harden plasmavmc image ingestion and internal execution paths

deployer/crates/deployer-ctl/src/power.rs

@@ -161,7 +161,9 @@ impl RedfishTarget {
 
         match action {
             PowerAction::Cycle => Ok(PowerState::Cycling),
-            PowerAction::On | PowerAction::Off | PowerAction::Refresh => self.refresh(&client).await,
+            PowerAction::On | PowerAction::Off | PowerAction::Refresh => {
+                self.refresh(&client).await
+            }
         }
     }
 
@@ -295,7 +297,12 @@ pub async fn request_reinstall(
 #[cfg(test)]
 mod tests {
     use super::*;
-    use axum::{extract::State, http::StatusCode, routing::{get, post}, Json, Router};
+    use axum::{
+        extract::State,
+        http::StatusCode,
+        routing::{get, post},
+        Json, Router,
+    };
     use serde_json::Value;
     use std::sync::{Arc, Mutex};
     use tokio::net::TcpListener;
@@ -303,7 +310,10 @@ mod tests {
     #[test]
     fn parse_redfish_short_reference_defaults_to_https() {
         let parsed = RedfishTarget::parse("redfish://lab-bmc/node01").unwrap();
-        assert_eq!(parsed.resource_url.as_str(), "https://lab-bmc/redfish/v1/Systems/node01");
+        assert_eq!(
+            parsed.resource_url.as_str(),
+            "https://lab-bmc/redfish/v1/Systems/node01"
+        );
     }
 
     #[test]
@@ -361,8 +371,14 @@ mod tests {
             addr
         ))
         .unwrap();
-        assert_eq!(target.perform(PowerAction::Refresh).await.unwrap(), PowerState::On);
+        assert_eq!(
-        assert_eq!(target.perform(PowerAction::Off).await.unwrap(), PowerState::On);
+            target.perform(PowerAction::Refresh).await.unwrap(),
+            PowerState::On
+        );
+        assert_eq!(
+            target.perform(PowerAction::Off).await.unwrap(),
+            PowerState::On
+        );
 
         let payloads = state.seen_payloads.lock().unwrap().clone();
         assert_eq!(payloads, vec![r#"{"ResetType":"ForceOff"}"#.to_string()]);

deployer/crates/deployer-ctl/src/remote.rs

@@ -31,5 +31,3 @@ pub async fn run_deployer_command(endpoint: &str, action: &str) -> Result<()> {
 
     Ok(())
 }
-
-

fiberlb/crates/fiberlb-server/src/config.rs

@@ -177,6 +177,10 @@ pub struct VipOwnershipConfig {
     /// Interface used for local VIP ownership.
     #[serde(default = "default_vip_ownership_interface")]
     pub interface: String,
+
+    /// Optional explicit `ip` command path used for local VIP ownership.
+    #[serde(default)]
+    pub ip_command: Option<String>,
 }
 
 fn default_vip_ownership_interface() -> String {
@@ -188,6 +192,7 @@ impl Default for VipOwnershipConfig {
         Self {
             enabled: false,
             interface: default_vip_ownership_interface(),
+            ip_command: None,
         }
     }
 }

fiberlb/crates/fiberlb-server/src/main.rs

@@ -41,26 +41,6 @@ struct Args {
     #[arg(long)]
     grpc_addr: Option<String>,
 
-    /// ChainFire endpoint for cluster coordination
-    #[arg(long, env = "FIBERLB_CHAINFIRE_ENDPOINT")]
-    chainfire_endpoint: Option<String>,
-
-    /// FlareDB endpoint for metadata and tenant data storage
-    #[arg(long, env = "FIBERLB_FLAREDB_ENDPOINT")]
-    flaredb_endpoint: Option<String>,
-
-    /// Metadata backend (flaredb, postgres, sqlite)
-    #[arg(long, env = "FIBERLB_METADATA_BACKEND")]
-    metadata_backend: Option<String>,
-
-    /// SQL database URL for metadata (required for postgres/sqlite backend)
-    #[arg(long, env = "FIBERLB_METADATA_DATABASE_URL")]
-    metadata_database_url: Option<String>,
-
-    /// Run in single-node mode (required when metadata backend is SQLite)
-    #[arg(long, env = "FIBERLB_SINGLE_NODE")]
-    single_node: bool,
-
     /// Log level (overrides config)
     #[arg(short, long)]
     log_level: Option<String>,
@@ -93,21 +73,6 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     if let Some(log_level) = args.log_level {
         config.log_level = log_level;
     }
-    if let Some(chainfire_endpoint) = args.chainfire_endpoint {
-        config.chainfire_endpoint = Some(chainfire_endpoint);
-    }
-    if let Some(flaredb_endpoint) = args.flaredb_endpoint {
-        config.flaredb_endpoint = Some(flaredb_endpoint);
-    }
-    if let Some(metadata_backend) = args.metadata_backend {
-        config.metadata_backend = parse_metadata_backend(&metadata_backend)?;
-    }
-    if let Some(metadata_database_url) = args.metadata_database_url {
-        config.metadata_database_url = Some(metadata_database_url);
-    }
-    if args.single_node {
-        config.single_node = true;
-    }
 
     // Initialize tracing
     tracing_subscriber::fmt()
@@ -194,12 +159,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
             )
         }
         MetadataBackend::Postgres | MetadataBackend::Sqlite => {
-            let database_url = config
+            let database_url = config.metadata_database_url.as_deref().ok_or_else(|| {
-                .metadata_database_url
-                .as_deref()
-                .ok_or_else(|| {
                 format!(
-                        "metadata_database_url is required when metadata_backend={} (env: FIBERLB_METADATA_DATABASE_URL)",
+                    "metadata_database_url is required when metadata_backend={}",
                     metadata_backend_name(config.metadata_backend)
                 )
             })?;
@@ -282,8 +244,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
         })?;
         let bgp = create_bgp_client(config.bgp.clone()).await?;
         let vip_owner: Option<Arc<dyn VipAddressOwner>> = if config.vip_ownership.enabled {
-            Some(Arc::new(KernelVipAddressOwner::new(
+            Some(Arc::new(KernelVipAddressOwner::with_ip_command(
                 config.vip_ownership.interface.clone(),
+                config.vip_ownership.ip_command.clone(),
             )))
         } else {
             None
@@ -439,19 +402,6 @@ async fn wait_for_shutdown_signal() -> Result<(), Box<dyn std::error::Error>> {
     Ok(())
 }
 
-fn parse_metadata_backend(value: &str) -> Result<MetadataBackend, Box<dyn std::error::Error>> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "flaredb" => Ok(MetadataBackend::FlareDb),
-        "postgres" => Ok(MetadataBackend::Postgres),
-        "sqlite" => Ok(MetadataBackend::Sqlite),
-        other => Err(format!(
-            "invalid metadata backend '{}'; expected one of: flaredb, postgres, sqlite",
-            other
-        )
-        .into()),
-    }
-}
-
 fn metadata_backend_name(backend: MetadataBackend) -> &'static str {
     match backend {
         MetadataBackend::FlareDb => "flaredb",

fiberlb/crates/fiberlb-server/src/metadata.rs

@@ -61,12 +61,12 @@ impl LbMetadataStore {
         endpoint: Option<String>,
         pd_endpoint: Option<String>,
     ) -> Result<Self> {
-        let endpoint = endpoint.unwrap_or_else(|| {
+        let endpoint = endpoint.unwrap_or_else(|| "127.0.0.1:2479".to_string());
-            std::env::var("FIBERLB_FLAREDB_ENDPOINT")
+        Self::connect_flaredb(endpoint, pd_endpoint).await
-                .unwrap_or_else(|_| "127.0.0.1:2479".to_string())
+    }
-        });
+
+    async fn connect_flaredb(endpoint: String, pd_endpoint: Option<String>) -> Result<Self> {
         let pd_endpoint = pd_endpoint
-            .or_else(|| std::env::var("FIBERLB_CHAINFIRE_ENDPOINT").ok())
             .map(|value| normalize_transport_addr(&value))
             .unwrap_or_else(|| endpoint.clone());
 

fiberlb/crates/fiberlb-server/src/vip_owner.rs

@@ -41,9 +41,17 @@ pub struct KernelVipAddressOwner {
 impl KernelVipAddressOwner {
     /// Create a kernel-backed VIP owner for the given interface.
     pub fn new(interface: impl Into<String>) -> Self {
+        Self::with_ip_command(interface, None::<String>)
+    }
+
+    /// Create a kernel-backed VIP owner with an optional explicit `ip` command path.
+    pub fn with_ip_command(
+        interface: impl Into<String>,
+        ip_command: Option<impl Into<String>>,
+    ) -> Self {
         Self {
             interface: interface.into(),
-            ip_command: resolve_ip_command(),
+            ip_command: resolve_ip_command(ip_command.map(Into::into)),
         }
     }
 
@@ -123,7 +131,14 @@ impl VipAddressOwner for KernelVipAddressOwner {
     }
 }
 
-fn resolve_ip_command() -> String {
+fn resolve_ip_command(configured: Option<String>) -> String {
+    if let Some(path) = configured {
+        let trimmed = path.trim();
+        if !trimmed.is_empty() {
+            return trimmed.to_string();
+        }
+    }
+
     if let Ok(path) = std::env::var("FIBERLB_IP_COMMAND") {
         let trimmed = path.trim();
         if !trimmed.is_empty() {
@@ -159,3 +174,37 @@ fn render_command_output(output: &std::process::Output) -> String {
 
     format!("exit status {}", output.status)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::sync::{Mutex, OnceLock};
+
+    fn env_lock() -> &'static Mutex<()> {
+        static LOCK: OnceLock<Mutex<()>> = OnceLock::new();
+        LOCK.get_or_init(|| Mutex::new(()))
+    }
+
+    #[test]
+    fn configured_ip_command_wins_over_env() {
+        let _guard = env_lock().lock().unwrap();
+        std::env::set_var("FIBERLB_IP_COMMAND", "/env/ip");
+
+        assert_eq!(
+            resolve_ip_command(Some("  /configured/ip  ".to_string())),
+            "/configured/ip"
+        );
+
+        std::env::remove_var("FIBERLB_IP_COMMAND");
+    }
+
+    #[test]
+    fn env_ip_command_is_used_when_config_is_absent() {
+        let _guard = env_lock().lock().unwrap();
+        std::env::set_var("FIBERLB_IP_COMMAND", " /env/ip ");
+
+        assert_eq!(resolve_ip_command(None), "/env/ip");
+
+        std::env::remove_var("FIBERLB_IP_COMMAND");
+    }
+}

flashdns/crates/flashdns-server/src/main.rs

@@ -1,27 +1,24 @@
 //! FlashDNS authoritative DNS server binary
 
+use anyhow::Result;
+use chainfire_client::Client as ChainFireClient;
+use clap::Parser;
 use flashdns_api::{RecordServiceServer, ZoneServiceServer};
 use flashdns_server::{
     config::{MetadataBackend, ServerConfig},
     dns::DnsHandler,
     metadata::DnsMetadataStore,
-    RecordServiceImpl,
+    RecordServiceImpl, ZoneServiceImpl,
-    ZoneServiceImpl,
 };
-use chainfire_client::Client as ChainFireClient;
 use iam_service_auth::AuthService;
 use metrics_exporter_prometheus::PrometheusBuilder;
+use std::path::PathBuf;
 use std::sync::Arc;
+use std::time::{SystemTime, UNIX_EPOCH};
 use tonic::transport::{Certificate, Identity, Server, ServerTlsConfig};
 use tonic::{Request, Status};
 use tonic_health::server::health_reporter;
 use tracing_subscriber::EnvFilter;
-use anyhow::Result;
-use clap::Parser;
-use std::path::PathBuf;
-use std::time::{SystemTime, UNIX_EPOCH};
-
-use config::{Config as Cfg, Environment, File, FileFormat};
 
 /// Command-line arguments for FlashDNS server.
 #[derive(Parser, Debug)]
@@ -39,26 +36,6 @@ struct CliArgs {
     #[arg(long)]
     dns_addr: Option<String>,
 
-    /// ChainFire endpoint for cluster coordination (overrides config)
-    #[arg(long, env = "FLASHDNS_CHAINFIRE_ENDPOINT")]
-    chainfire_endpoint: Option<String>,
-
-    /// FlareDB endpoint for metadata and tenant data storage (overrides config)
-    #[arg(long, env = "FLASHDNS_FLAREDB_ENDPOINT")]
-    flaredb_endpoint: Option<String>,
-
-    /// Metadata backend (flaredb, postgres, sqlite)
-    #[arg(long, env = "FLASHDNS_METADATA_BACKEND")]
-    metadata_backend: Option<String>,
-
-    /// SQL database URL for metadata (required for postgres/sqlite backend)
-    #[arg(long, env = "FLASHDNS_METADATA_DATABASE_URL")]
-    metadata_database_url: Option<String>,
-
-    /// Run in single-node mode (required when metadata backend is SQLite)
-    #[arg(long, env = "FLASHDNS_SINGLE_NODE")]
-    single_node: bool,
-
     /// Log level (overrides config)
     #[arg(short, long)]
     log_level: Option<String>,
@@ -72,54 +49,22 @@ struct CliArgs {
 async fn main() -> Result<(), Box<dyn std::error::Error>> {
     let cli_args = CliArgs::parse();
 
-    // Load configuration using config-rs
+    let mut config = if cli_args.config.exists() {
-    let mut settings = Cfg::builder()
-        // Layer 1: Application defaults. Serialize ServerConfig::default() into TOML.
-        .add_source(File::from_str(
-            toml::to_string(&ServerConfig::default())?.as_str(),
-            FileFormat::Toml,
-        ))
-        // Layer 2: Environment variables (e.g., FLASHDNS_GRPC_ADDR, FLASHDNS_LOG_LEVEL)
-        .add_source(
-            Environment::with_prefix("FLASHDNS")
-                .separator("__") // Use double underscore for nested fields
-        );
-
-    // Layer 3: Configuration file (if specified)
-    if cli_args.config.exists() {
         tracing::info!("Loading config from file: {}", cli_args.config.display());
-        settings = settings.add_source(File::from(cli_args.config.as_path()));
+        let contents = tokio::fs::read_to_string(&cli_args.config).await?;
+        toml::from_str(&contents)?
     } else {
-        tracing::info!("Config file not found, using defaults and environment variables.");
+        tracing::info!("Config file not found, using defaults.");
-    }
+        ServerConfig::default()
+    };
 
-    let mut config: ServerConfig = settings
+    // Apply command line overrides
-        .build()?
-        .try_deserialize()
-        .map_err(|e| anyhow::anyhow!("Failed to load configuration: {}", e))?;
-
-    // Apply command line overrides (Layer 4: highest precedence)
     if let Some(grpc_addr_str) = cli_args.grpc_addr {
         config.grpc_addr = grpc_addr_str.parse()?;
     }
     if let Some(dns_addr_str) = cli_args.dns_addr {
         config.dns_addr = dns_addr_str.parse()?;
     }
-    if let Some(chainfire_endpoint) = cli_args.chainfire_endpoint {
-        config.chainfire_endpoint = Some(chainfire_endpoint);
-    }
-    if let Some(flaredb_endpoint) = cli_args.flaredb_endpoint {
-        config.flaredb_endpoint = Some(flaredb_endpoint);
-    }
-    if let Some(metadata_backend) = cli_args.metadata_backend {
-        config.metadata_backend = parse_metadata_backend(&metadata_backend)?;
-    }
-    if let Some(metadata_database_url) = cli_args.metadata_database_url {
-        config.metadata_database_url = Some(metadata_database_url);
-    }
-    if cli_args.single_node {
-        config.single_node = true;
-    }
     if let Some(log_level) = cli_args.log_level {
         config.log_level = log_level;
     }
@@ -173,16 +118,15 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
                     config.chainfire_endpoint.clone(),
                 )
                 .await
-                    .map_err(|e| anyhow::anyhow!("Failed to initialize FlareDB metadata store: {}", e))?,
+                .map_err(|e| {
+                    anyhow::anyhow!("Failed to initialize FlareDB metadata store: {}", e)
+                })?,
             )
         }
         MetadataBackend::Postgres | MetadataBackend::Sqlite => {
-            let database_url = config
+            let database_url = config.metadata_database_url.as_deref().ok_or_else(|| {
-                .metadata_database_url
-                .as_deref()
-                .ok_or_else(|| {
                 anyhow::anyhow!(
-                        "metadata_database_url is required when metadata_backend={} (env: FLASHDNS_METADATA_DATABASE_URL)",
+                    "metadata_database_url is required when metadata_backend={}",
                     metadata_backend_name(config.metadata_backend)
                 )
             })?;
@@ -195,13 +139,18 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
             Arc::new(
                 DnsMetadataStore::new_sql(database_url, config.single_node)
                     .await
-                    .map_err(|e| anyhow::anyhow!("Failed to initialize SQL metadata store: {}", e))?,
+                    .map_err(|e| {
+                        anyhow::anyhow!("Failed to initialize SQL metadata store: {}", e)
+                    })?,
             )
         }
     };
 
     // Initialize IAM authentication service
-    tracing::info!("Connecting to IAM server at {}", config.auth.iam_server_addr);
+    tracing::info!(
+        "Connecting to IAM server at {}",
+        config.auth.iam_server_addr
+    );
     let auth_service = AuthService::new(&config.auth.iam_server_addr)
         .await
         .map_err(|e| anyhow::anyhow!("Failed to connect to IAM server: {}", e))?;
@@ -300,18 +249,6 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     Ok(())
 }
 
-fn parse_metadata_backend(value: &str) -> Result<MetadataBackend> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "flaredb" => Ok(MetadataBackend::FlareDb),
-        "postgres" => Ok(MetadataBackend::Postgres),
-        "sqlite" => Ok(MetadataBackend::Sqlite),
-        other => Err(anyhow::anyhow!(
-            "invalid metadata backend '{}'; expected one of: flaredb, postgres, sqlite",
-            other
-        )),
-    }
-}
-
 fn metadata_backend_name(backend: MetadataBackend) -> &'static str {
     match backend {
         MetadataBackend::FlareDb => "flaredb",
@@ -345,11 +282,7 @@ fn ensure_sql_backend_matches_url(backend: MetadataBackend, database_url: &str)
     }
 }
 
-async fn register_chainfire_membership(
+async fn register_chainfire_membership(endpoint: &str, service: &str, addr: String) -> Result<()> {
-    endpoint: &str,
-    service: &str,
-    addr: String,
-) -> Result<()> {
     let node_id =
         std::env::var("HOSTNAME").unwrap_or_else(|_| format!("{}-{}", service, std::process::id()));
     let ts = SystemTime::now()

flashdns/crates/flashdns-server/src/metadata.rs

@@ -57,12 +57,8 @@ impl DnsMetadataStore {
         endpoint: Option<String>,
         pd_endpoint: Option<String>,
     ) -> Result<Self> {
-        let endpoint = endpoint.unwrap_or_else(|| {
+        let endpoint = endpoint.unwrap_or_else(|| "127.0.0.1:2479".to_string());
-            std::env::var("FLASHDNS_FLAREDB_ENDPOINT")
-                .unwrap_or_else(|_| "127.0.0.1:2479".to_string())
-        });
         let pd_endpoint = pd_endpoint
-            .or_else(|| std::env::var("FLASHDNS_CHAINFIRE_ENDPOINT").ok())
             .map(|value| normalize_transport_addr(&value))
             .unwrap_or_else(|| endpoint.clone());
 

iam/crates/iam-server/src/config.rs

@@ -27,6 +27,14 @@ pub struct ServerConfig {
     /// Logging configuration
     #[serde(default)]
     pub logging: LoggingConfig,
+
+    /// Admin API policy configuration
+    #[serde(default)]
+    pub admin: AdminConfig,
+
+    /// Development-only safety valves
+    #[serde(default)]
+    pub dev: DevConfig,
 }
 
 impl ServerConfig {
@@ -102,6 +110,29 @@ impl ServerConfig {
             }
         }
 
+        if let Ok(value) = std::env::var("IAM_ALLOW_UNAUTHENTICATED_ADMIN")
+            .or_else(|_| std::env::var("PHOTON_IAM_ALLOW_UNAUTHENTICATED_ADMIN"))
+        {
+            let value = value.trim().to_ascii_lowercase();
+            config.admin.allow_unauthenticated =
+                matches!(value.as_str(), "1" | "true" | "yes" | "on");
+        }
+
+        if let Ok(value) = std::env::var("IAM_ALLOW_RANDOM_SIGNING_KEY")
+            .or_else(|_| std::env::var("PHOTON_IAM_ALLOW_RANDOM_SIGNING_KEY"))
+        {
+            let value = value.trim().to_ascii_lowercase();
+            config.dev.allow_random_signing_key =
+                matches!(value.as_str(), "1" | "true" | "yes" | "on");
+        }
+
+        if let Ok(value) = std::env::var("IAM_ALLOW_MEMORY_BACKEND")
+            .or_else(|_| std::env::var("PHOTON_IAM_ALLOW_MEMORY_BACKEND"))
+        {
+            let value = value.trim().to_ascii_lowercase();
+            config.dev.allow_memory_backend = matches!(value.as_str(), "1" | "true" | "yes" | "on");
+        }
+
         Ok(config)
     }
 
@@ -132,6 +163,8 @@ impl ServerConfig {
                 },
             },
             logging: LoggingConfig::default(),
+            admin: AdminConfig::default(),
+            dev: DevConfig::default(),
         }
     }
 }
@@ -226,6 +259,26 @@ pub struct ClusterConfig {
     pub chainfire_endpoint: Option<String>,
 }
 
+/// Admin API policy configuration
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct AdminConfig {
+    /// Allow admin APIs to run without an explicit admin token (dev only)
+    #[serde(default)]
+    pub allow_unauthenticated: bool,
+}
+
+/// Development-only safety valves
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct DevConfig {
+    /// Allow generating a random signing key when none is configured
+    #[serde(default)]
+    pub allow_random_signing_key: bool,
+
+    /// Allow the in-memory backend
+    #[serde(default)]
+    pub allow_memory_backend: bool,
+}
+
 /// Backend type
 #[derive(Debug, Clone, Copy, Default, Serialize, Deserialize)]
 #[serde(rename_all = "lowercase")]

iam/crates/iam-server/src/main.rs

@@ -64,8 +64,9 @@ fn load_admin_token() -> Option<String> {
         .filter(|value| !value.is_empty())
 }
 
-fn allow_unauthenticated_admin() -> bool {
+fn allow_unauthenticated_admin(config: &ServerConfig) -> bool {
-    std::env::var("IAM_ALLOW_UNAUTHENTICATED_ADMIN")
+    config.admin.allow_unauthenticated
+        || std::env::var("IAM_ALLOW_UNAUTHENTICATED_ADMIN")
             .or_else(|_| std::env::var("PHOTON_IAM_ALLOW_UNAUTHENTICATED_ADMIN"))
             .ok()
             .map(|value| {
@@ -334,7 +335,8 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
 
     // Create token service
     let signing_key = if config.authn.internal_token.signing_key.is_empty() {
-        let allow_random = std::env::var("IAM_ALLOW_RANDOM_SIGNING_KEY")
+        let allow_random = config.dev.allow_random_signing_key
+            || std::env::var("IAM_ALLOW_RANDOM_SIGNING_KEY")
                 .or_else(|_| std::env::var("PHOTON_IAM_ALLOW_RANDOM_SIGNING_KEY"))
                 .ok()
                 .map(|value| {
@@ -346,7 +348,7 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
                 .unwrap_or(false);
 
         if !allow_random {
-            return Err("No signing key configured. Set IAM_ALLOW_RANDOM_SIGNING_KEY=true for dev or configure authn.internal_token.signing_key.".into());
+            return Err("No signing key configured. Set dev.allow_random_signing_key=true (or IAM_ALLOW_RANDOM_SIGNING_KEY=true for legacy dev mode) or configure authn.internal_token.signing_key.".into());
         }
 
         warn!("No signing key configured, generating random key (dev-only)");
@@ -369,9 +371,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
 
     let token_service = Arc::new(InternalTokenService::new(token_config));
     let admin_token = load_admin_token();
-    if admin_token.is_none() && !allow_unauthenticated_admin() {
+    if admin_token.is_none() && !allow_unauthenticated_admin(&config) {
         return Err(
-            "IAM admin token not configured. Set IAM_ADMIN_TOKEN or explicitly allow dev mode with IAM_ALLOW_UNAUTHENTICATED_ADMIN=true."
+            "IAM admin token not configured. Set IAM_ADMIN_TOKEN or explicitly allow dev mode with admin.allow_unauthenticated=true."
                 .into(),
         );
     }
@@ -550,7 +552,8 @@ async fn create_backend(
 ) -> Result<Backend, Box<dyn std::error::Error>> {
     match config.store.backend {
         BackendKind::Memory => {
-            let allow_memory = std::env::var("IAM_ALLOW_MEMORY_BACKEND")
+            let allow_memory = config.dev.allow_memory_backend
+                || std::env::var("IAM_ALLOW_MEMORY_BACKEND")
                     .or_else(|_| std::env::var("PHOTON_IAM_ALLOW_MEMORY_BACKEND"))
                     .ok()
                     .map(|value| {
@@ -562,7 +565,7 @@ async fn create_backend(
                     .unwrap_or(false);
             if !allow_memory {
                 return Err(
-                    "In-memory IAM backend is disabled. Use FlareDB backend, or set IAM_ALLOW_MEMORY_BACKEND=true for tests/dev only."
+                    "In-memory IAM backend is disabled. Use FlareDB backend, or set dev.allow_memory_backend=true for tests/dev only."
                         .into(),
                 );
             }

k8shost/crates/k8shost-server/src/config.rs

@@ -91,6 +91,18 @@ impl Default for PrismNetConfig {
     }
 }
 
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct CreditServiceConfig {
+    #[serde(default)]
+    pub server_addr: Option<String>,
+}
+
+impl Default for CreditServiceConfig {
+    fn default() -> Self {
+        Self { server_addr: None }
+    }
+}
+
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
 pub struct ChainFireConfig {
     pub endpoint: Option<String>,
@@ -110,6 +122,8 @@ pub struct Config {
     pub flaredb: FlareDbConfig,
     pub chainfire: ChainFireConfig,
     pub iam: IamConfig,
+    #[serde(default)]
+    pub creditservice: CreditServiceConfig,
     pub fiberlb: FiberLbConfig,
     pub flashdns: FlashDnsConfig,
     pub prismnet: PrismNetConfig,

k8shost/crates/k8shost-server/src/main.rs

@@ -38,8 +38,8 @@ use tracing_subscriber::EnvFilter;
 #[command(about = "Kubernetes API server for PlasmaCloud's k8shost component")]
 struct Args {
     /// Configuration file path
-    #[arg(short, long)]
+    #[arg(short, long, default_value = "k8shost.toml")]
-    config: Option<PathBuf>,
+    config: PathBuf,
 
     /// Listen address for gRPC server (e.g., "[::]:6443")
     #[arg(long)]
@@ -91,17 +91,19 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     let args = Args::parse();
 
     // Load configuration
-    let mut settings = ::config::Config::builder()
+    let mut settings = ::config::Config::builder().add_source(::config::File::from_str(
-        .add_source(::config::File::from_str(
         toml::to_string(&Config::default())?.as_str(),
         ::config::FileFormat::Toml,
-        ))
+    ));
-        .add_source(::config::Environment::with_prefix("K8SHOST").separator("_"));
 
-    // Add config file if specified
+    if args.config.exists() {
-    if let Some(config_path) = &args.config {
+        info!("Loading config from file: {}", args.config.display());
-        info!("Loading config from file: {}", config_path.display());
+        settings = settings.add_source(::config::File::from(args.config.as_path()));
-        settings = settings.add_source(::config::File::from(config_path.as_path()));
+    } else {
+        info!(
+            "Config file not found: {}, using defaults",
+            args.config.display()
+        );
     }
 
     let loaded_config: Config = settings
@@ -136,6 +138,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
                 .iam_server_addr
                 .unwrap_or(loaded_config.iam.server_addr),
         },
+        creditservice: config::CreditServiceConfig {
+            server_addr: loaded_config.creditservice.server_addr,
+        },
         fiberlb: config::FiberLbConfig {
             server_addr: args
                 .fiberlb_server_addr
@@ -275,8 +280,14 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     let ipam_client = Arc::new(IpamClient::new(config.prismnet.server_addr.clone()));
 
     // Create service implementations with storage
+    let creditservice_endpoint = config.creditservice.server_addr.as_deref();
     let pod_service = Arc::new(
-        PodServiceImpl::new_with_credit_service(storage.clone(), auth_service.clone()).await,
+        PodServiceImpl::new_with_credit_service(
+            storage.clone(),
+            auth_service.clone(),
+            creditservice_endpoint,
+        )
+        .await,
     );
     let service_service = Arc::new(ServiceServiceImpl::new(
         storage.clone(),
@@ -290,7 +301,10 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     ));
 
     // Start scheduler in background with CreditService integration
-    let scheduler = Arc::new(scheduler::Scheduler::new_with_credit_service(storage.clone()).await);
+    let scheduler = Arc::new(
+        scheduler::Scheduler::new_with_credit_service(storage.clone(), creditservice_endpoint)
+            .await,
+    );
     tokio::spawn(async move {
         scheduler.run().await;
     });

k8shost/crates/k8shost-server/src/scheduler.rs

@@ -33,10 +33,11 @@ impl Scheduler {
     }
 
     /// Create a new scheduler with CreditService quota enforcement
-    pub async fn new_with_credit_service(storage: Arc<Storage>) -> Self {
+    pub async fn new_with_credit_service(storage: Arc<Storage>, endpoint: Option<&str>) -> Self {
         // Initialize CreditService client if endpoint is configured
-        let credit_service = match std::env::var("CREDITSERVICE_ENDPOINT") {
+        let credit_service = match endpoint {
-            Ok(endpoint) => match CreditServiceClient::connect(&endpoint).await {
+            Some(endpoint) if !endpoint.trim().is_empty() => {
+                match CreditServiceClient::connect(endpoint).await {
                     Ok(client) => {
                         info!(
                             "Scheduler: CreditService quota enforcement enabled: {}",
@@ -51,9 +52,12 @@ impl Scheduler {
                     );
                         None
                     }
-            },
+                }
-            Err(_) => {
+            }
-                info!("Scheduler: CREDITSERVICE_ENDPOINT not set, quota enforcement disabled");
+            _ => {
+                info!(
+                    "Scheduler: CreditService endpoint not configured, quota enforcement disabled"
+                );
                 None
             }
         };

k8shost/crates/k8shost-server/src/services/pod.rs

@@ -45,10 +45,15 @@ impl PodServiceImpl {
         }
     }
 
-    pub async fn new_with_credit_service(storage: Arc<Storage>, auth: Arc<AuthService>) -> Self {
+    pub async fn new_with_credit_service(
+        storage: Arc<Storage>,
+        auth: Arc<AuthService>,
+        endpoint: Option<&str>,
+    ) -> Self {
         // Initialize CreditService client if endpoint is configured
-        let credit_service = match std::env::var("CREDITSERVICE_ENDPOINT") {
+        let credit_service = match endpoint {
-            Ok(endpoint) => match CreditServiceClient::connect(&endpoint).await {
+            Some(endpoint) if !endpoint.trim().is_empty() => {
+                match CreditServiceClient::connect(endpoint).await {
                     Ok(client) => {
                         tracing::info!("CreditService admission control enabled: {}", endpoint);
                         Some(Arc::new(RwLock::new(client)))
@@ -60,9 +65,10 @@ impl PodServiceImpl {
                         );
                         None
                     }
-            },
+                }
-            Err(_) => {
+            }
-                tracing::info!("CREDITSERVICE_ENDPOINT not set, admission control disabled");
+            _ => {
+                tracing::info!("CreditService endpoint not configured, admission control disabled");
                 None
             }
         };

lightningstor/crates/lightningstor-server/src/config.rs

@@ -50,6 +50,75 @@ pub enum ObjectStorageBackend {
     Distributed,
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct S3AuthConfig {
+    #[serde(default = "default_s3_auth_enabled")]
+    pub enabled: bool,
+
+    #[serde(default = "default_s3_aws_region")]
+    pub aws_region: String,
+
+    #[serde(default = "default_s3_iam_cache_ttl_secs")]
+    pub iam_cache_ttl_secs: u64,
+
+    #[serde(default = "default_s3_default_org_id")]
+    pub default_org_id: Option<String>,
+
+    #[serde(default = "default_s3_default_project_id")]
+    pub default_project_id: Option<String>,
+
+    #[serde(default = "default_s3_max_auth_body_bytes")]
+    pub max_auth_body_bytes: usize,
+}
+
+impl Default for S3AuthConfig {
+    fn default() -> Self {
+        Self {
+            enabled: default_s3_auth_enabled(),
+            aws_region: default_s3_aws_region(),
+            iam_cache_ttl_secs: default_s3_iam_cache_ttl_secs(),
+            default_org_id: default_s3_default_org_id(),
+            default_project_id: default_s3_default_project_id(),
+            max_auth_body_bytes: default_s3_max_auth_body_bytes(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct S3PerformanceConfig {
+    #[serde(default = "default_s3_streaming_put_threshold_bytes")]
+    pub streaming_put_threshold_bytes: usize,
+
+    #[serde(default = "default_s3_inline_put_max_bytes")]
+    pub inline_put_max_bytes: usize,
+
+    #[serde(default = "default_s3_multipart_put_concurrency")]
+    pub multipart_put_concurrency: usize,
+
+    #[serde(default = "default_s3_multipart_fetch_concurrency")]
+    pub multipart_fetch_concurrency: usize,
+}
+
+impl Default for S3PerformanceConfig {
+    fn default() -> Self {
+        Self {
+            streaming_put_threshold_bytes: default_s3_streaming_put_threshold_bytes(),
+            inline_put_max_bytes: default_s3_inline_put_max_bytes(),
+            multipart_put_concurrency: default_s3_multipart_put_concurrency(),
+            multipart_fetch_concurrency: default_s3_multipart_fetch_concurrency(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct S3Config {
+    #[serde(default)]
+    pub auth: S3AuthConfig,
+
+    #[serde(default)]
+    pub performance: S3PerformanceConfig,
+}
+
 /// Server configuration
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct ServerConfig {
@@ -100,6 +169,10 @@ pub struct ServerConfig {
     /// Authentication configuration
     #[serde(default)]
     pub auth: AuthConfig,
+
+    /// S3 API runtime settings
+    #[serde(default)]
+    pub s3: S3Config,
 }
 
 /// Authentication configuration
@@ -114,6 +187,46 @@ fn default_iam_server_addr() -> String {
     "127.0.0.1:50051".to_string()
 }
 
