fix(torrents): close late Aria2 cleanup races

This commit is contained in:
NimBold
2026-07-30 23:30:51 +03:30
parent 292447d417
commit 2c1bd9cdf1
4 changed files with 310 additions and 44 deletions
+87 -9
View File
@@ -4884,16 +4884,22 @@ async fn remove_download(
) -> Result<(), String> {
log::info!("remove_download called for id: {}", id);
let preserve_resumable = preserve_resumable.unwrap_or(false);
let mut owned_paths = crate::download_ownership::owned_paths_for_id(&app_handle, &id)?;
let _control_guard = state.queue_manager.acquire_aria2_control(&id).await;
let control_guard = state.queue_manager.acquire_aria2_control(&id).await;
let active_kind = state.queue_manager.active_kind(&id).await;
let media_lifecycle_generation = state
let registered_lifecycle_generation = state
.queue_manager
.registered_lifecycle_generation(&id)
.await
.unwrap_or_default();
state.queue_manager.remove_from_pending(&id).await;
.await;
let media_lifecycle_generation = registered_lifecycle_generation.unwrap_or_default();
if let Some(generation) = registered_lifecycle_generation {
state
.queue_manager
.remove_from_pending_for_generation(&id, generation)
.await;
} else {
state.queue_manager.remove_from_pending(&id).await;
}
let gid = state.queue_manager.aria2_gid_for_download(&id);
if let Some(gid) = gid.as_deref() {
@@ -4909,7 +4915,6 @@ async fn remove_download(
error
);
}
owned_paths = crate::download_ownership::owned_paths_for_id(&app_handle, &id)?;
// Keep the current epoch and mapping alive until daemon removal is
// confirmed. If removal fails, terminal events from this lifecycle
// must still be accepted so the permit and mapping can be cleaned up.
@@ -4942,7 +4947,7 @@ async fn remove_download(
// There may be an addUri or retry handoff in flight with no mapped
// GID yet. Invalidate that pending lifecycle before acknowledging the
// removal so its late result cannot resurrect the download.
state.queue_manager.next_aria2_control_epoch(&id).await;
let removal_epoch = state.queue_manager.next_aria2_control_epoch(&id).await;
state.queue_manager.cancel_aria2_retries(&id).await;
let (tx, rx) = tokio::sync::oneshot::channel();
if matches!(active_kind, Some(crate::queue::TaskKind::Media)) {
@@ -4961,11 +4966,79 @@ async fn remove_download(
.await;
}
// The initial addUri call intentionally runs outside the control lock.
// Do not delete the guessed magnet path and ownership record until a
// late GID has been removed; otherwise the daemon can finish creating
// an output after cleanup and leave an untracked file behind.
drop(control_guard);
state.queue_manager.wait_for_aria2_dispatch(&id).await;
let _control_guard = state.queue_manager.acquire_aria2_control(&id).await;
let current_generation = state
.queue_manager
.registered_lifecycle_generation(&id)
.await;
let lifecycle_changed = registered_lifecycle_generation.is_some()
&& current_generation != registered_lifecycle_generation
|| registered_lifecycle_generation.is_none() && current_generation.is_some();
if lifecycle_changed {
state
.queue_manager
.release_permit_for_generation(&id, media_lifecycle_generation)
.await;
return Err("download lifecycle changed while waiting for aria2 dispatch".to_string());
}
if let Some(late_gid) = state.queue_manager.aria2_gid_for_download(&id) {
if let Err(error) = state
.queue_manager
.reconcile_aria2_torrent_ownership(&id, &late_gid)
.await
{
log::debug!(
"aria2 remove [{}]: could not resolve files for late gid {}: {}",
id,
late_gid,
error
);
}
let removal_result = async {
force_remove_aria2_gid(
state.aria2_port.load(std::sync::atomic::Ordering::Relaxed),
&state.aria2_secret,
&late_gid,
)
.await?;
wait_for_aria2_stopped(
state.aria2_port.load(std::sync::atomic::Ordering::Relaxed),
&state.aria2_secret,
&late_gid,
)
.await
}
.await;
if let Err(error) = removal_result {
state.queue_manager.allow_aria2_retries(&id).await;
return Err(error);
}
state.queue_manager.next_aria2_control_epoch(&id).await;
state.queue_manager.clear_aria2_retry_state(&id).await;
state.queue_manager.forget_aria2_gid(&id).await;
} else if !state
.queue_manager
.is_aria2_control_epoch_current(&id, removal_epoch)
.await
{
// A same-generation pause/resume control action may advance the
// epoch while the late add is being retired. Removal still owns
// the registered row, so fence that action before cleaning it.
