Archipelago — open-source initial import

This commit is contained in:
Archipelago
2026-08-12 10:55:50 +00:00
commit b67e1527a2
2068 changed files with 472303 additions and 0 deletions
+263
View File
@@ -0,0 +1,263 @@
//! iroh-blobs swarm provider — the DHT Phase 2 engine, gated behind the
//! `iroh-swarm` feature (heavy QUIC dep tree, off by default).
//!
//! Stands up a real iroh node: binds a QUIC [`Endpoint`], opens a persistent
//! blob [`FsStore`] under `data_dir/iroh-blobs`, and serves blobs over the
//! iroh-blobs protocol — so a node that *fetches* content also *seeds* it
//! afterwards. Content is addressed by BLAKE3 ([`Hash`]) and range-verified by
//! iroh on arrival.
//!
//! This provider is an optimization beneath the origin HTTP path: the [`super`]
//! swarm seam falls back to origin whenever [`try_fetch`](IrohProvider::try_fetch)
//! returns `Ok(false)` (no known seeds) or `Err` (transient swarm failure).
//!
//! ## Discovery boundary (Phase 3)
//! Downloading needs the [`EndpointId`]s of peers that hold the hash. That
//! discovery — design Phase 3, *signed Nostr advertisement events* mapping
//! `{content-hash → provider endpoint}` — is injected via [`ProviderDiscovery`].
//! Until it is wired, discovery yields nothing and every fetch defers to origin,
//! so enabling the feature is safe (never worse than today).
use std::path::Path;
use std::str::FromStr;
use std::sync::Arc;
use anyhow::Result;
use async_trait::async_trait;
use iroh::{endpoint::presets, protocol::Router, Endpoint, EndpointId};
use iroh_blobs::{store::fs::FsStore, BlobsProtocol, Hash};
use super::payment::PaymentPolicy;
use super::BlobProvider;
use crate::content_hash::{ContentDigest, HashAlg};
/// Resolves which peers are believed to hold a given content hash.
///
/// Phase 3 (signed Nostr advertisement events) provides the production impl
/// [`NostrSeedDiscovery`]; `None` discovery means "origin-only" — a safe
/// default. The query is async (it hits relays), so the trait is async.
#[async_trait]
pub trait ProviderDiscovery: Send + Sync {
/// Candidate seed endpoints for `hash` (may be empty).
async fn providers_for(&self, hash: &Hash) -> Vec<EndpointId>;
}
/// Production [`ProviderDiscovery`]: reads signed seed advertisements from Nostr
/// relays and parses the advertised endpoint-id strings into [`EndpointId`]s.
///
/// Unparseable ids are skipped (an advert from an incompatible/garbage peer must
/// not abort discovery). Reuses the node's existing relay list + Tor proxy.
pub struct NostrSeedDiscovery {
relays: Vec<String>,
tor_proxy: Option<String>,
}
impl NostrSeedDiscovery {
pub fn new(relays: Vec<String>, tor_proxy: Option<String>) -> Self {
Self { relays, tor_proxy }
}
}
#[async_trait]
impl ProviderDiscovery for NostrSeedDiscovery {
async fn providers_for(&self, hash: &Hash) -> Vec<EndpointId> {
let hex = hash.to_hex();
let ids = super::seed_advert::fetch_seed_endpoint_ids(
&self.relays,
self.tor_proxy.as_deref(),
&hex,
)
.await;
ids.into_iter()
.filter_map(|s| match EndpointId::from_str(&s) {
Ok(id) => Some(id),
Err(e) => {
tracing::debug!("swarm: skipping unparseable seed endpoint id {s}: {e}");
None
}
})
.collect()
}
}
/// Fetches content-addressed blobs from the iroh swarm, and seeds what it has.
#[allow(dead_code)] // constructed once Phase 3 discovery is wired into providers()
pub struct IrohProvider {
endpoint: Endpoint,
store: FsStore,
/// Kept alive so the node keeps accepting blob-protocol connections (seeds).
_router: Router,
discovery: Option<Arc<dyn ProviderDiscovery>>,
/// Where pricing/session/wallet state lives — for paid-fetch negotiation.
data_dir: std::path::PathBuf,
/// Willingness to pay swarm peers when fetching. Defaults to
/// [`PaymentPolicy::free`]: never pay (releases/catalog stay free), so a
/// seeder that prices a blob is skipped → origin. A future film fetch can
/// pass a real budget.
pay_policy: PaymentPolicy,
}
#[allow(dead_code)]
impl IrohProvider {
/// Bind an iroh endpoint, open the persistent blob store at
/// `data_dir/iroh-blobs`, and start serving blobs (seed capability).
pub async fn new(
data_dir: &Path,
discovery: Option<Arc<dyn ProviderDiscovery>>,
) -> Result<Self> {
let root = data_dir.join("iroh-blobs");
tokio::fs::create_dir_all(&root).await.ok();
let store = FsStore::load(&root)
.await
.map_err(|e| anyhow::anyhow!("open iroh blob store: {e}"))?;
let endpoint = Endpoint::bind(presets::N0)
.await
.map_err(|e| anyhow::anyhow!("bind iroh endpoint: {e}"))?;
// Serve blobs: a node that fetches a blob can then seed it to others.
// The event sender gates each request through the ecash `streaming` layer
// — free by default, paid only if the operator priced `content-download`
// (Networking Profits → Settings). It also hard-disables peer writes.
let event_sender =
super::paid::gated_event_sender(data_dir.to_path_buf(), (*store).clone());
let blobs = BlobsProtocol::new(&store, Some(event_sender));
// Shape-A paid negotiation rides a second ALPN on the same endpoint so a
// downloader can pay (open a session) before the blob-GET above serves it.
let paid =
super::paid_alpn::PaidBlobsProtocol::new(data_dir.to_path_buf(), (*store).clone());
let router = Router::builder(endpoint.clone())
.accept(iroh_blobs::ALPN, blobs)
.accept(super::paid_alpn::PAID_ALPN, paid)
.spawn();
Ok(Self {
endpoint,
store,
_router: router,
discovery,
data_dir: data_dir.to_path_buf(),
pay_policy: PaymentPolicy::free(),
})
}
/// This node's iroh endpoint id — what Phase 3 advertises as a seed address.
pub fn endpoint_id(&self) -> EndpointId {
self.endpoint.id()
}
/// Import a held PUBLIC blob into the seed store and advertise it on Nostr so
/// other nodes can fetch it from us. Call this only for releases/catalog
/// content (the design's privacy scope) — never private user blobs.
