chore: Release dkl version 1.2.3

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "dkl"
-version = "1.1.0"
+version = "1.2.3"
 edition = "2024"
 
 [profile.release]
@@ -23,13 +23,14 @@ eyre = "0.6.12"
 fastrand = "2.3.0"
 futures = "0.3.31"
 futures-util = "0.3.31"
+getrandom = "0.4.2"
 glob = "0.3.2"
 hex = "0.4.3"
 human-units = "0.5.3"
 log = "0.4.27"
 lz4 = "1.28.1"
 memchr = "2.8.0"
-nix = { version = "0.31.2", features = ["process", "signal", "user"] }
+nix = { version = "0.31.2", features = ["fs", "process", "signal", "user"] }
 openssl = "0.10.73"
 page_size = "0.6.0"
 reqwest = { version = "0.13.1", features = ["json", "stream", "native-tls", "socks"], default-features = false }
@@ -37,9 +38,9 @@ rpassword = "7.4.0"
 rust-argon2 = "3.0.0"
 serde = { version = "1.0.219", features = ["derive"] }
 serde_json = "1.0.140"
-serde_yaml = "0.9.34"
+serde_yaml = { version = "0.10.0", package = "serde_yaml_ng" }
 signal-hook = "0.4.4"
 tabled = "0.20.0"
 thiserror = "2.0.12"
-tokio = { version = "1.45.1", features = ["fs", "io-std", "macros", "process", "rt"] }
+tokio = { version = "1.45.1", features = ["fs", "io-std", "macros", "process", "rt", "signal"] }
 

Dockerfile

@@ -1,4 +1,4 @@
-from mcluseau/rust:1.94.0 as build
+from mcluseau/rust:1.97.0 as build
 
 workdir /app
 copy . .
@@ -10,6 +10,6 @@ run \
  && find target/release -maxdepth 1 -type f -executable -exec cp -v {} /dist/ +
 
 # ------------------------------------------------------------------------
-from alpine:3.23
+from alpine:3.24.0
 copy --from=build /dist/ /bin/
 

build-dist

@@ -1 +1 @@
-cargo install --path . --root dist
+cargo install --locked --path . --root dist

release.sh

@@ -2,15 +2,17 @@ set -ex
 tag=$(git describe --always)
 repo=novit.tech/direktil/dkl:$tag
 
-docker build --push --platform=linux/amd64,linux/arm64 . -t $repo
-
 publish() {
     arch=$1
     pf=$2
 
+    docker build --push --platform=$pf . -t $repo
+
+    for bin in dkl dls; do
     curl --user $(jq '.auths["novit.tech"].auth' ~/.docker/config.json -r |base64 -d) \
-        --upload-file <(docker run --rm --platform $pf $repo cat /bin/dkl) \
+        --upload-file <(docker run --rm --platform $pf $repo cat /bin/$bin) \
-        https://novit.tech/api/packages/direktil/generic/dkl/$tag/dkl.$arch
+        https://novit.tech/api/packages/direktil/generic/dkl/$tag/$bin.$arch
+    done
 }
 
 publish x86_64 linux/amd64

src/apply.rs

@@ -1,12 +1,20 @@
-use eyre::Result;
+use eyre::{Result, format_err, eyre};
 use log::info;
 use std::path::Path;
 use tokio::fs;
 
-use crate::base64_decode;
+use crate::File;
 
-pub async fn files(files: &[crate::File], root: &str, dry_run: bool) -> Result<()> {
+pub async fn files(files: &[File], root: &str, dry_run: bool) -> Result<()> {
-    for file in files {
+    for f in files {
+        if let Err(e) = file(f, root, dry_run).await {
+            return Err(format_err!("{}: {e}", f.path));
+        }
+    }
+    Ok(())
+}
+
+pub async fn file(file: &File, root: &str, dry_run: bool) -> Result<()> {
     let path = chroot(root, &file.path);
     let path = Path::new(&path);
 
@@ -14,60 +22,49 @@ pub async fn files(files: &[crate::File], root: &str, dry_run: bool) -> Result<(
         fs::create_dir_all(parent).await?;
     }
 
+    let kind = file.kind();
+    let content = kind.content()?;
+
     use crate::FileKind as K;
-        match &file.kind {
+    match kind.as_ref() {
-            K::Content(content) => {
+        K::Skip => {
+            info!("{}: kind is skip", file.path);
+            return Ok(());
+        }
+        K::Content(_) | K::Content64(_) | K::Parts(_) => {
+            let content = content.expect("this file kind should have content");
             if dry_run {
-                    info!(
+                info!("would create {} ({} bytes)", file.path, content.len());
-                        "would create {} ({} bytes from content)",
-                        file.path,
-                        content.len()
-                    );
             } else {
-                    fs::write(path, content.as_bytes()).await?;
+                fs::write(path, &content).await.map_err(|e| eyre!("write {}: {e}", path.display()))?;
             }
         }
-            K::Content64(content) => {
+        K::Dir => {
-                let content = base64_decode(content)?;
-                if dry_run {
-                    info!(
-                        "would create {} ({} bytes from content64)",
-                        file.path,
-                        content.len()
-                    );
-                } else {
-                    fs::write(path, content).await?
-                }
-            }
-            K::Dir(true) => {
             if dry_run {
                 info!("would create {} (directory)", file.path);
             } else {
-                    fs::create_dir(path).await?;
+                fs::create_dir(path).await.map_err(|e| eyre!("create dir {}: {e}", path.display()))?;
             }
         }
-            K::Dir(false) => {} // shouldn't happen, but semantic is to ignore
         K::Symlink(tgt) => {
             if dry_run {
                 info!("would create {} (symlink to {})", file.path, tgt);
             } else {
-                    fs::symlink(tgt, path).await?;
+                let _ = fs::remove_file(path).await; // we're ln --force
+                fs::symlink(tgt, path).await.map_err(|e| eyre!("symlink {} -> {tgt}: {e}", path.display()))?;
             }
         }
     }
 
     if dry_run {
-            continue;
+        return Ok(());
     }
 
-        match file.kind {
+    if !file.is_symlink() {
-            K::Symlink(_) => {}
+        set_perms(path, file.mode).await?;
-            _ => set_perms(path, file.mode).await?,
     }
 
     info!("created {}", file.path);
-    }
-
     Ok(())
 }
 

src/bin/dkl.rs

@@ -1,5 +1,5 @@
 use clap::{CommandFactory, Parser, Subcommand};
-use eyre::{format_err, Result};
+use eyre::{Result, format_err};
 use human_units::Duration;
 use log::{debug, error};
 use std::net::SocketAddr;
@@ -9,6 +9,8 @@ use tokio::fs;
 #[derive(Parser)]
 #[command()]
 struct Cli {
+    #[arg(long)]
+    log_to: Option<PathBuf>,
     #[command(subcommand)]
     command: Command,
 }
@@ -32,17 +34,17 @@ enum Command {
     Logger {
         /// Path where the logs are stored
         #[arg(long, short = 'p', default_value = "/var/log", env = "DKL_LOG_PATH")]
-        log_path: String,
+        log_path: PathBuf,
         /// Name of the log instead of the command's basename
         #[arg(long, short = 'n')]
-        log_name: Option<String>,
+        log_name: Option<PathBuf>,
         /// prefix log lines with time & stream
         #[arg(long)]
         with_prefix: bool,
         /// exec command in this cgroup
         #[arg(long)]
         cgroup: Option<String>,
-        command: String,
+        command: PathBuf,
         args: Vec<String>,
     },
     Log {
@@ -90,6 +92,11 @@ enum Command {
         #[command(subcommand)]
         cmd: CgCmd,
     },
+
+    Rc {
+        #[command(subcommand)]
+        cmd: RcCmd,
+    },
 }
 
 #[derive(Subcommand)]
@@ -104,16 +111,56 @@ enum CgCmd {
     },
 }
 
+#[derive(Subcommand)]
+enum RcCmd {
+    Run,
+    Ls,
+    Status,
+    ReloadConfig,
+    Start {
+        #[arg(add = completions(dkl::rc::complete))]
+        key: String,
+    },
+    Stop {
+        #[arg(add = completions(dkl::rc::complete))]
+        key: String,
+    },
+    Reload {
+        #[arg(add = completions(dkl::rc::complete))]
+        key: String,
+    },
+    Sig {
+        #[arg(add = completions(dkl::rc::complete))]
+        key: String,
+        signal: u32,
+    },
+    Ctl {
+        args: Vec<String>,
+    },
+}
+
 #[tokio::main(flavor = "current_thread")]
 async fn main() -> Result<()> {
     clap_complete::CompleteEnv::with_factory(Cli::command).complete();
 
     let cli = Cli::parse();
 
