librespot/discovery/src/lib.rs

544 lines
18 KiB
Rust
Raw Normal View History

2021-02-28 23:10:13 +00:00
//! Advertises this device to Spotify clients in the local network.
//!
//! This device will show up in the list of "available devices".
//! Once it is selected from the list, [`Credentials`] are received.
//! Those can be used to establish a new Session with [`librespot_core`].
//!
//! This library uses mDNS and DNS-SD so that other devices can find it,
//! and spawns an http server to answer requests of Spotify clients.
2021-02-28 23:10:13 +00:00
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
mod avahi;
2021-02-28 23:10:13 +00:00
mod server;
2022-01-06 08:48:11 +00:00
use std::{
borrow::Cow,
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
error::Error as StdError,
2022-01-06 08:48:11 +00:00
pin::Pin,
task::{Context, Poll},
};
2021-02-28 23:10:13 +00:00
use futures_core::Stream;
use thiserror::Error;
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
use tokio::sync::{mpsc, oneshot};
2021-02-28 23:10:13 +00:00
use self::server::DiscoveryServer;
2022-01-06 08:48:11 +00:00
pub use crate::core::Error;
use librespot_core as core;
2021-02-28 23:10:13 +00:00
/// Credentials to be used in [`librespot`](`librespot_core`).
pub use crate::core::authentication::Credentials;
/// Determining the icon in the list of available devices.
2022-01-06 08:48:11 +00:00
pub use crate::core::config::DeviceType;
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
pub enum DiscoveryEvent {
Credentials(Credentials),
ServerError(DiscoveryError),
ZeroconfError(DiscoveryError),
}
enum ZeroconfCmd {
Shutdown,
}
pub struct DnsSdHandle {
task_handle: tokio::task::JoinHandle<()>,
shutdown_tx: oneshot::Sender<ZeroconfCmd>,
}
impl DnsSdHandle {
async fn shutdown(self) {
log::debug!("Shutting down zeroconf responder");
let Self {
task_handle,
shutdown_tx,
} = self;
if shutdown_tx.send(ZeroconfCmd::Shutdown).is_err() {
log::warn!("Zeroconf responder unexpectedly disappeared");
} else {
let _ = task_handle.await;
log::debug!("Zeroconf responder stopped");
}
}
}
pub type DnsSdServiceBuilder = fn(
Cow<'static, str>,
Vec<std::net::IpAddr>,
u16,
mpsc::UnboundedSender<DiscoveryEvent>,
) -> Result<DnsSdHandle, Error>;
// Default goes first: This matches the behaviour when feature flags were exlusive, i.e. when there
// was only `feature = "with-dns-sd"` or `not(feature = "with-dns-sd")`
pub const BACKENDS: &[(
&str,
// If None, the backend is known but wasn't compiled.
Option<DnsSdServiceBuilder>,
)] = &[
#[cfg(feature = "with-avahi")]
("avahi", Some(launch_avahi)),
#[cfg(not(feature = "with-avahi"))]
("avahi", None),
#[cfg(feature = "with-dns-sd")]
("dns-sd", Some(launch_dns_sd)),
#[cfg(not(feature = "with-dns-sd"))]
("dns-sd", None),
#[cfg(feature = "with-libmdns")]
("libmdns", Some(launch_libmdns)),
#[cfg(not(feature = "with-libmdns"))]
("libmdns", None),
];
pub fn find(name: Option<&str>) -> Result<DnsSdServiceBuilder, Error> {
if let Some(ref name) = name {
match BACKENDS.iter().find(|(id, _)| name == id) {
Some((_id, Some(launch_svc))) => Ok(*launch_svc),
Some((_id, None)) => Err(Error::unavailable(format!(
"librespot built without '{}' support",
name
))),
None => Err(Error::not_found(format!(
"unknown zeroconf backend '{}'",
name
))),
}
} else {
BACKENDS
.iter()
.find_map(|(_, launch_svc)| *launch_svc)
.ok_or(Error::unavailable(
"librespot built without zeroconf backends",
))
}
}
2021-02-28 23:10:13 +00:00
/// Makes this device visible to Spotify clients in the local network.
