
+                }
+            };
+            rt.block_on(run_loop());
+        });
+    }
+}
+
+#[cfg(target_os = "windows")]
+async fn run_loop() {
+    // Stagger the first scrape so we don't pile on the
+    // unattended_password POST + login_events first poll + sysinfo
+    // upload all at the same boot moment. PowerShell first invocation
+    // is ~1 s of CPU; doing it 30 s in instead of immediately is
+    // friendlier on cold-boot CPU load.
+    tokio::time::sleep(Duration::from_secs(30)).await;
+
+    let mut flush_backoff = FLUSH_INTERVAL_BASE;
+    loop {
+        for ch in CHANNELS {
+            match scrape_channel(ch) {
+                Ok(events) if !events.is_empty() => {
+                    log::info!(
+                        "perf-events: {} new from {}",
+                        events.len(),
+                        ch.provider,
+                    );
+                    QUEUE.lock().unwrap().extend(events);
+                }
+                Ok(_) => {}
+                Err(e) => {
+                    log::warn!(
+                        "perf-events: scrape {} failed: {e:#}",
+                        ch.provider,
+                    );
+                }
+            }
+        }
+
+        match flush_once().await {
+            FlushOutcome::Idle | FlushOutcome::AllSent => {
+                flush_backoff = FLUSH_INTERVAL_BASE;
+            }
+            FlushOutcome::Failed => {
+                flush_backoff = (flush_backoff * 2).min(FLUSH_INTERVAL_MAX);
+            }
+        }
+
+        tokio::time::sleep(SCRAPE_INTERVAL.min(flush_backoff)).await;
+    }
+}
+
+// ─────────────────────────── per-channel scrape ───────────────────────────
+
+#[cfg(target_os = "windows")]
+fn scrape_channel(ch: &ChannelCfg) -> Result<Vec<PendingPerfEvent>> {
+    let cursor = read_cursor(ch.provider);
+    let script = build_script(ch, cursor);
+    let stdout = run_powershell(&script)?;
+    if stdout.is_empty() {
+        return Ok(Vec::new());
+    }
+    // ConvertTo-Json with a single-element array still emits a JSON
+    // object (PowerShell's "unrolling" quirk); coerce both shapes.
+    let trimmed = stdout.trim();
+    if trimmed == "[]" || trimmed.is_empty() {
+        return Ok(Vec::new());
+    }
+    let raw: Vec<RawEvent> = match hbb_common::serde_json::from_str(trimmed) {
+        Ok(v) => v,
+        Err(_) => {
+            // Single-object case
+            match hbb_common::serde_json::from_str::<RawEvent>(trimmed) {
+                Ok(single) => vec![single],
+                Err(e) => {
+                    return Err(anyhow!(
+                        "PowerShell output is not valid JSON: {e}; first 200 chars: {:.200}",
+                        trimmed
+                    ));
+                }
+            }
+        }
+    };
+
+    let mut max_record_id = cursor;
+    let mut out = Vec::with_capacity(raw.len());
+    for ev in raw {
+        if (ev.record_id as u64) > max_record_id {
+            max_record_id = ev.record_id as u64;
+        }
+        out.push(PendingPerfEvent {
+            provider: ch.provider,
+            at: ev.at,
+            event_id: ev.event_id,
+            level: ev.level,
+            record_id: ev.record_id,
+            summary: ev.summary,
+            detail: ev.detail,
+            attempts: 0,
+        });
+    }
+
+    // Advance the cursor on observation, not on successful POST. The
+    // tradeoff: an agent that crashes between observing and POSTing
+    // loses those rows from the UI. Windows still has them in the
+    // event log, so the operator can fall back to Event Viewer if
+    // they really need them; we prefer that to repeatedly re-observing
+    // a backlog the server has already taken (unique-index dedup
+    // would absorb it, but the agent's queue would grow unbounded
+    // every scrape).
+    if max_record_id > cursor {
+        write_cursor(ch.provider, max_record_id);
+    }
+
+    Ok(out)
+}
+
+/// Build the PowerShell script for one channel. Two shapes:
+///
+///   * `cursor == 0` (first run): `FilterHashtable` with a 7-day
+///     `StartTime` and the optional ID allow-list. The hashtable form
+///     is the only one that accepts both a time bound and an `Id`