+fn default_s3_auth_enabled() -> bool {
+    true
+}
+
+fn default_s3_aws_region() -> String {
+    "us-east-1".to_string()
+}
+
+fn default_s3_iam_cache_ttl_secs() -> u64 {
+    30
+}
+
+fn default_s3_default_org_id() -> Option<String> {
+    Some("default".to_string())
+}
+
+fn default_s3_default_project_id() -> Option<String> {
+    Some("default".to_string())
+}
+
+fn default_s3_max_auth_body_bytes() -> usize {
+    1024 * 1024 * 1024
+}
+
+fn default_s3_streaming_put_threshold_bytes() -> usize {
+    16 * 1024 * 1024
+}
+
+fn default_s3_inline_put_max_bytes() -> usize {
+    128 * 1024 * 1024
+}
+
+fn default_s3_multipart_put_concurrency() -> usize {
+    4
+}
+
+fn default_s3_multipart_fetch_concurrency() -> usize {
+    4
+}
+
 impl Default for AuthConfig {
     fn default() -> Self {
         Self {
@@ -139,6 +252,7 @@ impl Default for ServerConfig {
             sync_on_write: false,
             tls: None,
             auth: AuthConfig::default(),
+            s3: S3Config::default(),
         }
     }
 }

lightningstor/crates/lightningstor-server/src/main.rs

@@ -5,7 +5,7 @@ use clap::Parser;
 use iam_service_auth::AuthService;
 use lightningstor_api::{BucketServiceServer, ObjectServiceServer};
 use lightningstor_distributed::{
-    DistributedConfig, ErasureCodedBackend, RedundancyMode, ReplicatedBackend, RepairQueue,
+    DistributedConfig, ErasureCodedBackend, RedundancyMode, RepairQueue, ReplicatedBackend,
     StaticNodeRegistry,
 };
 use lightningstor_server::{
@@ -57,26 +57,6 @@ struct Args {
     #[arg(short, long)]
     log_level: Option<String>,
 
-    /// ChainFire endpoint for cluster coordination (overrides config)
-    #[arg(long, env = "LIGHTNINGSTOR_CHAINFIRE_ENDPOINT")]
-    chainfire_endpoint: Option<String>,
-
-    /// FlareDB endpoint for metadata and tenant data storage (overrides config)
-    #[arg(long, env = "LIGHTNINGSTOR_FLAREDB_ENDPOINT")]
-    flaredb_endpoint: Option<String>,
-
-    /// Metadata backend (flaredb, postgres, sqlite)
-    #[arg(long, env = "LIGHTNINGSTOR_METADATA_BACKEND")]
-    metadata_backend: Option<String>,
-
-    /// SQL database URL for metadata (required for postgres/sqlite backend)
-    #[arg(long, env = "LIGHTNINGSTOR_METADATA_DATABASE_URL")]
-    metadata_database_url: Option<String>,
-
-    /// Run in single-node mode (required when metadata backend is SQLite)
-    #[arg(long, env = "LIGHTNINGSTOR_SINGLE_NODE")]
-    single_node: bool,
-
     /// Data directory for object storage (overrides config)
     #[arg(long)]
     data_dir: Option<String>,
@@ -112,21 +92,6 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     if let Some(log_level) = args.log_level {
         config.log_level = log_level;
     }
-    if let Some(chainfire_endpoint) = args.chainfire_endpoint {
-        config.chainfire_endpoint = Some(chainfire_endpoint);
-    }
-    if let Some(flaredb_endpoint) = args.flaredb_endpoint {
-        config.flaredb_endpoint = Some(flaredb_endpoint);
-    }
-    if let Some(metadata_backend) = args.metadata_backend {
-        config.metadata_backend = parse_metadata_backend(&metadata_backend)?;
-    }
-    if let Some(metadata_database_url) = args.metadata_database_url {
-        config.metadata_database_url = Some(metadata_database_url);
-    }
-    if args.single_node {
-        config.single_node = true;
-    }
     if let Some(data_dir) = args.data_dir {
         config.data_dir = data_dir;
     }
@@ -187,12 +152,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
             )
         }
         MetadataBackend::Postgres | MetadataBackend::Sqlite => {
-            let database_url = config
+            let database_url = config.metadata_database_url.as_deref().ok_or_else(|| {
-                .metadata_database_url
-                .as_deref()
-                .ok_or_else(|| {
                 format!(
-                        "metadata_database_url is required when metadata_backend={} (env: LIGHTNINGSTOR_METADATA_DATABASE_URL)",
+                    "metadata_database_url is required when metadata_backend={}",
                     metadata_backend_name(config.metadata_backend)
                 )
             })?;
@@ -263,10 +225,11 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     let s3_addr: SocketAddr = config.s3_addr;
 
     // Start S3 HTTP server with shared state
-    let s3_router = s3::create_router_with_auth(
+    let s3_router = s3::create_router_with_auth_config(
         storage.clone(),
         metadata.clone(),
         Some(config.auth.iam_server_addr.clone()),
+        config.s3.clone(),
     );
     let s3_server = tokio::spawn(async move {
         tracing::info!("S3 HTTP server listening on {}", s3_addr);
@@ -341,19 +304,6 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     Ok(())
 }
 
-fn parse_metadata_backend(value: &str) -> Result<MetadataBackend, Box<dyn std::error::Error>> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "flaredb" => Ok(MetadataBackend::FlareDb),
-        "postgres" => Ok(MetadataBackend::Postgres),
-        "sqlite" => Ok(MetadataBackend::Sqlite),
-        other => Err(format!(
-            "invalid metadata backend '{}'; expected one of: flaredb, postgres, sqlite",
-            other
-        )
-        .into()),
-    }
-}
-
 fn metadata_backend_name(backend: MetadataBackend) -> &'static str {
     match backend {
         MetadataBackend::FlareDb => "flaredb",
@@ -442,7 +392,9 @@ async fn create_storage_backend(
         ObjectStorageBackend::LocalFs => {
             tracing::info!("Object storage backend: local_fs");
             Ok(StorageRuntime {
-                backend: Arc::new(LocalFsBackend::new(&config.data_dir, config.sync_on_write).await?),
+                backend: Arc::new(
+                    LocalFsBackend::new(&config.data_dir, config.sync_on_write).await?,
+                ),
                 repair_worker: None,
             })
         }

lightningstor/crates/lightningstor-server/src/metadata.rs

@@ -55,19 +55,18 @@ impl MetadataStore {
         endpoint: Option<String>,
         pd_endpoint: Option<String>,
     ) -> Result<Self> {
-        let endpoint = endpoint.unwrap_or_else(|| {
+        let endpoint = endpoint.unwrap_or_else(|| "127.0.0.1:2479".to_string());
-            std::env::var("LIGHTNINGSTOR_FLAREDB_ENDPOINT")
-                .unwrap_or_else(|_| "127.0.0.1:2479".to_string())
-        });
         let pd_endpoint = pd_endpoint
-            .or_else(|| std::env::var("LIGHTNINGSTOR_CHAINFIRE_ENDPOINT").ok())
             .map(|value| normalize_transport_addr(&value))
             .unwrap_or_else(|| endpoint.clone());
 
         let mut clients = Vec::with_capacity(FLAREDB_CLIENT_POOL_SIZE);
         for _ in 0..FLAREDB_CLIENT_POOL_SIZE {
-            let client =
+            let client = RdbClient::connect_with_pd_namespace(
-                RdbClient::connect_with_pd_namespace(endpoint.clone(), pd_endpoint.clone(), "lightningstor")
+                endpoint.clone(),
+                pd_endpoint.clone(),
+                "lightningstor",
+            )
             .await
             .map_err(|e| {
                 lightningstor_types::Error::StorageError(format!(
@@ -321,7 +320,11 @@ impl MetadataStore {
         Ok((results, next))
     }
 
-    async fn flaredb_put(clients: &[Arc<Mutex<RdbClient>>], key: &[u8], value: &[u8]) -> Result<()> {
+    async fn flaredb_put(
+        clients: &[Arc<Mutex<RdbClient>>],
+        key: &[u8],
+        value: &[u8],
+    ) -> Result<()> {
         let client = Self::flaredb_client_for_key(clients, key);
         let raw_result = {
             let mut c = client.lock().await;
@@ -443,7 +446,8 @@ impl MetadataStore {
         let client = Self::flaredb_scan_client(clients);
         let (mut items, next) = match {
             let mut c = client.lock().await;
-            c.raw_scan(start_key.clone(), end_key.clone(), fetch_limit).await
+            c.raw_scan(start_key.clone(), end_key.clone(), fetch_limit)
+                .await
         } {
             Ok((keys, values, next)) => {
                 let items = keys
@@ -697,7 +701,8 @@ impl MetadataStore {
                 .await
             }
             StorageBackend::Sql(sql) => {
-                let prefix_end = String::from_utf8(Self::prefix_end(prefix.as_bytes())).map_err(|e| {
+                let prefix_end =
+                    String::from_utf8(Self::prefix_end(prefix.as_bytes())).map_err(|e| {
                         lightningstor_types::Error::StorageError(format!(
                             "Failed to encode prefix end: {}",
                             e
@@ -908,7 +913,10 @@ impl MetadataStore {
     }
 
     fn multipart_bucket_prefix(bucket_id: &BucketId, prefix: &str) -> String {
-        format!("/lightningstor/multipart/by-bucket/{}/{}", bucket_id, prefix)
+        format!(
+            "/lightningstor/multipart/by-bucket/{}/{}",
+            bucket_id, prefix
+        )
     }
 
     fn multipart_object_key(object_id: &ObjectId) -> String {
@@ -955,7 +963,8 @@ impl MetadataStore {
     }
 
     pub async fn delete_replicated_repair_task(&self, task_id: &str) -> Result<()> {
-        self.delete_key(&Self::replicated_repair_task_key(task_id)).await
+        self.delete_key(&Self::replicated_repair_task_key(task_id))
+            .await
     }
 
     /// Save bucket metadata
@@ -1246,7 +1255,8 @@ impl MetadataStore {
             ))
             .await?;
         }
-        self.delete_key(&Self::multipart_upload_key(upload_id)).await
+        self.delete_key(&Self::multipart_upload_key(upload_id))
+            .await
     }
 
     pub async fn list_multipart_uploads(
@@ -1331,7 +1341,8 @@ impl MetadataStore {
     }
 
     pub async fn delete_object_multipart_upload(&self, object_id: &ObjectId) -> Result<()> {
-        self.delete_key(&Self::multipart_object_key(object_id)).await
+        self.delete_key(&Self::multipart_object_key(object_id))
+            .await
     }
 }
 
@@ -1380,13 +1391,11 @@ mod tests {
         store.delete_bucket(&bucket).await.unwrap();
         assert!(!store.bucket_cache.contains_key(&cache_key));
         assert!(!store.bucket_cache.contains_key(&cache_id_key));
-        assert!(
+        assert!(store
-            store
             .load_bucket("org-a", "project-a", "bench-bucket")
             .await
             .unwrap()
-                .is_none()
+            .is_none());
-        );
     }
 
     #[tokio::test]
@@ -1426,13 +1435,11 @@ mod tests {
             .await
             .unwrap();
         assert!(!store.object_cache.contains_key(&cache_key));
-        assert!(
+        assert!(store
-            store
             .load_object(&bucket.id, object.key.as_str(), None)
             .await
             .unwrap()
-                .is_none()
+            .is_none());
-        );
     }
 
     #[tokio::test]
@@ -1496,8 +1503,10 @@ mod tests {
         );
         store.save_bucket(&bucket).await.unwrap();
 
-        let upload_a = MultipartUpload::new(bucket.id.to_string(), ObjectKey::new("a/one.bin").unwrap());
+        let upload_a =
-        let upload_b = MultipartUpload::new(bucket.id.to_string(), ObjectKey::new("a/two.bin").unwrap());
+            MultipartUpload::new(bucket.id.to_string(), ObjectKey::new("a/one.bin").unwrap());
+        let upload_b =
+            MultipartUpload::new(bucket.id.to_string(), ObjectKey::new("a/two.bin").unwrap());
         let other_bucket = Bucket::new(
             BucketName::new("other-bucket").unwrap(),
             "org-a",
@@ -1505,8 +1514,10 @@ mod tests {
             "default",
         );
         store.save_bucket(&other_bucket).await.unwrap();
-        let upload_other =
+        let upload_other = MultipartUpload::new(
-            MultipartUpload::new(other_bucket.id.to_string(), ObjectKey::new("a/three.bin").unwrap());
+            other_bucket.id.to_string(),
+            ObjectKey::new("a/three.bin").unwrap(),
+        );
 
         store.save_multipart_upload(&upload_a).await.unwrap();
         store.save_multipart_upload(&upload_b).await.unwrap();
@@ -1543,10 +1554,7 @@ mod tests {
         assert_eq!(tasks[0].attempt_count, 1);
         assert_eq!(tasks[0].last_error.as_deref(), Some("transient failure"));
 
-        store
+        store.delete_replicated_repair_task(&task.id).await.unwrap();
-            .delete_replicated_repair_task(&task.id)
-            .await
-            .unwrap();
         assert!(store
             .list_replicated_repair_tasks(10)
             .await

lightningstor/crates/lightningstor-server/src/s3/auth.rs

@@ -3,6 +3,7 @@
 //! Implements simplified SigV4 authentication compatible with AWS S3 SDKs and aws-cli.
 //! Integrates with IAM for access key validation.
 
+use crate::config::S3AuthConfig;
 use crate::tenant::TenantContext;
 use axum::{
     body::{Body, Bytes},
@@ -23,8 +24,6 @@ use tracing::{debug, warn};
 use url::form_urlencoded;
 
 type HmacSha256 = Hmac<Sha256>;
-const DEFAULT_MAX_AUTH_BODY_BYTES: usize = 1024 * 1024 * 1024;
-
 #[derive(Clone, Debug)]
 pub(crate) struct VerifiedBodyBytes(pub Bytes);
 
@@ -49,6 +48,8 @@ pub struct AuthState {
     aws_region: String,
     /// AWS service name for SigV4 (e.g., s3)
     aws_service: String,
+    /// Maximum request body size to buffer during auth verification
+    max_auth_body_bytes: usize,
 }
 
 pub struct IamClient {
@@ -60,6 +61,8 @@ pub struct IamClient {
 enum IamClientMode {
     Env {
         credentials: std::collections::HashMap<String, String>,
+        default_org_id: Option<String>,
+        default_project_id: Option<String>,
     },
     Grpc {
         endpoint: String,
@@ -83,11 +86,11 @@ struct CachedCredential {
 impl IamClient {
     /// Create a new IAM client. If an endpoint is supplied, use the IAM gRPC API.
     pub fn new(iam_endpoint: Option<String>) -> Self {
-        let cache_ttl = std::env::var("LIGHTNINGSTOR_S3_IAM_CACHE_TTL_SECS")
+        Self::new_with_config(iam_endpoint, &S3AuthConfig::default())
-            .ok()
+    }
-            .and_then(|value| value.parse::<u64>().ok())
+
-            .map(StdDuration::from_secs)
+    pub fn new_with_config(iam_endpoint: Option<String>, config: &S3AuthConfig) -> Self {
-            .unwrap_or_else(|| StdDuration::from_secs(30));
+        let cache_ttl = StdDuration::from_secs(config.iam_cache_ttl_secs);
 
         if let Some(endpoint) = iam_endpoint
             .map(|value| normalize_iam_endpoint(&value))
@@ -106,6 +109,8 @@ impl IamClient {
         Self {
             mode: IamClientMode::Env {
                 credentials: Self::load_env_credentials(),
+                default_org_id: config.default_org_id.clone(),
+                default_project_id: config.default_project_id.clone(),
             },
             credential_cache: Arc::new(RwLock::new(HashMap::new())),
             cache_ttl,
@@ -160,32 +165,32 @@ impl IamClient {
     #[cfg(test)]
     fn env_credentials(&self) -> Option<&std::collections::HashMap<String, String>> {
         match &self.mode {
-            IamClientMode::Env { credentials } => Some(credentials),
+            IamClientMode::Env { credentials, .. } => Some(credentials),
             IamClientMode::Grpc { .. } => None,
         }
     }
 
-    fn env_default_tenant() -> (Option<String>, Option<String>) {
+    fn env_default_tenant(
-        let org_id = std::env::var("S3_TENANT_ORG_ID")
+        default_org_id: Option<String>,
-            .ok()
+        default_project_id: Option<String>,
-            .or_else(|| std::env::var("S3_ORG_ID").ok())
+    ) -> (Option<String>, Option<String>) {
-            .or_else(|| Some("default".to_string()));
+        (default_org_id, default_project_id)
-        let project_id = std::env::var("S3_TENANT_PROJECT_ID")
-            .ok()
-            .or_else(|| std::env::var("S3_PROJECT_ID").ok())
-            .or_else(|| Some("default".to_string()));
-        (org_id, project_id)
     }
 
     /// Validate access key and resolve the credential context.
     pub async fn get_credential(&self, access_key_id: &str) -> Result<ResolvedCredential, String> {
         match &self.mode {
-            IamClientMode::Env { credentials } => {
+            IamClientMode::Env {
+                credentials,
+                default_org_id,
+                default_project_id,
+            } => {
                 let secret_key = credentials
                     .get(access_key_id)
                     .cloned()
                     .ok_or_else(|| "Access key ID not found".to_string())?;
-                let (org_id, project_id) = Self::env_default_tenant();
+                let (org_id, project_id) =
+                    Self::env_default_tenant(default_org_id.clone(), default_project_id.clone());
                 Ok(ResolvedCredential {
                     secret_key,
                     principal_id: access_key_id.to_string(),
@@ -318,16 +323,21 @@ fn iam_admin_token() -> Option<String> {
 impl AuthState {
     /// Create new auth state with IAM integration
     pub fn new(iam_endpoint: Option<String>) -> Self {
-        let iam_client = Some(Arc::new(RwLock::new(IamClient::new(iam_endpoint))));
+        Self::new_with_config(iam_endpoint, &S3AuthConfig::default())
+    }
+
+    pub fn new_with_config(iam_endpoint: Option<String>, config: &S3AuthConfig) -> Self {
+        let iam_client = Some(Arc::new(RwLock::new(IamClient::new_with_config(
+            iam_endpoint,
+            config,
+        ))));
 
         Self {
             iam_client,
-            enabled: std::env::var("S3_AUTH_ENABLED")
+            enabled: config.enabled,
-                .unwrap_or_else(|_| "true".to_string())
+            aws_region: config.aws_region.clone(),
-                .parse()
-                .unwrap_or(true),
-            aws_region: std::env::var("AWS_REGION").unwrap_or_else(|_| "us-east-1".to_string()), // Default S3 region
             aws_service: "s3".to_string(),
+            max_auth_body_bytes: config.max_auth_body_bytes,
         }
     }
 
@@ -338,6 +348,7 @@ impl AuthState {
             enabled: false,
             aws_region: "us-east-1".to_string(),
             aws_service: "s3".to_string(),
+            max_auth_body_bytes: S3AuthConfig::default().max_auth_body_bytes,
         }
     }
 }
@@ -438,12 +449,8 @@ pub async fn sigv4_auth_middleware(
     let should_buffer_body = should_buffer_auth_body(payload_hash_header.as_deref());
 
     let body_bytes = if should_buffer_body {
-        let max_body_bytes = std::env::var("S3_MAX_AUTH_BODY_BYTES")
-            .ok()
-            .and_then(|value| value.parse::<usize>().ok())
-            .unwrap_or(DEFAULT_MAX_AUTH_BODY_BYTES);
         let (parts, body) = request.into_parts();
-        let body_bytes = match axum::body::to_bytes(body, max_body_bytes).await {
+        let body_bytes = match axum::body::to_bytes(body, auth_state.max_auth_body_bytes).await {
             Ok(b) => b,
             Err(e) => {
                 return error_response(

lightningstor/crates/lightningstor-server/src/s3/mod.rs

@@ -6,5 +6,8 @@ mod auth;
 mod router;
 mod xml;
 
-pub use auth::{AuthState, sigv4_auth_middleware};
+pub use auth::{sigv4_auth_middleware, AuthState};
-pub use router::{create_router, create_router_with_auth, create_router_with_state};
+pub use router::{
+    create_router, create_router_with_auth, create_router_with_auth_config,
+    create_router_with_state,
+};

lightningstor/crates/lightningstor-server/src/s3/router.rs

@@ -1,5 +1,6 @@
 //! S3 API router using Axum
 
+use crate::config::{S3Config, S3PerformanceConfig};
 use axum::{
     body::{Body, Bytes},
     extract::{Request, State},
@@ -14,8 +15,8 @@ use futures::{stream, stream::FuturesUnordered, StreamExt};
 use http_body_util::BodyExt;
 use md5::{Digest, Md5};
 use serde::Deserialize;
-use std::io;
 use sha2::Sha256;
+use std::io;
 use std::sync::Arc;
 use tokio::task::JoinHandle;
 
@@ -38,17 +39,24 @@ use super::xml::{
 pub struct S3State {
     pub storage: Arc<dyn StorageBackend>,
     pub metadata: Arc<MetadataStore>,
+    pub performance: S3PerformanceConfig,
 }
 
-// Keep streamed single-PUT parts aligned with the distributed backend's
-// large-object chunking so GET does not degrade into many small serial reads.
-const DEFAULT_STREAMING_PUT_THRESHOLD_BYTES: usize = 16 * 1024 * 1024;
-const DEFAULT_INLINE_PUT_MAX_BYTES: usize = 128 * 1024 * 1024;
-const DEFAULT_MULTIPART_PUT_CONCURRENCY: usize = 4;
-
 impl S3State {
     pub fn new(storage: Arc<dyn StorageBackend>, metadata: Arc<MetadataStore>) -> Self {
-        Self { storage, metadata }
+        Self::new_with_config(storage, metadata, S3PerformanceConfig::default())
+    }
+
+    pub fn new_with_config(
+        storage: Arc<dyn StorageBackend>,
+        metadata: Arc<MetadataStore>,
+        performance: S3PerformanceConfig,
+    ) -> Self {
+        Self {
+            storage,
+            metadata,
+            performance,
+        }
     }
 }
 
@@ -57,7 +65,12 @@ pub fn create_router_with_state(
     storage: Arc<dyn StorageBackend>,
     metadata: Arc<MetadataStore>,
 ) -> Router {
-    create_router_with_auth_state(storage, metadata, Arc::new(AuthState::new(None)))
+    create_router_with_auth_state(
+        storage,
+        metadata,
+        Arc::new(AuthState::new(None)),
+        S3PerformanceConfig::default(),
+    )
 }
 
 /// Create the S3-compatible HTTP router with auth and storage backends
@@ -66,15 +79,30 @@ pub fn create_router_with_auth(
     metadata: Arc<MetadataStore>,
     iam_endpoint: Option<String>,
 ) -> Router {
-    create_router_with_auth_state(storage, metadata, Arc::new(AuthState::new(iam_endpoint)))
+    create_router_with_auth_config(storage, metadata, iam_endpoint, S3Config::default())
+}
+
+pub fn create_router_with_auth_config(
+    storage: Arc<dyn StorageBackend>,
+    metadata: Arc<MetadataStore>,
+    iam_endpoint: Option<String>,
+    config: S3Config,
+) -> Router {
+    create_router_with_auth_state(
+        storage,
+        metadata,
+        Arc::new(AuthState::new_with_config(iam_endpoint, &config.auth)),
+        config.performance,
+    )
 }
 
 fn create_router_with_auth_state(
     storage: Arc<dyn StorageBackend>,
     metadata: Arc<MetadataStore>,
     auth_state: Arc<AuthState>,
+    performance: S3PerformanceConfig,
 ) -> Router {
-    let state = Arc::new(S3State::new(storage, metadata));
+    let state = Arc::new(S3State::new_with_config(storage, metadata, performance));
 
     Router::new()
         // Catch-all route for ALL operations (including root /)
@@ -126,28 +154,16 @@ fn request_tenant(extensions: &axum::http::Extensions) -> TenantContext {
         })
 }
 
-fn streaming_put_threshold_bytes() -> usize {
+fn streaming_put_threshold_bytes(state: &Arc<S3State>) -> usize {
-    std::env::var("LIGHTNINGSTOR_S3_STREAMING_PUT_THRESHOLD_BYTES")
+    state.performance.streaming_put_threshold_bytes
-        .ok()
-        .and_then(|value| value.parse::<usize>().ok())
-        .filter(|value| *value > 0)
-        .unwrap_or(DEFAULT_STREAMING_PUT_THRESHOLD_BYTES)
 }
 
-fn inline_put_max_bytes() -> usize {
+fn inline_put_max_bytes(state: &Arc<S3State>) -> usize {
-    std::env::var("LIGHTNINGSTOR_S3_INLINE_PUT_MAX_BYTES")
+    state.performance.inline_put_max_bytes
-        .ok()
-        .and_then(|value| value.parse::<usize>().ok())
-        .filter(|value| *value > 0)
-        .unwrap_or(DEFAULT_INLINE_PUT_MAX_BYTES)
 }
 
-fn multipart_put_concurrency() -> usize {
+fn multipart_put_concurrency(state: &Arc<S3State>) -> usize {
-    std::env::var("LIGHTNINGSTOR_S3_MULTIPART_PUT_CONCURRENCY")
+    state.performance.multipart_put_concurrency
-        .ok()
-        .and_then(|value| value.parse::<usize>().ok())
-        .filter(|value| *value > 0)
-        .unwrap_or(DEFAULT_MULTIPART_PUT_CONCURRENCY)
 }
 
 fn request_content_length(headers: &HeaderMap) -> Option<usize> {
@@ -751,13 +767,13 @@ async fn put_object(
             (body_len, etag, None, Some(body_bytes))
         }
         None => {
-            let prepared = if let Some(content_length) =
+            let prepared = if let Some(content_length) = content_length
-                content_length.filter(|content_length| *content_length <= inline_put_max_bytes())
+                .filter(|content_length| *content_length <= inline_put_max_bytes(&state))
             {
                 read_inline_put_body(
                     body,
                     verified_payload_hash.as_deref(),
-                    inline_put_max_bytes(),
+                    inline_put_max_bytes(&state),
                     Some(content_length),
                 )
                 .await
@@ -948,7 +964,7 @@ async fn stream_put_body(
 ) -> Result<PreparedPutBody, Response<Body>> {
     let verify_payload_hash = expected_payload_hash
         .filter(|expected_payload_hash| *expected_payload_hash != "UNSIGNED-PAYLOAD");
-    let threshold = streaming_put_threshold_bytes();
+    let threshold = streaming_put_threshold_bytes(state);
     let mut buffered = BytesMut::with_capacity(threshold);
     let mut full_md5 = Some(Md5::new());
     let mut full_sha256 = verify_payload_hash.map(|_| Sha256::new());
@@ -958,7 +974,7 @@ async fn stream_put_body(
     let mut scheduled_part_numbers = Vec::new();
     let mut completed_parts = Vec::new();
     let mut in_flight_uploads = FuturesUnordered::new();
-    let max_in_flight_uploads = multipart_put_concurrency();
+    let max_in_flight_uploads = multipart_put_concurrency(state);
 
     while let Some(frame) = body.frame().await {
         let frame = match frame {
@@ -1282,7 +1298,10 @@ fn multipart_object_body(state: Arc<S3State>, object: &Object, upload: Multipart
                     )));
                 }
 
-                return Ok(Some((bytes.slice(0..body_end), (storage, upload, idx, offset))));
+                return Ok(Some((
+                    bytes.slice(0..body_end),
+                    (storage, upload, idx, offset),
+                )));
             }
 
             Ok(None)
@@ -1374,7 +1393,11 @@ async fn get_object(
         );
     }
 
-    let multipart_upload = match state.metadata.load_object_multipart_upload(&object.id).await {
+    let multipart_upload = match state
+        .metadata
+        .load_object_multipart_upload(&object.id)
+        .await
+    {
         Ok(upload) => upload,
         Err(e) => {
             return error_response(
@@ -1385,7 +1408,10 @@ async fn get_object(
         }
     };
     let (body, content_length) = if let Some(upload) = multipart_upload {
-        (multipart_object_body(Arc::clone(&state), &object, upload), object.size as usize)
+        (
+            multipart_object_body(Arc::clone(&state), &object, upload),
+            object.size as usize,
+        )
     } else {
         let data = match state.storage.get_object(&object.id).await {
             Ok(data) => data,
@@ -1653,6 +1679,7 @@ mod tests {
             storage,
             metadata.clone(),
             Arc::new(AuthState::disabled()),
+            S3PerformanceConfig::default(),
         );
         std::mem::forget(tempdir);
         (router, metadata)
@@ -1982,7 +2009,8 @@ mod tests {
             .unwrap();
         assert_eq!(response.status(), StatusCode::OK);
 
-        let body = vec![b'x'; DEFAULT_STREAMING_PUT_THRESHOLD_BYTES + 1024];
+        let threshold = S3PerformanceConfig::default().streaming_put_threshold_bytes;
+        let body = vec![b'x'; threshold + 1024];
         let response = router
             .clone()
             .oneshot(
@@ -2197,10 +2225,12 @@ mod tests {
     async fn large_put_streams_multipart_parts_with_parallel_uploads() {
         let storage = Arc::new(DelayedMultipartStorage::new(Duration::from_millis(25)));
         let metadata = Arc::new(MetadataStore::new_in_memory());
+        let performance = S3PerformanceConfig::default();
         let router = create_router_with_auth_state(
             storage.clone(),
             metadata,
             Arc::new(AuthState::disabled()),
+            performance.clone(),
         );
 
         let response = router
@@ -2216,7 +2246,7 @@ mod tests {
             .unwrap();
         assert_eq!(response.status(), StatusCode::OK);
 
-        let body = vec![b'x'; (DEFAULT_STREAMING_PUT_THRESHOLD_BYTES * 2) + 4096];
+        let body = vec![b'x'; (performance.streaming_put_threshold_bytes * 2) + 4096];
         let response = router
             .clone()
             .oneshot(
@@ -2250,10 +2280,12 @@ mod tests {
     async fn moderate_put_with_content_length_stays_inline() {
         let storage = Arc::new(DelayedMultipartStorage::new(Duration::from_millis(25)));
         let metadata = Arc::new(MetadataStore::new_in_memory());
+        let performance = S3PerformanceConfig::default();
         let router = create_router_with_auth_state(
             storage.clone(),
             metadata.clone(),
             Arc::new(AuthState::disabled()),
+            performance.clone(),
         );
 
         let response = router
@@ -2269,7 +2301,7 @@ mod tests {
             .unwrap();
         assert_eq!(response.status(), StatusCode::OK);
 
-        let body = vec![b'y'; DEFAULT_STREAMING_PUT_THRESHOLD_BYTES + 4096];
+        let body = vec![b'y'; performance.streaming_put_threshold_bytes + 4096];
         let response = router
             .clone()
             .oneshot(

lightningstor/crates/lightningstor-server/src/s3/xml.rs

@@ -170,5 +170,8 @@ pub struct CompleteMultipartUploadResult {
 /// Convert to XML with declaration
 pub fn to_xml<T: Serialize>(value: &T) -> Result<String, quick_xml::DeError> {
     let xml = xml_to_string(value)?;
-    Ok(format!("<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n{}", xml))
+    Ok(format!(
+        "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n{}",
+        xml
+    ))
 }

nightlight/crates/nightlight-server/src/config.rs

@@ -6,6 +6,7 @@
 use anyhow::Result;
 use serde::{Deserialize, Serialize};
 use std::fs;
+use std::path::Path;
 
 /// Main server configuration
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -123,7 +124,7 @@ pub struct TlsConfig {
 
 impl Config {
     /// Load configuration from a YAML file
-    pub fn from_file(path: &str) -> Result<Self> {
+    pub fn from_file(path: impl AsRef<Path>) -> Result<Self> {
         let content = fs::read_to_string(path)?;
         let config = serde_yaml::from_str(&content)?;
         Ok(config)
@@ -136,27 +137,6 @@ impl Config {
         fs::write(path, content)?;
         Ok(())
     }
-
-    /// Apply environment variable overrides
-    ///
-    /// This allows NixOS service module to override configuration via environment variables.
-    /// Environment variables take precedence over configuration file values.
-    pub fn apply_env_overrides(&mut self) {
-        if let Ok(val) = std::env::var("NIGHTLIGHT_HTTP_ADDR") {
-            self.server.http_addr = val;
-        }
-        if let Ok(val) = std::env::var("NIGHTLIGHT_GRPC_ADDR") {
-            self.server.grpc_addr = val;
-        }
-        if let Ok(val) = std::env::var("NIGHTLIGHT_DATA_DIR") {
-            self.storage.data_dir = val;
-        }
-        if let Ok(val) = std::env::var("NIGHTLIGHT_RETENTION_DAYS") {
-            if let Ok(days) = val.parse() {
-                self.storage.retention_days = days;
-            }
-        }
-    }
 }
 
 impl Default for Config {

nightlight/crates/nightlight-server/src/grpc.rs

@@ -183,7 +183,11 @@ impl Admin for AdminServiceImpl {
         _request: Request<HealthRequest>,
     ) -> Result<Response<HealthResponse>, Status> {
         let storage_result = self.storage.stats().await;
-        let status = if storage_result.is_ok() { "ok" } else { "degraded" };
+        let status = if storage_result.is_ok() {
+            "ok"
+        } else {
+            "degraded"
+        };
         let storage_message = match &storage_result {
             Ok(_) => "storage ready".to_string(),
             Err(error) => error.to_string(),
@@ -253,7 +257,9 @@ impl Admin for AdminServiceImpl {
             version: env!("CARGO_PKG_VERSION").to_string(),
             commit: option_env!("GIT_COMMIT").unwrap_or("unknown").to_string(),
             build_time: option_env!("BUILD_TIME").unwrap_or("unknown").to_string(),
-            rust_version: option_env!("RUSTC_VERSION").unwrap_or("unknown").to_string(),
+            rust_version: option_env!("RUSTC_VERSION")
+                .unwrap_or("unknown")
+                .to_string(),
             target: format!("{}-{}", std::env::consts::ARCH, std::env::consts::OS),
         }))
     }
@@ -330,9 +336,7 @@ mod tests {
     use super::*;
     use crate::ingestion::IngestionService;
     use crate::storage::Storage;
-    use nightlight_api::nightlight::{
+    use nightlight_api::nightlight::{InstantQueryRequest, LabelValuesRequest, SeriesQueryRequest};
-        InstantQueryRequest, LabelValuesRequest, SeriesQueryRequest,
-    };
     use nightlight_api::prometheus::{Label, Sample, TimeSeries, WriteRequest};
 
     #[tokio::test]
@@ -380,7 +384,10 @@ mod tests {
             data.result[0].metric.get("__name__").map(String::as_str),
             Some("grpc_metric")
         );
-        assert_eq!(data.result[0].value.as_ref().map(|value| value.value), Some(12.5));
+        assert_eq!(
+            data.result[0].value.as_ref().map(|value| value.value),
+            Some(12.5)
+        );
     }
 
     #[tokio::test]
@@ -452,7 +459,10 @@ mod tests {
             .unwrap()
             .into_inner();
         assert_eq!(label_values.status, "success");
-        assert_eq!(label_values.data, vec!["api".to_string(), "worker".to_string()]);
+        assert_eq!(
+            label_values.data,
+            vec!["api".to_string(), "worker".to_string()]
+        );
     }
 
     #[tokio::test]
@@ -477,26 +487,33 @@ mod tests {
             .unwrap();
 
         let query = QueryService::from_storage(storage.queryable());
-        query.execute_instant_query("admin_metric", 2_000).await.unwrap();
+        query
+            .execute_instant_query("admin_metric", 2_000)
+            .await
+            .unwrap();
 