-    env_logger::builder()
+    {
-        .parse_filters("info")
+        let mut builder = env_logger::builder();
-        .parse_default_env()
+        builder.parse_filters("info").parse_default_env();
-        .init();
+
+        if let Some(log_to) = cli.log_to {
+            builder.target(env_logger::Target::Pipe(Box::new(
+                std::fs::OpenOptions::new()
+                    .create(true)
+                    .append(true)
+                    .open(log_to)
+                    .unwrap(),
+            )));
+        }
+
+        builder.init();
+    }
 
     use Command as C;
     match cli.command {
@@ -127,23 +174,24 @@ async fn main() -> Result<()> {
             apply_config(&config, &filters, &prefix, dry_run).await
         }
         C::Logger {
-            ref log_path,
+            log_path,
-            ref log_name,
+            log_name,
             with_prefix,
             cgroup,
             command,
             args,
         } => {
-            let command = command.as_str();
+            let log_name = log_name.unwrap_or_else(|| command.file_prefix().unwrap().into());
-            let log_name = log_name.as_deref().unwrap_or_else(|| basename(command));
 
-            dkl::logger::Logger {
+            let logger = dkl::logger::Logger {
                 log_path,
                 log_name,
                 with_prefix,
-            }
+                cgroup,
-            .run(cgroup, command, &args)
+            };
-            .await
+
+            let cmd = logger.setup(command, &args).await?;
+            logger.exec(cmd).await
         }
         C::Log {
             log_path,
@@ -187,6 +235,20 @@ async fn main() -> Result<()> {
                 cols,
             } => Ok(dkl::cgroup::ls(root, &exclude, cols.as_deref()).await?),
         },
+
+        C::Rc { cmd } => match cmd {
+            RcCmd::Run => Ok(dkl::rc::run().await?),
+            RcCmd::Ls => Ok(dkl::rc::ctl(["ls"]).await?),
+            RcCmd::Status => Ok(dkl::rc::ctl(["status"]).await?),
+            RcCmd::ReloadConfig => Ok(dkl::rc::ctl(["reload-config"]).await?),
+            RcCmd::Start { key } => Ok(dkl::rc::ctl(["start", &key]).await?),
+            RcCmd::Stop { key } => Ok(dkl::rc::ctl(["stop", &key]).await?),
+            RcCmd::Reload { key } => Ok(dkl::rc::ctl(["reload", &key]).await?),
+            RcCmd::Sig { key, signal } => {
+                Ok(dkl::rc::ctl(["sig", &key, &signal.to_string()]).await?)
+            }
+            RcCmd::Ctl { args } => Ok(dkl::rc::ctl(&args).await?),
+        },
     }
 }
 
@@ -298,10 +360,6 @@ fn parse_ts_arg(ts: Option<String>) -> Result<Option<dkl::logger::Timestamp>> {
     }
 }
 
-fn basename(path: &str) -> &str {
-    path.rsplit_once('/').map_or(path, |split| split.1)
-}
-
 fn parse_globs(filters: &[String]) -> Result<Vec<glob::Pattern>> {
     let mut errors = false;
     let filters = (filters.iter())
@@ -321,3 +379,25 @@ fn parse_globs(filters: &[String]) -> Result<Vec<glob::Pattern>> {
 
     Ok(filters)
 }
+
+use clap_complete::{ArgValueCandidates, CompletionCandidate};
+
+fn completions(f: impl AsyncFn() -> Vec<String> + Send + Sync + 'static) -> ArgValueCandidates {
+    let f = std::sync::Arc::new(f);
+
+    ArgValueCandidates::new(move || {
+        let f = f.clone();
+        std::thread::spawn(move || {
+            tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+                .unwrap()
+                .block_on(async move { f().await })
+        })
+        .join()
+        .into_iter()
+        .flatten()
+        .map(CompletionCandidate::new)
+        .collect::<Vec<_>>()
+    })
+}

src/bin/dls.rs

@@ -3,6 +3,7 @@ use clap::{CommandFactory, Parser, Subcommand};
 use eyre::format_err;
 use futures_util::Stream;
 use futures_util::StreamExt;
+use std::path::PathBuf;
 use std::time::{Duration, SystemTime};
 use tokio::fs;
 use tokio::io::{AsyncWrite, AsyncWriteExt};
@@ -21,6 +22,8 @@ struct Cli {
 
 #[derive(Subcommand)]
 enum Command {
+    #[command(subcommand)]
+    Config(Config),
     Clusters,
     Cluster {
         cluster: String,
@@ -40,6 +43,16 @@ enum Command {
     Hash {
         salt: String,
     },
+    Store {
+        store_path: PathBuf,
+        #[command(subcommand)]
+        op: StoreOp,
+    },
+}
+
+#[derive(Subcommand)]
+enum Config {
+    Upload { config_path: PathBuf },
 }
 
 #[derive(Subcommand)]
@@ -71,6 +84,10 @@ enum ClusterCommand {
         #[arg(default_value = "cluster")]
         name: String,
     },
+    SetCaCert {
+        ca: String,
+        cert_path: PathBuf,
+    },
     Token {
         #[arg(default_value = "admin")]
         name: String,
@@ -96,6 +113,12 @@ enum ClusterCommand {
     },
 }
 
+#[derive(Subcommand)]
+enum StoreOp {
+    Get { data_path: PathBuf },
+    Set { data_path: PathBuf, value: String },
+}
+
 #[tokio::main(flavor = "current_thread")]
 async fn main() -> eyre::Result<()> {
     clap_complete::CompleteEnv::with_factory(Cli::command).complete();
@@ -119,18 +142,26 @@ async fn main() -> eyre::Result<()> {
             let dls = dls();
             let cluster = dls.cluster(cluster);
 
+            let Some(command) = command else {
+                write_json(&cluster.config().await?);
+                return Ok(());
+            };
+
             use ClusterCommand as CC;
             match command {
-                None => write_json(&cluster.config().await?),
+                CC::CaCert { name } => write_raw(&cluster.ca_cert(&name).await?),
-                Some(CC::CaCert { name }) => write_raw(&cluster.ca_cert(&name).await?),
+                CC::SetCaCert { ca, cert_path } => {
-                Some(CC::Token { name }) => println!("{}", &cluster.token(&name).await?),
+                    let cert = tokio::fs::read(cert_path).await?;
-                Some(CC::Addons) => write_raw(&cluster.addons().await?),
+                    cluster.set_ca_cert(&ca, cert).await?;
-                Some(CC::SshSign {
+                }
+                CC::Token { name } => println!("{}", &cluster.token(&name).await?),
+                CC::Addons => write_raw(&cluster.addons().await?),
+                CC::SshSign {
                     user_public_key,
                     principal,
                     validity,
                     options,
-                }) => {
+                } => {
                     let pub_key = tokio::fs::read_to_string(user_public_key).await?;
                     let cert = cluster
                         .ssh_userca_sign(&dls::SshSignReq {
@@ -142,12 +173,12 @@ async fn main() -> eyre::Result<()> {
                         .await?;
                     write_raw(&cert);
                 }
-                Some(CC::KubeSign {
+                CC::KubeSign {
                     csr,
                     user,
                     group,
                     validity,
-                }) => {
+                } => {
                     let csr = tokio::fs::read_to_string(csr).await?;
                     let cert = cluster
                         .kube_sign(&dls::KubeSignReq {
@@ -231,6 +262,29 @@ async fn main() -> eyre::Result<()> {
             println!("hash (hex):    {}", hex::encode(&hash));
             println!("hash (base64): {}", dkl::base64_encode(&hash));
         }
+        C::Store { store_path, op } => {
+            let mut s = dls::store::Store::new(store_path);
+            s.unlock(&std::env::var("DLS_STORE_PW").unwrap()).await?;
+
+            match op {
+                StoreOp::Get { data_path } => {
+                    let mut data = std::io::Cursor::new(s.read(data_path).await?);
+                    tokio::io::copy(&mut data, &mut tokio::io::stdout()).await?;
+                }
+                StoreOp::Set { data_path, value } => s.write(data_path, value.as_bytes()).await?,
+            }
+        }
+        C::Config(cmd) => match cmd {
+            Config::Upload { config_path } => {
+                let cfg = fs::read(&config_path).await?;
+                let cfg: dls::Config = if config_path.ends_with(".yaml") {
+                    serde_yaml::from_slice(&cfg)?
+                } else {
+                    serde_json::from_slice(&cfg)?
+                };
+                dls().upload_config(&cfg).await?;
+            }
+        },
     };
 