///
/// `Discovery` implements the [`Stream`] trait. Every time this device
/// is selected in the list of available devices, it yields [`Credentials`].
pub struct Discovery {
server: DiscoveryServer,
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
/// An opaque handle to the DNS-SD service. Dropping this will unregister the service.
#[allow(unused)]
svc: DnsSdHandle,
event_rx: mpsc::UnboundedReceiver<DiscoveryEvent>,
2021-02-28 23:10:13 +00:00
}
/// A builder for [`Discovery`].
pub struct Builder {
server_config: server::Config,
port: u16,
zeroconf_ip: Vec<std::net::IpAddr>,
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
zeroconf_backend: Option<DnsSdServiceBuilder>,
2021-02-28 23:10:13 +00:00
}
/// Errors that can occur while setting up a [`Discovery`] instance.
#[derive(Debug, Error)]
pub enum DiscoveryError {
#[error("Creating SHA1 block cipher failed")]
AesError(#[from] aes::cipher::InvalidLength),
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
#[error("Setting up dns-sd failed: {0}")]
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
DnsSdError(#[source] Box<dyn StdError + Send + Sync>),
#[error("Creating SHA1 HMAC failed for base key {0:?}")]
HmacError(Vec<u8>),
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
#[error("Setting up the HTTP server failed: {0}")]
HttpServerError(#[from] hyper::Error),
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
#[error("Missing params for key {0}")]
ParamsError(&'static str),
}
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
#[cfg(feature = "with-avahi")]
impl From<zbus::Error> for DiscoveryError {
fn from(error: zbus::Error) -> Self {
Self::DnsSdError(Box::new(error))
}
}
impl From<DiscoveryError> for Error {
fn from(err: DiscoveryError) -> Self {
match err {
DiscoveryError::AesError(_) => Error::unavailable(err),
DiscoveryError::DnsSdError(_) => Error::unavailable(err),
DiscoveryError::HmacError(_) => Error::invalid_argument(err),
DiscoveryError::HttpServerError(_) => Error::unavailable(err),
DiscoveryError::ParamsError(_) => Error::invalid_argument(err),
}
}
}
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
#[allow(unused)]
const DNS_SD_SERVICE_NAME: &str = "_spotify-connect._tcp";
#[allow(unused)]
const TXT_RECORD: [&str; 2] = ["VERSION=1.0", "CPath=/"];
#[cfg(feature = "with-avahi")]
async fn avahi_task(
name: Cow<'static, str>,
port: u16,
entry_group: &mut Option<avahi::EntryGroupProxy<'_>>,
) -> Result<(), DiscoveryError> {
use self::avahi::{EntryGroupState, ServerProxy};
use futures_util::StreamExt;
let conn = zbus::Connection::system().await?;
// Wait for the daemon to show up.
