Add runtime Tor-egress self-check (fail-closed) for Nostr relays

Runtime-verified whether Apple's URLSession honors connectionProxyDictionary
SOCKS settings for Nostr relay traffic (plain HTTPS + URLSessionWebSocketTask),
since the app routes all Nostr traffic through Arti's local SOCKS5 proxy solely
via those keys and Apple does not officially support SOCKS for URLSession on iOS.

Verification harness (scripts/tor-egress-verification/): a minimal SOCKS5 CONNECT
proxy that logs arriving connections, plus a Swift probe that runs an HTTPS GET
and a WebSocket ping through a URLSession configured with the proxy dict. Result:
on macOS (kCFNetworkProxiesSOCKS* constants) and the iOS simulator (raw
"SOCKSProxy"/"SOCKSPort" string keys, the exact app path) BOTH HTTP and WebSocket
are proxied, do remote DNS through the proxy, and the proxied session is
fail-closed (every request errors when the SOCKS proxy is down). The feared
direct-egress leak did not reproduce on the tested platforms.

Defense-in-depth for the platform Apple does not guarantee (physical iOS):
add TorEgressVerifier, which performs a canary request through the proxied
session and asserts the exit is a Tor node (check.torproject.org IsTor==true),
positively detecting a direct egress even if the OS silently ignored the proxy.
Policy: verifiedTor allows (cached for a TTL); notTor refuses (leak detected,
never open relays); unreachable allows-with-warning (session stays fail-closed
by construction). Wired into TorManager.awaitEgressReady() and required by both
NostrRelayManager and GeoRelayDirectory before opening connections. Cache is
invalidated on Tor restart/dormant/shutdown. Never falls back to a direct
connection.

Probe is injected so the policy/caching is unit-tested offline; the live-network
harness lives under scripts/ and is not run by CI.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

bitchat/Nostr/GeoRelayDirectory.swift

@@ -54,7 +54,7 @@ private extension GeoRelayDirectoryDependencies {
             refreshCheckInterval: TransportConfig.geoRelayRefreshCheckIntervalSeconds,
             retryInitialSeconds: TransportConfig.geoRelayRetryInitialSeconds,
             retryMaxSeconds: TransportConfig.geoRelayRetryMaxSeconds,
-            awaitTorReady: { await TorManager.shared.awaitReady() },
+            awaitTorReady: { await TorManager.shared.awaitEgressReady() },
             makeFetchData: {
                 let session = TorURLSession.shared.session
                 return { request in

bitchat/Nostr/NostrRelayManager.swift

@@ -86,7 +86,10 @@ private extension NostrRelayManagerDependencies {
             torIsForeground: { TorManager.shared.isForeground() },
             awaitTorReady: { completion in
                 Task.detached {
-                    let ready = await TorManager.shared.awaitReady()
+                    // Require both Tor bootstrap AND a positive egress self-check
+                    // so relay sockets never open unless traffic is proven to
+                    // route through Tor (fail-closed).
+                    let ready = await TorManager.shared.awaitEgressReady()
                     await MainActor.run {
                         completion(ready)
                     }