     Ok(())

src/bootstrap.rs

@@ -250,4 +250,10 @@ pub struct Bootstrap {
     pub dev: String,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub seed: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub seed_ca: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub seed_servername: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub seed_proxy: Option<String>,
 }

src/cgroup.rs

@@ -64,8 +64,8 @@ pub async fn ls(
     }
 
     use tabled::settings::{
-        object::{Column, Row},
         Alignment, Modify,
+        object::{Column, Row},
     };
     let mut table = table.build();
     table.with(tabled::settings::Style::psql());

src/dls.rs

@@ -23,6 +23,18 @@ impl Client {
         }
     }
 
+    pub fn new_with_http_client(
+        base_url: String,
+        token: String,
+        http_client: reqwest::Client,
+    ) -> Self {
+        Self {
+            base_url,
+            token,
+            http_client,
+        }
+    }
+
     pub fn with_proxy(self, proxy: String) -> reqwest::Result<Self> {
         let proxy = reqwest::Proxy::all(proxy)?;
         Ok(Self {
@@ -65,6 +77,15 @@ impl Client {
         Ok(resp.bytes_stream())
     }
 
+    pub async fn upload_config(&self, config: &Config) -> Result<()> {
+        do_req(self.req(Method::POST, "configs")?.json(config), &self.token).await?;
+        Ok(())
+    }
+
+    pub async fn get_config(&self) -> Result<Config> {
+        self.get_json("config").await
+    }
+
     pub async fn get_json<T: serde::de::DeserializeOwned>(&self, path: impl Display) -> Result<T> {
         self.req_json(self.get(&path)?).await
     }
@@ -133,6 +154,18 @@ impl<'t> Cluster<'t> {
         Ok(resp.bytes().await.map_err(Error::Read)?.to_vec())
     }
 
+    pub async fn set_ca_cert(&self, ca: &str, cert: Vec<u8>) -> Result<()> {
+        let req = self.dls.req(
+            Method::PUT,
+            format!("clusters/{}/CAs/{ca}/certificate", self.name),
+        )?;
+        let req = req
+            .body(cert)
+            .header("Content-Type", "application/x-x509-ca-cert");
+        do_req(req, &self.dls.token).await?;
+        Ok(())
+    }
+
     pub async fn kube_sign(&self, sign_req: &KubeSignReq) -> Result<Vec<u8>> {
         let req = (self.dls).req(Method::POST, format!("clusters/{}/kube/sign", self.name))?;
         let req = req.json(sign_req);
@@ -154,7 +187,7 @@ impl<'t> Host<'t> {
     pub async fn asset(
         &self,
         asset_name: &str,
-    ) -> Result<impl Stream<Item = reqwest::Result<Bytes>>> {
+    ) -> Result<impl Stream<Item = reqwest::Result<Bytes>> + use<>> {
         let req = self.dls.get(format!("hosts/{}/{asset_name}", self.name))?;
         let resp = do_req(req, &self.dls.token).await?;
         Ok(resp.bytes_stream())
@@ -162,17 +195,23 @@ impl<'t> Host<'t> {
 }
 
 #[derive(Default, serde::Deserialize, serde::Serialize)]
-#[serde(rename_all = "PascalCase")]
+#[serde(default, rename_all = "PascalCase")]
 pub struct Config {
-    #[serde(default, deserialize_with = "deserialize_null_as_default")]
+    #[serde(deserialize_with = "deserialize_null_as_default", alias = "clusters")]
     pub clusters: Vec<ClusterConfig>,
-    #[serde(default, deserialize_with = "deserialize_null_as_default")]
+    #[serde(deserialize_with = "deserialize_null_as_default", alias = "hosts")]
     pub hosts: Vec<HostConfig>,
-    #[serde(default, deserialize_with = "deserialize_null_as_default")]
+    #[serde(
+        deserialize_with = "deserialize_null_as_default",
+        alias = "hosttemplates"
+    )]
     pub host_templates: Vec<HostConfig>,
-    #[serde(default, rename = "SSLConfig")]
+    #[serde(default, rename = "SSLConfig", alias = "sslconfig")]
     pub ssl_config: String,
-    #[serde(default, deserialize_with = "deserialize_null_as_default")]
+    #[serde(
+        deserialize_with = "deserialize_null_as_default",
+        alias = "extracacerts"
+    )]
     pub extra_ca_certs: Map<String, String>,
 }
 
@@ -190,40 +229,58 @@ where
 #[derive(serde::Deserialize, serde::Serialize)]
 #[serde(rename_all = "PascalCase")]
 pub struct ClusterConfig {
+    #[serde(alias = "name")]
     pub name: String,
+    #[serde(alias = "bootstrappods")]
     pub bootstrap_pods: String,
+    #[serde(alias = "addons")]
     pub addons: String,
 }
 
 #[derive(Default, serde::Deserialize, serde::Serialize)]
 #[serde(rename_all = "PascalCase")]
 pub struct HostConfig {
+    #[serde(alias = "name")]
     pub name: String,
-    #[serde(default, skip_serializing_if = "Option::is_none")]
+    #[serde(
+        default,
+        skip_serializing_if = "Option::is_none",
+        alias = "cluster_name",
+        alias = "clustername"
+    )]
     pub cluster_name: Option<String>,
 
-    #[serde(rename = "IPs")]
+    #[serde(rename = "IPs", alias = "ips")]
     pub ips: Vec<IpAddr>,
 
-    #[serde(default, skip_serializing_if = "Map::is_empty")]
+    #[serde(default, skip_serializing_if = "Map::is_empty", alias = "labels")]
     pub labels: Map<String, String>,
-    #[serde(default, skip_serializing_if = "Map::is_empty")]
+    #[serde(default, skip_serializing_if = "Map::is_empty", alias = "annotations")]
     pub annotations: Map<String, String>,
 
-    #[serde(rename = "IPXE", skip_serializing_if = "Option::is_none")]
+    #[serde(
+        rename = "IPXE",
+        skip_serializing_if = "Option::is_none",
+        alias = "ipxe"
+    )]
     pub ipxe: Option<String>,
 
+    #[serde(alias = "initrd")]
     pub initrd: String,
+    #[serde(alias = "kernel")]
     pub kernel: String,
+    #[serde(alias = "versions")]
     pub versions: Map<String, String>,
 
     /// initrd config template
+    #[serde(alias = "bootstrapconfig")]
     pub bootstrap_config: String,
     /// files to add to the final initrd config, with rendering
-    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    #[serde(default, skip_serializing_if = "Vec::is_empty", alias = "initrdfiles")]
     pub initrd_files: Vec<crate::File>,
 
     /// system config template
+    #[serde(alias = "config")]
     pub config: String,
 }
 

src/dls/store.rs

@@ -1,4 +1,13 @@
-pub fn hash_password(salt: &[u8], passphrase: &str) -> argon2::Result<[u8; 32]> {
+use openssl::symm::Mode;
+use std::borrow::Cow;
+use std::path::{Path, PathBuf};
+use tokio::{fs, io::AsyncWriteExt};
+
+pub type Salt = [u8; 16];
+pub type Key = [u8; 32];
+pub type Hash = Vec<u8>;
+
+pub fn hash_password(salt: &[u8], passphrase: &str) -> Result<Key> {
     let hash = argon2::hash_raw(
         passphrase.as_bytes(),
         salt,
@@ -6,12 +15,191 @@ pub fn hash_password(salt: &[u8], passphrase: &str) -> argon2::Result<[u8; 32]>
             variant: argon2::Variant::Argon2id,
             hash_length: 32,
             time_cost: 1,
-            mem_cost: 65536,
+            mem_cost: 64 << 10,
             thread_mode: argon2::ThreadMode::Parallel,
             lanes: 4,
             ..Default::default()
         },
-    )?;
+    )
+    .map_err(Error::Hash)?;
 