-        let admin = AdminServiceImpl::new(
+        let admin =
-            Arc::clone(&storage),
+            AdminServiceImpl::new(Arc::clone(&storage), ingestion.metrics(), query.metrics());
-            ingestion.metrics(),
-            query.metrics(),
-        );
         let stats = admin
             .stats(Request::new(StatsRequest {}))
             .await
             .unwrap()
             .into_inner();
 
-        assert_eq!(stats.storage.as_ref().map(|value| value.total_samples), Some(1));
         assert_eq!(
-            stats.ingestion
+            stats.storage.as_ref().map(|value| value.total_samples),
+            Some(1)
+        );
+        assert_eq!(
+            stats
+                .ingestion
                 .as_ref()
                 .map(|value| value.samples_ingested_total),
             Some(1)
         );
-        assert_eq!(stats.query.as_ref().map(|value| value.queries_total), Some(1));
+        assert_eq!(
+            stats.query.as_ref().map(|value| value.queries_total),
+            Some(1)
+        );
     }
 }

nightlight/crates/nightlight-server/src/ingestion.rs

@@ -15,8 +15,8 @@ use nightlight_api::prometheus::{Label, WriteRequest};
 use nightlight_types::Error;
 use prost::Message;
 use snap::raw::Decoder as SnappyDecoder;
-use std::sync::Arc;
 use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::Arc;
 use std::time::Instant;
 use tracing::{debug, error, info, warn};
 
@@ -113,9 +113,8 @@ impl IngestionService {
             }
 
             // Store series with samples in shared storage
-            let series_id = nightlight_types::SeriesId(
+            let series_id =
-                compute_series_fingerprint(&internal_labels)
+                nightlight_types::SeriesId(compute_series_fingerprint(&internal_labels));
-            );
 
             let time_series = nightlight_types::TimeSeries {
                 id: series_id,
@@ -178,11 +177,11 @@ impl IngestionMetrics {
 }
 
 /// Axum handler for /api/v1/write endpoint
-async fn handle_remote_write(
+async fn handle_remote_write(State(service): State<IngestionService>, body: Bytes) -> Response {
-    State(service): State<IngestionService>,
+    service
-    body: Bytes,
+        .metrics
-) -> Response {
+        .requests_total
-    service.metrics.requests_total.fetch_add(1, Ordering::Relaxed);
+        .fetch_add(1, Ordering::Relaxed);
 
     debug!("Received remote_write request, size: {} bytes", body.len());
 
@@ -225,17 +224,26 @@ async fn handle_remote_write(
         }
         Err(Error::Storage(msg)) if msg.contains("buffer full") => {
             warn!("Write buffer full, returning 429");
-            service.metrics.requests_failed.fetch_add(1, Ordering::Relaxed);
+            service
+                .metrics
+                .requests_failed
+                .fetch_add(1, Ordering::Relaxed);
             IngestionError::Backpressure.into_response()
         }
         Err(Error::InvalidLabel(msg)) => {
             warn!("Invalid labels: {}", msg);
-            service.metrics.requests_failed.fetch_add(1, Ordering::Relaxed);
+            service
+                .metrics
+                .requests_failed
+                .fetch_add(1, Ordering::Relaxed);
             IngestionError::InvalidLabels.into_response()
         }
         Err(e) => {
             error!("Failed to process write request: {}", e);
-            service.metrics.requests_failed.fetch_add(1, Ordering::Relaxed);
+            service
+                .metrics
+                .requests_failed
+                .fetch_add(1, Ordering::Relaxed);
             IngestionError::StorageError.into_response()
         }
     }
@@ -285,7 +293,11 @@ fn validate_labels(labels: Vec<Label>) -> Result<Vec<Label>, Error> {
         }
 
         // Label names must contain only [a-zA-Z0-9_]
-        if !label.name.chars().all(|c| c.is_ascii_alphanumeric() || c == '_') {
+        if !label
+            .name
+            .chars()
+            .all(|c| c.is_ascii_alphanumeric() || c == '_')
+        {
             return Err(Error::InvalidLabel(format!(
                 "Invalid label name '{}': must contain only [a-zA-Z0-9_]",
                 label.name
@@ -337,15 +349,12 @@ impl IntoResponse for IngestionError {
             IngestionError::InvalidProtobuf => {
                 (StatusCode::BAD_REQUEST, "Invalid protobuf encoding")
             }
-            IngestionError::InvalidLabels => {
+            IngestionError::InvalidLabels => (StatusCode::BAD_REQUEST, "Invalid metric labels"),
-                (StatusCode::BAD_REQUEST, "Invalid metric labels")
+            IngestionError::StorageError => (StatusCode::INTERNAL_SERVER_ERROR, "Storage error"),
-            }
+            IngestionError::Backpressure => (
-            IngestionError::StorageError => {
+                StatusCode::TOO_MANY_REQUESTS,
-                (StatusCode::INTERNAL_SERVER_ERROR, "Storage error")
+                "Write buffer full, retry later",
-            }
+            ),
-            IngestionError::Backpressure => {
-                (StatusCode::TOO_MANY_REQUESTS, "Write buffer full, retry later")
-            }
         };
 
         (status, message).into_response()

nightlight/crates/nightlight-server/src/main.rs

@@ -12,6 +12,7 @@ use std::time::Duration;
 
 use anyhow::Result;
 use axum::{routing::get, Router};
+use clap::Parser;
 use nightlight_api::nightlight::admin_server::AdminServer;
 use nightlight_api::nightlight::metric_query_server::MetricQueryServer;
 use tokio::time::MissedTickBehavior;
@@ -33,8 +34,18 @@ use storage::Storage;
 
 const DEFAULT_SNAPSHOT_INTERVAL_SECS: u64 = 30;
 
+#[derive(Parser, Debug)]
+#[command(name = "nightlight-server")]
+struct Args {
+    /// Configuration file path
+    #[arg(short, long, default_value = "nightlight.yaml")]
+    config: std::path::PathBuf,
+}
+
 #[tokio::main]
 async fn main() -> Result<()> {
+    let args = Args::parse();
+
     tracing_subscriber::fmt()
         .with_max_level(Level::INFO)
         .with_target(false)
@@ -44,17 +55,20 @@ async fn main() -> Result<()> {
     info!("Nightlight server starting");
     info!("Version: {}", env!("CARGO_PKG_VERSION"));
 
-    let mut config = match Config::from_file("config.yaml") {
+    let config = match Config::from_file(&args.config) {
         Ok(config) => {
-            info!("Configuration loaded from config.yaml");
+            info!("Configuration loaded from {}", args.config.display());
             config
         }
         Err(error) => {
-            info!("Failed to load config.yaml: {}, using defaults", error);
+            info!(
+                "Failed to load {}: {}, using defaults",
+                args.config.display(),
+                error
+            );
             Config::default()
         }
     };
-    config.apply_env_overrides();
 
     if config.tls.is_some() {
         warn!("Nightlight TLS configuration is currently ignored; starting plaintext listeners");
@@ -133,7 +147,9 @@ async fn main() -> Result<()> {
     info!("  - Admin.Health / Stats / BuildInfo");
 
     let shutdown = async {
-        tokio::signal::ctrl_c().await.expect("failed to install Ctrl+C handler");
+        tokio::signal::ctrl_c()
+            .await
+            .expect("failed to install Ctrl+C handler");
     };
     tokio::pin!(shutdown);
 
@@ -163,8 +179,9 @@ async fn maintenance_loop(
     config: StorageConfig,
     mut shutdown: tokio::sync::broadcast::Receiver<()>,
 ) {
-    let snapshot_interval_secs =
+    let snapshot_interval_secs = config
-        config.compaction_interval_seconds.clamp(5, DEFAULT_SNAPSHOT_INTERVAL_SECS);
+        .compaction_interval_seconds
+        .clamp(5, DEFAULT_SNAPSHOT_INTERVAL_SECS);
     let mut snapshot_interval = tokio::time::interval(Duration::from_secs(snapshot_interval_secs));
     snapshot_interval.set_missed_tick_behavior(MissedTickBehavior::Delay);
 

nix/modules/chainfire.nix

@@ -49,7 +49,7 @@ let
         raft_addr = raftAddr;
       })
     cfg.initialPeers;
-  chainfireConfigFile = tomlFormat.generate "chainfire.toml" {
+  generatedConfig = {
     node = {
       id = numericId cfg.nodeId;
       name = cfg.nodeId;
@@ -73,6 +73,7 @@ let
       role = cfg.raftRole;
     };
   };
+  chainfireConfigFile = tomlFormat.generate "chainfire.toml" (lib.recursiveUpdate generatedConfig cfg.settings);
 in
 {
   options.services.chainfire = {

nix/modules/fiberlb.nix

@@ -75,6 +75,8 @@ let
   fiberlbBaseConfig = {
     grpc_addr = "0.0.0.0:${toString cfg.port}";
     log_level = "info";
+    metadata_backend = cfg.metadataBackend;
+    single_node = cfg.singleNode;
     auth = {
       iam_server_addr =
         if cfg.iamAddr != null
@@ -93,6 +95,8 @@ let
     vip_ownership = {
       enabled = cfg.vipOwnership.enable;
       interface = cfg.vipOwnership.interface;
+    } // lib.optionalAttrs (cfg.vipOwnership.ipCommand != null) {
+      ip_command = cfg.vipOwnership.ipCommand;
     };
   } // lib.optionalAttrs cfg.bgp.enable {
     bgp =
@@ -127,6 +131,15 @@ let
       // lib.optionalAttrs (cfg.bgp.nextHop != null) {
         next_hop = cfg.bgp.nextHop;
       };
+  }
+  // lib.optionalAttrs (cfg.flaredbAddr != null) {
+    flaredb_endpoint = cfg.flaredbAddr;
+  }
+  // lib.optionalAttrs (normalizedDatabaseUrl != null) {
+    metadata_database_url = normalizedDatabaseUrl;
+  }
+  // lib.optionalAttrs (cfg.chainfireAddr != null) {
+    chainfire_endpoint = "http://${cfg.chainfireAddr}";
   };
   fiberlbConfigFile = tomlFormat.generate "fiberlb.toml" (lib.recursiveUpdate fiberlbBaseConfig cfg.settings);
   flaredbDependencies = lib.optional (cfg.metadataBackend == "flaredb") "flaredb.service";
@@ -222,6 +235,13 @@ in
         default = "lo";
         description = "Interface where FiberLB should claim VIP /32 addresses.";
       };
+
+      ipCommand = lib.mkOption {
+        type = lib.types.nullOr lib.types.str;
+        default = null;
+        description = "Optional explicit path to the `ip` command used for VIP ownership.";
+        example = "/run/current-system/sw/bin/ip";
+      };
     };
 
     vipDrain = {
@@ -367,22 +387,7 @@ in
         AmbientCapabilities = lib.optionals cfg.vipOwnership.enable [ "CAP_NET_ADMIN" ];
         CapabilityBoundingSet = lib.optionals cfg.vipOwnership.enable [ "CAP_NET_ADMIN" ];
 
-        # Environment variables for service endpoints
+        ExecStart = "${cfg.package}/bin/fiberlb --config ${fiberlbConfigFile}";
-        Environment = [
-          "RUST_LOG=info"
-          "FIBERLB_FLAREDB_ENDPOINT=${if cfg.flaredbAddr != null then cfg.flaredbAddr else "127.0.0.1:2479"}"
-          "FIBERLB_METADATA_BACKEND=${cfg.metadataBackend}"
-        ] ++ lib.optional (normalizedDatabaseUrl != null) "FIBERLB_METADATA_DATABASE_URL=${normalizedDatabaseUrl}"
-          ++ lib.optional cfg.singleNode "FIBERLB_SINGLE_NODE=true"
-          ++ lib.optional (cfg.chainfireAddr != null) "FIBERLB_CHAINFIRE_ENDPOINT=http://${cfg.chainfireAddr}";
-
-        # Start command
-        ExecStart = lib.concatStringsSep " " ([
-          "${cfg.package}/bin/fiberlb"
-          "--config ${fiberlbConfigFile}"
-          "--grpc-addr 0.0.0.0:${toString cfg.port}"
-          "--flaredb-endpoint ${if cfg.flaredbAddr != null then cfg.flaredbAddr else "127.0.0.1:2479"}"
-        ]);
       };
     };
   };

nix/modules/flashdns.nix

@@ -25,7 +25,7 @@ let
   // lib.optionalAttrs (cfg.databaseUrl != null) {
     metadata_database_url = cfg.databaseUrl;
   };
-  configFile = tomlFormat.generate "flashdns.toml" generatedConfig;
+  configFile = tomlFormat.generate "flashdns.toml" (lib.recursiveUpdate generatedConfig cfg.settings);
 in
 {
   options.services.flashdns = {

nix/modules/iam.nix

@@ -3,12 +3,19 @@
 let
   cfg = config.services.iam;
   tomlFormat = pkgs.formats.toml { };
-  iamConfigFile = tomlFormat.generate "iam.toml" {
+  generatedConfig = {
     server = {
       addr = "0.0.0.0:${toString cfg.port}";
       http_addr = "0.0.0.0:${toString cfg.httpPort}";
     };
     logging.level = "info";
+    admin = {
+      allow_unauthenticated = cfg.allowUnauthenticatedAdmin;
+    };
+    dev = {
+      allow_random_signing_key = cfg.allowRandomSigningKey;
+      allow_memory_backend = cfg.storeBackend == "memory";
+    };
     store = {
       backend = cfg.storeBackend;
       flaredb_endpoint =
@@ -25,6 +32,7 @@ let
       chainfire_endpoint = cfg.chainfireAddr;
     };
   };
+  iamConfigFile = tomlFormat.generate "iam.toml" (lib.recursiveUpdate generatedConfig cfg.settings);
 in
 {
   options.services.iam = {
@@ -87,6 +95,18 @@ in
       description = "Admin token injected as IAM_ADMIN_TOKEN for privileged IAM APIs.";
     };
 
+    allowRandomSigningKey = lib.mkOption {
+      type = lib.types.bool;
+      default = false;
+      description = "Allow IAM to generate a random signing key when authn.internal_token.signing_key is unset (dev only).";
+    };
+
+    allowUnauthenticatedAdmin = lib.mkOption {
+      type = lib.types.bool;
+      default = false;
+      description = "Allow privileged admin APIs without an admin token (dev only).";
+    };
+
     settings = lib.mkOption {
       type = lib.types.attrs;
       default = {};
@@ -119,20 +139,6 @@ in
       wants = [ "chainfire.service" "flaredb.service" ];
 
       environment = lib.mkMerge [
-        {
-          CHAINFIRE_ENDPOINT = if cfg.chainfireAddr != null then cfg.chainfireAddr else "127.0.0.1:2379";
-          FLAREDB_ENDPOINT = if cfg.flaredbAddr != null then cfg.flaredbAddr else "127.0.0.1:2479";
-          IAM_STORE_BACKEND = cfg.storeBackend;
-        }
-        (lib.mkIf (cfg.databaseUrl != null) {
-          IAM_DATABASE_URL = cfg.databaseUrl;
-        })
-        (lib.mkIf cfg.singleNode {
-          IAM_SINGLE_NODE = "true";
-        })
-        (lib.mkIf (cfg.storeBackend == "memory") {
-          IAM_ALLOW_MEMORY_BACKEND = "1";
-        })
         (lib.mkIf (cfg.adminToken != null) {
           IAM_ADMIN_TOKEN = cfg.adminToken;
         })

nix/modules/k8shost.nix

@@ -2,6 +2,59 @@
 
 let
   cfg = config.services.k8shost;
+  tomlFormat = pkgs.formats.toml { };
+  generatedConfig = {
+    server = {
+      addr = "0.0.0.0:${toString cfg.port}";
+      http_addr = "127.0.0.1:${toString cfg.httpPort}";
+      log_level = "info";
+    };
+    flaredb = {
+      pd_addr = cfg.flaredbPdAddr;
+      direct_addr = cfg.flaredbDirectAddr;
+    };
+    chainfire = {
+      endpoint = cfg.chainfireAddr;
+    };
+    iam = {
+      server_addr =
+        if cfg.iamAddr != null
+        then cfg.iamAddr
+        else "http://127.0.0.1:50080";
+    };
+    creditservice =
+      lib.optionalAttrs
+        (
+          cfg.creditserviceAddr != null
+          || ((config.services ? creditservice) && config.services.creditservice.enable)
+        )
+        {
+          server_addr =
+            if cfg.creditserviceAddr != null
+            then cfg.creditserviceAddr
+            else "http://127.0.0.1:${toString config.services.creditservice.grpcPort}";
+        };
+    fiberlb = {
+      server_addr =
+        if cfg.fiberlbAddr != null
+        then cfg.fiberlbAddr
+        else "http://127.0.0.1:50085";
+    };
+    flashdns = {
+      server_addr =
+        if cfg.flashdnsAddr != null
+        then cfg.flashdnsAddr
+        else "http://127.0.0.1:50084";
+    };
+    prismnet = {
+      server_addr =
+        if cfg.prismnetAddr != null
+        then cfg.prismnetAddr
+        else "http://127.0.0.1:50081";
+    };
+  };
+  configFile = tomlFormat.generate "k8shost.toml" (lib.recursiveUpdate generatedConfig cfg.settings);
+  configPath = "/etc/k8shost/k8shost.toml";
 in
 {
   options.services.k8shost = {
@@ -26,6 +79,13 @@ in
       example = "http://10.0.0.1:50080";
     };
 
+    creditserviceAddr = lib.mkOption {
+      type = lib.types.nullOr lib.types.str;
+      default = null;
+      description = "CreditService endpoint address (http://host:port) for pod admission and scheduler quota enforcement.";
+      example = "http://10.0.0.1:3010";
+    };
+
     chainfireAddr = lib.mkOption {
       type = lib.types.nullOr lib.types.str;
       default = null;
@@ -123,24 +183,10 @@ in
         ProtectHome = true;
         ReadWritePaths = [ cfg.dataDir ];
 
-        # Environment variables for service endpoints
+        ExecStart = "${cfg.package}/bin/k8shost-server --config ${configPath}";
-        Environment = [
+      };
-          "RUST_LOG=info"
+    };
-        ];
 
-        # Start command
+    environment.etc."k8shost/k8shost.toml".source = configFile;
-        ExecStart = lib.concatStringsSep " " ([
-          "${cfg.package}/bin/k8shost-server"
-          "--addr 0.0.0.0:${toString cfg.port}"
-          "--http-addr 127.0.0.1:${toString cfg.httpPort}"
-        ] ++ lib.optional (cfg.iamAddr != null) "--iam-server-addr ${cfg.iamAddr}"
-          ++ lib.optional (cfg.chainfireAddr != null) "--chainfire-endpoint ${cfg.chainfireAddr}"
-          ++ lib.optional (cfg.prismnetAddr != null) "--prismnet-server-addr ${cfg.prismnetAddr}"
-          ++ lib.optional (cfg.flaredbPdAddr != null) "--flaredb-pd-addr ${cfg.flaredbPdAddr}"
-          ++ lib.optional (cfg.flaredbDirectAddr != null) "--flaredb-direct-addr ${cfg.flaredbDirectAddr}"
-          ++ lib.optional (cfg.fiberlbAddr != null) "--fiberlb-server-addr ${cfg.fiberlbAddr}"
-          ++ lib.optional (cfg.flashdnsAddr != null) "--flashdns-server-addr ${cfg.flashdnsAddr}");
-      };
-    };
   };
 }

nix/modules/lightningstor.nix

@@ -73,6 +73,22 @@ let
     // lib.optionalAttrs (cfg.distributedRegistryEndpoint != null) {
       registry_endpoint = cfg.distributedRegistryEndpoint;
     };
+    s3 = {
+      auth = {
+        enabled = cfg.s3AuthEnabled;
+        aws_region = cfg.s3AwsRegion;
+        iam_cache_ttl_secs = cfg.s3IamCacheTtlSecs;
+        default_org_id = cfg.s3DefaultOrgId;
+        default_project_id = cfg.s3DefaultProjectId;
+        max_auth_body_bytes = cfg.s3MaxAuthBodyBytes;
+      };
+      performance = {
+        streaming_put_threshold_bytes = cfg.s3StreamingPutThresholdBytes;
+        inline_put_max_bytes = cfg.s3InlinePutMaxBytes;
+        multipart_put_concurrency = cfg.s3MultipartPutConcurrency;
+        multipart_fetch_concurrency = cfg.s3MultipartFetchConcurrency;
+      };
+    };
   }
   // lib.optionalAttrs (cfg.flaredbAddr != null) {
     flaredb_endpoint = cfg.flaredbAddr;
@@ -321,6 +337,54 @@ in
       description = "Maximum concurrent multipart GET part fetches.";
     };
 
+    s3AuthEnabled = lib.mkOption {
+      type = lib.types.bool;
+      default = true;
+      description = "Enable S3 Signature V4 authentication.";
+    };
+
+    s3AwsRegion = lib.mkOption {
+      type = lib.types.str;
+      default = "us-east-1";
+      description = "AWS region name exposed by the S3 compatibility layer.";
+    };
+
+    s3IamCacheTtlSecs = lib.mkOption {
+      type = lib.types.int;
+      default = 30;
+      description = "Seconds to cache IAM-backed S3 credential lookups.";
+    };
+
+    s3DefaultOrgId = lib.mkOption {
+      type = lib.types.nullOr lib.types.str;
+      default = "default";
+      description = "Default org ID used for static S3 credentials.";
+    };
+
+    s3DefaultProjectId = lib.mkOption {
+      type = lib.types.nullOr lib.types.str;
+      default = "default";
+      description = "Default project ID used for static S3 credentials.";
+    };
+
+    s3MaxAuthBodyBytes = lib.mkOption {
+      type = lib.types.int;
+      default = 1024 * 1024 * 1024;
+      description = "Maximum request body size buffered during S3 auth verification.";
+    };
+
+    s3AccessKeyId = lib.mkOption {
+      type = lib.types.nullOr lib.types.str;
+      default = null;
+      description = "Optional static S3 access key ID injected via environment for dev/test.";
+    };
+
+    s3SecretKey = lib.mkOption {
+      type = lib.types.nullOr lib.types.str;
+      default = null;
+      description = "Optional static S3 secret key injected via environment for dev/test.";
+    };
+
     databaseUrl = lib.mkOption {
       type = lib.types.nullOr lib.types.str;
       default = null;
@@ -360,6 +424,13 @@ in
   };
 
   config = lib.mkIf cfg.enable {
+    assertions = [
+      {
+        assertion = (cfg.s3AccessKeyId == null) == (cfg.s3SecretKey == null);
+        message = "services.lightningstor.s3AccessKeyId and services.lightningstor.s3SecretKey must be set together";
+      }
+    ];
+
     users.users.lightningstor = {
       isSystemUser = true;
       group = "lightningstor";
@@ -391,17 +462,12 @@ in
         ExecStart = execStart;
       };
 
-      environment = {
+      environment = lib.mkMerge [
-        RUST_LOG = "info";
+        (lib.mkIf (cfg.s3AccessKeyId != null) {
-        LIGHTNINGSTOR_S3_STREAMING_PUT_THRESHOLD_BYTES =
+          S3_ACCESS_KEY_ID = cfg.s3AccessKeyId;
-          toString cfg.s3StreamingPutThresholdBytes;
+          S3_SECRET_KEY = cfg.s3SecretKey;
-        LIGHTNINGSTOR_S3_INLINE_PUT_MAX_BYTES =
+        })
-          toString cfg.s3InlinePutMaxBytes;
+      ];
-        LIGHTNINGSTOR_S3_MULTIPART_PUT_CONCURRENCY =
-          toString cfg.s3MultipartPutConcurrency;
-        LIGHTNINGSTOR_S3_MULTIPART_FETCH_CONCURRENCY =
-          toString cfg.s3MultipartFetchConcurrency;
-      };
     };
   };
 }

nix/modules/nightlight.nix

@@ -2,6 +2,19 @@
 
 let
   cfg = config.services.nightlight;
+  yamlFormat = pkgs.formats.yaml { };
+  generatedConfig = {
+    server = {
+      grpc_addr = "0.0.0.0:${toString cfg.grpcPort}";
+      http_addr = "0.0.0.0:${toString cfg.httpPort}";
+    };
+    storage = {
+      data_dir = toString cfg.dataDir;
+      retention_days = cfg.retentionDays;
+    };
+  };
+  configFile = yamlFormat.generate "nightlight.yaml" (lib.recursiveUpdate generatedConfig cfg.settings);
+  configPath = "/etc/nightlight/nightlight.yaml";
 in
 {
   options.services.nightlight = {
@@ -79,19 +92,10 @@ in
         ProtectHome = true;
         ReadWritePaths = [ cfg.dataDir ];
 
-        # Start command
+        ExecStart = "${cfg.package}/bin/nightlight-server --config ${configPath}";
-        # Note: nightlight-server uses a config file, so we'll pass data-dir directly
+      };
-        # The config will be auto-generated or use defaults from Config::default()
+    };
-        ExecStart = "${cfg.package}/bin/nightlight-server";
 
-        # Environment variables for configuration
+    environment.etc."nightlight/nightlight.yaml".source = configFile;
-        Environment = [
-          "NIGHTLIGHT_HTTP_ADDR=0.0.0.0:${toString cfg.httpPort}"
-          "NIGHTLIGHT_GRPC_ADDR=0.0.0.0:${toString cfg.grpcPort}"
-          "NIGHTLIGHT_DATA_DIR=${cfg.dataDir}"
-          "NIGHTLIGHT_RETENTION_DAYS=${toString cfg.retentionDays}"
-        ];
-      };
-    };
   };
 }

nix/modules/plasmavmc.nix

@@ -27,7 +27,7 @@ let
     else if cfg.coronafsEndpoint != null then cfg.coronafsEndpoint
     else null;
   tomlFormat = pkgs.formats.toml { };
-  plasmavmcConfigFile = tomlFormat.generate "plasmavmc.toml" {
+  generatedConfig = {
     addr = "0.0.0.0:${toString cfg.port}";
     http_addr = "0.0.0.0:${toString cfg.httpPort}";
     log_level = "info";
@@ -37,7 +37,78 @@ let
         then cfg.iamAddr
         else "127.0.0.1:50080";
     };
+    kvm = {
+      qemu_path = "${pkgs.qemu}/bin/qemu-system-x86_64";
+      runtime_dir = "/run/libvirt/plasmavmc";
     };
+    storage = {
+      backend = "flaredb";
+      flaredb_endpoint =
+        if cfg.flaredbAddr != null
+        then cfg.flaredbAddr
+        else "127.0.0.1:2479";
+    } // lib.optionalAttrs (cfg.chainfireAddr != null) {
+      chainfire_endpoint = "http://${cfg.chainfireAddr}";
+    };
+    agent =
+      {
+        shared_live_migration = cfg.sharedLiveMigration;
+      }
+      // lib.optionalAttrs (cfg.mode != "server") {
+        node_id = cfg.nodeId;
+        node_name = cfg.nodeName;
+        heartbeat_interval_secs = cfg.heartbeatIntervalSeconds;
+      }
+      // lib.optionalAttrs (cfg.controlPlaneAddr != null) {
+        control_plane_addr = cfg.controlPlaneAddr;
+      }
+      // lib.optionalAttrs (cfg.advertiseAddr != null) {
+        advertise_endpoint = cfg.advertiseAddr;
+      };
+    integrations =
+      lib.optionalAttrs (cfg.prismnetAddr != null) {
+        prismnet_endpoint = "http://${cfg.prismnetAddr}";
+      };
+    watcher =
+      lib.optionalAttrs (cfg.chainfireAddr != null) {
+        enabled = true;
+      };
+    health =
+      lib.optionalAttrs (cfg.mode == "server") {
+        node_monitor_interval_secs = 5;
+        node_heartbeat_timeout_secs = 30;
+      };
+    artifacts =
+      {
+        image_cache_dir = "${toString cfg.dataDir}/images";
+      }
+      // lib.optionalAttrs (cfg.lightningstorAddr != null) {
+        lightningstor_endpoint = cfg.lightningstorAddr;
+      };
+    volumes = {
+      managed_volume_root = toString cfg.managedVolumeRoot;
+      coronafs =
+        (lib.optionalAttrs (effectiveCoronafsControllerEndpoint != null) {
+          controller_endpoint = effectiveCoronafsControllerEndpoint;
+        })
+        // (lib.optionalAttrs (effectiveCoronafsNodeEndpoint != null) {
+          node_endpoint = effectiveCoronafsNodeEndpoint;
+        })
+        // (lib.optionalAttrs (cfg.coronafsEndpoint != null) {
+          endpoint = cfg.coronafsEndpoint;
+        })
+        // {
+          node_local_attach = cfg.coronafsNodeLocalAttach || cfg.experimentalCoronafsNodeLocalAttach;
+        };
+      ceph =
+        (lib.optionalAttrs (cfg.cephMonitors != [ ]) {
+          monitors = cfg.cephMonitors;
+          cluster_id = cfg.cephClusterId;
+          user = cfg.cephUser;
+        });
+    };
+  };
+  plasmavmcConfigFile = tomlFormat.generate "plasmavmc.toml" (lib.recursiveUpdate generatedConfig cfg.settings);
 in
 {
   options.services.plasmavmc = {
@@ -286,64 +357,9 @@ in
           exit 1
         '';
 
-      environment = lib.mkMerge [
+      environment = lib.optionalAttrs (cfg.cephSecret != null) {
-        {
-          PLASMAVMC_MODE = cfg.mode;
-          PLASMAVMC_STORAGE_BACKEND = "flaredb";
-          PLASMAVMC_FLAREDB_ENDPOINT = if cfg.flaredbAddr != null then cfg.flaredbAddr else "127.0.0.1:2479";
-          PLASMAVMC_QEMU_PATH = "${pkgs.qemu}/bin/qemu-system-x86_64";
-          PLASMAVMC_RUNTIME_DIR = "/run/libvirt/plasmavmc";
-          PLASMAVMC_IMAGE_CACHE_DIR = "${toString cfg.dataDir}/images";
-          PLASMAVMC_MANAGED_VOLUME_ROOT = toString cfg.managedVolumeRoot;
-          PLASMAVMC_SHARED_LIVE_MIGRATION = lib.boolToString cfg.sharedLiveMigration;
-        }
-        (lib.mkIf (cfg.prismnetAddr != null) {
-          PRISMNET_ENDPOINT = "http://${cfg.prismnetAddr}";
-        })
-        (lib.mkIf (cfg.chainfireAddr != null) {
-          PLASMAVMC_CHAINFIRE_ENDPOINT = "http://${cfg.chainfireAddr}";
-          PLASMAVMC_STATE_WATCHER = "1";
-        })
-        (lib.mkIf (cfg.lightningstorAddr != null) {
-          PLASMAVMC_LIGHTNINGSTOR_ENDPOINT = cfg.lightningstorAddr;
-        })
-        (lib.mkIf (effectiveCoronafsControllerEndpoint != null) {
-          PLASMAVMC_CORONAFS_CONTROLLER_ENDPOINT = effectiveCoronafsControllerEndpoint;
-        })
-        (lib.mkIf (effectiveCoronafsNodeEndpoint != null) {
-          PLASMAVMC_CORONAFS_NODE_ENDPOINT = effectiveCoronafsNodeEndpoint;
-        })
-        (lib.mkIf (cfg.coronafsNodeLocalAttach || cfg.experimentalCoronafsNodeLocalAttach) {
-          PLASMAVMC_CORONAFS_NODE_LOCAL_ATTACH = "1";
-          PLASMAVMC_CORONAFS_ENABLE_EXPERIMENTAL_NODE_LOCAL_ATTACH = "1";
-        })
-        (lib.mkIf (cfg.coronafsEndpoint != null) {
-          PLASMAVMC_CORONAFS_ENDPOINT = cfg.coronafsEndpoint;
-        })
-        (lib.mkIf (cfg.cephMonitors != [ ]) {
-          PLASMAVMC_CEPH_MONITORS = lib.concatStringsSep "," cfg.cephMonitors;
-          PLASMAVMC_CEPH_CLUSTER_ID = cfg.cephClusterId;
-          PLASMAVMC_CEPH_USER = cfg.cephUser;
-        })
-        (lib.mkIf (cfg.cephSecret != null) {
         PLASMAVMC_CEPH_SECRET = cfg.cephSecret;
-        })
+      };
-        (lib.mkIf (cfg.mode != "server") {
-          PLASMAVMC_NODE_ID = cfg.nodeId;
-          PLASMAVMC_NODE_NAME = cfg.nodeName;
-          PLASMAVMC_NODE_HEARTBEAT_INTERVAL_SECS = toString cfg.heartbeatIntervalSeconds;
-        })
-        (lib.mkIf (cfg.controlPlaneAddr != null) {
-          PLASMAVMC_CONTROL_PLANE_ADDR = cfg.controlPlaneAddr;
-        })
-        (lib.mkIf (cfg.advertiseAddr != null) {
-          PLASMAVMC_ENDPOINT_ADVERTISE = cfg.advertiseAddr;
-        })
-        (lib.mkIf (cfg.mode == "server") {
-          PLASMAVMC_NODE_HEALTH_MONITOR_INTERVAL_SECS = "5";
-          PLASMAVMC_NODE_HEARTBEAT_TIMEOUT_SECS = "30";
-        })
-      ];
 
       serviceConfig = {
         Type = "simple";

nix/modules/prismnet.nix

@@ -3,17 +3,29 @@
 let
   cfg = config.services.prismnet;
   tomlFormat = pkgs.formats.toml { };
-  prismnetConfigFile = tomlFormat.generate "prismnet.toml" {
+  generatedConfig = {
     grpc_addr = "0.0.0.0:${toString cfg.port}";
     http_addr = "0.0.0.0:${toString cfg.httpPort}";
     log_level = "info";
+    metadata_backend = cfg.metadataBackend;
+    single_node = cfg.singleNode;
     auth = {
       iam_server_addr =
         if cfg.iamAddr != null
         then cfg.iamAddr
         else "127.0.0.1:50080";
     };
+  }
+  // lib.optionalAttrs (cfg.chainfireAddr != null) {
+    chainfire_endpoint = "http://${cfg.chainfireAddr}";
+  }
+  // lib.optionalAttrs (cfg.flaredbAddr != null) {
+    flaredb_endpoint = cfg.flaredbAddr;
+  }
+  // lib.optionalAttrs (cfg.databaseUrl != null) {
+    metadata_database_url = cfg.databaseUrl;
   };
+  prismnetConfigFile = tomlFormat.generate "prismnet.toml" (lib.recursiveUpdate generatedConfig cfg.settings);
 in
 {
   options.services.prismnet = {
@@ -108,22 +120,6 @@ in
       after = [ "network.target" "iam.service" "flaredb.service" "chainfire.service" ];
       wants = [ "iam.service" "flaredb.service" "chainfire.service" ];
 
-      environment = lib.mkMerge [
-        {
-          PRISMNET_FLAREDB_ENDPOINT = if cfg.flaredbAddr != null then cfg.flaredbAddr else "127.0.0.1:2479";
-          PRISMNET_METADATA_BACKEND = cfg.metadataBackend;
-        }
-        (lib.mkIf (cfg.databaseUrl != null) {
-          PRISMNET_METADATA_DATABASE_URL = cfg.databaseUrl;
-        })
-        (lib.mkIf cfg.singleNode {
-          PRISMNET_SINGLE_NODE = "1";
-        })
-        (lib.mkIf (cfg.chainfireAddr != null) {
-          PRISMNET_CHAINFIRE_ENDPOINT = "http://${cfg.chainfireAddr}";
-        })
-      ];
-
       serviceConfig = {
         Type = "simple";
         User = "prismnet";
@@ -143,12 +139,7 @@ in
         ReadWritePaths = [ cfg.dataDir ];
 