///
/// Importing makes us an actual seed: a node that downloaded a release from
/// the HTTP origin can now serve it to peers over iroh-blobs. The advert maps
/// `blake3_hex → this endpoint id`. Defensive check: the bytes we import must
/// hash to what we advertise, so a path/hash mismatch can never publish a lie.
pub async fn seed_and_advertise(
&self,
path: &Path,
blake3_hex: &str,
identity_dir: &Path,
relays: &[String],
tor_proxy: Option<&str>,
) -> Result<()> {
let expected = {
let raw = hex::decode(blake3_hex).map_err(|e| anyhow::anyhow!("blake3 hex: {e}"))?;
let arr: [u8; 32] = raw
.as_slice()
.try_into()
.map_err(|_| anyhow::anyhow!("blake3 digest must be 32 bytes"))?;
Hash::from_bytes(arr)
};
let info = self
.store
.blobs()
.add_path(path)
.await
.map_err(|e| anyhow::anyhow!("import blob into seed store: {e}"))?;
if info.hash != expected {
anyhow::bail!(
"imported blob hash {} != advertised {}",
info.hash.to_hex(),
blake3_hex
);
}
super::seed_advert::publish_seed_advert(
identity_dir,
relays,
tor_proxy,
blake3_hex,
&self.endpoint_id().to_string(),
)
.await
}
}
#[async_trait]
impl BlobProvider for IrohProvider {
fn name(&self) -> &str {
"iroh"
}
async fn try_fetch(&self, digest: &ContentDigest, dest: &Path) -> Result<bool> {
// iroh addresses content by BLAKE3. A sha256-only digest isn't fetchable
// from the swarm — defer to origin.
if digest.alg != HashAlg::Blake3 {
return Ok(false);
}
let raw = hex::decode(&digest.hex).map_err(|e| anyhow::anyhow!("digest hex: {e}"))?;
let arr: [u8; 32] = raw
.as_slice()
.try_into()
.map_err(|_| anyhow::anyhow!("blake3 digest must be 32 bytes"))?;
let hash = Hash::from_bytes(arr);
// Who has it? Without discovery (Phase 3) this is empty → origin wins.
let providers = match &self.discovery {
Some(d) => d.providers_for(&hash).await,
None => Vec::new(),
};
if providers.is_empty() {
return Ok(false);
}
// Shape-A: negotiate paid access with each candidate. Best-effort and
// additive — a peer is dropped only if it explicitly requires a payment
// we won't make under `pay_policy` (free by default → priced seeders are
// skipped). Connect/protocol failures keep the peer; the blob-GET gate is
// the real enforcement and a refused GET still falls back to origin.
let mut allowed = Vec::with_capacity(providers.len());
for peer in providers {
if super::paid_alpn::negotiate_access(
&self.endpoint,
&self.data_dir,
peer,
&digest.hex,
&self.pay_policy,
)
.await
{
allowed.push(peer);
}
}
if allowed.is_empty() {
return Ok(false);
}
// Fetch (range-verified by iroh) then export the verified blob to the
// staging path the caller expects. The seam re-verifies the digest.
let downloader = self.store.downloader(&self.endpoint);
downloader
.download(hash, allowed)
.await
.map_err(|e| anyhow::anyhow!("iroh swarm download: {e}"))?;
self.store
.blobs()
.export(hash, dest)
.await
.map_err(|e| anyhow::anyhow!("export blob to staging: {e}"))?;
Ok(true)
}
}
+377
View File
@@ -0,0 +1,377 @@
//! Swarm-assist content fetch — the *transport & swarm* tier of the DHT
//! distribution plan (`docs/dht-distribution-design.md` §4).
//!
//! ## Guiding principle: swarm-assist, origin ALWAYS wins
//! The peer swarm is an optimization layered *above* a proven HTTP path, never
//! in place of it. A node asks each available [`BlobProvider`] (e.g. an
//! iroh-blobs swarm) for content by its [`ContentDigest`]; the first peer that
//! serves bytes which **verify** against the digest wins. If no provider has it
//! — or the swarm is disabled, or every peer is offline — we fall back to the
//! origin HTTP download, which is the guaranteed source of truth. Worst case is
//! exactly today's behaviour.
//!
//! Peer-sourced bytes are UNTRUSTED, so this module verifies them against the
//! content digest before accepting. Origin bytes run through the caller's
//! existing verification (e.g. the SHA-256 gate in `update.rs`).
//!
//! The actual iroh-blobs provider is gated behind the `iroh-swarm` feature
//! (heavy QUIC dep tree); with the feature off, [`providers`] is empty and
//! every fetch goes straight to origin — byte-for-byte today's path.
use std::path::Path;
use std::sync::{Arc, OnceLock};
use anyhow::Result;
use async_trait::async_trait;
use tracing::{debug, info, warn};
use crate::content_hash::ContentDigest;
pub mod payment;
pub mod seed_advert;
#[cfg(feature = "iroh-swarm")]
pub mod iroh_provider;
#[cfg(feature = "iroh-swarm")]
pub mod paid;
#[cfg(feature = "iroh-swarm")]
pub mod paid_alpn;
/// Which source ultimately served the content.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum FetchSource {
/// A peer in the swarm served (and the bytes verified).
Swarm,
/// The origin HTTP fallback served.
Origin,
}
/// A source that may be able to serve content addressed by its digest.
#[async_trait]
pub trait BlobProvider: Send + Sync {
/// Short name for logging (e.g. "iroh").
fn name(&self) -> &str;
/// Try to fetch the content for `digest` into `dest`.
///
/// * `Ok(true)` — bytes written to `dest` (caller verifies the digest).
/// * `Ok(false)` — this provider does not have the content; try the next.
/// * `Err(_)` — a transient failure; try the next provider.
async fn try_fetch(&self, digest: &ContentDigest, dest: &Path) -> Result<bool>;
}
/// Process-wide swarm runtime, built once at startup by [`init`]. Holding the
/// providers here (rather than rebuilding per download) keeps the iroh endpoint
/// + blob store + protocol router alive for the life of the process, so a node
/// keeps *seeding* between downloads. Empty/inert unless the `iroh-swarm`
/// feature is built AND `swarm_enabled` is set.
struct SwarmRuntime {
providers: Vec<Arc<dyn BlobProvider>>,
/// Context for announcing held public blobs; `None` when seeding is off.
#[cfg(feature = "iroh-swarm")]
announce: Option<AnnounceCtx>,
}
#[cfg(feature = "iroh-swarm")]
struct AnnounceCtx {
iroh: Arc<iroh_provider::IrohProvider>,
relays: Vec<String>,
tor_proxy: Option<String>,
identity_dir: std::path::PathBuf,
}
static RUNTIME: OnceLock<SwarmRuntime> = OnceLock::new();
/// Build the swarm runtime once, at startup. Idempotent: a second call is a
/// no-op (the first registration wins). Safe to call unconditionally — when the
/// `iroh-swarm` feature is absent, or `enabled` is false, it registers an empty
/// runtime so every fetch goes straight to origin (today's path).