-    unsafe { Ok(hash.try_into().unwrap_unchecked()) }
+    Ok(hash.try_into().unwrap())
+}
+
+pub struct Store {
+    path: PathBuf,
+    key: Option<Key>,
+}
+
+impl Store {
+    pub fn new(path: impl Into<PathBuf>) -> Self {
+        Self {
+            path: path.into(),
+            key: None,
+        }
+    }
+
+    pub async fn unlock(&mut self, passphrase: &str) -> Result<()> {
+        let keys = self.read_keys().await?;
+
+        let salt = keys.salt;
+
+        let user_key = hash_password(&salt, passphrase)?;
+        let user_hash = openssl::sha::sha512(&user_key);
+
+        for nk in keys.keys {
+            if nk.hash != user_hash {
+                continue;
+            }
+
+            // build iv+data from salt
+            let mut enc_data = Vec::with_capacity(salt.len() + nk.enc_key.len());
+            enc_data.extend_from_slice(&salt);
+            enc_data.extend_from_slice(&nk.enc_key);
+
+            let key = crypt(&enc_data, &user_key, Mode::Decrypt)?;
+            let key = key.try_into().map_err(|_| Error::InvalidKey)?;
+
+            self.key = Some(key);
+            return Ok(());
+        }
+
+        Err(Error::KeyNotFound)
+    }
+
+    pub async fn read(&self, path: impl AsRef<Path>) -> Result<Vec<u8>> {
+        let enc_data = self.read_file(path.as_ref().with_extension("data")).await?;
+        self.decrypt(&enc_data)
+    }
+    pub async fn read_to_string(&self, path: impl AsRef<Path>) -> Result<String> {
+        let path = path.as_ref();
+        String::from_utf8(self.read(path).await?).map_err(|_| Error::InvalidUtf8(path.into()))
+    }
+
+    pub async fn write(&self, path: impl AsRef<Path>, data: &[u8]) -> Result<()> {
+        let enc_data = self.encrypt(data)?;
+        safe_write(self.path.join(&path).with_extension("data"), &enc_data).await
+    }
+
+    async fn read_keys(&self) -> Result<Keys<'_>> {
+        let keys = self.read_file(".keys").await?;
+        let Some(keys) = keys.strip_prefix(b"{json}") else {
+            return Err(Error::InvalidKeys);
+        };
+
+        serde_json::from_slice(keys).map_err(Error::KeysParse)
+    }
+
+    async fn read_file(&self, subpath: impl AsRef<Path>) -> Result<Vec<u8>> {
+        let path = self.path.join(subpath);
+        fs::read(&path).await.map_err(|e| Error::Read(path, e))
+    }
+
+    fn encrypt(&self, src: &[u8]) -> Result<Vec<u8>> {
+        self.crypt(src, Mode::Encrypt)
+    }
+    fn decrypt(&self, src: &[u8]) -> Result<Vec<u8>> {
+        self.crypt(src, Mode::Decrypt)
+    }
+    fn crypt(&self, src: &[u8], mode: Mode) -> Result<Vec<u8>> {
+        let key = self.key.as_ref().ok_or(Error::NotUnlocked)?;
+        crypt(src, key, mode)
+    }
+}
+
+async fn safe_write(path: impl AsRef<Path>, contents: &[u8]) -> Result<()> {
+    let path = path.as_ref();
+    let tmp = path.with_added_extension("new");
+
+    let tmp_err = |e| Error::Write(tmp.clone(), e);
+
+    let mut file = fs::OpenOptions::new()
+        .create(true)
+        .truncate(true)
+        .write(true)
+        .open(&tmp)
+        .await
+        .map_err(tmp_err)?;
+
+    file.write_all(contents).await.map_err(tmp_err)?;
+    file.sync_all().await.map_err(tmp_err)?;
+    file.shutdown().await.map_err(tmp_err)?;
+
+    fs::rename(tmp, &path)
+        .await
+        .map_err(|e| Error::Write(path.into(), e))
+}
+
+fn crypt(mut src: &[u8], key: &Key, mode: Mode) -> Result<Vec<u8>> {
+    use openssl::symm::{Cipher, Crypter};
+
+    let mut iv = [0u8; 16];
+    let mut dst: Vec<u8>;
+    let crypt_dst: &mut [u8];
+
+    match mode {
+        Mode::Encrypt => {
+            getrandom::fill(&mut iv).unwrap();
+            dst = vec![0u8; iv.len() + src.len()];
+            dst[..iv.len()].copy_from_slice(&iv);
+            crypt_dst = &mut dst[iv.len()..];
+        }
+        Mode::Decrypt => {
+            iv = src[..iv.len()]
+                .try_into()
+                .map_err(|_| Error::DecryptInputToSmall)?;
+            src = &src[iv.len()..];
+            dst = vec![0u8; src.len()];
+            crypt_dst = &mut dst;
+        }
+    }
+
+    let cipher = Cipher::aes_256_cfb128();
+    Crypter::new(cipher, mode, key, Some(&iv))
+        .expect("Failed to init AES")
+        .update(src, crypt_dst)
+        .expect("AES CFB encrypt/decrypt failed");
+
+    Ok(dst)
+}
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[error("hash error: {0}")]
+    Hash(argon2::Error),
+    #[error("read {0} failed: {1}")]
+    Read(PathBuf, std::io::Error),
+    #[error("write {0} failed: {1}")]
+    Write(PathBuf, std::io::Error),
+    #[error("read {0} failed: invalid UTF-8")]
+    InvalidUtf8(PathBuf),
+    #[error("invalid keys data")]
+    InvalidKeys,
+    #[error("invalid key")]
+    InvalidKey,
+    #[error("keys parse error: {0}")]
+    KeysParse(serde_json::Error),
+    #[error("key not found")]
+    KeyNotFound,
+    #[error("store not unlocked")]
+    NotUnlocked,
+    #[error("decrypt input too small")]
+    DecryptInputToSmall,
+}
+
+#[derive(Clone, Debug, serde::Serialize, serde::Deserialize)]
+#[serde(rename_all = "PascalCase")]
+struct Keys<'t> {
+    salt: Salt,
+    keys: Vec<NamedKey<'t>>,
+}
+
+#[derive(Clone, Debug, serde::Serialize, serde::Deserialize)]
+#[serde(rename_all = "PascalCase")]
+struct NamedKey<'t> {
+    name: Cow<'t, str>,
+    hash: Hash,
+    enc_key: Key,
 }

src/dynlay.rs

@@ -60,7 +60,7 @@ impl<'t> Dynlay<'t> {
 
         // mount layer
         info!("mounting layer");
-        sudo("mount", &["-t", "squashfs", lay_path, &mount_path_str]).await?;
+        sudo("mount", &[lay_path, &mount_path_str]).await?;
 
         let mut paths = spawn_walk_dir(mount_path.clone());
         while let Some(result) = paths.recv().await {

src/lib.rs

@@ -1,3 +1,5 @@
+use std::borrow::Cow;
+
 pub mod apply;
 pub mod bootstrap;
 pub mod cgroup;
@@ -9,7 +11,7 @@ pub mod logger;
 pub mod proxy;
 pub mod rc;
 
-#[derive(Debug, Default, serde::Deserialize, serde::Serialize)]
+#[derive(Debug, Default, Clone, serde::Deserialize, serde::Serialize)]
 pub struct Config {
     pub layers: Vec<String>,
     pub root_user: RootUser,
@@ -23,14 +25,14 @@ pub struct Config {
     pub users: Vec<User>,
 }
 
-#[derive(Debug, Default, serde::Deserialize, serde::Serialize)]
+#[derive(Debug, Default, Clone, serde::Deserialize, serde::Serialize)]
 pub struct RootUser {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub password_hash: Option<String>,
     pub authorized_keys: Vec<String>,
 }
 
-#[derive(Debug, Default, serde::Deserialize, serde::Serialize)]
+#[derive(Debug, Default, Clone, serde::Deserialize, serde::Serialize)]
 pub struct Mount {
     pub dev: String,
     pub path: String,
@@ -40,14 +42,14 @@ pub struct Mount {
     pub options: Option<String>,
 }
 
-#[derive(Debug, serde::Deserialize, serde::Serialize)]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
 pub struct Group {
     pub name: String,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub gid: Option<u32>,
 }
 
-#[derive(Debug, serde::Deserialize, serde::Serialize)]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
 pub struct User {
     pub name: String,
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -56,22 +58,45 @@ pub struct User {
     pub gid: Option<u32>,
 }
 
-#[derive(Debug, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
+#[derive(Default, Debug, Clone, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
 pub struct File {
     pub path: String,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub mode: Option<u32>,
-    #[serde(flatten)]
+    pub kind: Option<FileKind>,
-    pub kind: FileKind,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub content: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub content64: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub symlink: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub parts: Option<Vec<FilePart>>,
+    #[serde(default, skip_serializing_if = "is_false")]
+    pub dir: bool,
 }
 
-#[derive(Debug, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
+fn is_false(b: &bool) -> bool {
+    !b
+}
+
+#[derive(Default, Debug, Clone, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
 #[serde(rename_all = "lowercase")]
 pub enum FileKind {
+    #[default]
+    Skip,
     Content(String),
     Content64(String),
+    Parts(Vec<FilePart>),
     Symlink(String),
-    Dir(bool),
+    Dir,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
+#[serde(rename_all = "lowercase")]
+pub enum FilePart {
+    Content(String),
+    Content64(String),
 }
 