state.queue_manager.next_aria2_control_epoch(&id).await;
}
state.queue_manager.release_permit(&id).await;
state.queue_manager.clear_aria2_retry_state(&id).await;
state.queue_manager.forget_aria2_gid(&id).await;
}
let owned_paths = crate::download_ownership::owned_paths_for_id(&app_handle, &id)?;
use tauri::Emitter;
let _ = app_handle.emit(
"download-state",
@@ -5545,6 +5618,10 @@ async fn validate_torrent_enqueue(
let bytes = std::fs::read(path)
.map_err(|error| format!("could not read cached torrent metadata: {error}"))?;
let metadata = crate::torrent::parse_torrent_bytes(&bytes)?;
crate::torrent::validate_info_hash(
item.torrent_info_hash.as_deref(),
&metadata.info_hash,
)?;
crate::torrent::validate_selected_indices(
item.torrent_file_indices.as_deref(),
metadata.files.len(),
@@ -5559,7 +5636,8 @@ async fn validate_torrent_enqueue(
if item.torrent_file_indices.is_some() {
return Err("magnet file selection requires resolved torrent metadata".to_string());
}
crate::torrent::inspect_source(&item.url).map(|_| ())
let metadata = crate::torrent::inspect_source(&item.url)?;
crate::torrent::validate_info_hash(item.torrent_info_hash.as_deref(), &metadata.info_hash)
}
fn expected_torrent_output_paths(
+140 -23
View File
@@ -298,6 +298,11 @@ pub struct QueueManager<R: tauri::Runtime = tauri::Wry> {
/// download id -> spawn payload for aria2 transient-error re-addUri retries.
aria2_payloads: Mutex<HashMap<String, SpawnPayload>>,
/// Initial aria2 addUri handoffs that have not yet either published a GID
/// or removed a stale late GID. Removal waits for these handoffs before
/// deleting owned assets so a magnet cannot leave an orphaned output.
aria2_dispatch_inflight: Mutex<HashMap<String, HashSet<u64>>>,
aria2_dispatch_notify: Notify,
/// The daemon-wide download cap currently applied to aria2. This mirrors
/// successful RPC changes so the poller can avoid treating an intentional
@@ -376,6 +381,8 @@ impl<R: tauri::Runtime> QueueManager<R> {
aria2_gids: Arc::new(std::sync::RwLock::new(HashMap::new())),
pending_completion: Arc::new(Mutex::new(HashMap::new())),
aria2_payloads: Mutex::new(HashMap::new()),
aria2_dispatch_inflight: Mutex::new(HashMap::new()),
aria2_dispatch_notify: Notify::new(),
aria2_global_speed_limit: Arc::new(StdMutex::new(None)),
aria2_retry_strikes: Mutex::new(HashMap::new()),
aria2_retry_cancelled: Mutex::new(HashSet::new()),
@@ -1203,7 +1210,7 @@ impl<R: tauri::Runtime> QueueManager<R> {
}
}
async fn release_permit_for_generation(&self, id: &str, generation: u64) {
pub(crate) async fn release_permit_for_generation(&self, id: &str, generation: u64) {
let _admission_gate = self.admission_gate.lock().await;
let removed = {
let mut permits = self.active_permits.lock().await;
@@ -1392,6 +1399,9 @@ impl<R: tauri::Runtime> QueueManager<R> {
} else {
None
};
if let Some(epoch) = aria2_lifecycle_epoch {
self.begin_aria2_dispatch(&id, epoch).await;
}
self.emit_state(&id, DownloadStatus::Downloading);
drop(control_guard);
@@ -1399,7 +1409,8 @@ impl<R: tauri::Runtime> QueueManager<R> {
TaskKind::Aria2 => {
let lifecycle_epoch = aria2_lifecycle_epoch
.expect("aria2 dispatch must initialize a control epoch");
match self.spawner.add_uri(&id, &task.payload).await {
let add_result = self.spawner.add_uri(&id, &task.payload).await;
match add_result {
Ok(gid) => {
let control_guard = self.acquire_aria2_control(&id).await;
let cancelled = self.aria2_retry_cancelled.lock().await.contains(&id);
@@ -1414,6 +1425,23 @@ impl<R: tauri::Runtime> QueueManager<R> {
id,
gid
);
if task.payload.is_torrent {
if let Err(error) = self
.reconcile_aria2_torrent_ownership_for_payload(
&id,
&gid,
&task.payload,
)
.await
{
log::debug!(