         # Start command
-        ExecStart = lib.concatStringsSep " " [
+        ExecStart = "${cfg.package}/bin/prismnet-server --config ${prismnetConfigFile}";
-          "${cfg.package}/bin/prismnet-server"
-          "--config ${prismnetConfigFile}"
-          "--grpc-addr 0.0.0.0:${toString cfg.port}"
-          "--flaredb-endpoint ${if cfg.flaredbAddr != null then cfg.flaredbAddr else "127.0.0.1:2479"}"
-        ];
       };
     };
   };

nix/templates/plasmacloud-single-node.nix

@@ -15,17 +15,13 @@
   services.flaredb = {
     enable = true;
     dataDir = "/var/lib/flaredb";
-    settings = {
+    pdAddr = "127.0.0.1:${toString config.services.chainfire.port}";
-      chainfire_endpoint = "http://127.0.0.1:${toString config.services.chainfire.port}";
-    };
   };
 
   services.iam = {
     enable = true;
     dataDir = "/var/lib/iam";
-    settings = {
+    flaredbAddr = "127.0.0.1:${toString config.services.flaredb.port}";
-      flaredb_endpoint = "http://127.0.0.1:${toString config.services.flaredb.port}";
-    };
   };
 
   services.plasmavmc.enable = true;

nix/test-cluster/node01.nix

@@ -71,6 +71,8 @@
     port = 50080;
     chainfireAddr = config.photonTestCluster.chainfireControlPlaneAddrs;
     flaredbAddr = config.photonTestCluster.flaredbControlPlaneAddrs;
+    allowRandomSigningKey = true;
+    allowUnauthenticatedAdmin = true;
   };
 
   services.prismnet = {
@@ -162,13 +164,6 @@
     flashdnsAddr = "http://10.100.0.11:50084";
   };
 
-  systemd.services.iam.environment = {
+  services.lightningstor.s3AccessKeyId = "photoncloud-test";
-    IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
+  services.lightningstor.s3SecretKey = "photoncloud-test-secret";
-    IAM_ALLOW_UNAUTHENTICATED_ADMIN = "true";
-  };
-
-  systemd.services.lightningstor.environment = {
-    S3_ACCESS_KEY_ID = "photoncloud-test";
-    S3_SECRET_KEY = "photoncloud-test-secret";
-  };
 }

nix/test-cluster/node02.nix

@@ -64,10 +64,7 @@
     port = 50080;
     chainfireAddr = config.photonTestCluster.chainfireControlPlaneAddrs;
     flaredbAddr = config.photonTestCluster.flaredbControlPlaneAddrs;
-  };
+    allowRandomSigningKey = true;
-
+    allowUnauthenticatedAdmin = true;
-  systemd.services.iam.environment = {
-    IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
-    IAM_ALLOW_UNAUTHENTICATED_ADMIN = "true";
   };
 }

nix/test-cluster/node03.nix

@@ -64,10 +64,7 @@
     port = 50080;
     chainfireAddr = config.photonTestCluster.chainfireControlPlaneAddrs;
     flaredbAddr = config.photonTestCluster.flaredbControlPlaneAddrs;
-  };
+    allowRandomSigningKey = true;
-
+    allowUnauthenticatedAdmin = true;
-  systemd.services.iam.environment = {
-    IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
-    IAM_ALLOW_UNAUTHENTICATED_ADMIN = "true";
   };
 }

nix/test-cluster/storage-node01.nix

@@ -69,6 +69,8 @@
     port = 50080;
     chainfireAddr = config.photonTestCluster.chainfireControlPlaneAddrs;
     flaredbAddr = config.photonTestCluster.flaredbControlPlaneAddrs;
+    allowRandomSigningKey = true;
+    allowUnauthenticatedAdmin = true;
   };
 
   services.plasmavmc = {
@@ -124,13 +126,6 @@
     region = "test";
   };
 
-  systemd.services.iam.environment = {
+  services.lightningstor.s3AccessKeyId = "photoncloud-test";
-    IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
+  services.lightningstor.s3SecretKey = "photoncloud-test-secret";
-    IAM_ALLOW_UNAUTHENTICATED_ADMIN = "true";
-  };
-
-  systemd.services.lightningstor.environment = {
-    S3_ACCESS_KEY_ID = "photoncloud-test";
-    S3_SECRET_KEY = "photoncloud-test-secret";
-  };
 }

nix/test-cluster/storage-node02.nix

@@ -66,10 +66,7 @@
     port = 50080;
     chainfireAddr = config.photonTestCluster.chainfireControlPlaneAddrs;
     flaredbAddr = config.photonTestCluster.flaredbControlPlaneAddrs;
-  };
+    allowRandomSigningKey = true;
-
+    allowUnauthenticatedAdmin = true;
-  systemd.services.iam.environment = {
-    IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
-    IAM_ALLOW_UNAUTHENTICATED_ADMIN = "true";
   };
 }

nix/test-cluster/storage-node03.nix

@@ -66,10 +66,7 @@
     port = 50080;
     chainfireAddr = config.photonTestCluster.chainfireControlPlaneAddrs;
     flaredbAddr = config.photonTestCluster.flaredbControlPlaneAddrs;
-  };
+    allowRandomSigningKey = true;
-
+    allowUnauthenticatedAdmin = true;
-  systemd.services.iam.environment = {
-    IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
-    IAM_ALLOW_UNAUTHENTICATED_ADMIN = "true";
   };
 }

nix/tests/fiberlb-native-bgp-ecmp-drain-vm-smoke.nix

@@ -173,10 +173,7 @@ in
           port = 50080;
           httpPort = 8083;
           storeBackend = "memory";
-        };
+          allowRandomSigningKey = true;
-
-        systemd.services.iam.environment = {
-          IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
         };
 
         services.fiberlb = {
@@ -260,10 +257,7 @@ in
           port = 50080;
           httpPort = 8083;
           storeBackend = "memory";
-        };
+          allowRandomSigningKey = true;
-
-        systemd.services.iam.environment = {
-          IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
         };
 
         services.fiberlb = {

nix/tests/fiberlb-native-bgp-interop-vm-smoke.nix

@@ -260,10 +260,7 @@ in
           port = 50080;
           httpPort = 8083;
           storeBackend = "memory";
-        };
+          allowRandomSigningKey = true;
-
-        systemd.services.iam.environment = {
-          IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
         };
 
         services.fiberlb = {

nix/tests/fiberlb-native-bgp-multipath-vm-smoke.nix

@@ -165,10 +165,7 @@ in
           port = 50080;
           httpPort = 8083;
           storeBackend = "memory";
-        };
+          allowRandomSigningKey = true;
-
-        systemd.services.iam.environment = {
-          IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
         };
 
         services.fiberlb = {

nix/tests/fiberlb-native-bgp-vm-smoke.nix

@@ -120,10 +120,7 @@ in
           port = 50080;
           httpPort = 8083;
           storeBackend = "memory";
-        };
+          allowRandomSigningKey = true;
-
-        systemd.services.iam.environment = {
-          IAM_ALLOW_RANDOM_SIGNING_KEY = "1";
         };
 
         services.fiberlb = {

plasmavmc/Cargo.lock

@@ -2458,6 +2458,18 @@ version = "0.10.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1d87ecb2933e8aeadb3e3a02b828fed80a7528047e68b4f424523a0981a3a084"
 
+[[package]]
+name = "nix"
+version = "0.29.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "71e2746dc3a24dd78b3cfcb7be93368c6de9963d30f43a6a73998a9cf4b17b46"
+dependencies = [
+ "bitflags 2.11.0",
+ "cfg-if",
+ "cfg_aliases",
+ "libc",
+]
+
 [[package]]
 name = "nom"
 version = "7.1.3"
@@ -2807,6 +2819,7 @@ name = "plasmavmc-kvm"
 version = "0.1.0"
 dependencies = [
  "async-trait",
+ "nix",
  "plasmavmc-hypervisor",
  "plasmavmc-types",
  "serde",
@@ -2853,6 +2866,7 @@ dependencies = [
  "reqwest 0.12.28",
  "serde",
  "serde_json",
+ "sha2",
  "tempfile",
  "thiserror 1.0.69",
  "tokio",

plasmavmc/crates/plasmavmc-kvm/Cargo.toml

@@ -15,6 +15,7 @@ tracing = { workspace = true }
 thiserror = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
+nix = { version = "0.29", features = ["signal"] }
 
 [lints]
 workspace = true

plasmavmc/crates/plasmavmc-kvm/src/env.rs

@@ -115,14 +115,20 @@ mod tests {
     fn resolve_runtime_dir_defaults() {
         let _guard = env_test_lock().lock().unwrap();
         std::env::remove_var(ENV_RUNTIME_DIR);
-        assert_eq!(resolve_runtime_dir(), PathBuf::from("/run/libvirt/plasmavmc"));
+        assert_eq!(
+            resolve_runtime_dir(),
+            PathBuf::from("/run/libvirt/plasmavmc")
+        );
     }
 
     #[test]
     fn resolve_runtime_dir_from_env() {
         let _guard = env_test_lock().lock().unwrap();
         std::env::set_var(ENV_RUNTIME_DIR, "/tmp/plasmavmc-runtime");
-        assert_eq!(resolve_runtime_dir(), PathBuf::from("/tmp/plasmavmc-runtime"));
+        assert_eq!(
+            resolve_runtime_dir(),
+            PathBuf::from("/tmp/plasmavmc-runtime")
+        );
         std::env::remove_var(ENV_RUNTIME_DIR);
     }
 

plasmavmc/crates/plasmavmc-kvm/src/lib.rs

@@ -8,18 +8,19 @@ mod qmp;
 
 use async_trait::async_trait;
 use env::{
-    resolve_kernel_initrd, resolve_nbd_aio_mode, resolve_nbd_max_queues, resolve_qcow2_path, resolve_qemu_path,
+    resolve_kernel_initrd, resolve_nbd_aio_mode, resolve_nbd_max_queues, resolve_qcow2_path,
-    resolve_qmp_timeout_secs, resolve_runtime_dir, ENV_QCOW2_PATH,
+    resolve_qemu_path, resolve_qmp_timeout_secs, resolve_runtime_dir, ENV_QCOW2_PATH,
 };
+use nix::sys::signal::{kill as nix_kill, Signal};
+use nix::unistd::Pid;
 use plasmavmc_hypervisor::{BackendCapabilities, HypervisorBackend, UnsupportedReason};
 use plasmavmc_types::{
-    AttachedDisk, DiskAttachment, DiskBus, DiskCache, Error, HypervisorType, NetworkSpec,
+    AttachedDisk, DiskAttachment, DiskBus, DiskCache, Error, HypervisorType, NetworkSpec, NicModel,
-    NicModel, Result, VirtualMachine, VmHandle, VmSpec, VmState, VmStatus, VolumeFormat,
+    Result, VirtualMachine, VmHandle, VmSpec, VmState, VmStatus, VolumeFormat,
 };
 use qmp::QmpClient;
 use serde_json::{json, Value};
 use std::path::{Path, PathBuf};
-use std::process::Stdio;
 use std::time::Duration;
 use tokio::process::Command;
 use tokio::{net::UnixStream, time::Instant};
@@ -62,31 +63,6 @@ fn volume_format_name(format: VolumeFormat) -> &'static str {
     }
 }
 
-fn build_rbd_uri(pool: &str, image: &str, monitors: &[String], user: &str) -> String {
-    let mut uri = format!("rbd:{pool}/{image}");
-    if !user.is_empty() {
-        uri.push_str(&format!(":id={user}"));
-    }
-    if !monitors.is_empty() {
-        uri.push_str(&format!(":mon_host={}", monitors.join(";")));
-    }
-    uri
-}
-
-fn disk_source_arg(disk: &AttachedDisk) -> Result<(String, &'static str)> {
-    match &disk.attachment {
-        DiskAttachment::File { path, format } => Ok((path.clone(), volume_format_name(*format))),
-        DiskAttachment::Nbd { uri, format } => Ok((uri.clone(), volume_format_name(*format))),
-        DiskAttachment::CephRbd {
-            pool,
-            image,
-            monitors,
-            user,
-            ..
-        } => Ok((build_rbd_uri(pool, image, monitors, user), "raw")),
-    }
-}
-
 fn effective_disk_cache(disk: &AttachedDisk) -> DiskCache {
     match (&disk.attachment, disk.cache) {
         // Shared NBD-backed volumes perform better and behave more predictably
@@ -96,14 +72,6 @@ fn effective_disk_cache(disk: &AttachedDisk) -> DiskCache {
     }
 }
 
-fn disk_cache_mode(cache: DiskCache) -> &'static str {
-    match cache {
-        DiskCache::None => "none",
-        DiskCache::Writeback => "writeback",
-        DiskCache::Writethrough => "writethrough",
-    }
-}
-
 fn disk_aio_mode(disk: &AttachedDisk) -> Option<&'static str> {
     match (&disk.attachment, disk.cache) {
         (DiskAttachment::File { .. }, DiskCache::None) => Some("native"),
@@ -125,7 +93,8 @@ fn disk_queue_count(vm: &VirtualMachine, disk: &AttachedDisk) -> u16 {
         return 1;
     }
 
-    vm.spec.cpu
+    vm.spec
+        .cpu
         .vcpus
         .clamp(1, resolve_nbd_max_queues().max(1) as u32) as u16
 }
@@ -153,6 +122,169 @@ fn qmp_timeout() -> Duration {
     Duration::from_secs(resolve_qmp_timeout_secs())
 }
 
+fn disk_cache_json(cache: DiskCache) -> Value {
+    json!({
+        "direct": matches!(cache, DiskCache::None),
+        "no-flush": false
+    })
+}
+
+fn validate_ceph_component(field_name: &str, value: &str) -> Result<()> {
+    if value.is_empty() {
+        return Err(Error::HypervisorError(format!("{field_name} is required")));
+    }
+    let mut chars = value.chars();
+    let Some(first) = chars.next() else {
+        return Err(Error::HypervisorError(format!("{field_name} is required")));
+    };
+    if !first.is_ascii_alphanumeric() {
+        return Err(Error::HypervisorError(format!(
+            "{field_name} must start with an ASCII alphanumeric character"
+        )));
+    }
+    if chars.any(|ch| !(ch.is_ascii_alphanumeric() || matches!(ch, '.' | '_' | '-'))) {
+        return Err(Error::HypervisorError(format!(
+            "{field_name} contains unsupported characters"
+        )));
+    }
+    Ok(())
+}
+
+fn parse_host_port(authority: &str, default_port: u16) -> Result<(String, u16)> {
+    if let Some(rest) = authority.strip_prefix('[') {
+        let (host, tail) = rest
+            .split_once(']')
+            .ok_or_else(|| Error::HypervisorError("invalid IPv6 authority".into()))?;
+        let port = tail
+            .strip_prefix(':')
+            .and_then(|value| value.parse::<u16>().ok())
+            .unwrap_or(default_port);
+        return Ok((host.to_string(), port));
+    }
+
+    if let Some((host, port)) = authority.rsplit_once(':') {
+        if !host.is_empty() {
+            if let Ok(port) = port.parse::<u16>() {
+                return Ok((host.to_string(), port));
+            }
+        }
+    }
+
+    Ok((authority.to_string(), default_port))
+}
+
+fn parse_nbd_uri(uri: &str) -> Result<(String, u16, Option<String>)> {
+    let remainder = uri
+        .strip_prefix("nbd://")
+        .ok_or_else(|| Error::HypervisorError(format!("unsupported NBD URI: {uri}")))?;
+    if remainder.contains('@') || remainder.contains('?') || remainder.contains('#') {
+        return Err(Error::HypervisorError(format!(
+            "unsupported NBD URI components: {uri}"
+        )));
+    }
+    let (authority, path) = remainder.split_once('/').unwrap_or((remainder, ""));
+    let (host, port) = parse_host_port(authority, 10809)?;
+    if host.is_empty() {
+        return Err(Error::HypervisorError(format!(
+            "missing NBD host in URI: {uri}"
+        )));
+    }
+    let export = (!path.is_empty()).then(|| path.to_string());
+    Ok((host, port, export))
+}
+
+fn parse_ceph_monitor(monitor: &str) -> Result<Value> {
+    let (host, port) = parse_host_port(monitor, 6789)?;
+    if host.is_empty() {
+        return Err(Error::HypervisorError(format!(
+            "invalid Ceph monitor address: {monitor}"
+        )));
+    }
+    Ok(json!({
+        "type": "inet",
+        "host": host,
+        "port": port.to_string()
+    }))
+}
+
+fn disk_blockdev_arg(disk: &AttachedDisk, disk_id: &str) -> Result<String> {
+    let effective_cache = effective_disk_cache(disk);
+    let aio_mode = disk_aio_mode(disk);
+    let file = match &disk.attachment {
+        DiskAttachment::File { path, .. } => {
+            let mut file = json!({
+                "driver": "file",
+                "filename": path
+            });
+            if let Some(aio_mode) = aio_mode {
+                file["aio"] = json!(aio_mode);
+            }
+            file
+        }
+        DiskAttachment::Nbd { uri, .. } => {
+            let (host, port, export) = parse_nbd_uri(uri)?;
+            let mut nbd = json!({
+                "driver": "nbd",
+                "server": {
+                    "type": "inet",
+                    "host": host,
+                    "port": port.to_string()
+                }
+            });
+            if let Some(export) = export {
+                nbd["export"] = json!(export);
+            }
+            if let Some(aio_mode) = aio_mode {
+                nbd["aio"] = json!(aio_mode);
+            }
+            nbd
+        }
+        DiskAttachment::CephRbd {
+            pool,
+            image,
+            monitors,
+            user,
+            ..
+        } => {
+            validate_ceph_component("ceph pool", pool)?;
+            validate_ceph_component("ceph image", image)?;
+            if !user.is_empty() {
+                validate_ceph_component("ceph user", user)?;
+            }
+            let servers: Vec<Value> = monitors
+                .iter()
+                .map(|monitor| parse_ceph_monitor(monitor))
+                .collect::<Result<Vec<_>>>()?;
+            let mut rbd = json!({
+                "driver": "rbd",
+                "pool": pool,
+                "image": image,
+                "server": servers
+            });
+            if !user.is_empty() {
+                rbd["user"] = json!(user);
+            }
+            rbd
+        }
+    };
+
+    let format_driver = match &disk.attachment {
+        DiskAttachment::File { format, .. } | DiskAttachment::Nbd { format, .. } => {
+            volume_format_name(*format)
+        }
+        DiskAttachment::CephRbd { .. } => "raw",
+    };
+
+    Ok(json!({
+        "node-name": format!("drive-{disk_id}"),
+        "driver": format_driver,
+        "read-only": disk.read_only,
+        "cache": disk_cache_json(effective_cache),
+        "file": file
+    })
+    .to_string())
+}
+
 fn build_disk_args(vm: &VirtualMachine, disks: &[AttachedDisk]) -> Result<Vec<String>> {
     if disks.is_empty() && vm.spec.disks.is_empty() {
         let qcow_path = resolve_qcow2_path().ok_or_else(|| {
@@ -167,8 +299,20 @@ fn build_disk_args(vm: &VirtualMachine, disks: &[AttachedDisk]) -> Result<Vec<St
             )));
         }
         return Ok(vec![
-            "-drive".into(),
+            "-blockdev".into(),
-            format!("file={},if=virtio,format=qcow2", qcow_path.display()),
+            json!({
+                "node-name": "drive-root",
+                "driver": "qcow2",
+                "read-only": false,
+                "cache": disk_cache_json(DiskCache::Writeback),
+                "file": {
+                    "driver": "file",
+                    "filename": qcow_path.display().to_string()
+                }
+            })
+            .to_string(),
+            "-device".into(),
+            "virtio-blk-pci,drive=drive-root,id=disk-root".into(),
         ]);
     }
 
@@ -205,21 +349,12 @@ fn build_disk_args(vm: &VirtualMachine, disks: &[AttachedDisk]) -> Result<Vec<St
 
     for (index, disk) in disks.into_iter().enumerate() {
         let disk_id = sanitize_device_component(&disk.id, index);
-        let (source, format_name) = disk_source_arg(disk)?;
         if disk_uses_dedicated_iothread(disk) {
             args.push("-object".into());
             args.push(format!("iothread,id=iothread-{disk_id}"));
         }
-        let effective_cache = effective_disk_cache(disk);
+        args.push("-blockdev".into());
-        let mut drive_arg = format!(
+        args.push(disk_blockdev_arg(disk, &disk_id)?);
-            "file={source},if=none,format={format_name},id=drive-{disk_id},cache={}",
-            disk_cache_mode(effective_cache)
-        );
-        if let Some(aio_mode) = disk_aio_mode(disk) {
-            drive_arg.push_str(&format!(",aio={aio_mode}"));
-        }
-        args.push("-drive".into());
-        args.push(drive_arg);
 
         let bootindex = bootindex_suffix(disk.boot_index);
         let device_arg = match disk.bus {
@@ -359,33 +494,24 @@ async fn wait_for_qmp(qmp_socket: &Path, timeout: Duration) -> Result<()> {
 }
 
 fn kill_pid(pid: u32) -> Result<()> {
-    let status = std::process::Command::new("kill")
+    let pid = Pid::from_raw(pid as i32);
-        .arg("-9")
+    match nix_kill(pid, Signal::SIGKILL) {
-        .arg(pid.to_string())
+        Ok(()) => Ok(()),
-        .stdout(Stdio::null())
+        Err(nix::errno::Errno::ESRCH) => Ok(()),
-        .stderr(Stdio::null())
+        Err(error) => Err(Error::HypervisorError(format!(
-        .status()
+            "failed to send SIGKILL to pid {}: {error}",
-        .map_err(|e| Error::HypervisorError(format!("Failed to invoke kill -9: {e}")))?;
+            pid.as_raw()
-    if status.success() {
+        ))),
-        Ok(())
-    } else if !pid_running(pid) {
-        Ok(())
-    } else {
-        Err(Error::HypervisorError(format!(
-            "kill -9 exited with status: {status}"
-        )))
     }
 }
 
 fn pid_running(pid: u32) -> bool {
-    std::process::Command::new("kill")
+    match nix_kill(Pid::from_raw(pid as i32), None::<Signal>) {
-        .arg("-0")
+        Ok(()) => true,
-        .arg(pid.to_string())
+        Err(nix::errno::Errno::EPERM) => true,
-        .stdout(Stdio::null())
+        Err(nix::errno::Errno::ESRCH) => false,
-        .stderr(Stdio::null())
+        Err(_) => false,
-        .status()
+    }
-        .map(|status| status.success())
-        .unwrap_or(false)
 }
 
 fn vm_stopped_out_of_band(handle: &VmHandle, qmp_socket: &Path) -> bool {
@@ -569,7 +695,9 @@ impl HypervisorBackend for KvmBackend {
 
             match QmpClient::connect(&qmp_socket).await {
                 Ok(mut client) => match client.query_status().await {
-                    Ok(status) if matches!(status.actual_state, VmState::Stopped | VmState::Failed) => {
+                    Ok(status)
+                        if matches!(status.actual_state, VmState::Stopped | VmState::Failed) =>
+                    {
                         break;
                     }
                     Ok(_) => {}
@@ -1109,10 +1237,12 @@ mod tests {
         let args_joined = args.join(" ");
         assert!(args_joined.contains("vm-root.qcow2"));
         assert!(args_joined.contains("vm-data.qcow2"));
+        assert!(args_joined.contains("-blockdev"));
         assert!(args_joined.contains("bootindex=1"));
-        assert!(args_joined.contains("cache=writeback"));
+        assert!(args_joined.contains("\"cache\":{\"direct\":true,\"no-flush\":false}"));
-        assert!(args_joined.contains("cache=none,aio=native"));
+        assert!(args_joined.contains("\"cache\":{\"direct\":false,\"no-flush\":false}"));
-        assert!(args_joined.contains("cache=writeback,aio=threads"));
+        assert!(args_joined.contains("\"aio\":\"native\""));
+        assert!(args_joined.contains("\"aio\":\"threads\""));
     }
 
     #[test]
@@ -1135,6 +1265,7 @@ mod tests {
         let console = PathBuf::from("/tmp/console.log");
         let args = build_qemu_args(&vm, &disks, &qmp, &console, None, None).unwrap();
         let args_joined = args.join(" ");
+        assert!(args_joined.contains("\"driver\":\"nbd\""));
         assert!(args_joined.contains("-object iothread,id=iothread-root"));
         assert!(args_joined.contains("virtio-blk-pci,drive=drive-root,id=disk-root,iothread=iothread-root,num-queues=4,queue-size=1024,bootindex=1"));
     }
@@ -1159,7 +1290,8 @@ mod tests {
         let console = PathBuf::from("/tmp/console.log");
         let args = build_qemu_args(&vm, &disks, &qmp, &console, None, None).unwrap();
         let args_joined = args.join(" ");
-        assert!(args_joined.contains("cache=none,aio=io_uring"));
+        assert!(args_joined.contains("\"cache\":{\"direct\":true,\"no-flush\":false}"));
+        assert!(args_joined.contains("\"aio\":\"io_uring\""));
     }
 
     #[test]
@@ -1182,7 +1314,8 @@ mod tests {
         let console = PathBuf::from("/tmp/console.log");
         let args = build_qemu_args(&vm, &disks, &qmp, &console, None, None).unwrap();
         let args_joined = args.join(" ");
-        assert!(args_joined.contains("cache=none,aio=io_uring"));
+        assert!(args_joined.contains("\"cache\":{\"direct\":true,\"no-flush\":false}"));
+        assert!(args_joined.contains("\"aio\":\"io_uring\""));
     }
 
     #[test]
@@ -1205,10 +1338,34 @@ mod tests {
         let console = PathBuf::from("/tmp/console.log");
         let args = build_qemu_args(&vm, &disks, &qmp, &console, None, None).unwrap();
         let args_joined = args.join(" ");
-        assert!(args_joined.contains("cache=none,aio=threads"));
+        assert!(args_joined.contains("\"cache\":{\"direct\":true,\"no-flush\":false}"));
+        assert!(args_joined.contains("\"aio\":\"threads\""));
         std::env::remove_var(crate::env::ENV_NBD_AIO_MODE);
     }
 
+    #[test]
+    fn build_qemu_args_rejects_invalid_ceph_identifiers() {
+        let vm = VirtualMachine::new("vm1", "org", "proj", VmSpec::default());
+        let disks = vec![AttachedDisk {
+            id: "root".into(),
+            attachment: DiskAttachment::CephRbd {
+                pool: "pool,inject".into(),
+                image: "image".into(),
+                monitors: vec!["10.0.0.10:6789".into()],
+                user: "admin".into(),
+                secret: None,
+            },
+            bus: DiskBus::Virtio,
+            cache: DiskCache::None,
+            boot_index: Some(1),
+            read_only: false,
+        }];
+        let qmp = PathBuf::from("/tmp/qmp.sock");
+        let console = PathBuf::from("/tmp/console.log");
+        let error = build_qemu_args(&vm, &disks, &qmp, &console, None, None).unwrap_err();
+        assert!(error.to_string().contains("unsupported characters"));
+    }
+
     #[tokio::test]
     async fn wait_for_qmp_succeeds_after_socket_created() {
         let dir = tempfile::tempdir().unwrap();

plasmavmc/crates/plasmavmc-kvm/src/qmp.rs

@@ -33,7 +33,9 @@ impl QmpClient {
                     last_error = Some(format!("Failed to connect QMP: {e}"));
                 }
                 Err(e) => {
-                    return Err(Error::HypervisorError(format!("Failed to connect QMP: {e}")));
+                    return Err(Error::HypervisorError(format!(
+                        "Failed to connect QMP: {e}"
+                    )));
                 }
             }
 

plasmavmc/crates/plasmavmc-server/Cargo.toml

@@ -33,6 +33,7 @@ serde_json = { workspace = true }
 toml = { workspace = true }
 metrics-exporter-prometheus = { workspace = true }
 reqwest = { version = "0.12", default-features = false, features = ["rustls-tls"] }
+sha2 = "0.10"
 chainfire-client = { path = "../../../chainfire/chainfire-client" }
 creditservice-client = { path = "../../../creditservice/creditservice-client" }
 flaredb-client = { path = "../../../flaredb/crates/flaredb-client" }

plasmavmc/crates/plasmavmc-server/src/artifact_store.rs

@@ -1,3 +1,10 @@
+use crate::config::ArtifactStoreConfig;
+use crate::storage::ImageUploadPart;
+use reqwest::header::LOCATION;
+use reqwest::{Client as HttpClient, Url};
+use sha2::{Digest, Sha256};
+use std::collections::HashSet;
+use std::net::IpAddr;
 use std::path::{Path, PathBuf};
 use std::sync::Arc;
 use std::time::{Duration, Instant};
@@ -13,7 +20,6 @@ use lightningstor_api::proto::{
 };
 use lightningstor_api::{BucketServiceClient, ObjectServiceClient};
 use plasmavmc_types::ImageFormat;
-use reqwest::StatusCode as HttpStatusCode;
 use serde::Deserialize;
 use tokio::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::process::Command;
@@ -23,25 +29,30 @@ use tonic::metadata::MetadataValue;
 use tonic::transport::{Channel, Endpoint};
 use tonic::{Code, Request, Status};
 
-const DEFAULT_IMAGE_BUCKET: &str = "plasmavmc-images";
 const MAX_OBJECT_GRPC_MESSAGE_SIZE: usize = 1024 * 1024 * 1024;
 const OBJECT_GRPC_INITIAL_STREAM_WINDOW: u32 = 64 * 1024 * 1024;
 const OBJECT_GRPC_INITIAL_CONNECTION_WINDOW: u32 = 512 * 1024 * 1024;
 const OBJECT_GRPC_KEEPALIVE_INTERVAL: Duration = Duration::from_secs(30);
 const OBJECT_GRPC_KEEPALIVE_TIMEOUT: Duration = Duration::from_secs(10);
-const DEFAULT_MULTIPART_UPLOAD_PART_SIZE: usize = 32 * 1024 * 1024;
 const MIN_MULTIPART_UPLOAD_PART_SIZE: usize = 8 * 1024 * 1024;
 const MAX_MULTIPART_UPLOAD_PART_SIZE: usize = 128 * 1024 * 1024;
-const DEFAULT_MULTIPART_UPLOAD_CONCURRENCY: usize = 4;
 const MAX_MULTIPART_UPLOAD_CONCURRENCY: usize = 32;
-const DEFAULT_RAW_IMAGE_CONVERT_PARALLELISM: usize = 8;
+const MAX_IMPORT_REDIRECTS: usize = 5;
+const DEFAULT_HTTP_SEND_TIMEOUT: Duration = Duration::from_secs(15);
 
 #[derive(Clone)]
 pub struct ArtifactStore {
     channel: Channel,
     iam_client: Arc<IamClient>,
+    http_client: HttpClient,
     image_bucket: String,
     image_cache_dir: PathBuf,
+    multipart_upload_concurrency: usize,
+    multipart_upload_part_size: usize,
+    raw_image_convert_parallelism: usize,
+    max_image_import_size_bytes: u64,
+    allowed_https_hosts: Arc<HashSet<String>>,
+    qemu_img_path: PathBuf,
     project_tokens: Arc<DashMap<String, CachedToken>>,
     ensured_buckets: Arc<DashSet<String>>,
 }
@@ -50,6 +61,8 @@ pub(crate) struct ImportedImage {
     pub size_bytes: u64,
     pub checksum: String,
     pub format: ImageFormat,
+    pub source_type: String,
+    pub source_host: Option<String>,
 }
 
 #[derive(Deserialize)]
@@ -62,10 +75,24 @@ struct CachedToken {
     expires_at: Instant,
 }
 
+struct ValidatedImportUrl {
+    url: Url,
+    host: String,
+}
+
+struct ImportedImageSource {
+    source_type: String,
+    host: String,
+}
+
 impl ArtifactStore {
-    pub async fn from_env(iam_endpoint: &str) -> Result<Option<Self>, Box<dyn std::error::Error>> {
+    pub async fn from_config(
-        let Some(raw_endpoint) = std::env::var("PLASMAVMC_LIGHTNINGSTOR_ENDPOINT")
+        config: &ArtifactStoreConfig,
-            .ok()
+        iam_endpoint: &str,
+    ) -> Result<Option<Self>, Box<dyn std::error::Error>> {
+        let Some(raw_endpoint) = config
+            .lightningstor_endpoint
+            .as_ref()
             .map(|value| value.trim().to_string())
             .filter(|value| !value.is_empty())
         else {
@@ -87,18 +114,41 @@ impl ArtifactStore {
             .keep_alive_timeout(OBJECT_GRPC_KEEPALIVE_TIMEOUT)
             .connect_lazy();
 
-        let image_cache_dir = std::env::var("PLASMAVMC_IMAGE_CACHE_DIR")
+        let image_cache_dir = config.image_cache_dir.clone();
-            .map(PathBuf::from)
-            .unwrap_or_else(|_| PathBuf::from("/var/lib/plasmavmc/images"));
         tokio::fs::create_dir_all(&image_cache_dir).await?;
         ensure_cache_dir_permissions(&image_cache_dir).await?;
+        let http_client = HttpClient::builder()
+            .connect_timeout(Duration::from_secs(
+                config.image_import_connect_timeout_secs.max(1),
+            ))
+            .timeout(Duration::from_secs(config.image_import_timeout_secs.max(1)))
+            .redirect(reqwest::redirect::Policy::none())
+            .build()?;
+        let allowed_https_hosts = config
+            .allowed_https_hosts
+            .iter()
+            .map(|host| host.trim().to_ascii_lowercase())
+            .filter(|host| !host.is_empty())
+            .collect::<HashSet<_>>();
+        let qemu_img_path = resolve_binary_path(config.qemu_img_path.as_deref(), "qemu-img")?;
 
         Ok(Some(Self {
             channel,
             iam_client,
-            image_bucket: std::env::var("PLASMAVMC_IMAGE_BUCKET")
+            http_client,
-                .unwrap_or_else(|_| DEFAULT_IMAGE_BUCKET.to_string()),
+            image_bucket: config.image_bucket.clone(),
             image_cache_dir,
+            multipart_upload_concurrency: config
+                .multipart_upload_concurrency
+                .clamp(1, MAX_MULTIPART_UPLOAD_CONCURRENCY),
+            multipart_upload_part_size: config.multipart_upload_part_size.clamp(
+                MIN_MULTIPART_UPLOAD_PART_SIZE,
+                MAX_MULTIPART_UPLOAD_PART_SIZE,
+            ),
+            raw_image_convert_parallelism: config.raw_image_convert_parallelism.clamp(1, 64),
+            max_image_import_size_bytes: config.max_image_import_size_bytes.max(1),
+            allowed_https_hosts: Arc::new(allowed_https_hosts),
+            qemu_img_path,
             project_tokens: Arc::new(DashMap::new()),
             ensured_buckets: Arc::new(DashSet::new()),
         }))
@@ -116,50 +166,20 @@ impl ArtifactStore {
         self.ensure_bucket(&self.image_bucket, org_id, project_id, &token)
             .await?;
 
