1use crate::certs::{CertDirectoryLoadOptions, load_all_certs_from_directory};
16use crate::config::TlsReloadOptions;
17use crate::error::TlsRuntimeError;
18use crate::material::{ServerTlsMaterial, server_material_fingerprint};
19use crate::metrics::{record_tls_generation, record_tls_publication_fail, record_tls_reload_result, record_tls_reload_skipped};
20use crate::source::TlsSource;
21use rustls::pki_types::{CertificateDer, PrivateKeyDer};
22use rustls::server::{ClientHello, ResolvesServerCert, ResolvesServerCertUsingSni};
23use rustls::sign::CertifiedKey;
24use std::collections::HashMap;
25use std::io;
26use std::sync::atomic::{AtomicU64, Ordering};
27use std::sync::{Arc, RwLock};
28use tokio::sync::watch;
29use tokio::task::JoinHandle;
30use tokio::time::MissedTickBehavior;
31use tracing::{debug, info, warn};
32
33#[derive(Debug)]
34struct ResolverState {
35 cert_resolver: ResolvesServerCertUsingSni,
36 default_cert: Option<Arc<CertifiedKey>>,
37 cert_count: usize,
38 fingerprint: crate::fingerprint::TlsFingerprint,
39}
40
41impl ResolverState {
42 fn load_from_source(source: &TlsSource) -> Result<Self, TlsRuntimeError> {
43 let base_dir = source.validate_directory()?;
44 let cert_key_pairs = load_all_certs_from_directory(
45 CertDirectoryLoadOptions::builder(base_dir, &source.layout.server_cert_filename, &source.layout.server_key_filename)
46 .build(),
47 )?;
48 if cert_key_pairs.is_empty() {
49 return Err(TlsRuntimeError::Material("No valid certificates found in directory".to_string()));
50 }
51
52 Self::from_cert_key_pairs(cert_key_pairs)
53 }
54
55 fn from_cert_key_pairs(
56 cert_key_pairs: HashMap<String, (Vec<CertificateDer<'static>>, PrivateKeyDer<'static>)>,
57 ) -> Result<Self, TlsRuntimeError> {
58 let cert_count = cert_key_pairs.len();
59 let mut cert_resolver = ResolvesServerCertUsingSni::new();
60 let mut default_cert = None;
61 let mut entries = cert_key_pairs.into_iter().collect::<Vec<_>>();
62 entries.sort_by(|(left_domain, _), (right_domain, _)| left_domain.cmp(right_domain));
63 let material = ServerTlsMaterial::MultiCert {
64 cert_key_pairs: entries
65 .iter()
66 .map(|(domain, (certs, key))| (domain.clone(), (certs.clone(), key.clone_key())))
67 .collect(),
68 };
69 let fingerprint = server_material_fingerprint(&material);
70
71 for (domain, (certs, key)) in entries {
72 let signing_key = rustls::crypto::aws_lc_rs::sign::any_supported_type(&key)
73 .map_err(|e| io::Error::other(format!("unsupported private key type for {domain}: {e:?}")))?;
74 let certified_key = CertifiedKey::new(certs, signing_key);
75
76 if domain.as_str() == "default" {
77 default_cert = Some(Arc::new(certified_key.clone()));
78 } else {
79 cert_resolver
80 .add(&domain, certified_key)
81 .map_err(|e| io::Error::other(format!("failed to add certificate for {domain}: {e:?}")))?;
82 }
83 }
84
85 Ok(Self {
86 cert_resolver,
87 default_cert,
88 cert_count,
89 fingerprint,
90 })
91 }
92}
93
94#[derive(Debug)]
95pub struct ReloadableServerCertResolver {
96 source: TlsSource,
97 current: RwLock<ResolverState>,
98 generation: AtomicU64,
99}
100
101impl ReloadableServerCertResolver {
102 pub fn load_from_source(source: TlsSource) -> Result<Arc<Self>, TlsRuntimeError> {
103 let state = ResolverState::load_from_source(&source)?;
104 record_tls_generation("server_resolver", 1);
105 Ok(Arc::new(Self {
106 source,
107 current: RwLock::new(state),
108 generation: AtomicU64::new(1),
109 }))
110 }
111
112 pub fn load_from_directory(cert_dir: &str) -> Result<Arc<Self>, TlsRuntimeError> {