///
/// `relays` / `tor_proxy` come from the node's Nostr config and double as the
/// seed-advert transport; `data_dir` hosts the persistent iroh blob store under
/// `data_dir/iroh-blobs` and the node identity under `data_dir/identity`.
pub async fn init(
data_dir: &Path,
relays: &[String],
tor_proxy: Option<&str>,
enabled: bool,
) -> Result<()> {
if RUNTIME.get().is_some() {
return Ok(());
}
#[cfg(not(feature = "iroh-swarm"))]
{
let _ = (data_dir, relays, tor_proxy);
if enabled {
warn!("swarm: swarm_enabled set but binary built without the `iroh-swarm` feature — staying origin-only");
}
let _ = RUNTIME.set(SwarmRuntime {
providers: Vec::new(),
});
return Ok(());
}
#[cfg(feature = "iroh-swarm")]
{
if !enabled {
info!("swarm: disabled (swarm_enabled=false) — origin-only");
let _ = RUNTIME.set(SwarmRuntime {
providers: Vec::new(),
announce: None,
});
return Ok(());
}
let discovery: Arc<dyn iroh_provider::ProviderDiscovery> = Arc::new(
iroh_provider::NostrSeedDiscovery::new(relays.to_vec(), tor_proxy.map(str::to_string)),
);
let provider = Arc::new(iroh_provider::IrohProvider::new(data_dir, Some(discovery)).await?);
info!(
"swarm: iroh provider active (endpoint {}) — swarm-assist enabled, origin always wins",
provider.endpoint_id()
);
let providers: Vec<Arc<dyn BlobProvider>> = vec![provider.clone()];
let _ = RUNTIME.set(SwarmRuntime {
providers,
announce: Some(AnnounceCtx {
iroh: provider,
relays: relays.to_vec(),
tor_proxy: tor_proxy.map(str::to_string),
identity_dir: data_dir.join("identity"),
}),
});
Ok(())
}
}
/// The ordered list of swarm providers to consult before the origin.
///
/// Empty until [`init`] registers a provider (needs the `iroh-swarm` feature +
/// `swarm_enabled`). While empty, [`fetch_content_addressed`] goes straight to
/// origin — byte-for-byte today's path.
pub fn providers() -> Vec<Arc<dyn BlobProvider>> {
RUNTIME
.get()
.map(|r| r.providers.clone())
.unwrap_or_default()
}
/// Announce that this node now holds a PUBLIC release/catalog blob (addressed by
/// `blake3_hex`, bytes at `path`) so peers can fetch it from us: import it into
/// the seed store and publish a signed Nostr advert. Best-effort and inert
/// unless the iroh provider is active — a failure never affects the install.
///
/// **Scope:** call only for releases/catalog content, never private user blobs.
pub async fn announce_held_blob(_blake3_hex: &str, _path: &Path) {
#[cfg(feature = "iroh-swarm")]
{
let Some(rt) = RUNTIME.get() else { return };
let Some(ctx) = rt.announce.as_ref() else {
return;
};
if let Err(e) = ctx
.iroh
.seed_and_advertise(
_path,
_blake3_hex,
&ctx.identity_dir,
&ctx.relays,
ctx.tor_proxy.as_deref(),
)
.await
{
warn!("swarm: failed to announce held blob {_blake3_hex}: {e}");
}
}
}
/// Fetch content-addressed bytes: swarm-assist, origin always wins.
///
/// Tries each provider in order; the first to write bytes that VERIFY against
/// `digest` wins and returns [`FetchSource::Swarm`]. If none succeed, runs
/// `origin` (the guaranteed HTTP fallback) and returns [`FetchSource::Origin`].
/// A node that obtained bytes from the swarm has, by definition, a verified
/// copy it can itself seed afterwards.
pub async fn fetch_content_addressed<F, Fut>(
digest: &ContentDigest,
providers: &[Arc<dyn BlobProvider>],
dest: &Path,
origin: F,
) -> Result<FetchSource>
where
F: FnOnce() -> Fut,
Fut: std::future::Future<Output = Result<()>>,
{
for provider in providers {
match provider.try_fetch(digest, dest).await {
Ok(true) => match verify_dest(digest, dest).await {
Ok(()) => {
info!("swarm: {} served {} (verified)", provider.name(), digest);
return Ok(FetchSource::Swarm);
}
Err(e) => {
// A peer served bytes that don't match the digest — could be
// corruption or a malicious seed. Discard and try the next
// source; never let unverified peer bytes through.
warn!(
"swarm: {} served bytes failing verification for {}: {} — discarding",
provider.name(),
digest,
e
);
let _ = tokio::fs::remove_file(dest).await;
}
},
Ok(false) => debug!("swarm: {} does not have {}", provider.name(), digest),
Err(e) => debug!("swarm: {} failed for {}: {}", provider.name(), digest, e),
}
}
debug!(
"swarm: no provider served {} — falling back to origin",
digest
);
origin().await?;
Ok(FetchSource::Origin)
}
/// Read `dest` and verify it hashes to `digest`.
async fn verify_dest(digest: &ContentDigest, dest: &Path) -> Result<()> {
let bytes = tokio::fs::read(dest).await?;
digest.verify(&bytes)
}
#[cfg(test)]
mod tests {
use super::*;
use std::sync::atomic::{AtomicBool, Ordering};
fn digest_of(bytes: &[u8]) -> ContentDigest {
ContentDigest::parse(&format!(
"blake3:{}",
crate::content_hash::blake3_hex(bytes)
))
.unwrap()
}
/// Provider that writes a fixed payload (which may or may not match).
struct FixedProvider {
name: &'static str,
payload: Option<Vec<u8>>,
}
#[async_trait]
impl BlobProvider for FixedProvider {
fn name(&self) -> &str {
self.name
}
async fn try_fetch(&self, _d: &ContentDigest, dest: &Path) -> Result<bool> {
match &self.payload {
Some(p) => {
tokio::fs::write(dest, p).await?;
Ok(true)
}
None => Ok(false),
}
}
}
fn arc(p: FixedProvider) -> Arc<dyn BlobProvider> {
Arc::new(p)
}
#[tokio::test]
async fn swarm_hit_verifies_and_skips_origin() {
let dir = tempfile::tempdir().unwrap();
let dest = dir.path().join("out");
let content = b"hello swarm".to_vec();
let digest = digest_of(&content);
let providers = vec![arc(FixedProvider {
name: "good",
payload: Some(content.clone()),
})];
let origin_ran = AtomicBool::new(false);
let src = fetch_content_addressed(&digest, &providers, &dest, || async {
origin_ran.store(true, Ordering::SeqCst);
tokio::fs::write(&dest, b"from-origin").await?;
Ok(())
})
.await
.unwrap();
assert_eq!(src, FetchSource::Swarm);
assert!(
!origin_ran.load(Ordering::SeqCst),
"origin must not run on swarm hit"
);
assert_eq!(tokio::fs::read(&dest).await.unwrap(), content);
}
#[tokio::test]
async fn bad_swarm_bytes_are_discarded_and_origin_wins() {
let dir = tempfile::tempdir().unwrap();
let dest = dir.path().join("out");
let content = b"the real bytes".to_vec();
let digest = digest_of(&content);
// Provider claims a hit but serves tampered bytes.
