scv_server/
daemon.rs

1//! The daemon: the socket listener and its lock, supervision of components
2//! and delegated runs, and shutdown. [`run_stdio`] serves one connection
3//! without a daemon.
4
5use std::{path::Path, sync::Arc, time::Duration};
6
7use anyhow::{Context, Result, anyhow};
8use scv_client::Layout;
9use scv_tools::delegation::{self as delegations, DelegationRegistry};
10use tokio::{
11    net::{UnixListener, UnixStream},
12    sync::Mutex,
13};
14use tokio_util::{sync::CancellationToken, task::TaskTracker};
15
16use crate::{
17    components,
18    config::{ConfigOverrides, Instance},
19    confirm,
20    connection::run_managed,
21    restart,
22};
23
24/// Serve one session over stdin and stdout for the instance at `layout`.
25pub async fn run_stdio(layout: &Layout, overrides: ConfigOverrides) -> Result<()> {
26    let stdin = tokio::io::stdin();
27    let stdout = tokio::io::stdout();
28    let tasks = TaskTracker::new();
29    let registry = instance_delegations(layout);
30    // Without a daemon, a later `scv exec` is what cleans up after an earlier
31    // one that was killed; this runs alongside the session.
32    tokio::spawn(reconcile_delegations(Arc::clone(&registry)));
33    let result = run_managed(
34        stdin,
35        stdout,
36        Instance {
37            layout: layout.clone(),
38            overrides,
39        },
40        None,
41        registry,
42        CancellationToken::new(),
43        tasks.clone(),
44    )
45    .await;
46    tasks.close();
47    tasks.wait().await;
48    result
49}
50
51/// Run the authoritative server on the instance's Unix socket.
52pub async fn run_socket(layout: &Layout, overrides: ConfigOverrides) -> Result<()> {
53    let socket = layout.socket();
54    let path = socket.as_path();
55    let instance = Instance {
56        layout: layout.clone(),
57        overrides,
58    };
59    if let Some(parent) = path.parent() {
60        tokio::fs::create_dir_all(parent)
61            .await
62            .context("create SCV socket directory")?;
63        #[cfg(unix)]
64        {
65            use std::os::unix::fs::PermissionsExt;
66            std::fs::set_permissions(parent, std::fs::Permissions::from_mode(0o700))
67                .context("secure SCV socket directory")?;
68        }
69    }
70    let _lock = SocketLock::acquire(path)?;
71    // Nothing reads an older release's files; say so once rather than let
72    // them look like live configuration.
73    if let Ok(strays) = layout.strays() {
74        for stray in strays.into_iter().filter(|stray| stray.legacy) {
75            tracing::warn!(
76                "{} is from an older SCV layout and is not used; see `scv config show`",
77                stray.path.display()
78            );
79        }
80    }
81    if path.exists() {
82        if UnixStream::connect(path).await.is_ok() {
83            return Err(anyhow!(
84                "SCV server is already running at {}",
85                path.display()
86            ));
87        }
88        use std::os::unix::fs::FileTypeExt;
89        if !std::fs::symlink_metadata(path)?.file_type().is_socket() {
90            return Err(anyhow!(
91                "refusing to remove a non-socket at SCV socket path"
92            ));
93        }
94        tokio::fs::remove_file(path)
95            .await
96            .with_context(|| format!("remove stale SCV socket {}", path.display()))?;
97    }
98    let listener = UnixListener::bind(path)
99        .with_context(|| format!("bind SCV server socket {}", path.display()))?;
100    #[cfg(unix)]
101    {
102        use std::os::unix::fs::PermissionsExt;
103        std::fs::set_permissions(path, std::fs::Permissions::from_mode(0o600))
104            .context("secure SCV socket")?;
105    }
106    let hub = scv_channels::hub::Hub::new(Some(layout.last_owner()));
107    // Before any account starts: its recovery needs to know whether this
108    // start is a planned restart.
109    let startup = restart::startup(layout, &hub);
110    let components = Arc::new(Mutex::new(components::Components::with_hub(
111        instance.clone(),
112        std::env::current_dir()?,
113        Arc::clone(&hub),
114    )));
115    let registry = instance_delegations(layout);
116    // Descendants a delegated agent leaves behind reparent to the daemon, not init.
117    if !delegations::become_child_subreaper() {
118        tracing::debug!("SCV daemon is not a child subreaper on this platform");
119    }
120    let cancellation = CancellationToken::new();
121    let restarter = restart::Restarter::new(
122        instance.clone(),
123        Arc::clone(&hub),
124        Arc::clone(&registry),
125        &components,
126        cancellation.clone(),
127    );
128    let confirmer = confirm::Confirmer::new(
129        hub,
130        Arc::clone(&registry),
131        restarter.notifier().clone(),
132        cancellation.clone(),
133    );
134    {
135        let mut components = components.lock().await;
136        components.set_restarter(Arc::clone(&restarter));
137        components.set_confirmer(confirmer);
138    }
139    let notices = tokio::spawn(restart::announce(
140        layout.clone(),
141        startup,
142        restarter.notifier().clone(),
143        cancellation.clone(),
144    ));
145    let _notices_abort = AbortGuard(notices.abort_handle());
146    let monitor = tokio::spawn(restart::monitor(
147        restarter.notifier().clone(),
148        cancellation.clone(),
149    ));
150    let _monitor_abort = AbortGuard(monitor.abort_handle());
151    let delegation_registry = Arc::clone(&registry);
152    let delegation_cancel = cancellation.clone();
153    let mut delegation_task = tokio::spawn(async move {
154        // The first tick is immediate: orphans from before a restart go first.