// On error: Failed to listen for NameOwnerChanged signal => Fatal DBus issue
let bus = zbus::fdo::DBusProxy::new(&conn).await?;
let mut stream = bus
.receive_name_owner_changed_with_args(&[(0, "org.freedesktop.Avahi")])
.await?;
loop {
// Wait for Avahi daemon to be started
'wait_avahi: {
while let Poll::Ready(Some(_)) = futures_util::poll!(stream.next()) {
// Drain queued name owner changes, since we're going to connect in a second
}
// Ping after we connected to the signal since it might have shown up in the meantime
if let Ok(avahi_peer) =
zbus::fdo::PeerProxy::new(&conn, "org.freedesktop.Avahi", "/").await
{
if avahi_peer.ping().await.is_ok() {
log::debug!("Pinged Avahi: Available");
break 'wait_avahi;
}
}
log::warn!("Failed to connect to Avahi, zeroconf discovery will not work until avahi-daemon is started. Check that it is installed and running");
// If it didn't, wait for the signal
match stream.next().await {
Some(_signal) => {
log::debug!("Avahi appeared");
break 'wait_avahi;
}
// The stream ended, but this should never happen
None => {
return Err(zbus::Error::Failure("DBus disappeared".to_owned()).into());
}
}
}
// Connect to Avahi and publish the service
let avahi_server = ServerProxy::new(&conn).await?;
log::trace!("Connected to Avahi");
*entry_group = Some(avahi_server.entry_group_new().await?);
let mut entry_group_state_stream = entry_group
.as_mut()
.unwrap()
.receive_state_changed()
.await?;
entry_group
.as_mut()
.unwrap()
.add_service(
-1, // AVAHI_IF_UNSPEC
-1, // IPv4 and IPv6
0, // flags
&name,
DNS_SD_SERVICE_NAME, // type
"", // domain: let the server choose
"", // host: let the server choose
port,
&TXT_RECORD.map(|s| s.as_bytes()),
)
.await?;
entry_group.as_mut().unwrap().commit().await?;
log::debug!("Commited zeroconf service with name {}", &name);
'monitor_service: loop {
tokio::select! {
Some(state_changed) = entry_group_state_stream.next() => {
let (state, error) = match state_changed.args() {
Ok(sc) => (sc.state, sc.error),
Err(e) => {
log::warn!("Error on receiving EntryGroup state from Avahi: {}", e);
continue 'monitor_service;
}
};
match state {
EntryGroupState::Uncommited | EntryGroupState::Registering => {
// Not yet registered, ignore.
}
EntryGroupState::Established => {
log::info!("Published zeroconf service");
}
EntryGroupState::Collision => {
// This most likely means that librespot has unintentionally been started twice.
// Thus, don't retry with a new name, but abort.
//
// Note that the error would usually already be returned by
// entry_group.add_service above, so this state_changed handler
// won't be hit.
//
// EntryGroup has been withdrawn at this point already!
log::error!("zeroconf collision for name '{}'", &name);
return Err(zbus::Error::Failure(format!("zeroconf collision for name: {}", name)).into());
}
EntryGroupState::Failure => {
// TODO: Back off/treat as fatal?
// EntryGroup has been withdrawn at this point already!
// There seems to be no code in Avahi that actually sets this state.
log::error!("zeroconf failure: {}", error);
return Err(zbus::Error::Failure(format!("zeroconf failure: {}", error)).into());
}
}
}
_name_owner_change = stream.next() => {
break 'monitor_service;
}
}
}
// Avahi disappeared (or the service was immediately taken over by a
// new daemon) => drop all handles, and reconnect
log::info!("Avahi disappeared, trying to reconnect");
}
}
#[cfg(feature = "with-avahi")]
fn launch_avahi(
name: Cow<'static, str>,
_zeroconf_ip: Vec<std::net::IpAddr>,
port: u16,
status_tx: mpsc::UnboundedSender<DiscoveryEvent>,
) -> Result<DnsSdHandle, Error> {
let (shutdown_tx, shutdown_rx) = oneshot::channel();
let task_handle = tokio::spawn(async move {
let mut entry_group = None;
tokio::select! {
res = avahi_task(name, port, &mut entry_group) => {
if let Err(e) = res {
log::error!("Avahi error: {}", e);