bitchatTests/Nostr/TorEgressVerifierTests.swift

@@ -0,0 +1,137 @@
+//
+// TorEgressVerifierTests.swift
+// bitchatTests
+//
+// Unit tests for the runtime Tor-egress self-check policy/caching. The network
+// probe is injected, so these tests are deterministic and offline.
+//
+
+import Testing
+import Foundation
+@testable import bitchat
+import Tor
+
+@Suite(.serialized)
+struct TorEgressVerifierTests {
+
+    /// Deterministic, controllable clock + probe.
+    private final class Harness: @unchecked Sendable {
+        private let lock = NSLock()
+        private var _now = Date(timeIntervalSince1970: 1_000_000)
+        private var _result: TorEgressVerifier.ProbeResult = .verifiedTor
+        private(set) var probeCount = 0
+
+        var now: Date {
+            lock.lock(); defer { lock.unlock() }; return _now
+        }
+        func advance(_ seconds: TimeInterval) {
+            lock.lock(); _now = _now.addingTimeInterval(seconds); lock.unlock()
+        }
+        func setResult(_ r: TorEgressVerifier.ProbeResult) {
+            lock.lock(); _result = r; lock.unlock()
+        }
+        func makeProbe() -> @Sendable () async -> TorEgressVerifier.ProbeResult {
+            return { [self] in
+                lock.lock(); probeCount += 1; let r = _result; lock.unlock()
+                return r
+            }
+        }
+        func nowProvider() -> @Sendable () -> Date {
+            return { [self] in self.now }
+        }
+    }
+
+    private func makeVerifier(_ h: Harness, ttl: TimeInterval = 300, minRetry: TimeInterval = 5) -> TorEgressVerifier {
+        TorEgressVerifier(ttl: ttl, minRetryInterval: minRetry, now: h.nowProvider(), probe: h.makeProbe())
+    }
+
+    @Test("verifiedTor allows and is cached within TTL (single probe)")
+    func verifiedIsCached() async {
+        let h = Harness()
+        h.setResult(.verifiedTor)
+        let v = makeVerifier(h, ttl: 300)
+
+        #expect(await v.verify() == true)
+        // Second call within TTL must not re-probe.
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 1)
+    }
+
+    @Test("cache expires after TTL and re-probes")
+    func cacheExpires() async {
+        let h = Harness()
+        h.setResult(.verifiedTor)
+        let v = makeVerifier(h, ttl: 300)
+
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 1)
+
+        h.advance(301)
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 2)
+    }
+
+    @Test("notTor refuses (leak detected) and is never cached as allowed")
+    func notTorRefuses() async {
+        let h = Harness()
+        h.setResult(.notTor)
+        let v = makeVerifier(h, ttl: 300, minRetry: 0)
+
+        #expect(await v.verify() == false)
+        // A subsequent success recovers.
+        h.setResult(.verifiedTor)
+        #expect(await v.verify() == true)
+    }
+
+    @Test("unreachable allows (session stays fail-closed) but does not cache a Tor verification")
+    func unreachableAllowsButNotCached() async {
+        let h = Harness()
+        h.setResult(.unreachable("down"))
+        let v = makeVerifier(h, ttl: 300, minRetry: 0)
+
+        #expect(await v.verify() == true)
+        // Not a verified state: once it turns into a real leak, we must refuse.
+        h.setResult(.notTor)
+        #expect(await v.verify() == false)
+    }
+
+    @Test("minRetryInterval throttles re-probing when not verified")
+    func throttleReprobe() async {
+        let h = Harness()
+        h.setResult(.unreachable("down"))
+        let v = makeVerifier(h, ttl: 300, minRetry: 5)
+
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 1)
+        // Within minRetry window: reuse last decision, no new probe.
+        h.advance(1)
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 1)
+        // After the window: re-probe.
+        h.advance(5)
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 2)
+    }
+
+    @Test("invalidate forces a fresh probe")
+    func invalidateForcesReprobe() async {
+        let h = Harness()
+        h.setResult(.verifiedTor)
+        let v = makeVerifier(h, ttl: 300)
+
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 1)
+        await v.invalidate()
+        #expect(await v.verify() == true)
+        #expect(h.probeCount == 2)
+    }
+
+    @Test("lastProbeResult reflects the most recent outcome")
+    func lastResultTracked() async {
+        let h = Harness()
+        h.setResult(.notTor)
+        let v = makeVerifier(h, ttl: 300, minRetry: 0)
+        _ = await v.verify()
+        #expect(await v.lastProbeResult() == .notTor)
+    }
+}