let providers = vec![arc(FixedProvider {
name: "evil",
payload: Some(b"TAMPERED".to_vec()),
})];
let src = fetch_content_addressed(&digest, &providers, &dest, || async {
tokio::fs::write(&dest, &content).await?;
Ok(())
})
.await
.unwrap();
assert_eq!(
src,
FetchSource::Origin,
"tampered swarm bytes must not be accepted"
);
assert_eq!(tokio::fs::read(&dest).await.unwrap(), content);
}
#[tokio::test]
async fn no_providers_goes_straight_to_origin() {
let dir = tempfile::tempdir().unwrap();
let dest = dir.path().join("out");
let content = b"x".to_vec();
let digest = digest_of(&content);
let providers: Vec<Arc<dyn BlobProvider>> = vec![];
let src = fetch_content_addressed(&digest, &providers, &dest, || async {
tokio::fs::write(&dest, &content).await?;
Ok(())
})
.await
.unwrap();
assert_eq!(src, FetchSource::Origin);
}
#[tokio::test]
async fn falls_through_providers_in_order() {
let dir = tempfile::tempdir().unwrap();
let dest = dir.path().join("out");
let content = b"second wins".to_vec();
let digest = digest_of(&content);
let providers = vec![
arc(FixedProvider {
name: "miss",
payload: None,
}),
arc(FixedProvider {
name: "hit",
payload: Some(content.clone()),
}),
];
let src = fetch_content_addressed(&digest, &providers, &dest, || async {
tokio::fs::write(&dest, b"origin").await?;
Ok(())
})
.await
.unwrap();
assert_eq!(src, FetchSource::Swarm);
assert_eq!(tokio::fs::read(&dest).await.unwrap(), content);
}
}
+194
View File
@@ -0,0 +1,194 @@
//! Paid swarm serving — gate the iroh-blobs provider through the ecash
//! `streaming` payment layer (DHT distribution plan, Phase 4 step F).
//!
//! ## Free by default
//! Serving is **free unless the node operator turns it on** in
//! *Networking Profits → Settings* (which enables the `content-download`
//! streaming service). With that service disabled — the shipped default —
//! [`is_authorized`] returns `true` for everyone and behaviour is byte-for-byte
//! the old open seeder. When it is enabled, a peer must hold an active paid
//! session (opened out-of-band via the `streaming.pay` RPC with a Cashu token)
//! before the swarm will serve them; otherwise the request is refused and they
//! fall back to the HTTP origin.
//!
//! ## How it hooks in
//! iroh-blobs 0.103 lets a provider authorize each request: we pass an
//! [`EventSender`] (built here) to `BlobsProtocol::new`, set the [`EventMask`]
//! to intercept connections + GET requests, and answer each one with
//! `Ok(())` (serve) or `Err(AbortReason::Permission)` (refuse). Peer-initiated
//! writes (`push`) are hard-disabled so a peer can never mutate our store.
//!
//! Scope note: today every swarm blob is a public release/app component, so the
//! gate only ever charges if the operator explicitly priced `content-download`.
//! When IndeeHub films land on the same blob layer (Phase 4), they reuse this
//! exact path.
use std::collections::HashMap;
use std::path::{Path, PathBuf};
use iroh::EndpointId;
use iroh_blobs::api::blobs::BlobStatus;
use iroh_blobs::api::Store;
use iroh_blobs::provider::events::{
AbortReason, ConnectMode, EventMask, EventResult, EventSender, ObserveMode, ProviderMessage,
RequestMode, ThrottleMode,
};
use iroh_blobs::Hash;
use crate::streaming::gate::{self, GateResult};
/// The streaming pricing service that meters swarm blob serving. Enabling it in
/// the Settings UI is what flips swarm serving from free to paid.
const SERVICE_ID: &str = "content-download";
/// Build the gated [`EventSender`] for `BlobsProtocol` and spawn the task that
/// authorizes each blob GET through the ecash gate.
///
/// `data_dir` locates the pricing/session state; `store` is cloned in to look up
/// blob sizes for metering. The spawned task lives as long as the provider keeps
/// the returned sender alive (i.e. the life of the node).
pub fn gated_event_sender(data_dir: PathBuf, store: Store) -> EventSender {
// Intercept connections + read requests so we can allow/deny per peer & hash.
// `push` (peer writes into our store) is hard-disabled. `throttle`/`observe`
// stay off — we meter coarsely at request time, not per 16 KiB chunk.
let mask = EventMask {
connected: ConnectMode::Intercept,
get: RequestMode::Intercept,
get_many: RequestMode::Intercept,
push: RequestMode::Disabled,
observe: ObserveMode::None,
throttle: ThrottleMode::None,
};
let (sender, mut rx) = EventSender::channel(64, mask);
tokio::spawn(async move {
// connection_id → remote endpoint id, learned at ClientConnected and used
// to key the paying peer's streaming session on each request.
let mut peers: HashMap<u64, Option<EndpointId>> = HashMap::new();
while let Some(msg) = rx.recv().await {
match msg {
ProviderMessage::ClientConnected(m) => {
peers.insert(m.inner.connection_id, m.inner.endpoint_id);
// Accept the connection; gating happens per request.
let _ = m.tx.send(Ok(())).await;
}
ProviderMessage::ConnectionClosed(m) => {
peers.remove(&m.inner.connection_id);
}
ProviderMessage::GetRequestReceived(m) => {
let peer = peers.get(&m.inner.connection_id).copied().flatten();
let hash = m.inner.request.hash;
let verdict = authorize(&data_dir, &store, peer, &hash).await;
let _ = m.tx.send(verdict).await;
}
ProviderMessage::GetManyRequestReceived(m) => {
let peer = peers.get(&m.inner.connection_id).copied().flatten();
// A get-many is all-or-nothing here: authorize on the first hash.
let verdict = match m.inner.request.hashes.first().copied() {
Some(h) => authorize(&data_dir, &store, peer, &h).await,
None => Ok(()),
};
let _ = m.tx.send(verdict).await;
}
ProviderMessage::PushRequestReceived(m) => {
// Disabled in the mask; refuse defensively if one ever arrives.
let _ = m.tx.send(Err(AbortReason::Permission)).await;
}
// Notify-only variants, observe and throttle: nothing to gate.
_ => {}
}
}
});
sender
}
/// Authorize one blob GET, returning the iroh [`EventResult`]
/// (`Ok(())` = serve, `Err(Permission)` = refuse).
async fn authorize(
data_dir: &Path,
store: &Store,
peer: Option<EndpointId>,
hash: &Hash,
) -> EventResult {
// Cost = full blob size (coarse, request-time metering). If we don't hold the
// complete blob there's nothing to meter — let iroh serve what it can.
let size = match store.blobs().status(*hash).await {
Ok(BlobStatus::Complete { size }) => size,
_ => 0,
};
let peer_id = peer
.map(|e| e.to_string())
.unwrap_or_else(|| "anonymous".to_string());
if is_authorized(data_dir, &peer_id, size).await {
Ok(())
} else {
Err(AbortReason::Permission)
}
}
/// Pure allow/deny decision (no iroh types) — unit-testable without a live node.
async fn is_authorized(data_dir: &Path, peer_id: &str, size: u64) -> bool {
match gate::check_gate(data_dir, peer_id, SERVICE_ID, None, size).await {
// Service disabled (the default) → free for everyone. Or the peer holds an
// active paid session with remaining allotment.