let _ = status_tx.send(DiscoveryEvent::ZeroconfError(e));
}
},
_ = shutdown_rx => {
if let Some(entry_group) = entry_group.as_mut() {
if let Err(e) = entry_group.free().await {
log::warn!("Failed to un-publish zeroconf service: {}", e);
} else {
log::debug!("Un-published zeroconf service");
}
}
},
}
});
Ok(DnsSdHandle {
task_handle,
shutdown_tx,
})
}
#[cfg(feature = "with-dns-sd")]
fn launch_dns_sd(
name: Cow<'static, str>,
_zeroconf_ip: Vec<std::net::IpAddr>,
port: u16,
status_tx: mpsc::UnboundedSender<DiscoveryEvent>,
) -> Result<DnsSdHandle, Error> {
let (shutdown_tx, shutdown_rx) = oneshot::channel();
let task_handle = tokio::task::spawn_blocking(move || {
let inner = move || -> Result<(), DiscoveryError> {
let svc = dns_sd::DNSService::register(
Some(name.as_ref()),
DNS_SD_SERVICE_NAME,
None,
None,
port,
&TXT_RECORD,
)
.map_err(|e| DiscoveryError::DnsSdError(Box::new(e)))?;
let _ = shutdown_rx.blocking_recv();
std::mem::drop(svc);
Ok(())
};
if let Err(e) = inner() {
log::error!("dns_sd error: {}", e);
let _ = status_tx.send(DiscoveryEvent::ZeroconfError(e));
}
});
Ok(DnsSdHandle {
shutdown_tx,
task_handle,
})
}
#[cfg(feature = "with-libmdns")]
fn launch_libmdns(
name: Cow<'static, str>,
zeroconf_ip: Vec<std::net::IpAddr>,
port: u16,
status_tx: mpsc::UnboundedSender<DiscoveryEvent>,
) -> Result<DnsSdHandle, Error> {
let (shutdown_tx, shutdown_rx) = oneshot::channel();
let task_handle = tokio::task::spawn_blocking(move || {
let inner = move || -> Result<(), DiscoveryError> {
let svc = if !zeroconf_ip.is_empty() {
libmdns::Responder::spawn_with_ip_list(
&tokio::runtime::Handle::current(),
zeroconf_ip,
)
} else {
libmdns::Responder::spawn(&tokio::runtime::Handle::current())
}
.map_err(|e| DiscoveryError::DnsSdError(Box::new(e)))
.unwrap()
.register(
DNS_SD_SERVICE_NAME.to_owned(),
name.into_owned(),
port,
&TXT_RECORD,
);
let _ = shutdown_rx.blocking_recv();
std::mem::drop(svc);
Ok(())
};
if let Err(e) = inner() {
log::error!("libmdns error: {}", e);
let _ = status_tx.send(DiscoveryEvent::ZeroconfError(e));
}
});
Ok(DnsSdHandle {
shutdown_tx,
task_handle,
})
}
2021-02-28 23:10:13 +00:00
impl Builder {
/// Starts a new builder using the provided device and client IDs.
pub fn new<T: Into<String>>(device_id: T, client_id: T) -> Self {
2021-02-28 23:10:13 +00:00
Self {
server_config: server::Config {
name: "Librespot".into(),
device_type: DeviceType::default(),
is_group: false,
device_id: device_id.into(),
client_id: client_id.into(),
2021-02-28 23:10:13 +00:00
},
port: 0,
zeroconf_ip: vec![],
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
zeroconf_backend: None,
2021-02-28 23:10:13 +00:00
}
}
/// Sets the name to be displayed. Default is `"Librespot"`.
pub fn name(mut self, name: impl Into<Cow<'static, str>>) -> Self {
2021-02-28 23:10:13 +00:00
self.server_config.name = name.into();
self
}
/// Sets the device type which is visible as icon in other Spotify clients. Default is `Speaker`.
pub fn device_type(mut self, device_type: DeviceType) -> Self {
self.server_config.device_type = device_type;
self
}
/// Sets whether the device is a group. This affects the icon in Spotify clients. Default is `false`.
pub fn is_group(mut self, is_group: bool) -> Self {
self.server_config.is_group = is_group;
self
}
/// Set the ip addresses on which it should listen to incoming connections. The default is all interfaces.
pub fn zeroconf_ip(mut self, zeroconf_ip: Vec<std::net::IpAddr>) -> Self {
self.zeroconf_ip = zeroconf_ip;
self
}
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
/// Set the zeroconf (MDNS and DNS-SD) implementation to use.
pub fn zeroconf_backend(mut self, zeroconf_backend: DnsSdServiceBuilder) -> Self {
self.zeroconf_backend = Some(zeroconf_backend);
self
}
2021-02-28 23:10:13 +00:00
/// Sets the port on which it should listen to incoming connections.