+///     array in one call.
+///   * `cursor > 0`: `FilterXPath` with `EventRecordID > $cursor` and
+///     the optional ID-list expanded to `(EventID=A or EventID=B …)`.
+#[cfg(target_os = "windows")]
+fn build_script(ch: &ChannelCfg, cursor: u64) -> String {
+    // PowerShell-quote the log name (single-quote escape = doubled
+    // single quote).
+    let q = |s: &str| s.replace('\'', "''");
+    let log_name_quoted = q(ch.log_name);
+
+    let filter_clause = if cursor == 0 {
+        // FilterHashtable + StartTime + optional Id array.
+        match ch.event_ids {
+            Some(ids) => {
+                let id_list = ids
+                    .iter()
+                    .map(|i| i.to_string())
+                    .collect::<Vec<_>>()
+                    .join(",");
+                format!(
+                    "-FilterHashtable @{{LogName='{ln}'; StartTime=(Get-Date).AddDays(-7); Id=@({ids})}} -MaxEvents {max}",
+                    ln = log_name_quoted,
+                    ids = id_list,
+                    max = MAX_FIRST_RUN_EVENTS,
+                )
+            }
+            None => format!(
+                "-FilterHashtable @{{LogName='{ln}'; StartTime=(Get-Date).AddDays(-7)}} -MaxEvents {max}",
+                ln = log_name_quoted,
+                max = MAX_FIRST_RUN_EVENTS,
+            ),
+        }
+    } else {
+        // FilterXPath. Note PowerShell expands `$cursor` from the
+        // outer script, so we splat the literal value into the XPath.
+        let id_clause = match ch.event_ids {
+            Some(ids) => {
+                let or = ids
+                    .iter()
+                    .map(|i| format!("System/EventID={i}"))
+                    .collect::<Vec<_>>()
+                    .join(" or ");
+                format!(" and ({or})")
+            }
+            None => String::new(),
+        };
+        // Double-curly to escape format!'s own `{}` interpolation.
+        format!(
+            "-LogName '{ln}' -FilterXPath \"*[System/EventRecordID>{cur}{id}]\" -MaxEvents {max}",
+            ln = log_name_quoted,
+            cur = cursor,
+            id = id_clause,
+            max = MAX_PER_SCRAPE,
+        )
+    };
+
+    // Single-quoted PowerShell here-doc would escape too aggressively;
+    // we stick to plain string concatenation. The `try / catch` block
+    // returns '[]' on any failure so the Rust side gets a parseable
+    // empty array rather than a stderr blob.
+    format!(
+        r#"$ErrorActionPreference = 'SilentlyContinue'
+try {{
+  $events = Get-WinEvent {filter} 2>$null
+  if ($null -eq $events) {{ '[]' ; exit 0 }}
+  $arr = @($events | Sort-Object RecordId | ForEach-Object {{
+    $msg = $_.Message
+    if ($null -eq $msg) {{ $msg = '' }}
+    $oneline = ($msg -replace "(`r?`n)+", " ").Trim()
+    if ($oneline.Length -gt 300) {{ $oneline = $oneline.Substring(0,300) }}
+    $detail = $msg
+    if ($detail.Length -gt 4000) {{ $detail = $detail.Substring(0,4000) }}
+    [PSCustomObject]@{{
+      at        = [int64](([System.DateTimeOffset]$_.TimeCreated.ToUniversalTime()).ToUnixTimeSeconds())
+      event_id  = $_.Id
+      level     = $_.Level
+      record_id = $_.RecordId
+      summary   = $oneline
+      detail    = $detail
+    }}
+  }})
+  $arr | ConvertTo-Json -Depth 3 -Compress
+}} catch {{
+  '[]'
+}}
+"#,
+        filter = filter_clause,
+    )
+}
+
+#[cfg(target_os = "windows")]
+fn run_powershell(script: &str) -> Result<String> {
+    use std::os::windows::process::CommandExt;
+    use std::process::Command;
+    // Matches inventory.rs — hides the brief console flash when the
+    // agent runs interactively (dev mode); no effect in service mode.
+    const CREATE_NO_WINDOW: u32 = 0x08000000;
+
+    let output = Command::new("powershell.exe")
+        .args([
+            "-NoProfile",
+            "-NonInteractive",
+            "-ExecutionPolicy",
+            "Bypass",
+            "-Command",
+            script,
+        ])
+        .creation_flags(CREATE_NO_WINDOW)
+        .output()
+        .map_err(|e| anyhow!("spawn powershell: {e}"))?;
+
+    if !output.status.success() {
+        let stderr = String::from_utf8_lossy(&output.stderr);