Ok(GateResult::ServiceUnavailable)
| Ok(GateResult::Allowed { .. })
| Ok(GateResult::PaidAndAllowed { .. }) => true,
// Metered + no/exhausted session: the peer must pay out-of-band first
// (streaming.pay) before the swarm serves them — they fall back to origin.
Ok(_) => false,
// Never let a payment-layer fault break content distribution: fail OPEN
// (serve free) and log. Availability beats revenue when something breaks.
Err(e) => {
tracing::warn!("paid-gate: check errored ({e}); serving free");
true
}
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::streaming::pricing::{self, Metric, PricingConfig, ServicePricing};
fn content_download(enabled: bool) -> PricingConfig {
PricingConfig {
services: vec![ServicePricing {
service_id: SERVICE_ID.to_string(),
name: "Content Downloads".to_string(),
metric: Metric::Bytes,
step_size: 1_048_576,
price_per_step: 1,
min_steps: 0,
enabled,
description: String::new(),
accepted_mints: vec![],
}],
}
}
#[tokio::test]
async fn free_when_service_disabled_by_default() {
let dir = tempfile::tempdir().unwrap();
// No pricing file → defaults → content-download disabled → free for all.
assert!(is_authorized(dir.path(), "peer-a", 1_000_000).await);
}
#[tokio::test]
async fn free_when_service_explicitly_disabled() {
let dir = tempfile::tempdir().unwrap();
pricing::save_pricing(dir.path(), &content_download(false))
.await
.unwrap();
assert!(is_authorized(dir.path(), "peer-a", 1_048_576).await);
}
#[tokio::test]
async fn denied_when_metered_and_peer_has_not_paid() {
let dir = tempfile::tempdir().unwrap();
pricing::save_pricing(dir.path(), &content_download(true))
.await
.unwrap();
// Enabled service + no session/token → the swarm refuses; peer uses origin.
assert!(!is_authorized(dir.path(), "peer-b", 1_048_576).await);
}
}
+314
View File
@@ -0,0 +1,314 @@
//! Shape-A paid-blobs negotiation ALPN (`archy/paid-blobs/1`) — the on-wire
//! exchange that lets a downloader pay a seeder *before* fetching a gated blob
//! (DHT distribution plan §1, "shape A"). Gated behind `iroh-swarm`.
//!
//! ## Why a side ALPN
//! iroh-blobs carries the raw bytes; this tiny request/grant protocol rides a
//! second ALPN on the *same* endpoint so a downloader can discover the price and
//! deliver an ecash token first. The token opens a metered `streaming` session
//! keyed by the downloader's endpoint id — exactly the session the blob-GET gate
//! ([`super::paid`]) already checks. Same endpoint → same session → the GET is
//! then served.
//!
//! ```text
//! B ──(archy/paid-blobs/1)──▶ A PaidRequest { want: H, token: None }
//! B ◀─────────────────────── A PaymentRequired { price, accepted_mints }
//! B: auto_pay_token(...) ── builds a cashuA token (cross-mint aware)
//! B ──(archy/paid-blobs/1)──▶ A PaidRequest { want: H, token: Some(t) }
//! B ◀─────────────────────── A Granted (session now exists on A)
//! B ──(iroh-blobs ALPN)─────▶ A GET H → served (gate sees the session)
//! ```
//!
//! ## North star: origin always wins, releases stay free
//! Negotiation is **best-effort and additive**. A peer that doesn't speak this
//! ALPN, or any connect/protocol error, is treated as "proceed" — the blob-GET
//! gate is the real enforcement, and a denied GET just falls back to origin.
//! With the default [`PaymentPolicy::free`] a downloader never sends a token, so
//! a seeder that prices a blob is simply skipped → origin. Only films (a future
//! caller with a real budget) will actually pay.
use std::path::{Path, PathBuf};
use anyhow::Result;
use iroh::endpoint::Connection;
use iroh::protocol::{AcceptError, ProtocolHandler};
use iroh::{Endpoint, EndpointAddr, EndpointId};
use iroh_blobs::api::blobs::BlobStatus;
use iroh_blobs::api::Store;
use iroh_blobs::Hash;
use serde::{Deserialize, Serialize};
use super::payment::PaymentPolicy;
use crate::streaming::gate::{self, GateResult};
/// ALPN for the paid-blobs negotiation protocol.
pub const PAID_ALPN: &[u8] = b"archy/paid-blobs/1";
/// The streaming service that meters swarm blob serving (same id as [`super::paid`]).
const SERVICE_ID: &str = "content-download";
/// Cap on a single negotiation message (JSON). Requests/responses are tiny.
const MAX_MSG: usize = 64 * 1024;
/// A downloader's ask for one content-addressed blob, optionally with payment.
#[derive(Debug, Serialize, Deserialize)]
struct PaidRequest {
/// BLAKE3 hex of the wanted blob.
want: String,
/// A `cashuA` token, present on the paying retry.
#[serde(skip_serializing_if = "Option::is_none")]
token: Option<String>,
}
/// The seeder's verdict.
#[derive(Debug, Serialize, Deserialize)]
#[serde(tag = "status", rename_all = "snake_case")]
enum PaidResponse {
/// Fetch away — free, or a paid session is now active for this peer.
Granted,
/// Payment needed before serving. The downloader may pay and retry.
PaymentRequired {
price_sats: u64,
accepted_mints: Vec<String>,
},
/// Refused (bad request, insufficient/failed payment).
Denied { reason: String },
}
// ── Serve side ─────────────────────────────────────────────────────────────
/// Accept-side handler for [`PAID_ALPN`]. Registered on the provider's `Router`
/// alongside the iroh-blobs protocol.
#[derive(Clone)]
pub struct PaidBlobsProtocol {
data_dir: PathBuf,
store: Store,
}
impl std::fmt::Debug for PaidBlobsProtocol {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("PaidBlobsProtocol").finish()
}
}
impl PaidBlobsProtocol {
pub fn new(data_dir: PathBuf, store: Store) -> Self {
Self { data_dir, store }
}
/// Decide the verdict for a request from `peer`. Mirrors [`super::paid`]'s
/// policy: free when the service is disabled (default) or the peer holds an
/// active session; payment-required when metered and unpaid; fail-OPEN
/// (Granted) on an internal gate error so a fault never blocks distribution.
async fn decide(&self, peer: &str, req: &PaidRequest) -> PaidResponse {
let size = self.blob_size(&req.want).await;
match gate::check_gate(&self.data_dir, peer, SERVICE_ID, req.token.as_deref(), size).await {
Ok(GateResult::ServiceUnavailable)
| Ok(GateResult::Allowed { .. })
| Ok(GateResult::PaidAndAllowed { .. }) => PaidResponse::Granted,
Ok(GateResult::PaymentRequired {
minimum_sats,
pricing,
..