localPackages/Arti/Package.swift

@@ -30,6 +30,7 @@ let package = Package(
                 "TorManager.swift",
                 "TorURLSession.swift",
                 "TorNotifications.swift",
+                "TorEgressVerifier.swift",
             ],
             linkerSettings: [
                 .linkedLibrary("resolv"),

localPackages/Arti/Sources/TorEgressVerifier.swift

@@ -0,0 +1,174 @@
+import BitLogger
+import Foundation
+
+/// Runtime self-check that the proxied `URLSession` egress is *actually* routed
+/// through Tor — defense-in-depth for the case where a platform silently ignores
+/// `URLSessionConfiguration.connectionProxyDictionary` SOCKS settings and lets
+/// traffic egress directly (leaking the real IP while Tor appears enabled).
+///
+/// Runtime verification (see `scripts/tor-egress-verification/`) showed that
+/// macOS and the iOS simulator DO honor the SOCKS proxy for both plain HTTPS and
+/// `URLSessionWebSocketTask`, and that the proxied session is fail-closed (every
+/// request errors when the SOCKS proxy is down). Apple does not officially
+/// support SOCKS for URLSession on iOS, so on a physical device the behavior is
+/// not contractually guaranteed. This verifier closes that gap: before relay
+/// connections are opened under enforced Tor, it performs a canary request whose
+/// response positively reports whether the egress hit the network via Tor.
+///
+/// Policy (`verify()` return value):
+///   - `.verifiedTor`   → allow, and cache the positive result for `ttl`.
+///   - `.notTor`        → REFUSE. The canary reached the internet but the exit
+///                        is NOT a Tor node: a real leak. Never allow relays.
+///   - `.unreachable`   → allow-with-warning. The canary itself failed (endpoint
+///                        down / circuit not built). This does NOT egress direct:
+///                        the relay session is independently fail-closed, so the
+///                        worst case is that relay connections also fail. We do
+///                        not hard-block here to avoid taking Nostr fully offline
+///                        whenever the single canary host is unavailable.
+///
+/// The probe is injectable so the policy/caching logic is unit-tested without a
+/// live network (see `TorEgressVerifierTests`).
+public actor TorEgressVerifier {
+    public enum ProbeResult: Equatable, Sendable {
+        /// Canary succeeded and the exit is a Tor node.
+        case verifiedTor
+        /// Canary succeeded but the exit is NOT Tor — a direct-egress leak.
+        case notTor
+        /// Canary could not complete (endpoint down, no circuit, parse error).
+        case unreachable(String)
+    }
+
+    private let probe: @Sendable () async -> ProbeResult
+    private let now: @Sendable () -> Date
+    private let ttl: TimeInterval
+    /// Minimum spacing between probes when not currently verified, so a
+    /// persistent `.unreachable` cannot hammer the canary endpoint on every
+    /// reconnect burst.
+    private let minRetryInterval: TimeInterval
+
+    private var lastVerifiedAt: Date?
+    private var lastProbeAt: Date?
+    private var lastResult: ProbeResult?
+    private var inFlight: Task<Bool, Never>?
+
+    public init(
+        ttl: TimeInterval,
+        minRetryInterval: TimeInterval = 5.0,
+        now: @escaping @Sendable () -> Date = Date.init,
+        probe: @escaping @Sendable () async -> ProbeResult
+    ) {
+        self.ttl = ttl
+        self.minRetryInterval = minRetryInterval
+        self.now = now
+        self.probe = probe
+    }
+
+    /// Drop any cached verification (e.g. after a Tor restart or when the
+    /// network path changes). The next `verify()` re-probes.
+    public func invalidate() {
+        lastVerifiedAt = nil
+        lastProbeAt = nil
+        lastResult = nil
+    }
+
+    /// The most recent probe outcome, for diagnostics/tests.
+    public func lastProbeResult() -> ProbeResult? { lastResult }