-        let image_path = self.image_path(image_id);
+        let staging_path = self.staging_path(image_id)?;
-        let staging_path = self.image_cache_dir.join(format!("{image_id}.source"));
+        let ImportedImageSource { source_type, host } =
-
             self.materialize_source(source_url, &staging_path).await?;
-        if self
+        self.process_staged_image(
-            .can_reuse_qcow2_source(&staging_path, source_format)
+            org_id,
-            .await?
+            project_id,
-        {
+            image_id,
-            if tokio::fs::try_exists(&image_path).await.map_err(|e| {
+            &token,
-                Status::internal(format!("failed to inspect {}: {e}", image_path.display()))
+            &staging_path,
-            })? {
+            source_format,
-                let _ = tokio::fs::remove_file(&staging_path).await;
+            source_type,
-            } else {
+            Some(host),
-                tokio::fs::rename(&staging_path, &image_path)
+        )
         .await
-                    .map_err(|e| {
-                        Status::internal(format!(
-                            "failed to move qcow2 image {} into cache {}: {e}",
-                            staging_path.display(),
-                            image_path.display()
-                        ))
-                    })?;
-                ensure_cache_file_permissions(&image_path).await?;
-            }
-        } else {
-            // Normalize non-qcow2 inputs through qemu-img convert so the cached
-            // artifact has a stable qcow2 representation before upload.
-            self.convert_to_qcow2(&staging_path, &image_path).await?;
-            let _ = tokio::fs::remove_file(&staging_path).await;
-        }
-
-        let checksum = self.sha256sum(&image_path).await?;
-        let metadata = tokio::fs::metadata(&image_path).await.map_err(|e| {
-            Status::internal(format!("failed to stat {}: {e}", image_path.display()))
-        })?;
-        let image_key = image_object_key(org_id, project_id, image_id);
-        self.upload_file(&self.image_bucket, &image_key, &image_path, &token)
-            .await?;
-
-        Ok(ImportedImage {
-            size_bytes: metadata.len(),
-            checksum,
-            format: ImageFormat::Qcow2,
-        })
     }
 
     pub async fn materialize_image_cache(
@@ -171,8 +191,8 @@ impl ArtifactStore {
         let token = self.issue_project_token(org_id, project_id).await?;
         self.ensure_bucket(&self.image_bucket, org_id, project_id, &token)
             .await?;
-        let image_key = image_object_key(org_id, project_id, image_id);
+        let image_key = image_object_key(org_id, project_id, image_id)?;
-        let image_path = self.image_path(image_id);
+        let image_path = self.image_path(image_id)?;
         self.download_object_to_file(&self.image_bucket, &image_key, &image_path, &token)
             .await?;
         Ok(image_path)
@@ -187,7 +207,7 @@ impl ArtifactStore {
         let image_path = self
             .materialize_image_cache(org_id, project_id, image_id)
             .await?;
-        let raw_path = self.raw_image_path(image_id);
+        let raw_path = self.raw_image_path(image_id)?;
         self.convert_to_raw(&image_path, &raw_path).await?;
         Ok(raw_path)
     }
@@ -199,7 +219,7 @@ impl ArtifactStore {
         image_id: &str,
     ) -> Result<(), Status> {
         let token = self.issue_project_token(org_id, project_id).await?;
-        let image_key = image_object_key(org_id, project_id, image_id);
+        let image_key = image_object_key(org_id, project_id, image_id)?;
         let mut client = self.object_client().await?;
         let mut request = Request::new(DeleteObjectRequest {
             bucket: self.image_bucket.clone(),
@@ -213,7 +233,7 @@ impl ArtifactStore {
             Err(status) => return Err(Status::from_error(Box::new(status))),
         }
 
-        let image_path = self.image_path(image_id);
+        let image_path = self.image_path(image_id)?;
         if tokio::fs::try_exists(&image_path).await.map_err(|e| {
             Status::internal(format!("failed to inspect {}: {e}", image_path.display()))
         })? {
@@ -221,7 +241,7 @@ impl ArtifactStore {
                 Status::internal(format!("failed to remove {}: {e}", image_path.display()))
             })?;
         }
-        let raw_path = self.raw_image_path(image_id);
+        let raw_path = self.raw_image_path(image_id)?;
         if tokio::fs::try_exists(&raw_path).await.map_err(|e| {
             Status::internal(format!("failed to inspect {}: {e}", raw_path.display()))
         })? {
@@ -232,6 +252,156 @@ impl ArtifactStore {
         Ok(())
     }
 
+    pub fn minimum_upload_part_size(&self) -> u32 {
+        self.multipart_upload_part_size as u32
+    }
+
+    pub(crate) async fn begin_image_upload(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+    ) -> Result<(String, String), Status> {
+        let token = self.issue_project_token(org_id, project_id).await?;
+        self.ensure_bucket(&self.image_bucket, org_id, project_id, &token)
+            .await?;
+        let staging_key = staging_object_key(org_id, project_id, image_id)?;
+        let mut client = self.object_client().await?;
+        let mut request = Request::new(CreateMultipartUploadRequest {
+            bucket: self.image_bucket.clone(),
+            key: staging_key.clone(),
+            metadata: None,
+        });
+        attach_bearer(&mut request, &token)?;
+        let upload_id = client
+            .create_multipart_upload(request)
+            .await
+            .map_err(|status| Status::from_error(Box::new(status)))?
+            .into_inner()
+            .upload_id;
+        Ok((upload_id, staging_key))
+    }
+
+    pub(crate) async fn upload_image_part(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        staging_key: &str,
+        upload_id: &str,
+        part_number: u32,
+        body: Vec<u8>,
+    ) -> Result<ImageUploadPart, Status> {
+        if part_number == 0 {
+            return Err(Status::invalid_argument(
+                "part_number must be greater than zero",
+            ));
+        }
+        if body.is_empty() {
+            return Err(Status::invalid_argument(
+                "upload part body must not be empty",
+            ));
+        }
+
+        let token = self.issue_project_token(org_id, project_id).await?;
+        let mut client = self.object_client().await?;
+        let request_stream = tokio_stream::iter(vec![UploadPartRequest {
+            bucket: self.image_bucket.clone(),
+            key: staging_key.to_string(),
+            upload_id: upload_id.to_string(),
+            part_number,
+            body: body.clone().into(),
+            content_md5: String::new(),
+        }]);
+        let mut request = Request::new(request_stream);
+        attach_bearer(&mut request, &token)?;
+        let response = client
+            .upload_part(request)
+            .await
+            .map_err(|status| Status::from_error(Box::new(status)))?;
+        Ok(ImageUploadPart {
+            part_number,
+            etag: response.into_inner().etag,
+            size_bytes: body.len() as u64,
+        })
+    }
+
+    pub(crate) async fn complete_image_upload(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        staging_key: &str,
+        upload_id: &str,
+        parts: &[ImageUploadPart],
+        source_format: ImageFormat,
+    ) -> Result<ImportedImage, Status> {
+        if parts.is_empty() {
+            return Err(Status::failed_precondition(
+                "upload session does not contain any parts",
+            ));
+        }
+        let token = self.issue_project_token(org_id, project_id).await?;
+        self.ensure_bucket(&self.image_bucket, org_id, project_id, &token)
+            .await?;
+
+        let mut sorted_parts: Vec<CompletedPart> = parts
+            .iter()
+            .map(|part| CompletedPart {
+                part_number: part.part_number,
+                etag: part.etag.clone(),
+            })
+            .collect();
+        sorted_parts.sort_by_key(|part| part.part_number);
+
+        let mut client = self.object_client().await?;
+        let mut complete_request = Request::new(CompleteMultipartUploadRequest {
+            bucket: self.image_bucket.clone(),
+            key: staging_key.to_string(),
+            upload_id: upload_id.to_string(),
+            parts: sorted_parts,
+        });
+        attach_bearer(&mut complete_request, &token)?;
+        client
+            .complete_multipart_upload(complete_request)
+            .await
+            .map_err(|status| Status::from_error(Box::new(status)))?;
+
+        let staging_path = self.staging_path(image_id)?;
+        if tokio::fs::try_exists(&staging_path).await.unwrap_or(false) {
+            let _ = tokio::fs::remove_file(&staging_path).await;
+        }
+        self.download_object_to_file(&self.image_bucket, staging_key, &staging_path, &token)
+            .await?;
+        let result = self
+            .process_staged_image(
+                org_id,
+                project_id,
+                image_id,
+                &token,
+                &staging_path,
+                source_format,
+                "upload".to_string(),
+                None,
+            )
+            .await;
+        let _ = self
+            .delete_object_ignore_not_found(&self.image_bucket, staging_key, &token)
+            .await;
+        result
+    }
+
+    pub(crate) async fn abort_image_upload(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        staging_key: &str,
+        upload_id: &str,
+    ) -> Result<(), Status> {
+        let token = self.issue_project_token(org_id, project_id).await?;
+        self.abort_multipart_upload(&self.image_bucket, staging_key, upload_id, &token)
+            .await
+    }
+
     async fn ensure_bucket(
         &self,
         bucket: &str,
@@ -275,7 +445,7 @@ impl ArtifactStore {
         let metadata = tokio::fs::metadata(path)
             .await
             .map_err(|e| Status::internal(format!("failed to stat {path:?}: {e}")))?;
-        let multipart_part_size = multipart_upload_part_size();
+        let multipart_part_size = self.multipart_upload_part_size;
         if metadata.len() > multipart_part_size as u64 {
             return self
                 .upload_file_multipart(bucket, key, path, token, metadata.len())
@@ -336,7 +506,7 @@ impl ArtifactStore {
         size_bytes: u64,
     ) -> Result<(), Status> {
         let started = Instant::now();
-        let multipart_part_size = multipart_upload_part_size();
+        let multipart_part_size = self.multipart_upload_part_size;
         tracing::info!(
             bucket = bucket,
             key = key,
@@ -366,7 +536,7 @@ impl ArtifactStore {
         let mut part_number = 1u32;
         let mut completed_parts = Vec::new();
         let mut uploads = JoinSet::new();
-        let upload_concurrency = multipart_upload_concurrency();
+        let upload_concurrency = self.multipart_upload_concurrency;
 
         let enqueue_part_upload = |uploads: &mut JoinSet<Result<CompletedPart, Status>>,
                                    client: &ObjectServiceClient<Channel>,
@@ -569,76 +739,23 @@ impl ArtifactStore {
         Ok(())
     }
 
-    async fn materialize_source(&self, source_url: &str, path: &Path) -> Result<(), Status> {
+    async fn materialize_source(
+        &self,
+        source_url: &str,
+        path: &Path,
+    ) -> Result<ImportedImageSource, Status> {
         if let Some(parent) = path.parent() {
             tokio::fs::create_dir_all(parent)
                 .await
                 .map_err(|e| Status::internal(format!("failed to create {parent:?}: {e}")))?;
         }
 
-        if let Some(source_path) = source_url.strip_prefix("file://") {
+        let validated = self.validate_import_url(source_url).await?;
-            tokio::fs::copy(source_path, path).await.map_err(|e| {
+        self.download_https_source(&validated, path).await?;
-                Status::internal(format!("failed to copy image source {source_path}: {e}"))
+        Ok(ImportedImageSource {
-            })?;
+            source_type: "https".to_string(),
-            ensure_cache_file_permissions(path).await?;
+            host: validated.host,
-            return Ok(());
+        })
-        }
-
-        if source_url.starts_with('/') {
-            tokio::fs::copy(source_url, path).await.map_err(|e| {
-                Status::internal(format!("failed to copy image source {source_url}: {e}"))
-            })?;
-            ensure_cache_file_permissions(path).await?;
-            return Ok(());
-        }
-
-        if source_url.starts_with("http://") || source_url.starts_with("https://") {
-            let mut response = reqwest::get(source_url).await.map_err(|e| {
-                Status::unavailable(format!("failed to download image source: {e}"))
-            })?;
-            if response.status() != HttpStatusCode::OK {
-                return Err(Status::failed_precondition(format!(
-                    "image download failed with HTTP {}",
-                    response.status()
-                )));
-            }
-            let temp_path = path.with_extension("download");
-            let mut file = tokio::fs::File::create(&temp_path).await.map_err(|e| {
-                Status::internal(format!(
-                    "failed to create downloaded image {}: {e}",
-                    temp_path.display()
-                ))
-            })?;
-            while let Some(chunk) = response.chunk().await.map_err(|e| {
-                Status::unavailable(format!("failed to read image response body: {e}"))
-            })? {
-                file.write_all(&chunk).await.map_err(|e| {
-                    Status::internal(format!(
-                        "failed to write downloaded image {}: {e}",
-                        temp_path.display()
-                    ))
-                })?;
-            }
-            file.flush().await.map_err(|e| {
-                Status::internal(format!(
-                    "failed to flush downloaded image {}: {e}",
-                    temp_path.display()
-                ))
-            })?;
-            drop(file);
-            tokio::fs::rename(&temp_path, path).await.map_err(|e| {
-                Status::internal(format!(
-                    "failed to finalize downloaded image {}: {e}",
-                    path.display()
-                ))
-            })?;
-            ensure_cache_file_permissions(path).await?;
-            return Ok(());
-        }
-
-        Err(Status::invalid_argument(
-            "source_url must be file://, an absolute path, or http(s)://",
-        ))
     }
 
     async fn convert_to_qcow2(&self, source: &Path, destination: &Path) -> Result<(), Status> {
@@ -654,9 +771,9 @@ impl ArtifactStore {
                 .map_err(|e| Status::internal(format!("failed to create {parent:?}: {e}")))?;
         }
 
-        let parallelism = raw_image_convert_parallelism().to_string();
+        let parallelism = self.raw_image_convert_parallelism.to_string();
         let args = qemu_img_convert_to_qcow2_args(source, destination, &parallelism);
-        let status = Command::new("qemu-img")
+        let status = Command::new(&self.qemu_img_path)
             .args(args.iter().map(String::as_str))
             .status()
             .await
@@ -685,9 +802,9 @@ impl ArtifactStore {
                 .map_err(|e| Status::internal(format!("failed to create {parent:?}: {e}")))?;
         }
 
-        let parallelism = raw_image_convert_parallelism().to_string();
+        let parallelism = self.raw_image_convert_parallelism.to_string();
         let args = qemu_img_convert_to_raw_args(source, destination, &parallelism);
-        let status = Command::new("qemu-img")
+        let status = Command::new(&self.qemu_img_path)
             .args(args.iter().map(String::as_str))
             .status()
             .await
@@ -712,7 +829,7 @@ impl ArtifactStore {
             return Ok(false);
         }
 
-        let output = Command::new("qemu-img")
+        let output = Command::new(&self.qemu_img_path)
             .args(["info", "--output", "json", path.to_string_lossy().as_ref()])
             .output()
             .await
@@ -727,25 +844,262 @@ impl ArtifactStore {
     }
 
     async fn sha256sum(&self, path: &Path) -> Result<String, Status> {
-        let output = Command::new("sha256sum")
+        let mut file = tokio::fs::File::open(path)
-            .arg(path)
-            .output()
             .await
-            .map_err(|e| Status::internal(format!("failed to spawn sha256sum: {e}")))?;
+            .map_err(|e| Status::internal(format!("failed to open {}: {e}", path.display())))?;
-        if !output.status.success() {
+        let mut digest = Sha256::new();
-            return Err(Status::internal(format!(
+        let mut buffer = vec![0u8; 1024 * 1024];
-                "sha256sum failed for {} with status {}",
+        loop {
-                path.display(),
+            let read_now = file
-                output.status
+                .read(&mut buffer)
+                .await
+                .map_err(|e| Status::internal(format!("failed to read {}: {e}", path.display())))?;
+            if read_now == 0 {
+                break;
+            }
+            digest.update(&buffer[..read_now]);
+        }
+        Ok(format!("{:x}", digest.finalize()))
+    }
+
+    async fn process_staged_image(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        token: &str,
+        staging_path: &Path,
+        source_format: ImageFormat,
+        source_type: String,
+        source_host: Option<String>,
+    ) -> Result<ImportedImage, Status> {
+        let image_path = self.image_path(image_id)?;
+        if self
+            .can_reuse_qcow2_source(staging_path, source_format)
+            .await?
+        {
+            if tokio::fs::try_exists(&image_path).await.map_err(|e| {
+                Status::internal(format!("failed to inspect {}: {e}", image_path.display()))
+            })? {
+                let _ = tokio::fs::remove_file(staging_path).await;
+            } else {
+                tokio::fs::rename(staging_path, &image_path)
+                    .await
+                    .map_err(|e| {
+                        Status::internal(format!(
+                            "failed to move qcow2 image {} into cache {}: {e}",
+                            staging_path.display(),
+                            image_path.display()
+                        ))
+                    })?;
+                ensure_cache_file_permissions(&image_path).await?;
+            }
+        } else {
+            self.convert_to_qcow2(staging_path, &image_path).await?;
+            let _ = tokio::fs::remove_file(staging_path).await;
+        }
+
+        let checksum = self.sha256sum(&image_path).await?;
+        let metadata = tokio::fs::metadata(&image_path).await.map_err(|e| {
+            Status::internal(format!("failed to stat {}: {e}", image_path.display()))
+        })?;
+        let image_key = image_object_key(org_id, project_id, image_id)?;
+        self.upload_file(&self.image_bucket, &image_key, &image_path, token)
+            .await?;
+
+        Ok(ImportedImage {
+            size_bytes: metadata.len(),
+            checksum,
+            format: ImageFormat::Qcow2,
+            source_type,
+            source_host,
+        })
+    }
+
+    async fn validate_import_url(&self, source_url: &str) -> Result<ValidatedImportUrl, Status> {
+        if source_url.starts_with("file://") || source_url.starts_with('/') {
+            return Err(Status::invalid_argument(
+                "source_url must use https:// and may not reference local files",
+            ));
+        }
+
+        let url = Url::parse(source_url)
+            .map_err(|e| Status::invalid_argument(format!("invalid source_url: {e}")))?;
+        if url.scheme() != "https" {
+            return Err(Status::invalid_argument("source_url must use https://"));
+        }
+        if !url.username().is_empty() || url.password().is_some() {
+            return Err(Status::invalid_argument(
+                "source_url must not include embedded credentials",
+            ));
+        }
+        let host = url
+            .host_str()
+            .map(str::to_ascii_lowercase)
+            .ok_or_else(|| Status::invalid_argument("source_url must include a host"))?;
+        if host == "localhost" {
+            return Err(Status::invalid_argument(
+                "source_url host must not target loopback or local interfaces",
+            ));
+        }
+        if !self.allowed_https_hosts.is_empty() && !self.allowed_https_hosts.contains(&host) {
+            return Err(Status::permission_denied(format!(
+                "source_url host {host} is not in artifacts.allowed_https_hosts",
             )));
         }
-        let stdout = String::from_utf8(output.stdout)
+
-            .map_err(|e| Status::internal(format!("invalid sha256sum output: {e}")))?;
+        let port = url.port_or_known_default().unwrap_or(443);
-        stdout
+        let resolved = tokio::net::lookup_host((host.as_str(), port))
-            .split_whitespace()
+            .await
-            .next()
+            .map_err(|e| Status::unavailable(format!("failed to resolve source_url host: {e}")))?;
-            .map(str::to_string)
+        let mut found_any = false;
-            .ok_or_else(|| Status::internal("sha256sum output missing digest"))
+        for addr in resolved {
+            found_any = true;
+            let ip = addr.ip();
+            if !is_public_ip(ip) {
+                return Err(Status::permission_denied(format!(
+                    "source_url resolved to a non-public address: {ip}",
+                )));
+            }
+        }
+        if !found_any {
+            return Err(Status::failed_precondition(
+                "source_url host did not resolve to any public addresses",
+            ));
+        }
+
+        Ok(ValidatedImportUrl { url, host })
+    }
+
+    async fn download_https_source(
+        &self,
+        source: &ValidatedImportUrl,
+        path: &Path,
+    ) -> Result<(), Status> {
+        let temp_path = path.with_extension("download");
+        if tokio::fs::try_exists(&temp_path).await.unwrap_or(false) {
+            let _ = tokio::fs::remove_file(&temp_path).await;
+        }
+
+        let mut current = source.url.clone();
+        let mut redirects_remaining = MAX_IMPORT_REDIRECTS;
+        loop {
+            let response = tokio::time::timeout(
+                DEFAULT_HTTP_SEND_TIMEOUT,
+                self.http_client.get(current.clone()).send(),
+            )
+            .await
+            .map_err(|_| Status::deadline_exceeded("timed out waiting for source_url response"))?
+            .map_err(|e| Status::unavailable(format!("failed to download image source: {e}")))?;
+
+            if response.status().is_redirection() {
+                if redirects_remaining == 0 {
+                    return Err(Status::failed_precondition(
+                        "source_url redirect limit exceeded",
+                    ));
+                }
+                let location = response
+                    .headers()
+                    .get(LOCATION)
+                    .ok_or_else(|| {
+                        Status::failed_precondition(
+                            "source_url redirect response did not include a Location header",
+                        )
+                    })?
+                    .to_str()
+                    .map_err(|e| {
+                        Status::failed_precondition(format!(
+                            "invalid redirect Location header in source_url response: {e}"
+                        ))
+                    })?;
+                current = current.join(location).map_err(|e| {
+                    Status::failed_precondition(format!(
+                        "failed to resolve redirect target from source_url: {e}"
+                    ))
+                })?;
+                self.validate_import_url(current.as_ref()).await?;
+                redirects_remaining -= 1;
+                continue;
+            }
+
+            if !response.status().is_success() {
+                return Err(Status::failed_precondition(format!(
+                    "image download failed with HTTP {}",
+                    response.status()
+                )));
+            }
+
+            if let Some(content_length) = response.content_length() {
+                if content_length > self.max_image_import_size_bytes {
+                    return Err(Status::resource_exhausted(format!(
+                        "image download exceeds the configured maximum size of {} bytes",
+                        self.max_image_import_size_bytes
+                    )));
+                }
+            }
+
+            let mut file = tokio::fs::File::create(&temp_path).await.map_err(|e| {
+                Status::internal(format!(
+                    "failed to create downloaded image {}: {e}",
+                    temp_path.display()
+                ))
+            })?;
+            let mut response = response;
+            let mut downloaded = 0u64;
+            while let Some(chunk) = response.chunk().await.map_err(|e| {
+                Status::unavailable(format!("failed to read image response body: {e}"))
+            })? {
+                downloaded = downloaded.saturating_add(chunk.len() as u64);
+                if downloaded > self.max_image_import_size_bytes {
+                    let _ = tokio::fs::remove_file(&temp_path).await;
+                    return Err(Status::resource_exhausted(format!(
+                        "image download exceeds the configured maximum size of {} bytes",
+                        self.max_image_import_size_bytes
+                    )));
+                }
+                file.write_all(&chunk).await.map_err(|e| {
+                    Status::internal(format!(
+                        "failed to write downloaded image {}: {e}",
+                        temp_path.display()
+                    ))
+                })?;
+            }
+            file.flush().await.map_err(|e| {
+                Status::internal(format!(
+                    "failed to flush downloaded image {}: {e}",
+                    temp_path.display()
+                ))
+            })?;
+            drop(file);
+            tokio::fs::rename(&temp_path, path).await.map_err(|e| {
+                Status::internal(format!(
+                    "failed to finalize downloaded image {}: {e}",
+                    path.display()
+                ))
+            })?;
+            ensure_cache_file_permissions(path).await?;
+            return Ok(());
+        }
+    }
+
+    async fn delete_object_ignore_not_found(
+        &self,
+        bucket: &str,
+        key: &str,
+        token: &str,
+    ) -> Result<(), Status> {
+        let mut client = self.object_client().await?;
+        let mut request = Request::new(DeleteObjectRequest {
+            bucket: bucket.to_string(),
+            key: key.to_string(),
+            version_id: String::new(),
+        });
+        attach_bearer(&mut request, token)?;
+        match client.delete_object(request).await {
+            Ok(_) => Ok(()),
+            Err(status) if status.code() == Code::NotFound => Ok(()),
+            Err(status) => Err(Status::from_error(Box::new(status))),
+        }
     }
 
     async fn bucket_client(&self) -> Result<BucketServiceClient<Channel>, Status> {
@@ -832,12 +1186,22 @@ impl ArtifactStore {
         Ok(token)
     }
 
-    fn image_path(&self, image_id: &str) -> PathBuf {
+    fn image_path(&self, image_id: &str) -> Result<PathBuf, Status> {
-        self.image_cache_dir.join(format!("{image_id}.qcow2"))
+        Ok(self
+            .image_cache_dir
+            .join(format!("{}.qcow2", validated_image_id(image_id)?)))
     }
 
-    fn raw_image_path(&self, image_id: &str) -> PathBuf {
+    fn raw_image_path(&self, image_id: &str) -> Result<PathBuf, Status> {
-        self.image_cache_dir.join(format!("{image_id}.raw"))
+        Ok(self
+            .image_cache_dir
+            .join(format!("{}.raw", validated_image_id(image_id)?)))
+    }
+
+    fn staging_path(&self, image_id: &str) -> Result<PathBuf, Status> {
+        Ok(self
+            .image_cache_dir
+            .join(format!("{}.source", validated_image_id(image_id)?)))
     }
 
     async fn abort_multipart_upload(
@@ -915,35 +1279,6 @@ async fn next_uploaded_part(
     }
 }
 
-fn multipart_upload_concurrency() -> usize {
-    std::env::var("PLASMAVMC_LIGHTNINGSTOR_MULTIPART_CONCURRENCY")
-        .ok()
-        .and_then(|value| value.parse::<usize>().ok())
-        .map(|value| value.clamp(1, MAX_MULTIPART_UPLOAD_CONCURRENCY))
-        .unwrap_or(DEFAULT_MULTIPART_UPLOAD_CONCURRENCY)
-}
-
-fn multipart_upload_part_size() -> usize {
-    std::env::var("PLASMAVMC_LIGHTNINGSTOR_MULTIPART_PART_SIZE")
-        .ok()
-        .and_then(|value| value.parse::<usize>().ok())
-        .map(|value| {
-            value.clamp(
-                MIN_MULTIPART_UPLOAD_PART_SIZE,
-                MAX_MULTIPART_UPLOAD_PART_SIZE,
-            )
-        })
-        .unwrap_or(DEFAULT_MULTIPART_UPLOAD_PART_SIZE)
-}
-
-fn raw_image_convert_parallelism() -> usize {
-    std::env::var("PLASMAVMC_RAW_IMAGE_CONVERT_PARALLELISM")
-        .ok()
-        .and_then(|value| value.parse::<usize>().ok())
-        .map(|value| value.clamp(1, 64))
-        .unwrap_or(DEFAULT_RAW_IMAGE_CONVERT_PARALLELISM)
-}
-
 fn qemu_img_convert_to_qcow2_args(
     source: &Path,
     destination: &Path,
@@ -1008,8 +1343,74 @@ fn sanitize_identifier(value: &str) -> String {
         .collect()
 }
 
-fn image_object_key(org_id: &str, project_id: &str, image_id: &str) -> String {
+fn image_object_key(org_id: &str, project_id: &str, image_id: &str) -> Result<String, Status> {
-    format!("{org_id}/{project_id}/{image_id}.qcow2")
+    Ok(format!(
+        "{org_id}/{project_id}/{}.qcow2",
+        validated_image_id(image_id)?
+    ))
+}
+
+fn staging_object_key(org_id: &str, project_id: &str, image_id: &str) -> Result<String, Status> {
+    Ok(format!(
+        "{org_id}/{project_id}/uploads/{}.source",
+        validated_image_id(image_id)?
+    ))
+}
+
+fn validated_image_id(image_id: &str) -> Result<&str, Status> {
+    uuid::Uuid::parse_str(image_id)
+        .map(|_| image_id)
+        .map_err(|_| Status::invalid_argument("image_id must be a UUID"))
+}
+
+fn is_public_ip(ip: IpAddr) -> bool {
+    match ip {
+        IpAddr::V4(ip) => {
+            !(ip.is_private()
+                || ip.is_loopback()
+                || ip.is_link_local()
+                || ip.is_multicast()
+                || ip.is_broadcast()
+                || ip.is_documentation()
+                || ip.is_unspecified())
+        }
+        IpAddr::V6(ip) => {
+            !(ip.is_loopback()
+                || ip.is_unspecified()
+                || ip.is_multicast()
+                || ip.is_unique_local()
+                || ip.is_unicast_link_local())
+        }
+    }
+}
+
+fn resolve_binary_path(
+    configured_path: Option<&Path>,
+    binary_name: &str,
+) -> Result<PathBuf, Box<dyn std::error::Error>> {
+    let candidate = match configured_path {
+        Some(path) => path.to_path_buf(),
+        None => std::env::var_os("PATH")
+            .into_iter()
+            .flat_map(|paths| std::env::split_paths(&paths).collect::<Vec<_>>())
+            .map(|entry| entry.join(binary_name))
+            .find(|candidate| candidate.exists())
+            .ok_or_else(|| {
+                std::io::Error::new(
+                    std::io::ErrorKind::NotFound,
+                    format!("failed to locate {binary_name} in PATH"),
+                )
+            })?,
+    };
+    let metadata = std::fs::metadata(&candidate)?;
+    if !metadata.is_file() {
+        return Err(std::io::Error::new(
+            std::io::ErrorKind::InvalidInput,
+            format!("{} is not a regular file", candidate.display()),
+        )
+        .into());
+    }
+    Ok(candidate)
 }
 
 async fn ensure_cache_dir_permissions(path: &Path) -> Result<(), Status> {
@@ -1103,4 +1504,13 @@ mod tests {
             ]
         );
     }
+
+    #[test]
+    fn image_object_key_rejects_non_uuid_identifiers() {
+        assert!(image_object_key("org", "project", "../passwd").is_err());
+        assert_eq!(
+            image_object_key("org", "project", "11111111-1111-1111-1111-111111111111").unwrap(),
+            "org/project/11111111-1111-1111-1111-111111111111.qcow2"
+        );
+    }
 }

plasmavmc/crates/plasmavmc-server/src/config.rs

@@ -1,8 +1,9 @@
 //! Server configuration
 
+use plasmavmc_types::{FireCrackerConfig, KvmConfig};
 use serde::{Deserialize, Serialize};
 use std::net::SocketAddr;
-use plasmavmc_types::{FireCrackerConfig, KvmConfig};
+use std::path::PathBuf;
 