"aria2 dispatch cancellation [{}]: could not resolve torrent output paths for gid {}: {}",
id,
gid,
error
);
}
}
if let Err(error) = self.spawner.remove_uri(&gid).await {
log::warn!(
"aria2 dispatch cancellation [{}]: failed to remove late gid {}: {}",
@@ -1423,6 +1451,7 @@ impl<R: tauri::Runtime> QueueManager<R> {
);
}
self.ignore_aria2_gid(&gid).await;
self.finish_aria2_dispatch(&id, lifecycle_epoch).await;
if current_lifecycle {
self.clear_aria2_retry_state(&id).await;
self.release_permit(&id).await;
@@ -1430,6 +1459,7 @@ impl<R: tauri::Runtime> QueueManager<R> {
return;
}
let buffered_outcome = self.remember_gid(id.clone(), gid.clone()).await;
self.finish_aria2_dispatch(&id, lifecycle_epoch).await;
drop(control_guard);
if let Some(outcome) = buffered_outcome {
self.handle_aria2_event(&gid, outcome).await;
@@ -1454,6 +1484,7 @@ impl<R: tauri::Runtime> QueueManager<R> {
id
);
}
self.finish_aria2_dispatch(&id, lifecycle_epoch).await;
}
}
}
@@ -1755,23 +1786,55 @@ impl<R: tauri::Runtime> QueueManager<R> {
.find_map(|(gid, mapping)| (mapping.id == id).then(|| gid.clone()))
}
/// Refresh ownership from Aria2's resolved file list before a torrent
/// lifecycle is forgotten. Magnet metadata is not available when the row
/// is enqueued, so the user-provided display name is not a safe ownership
/// path until Aria2 reports the actual files.
pub async fn reconcile_aria2_torrent_ownership(
async fn begin_aria2_dispatch(&self, id: &str, epoch: u64) {
self.aria2_dispatch_inflight
.lock()
.await
.entry(id.to_string())
.or_default()
.insert(epoch);
}
async fn finish_aria2_dispatch(&self, id: &str, epoch: u64) {
let mut inflight = self.aria2_dispatch_inflight.lock().await;
let removed = if let Some(epochs) = inflight.get_mut(id) {
let removed = epochs.remove(&epoch);
if epochs.is_empty() {
inflight.remove(id);
}
removed
} else {
false
};
drop(inflight);
if removed {
self.aria2_dispatch_notify.notify_waiters();
}
}
pub async fn wait_for_aria2_dispatch(&self, id: &str) {
loop {
let notified = self.aria2_dispatch_notify.notified();
tokio::pin!(notified);
notified.as_mut().enable();
if !self
.aria2_dispatch_inflight
.lock()
.await
.contains_key(id)
{
return;
}
notified.await;
}
}
async fn reconcile_aria2_torrent_ownership_for_payload(
&self,
id: &str,
gid: &str,
payload: &SpawnPayload,
) -> Result<(), String> {
let payload = self.aria2_payloads.lock().await.get(id).cloned();
let Some(payload) = payload.filter(|payload| payload.is_torrent) else {
return Ok(());
};
let mapping = self
.aria2_gid_mapping(gid)
.filter(|mapping| mapping.id == id)
.ok_or_else(|| "aria2 torrent ownership has no current gid mapping".to_string())?;
let state = self.app_handle.state::<crate::AppState>();
let result = crate::rpc_call(
state.aria2_port.load(std::sync::atomic::Ordering::Relaxed),
@@ -1780,14 +1843,15 @@ impl<R: tauri::Runtime> QueueManager<R> {
serde_json::json!([gid]),
)
.await?;
if !self.is_current_aria2_gid_mapping(gid, &mapping)
|| !self
.is_aria2_control_epoch_current(id, mapping.epoch)
.await
{
return Err("aria2 torrent lifecycle changed while resolving output paths".to_string());
}
self.persist_aria2_torrent_ownership(id, payload, result)
}
fn persist_aria2_torrent_ownership(
&self,
id: &str,
payload: &SpawnPayload,
result: serde_json::Value,
) -> Result<(), String> {
let paths = result
.as_array()
.into_iter()
@@ -1841,6 +1905,41 @@ impl<R: tauri::Runtime> QueueManager<R> {
)
}
/// Refresh ownership from Aria2's resolved file list before a torrent
/// lifecycle is forgotten. Magnet metadata is not available when the row
/// is enqueued, so the user-provided display name is not a safe ownership
/// path until Aria2 reports the actual files.