}) => PaidResponse::PaymentRequired {
price_sats: minimum_sats,
accepted_mints: pricing.accepted_mints,
},
Ok(GateResult::InsufficientPayment {
provided_sats,
minimum_sats,
}) => PaidResponse::Denied {
reason: format!("insufficient payment: {provided_sats} < {minimum_sats} sats"),
},
Ok(GateResult::PaymentFailed { reason }) => PaidResponse::Denied { reason },
// Availability beats revenue: a gate fault serves free, matching the
// blob-GET gate's fail-open behaviour.
Err(e) => {
tracing::warn!("paid-alpn: gate errored ({e}); granting free");
PaidResponse::Granted
}
}
}
/// Full size of a held blob (for metering); 0 if we don't hold it complete.
async fn blob_size(&self, blake3_hex: &str) -> u64 {
let Ok(raw) = hex::decode(blake3_hex) else {
return 0;
};
let Ok(arr) = <[u8; 32]>::try_from(raw.as_slice()) else {
return 0;
};
match self.store.blobs().status(Hash::from_bytes(arr)).await {
Ok(BlobStatus::Complete { size }) => size,
_ => 0,
}
}
}
impl ProtocolHandler for PaidBlobsProtocol {
async fn accept(&self, connection: Connection) -> Result<(), AcceptError> {
let peer = connection.remote_id().to_string();
// One bi-stream per request (a paying downloader opens a second one).
loop {
let (mut send, mut recv) = match connection.accept_bi().await {
Ok(s) => s,
// Connection closed by the peer — normal end of negotiation.
Err(_) => break,
};
let buf = recv
.read_to_end(MAX_MSG)
.await
.map_err(AcceptError::from_err)?;
let response = match serde_json::from_slice::<PaidRequest>(&buf) {
Ok(req) => self.decide(&peer, &req).await,
Err(e) => PaidResponse::Denied {
reason: format!("bad request: {e}"),
},
};
let bytes = serde_json::to_vec(&response).map_err(AcceptError::from_err)?;
send.write_all(&bytes)
.await
.map_err(AcceptError::from_err)?;
send.finish().map_err(AcceptError::from_err)?;
}
Ok(())
}
}
// ── Fetch side ───────────────────────────────────────────────────────────────
/// Negotiate access to `blake3_hex` from `peer` before fetching. Returns whether
/// the caller should proceed to download from this peer.
///
/// Best-effort: any connect/protocol failure returns `true` (proceed — the
/// blob-GET gate is the real enforcement, and a denied GET falls back to origin).
/// Returns `false` only when the seeder explicitly requires a payment we won't or
/// can't make under `policy`.
pub async fn negotiate_access(
endpoint: &Endpoint,
data_dir: &Path,
peer: EndpointId,
blake3_hex: &str,
policy: &PaymentPolicy,
) -> bool {
match negotiate_inner(endpoint, data_dir, peer, blake3_hex, policy).await {
Ok(proceed) => proceed,
Err(e) => {
tracing::debug!(
"paid-alpn: negotiation with {peer} failed ({e}) — proceeding (gate decides)"
);
true
}
}
}
async fn negotiate_inner(
endpoint: &Endpoint,
data_dir: &Path,
peer: EndpointId,
blake3_hex: &str,
policy: &PaymentPolicy,
) -> Result<bool> {
let conn = endpoint.connect(EndpointAddr::new(peer), PAID_ALPN).await?;
// First ask with no token.
let resp = exchange(
&conn,
&PaidRequest {
want: blake3_hex.to_string(),
token: None,
},
)
.await?;
match resp {
PaidResponse::Granted => Ok(true),
PaidResponse::Denied { .. } => Ok(false),
PaidResponse::PaymentRequired {
price_sats,
accepted_mints,
} => {
// Build a token within budget (cross-mint aware); None ⇒ use origin.
match super::payment::auto_pay_token(data_dir, policy, &accepted_mints, price_sats)
.await?
{
None => Ok(false),
Some(token) => {
let resp2 = exchange(
&conn,
&PaidRequest {
want: blake3_hex.to_string(),
token: Some(token),
},
)
.await?;
Ok(matches!(resp2, PaidResponse::Granted))
}
}
}
}
}
/// One request/response round trip on a fresh bi-stream.
async fn exchange(conn: &Connection, req: &PaidRequest) -> Result<PaidResponse> {
let (mut send, mut recv) = conn.open_bi().await?;
send.write_all(&serde_json::to_vec(req)?).await?;
send.finish()?;
let buf = recv.read_to_end(MAX_MSG).await?;
Ok(serde_json::from_slice(&buf)?)
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn request_round_trips_and_omits_absent_token() {
let req = PaidRequest {
want: "abcd".into(),
token: None,
};
let json = serde_json::to_string(&req).unwrap();
assert!(
!json.contains("token"),
"absent token must be omitted: {json}"
);
let back: PaidRequest = serde_json::from_str(&json).unwrap();
assert_eq!(back.want, "abcd");
assert!(back.token.is_none());
}
#[test]
fn request_with_token_round_trips() {
let req = PaidRequest {
want: "ff".into(),
token: Some("cashuAbc".into()),
};
let back: PaidRequest =
serde_json::from_str(&serde_json::to_string(&req).unwrap()).unwrap();
assert_eq!(back.token.as_deref(), Some("cashuAbc"));
}
#[test]
fn response_tagged_serialization() {
let granted = serde_json::to_string(&PaidResponse::Granted).unwrap();
assert_eq!(granted, r#"{"status":"granted"}"#);
let pr = serde_json::to_string(&PaidResponse::PaymentRequired {
price_sats: 7,
accepted_mints: vec!["https://m".into()],
})
.unwrap();
let back: PaidResponse = serde_json::from_str(&pr).unwrap();
match back {
PaidResponse::PaymentRequired {
price_sats,
accepted_mints,
} => {
assert_eq!(price_sats, 7);
assert_eq!(accepted_mints, vec!["https://m".to_string()]);
}
other => panic!("expected PaymentRequired, got {other:?}"),
}
}
}
+167
View File
@@ -0,0 +1,167 @@
//! Fetch-side auto-pay — the *downloader's* decision layer for paid swarm
//! content (plan §1 "fetch side" + §2a cross-mint).
//!
//! When a swarm seeder gates a blob behind payment (its `PaymentRequired`
//! advertises a price and a set of `accepted_mints`), a downloading node uses
//! this layer to decide whether to pay and, if so, to build a `cashuA` token
//! denominated in one of the seeder's accepted mints — auto-swapping across
//! mints when needed (see [`crate::wallet::ecash::build_payment_token`]).
//!