+
+    /// Returns `true` when it is safe to open relay connections through the
+    /// proxied session, `false` only when a non-Tor egress was positively
+    /// detected. See the type doc for the full policy.
+    public func verify() async -> Bool {
+        if isFreshlyVerified() { return true }
+        // Throttle re-probes when the last attempt did not verify.
+        if let last = lastProbeAt,
+           let result = lastResult,
+           now().timeIntervalSince(last) < minRetryInterval {
+            return decision(for: result)
+        }
+        if let inFlight { return await inFlight.value }
+
+        let task = Task<Bool, Never> { await self.runProbe() }
+        inFlight = task
+        let allowed = await task.value
+        inFlight = nil
+        return allowed
+    }
+
+    private func isFreshlyVerified() -> Bool {
+        guard let last = lastVerifiedAt else { return false }
+        return now().timeIntervalSince(last) < ttl
+    }
+
+    private func decision(for result: ProbeResult) -> Bool {
+        switch result {
+        case .verifiedTor: return true
+        case .unreachable: return true
+        case .notTor: return false
+        }
+    }
+
+    private func runProbe() async -> Bool {
+        let result = await probe()
+        lastProbeAt = now()
+        lastResult = result
+        switch result {
+        case .verifiedTor:
+            lastVerifiedAt = now()
+            return true
+        case .notTor:
+            lastVerifiedAt = nil
+            SecureLogger.error(
+                "🧅 Tor egress self-check FAILED: request exited via a NON-Tor address — refusing relay connections (possible IP leak)",
+                category: .session
+            )
+            return false
+        case .unreachable(let why):
+            lastVerifiedAt = nil
+            SecureLogger.warning(
+                "🧅 Tor egress self-check could not complete (\(why)); relay session remains fail-closed via SOCKS proxy",
+                category: .session
+            )
+            return true
+        }
+    }
+}
+
+// MARK: - Live probe
+
+public extension TorEgressVerifier {
+    /// Default canary: fetch Tor Project's connectivity check API through the
+    /// shared proxied session and assert `IsTor == true`. Because the response
+    /// is served from the *exit's* vantage point, a silent direct egress is
+    /// caught here as `.notTor`. `check.torproject.org` is clearnet, so this
+    /// works without onion-service support.
+    ///
+    /// Follow-up (see PR): make the canary endpoint configurable and add an
+    /// onion-service canary so verification does not depend on a single host.
+    static func liveProbe(
+        endpoint: URL = URL(string: "https://check.torproject.org/api/ip")!,
+        timeout: TimeInterval = 20
+    ) -> @Sendable () async -> ProbeResult {
+        return {
+            var request = URLRequest(url: endpoint)
+            request.timeoutInterval = timeout
+            request.cachePolicy = .reloadIgnoringLocalAndRemoteCacheData
+            let session = TorURLSession.shared.session
+            do {
+                let (data, response) = try await session.data(for: request)
+                guard let http = response as? HTTPURLResponse,
+                      (200..<300).contains(http.statusCode) else {
+                    return .unreachable("http status \((response as? HTTPURLResponse)?.statusCode ?? -1)")
+                }
+                guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any] else {
+                    return .unreachable("unparseable canary response")
+                }
+                if let isTor = json["IsTor"] as? Bool {
+                    return isTor ? .verifiedTor : .notTor
+                }
+                return .unreachable("canary response missing IsTor")
+            } catch {
+                return .unreachable(error.localizedDescription)
+            }
+        }
+    }
+}