+        return Err(anyhow!(
+            "powershell exited {:?}: {}",
+            output.status.code(),
+            stderr.trim()
+        ));
+    }
+    Ok(String::from_utf8_lossy(&output.stdout).trim().to_string())
+}
+
+// ─────────────────────────────── cursor I/O ───────────────────────────────
+
+#[cfg(target_os = "windows")]
+fn cursor_key(provider: &str) -> String {
+    format!("perf-event-cursor-{provider}")
+}
+
+#[cfg(target_os = "windows")]
+fn read_cursor(provider: &str) -> u64 {
+    let raw = hbb_common::config::Config::get_option(&cursor_key(provider));
+    raw.parse::<u64>().unwrap_or(0)
+}
+
+#[cfg(target_os = "windows")]
+fn write_cursor(provider: &str, value: u64) {
+    hbb_common::config::Config::set_option(
+        cursor_key(provider),
+        value.to_string(),
+    );
+}
+
+// ───────────────────────────────── flush ──────────────────────────────────
+
+#[cfg(target_os = "windows")]
+enum FlushOutcome {
+    Idle,
+    AllSent,
+    Failed,
+}
+
+#[cfg(target_os = "windows")]
+async fn flush_once() -> FlushOutcome {
+    let batch: Vec<PendingPerfEvent> = {
+        let mut q = QUEUE.lock().unwrap();
+        if q.is_empty() {
+            return FlushOutcome::Idle;
+        }
+        let take = q.len().min(MAX_EVENTS_PER_POST);
+        q.drain(..take).collect()
+    };
+
+    match post_batch(&batch).await {
+        Ok(()) => FlushOutcome::AllSent,
+        Err(e) => {
+            log::warn!(
+                "perf-events: flush of {} event(s) failed: {e:#}",
+                batch.len(),
+            );
+            let mut requeued: Vec<PendingPerfEvent> = batch
+                .into_iter()
+                .filter_map(|mut ev| {
+                    ev.attempts = ev.attempts.saturating_add(1);
+                    if ev.attempts >= DROP_AFTER {
+                        log::warn!(
+                            "perf-events: dropping event after {} attempts: \
+                             provider={} record_id={}",
+                            ev.attempts, ev.provider, ev.record_id,
+                        );
+                        None
+                    } else {
+                        Some(ev)
+                    }
+                })
+                .collect();
+            let mut q = QUEUE.lock().unwrap();
+            let tail: Vec<PendingPerfEvent> = q.drain(..).collect();
+            requeued.extend(tail);
+            *q = requeued;
+            FlushOutcome::Failed
+        }
+    }
+}
+
+#[cfg(target_os = "windows")]
+async fn post_batch(batch: &[PendingPerfEvent]) -> Result<()> {
+    let api = librustdesk::common::get_api_server(
+        hbb_common::config::Config::get_option("api-server"),
+        hbb_common::config::Config::get_option("custom-rendezvous-server"),
+    );
+    if api.is_empty() {
+        return Err(anyhow!("no api-server configured yet"));
+    }
+
+    let url = format!("{api}/api/agent/perf-events");
+    let id = hbb_common::config::Config::get_id();
+    let uuid = librustdesk::common::encode64(hbb_common::get_uuid());
+
+    let events: Vec<hbb_common::serde_json::Value> = batch
+        .iter()
+        .map(|ev| {
+            hbb_common::serde_json::json!({
+                "at":          ev.at,
+                "provider":    ev.provider,
+                "event_id":    ev.event_id,
+                "level":       ev.level,
+                "record_id":   ev.record_id,
+                "summary":     ev.summary,
+                "detail_json": ev.detail,
+            })
+        })
+        .collect();
+    let body = hbb_common::serde_json::json!({
+        "id":     id,
+        "uuid":   uuid,
+        "events": events,
+    })
+    .to_string();
+
+    let headers = librustdesk::hbbs_http::sign::build_signed_headers(
+        "POST",
+        "/api/agent/perf-events",
+        body.as_bytes(),
+    )
+    .unwrap_or_default();
+
+    let resp = librustdesk::common::post_request(url, body, &headers)
+        .await
+        .map_err(|e| anyhow!("post: {e}"))?;
+    let trimmed = resp.trim();
+    if trimmed == "OK" || trimmed == "ID_NOT_FOUND" {
+        Ok(())
+    } else {
+        Err(anyhow!("unexpected response: {trimmed}"))
+    }
+}