113 Self::load_from_source(TlsSource::from_directory(cert_dir))
114 }
115
116 pub fn reload(&self) -> Result<Option<usize>, TlsRuntimeError> {
117 let new_state = ResolverState::load_from_source(&self.source)?;
118
119 match self.current.write() {
120 Ok(mut guard) => {
121 if guard.fingerprint == new_state.fingerprint {
122 return Ok(None);
123 }
124 let cert_count = new_state.cert_count;
125 *guard = new_state;
126 self.generation.fetch_add(1, Ordering::Relaxed);
127 Ok(Some(cert_count))
128 }
129 Err(poisoned) => {
130 let mut guard = poisoned.into_inner();
131 if guard.fingerprint == new_state.fingerprint {
132 return Ok(None);
133 }
134 let cert_count = new_state.cert_count;
135 *guard = new_state;
136 self.generation.fetch_add(1, Ordering::Relaxed);
137 Ok(Some(cert_count))
138 }
139 }
140 }
141
142 pub fn generation(&self) -> u64 {
143 self.generation.load(Ordering::Relaxed)
144 }
145}
146
147impl ResolvesServerCert for ReloadableServerCertResolver {
148 fn resolve(&self, client_hello: ClientHello<'_>) -> Option<Arc<CertifiedKey>> {
149 let guard = match self.current.read() {
150 Ok(guard) => guard,
151 Err(poisoned) => poisoned.into_inner(),
152 };
153
154 guard
155 .cert_resolver
156 .resolve(client_hello)
157 .or_else(|| guard.default_cert.clone())
158 }
159}
160
161pub fn spawn_server_cert_reload_loop(
162 protocol: &'static str,
163 resolver: Arc<ReloadableServerCertResolver>,
164 options: TlsReloadOptions,
165 mut shutdown_rx: watch::Receiver<bool>,
166) -> Option<JoinHandle<()>> {
167 if !options.enabled {
168 debug!(protocol, "TLS certificate hot reload is disabled");
169 return None;
170 }
171
172 info!(
173 protocol,
174 cert_dir = %resolver.source.base_dir.display(),
175 "TLS certificate hot reload enabled, checking every {}s",
176 options.interval.as_secs()
177 );
178
179 Some(tokio::spawn(async move {
180 let mut interval = tokio::time::interval(options.interval);
181 interval.set_missed_tick_behavior(MissedTickBehavior::Delay);
182 interval.tick().await;
183
184 loop {
185 tokio::select! {
186 changed = shutdown_rx.changed() => {
187 match changed {
188 Ok(()) => {
189 if *shutdown_rx.borrow() {
190 info!(protocol, cert_dir = %resolver.source.base_dir.display(), "TLS certificate hot reload task stopped");
191 break;
192 }
193 continue;
194 }
195 Err(_) => {
196 info!(
197 protocol,
198 cert_dir = %resolver.source.base_dir.display(),
199 "TLS certificate hot reload task stopped because the shutdown channel closed"
200 );
201 break;
202 }
203 }
204 }
205 _ = interval.tick() => {}
206 }
207
208 match resolver.reload() {
209 Ok(Some(cert_count)) => {
210 record_tls_reload_result(protocol, "ok", None, Some(resolver.generation()));
211 info!(
212 protocol,
213 cert_dir = %resolver.source.base_dir.display(),
214 cert_count,
215 "TLS certificates reloaded successfully"
216 );
217 }
218 Ok(None) => {
219 record_tls_reload_skipped(protocol, "unchanged");
220 debug!(
221 protocol,
222 cert_dir = %resolver.source.base_dir.display(),
223 "TLS certificate material unchanged; skipping reload"
224 );
225 }
226 Err(e) => {
227 record_tls_publication_fail(protocol);
228 warn!(
229 protocol,
230 cert_dir = %resolver.source.base_dir.display(),
231 "TLS certificate reload failed (will retry): {}",
232 e
233 );
234 }
235 }
236 }
237 }))
238}
239
240#[cfg(test)]
241mod tests {
242 use super::*;
243 use rcgen::generate_simple_self_signed;
244 use std::fs;
245 use tempfile::TempDir;
246