 /// TLS configuration
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -42,12 +43,332 @@ pub struct ServerConfig {
     /// Configuration for FireCracker backend
     #[serde(default)]
     pub firecracker: FireCrackerConfig,
+    /// Durable/shared state configuration
+    #[serde(default)]
+    pub storage: StorageRuntimeConfig,
+    /// Control-plane and agent heartbeat configuration
+    #[serde(default)]
+    pub agent: AgentRuntimeConfig,
+    /// External service integrations
+    #[serde(default)]
+    pub integrations: IntegrationConfig,
+    /// State watcher and VM watch polling configuration
+    #[serde(default)]
+    pub watcher: WatcherRuntimeConfig,
+    /// Background health monitoring and failover configuration
+    #[serde(default)]
+    pub health: HealthRuntimeConfig,
+    /// Artifact storage and image import/export configuration
+    #[serde(default)]
+    pub artifacts: ArtifactStoreConfig,
+    /// Persistent volume runtime configuration
+    #[serde(default)]
+    pub volumes: VolumeRuntimeConfig,
+    /// Default hypervisor used when requests do not specify one
+    #[serde(default)]
+    pub default_hypervisor: DefaultHypervisor,
 }
 
 fn default_http_addr() -> SocketAddr {
     "127.0.0.1:8084".parse().unwrap()
 }
 
+#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, Default)]
+#[serde(rename_all = "lowercase")]
+pub enum StorageBackendKind {
+    #[default]
+    Flaredb,
+    File,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct StorageRuntimeConfig {
+    #[serde(default)]
+    pub backend: StorageBackendKind,
+    #[serde(default)]
+    pub flaredb_endpoint: Option<String>,
+    #[serde(default)]
+    pub chainfire_endpoint: Option<String>,
+    #[serde(default)]
+    pub state_path: Option<PathBuf>,
+}
+
+impl Default for StorageRuntimeConfig {
+    fn default() -> Self {
+        Self {
+            backend: StorageBackendKind::default(),
+            flaredb_endpoint: None,
+            chainfire_endpoint: None,
+            state_path: None,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct AgentRuntimeConfig {
+    #[serde(default)]
+    pub control_plane_addr: Option<String>,
+    #[serde(default)]
+    pub advertise_endpoint: Option<String>,
+    #[serde(default)]
+    pub node_id: Option<String>,
+    #[serde(default)]
+    pub node_name: Option<String>,
+    #[serde(default = "default_agent_heartbeat_interval_secs")]
+    pub heartbeat_interval_secs: u64,
+    #[serde(default = "default_shared_live_migration")]
+    pub shared_live_migration: bool,
+}
+
+fn default_agent_heartbeat_interval_secs() -> u64 {
+    5
+}
+
+fn default_shared_live_migration() -> bool {
+    true
+}
+
+impl Default for AgentRuntimeConfig {
+    fn default() -> Self {
+        Self {
+            control_plane_addr: None,
+            advertise_endpoint: None,
+            node_id: None,
+            node_name: None,
+            heartbeat_interval_secs: default_agent_heartbeat_interval_secs(),
+            shared_live_migration: default_shared_live_migration(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct IntegrationConfig {
+    #[serde(default)]
+    pub prismnet_endpoint: Option<String>,
+    #[serde(default)]
+    pub creditservice_endpoint: Option<String>,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct WatcherRuntimeConfig {
+    #[serde(default)]
+    pub enabled: bool,
+    #[serde(default = "default_state_watcher_poll_interval_ms")]
+    pub poll_interval_ms: u64,
+    #[serde(default = "default_vm_watch_poll_interval_ms")]
+    pub vm_watch_poll_interval_ms: u64,
+}
+
+fn default_state_watcher_poll_interval_ms() -> u64 {
+    1000
+}
+
+fn default_vm_watch_poll_interval_ms() -> u64 {
+    500
+}
+
+impl Default for WatcherRuntimeConfig {
+    fn default() -> Self {
+        Self {
+            enabled: false,
+            poll_interval_ms: default_state_watcher_poll_interval_ms(),
+            vm_watch_poll_interval_ms: default_vm_watch_poll_interval_ms(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct HealthRuntimeConfig {
+    #[serde(default)]
+    pub vm_monitor_interval_secs: Option<u64>,
+    #[serde(default)]
+    pub node_monitor_interval_secs: Option<u64>,
+    #[serde(default = "default_node_heartbeat_timeout_secs")]
+    pub node_heartbeat_timeout_secs: u64,
+    #[serde(default)]
+    pub auto_restart: bool,
+    #[serde(default)]
+    pub failover_enabled: bool,
+    #[serde(default = "default_failover_min_interval_secs")]
+    pub failover_min_interval_secs: u64,
+}
+
+fn default_node_heartbeat_timeout_secs() -> u64 {
+    60
+}
+
+fn default_failover_min_interval_secs() -> u64 {
+    60
+}
+
+impl Default for HealthRuntimeConfig {
+    fn default() -> Self {
+        Self {
+            vm_monitor_interval_secs: None,
+            node_monitor_interval_secs: None,
+            node_heartbeat_timeout_secs: default_node_heartbeat_timeout_secs(),
+            auto_restart: false,
+            failover_enabled: false,
+            failover_min_interval_secs: default_failover_min_interval_secs(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ArtifactStoreConfig {
+    #[serde(default)]
+    pub lightningstor_endpoint: Option<String>,
+    #[serde(default = "default_image_cache_dir")]
+    pub image_cache_dir: PathBuf,
+    #[serde(default = "default_image_bucket")]
+    pub image_bucket: String,
+    #[serde(default = "default_multipart_upload_concurrency")]
+    pub multipart_upload_concurrency: usize,
+    #[serde(default = "default_multipart_upload_part_size")]
+    pub multipart_upload_part_size: usize,
+    #[serde(default = "default_raw_image_convert_parallelism")]
+    pub raw_image_convert_parallelism: usize,
+    #[serde(default = "default_image_import_connect_timeout_secs")]
+    pub image_import_connect_timeout_secs: u64,
+    #[serde(default = "default_image_import_timeout_secs")]
+    pub image_import_timeout_secs: u64,
+    #[serde(default = "default_max_image_import_size_bytes")]
+    pub max_image_import_size_bytes: u64,
+    #[serde(default)]
+    pub allowed_https_hosts: Vec<String>,
+    #[serde(default)]
+    pub qemu_img_path: Option<PathBuf>,
+}
+
+fn default_image_cache_dir() -> PathBuf {
+    PathBuf::from("/var/lib/plasmavmc/images")
+}
+
+fn default_image_bucket() -> String {
+    "plasmavmc-images".to_string()
+}
+
+fn default_multipart_upload_concurrency() -> usize {
+    4
+}
+
+fn default_multipart_upload_part_size() -> usize {
+    32 * 1024 * 1024
+}
+
+fn default_raw_image_convert_parallelism() -> usize {
+    8
+}
+
+fn default_image_import_connect_timeout_secs() -> u64 {
+    5
+}
+
+fn default_image_import_timeout_secs() -> u64 {
+    60 * 60
+}
+
+fn default_max_image_import_size_bytes() -> u64 {
+    64 * 1024 * 1024 * 1024
+}
+
+impl Default for ArtifactStoreConfig {
+    fn default() -> Self {
+        Self {
+            lightningstor_endpoint: None,
+            image_cache_dir: default_image_cache_dir(),
+            image_bucket: default_image_bucket(),
+            multipart_upload_concurrency: default_multipart_upload_concurrency(),
+            multipart_upload_part_size: default_multipart_upload_part_size(),
+            raw_image_convert_parallelism: default_raw_image_convert_parallelism(),
+            image_import_connect_timeout_secs: default_image_import_connect_timeout_secs(),
+            image_import_timeout_secs: default_image_import_timeout_secs(),
+            max_image_import_size_bytes: default_max_image_import_size_bytes(),
+            allowed_https_hosts: Vec::new(),
+            qemu_img_path: None,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct CoronaFsConfig {
+    #[serde(default)]
+    pub controller_endpoint: Option<String>,
+    #[serde(default)]
+    pub node_endpoint: Option<String>,
+    #[serde(default)]
+    pub endpoint: Option<String>,
+    #[serde(default)]
+    pub node_local_attach: bool,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct CephConfig {
+    #[serde(default)]
+    pub monitors: Vec<String>,
+    #[serde(default = "default_ceph_cluster_id")]
+    pub cluster_id: String,
+    #[serde(default = "default_ceph_user")]
+    pub user: String,
+    #[serde(default)]
+    pub secret: Option<String>,
+}
+
+fn default_ceph_cluster_id() -> String {
+    "default".to_string()
+}
+
+fn default_ceph_user() -> String {
+    "admin".to_string()
+}
+
+impl Default for CephConfig {
+    fn default() -> Self {
+        Self {
+            monitors: Vec::new(),
+            cluster_id: default_ceph_cluster_id(),
+            user: default_ceph_user(),
+            secret: None,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct VolumeRuntimeConfig {
+    #[serde(default = "default_managed_volume_root")]
+    pub managed_volume_root: PathBuf,
+    #[serde(default)]
+    pub ceph: CephConfig,
+    #[serde(default)]
+    pub coronafs: CoronaFsConfig,
+    #[serde(default)]
+    pub qemu_img_path: Option<PathBuf>,
+}
+
+fn default_managed_volume_root() -> PathBuf {
+    PathBuf::from("/var/lib/plasmavmc/managed-volumes")
+}
+
+impl Default for VolumeRuntimeConfig {
+    fn default() -> Self {
+        Self {
+            managed_volume_root: default_managed_volume_root(),
+            ceph: CephConfig::default(),
+            coronafs: CoronaFsConfig::default(),
+            qemu_img_path: None,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, Default)]
+#[serde(rename_all = "lowercase")]
+pub enum DefaultHypervisor {
+    #[default]
+    Kvm,
+    Firecracker,
+    Mvisor,
+}
+
 /// Authentication configuration
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct AuthConfig {
@@ -78,6 +399,14 @@ impl Default for ServerConfig {
             auth: AuthConfig::default(),
             kvm: KvmConfig::default(),
             firecracker: FireCrackerConfig::default(),
+            storage: StorageRuntimeConfig::default(),
+            agent: AgentRuntimeConfig::default(),
+            integrations: IntegrationConfig::default(),
+            watcher: WatcherRuntimeConfig::default(),
+            health: HealthRuntimeConfig::default(),
+            artifacts: ArtifactStoreConfig::default(),
+            volumes: VolumeRuntimeConfig::default(),
+            default_hypervisor: DefaultHypervisor::default(),
         }
     }
 }

plasmavmc/crates/plasmavmc-server/src/main.rs

@@ -15,7 +15,7 @@ use plasmavmc_api::proto::{
 use plasmavmc_firecracker::FireCrackerBackend;
 use plasmavmc_hypervisor::HypervisorRegistry;
 use plasmavmc_kvm::KvmBackend;
-use plasmavmc_server::config::ServerConfig;
+use plasmavmc_server::config::{AgentRuntimeConfig, ServerConfig};
 use plasmavmc_server::watcher::{StateSynchronizer, StateWatcher, WatcherConfig};
 use plasmavmc_server::VmServiceImpl;
 use std::net::SocketAddr;
@@ -83,16 +83,19 @@ async fn start_agent_heartbeat(
     supported_volume_drivers: Vec<i32>,
     supported_storage_classes: Vec<String>,
     shared_live_migration: bool,
+    agent_config: &AgentRuntimeConfig,
 ) {
-    let Some(control_plane_addr) = std::env::var("PLASMAVMC_CONTROL_PLANE_ADDR")
+    let Some(control_plane_addr) = agent_config
-        .ok()
+        .control_plane_addr
+        .as_ref()
         .map(|value| value.trim().to_string())
         .filter(|value| !value.is_empty())
     else {
         return;
     };
-    let Some(node_id) = std::env::var("PLASMAVMC_NODE_ID")
+    let Some(node_id) = agent_config
-        .ok()
+        .node_id
+        .as_ref()
         .map(|value| value.trim().to_string())
         .filter(|value| !value.is_empty())
     else {
@@ -100,19 +103,19 @@ async fn start_agent_heartbeat(
     };
 
     let endpoint = normalize_endpoint(&control_plane_addr);
-    let advertise_endpoint = std::env::var("PLASMAVMC_ENDPOINT_ADVERTISE")
+    let advertise_endpoint = agent_config
-        .ok()
+        .advertise_endpoint
+        .as_ref()
         .map(|value| value.trim().to_string())
         .filter(|value| !value.is_empty())
         .unwrap_or_else(|| local_addr.to_string());
-    let node_name = std::env::var("PLASMAVMC_NODE_NAME")
+    let node_name = agent_config
-        .ok()
+        .node_name
+        .as_ref()
+        .cloned()
         .filter(|value| !value.trim().is_empty())
         .unwrap_or_else(|| node_id.clone());
-    let heartbeat_secs = std::env::var("PLASMAVMC_NODE_HEARTBEAT_INTERVAL_SECS")
+    let heartbeat_secs = agent_config.heartbeat_interval_secs.max(1);
-        .ok()
-        .and_then(|value| value.parse::<u64>().ok())
-        .unwrap_or(5);
 
     tokio::spawn(async move {
         let mut ticker = tokio::time::interval(Duration::from_secs(heartbeat_secs));
@@ -222,7 +225,18 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     let registry = Arc::new(HypervisorRegistry::new());
 
     // Register KVM backend (always available)
-    let kvm_backend = Arc::new(KvmBackend::with_defaults());
+    let kvm_backend = Arc::new(KvmBackend::new(
+        config
+            .kvm
+            .qemu_path
+            .clone()
+            .unwrap_or_else(|| PathBuf::from("/usr/bin/qemu-system-x86_64")),
+        config
+            .kvm
+            .runtime_dir
+            .clone()
+            .unwrap_or_else(|| PathBuf::from("/run/libvirt/plasmavmc")),
+    ));
     registry.register(kvm_backend);
 
     // Register FireCracker backend if kernel/rootfs paths are configured (config or env)
@@ -285,17 +299,18 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
             registry,
             auth_service.clone(),
             config.auth.iam_server_addr.clone(),
+            &config,
         )
         .await?,
     );
 
     // Optional: start state watcher for multi-instance HA sync
-    if std::env::var("PLASMAVMC_STATE_WATCHER")
+    if config.watcher.enabled {
-        .map(|v| matches!(v.as_str(), "1" | "true" | "yes"))
+        let watcher_config = WatcherConfig {
-        .unwrap_or(false)
+            poll_interval: Duration::from_millis(config.watcher.poll_interval_ms.max(100)),
-    {
+            buffer_size: 256,
-        let config = WatcherConfig::default();
+        };
-        let (watcher, rx) = StateWatcher::new(vm_service.store(), config);
+        let (watcher, rx) = StateWatcher::new(vm_service.store(), watcher_config);
         let synchronizer = StateSynchronizer::new(vm_service.clone());
         tokio::spawn(async move {
             if let Err(e) = watcher.start().await {
@@ -305,13 +320,14 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
         tokio::spawn(async move {
             synchronizer.run(rx).await;
         });
-        tracing::info!("State watcher enabled (PLASMAVMC_STATE_WATCHER)");
+        tracing::info!("State watcher enabled");
     }
 
     // Optional: start health monitor to refresh VM status periodically
-    if let Some(secs) = std::env::var("PLASMAVMC_HEALTH_MONITOR_INTERVAL_SECS")
+    if let Some(secs) = config
-        .ok()
+        .health
-        .and_then(|v| v.parse::<u64>().ok())
+        .vm_monitor_interval_secs
+        .filter(|secs| *secs > 0)
     {
         if secs > 0 {
             vm_service
@@ -321,18 +337,15 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     }
 
     // Optional: start node health monitor to detect stale heartbeats
-    if let Some(interval_secs) = std::env::var("PLASMAVMC_NODE_HEALTH_MONITOR_INTERVAL_SECS")
+    if let Some(interval_secs) = config
-        .ok()
+        .health
-        .and_then(|v| v.parse::<u64>().ok())
+        .node_monitor_interval_secs
+        .filter(|secs| *secs > 0)
     {
         if interval_secs > 0 {
-            let timeout_secs = std::env::var("PLASMAVMC_NODE_HEARTBEAT_TIMEOUT_SECS")
-                .ok()
-                .and_then(|v| v.parse::<u64>().ok())
-                .unwrap_or(60);
             vm_service.clone().start_node_health_monitor(
                 Duration::from_secs(interval_secs),
-                Duration::from_secs(timeout_secs),
+                Duration::from_secs(config.health.node_heartbeat_timeout_secs.max(1)),
             );
         }
     }
@@ -369,6 +382,7 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
         heartbeat_volume_drivers,
         heartbeat_storage_classes,
         shared_live_migration,
+        &config.agent,
     )
     .await;
 

plasmavmc/crates/plasmavmc-server/src/storage.rs

@@ -1,7 +1,8 @@
 //! Storage abstraction for VM persistence
 
+use crate::config::{StorageBackendKind, StorageRuntimeConfig};
 use async_trait::async_trait;
-use plasmavmc_types::{Image, Node, VirtualMachine, VmHandle, Volume};
+use plasmavmc_types::{Image, ImageFormat, Node, VirtualMachine, VmHandle, Volume};
 use std::path::PathBuf;
 use thiserror::Error;
 
@@ -16,14 +17,10 @@ pub enum StorageBackend {
 }
 
 impl StorageBackend {
-    pub fn from_env() -> Self {
+    pub fn from_config(config: &StorageRuntimeConfig) -> Self {
-        match std::env::var("PLASMAVMC_STORAGE_BACKEND")
+        match config.backend {
-            .as_deref()
+            StorageBackendKind::Flaredb => Self::FlareDB,
-            .unwrap_or("flaredb")
+            StorageBackendKind::File => Self::File,
-        {
-            "flaredb" => Self::FlareDB,
-            "file" => Self::File,
-            _ => Self::FlareDB,
         }
     }
 }
@@ -48,6 +45,29 @@ pub enum StorageError {
 /// Result type for storage operations
 pub type StorageResult<T> = Result<T, StorageError>;
 
+#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+pub struct ImageUploadPart {
+    pub part_number: u32,
+    pub etag: String,
+    pub size_bytes: u64,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+pub struct ImageUploadSession {
+    pub session_id: String,
+    pub org_id: String,
+    pub project_id: String,
+    pub image_id: String,
+    pub upload_id: String,
+    pub staging_key: String,
+    pub source_format: ImageFormat,
+    #[serde(default)]
+    pub parts: Vec<ImageUploadPart>,
+    pub committed_size_bytes: u64,
+    pub created_at: u64,
+    pub updated_at: u64,
+}
+
 /// Storage trait for VM persistence
 #[async_trait]
 pub trait VmStore: Send + Sync {
@@ -126,6 +146,35 @@ pub trait VmStore: Send + Sync {
     /// List images for a tenant
     async fn list_images(&self, org_id: &str, project_id: &str) -> StorageResult<Vec<Image>>;
 
+    /// Save an in-progress image upload session.
+    async fn save_image_upload_session(&self, session: &ImageUploadSession) -> StorageResult<()>;
+
+    /// Load an in-progress image upload session.
+    async fn load_image_upload_session(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        session_id: &str,
+    ) -> StorageResult<Option<ImageUploadSession>>;
+
+    /// Delete an image upload session.
+    async fn delete_image_upload_session(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        session_id: &str,
+    ) -> StorageResult<()>;
+
+    /// List image upload sessions for an image.
+    async fn list_image_upload_sessions(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+    ) -> StorageResult<Vec<ImageUploadSession>>;
+
     /// Save a persistent volume
     async fn save_volume(&self, volume: &Volume) -> StorageResult<()>;
 
@@ -191,6 +240,27 @@ fn image_prefix(org_id: &str, project_id: &str) -> String {
     format!("/plasmavmc/images/{}/{}/", org_id, project_id)
 }
 
+/// Build key for image upload session metadata
+fn image_upload_session_key(
+    org_id: &str,
+    project_id: &str,
+    image_id: &str,
+    session_id: &str,
+) -> String {
+    format!(
+        "/plasmavmc/image-upload-sessions/{}/{}/{}/{}",
+        org_id, project_id, image_id, session_id
+    )
+}
+
+/// Build prefix for image upload session listing
+fn image_upload_session_prefix(org_id: &str, project_id: &str, image_id: &str) -> String {
+    format!(
+        "/plasmavmc/image-upload-sessions/{}/{}/{}/",
+        org_id, project_id, image_id
+    )
+}
+
 /// Build key for volume metadata
 fn volume_key(org_id: &str, project_id: &str, volume_id: &str) -> String {
     format!("/plasmavmc/volumes/{}/{}/{}", org_id, project_id, volume_id)
@@ -207,14 +277,20 @@ pub struct FlareDBStore {
 }
 
 impl FlareDBStore {
-    /// Create a new FlareDB store
+    pub async fn new_with_config(config: &StorageRuntimeConfig) -> StorageResult<Self> {
-    pub async fn new(endpoint: Option<String>) -> StorageResult<Self> {
+        Self::new_with_endpoints(
-        let endpoint = endpoint.unwrap_or_else(|| {
+            config.flaredb_endpoint.clone(),
-            std::env::var("PLASMAVMC_FLAREDB_ENDPOINT")
+            config.chainfire_endpoint.clone(),
-                .unwrap_or_else(|_| "127.0.0.1:2479".to_string())
+        )
-        });
+        .await
-        let pd_endpoint = std::env::var("PLASMAVMC_CHAINFIRE_ENDPOINT")
+    }
-            .ok()
+
+    pub async fn new_with_endpoints(
+        endpoint: Option<String>,
+        pd_endpoint: Option<String>,
+    ) -> StorageResult<Self> {
+        let endpoint = endpoint.unwrap_or_else(|| "127.0.0.1:2479".to_string());
+        let pd_endpoint = pd_endpoint
             .map(|value| normalize_transport_addr(&value))
             .unwrap_or_else(|| endpoint.clone());
 
@@ -561,6 +637,58 @@ impl VmStore for FlareDBStore {
         Ok(images)
     }
 
+    async fn save_image_upload_session(&self, session: &ImageUploadSession) -> StorageResult<()> {
+        let key = image_upload_session_key(
+            &session.org_id,
+            &session.project_id,
+            &session.image_id,
+            &session.session_id,
+        );
+        let value = serde_json::to_vec(session)?;
+        self.cas_put(&key, value).await
+    }
+
+    async fn load_image_upload_session(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        session_id: &str,
+    ) -> StorageResult<Option<ImageUploadSession>> {
+        let key = image_upload_session_key(org_id, project_id, image_id, session_id);
+        match self.cas_get(&key).await? {
+            Some(data) => Ok(Some(serde_json::from_slice(&data)?)),
+            None => Ok(None),
+        }
+    }
+
+    async fn delete_image_upload_session(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        session_id: &str,
+    ) -> StorageResult<()> {
+        let key = image_upload_session_key(org_id, project_id, image_id, session_id);
+        self.cas_delete(&key).await
+    }
+
+    async fn list_image_upload_sessions(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+    ) -> StorageResult<Vec<ImageUploadSession>> {
+        let prefix = image_upload_session_prefix(org_id, project_id, image_id);
+        let mut sessions = Vec::new();
+        for value in self.cas_scan_values(&prefix).await? {
+            if let Ok(session) = serde_json::from_slice::<ImageUploadSession>(&value) {
+                sessions.push(session);
+            }
+        }
+        Ok(sessions)
+    }
+
     async fn save_volume(&self, volume: &Volume) -> StorageResult<()> {
         let key = volume_key(&volume.org_id, &volume.project_id, &volume.id);
         let value = serde_json::to_vec(volume)?;
@@ -653,6 +781,8 @@ struct PersistedState {
     #[serde(default)]
     images: Vec<Image>,
     #[serde(default)]
+    image_upload_sessions: Vec<ImageUploadSession>,
+    #[serde(default)]
     volumes: Vec<Volume>,
 }
 
@@ -841,6 +971,71 @@ impl VmStore for FileStore {
             .collect())
     }
 
+    async fn save_image_upload_session(&self, session: &ImageUploadSession) -> StorageResult<()> {
+        let mut state = self.load_state().unwrap_or_default();
+        state.image_upload_sessions.retain(|existing| {
+            !(existing.org_id == session.org_id
+                && existing.project_id == session.project_id
+                && existing.image_id == session.image_id
+                && existing.session_id == session.session_id)
+        });
+        state.image_upload_sessions.push(session.clone());
+        self.save_state(&state)?;
+        Ok(())
+    }
+
+    async fn load_image_upload_session(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        session_id: &str,
+    ) -> StorageResult<Option<ImageUploadSession>> {
+        let state = self.load_state().unwrap_or_default();
+        Ok(state.image_upload_sessions.into_iter().find(|session| {
+            session.org_id == org_id
+                && session.project_id == project_id
+                && session.image_id == image_id
+                && session.session_id == session_id
+        }))
+    }
+
+    async fn delete_image_upload_session(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+        session_id: &str,
+    ) -> StorageResult<()> {
+        let mut state = self.load_state().unwrap_or_default();
+        state.image_upload_sessions.retain(|session| {
+            !(session.org_id == org_id
+                && session.project_id == project_id
+                && session.image_id == image_id
+                && session.session_id == session_id)
+        });
+        self.save_state(&state)?;
+        Ok(())
+    }
+
+    async fn list_image_upload_sessions(
+        &self,
+        org_id: &str,
+        project_id: &str,
+        image_id: &str,
+    ) -> StorageResult<Vec<ImageUploadSession>> {
+        let state = self.load_state().unwrap_or_default();
+        Ok(state
+            .image_upload_sessions
+            .into_iter()
+            .filter(|session| {
+                session.org_id == org_id
+                    && session.project_id == project_id
+                    && session.image_id == image_id
+            })
+            .collect())
+    }
+
     async fn save_volume(&self, volume: &Volume) -> StorageResult<()> {
         let mut state = self.load_state().unwrap_or_default();
         state.volumes.retain(|existing| existing.id != volume.id);
@@ -972,4 +1167,68 @@ mod tests {
             .expect("current volume should remain");
         assert_eq!(loaded, current);
     }
+
+    #[tokio::test]
+    async fn filestore_round_trips_image_upload_sessions() {
+        let tempdir = tempdir().unwrap();
+        let store = FileStore::new(Some(tempdir.path().join("state.json")));
+        let session = ImageUploadSession {
+            session_id: "11111111-1111-1111-1111-111111111111".to_string(),
+            org_id: "org-1".to_string(),
+            project_id: "project-1".to_string(),
+            image_id: "22222222-2222-2222-2222-222222222222".to_string(),
+            upload_id: "multipart-1".to_string(),
+            staging_key: "org-1/project-1/uploads/22222222-2222-2222-2222-222222222222.source"
+                .to_string(),
+            source_format: ImageFormat::Raw,
+            parts: vec![ImageUploadPart {
+                part_number: 1,
+                etag: "etag-1".to_string(),
+                size_bytes: 4096,
+            }],
+            committed_size_bytes: 4096,
+            created_at: 1,
+            updated_at: 2,
+        };
+
+        store.save_image_upload_session(&session).await.unwrap();
+
+        let loaded = store
+            .load_image_upload_session(
+                &session.org_id,
+                &session.project_id,
+                &session.image_id,
+                &session.session_id,
+            )
+            .await
+            .unwrap()
+            .expect("session should exist");
+        assert_eq!(loaded, session);
+
+        let listed = store
+            .list_image_upload_sessions(&session.org_id, &session.project_id, &session.image_id)
+            .await
+            .unwrap();
+        assert_eq!(listed, vec![session.clone()]);
+
+        store
+            .delete_image_upload_session(
+                &session.org_id,
+                &session.project_id,
+                &session.image_id,
+                &session.session_id,
+            )
+            .await
+            .unwrap();
+        assert!(store
+            .load_image_upload_session(
+                &session.org_id,
+                &session.project_id,
+                &session.image_id,
+                &session.session_id,
+            )
+            .await
+            .unwrap()
+            .is_none());
+    }
 }

plasmavmc/crates/plasmavmc-server/src/vm_service.rs

@@ -1,8 +1,9 @@
 //! VM Service implementation
 
 use crate::artifact_store::ArtifactStore;
+use crate::config::{DefaultHypervisor, ServerConfig};
 use crate::prismnet_client::PrismNETClient;
-use crate::storage::{FileStore, FlareDBStore, StorageBackend, VmStore};
+use crate::storage::{FileStore, FlareDBStore, ImageUploadSession, StorageBackend, VmStore};
 use crate::volume_manager::VolumeManager;
 use crate::watcher::StateSink;
 use creditservice_client::{Client as CreditServiceClient, ResourceType as CreditResourceType};
@@ -11,13 +12,15 @@ use iam_client::client::IamClientConfig;
 use iam_client::IamClient;
 use iam_service_auth::{
     get_tenant_context, resolve_tenant_ids_from_context, resource_for_tenant, AuthService,
+    TenantContext,
 };
-use iam_types::{PolicyBinding, PrincipalRef, Scope};
+use iam_types::{PolicyBinding, PrincipalKind, PrincipalRef, Scope};
 use plasmavmc_api::proto::{
     disk_source::Source as ProtoDiskSourceKind, image_service_server::ImageService,
     node_service_server::NodeService, vm_service_client::VmServiceClient,
-    vm_service_server::VmService, volume_service_server::VolumeService,
+    vm_service_server::VmService, volume_service_server::VolumeService, AbortImageUploadRequest,
-    Architecture as ProtoArchitecture, AttachDiskRequest, AttachNicRequest, CephRbdBacking,
+    Architecture as ProtoArchitecture, AttachDiskRequest, AttachNicRequest,
+    BeginImageUploadRequest, BeginImageUploadResponse, CephRbdBacking, CompleteImageUploadRequest,
     CordonNodeRequest, CreateImageRequest, CreateVmRequest, CreateVolumeRequest,
     DeleteImageRequest, DeleteVmRequest, DeleteVolumeRequest, DetachDiskRequest, DetachNicRequest,
     DiskBus as ProtoDiskBus, DiskCache as ProtoDiskCache, DiskSource as ProtoDiskSource,
@@ -30,9 +33,9 @@ use plasmavmc_api::proto::{
     NodeState as ProtoNodeState, OsType as ProtoOsType, PrepareVmMigrationRequest, RebootVmRequest,
     RecoverVmRequest, RegisterExternalVolumeRequest, ResetVmRequest, ResizeVolumeRequest,
     StartVmRequest, StopVmRequest, UncordonNodeRequest, UpdateImageRequest, UpdateVmRequest,
-    VirtualMachine, Visibility as ProtoVisibility, VmEvent, VmEventType as ProtoVmEventType,
+    UploadImagePartRequest, UploadImagePartResponse, VirtualMachine, Visibility as ProtoVisibility,
-    VmSpec as ProtoVmSpec, VmState as ProtoVmState, VmStatus as ProtoVmStatus,
+    VmEvent, VmEventType as ProtoVmEventType, VmSpec as ProtoVmSpec, VmState as ProtoVmState,
-    Volume as ProtoVolume, VolumeBacking as ProtoVolumeBacking,
+    VmStatus as ProtoVmStatus, Volume as ProtoVolume, VolumeBacking as ProtoVolumeBacking,
     VolumeDriverKind as ProtoVolumeDriverKind, VolumeFormat as ProtoVolumeFormat,
     VolumeStatus as ProtoVolumeStatus, WatchVmRequest,
 };
@@ -91,6 +94,12 @@ pub struct VmServiceImpl {
     credit_service: Option<Arc<RwLock<CreditServiceClient>>>,
     /// Local node identifier (optional)
     local_node_id: Option<String>,
+    shared_live_migration: bool,
+    default_hypervisor: HypervisorType,
+    watch_poll_interval: Duration,
+    auto_restart: bool,
+    failover_enabled: bool,
+    failover_min_interval_secs: u64,
     artifact_store: Option<Arc<ArtifactStore>>,
     volume_manager: Arc<VolumeManager>,
     iam_client: Arc<IamClient>,
@@ -149,33 +158,34 @@ impl VmServiceImpl {
         hypervisor_registry: Arc<HypervisorRegistry>,
         auth: Arc<AuthService>,
         iam_endpoint: impl Into<String>,
+        config: &ServerConfig,
     ) -> Result<Self, Box<dyn std::error::Error>> {
-        let backend = StorageBackend::from_env();
+        let backend = StorageBackend::from_config(&config.storage);
         let store: Arc<dyn VmStore> = match backend {
-            StorageBackend::FlareDB => match FlareDBStore::new(None).await {
+            StorageBackend::FlareDB => match FlareDBStore::new_with_config(&config.storage).await {
                 Ok(flaredb_store) => Arc::new(flaredb_store),
                 Err(e) => {
                     tracing::warn!(
                         "Failed to connect to FlareDB, falling back to file storage: {}",
                         e
                     );
-                    Arc::new(FileStore::new(None))
+                    Arc::new(FileStore::new(config.storage.state_path.clone()))
                 }
             },
             StorageBackend::File => {
-                let file_store = FileStore::new(None);
+                let file_store = FileStore::new(config.storage.state_path.clone());
                 Arc::new(file_store)
             }
         };
 
-        let prismnet_endpoint = std::env::var("PRISMNET_ENDPOINT").ok();
+        let prismnet_endpoint = config.integrations.prismnet_endpoint.clone();
         if let Some(ref endpoint) = prismnet_endpoint {
             tracing::info!("PrismNET integration enabled: {}", endpoint);
         }
 
         // Initialize CreditService client if endpoint is configured
-        let credit_service = match std::env::var("CREDITSERVICE_ENDPOINT") {
+        let credit_service = match config.integrations.creditservice_endpoint.as_deref() {
-            Ok(endpoint) => match CreditServiceClient::connect(&endpoint).await {
+            Some(endpoint) => match CreditServiceClient::connect(endpoint).await {
                 Ok(client) => {
                     tracing::info!("CreditService admission control enabled: {}", endpoint);
                     Some(Arc::new(RwLock::new(client)))
@@ -188,13 +198,17 @@ impl VmServiceImpl {
                     None
                 }
             },
-            Err(_) => {
+            None => {
-                tracing::info!("CREDITSERVICE_ENDPOINT not set, admission control disabled");
+                tracing::info!("CreditService endpoint not configured, admission control disabled");
                 None
             }
         };
 
-        let local_node_id = std::env::var("PLASMAVMC_NODE_ID").ok();
+        let local_node_id = config
+            .agent
+            .node_id
+            .clone()
+            .filter(|value| !value.trim().is_empty());
         if let Some(ref node_id) = local_node_id {
             tracing::info!("Local node ID: {}", node_id);
         }
@@ -206,13 +220,25 @@ impl VmServiceImpl {
             iam_config = iam_config.without_tls();
         }
         let iam_client = Arc::new(IamClient::connect(iam_config).await?);
-        let artifact_store = ArtifactStore::from_env(&normalized_iam_endpoint)
+        let artifact_store =
+            ArtifactStore::from_config(&config.artifacts, &normalized_iam_endpoint)
                 .await?
                 .map(Arc::new);
         if artifact_store.is_some() {
             tracing::info!("LightningStor artifact backing enabled for VM disks");
         }
-        let volume_manager = Arc::new(VolumeManager::new(store.clone(), artifact_store.clone()));
+        let volume_manager = Arc::new(VolumeManager::new_with_config(
+            store.clone(),
+            artifact_store.clone(),
+            &config.volumes,
+            local_node_id.clone(),
+        )?);
+
+        let default_hypervisor = match config.default_hypervisor {
+            DefaultHypervisor::Kvm => HypervisorType::Kvm,
+            DefaultHypervisor::Firecracker => HypervisorType::Firecracker,
+            DefaultHypervisor::Mvisor => HypervisorType::Mvisor,
+        };
 
         let svc = Self {
             hypervisor_registry,
@@ -224,6 +250,14 @@ impl VmServiceImpl {
             prismnet_endpoint,
             credit_service,
             local_node_id,
+            shared_live_migration: config.agent.shared_live_migration,
+            default_hypervisor,
+            watch_poll_interval: Duration::from_millis(
+                config.watcher.vm_watch_poll_interval_ms.max(100),
+            ),
+            auto_restart: config.health.auto_restart,
+            failover_enabled: config.health.failover_enabled,
+            failover_min_interval_secs: config.health.failover_min_interval_secs.max(1),
             artifact_store,
             volume_manager,
             iam_client,
@@ -242,10 +276,7 @@ impl VmServiceImpl {
     }
 
     pub fn shared_live_migration(&self) -> bool {
-        std::env::var("PLASMAVMC_SHARED_LIVE_MIGRATION")
+        self.shared_live_migration
-            .ok()
-            .map(|value| matches!(value.as_str(), "1" | "true" | "yes"))
-            .unwrap_or(true)
     }
 
     pub fn store(&self) -> Arc<dyn VmStore> {
@@ -256,24 +287,12 @@ impl VmServiceImpl {
         Status::internal(err.to_string())
     }
 
-    fn map_hv(typ: ProtoHypervisorType) -> HypervisorType {
+    fn map_hv(&self, typ: ProtoHypervisorType) -> HypervisorType {
         match typ {
             ProtoHypervisorType::Kvm => HypervisorType::Kvm,
             ProtoHypervisorType::Firecracker => HypervisorType::Firecracker,
             ProtoHypervisorType::Mvisor => HypervisorType::Mvisor,
-            ProtoHypervisorType::Unspecified => {
+            ProtoHypervisorType::Unspecified => self.default_hypervisor,
-                // Use environment variable for default, fallback to KVM
-                match std::env::var("PLASMAVMC_HYPERVISOR")
-                    .as_deref()
-                    .map(|s| s.to_lowercase())
-                    .as_deref()
-                {
-                    Ok("firecracker") => HypervisorType::Firecracker,
-                    Ok("kvm") => HypervisorType::Kvm,
-                    Ok("mvisor") => HypervisorType::Mvisor,
-                    _ => HypervisorType::Kvm, // Default to KVM for backwards compatibility
-                }
-            }
         }
     }
 
@@ -292,13 +311,81 @@ impl VmServiceImpl {
             .as_secs()
     }
 
-    fn watch_poll_interval() -> Duration {
+    fn require_uuid(value: &str, field_name: &str) -> Result<(), Status> {
-        let poll_interval_ms = std::env::var("PLASMAVMC_VM_WATCH_POLL_INTERVAL_MS")
+        Uuid::parse_str(value)
-            .ok()
+            .map(|_| ())
-            .and_then(|value| value.parse::<u64>().ok())
+            .map_err(|_| Status::invalid_argument(format!("{field_name} must be a UUID")))
-            .unwrap_or(500)
+    }
-            .max(100);
+
-        Duration::from_millis(poll_interval_ms)
+    fn validate_disk_reference(disk: &plasmavmc_types::DiskSpec) -> Result<(), Status> {
+        match &disk.source {
+            DiskSource::Image { image_id } => Self::require_uuid(image_id, "image_id"),
+            DiskSource::Volume { volume_id } => Self::require_uuid(volume_id, "volume_id"),
+            DiskSource::Blank => Ok(()),
+        }
+    }
+
+    fn validate_vm_disk_references(spec: &plasmavmc_types::VmSpec) -> Result<(), Status> {
+        for disk in &spec.disks {
+            Self::validate_disk_reference(disk)?;
+        }
+        Ok(())
+    }
+
+    fn ensure_internal_rpc(tenant: &TenantContext) -> Result<(), Status> {
+        if tenant.principal_kind != PrincipalKind::ServiceAccount
+            || !tenant.principal_id.starts_with("plasmavmc-")
+        {
+            return Err(Status::permission_denied(
+                "this RPC is restricted to internal PlasmaVMC service accounts",
+            ));
+        }
+        Ok(())
+    }
+
+    fn expected_migration_listener_uri(&self, vm_id: &str) -> Result<String, Status> {
+        let node_id = self.local_node_id.as_deref().ok_or_else(|| {
+            Status::failed_precondition("local node id is required for migration preparation")
+        })?;
+        let mut hasher = std::collections::hash_map::DefaultHasher::new();
+        node_id.hash(&mut hasher);
+        vm_id.hash(&mut hasher);
+        let port = 4400 + (hasher.finish() % 1000) as u16;
+        Ok(format!("tcp:0.0.0.0:{port}"))
+    }
+
+    fn apply_imported_image_metadata(
+        image: &mut Image,
+        imported: &crate::artifact_store::ImportedImage,
+        source_format: ImageFormat,
+    ) {
+        image.status = ImageStatus::Available;
+        image.format = imported.format;
+        image.size_bytes = imported.size_bytes;
+        image.checksum = imported.checksum.clone();
+        image.updated_at = Self::now_epoch();
+        image
+            .metadata
+            .insert("source_type".to_string(), imported.source_type.clone());
+        if let Some(host) = &imported.source_host {
+            image
+                .metadata
+                .insert("source_host".to_string(), host.clone());
+        }
+        image.metadata.insert(
+            "artifact_key".to_string(),
+            format!("{}/{}/{}.qcow2", image.org_id, image.project_id, image.id),
+        );
+        if source_format != image.format {
+            image.metadata.insert(
+                "source_format".to_string(),
+                format!("{source_format:?}").to_lowercase(),
+            );
+        }
+    }
+
+    fn watch_poll_interval(&self) -> Duration {
+        self.watch_poll_interval
     }
 
     fn vm_values_differ(
@@ -954,7 +1041,10 @@ impl VmServiceImpl {
         }
     }
 