localPackages/Arti/Sources/TorManager.swift

@@ -59,6 +59,14 @@ public final class TorManager: ObservableObject {
     private var socksReady: Bool = false { didSet { recomputeReady() } }
     private var restarting: Bool = false
 
+    /// Runtime egress self-check: proves the proxied session actually exits via
+    /// Tor before relay connections are opened (defense-in-depth against a
+    /// platform silently ignoring the SOCKS proxy). Cached for a few minutes.
+    public let egressVerifier = TorEgressVerifier(
+        ttl: 300,
+        probe: TorEgressVerifier.liveProbe()
+    )
+
     // Whether the app must enforce Tor for all connections (fail-closed).
     public var torEnforced: Bool {
         #if BITCHAT_DEV_ALLOW_CLEARNET
@@ -125,6 +133,22 @@ public final class TorManager: ObservableObject {
         return await MainActor.run(body: { self.networkPermitted })
     }
 
+    /// Like `awaitReady`, but additionally requires that a canary request
+    /// through the proxied session positively verifies Tor egress. Returns
+    /// `false` if Tor never became ready, or if the egress self-check
+    /// positively detected a non-Tor (direct) egress. Callers must fail closed
+    /// on `false` — never fall back to a direct connection.
+    nonisolated
+    public func awaitEgressReady(timeout: TimeInterval = 75.0) async -> Bool {
+        let ready = await awaitReady(timeout: timeout)
+        guard ready else { return false }
+        // Clearnet dev builds don't route through Tor, so the canary would
+        // (correctly) report non-Tor; skip it there.
+        let enforced = await MainActor.run { self.torEnforced }
+        guard enforced else { return true }
+        return await egressVerifier.verify()
+    }
+
     // MARK: - Filesystem
 
     func dataDirectoryURL() -> URL? {
@@ -325,6 +349,8 @@ public final class TorManager: ObservableObject {
             self.socksReady = false
             self.isStarting = false
         }
+        // Force a fresh egress self-check once Tor comes back.
+        Task { await egressVerifier.invalidate() }
     }
 
     public func shutdownCompletely() {
@@ -353,6 +379,7 @@ public final class TorManager: ObservableObject {
                 // Note: Don't clear startedAt here - it will be set fresh on next startIfNeeded()
                 // Clearing it here races with startup and defeats the grace period
             }
+            await self.egressVerifier.invalidate()
         }
     }
 
@@ -368,6 +395,8 @@ public final class TorManager: ObservableObject {
             self.isDormant = false
             self.lastRestartAt = Date()
         }
+        // New Arti instance means new circuits; re-verify egress after restart.
+        await egressVerifier.invalidate()
 
         _ = arti_stop()
 