247 fn cert_key_pair(san: &str) -> (Vec<CertificateDer<'static>>, PrivateKeyDer<'static>) {
248 let cert = generate_simple_self_signed(vec![san.to_string()]).expect("cert should generate");
249 (
250 vec![cert.cert.der().clone()],
251 PrivateKeyDer::try_from(cert.signing_key.serialize_der()).expect("key should convert"),
252 )
253 }
254
255 fn clone_cert_key_pair(
256 cert_key_pair: &(Vec<CertificateDer<'static>>, PrivateKeyDer<'static>),
257 ) -> (Vec<CertificateDer<'static>>, PrivateKeyDer<'static>) {
258 (cert_key_pair.0.clone(), cert_key_pair.1.clone_key())
259 }
260
261 fn write_default_cert(dir: &std::path::Path, san: &str) {
262 let cert = generate_simple_self_signed(vec![san.to_string()]).expect("cert should generate");
263 fs::write(dir.join(rustfs_config::RUSTFS_TLS_CERT), cert.cert.pem()).expect("cert should write");
264 fs::write(dir.join(rustfs_config::RUSTFS_TLS_KEY), cert.signing_key.serialize_pem()).expect("key should write");
265 }
266
267 #[test]
268 fn reload_replaces_default_certificate() {
269 let temp_dir = TempDir::new().expect("tempdir should create");
270 write_default_cert(temp_dir.path(), "localhost");
271
272 let resolver = ReloadableServerCertResolver::load_from_directory(temp_dir.path().to_str().expect("path should utf8"))
273 .expect("resolver should load");
274 let before = {
275 let guard = resolver.current.read().expect("lock should acquire");
276 guard.default_cert.as_ref().expect("default cert should exist").clone()
277 };
278
279 write_default_cert(temp_dir.path(), "rotated.local");
280
281 let cert_count = resolver.reload().expect("reload should succeed");
282 assert_eq!(cert_count, Some(1));
283
284 let after = {
285 let guard = resolver.current.read().expect("lock should acquire");
286 guard.default_cert.as_ref().expect("default cert should exist").clone()
287 };
288
289 assert_ne!(before.cert[0].as_ref(), after.cert[0].as_ref());
290 }
291
292 #[test]
293 fn reload_skips_when_material_is_unchanged() {
294 let temp_dir = TempDir::new().expect("tempdir should create");
295 write_default_cert(temp_dir.path(), "localhost");
296
297 let resolver = ReloadableServerCertResolver::load_from_directory(temp_dir.path().to_str().expect("path should utf8"))
298 .expect("resolver should load");
299 let outcome = resolver.reload().expect("reload should succeed");
300 assert_eq!(outcome, None);
301 }
302
303 #[test]
304 fn resolver_state_fingerprint_is_stable_across_domain_ordering() {
305 let default_cert = cert_key_pair("localhost");
306 let api_cert = cert_key_pair("api.example.com");
307 let web_cert = cert_key_pair("web.example.com");
308
309 let mut first = HashMap::new();
310 first.insert("default".to_string(), clone_cert_key_pair(&default_cert));
311 first.insert("api.example.com".to_string(), clone_cert_key_pair(&api_cert));
312 first.insert("web.example.com".to_string(), clone_cert_key_pair(&web_cert));
313
314 let mut second = HashMap::new();
315 second.insert("web.example.com".to_string(), clone_cert_key_pair(&web_cert));
316 second.insert("default".to_string(), clone_cert_key_pair(&default_cert));
317 second.insert("api.example.com".to_string(), clone_cert_key_pair(&api_cert));
318
319 let first_state = ResolverState::from_cert_key_pairs(first).expect("first state should build");
320 let second_state = ResolverState::from_cert_key_pairs(second).expect("second state should build");
321
322 assert_eq!(first_state.cert_count, 3);
323 assert_eq!(second_state.cert_count, 3);
324 assert_eq!(first_state.fingerprint, second_state.fingerprint);
325 }
326}