155        let mut interval = tokio::time::interval(DELEGATION_RECONCILE_INTERVAL);
156        loop {
157            tokio::select! {
158                biased;
159                () = delegation_cancel.cancelled() => break,
160                _ = interval.tick() => {
161                    reconcile_delegations(Arc::clone(&delegation_registry)).await;
162                    let zombies = delegations::reap_orphaned_zombies();
163                    if zombies > 0 {
164                        tracing::debug!("Reaped {zombies} exited orphan processes");
165                    }
166                }
167            }
168        }
169    });
170    let _delegation_abort = AbortGuard(delegation_task.abort_handle());
171    let tasks = TaskTracker::new();
172    let mut clients = tokio::task::JoinSet::new();
173    let refresh_components = components.clone();
174    let refresh_cancel = cancellation.clone();
175    let mut refresh_task = tokio::spawn(async move {
176        let mut refresh = tokio::time::interval(Duration::from_secs(2));
177        loop {
178            tokio::select! {
179                biased;
180                () = refresh_cancel.cancelled() => break,
181                _ = refresh.tick() => {
182                    tokio::select! {
183                        biased;
184                        () = refresh_cancel.cancelled() => break,
185                        result = async { refresh_components.lock().await.reconcile().await } => {
186                            if result.is_err() { tracing::warn!("Component account discovery failed"); }
187                        }
188                    }
189                }
190            }
191        }
192    });
193    let _refresh_abort = AbortGuard(refresh_task.abort_handle());
194    let mut terminate = tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())?;
195    let result = loop {
196        tokio::select! {
197            accepted = listener.accept() => {
198                let (stream, _) = match accepted { Ok(value) => value, Err(error) => break Err(error.into()) };
199                let instance = instance.clone();
200                let components = components.clone();
201                let registry = Arc::clone(&registry);
202                let cancellation = cancellation.clone();
203                let tasks = tasks.clone();
204                clients.spawn(async move {
205                    let (reader, writer) = stream.into_split();
206                    if let Err(error) = run_managed(reader, writer, instance, Some(components), registry, cancellation, tasks).await {
207                        tracing::warn!(error = format!("{error:#}"), "SCV socket client stopped");
208                    }
209                });
210            }
211            _ = clients.join_next(), if !clients.is_empty() => {},
212            _ = tokio::signal::ctrl_c() => break Ok(()),
213            _ = terminate.recv() => break Ok(()),
214        }
215    };
216    drop(listener);
217    cancellation.cancel();
218    let _ = (&mut refresh_task).await;
219    let _ = (&mut delegation_task).await;
220    components.lock().await.shutdown().await;
221    if tokio::time::timeout(Duration::from_secs(8), async {
222        while clients.join_next().await.is_some() {}
223    })
224    .await
225    .is_err()
226    {
227        clients.abort_all();
228        while clients.join_next().await.is_some() {}
229    }
230    tasks.close();
231    tasks.wait().await;
232    let _ = tokio::fs::remove_file(path).await;
233    restart::clean_shutdown(layout);
234    result
235}
236
237pub(crate) const DELEGATION_RECONCILE_INTERVAL: Duration = Duration::from_secs(60);
238
239/// The delegation registry for this process's SCV instance.
240pub(crate) fn instance_delegations(layout: &Layout) -> Arc<DelegationRegistry> {
241    Arc::new(DelegationRegistry::new(layout))
242}
243
244/// Stop orphaned delegations of this instance and log what was stopped.
245pub(crate) async fn reconcile_delegations(registry: Arc<DelegationRegistry>) {
246    let report = registry.reconcile().await;
247    if !report.reaped.is_empty() {
248        tracing::info!(
249            "Reaped {} orphaned delegations: {}",
250            report.reaped.len(),
251            report.reaped.join(", ")
252        );
253    }
254    if report.removed > 0 {
255        tracing::debug!(
256            "Removed {} delegation records whose processes had exited",
257            report.removed
258        );
259    }
260    if report.stale_markers > 0 {
261        tracing::debug!(
262            "Removed {} conversation markers whose SCV process had exited",
263            report.stale_markers
264        );
265    }
266}
267
268/// A persistent advisory lock closes the stale-socket unlink/bind race.
269pub(crate) struct SocketLock(std::fs::File);
270impl SocketLock {
271    pub(crate) fn acquire(socket: &Path) -> Result<Self> {
272        use std::os::unix::{fs::OpenOptionsExt, io::AsRawFd};
273        let file = std::fs::OpenOptions::new()
274            .read(true)
275            .write(true)
276            .create(true)
277            .truncate(false)
278            .mode(0o600)
279            .custom_flags(libc::O_NOFOLLOW)
280            .open(socket.with_extension("lock"))?;
281        // SAFETY: flock operates on this owned, live file descriptor.
282        if unsafe { libc::flock(file.as_raw_fd(), libc::LOCK_EX | libc::LOCK_NB) } != 0 {
283            return Err(anyhow!("SCV daemon already owns this socket"));
284        }
285        Ok(Self(file))
286    }
287}
288impl Drop for SocketLock {
289    fn drop(&mut self) {
290        use std::os::unix::io::AsRawFd;
291        // SAFETY: the descriptor remains live until this drop returns.
292        unsafe {
293            libc::flock(self.0.as_raw_fd(), libc::LOCK_UN);
294        }
295    }
296}
297
298/// Aborts a task when dropped, so a task the daemon spawned never outlives it.
299pub(crate) struct AbortGuard(pub(crate) tokio::task::AbortHandle);
300impl Drop for AbortGuard {
301    fn drop(&mut self) {
302        self.0.abort();
303    }
304}
305
306#[cfg(test)]
307mod tests;
scv_server/daemon.rs

scv_server/
daemon.rs