-    fn proto_vm_to_types(vm: &VirtualMachine) -> Result<plasmavmc_types::VirtualMachine, Status> {
+    fn proto_vm_to_types(
+        &self,
+        vm: &VirtualMachine,
+    ) -> Result<plasmavmc_types::VirtualMachine, Status> {
         let spec = Self::proto_spec_to_types(vm.spec.clone());
         let mut typed = plasmavmc_types::VirtualMachine::new(
             vm.name.clone(),
@@ -965,7 +1055,7 @@ impl VmServiceImpl {
         typed.id = VmId::from_uuid(
             Uuid::parse_str(&vm.id).map_err(|e| Status::internal(format!("invalid VM id: {e}")))?,
         );
-        typed.hypervisor = Self::map_hv(
+        typed.hypervisor = self.map_hv(
             ProtoHypervisorType::try_from(vm.hypervisor).unwrap_or(ProtoHypervisorType::Kvm),
         );
         typed.node_id = if vm.node_id.is_empty() {
@@ -1690,9 +1780,6 @@ impl VmServiceImpl {
             .map(|entry| (entry.key().clone(), entry.value().clone()))
             .collect();
 
-        let auto_restart = std::env::var("PLASMAVMC_AUTO_RESTART")
-            .map(|v| matches!(v.as_str(), "1" | "true" | "yes"))
-            .unwrap_or(false);
         let local_node = self.local_node_id.as_deref();
 
         for (key, mut vm) in entries {
@@ -1733,7 +1820,7 @@ impl VmServiceImpl {
 
             match backend.status(&handle).await {
                 Ok(status) => {
-                    if auto_restart
+                    if self.auto_restart
                         && vm.state == VmState::Running
                         && matches!(status.actual_state, VmState::Stopped | VmState::Error)
                     {
@@ -1812,18 +1899,10 @@ impl VmServiceImpl {
             return;
         }
 
-        let failover_enabled = std::env::var("PLASMAVMC_FAILOVER_CONTROLLER")
-            .map(|v| matches!(v.as_str(), "1" | "true" | "yes"))
-            .unwrap_or(false);
-        let min_interval_secs = std::env::var("PLASMAVMC_FAILOVER_MIN_INTERVAL_SECS")
-            .ok()
-            .and_then(|v| v.parse::<u64>().ok())
-            .unwrap_or(60);
-
         let mut failed_over: HashSet<String> = HashSet::new();
-        if failover_enabled {
+        if self.failover_enabled {
             failed_over = self
-                .failover_vms_on_unhealthy(&unhealthy, &nodes, min_interval_secs)
+                .failover_vms_on_unhealthy(&unhealthy, &nodes, self.failover_min_interval_secs)
                 .await;
         }
 
@@ -2047,6 +2126,43 @@ mod tests {
             .unwrap();
         assert_eq!(deleted.event_type, ProtoVmEventType::Deleted as i32);
     }
+
+    #[test]
+    fn internal_rpc_guard_requires_plasmavmc_service_account() {
+        let internal = TenantContext {
+            org_id: "org".to_string(),
+            project_id: "project".to_string(),
+            principal_id: "plasmavmc-org-project".to_string(),
+            principal_name: "plasmavmc".to_string(),
+            principal_kind: PrincipalKind::ServiceAccount,
+            node_id: None,
+        };
+        let external = TenantContext {
+            principal_id: "user-1".to_string(),
+            principal_kind: PrincipalKind::User,
+            ..internal.clone()
+        };
+
+        assert!(VmServiceImpl::ensure_internal_rpc(&internal).is_ok());
+        assert!(VmServiceImpl::ensure_internal_rpc(&external).is_err());
+    }
+
+    #[test]
+    fn vm_disk_reference_validation_requires_uuid_identifiers() {
+        let mut spec = VmSpec::default();
+        spec.disks.push(plasmavmc_types::DiskSpec {
+            id: "root".to_string(),
+            source: DiskSource::Image {
+                image_id: "../passwd".to_string(),
+            },
+            size_gib: 10,
+            bus: DiskBus::Virtio,
+            cache: DiskCache::Writeback,
+            boot_index: Some(1),
+        });
+
+        assert!(VmServiceImpl::validate_vm_disk_references(&spec).is_err());
+    }
 }
 
 impl StateSink for VmServiceImpl {
@@ -2121,13 +2237,14 @@ impl VmService for VmServiceImpl {
             "CreateVm request"
         );
 
-        let hv = Self::map_hv(
+        let hv = self.map_hv(
             ProtoHypervisorType::try_from(req.hypervisor).unwrap_or(ProtoHypervisorType::Kvm),
         );
         if req.spec.is_none() {
             return Err(Status::invalid_argument("spec is required"));
         }
         let spec = Self::proto_spec_to_types(req.spec.clone());
+        Self::validate_vm_disk_references(&spec)?;
         if self.is_control_plane_scheduler() {
             if let Some(target) = self
                 .select_target_node(hv, &req.org_id, &req.project_id, &spec)
@@ -2137,7 +2254,7 @@ impl VmService for VmServiceImpl {
                     let forwarded = self
                         .forward_create_to_node(endpoint, &req.org_id, &req.project_id, &req)
                         .await?;
-                    let forwarded_vm = Self::proto_vm_to_types(&forwarded)?;
+                    let forwarded_vm = self.proto_vm_to_types(&forwarded)?;
                     let key = TenantKey::new(
                         &forwarded_vm.org_id,
                         &forwarded_vm.project_id,
@@ -2395,7 +2512,7 @@ impl VmService for VmServiceImpl {
                             .await
                             .map_err(|status| Status::from_error(Box::new(status)))?
                             .into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key.clone(), typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         return Ok(Response::new(remote_vm));
@@ -2700,7 +2817,7 @@ impl VmService for VmServiceImpl {
                             .await
                             .map_err(|status| Status::from_error(Box::new(status)))?
                             .into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key, typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         return Ok(Response::new(remote_vm));
@@ -2802,7 +2919,7 @@ impl VmService for VmServiceImpl {
                             .await
                             .map_err(|status| Status::from_error(Box::new(status)))?
                             .into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key, typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         return Ok(Response::new(remote_vm));
@@ -2889,7 +3006,7 @@ impl VmService for VmServiceImpl {
                             .await
                             .map_err(|status| Status::from_error(Box::new(status)))?
                             .into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key, typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         return Ok(Response::new(remote_vm));
@@ -2970,7 +3087,7 @@ impl VmService for VmServiceImpl {
                             .await
                             .map_err(|status| Status::from_error(Box::new(status)))?
                             .into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key, typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         return Ok(Response::new(remote_vm));
@@ -3061,7 +3178,7 @@ impl VmService for VmServiceImpl {
                             .await
                             .map_err(|status| Status::from_error(Box::new(status)))?
                             .into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key, typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         return Ok(Response::new(remote_vm));
@@ -3171,7 +3288,7 @@ impl VmService for VmServiceImpl {
                 return match client.recover_vm(recover_req).await {
                     Ok(remote_vm) => {
                         let remote_vm = remote_vm.into_inner();
-                        let typed_vm = Self::proto_vm_to_types(&remote_vm)?;
+                        let typed_vm = self.proto_vm_to_types(&remote_vm)?;
                         self.vms.insert(key, typed_vm.clone());
                         self.persist_vm(&typed_vm).await;
                         Ok(Response::new(remote_vm))
@@ -3352,6 +3469,7 @@ impl VmService for VmServiceImpl {
         request: Request<PrepareVmMigrationRequest>,
     ) -> Result<Response<VirtualMachine>, Status> {
         let tenant = get_tenant_context(&request)?;
+        Self::ensure_internal_rpc(&tenant)?;
         let (org_id, project_id) = resolve_tenant_ids_from_context(
             &tenant,
             &request.get_ref().org_id,
@@ -3378,6 +3496,12 @@ impl VmService for VmServiceImpl {
         if req.listen_uri.is_empty() {
             return Err(Status::invalid_argument("listen_uri is required"));
         }
+        let expected_listen_uri = self.expected_migration_listener_uri(&req.vm_id)?;
+        if req.listen_uri != expected_listen_uri {
+            return Err(Status::invalid_argument(format!(
+                "listen_uri must exactly match {expected_listen_uri}",
+            )));
+        }
 
         self.ensure_destination_slot_available(&req.org_id, &req.project_id, &req.vm_id)
             .await?;
@@ -3385,7 +3509,7 @@ impl VmService for VmServiceImpl {
         let vm_uuid = Uuid::parse_str(&req.vm_id)
             .map_err(|_| Status::invalid_argument("vm_id must be a UUID"))?;
 
-        let hv = Self::map_hv(
+        let hv = self.map_hv(
             ProtoHypervisorType::try_from(req.hypervisor).unwrap_or(ProtoHypervisorType::Kvm),
         );
         let backend = self
@@ -3399,6 +3523,7 @@ impl VmService for VmServiceImpl {
         }
 
         let spec = Self::proto_spec_to_types(req.spec);
+        Self::validate_vm_disk_references(&spec)?;
         let name = if req.name.is_empty() {
             req.vm_id.clone()
         } else {
@@ -3440,6 +3565,7 @@ impl VmService for VmServiceImpl {
         request: Request<RecoverVmRequest>,
     ) -> Result<Response<VirtualMachine>, Status> {
         let tenant = get_tenant_context(&request)?;
+        Self::ensure_internal_rpc(&tenant)?;
         let (org_id, project_id) = resolve_tenant_ids_from_context(
             &tenant,
             &request.get_ref().org_id,
@@ -3469,7 +3595,7 @@ impl VmService for VmServiceImpl {
         let vm_uuid = Uuid::parse_str(&req.vm_id)
             .map_err(|_| Status::invalid_argument("vm_id must be a UUID"))?;
 
-        let hv = Self::map_hv(
+        let hv = self.map_hv(
             ProtoHypervisorType::try_from(req.hypervisor).unwrap_or(ProtoHypervisorType::Kvm),
         );
         let backend = self
@@ -3481,6 +3607,7 @@ impl VmService for VmServiceImpl {
             .await?;
 
         let spec = Self::proto_spec_to_types(req.spec);
+        Self::validate_vm_disk_references(&spec)?;
         let name = if req.name.is_empty() {
             req.vm_id.clone()
         } else {
@@ -3584,6 +3711,7 @@ impl VmService for VmServiceImpl {
             .disk
             .ok_or_else(|| Status::invalid_argument("disk spec required"))?;
         let mut disk_spec = Self::proto_disk_to_types(proto_disk);
+        Self::validate_disk_reference(&disk_spec)?;
         if vm.spec.disks.iter().any(|disk| disk.id == disk_spec.id) {
             return Err(Status::already_exists("disk already attached"));
         }
@@ -3858,7 +3986,7 @@ impl VmService for VmServiceImpl {
             .ensure_vm_loaded(&req.org_id, &req.project_id, &req.vm_id)
             .await
             .ok_or_else(|| Status::not_found("VM not found"))?;
-        let poll_interval = Self::watch_poll_interval();
+        let poll_interval = self.watch_poll_interval();
         let store = Arc::clone(&self.store);
         let org_id = req.org_id.clone();
         let project_id = req.project_id.clone();
@@ -3948,6 +4076,9 @@ impl VolumeService for VmServiceImpl {
                 "size_gib must be greater than zero",
             ));
         }
+        if !req.image_id.trim().is_empty() {
+            Self::require_uuid(&req.image_id, "image_id")?;
+        }
 
         let driver = Self::map_volume_driver(
             ProtoVolumeDriverKind::try_from(req.driver).unwrap_or(ProtoVolumeDriverKind::Managed),
@@ -4206,6 +4337,9 @@ impl ImageService for VmServiceImpl {
         if req.source_url.trim().is_empty() {
             return Err(Status::invalid_argument("source_url is required"));
         }
+        if !req.source_url.starts_with("https://") {
+            return Err(Status::invalid_argument("source_url must use https://"));
+        }
         let Some(store) = self.artifact_store.as_ref() else {
             return Err(Status::failed_precondition(
                 "LightningStor artifact backing is required for image imports",
@@ -4245,20 +4379,7 @@ impl ImageService for VmServiceImpl {
             .await
         {
             Ok(imported) => {
-                image.status = ImageStatus::Available;
+                Self::apply_imported_image_metadata(&mut image, &imported, source_format);
-                image.format = imported.format;
-                image.size_bytes = imported.size_bytes;
-                image.checksum = imported.checksum;
-                image.updated_at = Self::now_epoch();
-                image
-                    .metadata
-                    .insert("source_url".to_string(), req.source_url.clone());
-                if source_format != image.format {
-                    image.metadata.insert(
-                        "source_format".to_string(),
-                        format!("{source_format:?}").to_lowercase(),
-                    );
-                }
                 self.images.insert(key, image.clone());
                 self.persist_image(&image).await;
                 Ok(Response::new(Self::types_image_to_proto(&image)))
@@ -4276,6 +4397,332 @@ impl ImageService for VmServiceImpl {
         }
     }
 
+    async fn begin_image_upload(
+        &self,
+        request: Request<BeginImageUploadRequest>,
+    ) -> Result<Response<BeginImageUploadResponse>, Status> {
+        let tenant = get_tenant_context(&request)?;
+        let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
+        self.auth
+            .authorize(
+                &tenant,
+                ACTION_IMAGE_CREATE,
+                &resource_for_tenant("image", "*", &org_id, &project_id),
+            )
+            .await?;
+
+        let mut req = request.into_inner();
+        if req.name.trim().is_empty() {
+            return Err(Status::invalid_argument("name is required"));
+        }
+        let Some(store) = self.artifact_store.as_ref() else {
+            return Err(Status::failed_precondition(
+                "LightningStor artifact backing is required for image uploads",
+            ));
+        };
+
+        let source_format = Self::map_image_format(
+            ProtoImageFormat::try_from(req.format).unwrap_or(ProtoImageFormat::Qcow2),
+        );
+
+        let mut image = Image::new(req.name, &org_id, &project_id);
+        image.visibility = Self::map_visibility(
+            ProtoVisibility::try_from(req.visibility).unwrap_or(ProtoVisibility::Private),
+        );
+        image.os_type = Self::map_os_type(
+            ProtoOsType::try_from(req.os_type).unwrap_or(ProtoOsType::Unspecified),
+        );
+        image.os_version = req.os_version;
+        image.architecture = Self::map_architecture(req.architecture);
+        image.min_disk_gib = req.min_disk_gib;
+        image.min_memory_mib = req.min_memory_mib;
+        image.metadata = std::mem::take(&mut req.metadata);
+        image.status = ImageStatus::Uploading;
+
+        let key = TenantKey::new(&org_id, &project_id, &image.id);
+        self.images.insert(key.clone(), image.clone());
+        self.persist_image(&image).await;
+
+        let begin_result = store
+            .begin_image_upload(&org_id, &project_id, &image.id)
+            .await;
+        match begin_result {
+            Ok((upload_id, staging_key)) => {
+                let now = Self::now_epoch();
+                let session = ImageUploadSession {
+                    session_id: Uuid::new_v4().to_string(),
+                    org_id: org_id.clone(),
+                    project_id: project_id.clone(),
+                    image_id: image.id.clone(),
+                    upload_id,
+                    staging_key,
+                    source_format,
+                    parts: Vec::new(),
+                    committed_size_bytes: 0,
+                    created_at: now,
+                    updated_at: now,
+                };
+                if let Err(error) = self.store.save_image_upload_session(&session).await {
+                    let _ = store
+                        .abort_image_upload(
+                            &org_id,
+                            &project_id,
+                            &session.staging_key,
+                            &session.upload_id,
+                        )
+                        .await;
+                    image.status = ImageStatus::Error;
+                    image.updated_at = Self::now_epoch();
+                    image.metadata.insert(
+                        "last_error".to_string(),
+                        format!("failed to persist image upload session: {error}"),
+                    );
+                    self.images.insert(key, image.clone());
+                    self.persist_image(&image).await;
+                    return Err(Status::internal(format!(
+                        "failed to persist image upload session: {error}"
+                    )));
+                }
+                Ok(Response::new(BeginImageUploadResponse {
+                    image: Some(Self::types_image_to_proto(&image)),
+                    upload_session_id: session.session_id,
+                    minimum_part_size_bytes: store.minimum_upload_part_size(),
+                }))
+            }
+            Err(error) => {
+                image.status = ImageStatus::Error;
+                image.updated_at = Self::now_epoch();
+                image
+                    .metadata
+                    .insert("last_error".to_string(), error.message().to_string());
+                self.images.insert(key, image.clone());
+                self.persist_image(&image).await;
+                Err(error)
+            }
+        }
+    }
+
+    async fn upload_image_part(
+        &self,
+        request: Request<UploadImagePartRequest>,
+    ) -> Result<Response<UploadImagePartResponse>, Status> {
+        let tenant = get_tenant_context(&request)?;
+        let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
+        let req = request.into_inner();
+        Self::require_uuid(&req.image_id, "image_id")?;
+        Self::require_uuid(&req.upload_session_id, "upload_session_id")?;
+        self.auth
+            .authorize(
+                &tenant,
+                ACTION_IMAGE_CREATE,
+                &resource_for_tenant("image", req.image_id.clone(), &org_id, &project_id),
+            )
+            .await?;
+
+        let Some(store) = self.artifact_store.as_ref() else {
+            return Err(Status::failed_precondition(
+                "LightningStor artifact backing is required for image uploads",
+            ));
+        };
+        let Some(mut session) = self
+            .store
+            .load_image_upload_session(&org_id, &project_id, &req.image_id, &req.upload_session_id)
+            .await
+            .map_err(|error| {
+                Status::internal(format!("failed to load image upload session: {error}"))
+            })?
+        else {
+            return Err(Status::not_found("image upload session not found"));
+        };
+        let Some(image) = self
+            .ensure_image_loaded(&org_id, &project_id, &req.image_id)
+            .await
+        else {
+            return Err(Status::not_found("image not found"));
+        };
+        if image.status != ImageStatus::Uploading {
+            return Err(Status::failed_precondition(
+                "image is not accepting uploaded parts",
+            ));
+        }
+
+        let part = store
+            .upload_image_part(
+                &org_id,
+                &project_id,
+                &session.staging_key,
+                &session.upload_id,
+                req.part_number,
+                req.body,
+            )
+            .await?;
+        session
+            .parts
+            .retain(|existing| existing.part_number != part.part_number);
+        session.parts.push(part);
+        session.parts.sort_by_key(|part| part.part_number);
+        session.committed_size_bytes = session.parts.iter().map(|part| part.size_bytes).sum();
+        session.updated_at = Self::now_epoch();
+        self.store
+            .save_image_upload_session(&session)
+            .await
+            .map_err(|error| {
+                Status::internal(format!("failed to persist image upload session: {error}"))
+            })?;
+
+        Ok(Response::new(UploadImagePartResponse {
+            part_number: req.part_number,
+            committed_size_bytes: session.committed_size_bytes,
+        }))
+    }
+
+    async fn complete_image_upload(
+        &self,
+        request: Request<CompleteImageUploadRequest>,
+    ) -> Result<Response<ProtoImage>, Status> {
+        let tenant = get_tenant_context(&request)?;
+        let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
+        let req = request.into_inner();
+        Self::require_uuid(&req.image_id, "image_id")?;
+        Self::require_uuid(&req.upload_session_id, "upload_session_id")?;
+        self.auth
+            .authorize(
+                &tenant,
+                ACTION_IMAGE_CREATE,
+                &resource_for_tenant("image", req.image_id.clone(), &org_id, &project_id),
+            )
+            .await?;
+
+        let Some(store) = self.artifact_store.as_ref() else {
+            return Err(Status::failed_precondition(
+                "LightningStor artifact backing is required for image uploads",
+            ));
+        };
+        let key = TenantKey::new(&org_id, &project_id, &req.image_id);
+        let Some(mut image) = self
+            .ensure_image_loaded(&org_id, &project_id, &req.image_id)
+            .await
+        else {
+            return Err(Status::not_found("image not found"));
+        };
+        let Some(session) = self
+            .store
+            .load_image_upload_session(&org_id, &project_id, &req.image_id, &req.upload_session_id)
+            .await
+            .map_err(|error| {
+                Status::internal(format!("failed to load image upload session: {error}"))
+            })?
+        else {
+            return Err(Status::not_found("image upload session not found"));
+        };
+
+        match store
+            .complete_image_upload(
+                &org_id,
+                &project_id,
+                &req.image_id,
+                &session.staging_key,
+                &session.upload_id,
+                &session.parts,
+                session.source_format,
+            )
+            .await
+        {
+            Ok(imported) => {
+                Self::apply_imported_image_metadata(&mut image, &imported, session.source_format);
+                self.images.insert(key, image.clone());
+                self.persist_image(&image).await;
+                let _ = self
+                    .store
+                    .delete_image_upload_session(
+                        &org_id,
+                        &project_id,
+                        &req.image_id,
+                        &req.upload_session_id,
+                    )
+                    .await;
+                Ok(Response::new(Self::types_image_to_proto(&image)))
+            }
+            Err(error) => {
+                image.status = ImageStatus::Error;
+                image.updated_at = Self::now_epoch();
+                image
+                    .metadata
+                    .insert("last_error".to_string(), error.message().to_string());
+                self.images.insert(key, image.clone());
+                self.persist_image(&image).await;
+                Err(error)
+            }
+        }
+    }
+
+    async fn abort_image_upload(
+        &self,
+        request: Request<AbortImageUploadRequest>,
+    ) -> Result<Response<Empty>, Status> {
+        let tenant = get_tenant_context(&request)?;
+        let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
+        let req = request.into_inner();
+        Self::require_uuid(&req.image_id, "image_id")?;
+        Self::require_uuid(&req.upload_session_id, "upload_session_id")?;
+        self.auth
+            .authorize(
+                &tenant,
+                ACTION_IMAGE_CREATE,
+                &resource_for_tenant("image", req.image_id.clone(), &org_id, &project_id),
+            )
+            .await?;
+
+        let Some(store) = self.artifact_store.as_ref() else {
+            return Err(Status::failed_precondition(
+                "LightningStor artifact backing is required for image uploads",
+            ));
+        };
+        let Some(session) = self
+            .store
+            .load_image_upload_session(&org_id, &project_id, &req.image_id, &req.upload_session_id)
+            .await
+            .map_err(|error| {
+                Status::internal(format!("failed to load image upload session: {error}"))
+            })?
+        else {
+            return Err(Status::not_found("image upload session not found"));
+        };
+        store
+            .abort_image_upload(
+                &org_id,
+                &project_id,
+                &session.staging_key,
+                &session.upload_id,
+            )
+            .await?;
+        self.store
+            .delete_image_upload_session(
+                &org_id,
+                &project_id,
+                &req.image_id,
+                &req.upload_session_id,
+            )
+            .await
+            .map_err(|error| {
+                Status::internal(format!("failed to delete image upload session: {error}"))
+            })?;
+        if let Some(mut image) = self
+            .ensure_image_loaded(&org_id, &project_id, &req.image_id)
+            .await
+        {
+            let key = TenantKey::new(&org_id, &project_id, &req.image_id);
+            image.status = ImageStatus::Error;
+            image.updated_at = Self::now_epoch();
+            image
+                .metadata
+                .insert("last_error".to_string(), "upload aborted".to_string());
+            self.images.insert(key, image.clone());
+            self.persist_image(&image).await;
+        }
+        Ok(Response::new(Empty {}))
+    }
+
     async fn get_image(
         &self,
         request: Request<GetImageRequest>,
@@ -4283,6 +4730,7 @@ impl ImageService for VmServiceImpl {
         let tenant = get_tenant_context(&request)?;
         let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
         let req = request.into_inner();
+        Self::require_uuid(&req.image_id, "image_id")?;
         self.auth
             .authorize(
                 &tenant,
@@ -4337,6 +4785,7 @@ impl ImageService for VmServiceImpl {
         let tenant = get_tenant_context(&request)?;
         let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
         let req = request.into_inner();
+        Self::require_uuid(&req.image_id, "image_id")?;
         self.auth
             .authorize(
                 &tenant,
@@ -4378,6 +4827,7 @@ impl ImageService for VmServiceImpl {
         let tenant = get_tenant_context(&request)?;
         let (org_id, project_id) = Self::resolve_image_tenant(&tenant, &request.get_ref().org_id)?;
         let req = request.into_inner();
+        Self::require_uuid(&req.image_id, "image_id")?;
         self.auth
             .authorize(
                 &tenant,
@@ -4396,6 +4846,31 @@ impl ImageService for VmServiceImpl {
         }
 
         if let Some(store) = self.artifact_store.as_ref() {
+            if let Ok(sessions) = self
+                .store
+                .list_image_upload_sessions(&org_id, &project_id, &req.image_id)
+                .await
+            {
+                for session in sessions {
+                    let _ = store
+                        .abort_image_upload(
+                            &org_id,
+                            &project_id,
+                            &session.staging_key,
+                            &session.upload_id,
+                        )
+                        .await;
+                    let _ = self
+                        .store
+                        .delete_image_upload_session(
+                            &org_id,
+                            &project_id,
+                            &req.image_id,
+                            &session.session_id,
+                        )
+                        .await;
+                }
+            }
             store
                 .delete_image(&org_id, &project_id, &req.image_id)
                 .await?;
@@ -4540,7 +5015,7 @@ impl NodeService for VmServiceImpl {
                 .hypervisors
                 .iter()
                 .filter_map(|h| ProtoHypervisorType::try_from(*h).ok())
-                .map(Self::map_hv)
+                .map(|hypervisor| self.map_hv(hypervisor))
                 .collect();
         }
         if !req.supported_volume_drivers.is_empty() {

plasmavmc/crates/plasmavmc-server/src/volume_manager.rs

@@ -1,4 +1,5 @@
 use crate::artifact_store::ArtifactStore;
+use crate::config::{CephConfig, CoronaFsConfig, VolumeRuntimeConfig};
 use crate::storage::VmStore;
 use plasmavmc_types::{
     AttachedDisk, DiskAttachment, DiskSource, DiskSpec, VirtualMachine, Volume, VolumeBacking,
@@ -18,6 +19,7 @@ const AUTO_DELETE_VOLUME_SOURCE_METADATA_KEY: &str = "plasmavmc.auto_delete_sour
 const CORONAFS_IMAGE_SOURCE_ID_METADATA_KEY: &str = "plasmavmc.coronafs_image_source_id";
 const CORONAFS_IMAGE_SEED_PENDING_METADATA_KEY: &str = "plasmavmc.coronafs_image_seed_pending";
 const VOLUME_METADATA_CAS_RETRIES: usize = 16;
+const CEPH_IDENTIFIER_MAX_LEN: usize = 128;
 
 #[derive(Clone, Debug)]
 struct CephClusterConfig {
@@ -116,6 +118,7 @@ pub struct VolumeManager {
     artifact_store: Option<Arc<ArtifactStore>>,
     managed_root: PathBuf,
     supported_storage_classes: Vec<String>,
+    qemu_img_path: PathBuf,
     ceph_cluster: Option<CephClusterConfig>,
     coronafs_controller: Option<CoronaFsClient>,
     coronafs_node: Option<CoronaFsClient>,
@@ -124,31 +127,43 @@ pub struct VolumeManager {
 }
 
 impl VolumeManager {
-    pub fn new(store: Arc<dyn VmStore>, artifact_store: Option<Arc<ArtifactStore>>) -> Self {
+    pub fn new_with_config(
-        let managed_root = std::env::var("PLASMAVMC_MANAGED_VOLUME_ROOT")
+        store: Arc<dyn VmStore>,
-            .map(PathBuf::from)
+        artifact_store: Option<Arc<ArtifactStore>>,
-            .unwrap_or_else(|_| PathBuf::from("/var/lib/plasmavmc/managed-volumes"));
+        config: &VolumeRuntimeConfig,
-        let ceph_cluster = std::env::var("PLASMAVMC_CEPH_MONITORS")
+        local_node_id: Option<String>,
-            .ok()
+    ) -> Result<Self, Box<dyn std::error::Error>> {
-            .filter(|value| !value.trim().is_empty())
+        let managed_root = config.managed_volume_root.clone();
-            .map(|monitors| CephClusterConfig {
+        let ceph_cluster = ceph_cluster_from_config(&config.ceph);
-                cluster_id: std::env::var("PLASMAVMC_CEPH_CLUSTER_ID")
+        let (coronafs_controller, coronafs_node) =
-                    .unwrap_or_else(|_| "default".to_string()),
+            resolve_coronafs_clients_with_config(&config.coronafs);
-                monitors: monitors
+        let coronafs_node_local_attach = config.coronafs.node_local_attach;
-                    .split(',')
+        let qemu_img_path = resolve_binary_path(config.qemu_img_path.as_deref(), "qemu-img")?;
-                    .map(str::trim)
-                    .filter(|item| !item.is_empty())
-                    .map(ToOwned::to_owned)
-                    .collect(),
-                user: std::env::var("PLASMAVMC_CEPH_USER").unwrap_or_else(|_| "admin".to_string()),
-                secret: std::env::var("PLASMAVMC_CEPH_SECRET").ok(),
-            });
-        let (coronafs_controller, coronafs_node) = resolve_coronafs_clients();
-        let coronafs_node_local_attach = coronafs_node_local_attach_enabled();
-        let local_node_id = std::env::var("PLASMAVMC_NODE_ID")
-            .ok()
-            .filter(|value| !value.trim().is_empty());
 
+        Ok(Self::build(
+            store,
+            artifact_store,
+            managed_root,
+            qemu_img_path,
+            ceph_cluster,
+            coronafs_controller,
+            coronafs_node,
+            coronafs_node_local_attach,
+            local_node_id,
+        ))
+    }
+
+    fn build(
+        store: Arc<dyn VmStore>,
+        artifact_store: Option<Arc<ArtifactStore>>,
+        managed_root: PathBuf,
+        qemu_img_path: PathBuf,
+        ceph_cluster: Option<CephClusterConfig>,
+        coronafs_controller: Option<CoronaFsClient>,
+        coronafs_node: Option<CoronaFsClient>,
+        coronafs_node_local_attach: bool,
+        local_node_id: Option<String>,
+    ) -> Self {
         Self {
             store,
             artifact_store,
@@ -161,6 +176,7 @@ impl VolumeManager {
                 classes.push("ceph-rbd".to_string());
                 classes
             },
+            qemu_img_path,
             ceph_cluster,
             coronafs_controller,
             coronafs_node,
@@ -320,6 +336,7 @@ impl VolumeManager {
             Some(export) => export,
             None => controller.ensure_export_read_only(volume_id).await?,
         };
+        validate_coronafs_export(&export)?;
         node_client
             .materialize_from_export(
                 volume_id,
@@ -377,6 +394,9 @@ impl VolumeManager {
         {
             return Ok(existing);
         }
+        if let Some(image_id) = image_id {
+            validate_uuid(image_id, "image_id")?;
+        }
 
         let path = self.managed_volume_path(volume_id);
         let mut metadata = metadata;
@@ -503,6 +523,9 @@ impl VolumeManager {
                 "Ceph RBD support is not configured on this node",
             ));
         }
+        validate_ceph_identifier("ceph_rbd.cluster_id", cluster_id)?;
+        validate_ceph_identifier("ceph_rbd.pool", pool)?;
+        validate_ceph_identifier("ceph_rbd.image", image)?;
         let volume_id = Uuid::new_v4().to_string();
         let mut volume = Volume::new(volume_id, name.to_string(), org_id, project_id, size_gib);
         volume.driver = VolumeDriverKind::CephRbd;
@@ -943,6 +966,7 @@ impl VolumeManager {
         let Some(local_volume) = node_client.get_volume_optional(volume_id).await? else {
             return Ok(());
         };
+        validate_coronafs_volume_response(&local_volume)?;
         if local_volume.export.is_some() {
             node_client.release_export(volume_id).await?;
         }
@@ -975,6 +999,7 @@ impl VolumeManager {
             Some(export) => export,
             None => controller.ensure_export(volume_id).await?,
         };
+        validate_coronafs_export(&export)?;
         tracing::info!(
             volume_id,
             node_endpoint = %node_client.endpoint,
@@ -983,7 +1008,7 @@ impl VolumeManager {
             export_uri = %export.uri,
             "Syncing node-local CoronaFS volume back to controller"
         );
-        sync_local_coronafs_volume_to_export(&local_volume, &export.uri).await
+        sync_local_coronafs_volume_to_export(&self.qemu_img_path, &local_volume, &export.uri).await
     }
 
     async fn materialize_pending_coronafs_image_seed_on_node(
@@ -1185,6 +1210,7 @@ impl VolumeManager {
                 if self.coronafs_attachment_backend().is_some() {
                     let (coronafs_volume, coronafs) =
                         self.load_coronafs_volume_for_attachment(volume).await?;
+                    validate_coronafs_volume_response(&coronafs_volume)?;
                     if coronafs.supports_local_backing_file().await
                         && !should_prefer_coronafs_export_attachment(&coronafs_volume)
                         && coronafs_local_target_ready(&coronafs_volume.path).await
@@ -1226,6 +1252,9 @@ impl VolumeManager {
                         cluster_id
                     )));
                 }
+                validate_ceph_identifier("ceph cluster_id", cluster_id)?;
+                validate_ceph_identifier("ceph pool", pool)?;
+                validate_ceph_identifier("ceph image", image)?;
                 DiskAttachment::CephRbd {
                     pool: pool.clone(),
                     image: image.clone(),
@@ -1266,7 +1295,7 @@ impl VolumeManager {
                 .await
                 .map_err(|e| Status::internal(format!("failed to create volume dir: {e}")))?;
         }
-        let status = Command::new("qemu-img")
+        let status = Command::new(&self.qemu_img_path)
             .args([
                 "convert",
                 "-O",
@@ -1307,7 +1336,7 @@ impl VolumeManager {
             .materialize_image_cache(org_id, project_id, image_id)
             .await?;
         let requested_size = gib_to_bytes(size_gib);
-        let image_info = inspect_qemu_image(&image_path).await?;
+        let image_info = inspect_qemu_image(&self.qemu_img_path, &image_path).await?;
         if requested_size < image_info.virtual_size {
             return Err(Status::failed_precondition(format!(
                 "requested volume {} GiB is smaller than image virtual size {} bytes",
@@ -1413,6 +1442,7 @@ impl VolumeManager {
                 &raw_image_path,
                 Path::new(&volume.path),
                 requested_size,
+                &self.qemu_img_path,
             )
             .await?;
             return Ok(CoronaFsProvisionOutcome {
@@ -1431,7 +1461,7 @@ impl VolumeManager {
                 export_uri = %export.uri,
                 "Populating CoronaFS-backed VM volume over NBD from image cache"
             );
-            let status = Command::new("qemu-img")
+            let status = Command::new(&self.qemu_img_path)
                 .args([
                     "convert",
                     "-t",
@@ -1481,7 +1511,7 @@ impl VolumeManager {
                 .await
                 .map_err(|e| Status::internal(format!("failed to create volume dir: {e}")))?;
         }
-        let status = Command::new("qemu-img")
+        let status = Command::new(&self.qemu_img_path)
             .args([
                 "create",
                 "-f",
@@ -1508,7 +1538,7 @@ impl VolumeManager {
         format: VolumeFormat,
         size_gib: u64,
     ) -> Result<(), Status> {
-        let status = Command::new("qemu-img")
+        let status = Command::new(&self.qemu_img_path)
             .args([
                 "resize",
                 "-f",
@@ -1542,8 +1572,8 @@ impl VolumeManager {
     }
 }
 