 // ------------------------------------------------------------------------
@@ -86,11 +111,71 @@ impl Config {
 }
 
 pub fn base64_decode(s: &str) -> Result<Vec<u8>, base64::DecodeError> {
-    use base64::{Engine as _, prelude::BASE64_STANDARD_NO_PAD as B64};
+    use base64::{prelude::BASE64_STANDARD_NO_PAD as B64, Engine as _};
     B64.decode(s.trim_end_matches('='))
 }
 
 pub fn base64_encode(b: &[u8]) -> String {
-    use base64::{Engine as _, prelude::BASE64_STANDARD as B64};
+    use base64::{prelude::BASE64_STANDARD as B64, Engine as _};
     B64.encode(b)
 }
+
+impl<'t> File {
+    pub fn kind(&'t self) -> Cow<'t, FileKind> {
+        self.kind.as_ref().map(Cow::Borrowed).unwrap_or_else(|| {
+            use FileKind::*;
+            Cow::Owned(if let Some(ref s) = self.content {
+                Content(s.clone())
+            } else if let Some(ref s) = self.content64 {
+                Content64(s.clone())
+            } else if let Some(ref s) = self.symlink {
+                Symlink(s.clone())
+            } else if let Some(ref p) = self.parts {
+                Parts(p.clone())
+            } else if self.dir {
+                Dir
+            } else {
+                Skip
+            })
+        })
+    }
+
+    pub fn is_symlink(&self) -> bool {
+        if let Some(ref kind) = self.kind {
+            matches!(kind, FileKind::Symlink(_))
+        } else {
+            self.symlink.is_some()
+        }
+    }
+}
+
+impl FileKind {
+    pub fn content<'t>(&'t self) -> Result<Option<Cow<'t, [u8]>>, base64::DecodeError> {
+        use FileKind::*;
+        Ok(match self {
+            Content(content) => Some(Cow::Borrowed(content.as_bytes())),
+            Content64(content) => {
+                let content = base64_decode(content)?;
+                Some(Cow::Owned(content))
+            }
+            Parts(parts) => {
+                let mut assembly = Vec::new();
+                for part in parts {
+                    assembly.extend(part.content()?.into_iter());
+                }
+                Some(Cow::Owned(assembly))
+            }
+            _ => None,
+        })
+    }
+}
+
+impl FilePart {
+    pub fn content(&self) -> Result<Cow<'_, [u8]>, base64::DecodeError> {
+        use FilePart::*;
+        Ok(match self {
+            Content(content) => Cow::Borrowed(content.as_bytes()),
+            Content64(content) => Cow::Owned(base64_decode(content)?),
+        })
+    }
+}

src/logger.rs

@@ -1,15 +1,16 @@
 use async_compression::tokio::write::{ZstdDecoder, ZstdEncoder};
 use chrono::{DurationRound, TimeDelta, Utc};
-use eyre::{format_err, Result};
+use eyre::{Result, format_err};
 use log::{debug, error, warn};
+use std::ffi::OsStr;
 use std::path::{Path, PathBuf};
 use std::process::Stdio;
 use tokio::{
     fs::File,
     io::{self, AsyncBufReadExt, AsyncRead, AsyncWrite, AsyncWriteExt, BufReader, BufWriter},
-    process,
+    process::{Child, Command},
     sync::mpsc,
-    time::{sleep, Duration},
+    time::{Duration, sleep},
 };
 
 use crate::{cgroup, fs};
@@ -22,67 +23,92 @@ const TRUNC_DELTA: TimeDelta = TimeDelta::hours(1);
 const FLUSH_INTERVAL: Duration = Duration::from_secs(1);
 const WRITE_RETRY_DELAY: Duration = Duration::from_secs(1);
 
-pub struct Logger<'t> {
+pub struct Logger {
-    pub log_path: &'t str,
+    pub log_path: PathBuf,
-    pub log_name: &'t str,
+    pub log_name: PathBuf,
     pub with_prefix: bool,
+    pub cgroup: Option<String>,
 }
 
-impl<'t> Logger<'t> {
+impl Logger {
-    pub async fn run(&self, cgroup: Option<String>, command: &str, args: &[String]) -> Result<()> {
+    pub async fn setup<I, S>(&self, command: impl AsRef<OsStr>, args: I) -> fs::Result<Command>
+    where
+        I: IntoIterator<Item = S>,
+        S: AsRef<OsStr>,
+    {
         // make sure we can at least open the log before starting the command
-        let archives_path = &format!("{path}/archives", path = self.log_path);
+        let archives_path = &self.log_path.join("archives");
-        (fs::create_dir_all(archives_path).await)
+        fs::create_dir_all(archives_path).await?;
-            .map_err(|e| format_err!("failed to create archives dir: {e}"))?;
-        let archives_read_dir = (fs::read_dir(archives_path).await)
-            .map_err(|e| format_err!("failed to list archives: {e}"))?;
 
-        let mut prev_stamp = trunc_ts(Utc::now());
+        let archives_read_dir = fs::read_dir(archives_path).await?;
-        let mut current_log = BufWriter::new(self.open_log(prev_stamp).await?);
+
+        let prev_stamp = trunc_ts(Utc::now());
 
         tokio::spawn(compress_archives(
             archives_read_dir,
-            self.log_name.to_string(),
+            self.log_name.clone(),
             prev_stamp.format(TS_FORMAT).to_string(),
         ));
 
-        // start the command
+        // create the command
-        let mut cmd = process::Command::new(command);
+        let mut cmd = Command::new(command);
-        cmd.args(args).stdout(Stdio::piped()).stderr(Stdio::piped());
-        if let Some(cgroup) = cgroup.as_deref() {
-            let mut cg_path = PathBuf::from(cgroup::ROOT);
-            cg_path.push(cgroup);
-            cg_path.push(self.log_name);
 
-            use std::io::ErrorKind as K;
+        cmd.args(args);
-            match tokio::fs::create_dir(&cg_path).await {
+
-                Ok(_) => debug!("created dir {}", cg_path.display()),
+        if let Some(cgroup) = self.cgroup.as_deref() {
-                Err(e) if e.kind() == K::AlreadyExists => {
+            let cg_path = PathBuf::from(cgroup::ROOT)
-                    debug!("existing dir {}", cg_path.display())
+                .join(cgroup)
-                }
+                .join(&self.log_name);
-                Err(e) => return Err(fs::Error::CreateDir(cg_path, e).into()),
+
-            }
+            fs::create_dir_all(&cg_path).await?;
 
             let procs_file = cg_path.join("cgroup.procs");
             debug!("procs file {}", procs_file.display());
-            fs::write(&procs_file, b"0").await?;
+
+            unsafe { cmd.pre_exec(move || std::fs::write(&procs_file, b"0")) };
         }
 
-        let mut child = cmd.spawn().map_err(|e| format_err!("exec failed: {e}"))?;
+        Ok(cmd)
+    }
 
-        let (tx, mut rx) = mpsc::channel(8);
+    pub fn spawn(self, mut cmd: Command) -> std::io::Result<Child> {
+        // setup outputs for capture
+        cmd.stdout(Stdio::piped()).stderr(Stdio::piped());
+
+        // spawn
+        let mut child = cmd.spawn()?;
+
+        // capture outputs
+        let (tx, rx) = mpsc::channel(8);
 
         tokio::spawn(copy("stdout", child.stdout.take().unwrap(), tx.clone()));
         tokio::spawn(copy("stderr", child.stderr.take().unwrap(), tx));
 
+        // log outputs
+        tokio::spawn(self.log_stream(rx));
+
+        Ok(child)
+    }
+
+    // TODO: Result<!> when stable
+    pub async fn exec(self, cmd: Command) -> Result<()> {
+        let mut child = self.spawn(cmd)?;
+
         // forward signals
         if let Some(child_pid) = child.id() {
             forward_signals_to(child_pid as i32);
         }
 
-        // handle output
+        let status = child.wait().await?;
+        std::process::exit(status.code().unwrap_or(-1));
+    }
+
+    async fn log_stream(self, mut rx: mpsc::Receiver<LogItem>) {
         let mut flush_ticker = tokio::time::interval(FLUSH_INTERVAL);
 
+        let mut prev_stamp = trunc_ts(Utc::now());
+        let mut current_log = BufWriter::new(self.eventually_open_log(prev_stamp).await);
+
         loop {
             tokio::select!(
                 r = rx.recv() => {
@@ -102,15 +128,11 @@ impl<'t> Logger<'t> {
             );
         }
 
-        let status = child.wait().await?;
-
         // finalize
-        while let Err(e) = current_log.flush().await {
+        while let Err(e) = current_log.shutdown().await {
-            error!("final log flush failed: {e}");
+            error!("final log shutdown failed: {e}");
             sleep(WRITE_RETRY_DELAY).await;
         }
-
-        std::process::exit(status.code().unwrap_or(-1));
     }
 
     async fn log_item(
@@ -143,6 +165,18 @@ impl<'t> Logger<'t> {
         Ok(())
     }
 