scripts/tor-egress-verification/proxy_probe.swift

@@ -0,0 +1,90 @@
+// Standalone harness: does Apple's URLSession honor connectionProxyDictionary
+// SOCKS settings for a plain HTTPS GET and for URLSessionWebSocketTask?
+//
+// Build:  swiftc -O proxy_probe.swift -o proxy_probe
+// Usage:  proxy_probe <http|ws> <cf|raw> <proxyPort> [targetURL]
+//
+// Prints a single RESULT line: RESULT <mode> <keyStyle> <outcome> <detail>
+// The caller correlates this with the SOCKS proxy's connection log to decide
+// whether the request was proxied.
+#if canImport(CFNetwork)
+import CFNetwork
+#endif
+import Foundation
+
+let args = CommandLine.arguments
+guard args.count >= 4 else {
+    FileHandle.standardError.write(Data("usage: proxy_probe <http|ws> <cf|raw> <proxyPort> [targetURL]\n".utf8))
+    exit(2)
+}
+let mode = args[1]
+let keyStyle = args[2]
+let proxyPort = Int(args[3]) ?? 19999
+let host = "127.0.0.1"
+
+func makeProxyDict() -> [AnyHashable: Any] {
+    switch keyStyle {
+#if os(macOS)
+    case "cf":
+        // The exact constants the app uses on macOS.
+        return [
+            kCFNetworkProxiesSOCKSEnable as String: 1,
+            kCFNetworkProxiesSOCKSProxy as String: host,
+            kCFNetworkProxiesSOCKSPort as String: proxyPort
+        ]
+#endif
+    default:
+        // The exact raw string keys the app uses on iOS.
+        return [
+            "SOCKSEnable": 1,
+            "SOCKSProxy": host,
+            "SOCKSPort": proxyPort
+        ]
+    }
+}
+
+let cfg = URLSessionConfiguration.ephemeral
+cfg.waitsForConnectivity = false
+cfg.timeoutIntervalForRequest = 20
+cfg.connectionProxyDictionary = makeProxyDict()
+let session = URLSession(configuration: cfg)
+
+func emit(_ outcome: String, _ detail: String) {
+    print("RESULT \(mode) \(keyStyle) \(outcome) \(detail)")
+    exit(outcome == "ERROR" ? 1 : 0)
+}
+
+let sem = DispatchSemaphore(value: 0)
+
+if mode == "http" {
+    let target = URL(string: args.count >= 5 ? args[4] : "https://raw.githubusercontent.com/permissionlesstech/georelays/refs/heads/main/nostr_relays.csv")!
+    let task = session.dataTask(with: target) { data, resp, err in
+        if let err = err {
+            emit("ERROR", "\(err.localizedDescription)")
+        } else if let http = resp as? HTTPURLResponse {
+            emit("OK", "status=\(http.statusCode) bytes=\(data?.count ?? 0)")
+        } else {
+            emit("OK", "bytes=\(data?.count ?? 0)")
+        }
+    }
+    task.resume()
+} else {
+    // WebSocket
+    let target = URL(string: args.count >= 5 ? args[4] : "wss://relay.damus.io")!
+    let ws = session.webSocketTask(with: target)
+    ws.resume()
+    // A successful ping proves the TLS+WS handshake completed end-to-end.
+    ws.sendPing { err in
+        if let err = err {
+            emit("ERROR", "\(err.localizedDescription)")
+        } else {
+            emit("OK", "ws-ping-ok")
+        }
+    }
+}
+
+// Global watchdog so we never hang.
+DispatchQueue.global().asyncAfter(deadline: .now() + 25) {
+    emit("ERROR", "timeout")
+}
+sem.wait()