-async fn inspect_qemu_image(path: &Path) -> Result<QemuImageInfo, Status> {
+async fn inspect_qemu_image(qemu_img_path: &Path, path: &Path) -> Result<QemuImageInfo, Status> {
-    let output = Command::new("qemu-img")
+    let output = Command::new(qemu_img_path)
         .args(["info", "--output", "json", path.to_string_lossy().as_ref()])
         .output()
         .await
@@ -1585,6 +1615,7 @@ async fn clone_local_raw_into_coronafs_volume(
     source: &Path,
     destination: &Path,
     requested_size: u64,
+    _qemu_img_path: &Path,
 ) -> Result<(), Status> {
     let temp_path = destination.with_extension("clone.tmp");
     if let Some(parent) = temp_path.parent() {
@@ -1595,33 +1626,13 @@ async fn clone_local_raw_into_coronafs_volume(
     if tokio::fs::try_exists(&temp_path).await.unwrap_or(false) {
         let _ = tokio::fs::remove_file(&temp_path).await;
     }
-
+    tokio::fs::copy(source, &temp_path).await.map_err(|e| {
-    let copy_output = Command::new("cp")
+        Status::internal(format!(
-        .args([
+            "failed to clone raw image {} into {}: {e}",
-            "--reflink=auto",
-            "--sparse=always",
-            source.to_string_lossy().as_ref(),
-            temp_path.to_string_lossy().as_ref(),
-        ])
-        .output()
-        .await
-        .map_err(|e| Status::internal(format!("failed to spawn raw clone copy: {e}")))?;
-    if !copy_output.status.success() {
-        let stderr = String::from_utf8_lossy(&copy_output.stderr)
-            .trim()
-            .to_string();
-        return Err(Status::internal(format!(
-            "failed to clone raw image {} into {} with status {}{}",
             source.display(),
-            temp_path.display(),
+            temp_path.display()
-            copy_output.status,
+        ))
-            if stderr.is_empty() {
+    })?;
-                String::new()
-            } else {
-                format!(": {stderr}")
-            }
-        )));
-    }
 
     let file = tokio::fs::OpenOptions::new()
         .write(true)
@@ -1656,11 +1667,12 @@ async fn clone_local_raw_into_coronafs_volume(
 }
 
 async fn sync_local_coronafs_volume_to_export(
+    qemu_img_path: &Path,
     local_volume: &CoronaFsVolumeResponse,
     export_uri: &str,
 ) -> Result<(), Status> {
     let local_format = local_volume.format.unwrap_or(CoronaFsVolumeFormat::Raw);
-    let status = Command::new("qemu-img")
+    let status = Command::new(qemu_img_path)
         .args([
             "convert",
             "-t",
@@ -1773,17 +1785,121 @@ fn gib_to_bytes(size_gib: u64) -> u64 {
     size_gib.saturating_mul(1024 * 1024 * 1024)
 }
 
-fn coronafs_node_local_attach_enabled() -> bool {
+fn validate_uuid(value: &str, field_name: &str) -> Result<(), Status> {
-    coronafs_node_local_attach_enabled_from_values(
+    Uuid::parse_str(value)
-        std::env::var("PLASMAVMC_CORONAFS_NODE_LOCAL_ATTACH")
+        .map(|_| ())
-            .ok()
+        .map_err(|_| Status::invalid_argument(format!("{field_name} must be a UUID")))
-            .as_deref(),
-        std::env::var("PLASMAVMC_CORONAFS_ENABLE_EXPERIMENTAL_NODE_LOCAL_ATTACH")
-            .ok()
-            .as_deref(),
-    )
 }
 
+fn validate_ceph_identifier(field_name: &str, value: &str) -> Result<(), Status> {
+    let trimmed = value.trim();
+    if trimmed.is_empty() {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} is required"
+        )));
+    }
+    if trimmed.len() > CEPH_IDENTIFIER_MAX_LEN {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} exceeds {CEPH_IDENTIFIER_MAX_LEN} characters"
+        )));
+    }
+    let mut chars = trimmed.chars();
+    let Some(first) = chars.next() else {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} is required"
+        )));
+    };
+    if !first.is_ascii_alphanumeric() {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} must start with an ASCII alphanumeric character"
+        )));
+    }
+    if chars.any(|ch| !(ch.is_ascii_alphanumeric() || matches!(ch, '.' | '_' | '-'))) {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} contains unsupported characters"
+        )));
+    }
+    Ok(())
+}
+
+fn validate_absolute_path(path: &str, field_name: &str) -> Result<(), Status> {
+    let candidate = Path::new(path);
+    if !candidate.is_absolute() {
+        return Err(Status::failed_precondition(format!(
+            "{field_name} must be an absolute path"
+        )));
+    }
+    Ok(())
+}
+
+fn validate_nbd_uri(uri: &str, field_name: &str) -> Result<(), Status> {
+    let parsed = reqwest::Url::parse(uri)
+        .map_err(|e| Status::failed_precondition(format!("{field_name} is invalid: {e}")))?;
+    if parsed.scheme() != "nbd" {
+        return Err(Status::failed_precondition(format!(
+            "{field_name} must use nbd://"
+        )));
+    }
+    if !parsed.username().is_empty() || parsed.password().is_some() {
+        return Err(Status::failed_precondition(format!(
+            "{field_name} must not contain embedded credentials"
+        )));
+    }
+    if parsed.host_str().is_none() || parsed.port().is_none() {
+        return Err(Status::failed_precondition(format!(
+            "{field_name} must include host and port"
+        )));
+    }
+    Ok(())
+}
+
+fn validate_coronafs_export(export: &CoronaFsExport) -> Result<(), Status> {
+    validate_nbd_uri(&export.uri, "coronafs export uri")
+}
+
+fn validate_coronafs_volume_response(volume: &CoronaFsVolumeResponse) -> Result<(), Status> {
+    validate_absolute_path(&volume.path, "coronafs volume path")?;
+    if let Some(export) = &volume.export {
+        validate_coronafs_export(export)?;
+    }
+    if let Some(source) = &volume.materialized_from {
+        if source.starts_with("nbd://") {
+            validate_nbd_uri(source, "coronafs materialized_from")?;
+        }
+    }
+    Ok(())
+}
+
+fn resolve_binary_path(
+    configured_path: Option<&Path>,
+    binary_name: &str,
+) -> Result<PathBuf, Box<dyn std::error::Error>> {
+    let candidate = match configured_path {
+        Some(path) => path.to_path_buf(),
+        None => std::env::var_os("PATH")
+            .into_iter()
+            .flat_map(|paths| std::env::split_paths(&paths).collect::<Vec<_>>())
+            .map(|entry| entry.join(binary_name))
+            .find(|candidate| candidate.exists())
+            .ok_or_else(|| {
+                std::io::Error::new(
+                    std::io::ErrorKind::NotFound,
+                    format!("failed to locate {binary_name} in PATH"),
+                )
+            })?,
+    };
+    let metadata = std::fs::metadata(&candidate)?;
+    if !metadata.is_file() {
+        return Err(std::io::Error::new(
+            std::io::ErrorKind::InvalidInput,
+            format!("{} is not a regular file", candidate.display()),
+        )
+        .into());
+    }
+    Ok(candidate)
+}
+
+#[cfg(test)]
 fn coronafs_node_local_attach_enabled_from_values(
     stable_value: Option<&str>,
     legacy_value: Option<&str>,
@@ -1794,6 +1910,23 @@ fn coronafs_node_local_attach_enabled_from_values(
         .unwrap_or(false)
 }
 
+fn ceph_cluster_from_config(config: &CephConfig) -> Option<CephClusterConfig> {
+    if config.monitors.is_empty() {
+        None
+    } else {
+        Some(CephClusterConfig {
+            cluster_id: config.cluster_id.clone(),
+            monitors: config.monitors.clone(),
+            user: config.user.clone(),
+            secret: config
+                .secret
+                .clone()
+                .or_else(|| std::env::var("PLASMAVMC_CEPH_SECRET").ok()),
+        })
+    }
+}
+
+#[cfg(test)]
 fn parse_truthy(value: &str) -> bool {
     matches!(
         value.trim().to_ascii_lowercase().as_str(),
@@ -1801,16 +1934,21 @@ fn parse_truthy(value: &str) -> bool {
     )
 }
 
-fn resolve_coronafs_clients() -> (Option<CoronaFsClient>, Option<CoronaFsClient>) {
+fn resolve_coronafs_clients_with_config(
+    config: &CoronaFsConfig,
+) -> (Option<CoronaFsClient>, Option<CoronaFsClient>) {
     let (controller_endpoint, node_endpoint) = resolve_coronafs_endpoints(
-        std::env::var("PLASMAVMC_CORONAFS_CONTROLLER_ENDPOINT")
+        config
-            .ok()
+            .controller_endpoint
+            .clone()
             .and_then(normalize_coronafs_endpoint),
-        std::env::var("PLASMAVMC_CORONAFS_NODE_ENDPOINT")
+        config
-            .ok()
+            .node_endpoint
+            .clone()
             .and_then(normalize_coronafs_endpoint),
-        std::env::var("PLASMAVMC_CORONAFS_ENDPOINT")
+        config
-            .ok()
+            .endpoint
+            .clone()
             .and_then(normalize_coronafs_endpoint),
     );
     (
@@ -1911,14 +2049,19 @@ impl CoronaFsClient {
             backing_file: None,
             backing_format: None,
         };
-        self.try_request("create CoronaFS volume", |endpoint, http| {
+        let volume = self
+            .try_request("create CoronaFS volume", |endpoint, http| {
                 http.put(format!("{endpoint}/v1/volumes/{volume_id}"))
                     .json(&request)
             })
             .await?
             .json::<CoronaFsVolumeResponse>()
             .await
-        .map_err(|e| Status::internal(format!("failed to decode CoronaFS create response: {e}")))
+            .map_err(|e| {
+                Status::internal(format!("failed to decode CoronaFS create response: {e}"))
+            })?;
+        validate_coronafs_volume_response(&volume)?;
+        Ok(volume)
     }
 
     async fn create_image_backed(
@@ -1934,7 +2077,8 @@ impl CoronaFsClient {
             backing_file: Some(backing_file.to_string_lossy().into_owned()),
             backing_format: Some(backing_format),
         };
-        self.try_request("create CoronaFS image-backed volume", |endpoint, http| {
+        let volume = self
+            .try_request("create CoronaFS image-backed volume", |endpoint, http| {
                 http.put(format!("{endpoint}/v1/volumes/{volume_id}"))
                     .json(&request)
             })
@@ -1945,17 +2089,24 @@ impl CoronaFsClient {
                 Status::internal(format!(
                     "failed to decode CoronaFS image-backed create response: {e}"
                 ))
-        })
+            })?;
+        validate_coronafs_volume_response(&volume)?;
+        Ok(volume)
     }
 
     async fn get_volume(&self, volume_id: &str) -> Result<CoronaFsVolumeResponse, Status> {
-        self.try_request("inspect CoronaFS volume", |endpoint, http| {
+        let volume = self
+            .try_request("inspect CoronaFS volume", |endpoint, http| {
                 http.get(format!("{endpoint}/v1/volumes/{volume_id}"))
             })
             .await?
             .json::<CoronaFsVolumeResponse>()
             .await
-        .map_err(|e| Status::internal(format!("failed to decode CoronaFS inspect response: {e}")))
+            .map_err(|e| {
+                Status::internal(format!("failed to decode CoronaFS inspect response: {e}"))
+            })?;
+        validate_coronafs_volume_response(&volume)?;
+        Ok(volume)
     }
 
     async fn get_volume_optional(
@@ -2001,9 +2152,11 @@ impl CoronaFsClient {
             .map_err(|e| {
                 Status::internal(format!("failed to decode CoronaFS export response: {e}"))
             })?;
-        response.export.ok_or_else(|| {
+        let export = response.export.ok_or_else(|| {
             Status::internal("CoronaFS export response did not include an export URI")
-        })
+        })?;
+        validate_coronafs_export(&export)?;
+        Ok(export)
     }
 
     async fn materialize_from_export(
@@ -2014,13 +2167,15 @@ impl CoronaFsClient {
         format: Option<CoronaFsVolumeFormat>,
         lazy: bool,
     ) -> Result<CoronaFsVolumeResponse, Status> {
+        validate_nbd_uri(source_uri, "coronafs materialize source_uri")?;
         let request = CoronaFsMaterializeRequest {
             source_uri: source_uri.to_string(),
             size_bytes,
             format,
             lazy,
         };
-        self.try_request("materialize CoronaFS volume", |endpoint, http| {
+        let volume = self
+            .try_request("materialize CoronaFS volume", |endpoint, http| {
                 http.post(format!("{endpoint}/v1/volumes/{volume_id}/materialize"))
                     .json(&request)
             })
@@ -2031,7 +2186,9 @@ impl CoronaFsClient {
                 Status::internal(format!(
                     "failed to decode CoronaFS materialize response: {e}"
                 ))
-        })
+            })?;
+        validate_coronafs_volume_response(&volume)?;
+        Ok(volume)
     }
 
     async fn resize_volume(&self, volume_id: &str, size_bytes: u64) -> Result<(), Status> {
@@ -2210,7 +2367,12 @@ mod tests {
             .unwrap();
         tokio::fs::write(&source, b"raw-seed").await.unwrap();
 
-        clone_local_raw_into_coronafs_volume(&source, &destination, 4096)
+        clone_local_raw_into_coronafs_volume(
+            &source,
+            &destination,
+            4096,
+            Path::new("/usr/bin/qemu-img"),
+        )
         .await
         .unwrap();
 
@@ -2398,4 +2560,27 @@ mod tests {
             .remove(CORONAFS_IMAGE_SEED_PENDING_METADATA_KEY);
         assert!(!volume_has_pending_coronafs_image_seed(&volume));
     }
+
+    #[test]
+    fn ceph_identifier_validation_rejects_option_injection_characters() {
+        assert!(validate_ceph_identifier("pool", "pool-alpha").is_ok());
+        assert!(validate_ceph_identifier("pool", "pool,inject").is_err());
+        assert!(validate_ceph_identifier("image", "image:inject").is_err());
+    }
+
+    #[test]
+    fn coronafs_export_validation_requires_nbd_uri() {
+        let valid = CoronaFsExport {
+            uri: "nbd://10.0.0.1:11000".to_string(),
+            port: 11000,
+            pid: None,
+        };
+        let invalid = CoronaFsExport {
+            uri: "http://10.0.0.1:11000".to_string(),
+            ..valid.clone()
+        };
+
+        assert!(validate_coronafs_export(&valid).is_ok());
+        assert!(validate_coronafs_export(&invalid).is_err());
+    }
 }

plasmavmc/crates/plasmavmc-server/src/watcher.rs

@@ -64,13 +64,8 @@ pub struct WatcherConfig {
 
 impl Default for WatcherConfig {
     fn default() -> Self {
-        let poll_interval_ms = std::env::var("PLASMAVMC_STATE_WATCHER_POLL_INTERVAL_MS")
-            .ok()
-            .and_then(|value| value.parse::<u64>().ok())
-            .unwrap_or(1000)
-            .max(100);
         Self {
-            poll_interval: Duration::from_millis(poll_interval_ms),
+            poll_interval: Duration::from_millis(1000),
             buffer_size: 256,
         }
     }

plasmavmc/crates/plasmavmc-types/src/config.rs

@@ -4,8 +4,7 @@ use serde::{Deserialize, Serialize};
 use std::path::PathBuf;
 
 /// FireCracker backend configuration
-#[derive(Debug, Clone, Serialize, Deserialize)]
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
-#[derive(Default)]
 pub struct FireCrackerConfig {
     /// Path to the Firecracker binary
     pub firecracker_path: Option<PathBuf>,
@@ -27,11 +26,11 @@ pub struct FireCrackerConfig {
     pub use_jailer: Option<bool>,
 }
 
-
 /// KVM backend configuration
-#[derive(Debug, Clone, Serialize, Deserialize)]
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
-#[derive(Default)]
 pub struct KvmConfig {
-    // Add KVM specific configuration fields here if any
+    /// Path to the QEMU binary
+    pub qemu_path: Option<PathBuf>,
+    /// Runtime directory used for QMP sockets and console logs
+    pub runtime_dir: Option<PathBuf>,
 }
-

plasmavmc/proto/plasmavmc.proto

@@ -44,6 +44,10 @@ service VmService {
 
 service ImageService {
   rpc CreateImage(CreateImageRequest) returns (Image);
+  rpc BeginImageUpload(BeginImageUploadRequest) returns (BeginImageUploadResponse);
+  rpc UploadImagePart(UploadImagePartRequest) returns (UploadImagePartResponse);
+  rpc CompleteImageUpload(CompleteImageUploadRequest) returns (Image);
+  rpc AbortImageUpload(AbortImageUploadRequest) returns (Empty);
   rpc GetImage(GetImageRequest) returns (Image);
   rpc ListImages(ListImagesRequest) returns (ListImagesResponse);
   rpc UpdateImage(UpdateImageRequest) returns (Image);
@@ -475,6 +479,50 @@ message CreateImageRequest {
   string source_url = 11;
 }
 
+message BeginImageUploadRequest {
+  string name = 1;
+  string org_id = 2;
+  Visibility visibility = 3;
+  ImageFormat format = 4;
+  OsType os_type = 5;
+  string os_version = 6;
+  Architecture architecture = 7;
+  uint32 min_disk_gib = 8;
+  uint32 min_memory_mib = 9;
+  map<string, string> metadata = 10;
+}
+
+message BeginImageUploadResponse {
+  Image image = 1;
+  string upload_session_id = 2;
+  uint32 minimum_part_size_bytes = 3;
+}
+
+message UploadImagePartRequest {
+  string org_id = 1;
+  string image_id = 2;
+  string upload_session_id = 3;
+  uint32 part_number = 4;
+  bytes body = 5;
+}
+
+message UploadImagePartResponse {
+  uint32 part_number = 1;
+  uint64 committed_size_bytes = 2;
+}
+
+message CompleteImageUploadRequest {
+  string org_id = 1;
+  string image_id = 2;
+  string upload_session_id = 3;
+}
+
+message AbortImageUploadRequest {
+  string org_id = 1;
+  string image_id = 2;
+  string upload_session_id = 3;
+}
+
 message GetImageRequest {
   string org_id = 1;
   string image_id = 2;

prismnet/crates/prismnet-server/src/config.rs

@@ -20,6 +20,36 @@ pub struct TlsConfig {
     pub require_client_cert: bool,
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
+#[serde(rename_all = "lowercase")]
+pub enum OvnMode {
+    Mock,
+    Real,
+}
+
+impl Default for OvnMode {
+    fn default() -> Self {
+        Self::Mock
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct OvnConfig {
+    #[serde(default)]
+    pub mode: OvnMode,
+    #[serde(default = "default_ovn_nb_addr")]
+    pub nb_addr: String,
+}
+
+impl Default for OvnConfig {
+    fn default() -> Self {
+        Self {
+            mode: OvnMode::default(),
+            nb_addr: default_ovn_nb_addr(),
+        }
+    }
+}
+
 /// Metadata storage backend
 #[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
 #[serde(rename_all = "lowercase")]
@@ -74,6 +104,10 @@ pub struct ServerConfig {
     /// Authentication configuration
     #[serde(default)]
     pub auth: AuthConfig,
+
+    /// OVN integration settings
+    #[serde(default)]
+    pub ovn: OvnConfig,
 }
 
 /// Authentication configuration
@@ -100,6 +134,10 @@ fn default_http_addr() -> SocketAddr {
     "127.0.0.1:8087".parse().unwrap()
 }
 
+fn default_ovn_nb_addr() -> String {
+    "tcp:127.0.0.1:6641".to_string()
+}
+
 impl Default for ServerConfig {
     fn default() -> Self {
         Self {
@@ -113,6 +151,7 @@ impl Default for ServerConfig {
             log_level: "info".to_string(),
             tls: None,
             auth: AuthConfig::default(),
+            ovn: OvnConfig::default(),
         }
     }
 }

prismnet/crates/prismnet-server/src/main.rs

@@ -11,9 +11,8 @@ use prismnet_api::{
     subnet_service_server::SubnetServiceServer, vpc_service_server::VpcServiceServer,
 };
 use prismnet_server::{
-    config::MetadataBackend,
+    config::MetadataBackend, IpamServiceImpl, NetworkMetadataStore, OvnClient, PortServiceImpl,
-    IpamServiceImpl, NetworkMetadataStore, OvnClient, PortServiceImpl, SecurityGroupServiceImpl,
+    SecurityGroupServiceImpl, ServerConfig, SubnetServiceImpl, VpcServiceImpl,
-    ServerConfig, SubnetServiceImpl, VpcServiceImpl,
 };
 use std::net::SocketAddr;
 use std::path::PathBuf;
@@ -36,26 +35,6 @@ struct Args {
     #[arg(long)]
     grpc_addr: Option<String>,
 
-    /// ChainFire endpoint for cluster coordination (optional)
-    #[arg(long)]
-    chainfire_endpoint: Option<String>,
-
-    /// FlareDB endpoint for metadata and tenant data storage
-    #[arg(long)]
-    flaredb_endpoint: Option<String>,
-
-    /// Metadata backend (flaredb, postgres, sqlite)
-    #[arg(long)]
-    metadata_backend: Option<String>,
-
-    /// SQL database URL for metadata (required for postgres/sqlite backend)
-    #[arg(long)]
-    metadata_database_url: Option<String>,
-
-    /// Run in single-node mode (required when metadata backend is SQLite)
-    #[arg(long)]
-    single_node: bool,
-
     /// Log level (overrides config)
     #[arg(short, long)]
     log_level: Option<String>,
@@ -88,58 +67,6 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     if let Some(log_level) = args.log_level {
         config.log_level = log_level;
     }
-    if let Some(chainfire_endpoint) = args.chainfire_endpoint {
-        config.chainfire_endpoint = Some(chainfire_endpoint);
-    }
-    if let Some(flaredb_endpoint) = args.flaredb_endpoint {
-        config.flaredb_endpoint = Some(flaredb_endpoint);
-    }
-    if let Some(metadata_backend) = args.metadata_backend {
-        config.metadata_backend = parse_metadata_backend(&metadata_backend)?;
-    }
-    if let Some(metadata_database_url) = args.metadata_database_url {
-        config.metadata_database_url = Some(metadata_database_url);
-    }
-    if args.single_node {
-        config.single_node = true;
-    }
-    if config.chainfire_endpoint.is_none() {
-        if let Ok(chainfire_endpoint) = std::env::var("PRISMNET_CHAINFIRE_ENDPOINT") {
-            let trimmed = chainfire_endpoint.trim();
-            if !trimmed.is_empty() {
-                config.chainfire_endpoint = Some(trimmed.to_string());
-            }
-        }
-    }
-    if config.flaredb_endpoint.is_none() {
-        if let Ok(flaredb_endpoint) = std::env::var("PRISMNET_FLAREDB_ENDPOINT") {
-            let trimmed = flaredb_endpoint.trim();
-            if !trimmed.is_empty() {
-                config.flaredb_endpoint = Some(trimmed.to_string());
-            }
-        }
-    }
-    if let Ok(metadata_backend) = std::env::var("PRISMNET_METADATA_BACKEND") {
-        let trimmed = metadata_backend.trim();
-        if !trimmed.is_empty() {
-            config.metadata_backend = parse_metadata_backend(trimmed)?;
-        }
-    }
-    if config.metadata_database_url.is_none() {
-        if let Ok(metadata_database_url) = std::env::var("PRISMNET_METADATA_DATABASE_URL") {
-            let trimmed = metadata_database_url.trim();
-            if !trimmed.is_empty() {
-                config.metadata_database_url = Some(trimmed.to_string());
-            }
-        }
-    }
-    if !config.single_node {
-        if let Ok(single_node) = std::env::var("PRISMNET_SINGLE_NODE") {
-            let parsed = single_node.trim().to_ascii_lowercase();
-            config.single_node = matches!(parsed.as_str(), "1" | "true" | "yes" | "on");
-        }
-    }
-
     // Initialize tracing
     tracing_subscriber::fmt()
         .with_env_filter(
@@ -192,12 +119,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
             )
         }
         MetadataBackend::Postgres | MetadataBackend::Sqlite => {
-            let database_url = config
+            let database_url = config.metadata_database_url.as_deref().ok_or_else(|| {
-                .metadata_database_url
-                .as_deref()
-                .ok_or_else(|| {
                 anyhow!(
-                        "metadata_database_url is required when metadata_backend={} (env: PRISMNET_METADATA_DATABASE_URL)",
+                    "metadata_database_url is required when metadata_backend={}",
                     metadata_backend_name(config.metadata_backend)
                 )
             })?;
@@ -216,8 +140,10 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     };
 
     // Initialize OVN client (default: mock)
-    let ovn =
+    let ovn = Arc::new(
-        Arc::new(OvnClient::from_env().map_err(|e| anyhow!("Failed to init OVN client: {}", e))?);
+        OvnClient::from_config(&config.ovn)
+            .map_err(|e| anyhow!("Failed to init OVN client: {}", e))?,
+    );
 
     // Initialize IAM authentication service
     tracing::info!(
@@ -374,19 +300,6 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     Ok(())
 }
 
-fn parse_metadata_backend(value: &str) -> Result<MetadataBackend, Box<dyn std::error::Error>> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "flaredb" => Ok(MetadataBackend::FlareDb),
-        "postgres" => Ok(MetadataBackend::Postgres),
-        "sqlite" => Ok(MetadataBackend::Sqlite),
-        other => Err(format!(
-            "invalid metadata backend '{}'; expected one of: flaredb, postgres, sqlite",
-            other
-        )
-        .into()),
-    }
-}
-
 fn metadata_backend_name(backend: MetadataBackend) -> &'static str {
     match backend {
         MetadataBackend::FlareDb => "flaredb",

prismnet/crates/prismnet-server/src/metadata.rs

@@ -60,12 +60,8 @@ impl NetworkMetadataStore {
         endpoint: Option<String>,
         pd_endpoint: Option<String>,
     ) -> Result<Self> {
-        let endpoint = endpoint.unwrap_or_else(|| {
+        let endpoint = endpoint.unwrap_or_else(|| "127.0.0.1:2479".to_string());
-            std::env::var("PRISMNET_FLAREDB_ENDPOINT")
-                .unwrap_or_else(|_| "127.0.0.1:2479".to_string())
-        });
         let pd_endpoint = pd_endpoint
-            .or_else(|| std::env::var("PRISMNET_CHAINFIRE_ENDPOINT").ok())
             .map(|value| normalize_transport_addr(&value))
             .unwrap_or_else(|| endpoint.clone());
 
@@ -168,7 +164,9 @@ impl NetworkMetadataStore {
         )
         .execute(pool)
         .await
-        .map_err(|e| MetadataError::Storage(format!("Failed to initialize Postgres schema: {}", e)))?;
+        .map_err(|e| {
+            MetadataError::Storage(format!("Failed to initialize Postgres schema: {}", e))
+        })?;
         Ok(())
     }
 
@@ -181,7 +179,9 @@ impl NetworkMetadataStore {
         )
         .execute(pool)
         .await
-        .map_err(|e| MetadataError::Storage(format!("Failed to initialize SQLite schema: {}", e)))?;
+        .map_err(|e| {
+            MetadataError::Storage(format!("Failed to initialize SQLite schema: {}", e))
+        })?;
         Ok(())
     }
 
@@ -208,9 +208,7 @@ impl NetworkMetadataStore {
                     .bind(value)
                     .execute(pool.as_ref())
                     .await
-                    .map_err(|e| {
+                    .map_err(|e| MetadataError::Storage(format!("Postgres put failed: {}", e)))?;
-                        MetadataError::Storage(format!("Postgres put failed: {}", e))
-                    })?;
                 }
                 SqlStorageBackend::Sqlite(pool) => {
                     sqlx::query(

prismnet/crates/prismnet-server/src/ovn/client.rs

@@ -1,9 +1,12 @@
 use std::sync::Arc;
 
-use prismnet_types::{DhcpOptions, Port, PortId, SecurityGroupId, SecurityGroupRule, SecurityGroupRuleId, VpcId};
+use prismnet_types::{
+    DhcpOptions, Port, PortId, SecurityGroupId, SecurityGroupRule, SecurityGroupRuleId, VpcId,
+};
 use tokio::process::Command;
 use tokio::sync::Mutex;
 
+use crate::config::{OvnConfig, OvnMode as ConfigOvnMode};
 use crate::ovn::mock::MockOvnState;
 
 /// OVN client mode
@@ -31,22 +34,19 @@ pub struct OvnClient {
 }
 
 impl OvnClient {
-    /// Build an OVN client from environment variables (default: mock)
+    /// Build an OVN client from configuration (default: mock)
-    /// - PRISMNET_OVN_MODE: "mock" (default) or "real"
+    pub fn from_config(config: &OvnConfig) -> OvnResult<Self> {
-    /// - PRISMNET_OVN_NB_ADDR: ovsdb northbound address (real mode only)
+        match config.mode {
-    pub fn from_env() -> OvnResult<Self> {
+            ConfigOvnMode::Mock => Ok(Self::new_mock()),
-        let mode = std::env::var("PRISMNET_OVN_MODE").unwrap_or_else(|_| "mock".to_string());
+            ConfigOvnMode::Real => {
-        match mode.to_lowercase().as_str() {
+                if config.nb_addr.trim().is_empty() {
-            "mock" => Ok(Self::new_mock()),
+                    Err(OvnError::InvalidArgument(
-            "real" => {
+                        "OVN nb_addr must not be empty in real mode".to_string(),
-                let nb_addr = std::env::var("PRISMNET_OVN_NB_ADDR")
+                    ))
-                    .unwrap_or_else(|_| "tcp:127.0.0.1:6641".to_string());
+                } else {
-                Ok(Self::new_real(nb_addr))
+                    Ok(Self::new_real(config.nb_addr.clone()))
+                }
             }
-            other => Err(OvnError::InvalidArgument(format!(
-                "Unknown OVN mode: {}",
-                other
-            ))),
         }
     }
 
@@ -305,10 +305,7 @@ impl OvnClient {
                 }
 
                 if !options.dns_servers.is_empty() {
-                    opts.push(format!(
+                    opts.push(format!("dns_server={{{}}}", options.dns_servers.join(",")));
-                        "dns_server={{{}}}",
-                        options.dns_servers.join(",")
-                    ));
                 }
 
                 opts.push(format!("lease_time={}", options.lease_time));
@@ -389,7 +386,10 @@ impl OvnClient {
 
                 if !output.status.success() {
                     let stderr = String::from_utf8_lossy(&output.stderr).to_string();
-                    return Err(OvnError::Command(format!("lr-add query failed: {}", stderr)));
+                    return Err(OvnError::Command(format!(
+                        "lr-add query failed: {}",
+                        stderr
+                    )));
                 }
 
                 let uuid = String::from_utf8_lossy(&output.stdout).trim().to_string();
@@ -456,11 +456,7 @@ impl OvnClient {
                 let ls_name = Self::logical_switch_name(switch_id);
 
                 // ovn-nbctl lsp-add <switch> <port-name>
-                self.run_nbctl(vec![
+                self.run_nbctl(vec!["lsp-add".into(), ls_name, switch_port_name.clone()])
-                    "lsp-add".into(),
-                    ls_name,
-                    switch_port_name.clone(),
-                ])
                     .await?;
 
                 // ovn-nbctl lsp-set-type <port> router
@@ -698,10 +694,7 @@ mod tests {
         let client = OvnClient::new_mock();
 
         // Create a router
-        let router_id = client
+        let router_id = client.create_logical_router("test-router").await.unwrap();
-            .create_logical_router("test-router")
-            .await
-            .unwrap();
 
         assert!(!router_id.is_empty());
         assert!(router_id.starts_with("router-"));
@@ -732,10 +725,7 @@ mod tests {
             .unwrap();
 
         // Create router
-        let router_id = client
+        let router_id = client.create_logical_router("test-router").await.unwrap();
-            .create_logical_router("test-router")
-            .await
-            .unwrap();
 
         // Add router port
         let mac = "02:00:00:00:00:01";
@@ -760,10 +750,7 @@ mod tests {
         let client = OvnClient::new_mock();
 
         // Create router
-        let router_id = client
+        let router_id = client.create_logical_router("test-router").await.unwrap();
-            .create_logical_router("test-router")
-            .await
-            .unwrap();
 
         // Configure SNAT
         let external_ip = "203.0.113.10";
@@ -791,10 +778,7 @@ mod tests {
             .unwrap();
 
         // Create router
-        let router_id = client
+        let router_id = client.create_logical_router("test-router").await.unwrap();
-            .create_logical_router("test-router")
-            .await
-            .unwrap();
 
         // Add router port
         let mac = "02:00:00:00:00:01";
@@ -840,10 +824,7 @@ mod tests {
             .unwrap();
 
         // Create router
-        let router_id = client
+        let router_id = client.create_logical_router("test-router").await.unwrap();
-            .create_logical_router("test-router")
-            .await
-            .unwrap();
 
         // Add router ports to both switches
         let port1_id = client
@@ -876,10 +857,7 @@ mod tests {
             .unwrap();
 
         // Step 2: Create router
-        let router_id = client
+        let router_id = client.create_logical_router("vpc-router").await.unwrap();
-            .create_logical_router("vpc-router")
-            .await
-            .unwrap();
 
         // Step 3: Attach router to switch
         let mac = "02:00:00:00:00:01";
@@ -920,10 +898,7 @@ mod tests {
         let client = OvnClient::new_mock();
 
         // Create router
-        let router_id = client
+        let router_id = client.create_logical_router("test-router").await.unwrap();
-            .create_logical_router("test-router")
-            .await
-            .unwrap();
 
         // Configure multiple SNAT rules
         client