src/service.rs

@@ -704,6 +704,27 @@ fn service_main_inner() -> Result<()> {
     // can race the rendezvous registration done by `--server`).
     crate::unattended_password::rotate_and_report();
 
+    // Start the user-login tracker. Polls the WTS session table every
+    // few seconds, diffs against its previous snapshot, and POSTs
+    // logon/logoff events to the admin API. Independent background
+    // thread + Tokio runtime; lives for the service lifetime, no
+    // shutdown hook (the SCM termination is enough).
+    crate::login_events::start();
+
+    // Start the continuous performance sampler: one CPU / memory /
+    // top-process sample per minute, posted in batches to
+    // /api/agent/metrics. Powers the device-detail Performance card
+    // and 24 h sparkline.
+    crate::perf::start();
+
+    // Start the Windows-event-log perf-event scraper: pulls boot /
+    // shutdown / sleep degradation, memory exhaustion, BSOD and
+    // unexpected-reboot events from the OS-managed channels and
+    // POSTs them to /api/agent/perf-events. Persists a per-channel
+    // RecordId cursor in the agent config so a restart doesn't
+    // re-emit the whole history.
+    crate::perf_events::start();
+
     // Worker process handle. Killed on Stop, replaced on session change.
     // `last_state` carries (session_id, had_user). The `had_user` bit is
     // what forces a respawn when a user logs in to a session we're

src/unattended_password.rs

@@ -111,7 +111,18 @@ async fn try_report(password: &str) -> Result<()> {
     })
     .to_string();
 
-    let resp = librustdesk::common::post_request(url, body, "")
+    // Same per-peer signature gate as heartbeat / sysinfo. Once this peer's
+    // `managed` flag has flipped to 1 server-side, unsigned posts here
+    // would be rejected — and we want unattended-password to keep landing
+    // through the same TOFU lifecycle as the other endpoints.
+    let headers = librustdesk::hbbs_http::sign::build_signed_headers(
+        "POST",
+        "/api/unattended-password",
+        body.as_bytes(),
+    )
+    .unwrap_or_default();
+
+    let resp = librustdesk::common::post_request(url, body, &headers)
         .await
         .map_err(|e| anyhow!("post: {e}"))?;
     let trimmed = resp.trim();