+    async fn eventually_open_log(&self, ts: Timestamp) -> File {
+        loop {
+            match self.open_log(ts).await {
+                Ok(log) => break log,
+                Err(e) => {
+                    error!("open log failed: {e}");
+                    sleep(WRITE_RETRY_DELAY).await;
+                }
+            }
+        }
+    }
+
     async fn open_log(&self, ts: Timestamp) -> Result<File> {
         let log_file = &self.archive_path(ts);
 
@@ -153,8 +187,9 @@ impl<'t> Logger<'t> {
             .open(log_file)
             .await?;
 
-        let link_src = &PathBuf::from(self.log_path)
+        let link_src = &self
-            .join(self.log_name)
+            .log_path
+            .join(&self.log_name)
             .with_added_extension("log");
         let link_tgt = &self.archive_rel_path(ts);
 
@@ -171,23 +206,21 @@ impl<'t> Logger<'t> {
     }
 
     fn archive_path(&self, ts: Timestamp) -> PathBuf {
-        PathBuf::from(self.log_path).join(self.archive_rel_path(ts))
+        self.log_path.join(self.archive_rel_path(ts))
     }
     fn archive_rel_path(&self, ts: Timestamp) -> PathBuf {
         PathBuf::from("archives").join(self.archive_file(ts))
     }
-    fn archive_file(&self, ts: Timestamp) -> String {
+    fn archive_file(&self, ts: Timestamp) -> PathBuf {
-        format!(
+        self.log_name
-            "{name}.{ts}.log",
+            .with_added_extension(ts.format(TS_FORMAT).to_string())
-            name = self.log_name,
+            .with_added_extension("log")
-            ts = ts.format(TS_FORMAT),
-        )
     }
 }
 
 fn forward_signals_to(pid: i32) {
     use nix::{
-        sys::signal::{kill, Signal},
+        sys::signal::{Signal, kill},
         unistd::Pid,
     };
     use signal_hook::{consts::*, low_level::register};
@@ -266,7 +299,7 @@ async fn copy(stream_name: &'static str, out: impl AsyncRead + Unpin, tx: mpsc::
             line.extend_from_slice(buf);
             out.consume(len);
             send_line!();
-        } else if buf.len() > remaining {
+        } else if buf.len() >= remaining {
             line.extend_from_slice(&buf[..remaining]);
             out.consume(remaining);
             send_line!();
@@ -287,7 +320,12 @@ pub fn trunc_ts(ts: Timestamp) -> Timestamp {
         .expect("duration_trunc failed")
 }
 
-async fn compress_archives(mut read_dir: fs::ReadDir, log_name: String, exclude_ts: String) {
+async fn compress_archives(
+    mut read_dir: fs::ReadDir,
+    log_name: impl AsRef<Path>,
+    exclude_ts: String,
+) {
+    let log_name = log_name.as_ref();
     loop {
         let Ok(Some(entry)) =
             (read_dir.next_entry().await).inspect_err(|e| error!("archive dir read failed: {e}"))

src/rc.rs

@@ -1 +1,495 @@
+use eyre::format_err;
+use log::{error, info, warn};
+use nix::sys::signal::Signal;
+use std::collections::{BTreeMap as Map, BTreeSet as Set};
+use std::path::PathBuf;
+use std::sync::LazyLock;
+use tokio::{
+    io::{AsyncBufReadExt, AsyncReadExt, AsyncWriteExt, BufReader, copy},
+    net::{UnixListener, UnixStream},
+    sync::{RwLock, mpsc, watch},
+};
 