pub async fn reconcile_aria2_torrent_ownership(
&self,
id: &str,
gid: &str,
) -> Result<(), String> {
let payload = self.aria2_payloads.lock().await.get(id).cloned();
let Some(payload) = payload.filter(|payload| payload.is_torrent) else {
return Ok(());
};
let mapping = self
.aria2_gid_mapping(gid)
.filter(|mapping| mapping.id == id)
.ok_or_else(|| "aria2 torrent ownership has no current gid mapping".to_string())?;
let state = self.app_handle.state::<crate::AppState>();
let result = crate::rpc_call(
state.aria2_port.load(std::sync::atomic::Ordering::Relaxed),
&state.aria2_secret,
"aria2.getFiles",
serde_json::json!([gid]),
)
.await?;
if !self.is_current_aria2_gid_mapping(gid, &mapping)
|| !self
.is_aria2_control_epoch_current(id, mapping.epoch)
.await
{
return Err("aria2 torrent lifecycle changed while resolving output paths".to_string());
}
self.persist_aria2_torrent_ownership(id, &payload, result)
}
/// Capture the GID ownership token used by the poller. The mapping's
/// epoch must still be current when an asynchronous status snapshot is
/// emitted; otherwise a late response from an older GID can be attributed
@@ -2467,6 +2566,17 @@ impl<R: tauri::Runtime> QueueManager<R> {
}
removed
}
pub async fn remove_from_pending_for_generation(&self, id: &str, generation: u64) -> bool {
let mut pending = self.pending.lock().await;
let before = pending.len();
pending.retain(|task| !(task.id == id && task.lifecycle_generation == generation));
let removed = pending.len() < before;
if removed {
self.notify.notify_one();
}
removed
}
}
fn automatic_retry_limit(max_tries: Option<i32>) -> usize {
@@ -2950,7 +3060,14 @@ impl SidecarSpawner for ProductionSpawner {
)
.await?;
match result.as_str() {
Some(returned_gid) if returned_gid == gid => Ok(()),
Some(returned_gid) if returned_gid == gid => {
crate::wait_for_aria2_stopped(
state.aria2_port.load(std::sync::atomic::Ordering::Relaxed),
&state.aria2_secret,
gid,
)
.await
}
Some(returned_gid) => Err(format!(
"aria2.forceRemove returned unexpected gid {returned_gid}, expected {gid}"
)),
+80 -11
View File
@@ -292,6 +292,55 @@ fn parse_info(info: &BencodeValue) -> Result<ParsedTorrent, String> {
Ok(ParsedTorrent { name, total_bytes, files, info_hash })
}
fn canonical_btih(value: &str) -> Option<String> {
let normalized = value.trim().to_ascii_lowercase();
if normalized.len() == 40 && normalized.bytes().all(|byte| byte.is_ascii_hexdigit()) {
return Some(normalized);
}
if normalized.len() != 32 {
return None;
}
let mut decoded = Vec::with_capacity(20);
let mut accumulator = 0u64;
let mut bits = 0u8;
for byte in normalized.bytes() {
let value = match byte {
b'a'..=b'z' => byte - b'a',
b'2'..=b'7' => byte - b'2' + 26,
_ => return None,
} as u64;
accumulator = (accumulator << 5) | value;
bits += 5;
while bits >= 8 {
bits -= 8;
decoded.push(((accumulator >> bits) & 0xff) as u8);
}
if bits == 0 {
accumulator = 0;
} else {
accumulator &= (1u64 << bits) - 1;
}
}
if bits != 0 || decoded.len() != 20 {
return None;
}
Some(decoded.iter().map(|byte| format!("{byte:02x}")).collect())