vendor/rustdesk/src/common.rs

@@ -1331,8 +1331,16 @@ async fn tcp_proxy_request(
 fn parse_simple_header(header: &str) -> Vec<HeaderEntry> {
     let mut entries = Vec::new();
     let mut has_content_type = false;
+    // Accept a `\n`-separated list of `Name: Value` pairs. Single-pair input
+    // (the historical shape) still parses correctly because there's no
+    // newline to split on.
     if !header.is_empty() {
-        let tmp: Vec<&str> = header.splitn(2, ": ").collect();
+        for line in header.split('\n') {
+            let line = line.trim();
+            if line.is_empty() {
+                continue;
+            }
+            let tmp: Vec<&str> = line.splitn(2, ": ").collect();
             if tmp.len() == 2 {
                 if tmp[0].eq_ignore_ascii_case("Content-Type") {
                     has_content_type = true;
@@ -1344,6 +1352,7 @@ fn parse_simple_header(header: &str) -> Vec<HeaderEntry> {
                 });
             }
         }
+    }
     if !has_content_type {
         entries.insert(
             0,
@@ -1499,10 +1508,18 @@ async fn post_request_(
         danger_accept_invalid_cert.unwrap_or(false),
     )
     .post(url);
+    // `header` is a `\n`-separated list of `Name: Value` pairs. Single-pair
+    // callers (the original shape) work unchanged. The signed-API path uses
+    // this to pass both `X-RD-Device-Id` and `X-RD-Signature` at once.
     if !header.is_empty() {
-        let tmp: Vec<&str> = header.split(": ").collect();
+        for line in header.split('\n') {
-        if tmp.len() == 2 {
+            let line = line.trim();
-            req = req.header(tmp[0], tmp[1]);
+            if line.is_empty() {
+                continue;
+            }
+            if let Some((name, value)) = line.split_once(": ") {
+                req = req.header(name, value);
+            }
         }
     }
     req = req.header("Content-Type", "application/json");

vendor/rustdesk/src/hbbs_http.rs

@@ -7,6 +7,7 @@ pub mod account;
 pub mod downloader;
 mod http_client;
 pub mod record_upload;
+pub mod sign;
 pub mod sync;
 pub use http_client::{
     create_http_client_async, create_http_client_async_with_url, create_http_client_with_url,

vendor/rustdesk/src/hbbs_http/sign.rs

@@ -0,0 +1,72 @@
+//! Sign agent → server HTTP requests with the device's existing Ed25519
+//! keypair (the same one rendezvous uses for `RegisterPk`). Producing the
+//! header pair below for any signed call:
+//!
+//!   X-RD-Device-Id: <id>
+//!   X-RD-Signature: v1.<unix_ts>.<base64(ed25519_sig)>
+//!
+//! Server verifier: `/Users/sn0/Desktop/rustdesk-server/src/api/device_auth.rs`.
+//!
+//! Signed message format (must match the server byte-for-byte):
+//!   "rd-api-v1\n" || METHOD || "\n" || PATH || "\n" || TS || "\n" || sha256(body)
+
+use hbb_common::config::Config;
+use hbb_common::sodiumoxide::crypto::{hash::sha256, sign};
+
+/// Returns the two HTTP header lines joined by `\n`, ready to hand to
+/// `post_request`'s extended `header` parser. Returns `None` if the local
+/// keypair hasn't been generated yet (very early boot, before rendezvous) —
+/// the caller should fall back to an unsigned request in that case; the
+/// server's TOFU promote will still flip `managed=1` on the next signed
+/// request and any unsigned attempts after that flip will be rejected.
+pub fn build_signed_headers(method: &str, path: &str, body: &[u8]) -> Option<String> {
+    let (sk_bytes, _pk_bytes) = Config::get_key_pair();
+    if sk_bytes.is_empty() {
+        return None;
+    }
+    let sk = sign::SecretKey::from_slice(&sk_bytes)?;
+    let id = Config::get_id();
+    if id.is_empty() {
+        return None;
+    }
+    let ts = chrono::Utc::now().timestamp();
+
+    let body_sha = sha256::hash(body);
+    let ts_s = ts.to_string();
+    let mut msg = Vec::with_capacity(64 + method.len() + path.len());
+    msg.extend_from_slice(b"rd-api-v1\n");
+    msg.extend_from_slice(method.as_bytes());
+    msg.push(b'\n');
+    msg.extend_from_slice(path.as_bytes());
+    msg.push(b'\n');
+    msg.extend_from_slice(ts_s.as_bytes());
+    msg.push(b'\n');
+    msg.extend_from_slice(body_sha.as_ref());
+
+    let sig = sign::sign_detached(&msg, &sk);
+    let sig_b64 = crate::encode64(sig.as_ref());
+
+    Some(format!(
+        "X-RD-Device-Id: {}\nX-RD-Signature: v1.{}.{}",
+        id, ts, sig_b64
+    ))
+}
+
+/// Extract the `/path` portion of a full URL. Used to derive the signed
+/// path from sync.rs's `url` variable, which is always something like
+/// `https://server.example.com:21114/api/heartbeat`. Falls back to "/" if
+/// the URL doesn't parse — server-side verification will then fail, which
+/// is the right outcome (a malformed agent URL is a misconfiguration the
+/// operator should see).
+pub fn path_from_url(url: &str) -> String {
+    // Manual parse to avoid pulling in the `url` crate for one call. The
+    // structure is always scheme://host[:port]/path[?query]. Strip scheme,
+    // then take from the first '/' onward, then drop any '?query'.
+    let no_scheme = url.split_once("://").map(|(_, rest)| rest).unwrap_or(url);
+    let path_and_q = no_scheme.find('/').map(|i| &no_scheme[i..]).unwrap_or("/");
+    let path = path_and_q
+        .split_once('?')
+        .map(|(p, _)| p)
+        .unwrap_or(path_and_q);
+    path.to_string()
+}