+use crate::{cgroup, fs};
+
+mod runner;
+
+use runner::{Child, State};
+
+const CFG_PATH: &str = "/etc/direktil/rc.yaml";
+const SOCK_PATH: &str = "/run/dkl-rc/ctl.sock"; // Path::new when stable
+
+#[derive(Default, serde::Serialize, serde::Deserialize)]
+pub struct Config {
+    #[serde(default, skip_serializing_if = "Map::is_empty")]
+    pub cgroups: Map<String, CgroupConfig>,
+}
+
+#[derive(serde::Serialize, serde::Deserialize)]
+pub struct CgroupConfig {
+    pub controllers: String,
+    #[serde(default, skip_serializing_if = "Map::is_empty")]
+    pub settings: Map<String, String>,
+    #[serde(default, skip_serializing_if = "Map::is_empty")]
+    pub services: Map<String, Service>,
+}
+
+pub type Service = Vec<String>;
+
+static MANAGER: LazyLock<RwLock<Manager>> = LazyLock::new(|| RwLock::new(Manager::default()));
+
+type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Debug, thiserror::Error)]
+enum Error {
+    #[error("invalid command: {0:?}")]
+    InvalidCommand(String),
+    #[error("config read failed: {0}")]
+    ConfigRead(fs::Error),
+    #[error("config parse failed: {0}")]
+    ConfigParse(serde_yaml::Error),
+    #[error("cgroup setup failed: {0}")]
+    CgroupSetup(fs::Error),
+    #[error("invalid key (cgroup/service)")]
+    InvalidKey,
+    #[error("unknown cgroup: {0:?}")]
+    UnknownCgroup(String),
+    #[error("unknown service: {0:?}")]
+    UnknownService(String),
+    #[error("invalid signal: {0:?}")]
+    InvalidSignal(String),
+    #[error("process exited")]
+    ProcessExited,
+    #[error("nothing running under {0:?}")]
+    NotRunning(String),
+    #[error("kill failed: {0:?}")]
+    KillFailed(nix::Error),
+    #[error("service runner is dead")]
+    RunnerDead,
+}
+
+pub async fn run() -> eyre::Result<()> {
+    info!("starting");
+
+    tokio::spawn(wait_terminate());
+
+    let _ = reload_config().await;
+    tokio::spawn(wait_reload());
+
+    if let Some(sock_dir) = PathBuf::from(SOCK_PATH).parent() {
+        let _ = tokio::fs::DirBuilder::new()
+            .mode(0o700)
+            .create(sock_dir)
+            .await;
+    }
+
+    let _ = tokio::fs::remove_file(SOCK_PATH).await;
+    let listener = UnixListener::bind(SOCK_PATH)?;
+
+    loop {
+        let Ok((conn, _)) = listener.accept().await else {
+            warn!("listener closed");
+            break;
+        };
+
+        tokio::spawn(async move { handle(conn).await });
+    }
+
+    cleanup().await;
+    Ok(())
+}
+
+async fn cleanup() {
+    let _ = tokio::fs::remove_file(SOCK_PATH).await;
+}
+
+pub async fn ctl<I, S>(args: I) -> eyre::Result<()>
+where
+    I: IntoIterator<Item = S>,
+    S: Into<String>,
+{
+    let args: Vec<_> = args.into_iter().map(|s| s.into()).collect();
+    let args = format!("{}\n", args.join(" "));
+
+    match ctl_exec(args.as_bytes()).await {
+        Ok(mut rd) => {
+            copy(&mut rd, &mut tokio::io::stdout()).await?;
+            std::process::exit(0);
+        }
+        Err(e) => {
+            eprintln!("{e}");
+            std::process::exit(1);
+        }
+    }
+}
+
+async fn ctl_exec(request: &[u8]) -> eyre::Result<BufReader<UnixStream>> {
+    let mut conn = UnixStream::connect(SOCK_PATH)
+        .await
+        .map_err(|e| format_err!("{SOCK_PATH}: {e}"))?;
+
+    conn.write_all(request).await?;
+
+    let mut rd = BufReader::with_capacity(64, conn);
+
+    let mut code = String::new();
+    rd.read_line(&mut code).await?;
+    let code: i32 = code.trim_ascii().parse()?;
+
+    if code != 0 {
+        let mut err = String::new();
+        rd.read_to_string(&mut err).await?;
+        return Err(format_err!("{}", err.trim_ascii_end()));
+    }
+
+    Ok(rd)
+}
+
+async fn handle(mut conn: UnixStream) {
+    let (rd, mut wr) = conn.split();
+    let mut rd = BufReader::with_capacity(64, rd).lines();
+
+    let Ok(Some(line)) = rd.next_line().await else {
+        return;
+    };
+
+    let mut line = line.split_ascii_whitespace();
+
+    macro_rules! next {
+        () => {{
+            match line.next() {
+                Some(v) => v,
+                None => return,
+            }
+        }};
+    }
+
+    let r = match next!() {
+        "ls" => Ok(Some(ls().await)),
+        "status" => Ok(Some(status().await)),
+        "reload-config" => reload_config().await.map(|_| None),
+        "start" => start(next!()).await.map(|_| None),
+        "stop" => stop(next!()).await.map(|_| None),
+        "reload" => reload(next!()).await.map(|_| None),
+        "sig" => sig(next!(), next!()).await.map(|_| None),
+        cmd => Err(Error::InvalidCommand(cmd.into())),
+    };
+
+    let _ = match r {
+        Ok(None) => wr.write_all(b"0\n").await,
+        Ok(Some(s)) => wr.write_all(format!("0\n{s}\n").as_bytes()).await,
+        Err(e) => wr.write_all(format!("1\n{e}\n").as_bytes()).await,
+    };
+
+    let _ = wr.shutdown().await;
+}
+
+async fn wait_terminate() {
+    use tokio::signal::unix::{SignalKind, signal};
+    let Ok(mut sig) = signal(SignalKind::terminate())
+        .inspect_err(|e| error!("failed to listen to SIGTERM (will be ignored): {e}"))
+    else {
+        return;
+    };
+
+    sig.recv().await;
+
+    info!("SIGTERM received, terminating");
+    MANAGER.write().await.terminate().await;
+
+    cleanup().await;
+    log::logger().flush();
+    std::process::exit(0);
+}
+
+async fn wait_reload() {
+    use tokio::signal::unix::{SignalKind, signal};
+    let Ok(mut sig) = signal(SignalKind::hangup())
+        .inspect_err(|e| error!("failed to listen to SIGHUP (will be ignored): {e}"))
+    else {
+        return;
+    };
+
+    loop {
+        sig.recv().await;
+        let _ = reload_config().await;
+    }
+}
+
+async fn reload_config() -> Result<()> {
+    let cfg = (fs::read(CFG_PATH).await)
+        .map_err(Error::ConfigRead)
+        .inspect_err(|e| error!("{e}"))?;
+
+    let cfg = serde_yaml::from_slice::<Config>(&cfg)
+        .map_err(Error::ConfigParse)
+        .inspect_err(|e| error!("{CFG_PATH}: {e}"))?;
+
+    info!("applying new config");
+    let r = MANAGER.write().await.apply_config(cfg).await;
+    match &r {
+        Ok(_) => info!("applied new config"),
+        Err(e) => info!("failed to apply new config: {e}"),
+    }
+    r
+}
+
+async fn ls() -> String {
+    let mut keys = String::new();
+    for (i, k) in MANAGER.read().await.runners.keys().enumerate() {
+        if i != 0 {
+            keys.push('\n');
+        }
+        keys.push_str(k);
+    }
+    keys
+}
+
+async fn status() -> String {
+    let status = MANAGER.read().await.status();
+
+    let mut table = tabled::builder::Builder::new();
+    table.push_record(["cgroup", "service", "PID", "state", "msg"]);
+
+    for (cg_svc, child) in status {
+        let (cg, svc) = cg_svc.split_once('/').unwrap();
+        let pid = child.pid.map(|p| p.to_string());
+        table.push_record([
+            cg,
+            svc,
+            pid.as_deref().unwrap_or("◌"),
+            &format!("{:?}", child.state),
+            child.msg.as_deref().unwrap_or("◌"),
+        ]);
+    }
+
+    (table.build())
+        .with(tabled::settings::Style::psql())
+        .to_string()
+}
+
+async fn start(key: &str) -> Result<()> {
+    MANAGER.write().await.start(key).await
+}
+
+async fn stop(key: &str) -> Result<()> {
+    MANAGER.write().await.stop(key).await
+}
+
+async fn reload(key: &str) -> Result<()> {
+    MANAGER.read().await.reload(key).await
+}
+
+async fn sig(key: &str, sig: &str) -> Result<()> {
+    let sig: Signal = sig.parse().map_err(|_| Error::InvalidSignal(sig.into()))?;
+    signal(key, sig).await
+}
+
+async fn child_for(key: &str) -> Result<Child> {
+    MANAGER.read().await.child_for(key)
+}
+
+async fn signal(key: &str, sig: Signal) -> Result<()> {
+    child_for(key).await?.kill(sig)
+}
+
+fn child_key(cg: &str, svc: &str) -> String {
+    [cg, svc].join("/")
+}
+fn split_key(key: &str) -> Result<(&str, &str)> {
+    key.split_once('/').ok_or(Error::InvalidKey)
+}
+
+#[derive(Default)]
+struct Manager {
+    cfg: Config,
+    procs: Map<String, watch::Receiver<Child>>,
+    runners: Map<String, mpsc::Sender<runner::Cmd>>,
+}
+
+impl Manager {
+    fn status(&self) -> Vec<(String, Child)> {
+        (self.procs.iter())
+            .map(|(n, c)| (n.clone(), c.borrow().clone()))
+            .collect()
+    }
+
+    fn child_for(&self, key: &str) -> Result<Child> {
+        (self.procs.get(key))
+            .map(|c| c.borrow().clone())
+            .ok_or_else(|| Error::NotRunning(key.into()))
+    }
+
+    async fn apply_config(&mut self, new_cfg: Config) -> Result<()> {
+        // create and configure cgroups
+        for (name, cg) in &new_cfg.cgroups {
+            let cg_path = PathBuf::from(cgroup::ROOT).join(name);
+            fs::create_dir_all(&cg_path)
+                .await
+                .map_err(Error::CgroupSetup)?;
+
+            fs::write(
+                cg_path.join("cgroup.subtree_control"),
+                cg.controllers.as_bytes(),
+            )
+            .await
+            .map_err(Error::CgroupSetup)?;
+
+            for (setting, value) in &cg.settings {
+                fs::write(cg_path.join(setting), value.as_bytes())
+                    .await
+                    .map_err(Error::CgroupSetup)?;
+            }
+        }
+
+        let new_svcs: Set<_> = new_cfg.service_keys().collect();
+
+        // stop removed services
+        let to_stop = Map::from_iter(self.runners.extract_if(.., |k, _| !new_svcs.contains(k)));
+        let mut stopped = Set::new();
+        for (key, runner_cmd) in to_stop {
+            if runner_cmd.send(runner::Cmd::Stop).await.is_err() {
+                // runner already dead
+                continue;
+            }
+            stopped.insert(key);
+        }
+
+        // start added services
+        for (key, cg, svc, service) in new_cfg.services() {
+            if self.runners.contains_key(&key) {
+                continue;
+            };
+
+            let cmd = self.spawn_runner(key, cg, svc, service.clone());
+            if let Err(e) = cmd.send(runner::Cmd::Start).await {
+                error!("runner instantly died: {e}");
+            }
+        }
+
+        // wait & cleanup stopped
+        for key in stopped {
+            let Some(mut child_rx) = self.procs.remove(&key) else {
+                continue;
+            };
+            let _ = child_rx
+                .wait_for(|c| matches!(c.state, State::Finalized))
+                .await;
+        }
+
+        self.cfg = new_cfg;
+        Ok(())
+    }
+
+    async fn terminate(&mut self) {
+        self.runners.clear();
+
+        for child in self.procs.values_mut() {
+            let _ = child
+                .wait_for(|c| matches!(c.state, State::Finalized))
+                .await;
+        }
+
+        self.procs.clear();
+    }
+
+    fn runner(&mut self, key: &str) -> Result<mpsc::Sender<runner::Cmd>> {
+        if let Some(c) = self.runners.get(key) {
+            return Ok(c.clone());
+        }
+
+        let (cg, svc) = split_key(key)?;
+        let service = self.cfg.service(key)?;
+
+        Ok(self.spawn_runner(key.into(), cg, svc, service.clone()))
+    }
+
+    fn spawn_runner(
+        &mut self,
+        key: String,
+        cg: &str,
+        svc: &str,
+        service: Service,
+    ) -> mpsc::Sender<runner::Cmd> {
+        let (runner, child_rx, cmds_tx) = runner::new(cg, svc, service);
+
+        tokio::spawn(runner.run());
+
+        self.procs.insert(key.clone(), child_rx);
+        self.runners.insert(key, cmds_tx.clone());
+
+        cmds_tx
+    }
+
+    async fn cmd(&mut self, key: &str, cmd: runner::Cmd) -> Result<()> {
+        if self.runner(key)?.send(cmd).await.is_err() {
+            // runner died
+            self.runners.remove(key);
+            return Err(Error::RunnerDead);
+        }
+        Ok(())
+    }
+
+    async fn start(&mut self, key: &str) -> Result<()> {
+        self.cmd(key, runner::Cmd::Start).await
+    }
+
+    async fn stop(&mut self, key: &str) -> Result<()> {
+        self.cmd(key, runner::Cmd::Stop).await
+    }
+
+    async fn reload(&self, key: &str) -> Result<()> {
+        let proc = (self.procs.get(key)) //
+            .ok_or_else(|| Error::UnknownService(key.into()))?;
+        proc.borrow().reload()
+    }
+}
+
+impl Config {
+    fn cgroup(&self, cg: &str) -> Result<&CgroupConfig> {
+        self.cgroups
+            .get(cg)
+            .ok_or_else(|| Error::UnknownCgroup(cg.into()))
+    }
+
+    fn service(&self, key: &str) -> Result<&Service> {
+        let (cg, svc) = split_key(key)?;
+        self.cgroup(cg)?.service(svc)
+    }
+
+    fn service_keys(&self) -> impl Iterator<Item = String> {
+        (self.cgroups.iter())
+            .map(|(cg_name, cg)| cg.services.keys().map(move |n| child_key(cg_name, n)))
+            .flatten()
+    }
+
+    fn services(&self) -> impl Iterator<Item = (String, &String, &String, &Service)> {
+        (self.cgroups.iter())
+            .map(|(cg_name, cg)| {
+                cg.services
+                    .iter()
+                    .map(move |(n, service)| (child_key(cg_name, n), cg_name, n, service))
+            })
+            .flatten()
+    }
+}
+
+impl CgroupConfig {
+    fn service(&self, svc: &str) -> Result<&Vec<String>> {
+        self.services
+            .get(svc)
+            .ok_or_else(|| Error::UnknownService(svc.into()))
+    }
+}
+
+pub async fn complete() -> Vec<String> {
+    let mut r = vec![];
+    let Ok(rd) = ctl_exec(b"ls\n").await else {
+        return r;
+    };
+    let mut rd = rd.lines();
+    while let Some(line) = rd.next_line().await.ok().flatten() {
+        r.push(line);
+    }
+    r
+}