//! ## North star: origin always wins
//! Paying is strictly an optimization. If the price is over budget, the wallet
//! can't cover it, no trusted mint is reachable, or a swap would cost too much,
//! this layer returns `None` and the caller falls back to the free HTTP origin —
//! exactly today's path. A wallet/mint problem must never block content.
//!
//! ## Scope / what's NOT here
//! This builds the *token*; it does not yet carry it to the seeder. The on-wire
//! exchange (a downloader presenting the token to a paid seeder, then streaming
//! the blob) is the in-band paid-blobs ALPN — "shape (A)" in the design doc —
//! which is deferred. Today's seeder side (`swarm::paid`) only allow/deny-gates
//! iroh-blobs requests; once shape (A) lands, the provider's fetch path calls
//! [`auto_pay_token`] on a `PaymentRequired` and retries with the token.
use std::path::Path;
use anyhow::Result;
use tracing::debug;
use crate::wallet::ecash;
/// A downloader's willingness to pay swarm peers for a single fetch.
#[derive(Debug, Clone, Copy)]
pub struct PaymentPolicy {
/// Maximum total sats to spend for this content. `0` disables paying
/// entirely (origin-only) — the safe default.
pub budget_sats: u64,
/// Maximum cross-mint swap fee tolerated when we must swap into the
/// seeder's mint. Ignored when we already hold the right mint.
pub max_fee_sats: u64,
}
impl PaymentPolicy {
/// The default: never pay, always use the free origin. The production caller
/// is the deferred in-band paid-blobs ALPN (shape A); used by tests today.
#[allow(dead_code)]
pub fn free() -> Self {
Self {
budget_sats: 0,
max_fee_sats: 0,
}
}
/// A budget-capped policy.
pub fn with_budget(budget_sats: u64, max_fee_sats: u64) -> Self {
Self {
budget_sats,
max_fee_sats,
}
}
/// Whether a seeder's `price_sats` is worth paying under this policy. A zero
/// price is treated as "not a real paid request" (use origin / free path).
pub fn affords(&self, price_sats: u64) -> bool {
price_sats > 0 && price_sats <= self.budget_sats
}
}
/// Decide whether to pay a seeder `price_sats`, and if so build a `cashuA` token
/// denominated in one of its `accepted_mints` (auto-swapping if needed).
///
/// * `Ok(Some(token))` — pay the seeder with this token.
/// * `Ok(None)` — decline (over budget, unpayable, or swap too costly);
/// the caller should fall back to the free origin.
///
/// Never returns `Err` for a wallet/mint problem: those degrade to `Ok(None)`
/// so a payment failure can never block content.
pub async fn auto_pay_token(
data_dir: &Path,
policy: &PaymentPolicy,
accepted_mints: &[String],
price_sats: u64,
) -> Result<Option<String>> {
if !policy.affords(price_sats) {
debug!(
"auto-pay: price {} sats over budget {} (or zero) — using origin",
price_sats, policy.budget_sats
);
return Ok(None);
}
match ecash::build_payment_token(data_dir, accepted_mints, price_sats, policy.max_fee_sats)
.await
{
Ok(token) => Ok(Some(token)),
Err(e) => {
// Unpayable within balance/trust/fee — not an error, just decline.
debug!("auto-pay: declined ({}) — falling back to origin", e);
Ok(None)
}
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn free_policy_never_affords() {
let p = PaymentPolicy::free();
assert!(!p.affords(1));
assert!(!p.affords(0));
}
#[test]
fn budget_policy_affordability() {
let p = PaymentPolicy::with_budget(100, 5);
assert!(p.affords(100)); // exactly at budget
assert!(p.affords(1));
assert!(!p.affords(101)); // over budget
assert!(!p.affords(0)); // zero price is never a real paid request
}
#[tokio::test]
async fn over_budget_declines_without_touching_wallet() {
let tmp = tempfile::tempdir().unwrap();
// Price exceeds budget → None, and no wallet/mint interaction occurs.
let out = auto_pay_token(
tmp.path(),
&PaymentPolicy::with_budget(50, 5),
&["https://seeder.example.com".into()],
100,
)
.await
.unwrap();
assert!(out.is_none());
}
#[tokio::test]
async fn zero_budget_is_origin_only() {
let tmp = tempfile::tempdir().unwrap();
let out = auto_pay_token(
tmp.path(),
&PaymentPolicy::free(),
&["https://seeder.example.com".into()],
10,
)
.await
.unwrap();
assert!(out.is_none());
}
#[tokio::test]
async fn unpayable_within_budget_declines_gracefully() {
let tmp = tempfile::tempdir().unwrap();
// Within budget, but empty wallet + untrusted seeder mint → build fails;
// auto_pay degrades to None (origin) rather than erroring.
let out = auto_pay_token(
tmp.path(),
&PaymentPolicy::with_budget(1000, 10),
&["https://untrusted.example.com".into()],
100,
)
.await
.unwrap();
assert!(out.is_none());
}
}
+233
View File
@@ -0,0 +1,233 @@
//! Phase 3 discovery — signed Nostr "seed advertisement" events.
//!
//! A node that holds a PUBLIC release / app-image blob (addressed by BLAKE3)
//! announces "I can seed hash H from iroh endpoint E" as a signed, NIP-33
//! addressable Nostr event. **Scope is releases/catalog content ONLY** — never
//! private user blobs (decided 2026-06-16): smallest privacy surface, covers
//! the OTA + app-install use-cases. Discovery queries these events to find
//! swarm seeds for a hash; the iroh provider then dials those endpoints.
//!
//! Event shape (NIP-33 addressable, kind [`ARCHIPELAGO_SEED_KIND`]):
//! - `d` tag = blake3 hex of the content → one current advert per (author, hash)
//! - content = `{"v":1,"endpoint_id":"<iroh endpoint id>"}`
//! - author pubkey = the node's seed-derived Nostr identity (signs the event)
//!
//! Endpoint ids stay opaque strings here so this protocol layer builds/parses/
//! publishes/queries WITHOUT the heavy iroh dep; only the `iroh-swarm`
//! discovery glue parses the string into an `iroh::EndpointId`.
// The publish/query path that calls these lives behind `iroh-swarm` (it needs
// the node's iroh EndpointId), so in the default build they're exercised only
// by unit tests — allow them to stand without a production caller.
#![allow(dead_code)]
use std::path::Path;
use std::time::Duration;
use nostr_sdk::{Event, EventBuilder, Filter, Keys, Kind, Tag};
use serde::{Deserialize, Serialize};
/// How long to wait for relay connects / event fetches. Matches the rest of the
/// Nostr discovery path so the swarm never stalls the download longer than node
/// discovery already might.
const RELAY_CONNECT_TIMEOUT: Duration = Duration::from_secs(10);
const RELAY_FETCH_TIMEOUT: Duration = Duration::from_secs(15);
/// NIP-33 addressable kind for Archipelago seed advertisements.