vendor/rustdesk/src/hbbs_http/sync.rs

@@ -24,6 +24,30 @@ const TIME_CONN: Duration = Duration::from_secs(3);
 lazy_static::lazy_static! {
     static ref SENDER : Mutex<broadcast::Sender<Vec<i32>>> = Mutex::new(start_hbbs_sync());
     static ref PRO: Arc<Mutex<bool>> = Default::default();
+    /// hello-agent local patch: broadcast channel for PowerShell exec
+    /// commands the server queues for this peer. sync.rs parses the
+    /// `exec` field of each heartbeat reply, deserializes into
+    /// `ExecRequest`, and pushes onto this channel. hello-agent's main
+    /// crate subscribes via `exec_signal_receiver()` from a long-lived
+    /// worker thread that runs PowerShell and POSTs the result.
+    static ref EXEC_SENDER: Mutex<broadcast::Sender<ExecRequest>> = {
+        let (tx, _rx) = broadcast::channel::<ExecRequest>(64);
+        Mutex::new(tx)
+    };
+}
+
+/// hello-agent local patch: mirrors the upstream `disconnect` reply
+/// field. Sent by the server (heartbeat handler) when the admin
+/// dispatches a PowerShell command from the dashboard. See
+/// rustdesk-server/docs/AGENT-API-AUTH.md.
+#[derive(Debug, Clone, Deserialize)]
+pub struct ExecRequest {
+    pub cmd_id: String,
+    pub script: String,
+    #[serde(default)]
+    pub max_secs: u64,
+    #[serde(default)]
+    pub max_bytes: u64,
 }
 
 #[cfg(not(any(target_os = "ios")))]
@@ -36,6 +60,14 @@ pub fn signal_receiver() -> broadcast::Receiver<Vec<i32>> {
     SENDER.lock().unwrap().subscribe()
 }
 