scripts/tor-egress-verification/run_verification.sh

@@ -0,0 +1,73 @@
+#!/usr/bin/env bash
+# Orchestrates the Tor-egress proxy-honoring verification on macOS.
+#
+# For each (request-type x key-style) it runs two experiments:
+#   A) proxy UP   — did a connection arrive at the SOCKS proxy? (log grows)
+#   B) proxy DOWN — pointed at a dead port; does the request still SUCCEED?
+#      If it succeeds with no proxy, egress went DIRECT (proxy ignored).
+#      If it fails, the proxy setting is being enforced (fail-closed).
+#
+# Discriminator: PROXIED  = connection observed at proxy AND fails when proxy down
+#                DIRECT   = no connection at proxy OR succeeds when proxy down
+set -u
+DIR="$(cd "$(dirname "$0")" && pwd)"
+PORT=19999
+DEADPORT=19998   # nothing listens here
+LOG="$(mktemp -t sockslog)"
+BIN="$(mktemp -t proxyprobe)"
+
+echo "== building swift probe =="
+swiftc -O "$DIR/proxy_probe.swift" -o "$BIN" || { echo "swiftc failed"; exit 1; }
+
+echo "== starting SOCKS proxy on $PORT =="
+: > "$LOG"
+python3 "$DIR/socks5_probe_proxy.py" "$PORT" "$LOG" >/tmp/socksproxy.out 2>&1 &
+PROXY_PID=$!
+trap 'kill $PROXY_PID 2>/dev/null' EXIT
+# wait for READY
+for _ in $(seq 1 50); do
+  grep -q READY /tmp/socksproxy.out 2>/dev/null && break
+  sleep 0.1
+done
+
+run_case() {
+  local mode="$1" key="$2"
+  # Experiment A: proxy up, watch log
+  local before after target
+  before=$(wc -l < "$LOG" | tr -d ' ')
+  local outA
+  outA=$("$BIN" "$mode" "$key" "$PORT" 2>/dev/null | grep '^RESULT' || echo "RESULT $mode $key ERROR no-output")
+  sleep 0.3
+  after=$(wc -l < "$LOG" | tr -d ' ')
+  local proxied="NO"
+  if [ "$after" -gt "$before" ]; then proxied="YES"; fi
+  local newlines
+  newlines=$(tail -n +"$((before+1))" "$LOG" | tr '\t' ' ' | tr '\n' '|')
+
+  # Experiment B: proxy down (dead port), same request
+  local outB
+  outB=$("$BIN" "$mode" "$key" "$DEADPORT" 2>/dev/null | grep '^RESULT' || echo "RESULT $mode $key ERROR no-output")
+
+  echo "----------------------------------------"
+  echo "CASE mode=$mode key=$key"
+  echo "  A(proxy up):   $outA   | connection_at_proxy=$proxied  [$newlines]"
+  echo "  B(proxy down): $outB"
+  # verdict
+  local a_ok b_ok
+  a_ok=$(echo "$outA" | awk '{print $4}')
+  b_ok=$(echo "$outB" | awk '{print $4}')
+  local verdict="UNKNOWN"
+  if [ "$proxied" = "YES" ] && [ "$b_ok" = "ERROR" ]; then verdict="PROXIED (enforced)"; fi
+  if [ "$proxied" = "NO" ] && [ "$b_ok" = "OK" ]; then verdict="DIRECT (proxy ignored)"; fi
+  if [ "$proxied" = "YES" ] && [ "$b_ok" = "OK" ]; then verdict="AMBIGUOUS (uses proxy if up, but egresses direct if down)"; fi
+  if [ "$proxied" = "NO" ] && [ "$b_ok" = "ERROR" ]; then verdict="BLOCKED both (network/target issue?)"; fi
+  echo "  VERDICT: $verdict"
+}
+
+for mode in http ws; do
+  for key in cf raw; do
+    run_case "$mode" "$key"
+  done
+done
+echo "========================================"
+echo "raw proxy log:"; cat "$LOG" | tr '\t' ' '

