From 386d422a1be8f7c28abd6f461944ee609839f95f Mon Sep 17 00:00:00 2001
From: Nick Mathewson <nickm@torproject.org>
Date: Tue, 22 Aug 2023 11:33:27 -0400
Subject: [PATCH] hss: Retry rendezvous circuit attempts.

Closes #1013.
---
 Cargo.lock                                    |  1 +
 crates/tor-hsservice/Cargo.toml               |  1 +
 .../tor-hsservice/src/svc/rend_handshake.rs   | 34 ++++++++++++++-----
 3 files changed, 28 insertions(+), 8 deletions(-)

diff --git a/Cargo.lock b/Cargo.lock
index 7a049cb40..b0ba9114c 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -4639,6 +4639,7 @@ dependencies = [
  "postage",
  "rand 0.8.5",
  "rand_core 0.6.4",
+ "retry-error",
  "safelog",
  "serde",
  "serde_json",
diff --git a/crates/tor-hsservice/Cargo.toml b/crates/tor-hsservice/Cargo.toml
index 3006e60b0..dbd9cd679 100644
--- a/crates/tor-hsservice/Cargo.toml
+++ b/crates/tor-hsservice/Cargo.toml
@@ -35,6 +35,7 @@ once_cell = "1"
 postage = { version = "0.5.0", default-features = false, features = ["futures-traits"] }
 rand = "0.8.5"
 rand_core = "0.6.2"
+retry-error = { version = "0.5.0", path = "../retry-error" }
 safelog = { path = "../safelog", version = "0.3.3" }
 serde = { version = "1.0.103", features = ["derive"] }
 thiserror = "1"
diff --git a/crates/tor-hsservice/src/svc/rend_handshake.rs b/crates/tor-hsservice/src/svc/rend_handshake.rs
index 0673bc04c..9a6018930 100644
--- a/crates/tor-hsservice/src/svc/rend_handshake.rs
+++ b/crates/tor-hsservice/src/svc/rend_handshake.rs
@@ -5,6 +5,7 @@
 use std::sync::Arc;
 
 use futures::{stream::BoxStream, StreamExt as _};
+use retry_error::RetryError;
 use tor_cell::relaycell::{
     hs::intro_payload::{IntroduceHandshakePayload, OnionKey},
     msg::{Introduce2, Rendezvous1},
@@ -59,9 +60,9 @@ pub(crate) enum EstablishSessionError {
     /// Got an onion key with an unrecognized type (not ntor).
     #[error("Received an unsupported type of onion key")]
     UnsupportedOnionKey,
-    /// Encountered an error while trying to build a circuit to the rendezvous point.
+    /// Unable to build a circuit to the rendezvous point.
     #[error("Could not establish circuit to rendezvous point")]
-    RendCirc(#[source] tor_circmgr::Error),
+    RendCirc(#[source] RetryError<tor_circmgr::Error>),
     /// Encountered a failure while trying to add a virtual hop to the circuit.
     #[error("Could not add virtual hop to circuit")]
     VirtualHop(#[source] tor_proto::Error),
@@ -226,13 +227,30 @@ impl IntroRequest {
             )
         };
 
+        let max_n_attempts = netdir.params().hs_service_rendezvous_failures_max;
+        let mut circuit = None;
+        let mut retry_err: RetryError<tor_circmgr::Error> =
+            RetryError::in_attempt_to("Establish a circuit to a rendezvous point");
+
         // Open circuit to rendezvous point.
-        let circuit = hs_pool
-            .get_or_launch_specific(&netdir, HsCircKind::SvcRend, rend_point)
-            .await
-            .map_err(E::RendCirc)?;
-        // TODO HSS: Maybe we should retry a couple of time if the failure is not
-        // the fault of the rend_point?
+        for _attempt in 1..=max_n_attempts.into() {
+            match hs_pool
+                .get_or_launch_specific(&netdir, HsCircKind::SvcRend, rend_point.clone())
+                .await
+            {
+                Ok(circ) => {
+                    circuit = Some(circ);
+                    break;
+                }
+                Err(e) => {
+                    retry_err.push(e);
+                    // Note that we do not sleep on errors: if there is any
+                    // error that will be solved by waiting, it would probably
+                    // require waiting too long to satisfy the client.
+                }
+            }
+        }
+        let circuit = circuit.ok_or_else(|| E::RendCirc(retry_err))?;
 
         // We'll need parameters to extend the virtual hop.
         let params = circparameters_from_netparameters(netdir.params());