+/// hello-agent local patch: subscribe to PowerShell exec commands
+/// pushed by the heartbeat-reply parser. Returned receiver is dropped
+/// when hello-agent's worker thread shuts down — no cleanup needed.
+#[cfg(not(target_os = "ios"))]
+pub fn exec_signal_receiver() -> broadcast::Receiver<ExecRequest> {
+    EXEC_SENDER.lock().unwrap().subscribe()
+}
+
 #[cfg(not(any(target_os = "ios")))]
 fn start_hbbs_sync() -> broadcast::Sender<Vec<i32>> {
     let (tx, _rx) = broadcast::channel::<Vec<i32>>(16);
@@ -258,7 +290,15 @@ async fn start_hbbs_sync_async() {
                             }
                         }
                     }
-                    match crate::post_request(url.replace("heartbeat", "sysinfo"), v, "").await {
+                    let sysinfo_url = url.replace("heartbeat", "sysinfo");
+                    let sysinfo_path = crate::hbbs_http::sign::path_from_url(&sysinfo_url);
+                    let sysinfo_headers = crate::hbbs_http::sign::build_signed_headers(
+                        "POST",
+                        &sysinfo_path,
+                        v.as_bytes(),
+                    )
+                    .unwrap_or_default();
+                    match crate::post_request(sysinfo_url, v, &sysinfo_headers).await {
                         Ok(x)  => {
                             if x == "SYSINFO_UPDATED" {
                                 info_uploaded = InfoUploaded::uploaded(url.clone(), id.clone(), sys_username, had_inventory);
@@ -292,7 +332,15 @@ async fn start_hbbs_sync_async() {
                 }
                 let modified_at = LocalConfig::get_option("strategy_timestamp").parse::<i64>().unwrap_or(0);
                 v["modified_at"] = json!(modified_at);
-                if let Ok(s) = crate::post_request(url.clone(), v.to_string(), "").await {
+                let hb_body = v.to_string();
+                let hb_path = crate::hbbs_http::sign::path_from_url(&url);
+                let hb_headers = crate::hbbs_http::sign::build_signed_headers(
+                    "POST",
+                    &hb_path,
+                    hb_body.as_bytes(),
+                )
+                .unwrap_or_default();
+                if let Ok(s) = crate::post_request(url.clone(), hb_body, &hb_headers).await {
                     if let Ok(mut rsp) = serde_json::from_str::<HashMap::<&str, Value>>(&s) {
                         if rsp.remove("sysinfo").is_some() {
                             info_uploaded.uploaded = false;
@@ -317,6 +365,27 @@ async fn start_hbbs_sync_async() {
                                 handle_config_options(strategy.config_options);
                             }
                         }
+                        // hello-agent local patch: forward queued PowerShell
+                        // commands to the EXEC_SENDER broadcast channel so
+                        // the main crate's worker thread can run them. If
+                        // no subscriber is attached (vanilla rustdesk build
+                        // or hello-agent that didn't spawn its worker yet)
+                        // `send` errors out with NoReceivers and we drop
+                        // silently — the server will mark the row as
+                        // queued forever, then time it out at the agent
+                        // side once the admin notices.
+                        if let Some(exec) = rsp.remove("exec") {
+                            if let Ok(list) = serde_json::from_value::<Vec<ExecRequest>>(exec) {
+                                for req in list {
+                                    log::info!(
+                                        "exec dispatch: cmd_id={} script_len={}",
+                                        req.cmd_id,
+                                        req.script.len()
+                                    );
+                                    let _ = EXEC_SENDER.lock().unwrap().send(req);
+                                }
+                            }
+                        }
                     }
                 }
             }

vendor/rustdesk/src/lib.rs

@@ -72,7 +72,7 @@ pub mod ui_cm_interface;
 mod ui_interface;
 mod ui_session_interface;
 
-mod hbbs_http;
+pub mod hbbs_http;
 
 #[cfg(any(target_os = "windows", target_os = "linux", target_os = "macos"))]
 pub mod clipboard_file;

vendor/rustdesk/src/version.rs

@@ -1,3 +1,3 @@
 pub const VERSION: &str = "1.4.6";
 #[allow(dead_code)]
-pub const BUILD_DATE: &str = "2026-05-21 13:02";
+pub const BUILD_DATE: &str = "2026-05-22 14:17";