}
pub fn validate_info_hash(expected: Option<&str>, actual: &str) -> Result<(), String> {
let Some(expected) = expected else {
return Ok(());
};
let expected = canonical_btih(expected)
.ok_or_else(|| "torrent metadata has an invalid expected info hash".to_string())?;
if expected != actual {
return Err("torrent metadata changed before it was queued".to_string());
}
Ok(())
}
pub fn parse_torrent_bytes(bytes: &[u8]) -> Result<ParsedTorrent, String> {
if bytes.is_empty() || bytes.len() > MAX_TORRENT_BYTES {
return Err(format!("torrent metadata must be between 1 byte and {MAX_TORRENT_BYTES} bytes"));
@@ -319,17 +368,7 @@ fn magnet_metadata(source: &str) -> Result<ParsedTorrent, String> {
return None;
}
let value = value.strip_prefix("urn:btih:")?;
let normalized = value.trim().to_ascii_lowercase();
if (normalized.len() == 40 && normalized.bytes().all(|byte| byte.is_ascii_hexdigit()))
|| (normalized.len() == 32
&& normalized
.bytes()
.all(|byte| byte.is_ascii_lowercase() || matches!(byte, b'2'..=b'7')))
{
Some(normalized)
} else {
None
}
canonical_btih(value)
})
.ok_or_else(|| "magnet URI has no valid BitTorrent info hash".to_string())?;
let name = parsed
@@ -542,6 +581,30 @@ mod tests {
assert!(parsed.files.is_empty());
}
#[test]
fn canonicalizes_base32_magnet_hashes_to_hex() {
let parsed = inspect_source(
"magnet:?xt=urn:btih:AERUKZ4JVPG66AJDIVTYTK6N54ASGRLH&dn=Base32",
)
.expect("a valid Base32 hash should parse");
assert_eq!(parsed.info_hash, "0123456789abcdef0123456789abcdef01234567");
}
#[test]
fn validates_expected_hashes_across_hex_and_base32_encodings() {
validate_info_hash(
Some("AERUKZ4JVPG66AJDIVTYTK6N54ASGRLH"),
"0123456789abcdef0123456789abcdef01234567",
)
.expect("equivalent Base32 and hexadecimal hashes should match");
assert!(validate_info_hash(
Some("0123456789abcdef0123456789abcdef01234567"),
"fedcba9876543210fedcba9876543210fedcba98",
)
.is_err());
validate_info_hash(None, "not-used").expect("missing legacy identity should remain compatible");
}
#[test]
fn rejects_malformed_base32_magnet_hashes() {
let error = inspect_source(
@@ -549,6 +612,12 @@ mod tests {
)
.expect_err("a 32-character hash with non-base32 characters must be rejected");
assert!(error.contains("valid BitTorrent info hash"));
let error = inspect_source(
"magnet:?xt=urn:btih:00000000000000000000000000000000&dn=Invalid",
)
.expect_err("characters outside RFC 4648 Base32 must be rejected");
assert!(error.contains("valid BitTorrent info hash"));
}
#[test]
+3 -1
View File
@@ -2043,7 +2043,9 @@ async fn late_aria2_gid_after_cancellation_is_removed_without_leaking_permit() {
manager.release_registered_id("late").await;
manager.release_permit("late").await;
tokio::time::sleep(Duration::from_millis(100)).await;
timeout(Duration::from_secs(1), manager.wait_for_aria2_dispatch("late"))
.await
.expect("late dispatch must be fully removed before it is considered finished");
assert!(manager.aria2_gid_for_download("late").is_none());
assert_eq!(manager.available_permits(), 1);