/// The default value `0` means any port.
pub fn port(mut self, port: u16) -> Self {
self.port = port;
self
}
/// Sets up the [`Discovery`] instance.
///
/// # Errors
/// If setting up the mdns service or creating the server fails, this function returns an error.
pub fn launch(self) -> Result<Discovery, Error> {
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
let name = self.server_config.name.clone();
let zeroconf_ip = self.zeroconf_ip;
2022-03-20 03:12:24 +00:00
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
let (event_tx, event_rx) = mpsc::unbounded_channel();
let mut port = self.port;
let server = DiscoveryServer::new(self.server_config, &mut port, event_tx.clone())?;
2021-02-28 23:10:13 +00:00
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
let launch_svc = self.zeroconf_backend.unwrap_or(find(None)?);
let svc = launch_svc(name, zeroconf_ip, port, event_tx)?;
Ok(Discovery {
server,
svc,
event_rx,
})
}
}
impl Discovery {
/// Starts a [`Builder`] with the provided device id.
pub fn builder<T: Into<String>>(device_id: T, client_id: T) -> Builder {
Builder::new(device_id, client_id)
}
/// Create a new instance with the specified device id and default paramaters.
pub fn new<T: Into<String>>(device_id: T, client_id: T) -> Result<Self, Error> {
Self::builder(device_id, client_id).launch()
2021-02-28 23:10:13 +00:00
}
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
pub async fn shutdown(self) {
tokio::join!(self.server.shutdown(), self.svc.shutdown(),);
}
2021-02-28 23:10:13 +00:00
}
impl Stream for Discovery {
type Item = Credentials;
fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
Discovery: Refactor and add Avahi DBus backend (#1347) * discovery: use opaque error type for DnsSdError This helps to decouple discovery and core by not leaking implementation details of the zeroconf backend into Error conversion impls in core. * discovery: map all MDNS/DNS-SD errors to DiscoveryError::DnsSdError previously, libmdns errors would use a generic conversion from std::io::Error to core::Error * discovery: use an opaque type for the handle to the DNS-SD service * discovery: make features additive i.e. add with-libmdns instead of using not(with-dns-sd). The logic is such that enabling with-dns-sd in addition to the default with-libmdns will still end up using dns-sd, as before. If only with-libmdns is enabled, that will be the default. If none of the features is enabled, attempting to build a `Discovery` will yield an error. * discovery: add --zeroconf-backend CLI flag * discovery: Add minimal Avahi zeroconf backend * bump MSRV to 1.75 required by zbus >= 4 * discovery: ensure that server and dns-sd backend shutdown gracefully Previously, on drop the the shutdown_tx/close_tx, it wasn't guaranteed the corresponding tasks would continue to be polled until they actually completed their shutdown. Since dns_sd::Service is not Send and non-async, and because libmdns is non-async, put them on their own threads. * discovery: use a shared channel for server and zeroconf status messages * discovery: add Avahi reconnection logic This deals gracefully with the case where the Avahi daemon is restarted or not running initially. * discovery: allow running when compiled without zeroconf backend... ...but exit with an error if there's no way to authenticate * better error messages for invalid options with no short flag
2024-10-26 14:45:02 +00:00
match Pin::new(&mut self.event_rx).poll_recv(cx) {
// Yields credentials
Poll::Ready(Some(DiscoveryEvent::Credentials(creds))) => Poll::Ready(Some(creds)),
// Also terminate the stream on fatal server or MDNS/DNS-SD errors.
Poll::Ready(Some(
DiscoveryEvent::ServerError(_) | DiscoveryEvent::ZeroconfError(_),
)) => Poll::Ready(None),
Poll::Ready(None) => Poll::Ready(None),
Poll::Pending => Poll::Pending,
}
2021-02-28 23:10:13 +00:00
}
}