/// Distinct from the node-discovery app-data kind (30078).
pub const ARCHIPELAGO_SEED_KIND: u16 = 30081;
#[derive(Debug, Clone, Serialize, Deserialize)]
struct AdvertContent {
v: u8,
endpoint_id: String,
}
/// Build the (unsigned) advertisement event for `blake3_hex` served from
/// `endpoint_id`. Sign with the node's Nostr key (`.sign_with_keys()` /
/// `.sign()`) or publish via `client.send_event_builder()`.
pub fn advertisement_builder(blake3_hex: &str, endpoint_id: &str) -> EventBuilder {
let content = serde_json::to_string(&AdvertContent {
v: 1,
endpoint_id: endpoint_id.to_string(),
})
.expect("serialize advert content");
EventBuilder::new(Kind::Custom(ARCHIPELAGO_SEED_KIND), content)
.tag(Tag::identifier(blake3_hex.to_string()))
}
/// Filter matching all current seed advertisements for `blake3_hex` (one per
/// advertising node; NIP-33 latest-replaces per author).
pub fn advertisement_filter(blake3_hex: &str) -> Filter {
Filter::new()
.kind(Kind::Custom(ARCHIPELAGO_SEED_KIND))
.identifier(blake3_hex.to_string())
}
/// Extract the advertised endpoint id from an event, or `None` if it is the
/// wrong kind or malformed.
pub fn parse_endpoint_id(event: &Event) -> Option<String> {
if event.kind != Kind::Custom(ARCHIPELAGO_SEED_KIND) {
return None;
}
serde_json::from_str::<AdvertContent>(&event.content)
.ok()
.map(|c| c.endpoint_id)
.filter(|s| !s.is_empty())
}
/// Collect the unique advertised endpoint ids across a set of events, skipping
/// malformed ones. Order-preserving, de-duplicated.
pub fn endpoint_ids_from_events<'a>(events: impl IntoIterator<Item = &'a Event>) -> Vec<String> {
let mut seen = std::collections::HashSet::new();
let mut out = Vec::new();
for ev in events {
if let Some(id) = parse_endpoint_id(ev) {
if seen.insert(id.clone()) {
out.push(id);
}
}
}
out
}
/// Query `relays` for the current seed advertisements for `blake3_hex` and
/// return the de-duplicated endpoint-id strings (opaque here; the `iroh-swarm`
/// glue parses them into `iroh::EndpointId`).
///
/// Best-effort by design: an empty relay list, a connect timeout, or a fetch
/// failure all yield an empty list — never an error. The swarm seam treats "no
/// providers" as "use origin", so discovery problems can only ever degrade to
/// today's HTTP path, never block it.
pub async fn fetch_seed_endpoint_ids(
relays: &[String],
tor_proxy: Option<&str>,
blake3_hex: &str,
) -> Vec<String> {
if relays.is_empty() {
return Vec::new();
}
// Query anonymously — discovery reads public adverts and must not link the
// query back to this node's seed identity.
let anon = Keys::generate();
let client = match crate::nostr_discovery::build_nostr_client(anon, tor_proxy) {
Ok(c) => c,
Err(e) => {
tracing::warn!("seed-advert: build relay client failed: {e}");
return Vec::new();
}
};
for url in relays {
let _ = client.add_relay(url).await;
}
if tokio::time::timeout(RELAY_CONNECT_TIMEOUT, client.connect())
.await
.is_err()
{
tracing::warn!("seed-advert: relay connect timed out, continuing anyway");
}
let events = client
.fetch_events(advertisement_filter(blake3_hex), RELAY_FETCH_TIMEOUT)
.await
.map(|e| e.to_vec())
.unwrap_or_default();
client.disconnect().await;
endpoint_ids_from_events(events.iter())
}
/// Publish a signed advertisement — "this node can seed `blake3_hex` from
/// `endpoint_id`" — to `relays`, signed with the node's seed-derived Nostr key.
///
/// **Caller must restrict this to PUBLIC releases/catalog blobs** (the design's
/// privacy scope, decided 2026-06-16) — never private user content. Best-effort:
/// relay failures are logged, not fatal, since seeding is an optimization.
pub async fn publish_seed_advert(
identity_dir: &Path,
relays: &[String],
tor_proxy: Option<&str>,
blake3_hex: &str,
endpoint_id: &str,
) -> anyhow::Result<()> {
if relays.is_empty() {
return Ok(());
}
let keys = crate::nostr_discovery::load_or_create_nostr_keys(identity_dir).await?;
let client = crate::nostr_discovery::build_nostr_client(keys, tor_proxy)?;
for url in relays {
let _ = client.add_relay(url).await;
}
if tokio::time::timeout(RELAY_CONNECT_TIMEOUT, client.connect())
.await
.is_err()
{
tracing::warn!("seed-advert: publish relay connect timed out, continuing anyway");
}
let _ = client
.send_event_builder(advertisement_builder(blake3_hex, endpoint_id))
.await;
client.disconnect().await;
tracing::info!("seed-advert: announced {blake3_hex} seedable from {endpoint_id}");
Ok(())
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn build_sign_parse_roundtrip() {
let keys = Keys::generate();
let hash = "a".repeat(64);
let endpoint = "node-example-endpoint-id";
let event = advertisement_builder(&hash, endpoint)
.sign_with_keys(&keys)
.unwrap();
assert_eq!(event.kind, Kind::Custom(ARCHIPELAGO_SEED_KIND));
assert_eq!(parse_endpoint_id(&event).as_deref(), Some(endpoint));
}
#[test]
fn filter_targets_the_hash_dtag_and_kind() {
let hash = "b".repeat(64);
let json = serde_json::to_string(&advertisement_filter(&hash)).unwrap();
assert!(json.contains(&hash), "filter must target the hash d-tag");
assert!(
json.contains("30081"),
"filter must constrain the seed kind"
);
}
#[test]
fn parse_rejects_wrong_kind_and_empty_endpoint() {
let keys = Keys::generate();
let wrong_kind = EventBuilder::new(Kind::Custom(1), "{}")
.sign_with_keys(&keys)
.unwrap();
assert_eq!(parse_endpoint_id(&wrong_kind), None);
let empty_endpoint = advertisement_builder(&"c".repeat(64), "")
.sign_with_keys(&keys)
.unwrap();
assert_eq!(parse_endpoint_id(&empty_endpoint), None);
}
#[test]
fn dedups_endpoint_ids_across_events() {
let a = Keys::generate();
let b = Keys::generate();
let hash = "d".repeat(64);
let e1 = advertisement_builder(&hash, "endpoint-A")
.sign_with_keys(&a)
.unwrap();
let e2 = advertisement_builder(&hash, "endpoint-A")
.sign_with_keys(&b)
.unwrap();
let e3 = advertisement_builder(&hash, "endpoint-B")
.sign_with_keys(&b)
.unwrap();
let ids = endpoint_ids_from_events([&e1, &e2, &e3]);
assert_eq!(
ids,
vec!["endpoint-A".to_string(), "endpoint-B".to_string()]
);
}
}