src/rc/runner.rs

@@ -0,0 +1,263 @@
+use log::{error, warn};
+use nix::{
+    sys::signal::{Signal, kill},
+    unistd::Pid,
+};
+use std::num::NonZero;
+use tokio::{
+    process, select,
+    sync::{mpsc, watch},
+    time::{Duration, Instant, sleep, sleep_until},
+};
+
+use super::{Error, Result, Service};
+use crate::logger::Logger;
+
+const LOG_PATH: &str = "/var/log";
+
+const TERM_DELAY: Duration = Duration::from_secs(30);
+const KILL_DELAY: Duration = Duration::from_secs(10);
+const RESTART_DELAY: Duration = Duration::from_secs(8);
+
+#[derive(Debug, Clone, Copy)]
+pub enum Cmd {
+    Start,
+    Stop,
+}
+
+#[derive(Default, Clone, Copy, Debug)]
+pub enum State {
+    #[default]
+    NeverStarted,
+    Starting,
+    Running,
+    Crashed,
+    Stopping,
+    Stopped,
+    Finalized,
+}
+
+pub fn new(
+    cg: impl Into<String>,
+    svc: impl Into<String>,
+    service: Service,
+) -> (Runner, watch::Receiver<Child>, mpsc::Sender<Cmd>) {
+    let (manager, child_rx) = ProcessManager::new(service);
+    let (cmds_tx, cmds_rx) = mpsc::channel(1);
+    let r = Runner {
+        cg: cg.into(),
+        svc: svc.into(),
+        cmds_rx,
+        manager,
+    };
+    (r, child_rx, cmds_tx)
+}
+
+pub struct Runner {
+    cg: String,
+    svc: String,
+    cmds_rx: mpsc::Receiver<Cmd>,
+    manager: ProcessManager,
+}
+
+impl Runner {
+    pub async fn run(mut self) {
+        self.manager.update(State::NeverStarted);
+
+        loop {
+            let cmd = select! {
+                cmd = self.manager.manage() => {
+                    cmd
+                }
+                cmd = self.cmds_rx.recv() => {
+                    let Some(cmd) = cmd else {
+                        break; // command side dropped
+                    };
+                    Some(cmd)
+                }
+            };
+
+            if let Some(cmd) = cmd {
+                self.process_cmd(cmd).await;
+            }
+        }
+
+        self.process_cmd(Cmd::Stop).await;
+        self.manager.update(State::Finalized);
+    }
+
+    async fn process_cmd(&mut self, cmd: Cmd) {
+        let cg = &self.cg;
+        let svc = &self.svc;
+
+        match cmd {
+            Cmd::Start => {
+                self.manager.start(cg, svc).await;
+            }
+            Cmd::Stop => {
+                self.manager.stop().await;
+            }
+        }
+    }
+}
+
+struct ProcessManager {
+    service: Service,
+    child_tx: watch::Sender<Child>,
+    process: Option<process::Child>,
+    restart_deadline: Option<Instant>,
+}
+
+impl ProcessManager {
+    fn new(service: Service) -> (Self, watch::Receiver<Child>) {
+        let (child_tx, child_rx) = watch::channel(Child::default());
+        let pm = Self {
+            service,
+            child_tx,
+            process: None,
+            restart_deadline: None,
+        };
+        (pm, child_rx)
+    }
+
+    /// runs a management iteration (ie: waiting for the child or a restart deadline).
+    async fn manage(&mut self) -> Option<Cmd> {
+        if let Some(process) = self.process.as_mut() {
+            let msg = match process.wait().await {
+                Ok(status) => status.to_string(),
+                Err(e) => e.to_string(),
+            };
+            self.crashed(msg);
+            self.process = None;
+            self.restart_deadline = Some(Instant::now() + RESTART_DELAY);
+            None
+        } else if let Some(deadline) = self.restart_deadline {
+            sleep_until(deadline).await;
+            Some(Cmd::Start)
+        } else {
+            std::future::pending().await
+        }
+    }
+
+    async fn start(&mut self, cg: &str, svc: &str) {
+        if self.process.is_some() {
+            return;
+        }
+
+        self.update(State::Starting);
+
+        let logger = Logger {
+            log_path: LOG_PATH.into(),
+            log_name: svc.into(),
+            with_prefix: false,
+            cgroup: Some(cg.into()),
+        };
+
+        let mut args = self.service.iter();
+        let Some(cmd) = args.next() else {
+            error!("{cg}/{svc}: empty command");
+            return;
+        };
+
+        let Ok(cmd) = (logger.setup(cmd, args).await)
+            .inspect_err(|e| self.crashed(format!("setup failed: {e}")))
+        else {
+            return;
+        };
+
+        let Ok(child) = logger
+            .spawn(cmd)
+            .inspect_err(|e| self.crashed(format!("exec failed: {e}")))
+        else {
+            return;
+        };
+
+        self.process = Some(child);
+        self.restart_deadline = None;
+        self.update(State::Running);
+    }
+
+    async fn stop(&mut self) {
+        self.restart_deadline = None;
+
+        let Some(mut process) = self.process.take() else {
+            return;
+        };
+        let Some(pid) = process.id() else {
+            let _ = process.wait().await; // already dead, reap it
+            self.update(State::Stopped);
+            return;
+        };
+
+        let pid = pid as i32;
+        self.update_full(pid, State::Stopping, None);
+        let pid = Pid::from_raw(pid);
+
+        let _ = kill(pid, Signal::SIGTERM).inspect_err(|e| error!("kill -TERM {pid} failed: {e}"));
+
+        select! {
+            _ = process.wait() => {
+                self.update(State::Stopped);
+                return
+            },
+            _ = sleep(TERM_DELAY) => {
+                warn!("process {pid} did not exit during the grace period, killing");
+                let _ = process.kill().await.inspect_err(|e| error!("kill -KILL {pid} failed: {e}"));
+            }
+        }
+
+        select! {
+            _ = process.wait() => {
+                self.update(State::Stopped);
+                return
+            },
+            _ = sleep(KILL_DELAY) => {
+                error!("process {pid} still alive after SIGKILL");
+            }
+        }
+    }
+
+    fn process_pid(&self) -> i32 {
+        (self.process.as_ref())
+            .and_then(|c| Some(c.id()? as i32))
+            .unwrap_or(0)
+    }
+
+    fn update(&self, state: State) {
+        let pid = self.process_pid();
+        self.update_full(pid, state, None);
+    }
+
+    fn update_full(&self, pid: i32, state: State, msg: Option<String>) {
+        self.child_tx.send_replace(Child {
+            pid: NonZero::new(pid),
+            state,
+            msg,
+        });
+    }
+
+    fn crashed(&self, msg: String) {
+        let pid = self.process_pid();
+        self.update_full(pid, State::Crashed, Some(msg));
+    }
+}
+
+#[derive(Clone, Default)]
+pub struct Child {
+    pub pid: Option<NonZero<i32>>,
+    pub state: State,
+    pub msg: Option<String>,
+}
+
+impl Child {
+    pub fn reload(&self) -> Result<()> {
+        self.kill(Signal::SIGHUP)
+    }
+
+    pub fn kill(&self, sig: Signal) -> Result<()> {
+        let Some(pid) = self.pid else {
+            return Err(Error::ProcessExited);
+        };
+        kill(Pid::from_raw(pid.into()), sig).map_err(|e| Error::KillFailed(e))
+    }
+}