scripts/tor-egress-verification/socks5_probe_proxy.py

@@ -0,0 +1,166 @@
+#!/usr/bin/env python3
+"""
+Minimal threaded SOCKS5 CONNECT proxy used to verify whether Apple's
+URLSession actually honors `connectionProxyDictionary` SOCKS settings for
+different request types (plain HTTPS vs URLSessionWebSocketTask).
+
+Behavior:
+  - Speaks enough SOCKS5 (no-auth) to complete a CONNECT and then relays
+    bytes bidirectionally to the real destination.
+  - Every accepted CONNECT is appended to a log file as one line:
+        <iso8601>\tCONNECT\t<host>:<port>
+  - Any raw connection that is NOT valid SOCKS5 is logged as:
+        <iso8601>\tNON_SOCKS\t<first-bytes-hex>
+    (this catches the feared case where URLSession sends a raw TLS/HTTP
+     ClientHello straight at the proxy port instead of a SOCKS greeting).
+
+If a request egresses DIRECTLY (proxy ignored), nothing is logged at all.
+
+Usage: socks5_probe_proxy.py <listen_port> <log_file>
+"""
+import selectors
+import socket
+import sys
+import threading
+from datetime import datetime, timezone
+
+LOG_LOCK = threading.Lock()
+
+
+def log(logfile, kind, detail):
+    line = f"{datetime.now(timezone.utc).isoformat()}\t{kind}\t{detail}\n"
+    with LOG_LOCK:
+        with open(logfile, "a") as f:
+            f.write(line)
+    sys.stderr.write("[proxy] " + line)
+    sys.stderr.flush()
+
+
+def recv_exact(sock, n):
+    buf = b""
+    while len(buf) < n:
+        chunk = sock.recv(n - len(buf))
+        if not chunk:
+            return None
+        buf += chunk
+    return buf
+
+
+def handle(client, logfile):
+    client.settimeout(15)
+    try:
+        # SOCKS5 greeting: VER=0x05, NMETHODS, METHODS...
+        head = recv_exact(client, 2)
+        if not head:
+            return
+        if head[0] != 0x05:
+            # Not SOCKS5 at all — this is the smoking gun for a direct egress
+            # that mistakenly hit the proxy port. Log the first bytes.
+            rest = b""
+            try:
+                client.setblocking(False)
+                rest = client.recv(64)
+            except Exception:
+                pass
+            log(logfile, "NON_SOCKS", (head + rest).hex())
+            return
+        nmethods = head[1]
+        if nmethods:
+            recv_exact(client, nmethods)
+        # Reply: no authentication required
+        client.sendall(b"\x05\x00")
+
+        # Request: VER, CMD, RSV, ATYP, ADDR, PORT
+        req = recv_exact(client, 4)
+        if not req or req[1] != 0x01:  # only CONNECT
+            client.sendall(b"\x05\x07\x00\x01\x00\x00\x00\x00\x00\x00")
+            return
+        atyp = req[3]
+        if atyp == 0x01:  # IPv4
+            addr = socket.inet_ntoa(recv_exact(client, 4))
+        elif atyp == 0x03:  # domain
+            ln = recv_exact(client, 1)[0]
+            addr = recv_exact(client, ln).decode("ascii", errors="replace")
+        elif atyp == 0x04:  # IPv6
+            addr = socket.inet_ntop(socket.AF_INET6, recv_exact(client, 16))
+        else:
+            client.sendall(b"\x05\x08\x00\x01\x00\x00\x00\x00\x00\x00")
+            return
+        port = int.from_bytes(recv_exact(client, 2), "big")
+
+        log(logfile, "CONNECT", f"{addr}:{port}")
+
+        # Connect to the real destination and reply success.
+        try:
+            remote = socket.create_connection((addr, port), timeout=15)
+        except Exception as e:
+            log(logfile, "CONNECT_FAIL", f"{addr}:{port} {e}")
+            client.sendall(b"\x05\x01\x00\x01\x00\x00\x00\x00\x00\x00")
+            return
+        client.sendall(b"\x05\x00\x00\x01\x00\x00\x00\x00\x00\x00")
+
+        relay(client, remote)
+    except Exception:
+        pass
+    finally:
+        try:
+            client.close()
+        except Exception:
+            pass
+
+
+def relay(a, b):
+    a.setblocking(False)
+    b.setblocking(False)
+    sel = selectors.DefaultSelector()
+    sel.register(a, selectors.EVENT_READ, b)
+    sel.register(b, selectors.EVENT_READ, a)
+    try:
+        while True:
+            events = sel.select(timeout=30)
+            if not events:
+                break
+            for key, _ in events:
+                src = key.fileobj
+                dst = key.data
+                try:
+                    data = src.recv(65536)
+                except (BlockingIOError, InterruptedError):
+                    continue
+                except Exception:
+                    return
+                if not data:
+                    return
+                try:
+                    dst.sendall(data)
+                except Exception:
+                    return
+    finally:
+        sel.close()
+        for s in (a, b):
+            try:
+                s.close()
+            except Exception:
+                pass
+
+
+def main():
+    if len(sys.argv) != 3:
+        print("usage: socks5_probe_proxy.py <port> <logfile>", file=sys.stderr)
+        sys.exit(2)
+    port = int(sys.argv[1])
+    logfile = sys.argv[2]
+    srv = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+    srv.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+    srv.bind(("127.0.0.1", port))
+    srv.listen(64)
+    sys.stderr.write(f"[proxy] listening on 127.0.0.1:{port}, log={logfile}\n")
+    sys.stderr.flush()
+    print("READY", flush=True)
+    while True:
+        client, _ = srv.accept()
+        threading.Thread(target=handle, args=(client, logfile), daemon=True).start()
+
+
+if __name__ == "__main__":
+    main()