diff --git a/backend/fleet/src/Fleet.Server.Abstractions/ServerContracts.cs b/backend/fleet/src/Fleet.Server.Abstractions/ServerContracts.cs index 298896d..3b1d896 100644 --- a/backend/fleet/src/Fleet.Server.Abstractions/ServerContracts.cs +++ b/backend/fleet/src/Fleet.Server.Abstractions/ServerContracts.cs @@ -6,6 +6,17 @@ namespace Fleet.Server; public sealed record EnrollmentContext(string? RemoteIp, JsonElement? ProductMeta); +/// +/// 节点的真实来源地址：前面有反向代理时，套接字上只有代理的地址，产品在这里决定信不信转发头。 +/// 没有注册就用套接字地址。 +/// +public interface IClientAddressSource +{ + /// 套接字上看到的对端地址。 + /// 按名字取请求头的所有值。 + string? Resolve(System.Net.IPAddress? socket, Func> header); +} + /// /// 服务器的身份密钥：操作员用 passkey 把它的公钥签进操作员集合。有它的时候，认证挑战会带上服务器的签名，节点据此确认对面是真的服务器， /// 并在这条链路上用应用层加密。没有注册就是旧的行为（没有服务器身份）。 @@ -38,7 +49,8 @@ public enum NodeEventKind Heartbeat, } - public sealed record NodeEvent(NodeEventKind Kind, NodeRecord Node, long AtMs, HelloParams? Hello = null, JsonElement? Extra = null, string? Reason = null); + /// 这条连接的来源地址（经过 ）；没有就是 null。 +public sealed record NodeEvent(NodeEventKind Kind, NodeRecord Node, long AtMs, HelloParams? Hello = null, JsonElement? Extra = null, string? Reason = null, string? RemoteIp = null); public interface INodeEventSink { diff --git a/backend/fleet/src/Fleet.Server.Enrollment/FleetEnrollmentEndpoints.cs b/backend/fleet/src/Fleet.Server.Enrollment/FleetEnrollmentEndpoints.cs index ccd3567..9b3e948 100644 --- a/backend/fleet/src/Fleet.Server.Enrollment/FleetEnrollmentEndpoints.cs +++ b/backend/fleet/src/Fleet.Server.Enrollment/FleetEnrollmentEndpoints.cs @@ -73,7 +73,7 @@ public static class FleetEnrollmentEndpoints return Problem(StatusCodes.Status400BadRequest, FleetErrors.InvalidParams, "request body is not valid"); } - Result result = await service.EnrollAsync(request, http.Connection.RemoteIpAddress?.ToString(), http.RequestAborted); + Result result = await service.EnrollAsync(request, ClientIp(http), http.RequestAborted); if (result.TryGet(out EnrollResponse? response, out FleetError? error)) { return Results.Json(response, EnrollmentJsonContext.Default.EnrollResponse, statusCode: StatusCodes.Status201Created); @@ -88,6 +88,11 @@ public static class FleetEnrollmentEndpoints return Problem(status, error.Code, error.Message); } + private static string? ClientIp(HttpContext http) => + http.RequestServices.GetService() is { } source + ? source.Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? "")) + : http.Connection.RemoteIpAddress?.ToString(); + private static async Task ReadBodyAsync(HttpContext http, CancellationToken ct) { if (http.Request.ContentLength > MaxBodyBytes) diff --git a/backend/fleet/src/Fleet.Server.Link/FleetLinkHost.cs b/backend/fleet/src/Fleet.Server.Link/FleetLinkHost.cs index d4d4b82..976df47 100644 --- a/backend/fleet/src/Fleet.Server.Link/FleetLinkHost.cs +++ b/backend/fleet/src/Fleet.Server.Link/FleetLinkHost.cs @@ -67,9 +67,9 @@ public sealed class FleetLinkHost : IFleetConnections, IStateNotifier _connections.TryGetValue(node, out NodeConnection? c) ? c.Session : null; /// 处理一条 WebSocket 连接，直到连接结束。 - public async Task RunConnectionAsync(WebSocket socket, CancellationToken ct) + public async Task RunConnectionAsync(WebSocket socket, string? remoteIp, CancellationToken ct) { - using var connection = new NodeConnection(this, socket); + using var connection = new NodeConnection(this, socket, remoteIp); await connection.RunAsync(ct); } diff --git a/backend/fleet/src/Fleet.Server.Link/FleetServerBuilder.cs b/backend/fleet/src/Fleet.Server.Link/FleetServerBuilder.cs index 0c99b75..6361887 100644 --- a/backend/fleet/src/Fleet.Server.Link/FleetServerBuilder.cs +++ b/backend/fleet/src/Fleet.Server.Link/FleetServerBuilder.cs @@ -128,7 +128,10 @@ public static class FleetServerExtensions } using System.Net.WebSockets.WebSocket socket = await http.WebSockets.AcceptWebSocketAsync(); - await http.RequestServices.GetRequiredService().RunConnectionAsync(socket, http.RequestAborted); + string? remote = http.RequestServices.GetService() is { } source + ? source.Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? "")) + : http.Connection.RemoteIpAddress?.ToString(); + await http.RequestServices.GetRequiredService().RunConnectionAsync(socket, remote, http.RequestAborted); }); return app; } diff --git a/backend/fleet/src/Fleet.Server.Link/NodeConnection.State.cs b/backend/fleet/src/Fleet.Server.Link/NodeConnection.State.cs index 06feb0e..f4511e1 100644 --- a/backend/fleet/src/Fleet.Server.Link/NodeConnection.State.cs +++ b/backend/fleet/src/Fleet.Server.Link/NodeConnection.State.cs @@ -102,7 +102,7 @@ internal sealed partial class NodeConnection private async Task FireAsync(NodeEventKind kind, CancellationToken ct, HelloParams? hello = null, JsonElement? extra = null) { - var nodeEvent = new NodeEvent(kind, _node!, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), hello, extra, kind == NodeEventKind.Offline ? _closeReason : null); + var nodeEvent = new NodeEvent(kind, _node!, _host.Time.GetUtcNow().ToUnixTimeMilliseconds(), hello, extra, kind == NodeEventKind.Offline ? _closeReason : null, _remoteIp); foreach (INodeEventSink sink in _host.Sinks) { try diff --git a/backend/fleet/src/Fleet.Server.Link/NodeConnection.cs b/backend/fleet/src/Fleet.Server.Link/NodeConnection.cs index 95867d6..9e867ea 100644 --- a/backend/fleet/src/Fleet.Server.Link/NodeConnection.cs +++ b/backend/fleet/src/Fleet.Server.Link/NodeConnection.cs @@ -34,10 +34,13 @@ internal sealed partial class NodeConnection : IDisposable private bool _helloDone; private int _violations; - public NodeConnection(FleetLinkHost host, WebSocket socket) + private readonly string? _remoteIp; + + public NodeConnection(FleetLinkHost host, WebSocket socket, string? remoteIp = null) { _host = host; _socket = socket; + _remoteIp = remoteIp; } public NodeId NodeId => _node!.Id; diff --git a/backend/src/MoShiro.Backend.Api/ApiModule.cs b/backend/src/MoShiro.Backend.Api/ApiModule.cs index 60bfc5e..552b49b 100644 --- a/backend/src/MoShiro.Backend.Api/ApiModule.cs +++ b/backend/src/MoShiro.Backend.Api/ApiModule.cs @@ -20,6 +20,7 @@ public static class ApiModule services.AddSingleton(); services.AddSingleton(); services.AddSingleton(); + services.AddSingleton(p => p.GetRequiredService()); services.Configure(options => { JsonSerializerOptions json = options.SerializerOptions; diff --git a/backend/src/MoShiro.Backend.Api/Common/ClientAddress.cs b/backend/src/MoShiro.Backend.Api/Common/ClientAddress.cs index bf61a94..82e8d08 100644 --- a/backend/src/MoShiro.Backend.Api/Common/ClientAddress.cs +++ b/backend/src/MoShiro.Backend.Api/Common/ClientAddress.cs @@ -11,7 +11,7 @@ namespace MoShiro.Backend.Api; /// 且从右往左跳过所有可信代理，取第一个不可信的地址（客户端自己伪造的在最左边，不会被采信）。 /// 没配置就一律用套接字地址：宁可限流和审计里是代理的地址，也不让客户端自己报地址。 /// -public sealed class ClientAddressResolver +public sealed class ClientAddressResolver : Fleet.Server.IClientAddressSource { private readonly IPNetwork[] _trusted; private readonly string _header; @@ -43,9 +43,13 @@ public sealed class ClientAddressResolver return _trusted.Any(n => n.Contains(normalized)); } - public string Resolve(HttpContext http) + string? Fleet.Server.IClientAddressSource.Resolve(IPAddress? socket, Func> header) => Resolve(socket, header); + + public string Resolve(HttpContext http) => + Resolve(http.Connection.RemoteIpAddress, name => http.Request.Headers[name].Select(v => v ?? "")); + + private string Resolve(IPAddress? remote, Func> headers) { - IPAddress? remote = http.Connection.RemoteIpAddress; if (remote is null) { return "unknown"; @@ -56,7 +60,7 @@ public sealed class ClientAddressResolver return remote.ToString(); } - string[] chain = [.. http.Request.Headers[_header].SelectMany(v => (v ?? "").Split(',', StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))]; + string[] chain = [.. headers(_header).SelectMany(v => (v ?? "").Split(',', StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))]; for (int i = chain.Length - 1; i >= 0; i--) { if (!IPAddress.TryParse(chain[i], out IPAddress? hop)) diff --git a/backend/src/MoShiro.Backend.Application/Agents/MoShiroFleetHooks.cs b/backend/src/MoShiro.Backend.Application/Agents/MoShiroFleetHooks.cs index 75010cf..f3dbfdc 100644 --- a/backend/src/MoShiro.Backend.Application/Agents/MoShiroFleetHooks.cs +++ b/backend/src/MoShiro.Backend.Application/Agents/MoShiroFleetHooks.cs @@ -115,10 +115,15 @@ public sealed class MoShiroNodeEventSink : INodeEventSink await _events.AddAsync(id, "agent_offline", nodeEvent.AtMs, null, ct); break; case NodeEventKind.Hello when nodeEvent.Hello is { } hello: - await ApplyHelloAsync(id, hello, ct); + await ApplyHelloAsync(id, hello, nodeEvent.RemoteIp, ct); break; case NodeEventKind.Heartbeat when nodeEvent.Extra is { } extra: await _nodes.SetVersionAsync(id.Value, NodeKinds.Agent, nodeEvent.Node.NodeVersion, ct); + if (_nodes.ShouldRetryGeo(id.Value) && await _agents.GetAsync(id, ct) is { } unlocated) + { + _nodes.ObserveAgentInBackground(id.Value, GeoHostOf(unlocated, nodeEvent.RemoteIp), null); + } + try { AgentRuntimeState state = _runtime.Get(id); @@ -145,6 +150,12 @@ public sealed class MoShiroNodeEventSink : INodeEventSink } } + /// 查国家和运营商用哪个地址：对外地址优先，其次 IPv6 地址，都没有（NAT 后面、或者没有上报公网地址）才用这条连接的来源地址。 + private static string? GeoHostOf(AgentRecord? agent, string? remoteIp) => + !string.IsNullOrEmpty(agent?.PublicHost) ? agent.PublicHost + : !string.IsNullOrEmpty(agent?.Ipv6Host) ? agent.Ipv6Host + : remoteIp; + /// Probe 自己连上来了：hello 里带着它的系统、架构和构建信息；上线下线只推给 SSE，不记事件。 private async ValueTask OnProbeEventAsync(NodeId id, NodeEvent nodeEvent, CancellationToken ct) { @@ -213,7 +224,7 @@ public sealed class MoShiroNodeEventSink : INodeEventSink } } - private async ValueTask ApplyHelloAsync(NodeId id, HelloParams hello, CancellationToken ct) + private async ValueTask ApplyHelloAsync(NodeId id, HelloParams hello, string? remoteIp, CancellationToken ct) { var kinds = new List(); foreach (string capability in hello.Capabilities) @@ -278,7 +289,7 @@ public sealed class MoShiroNodeEventSink : INodeEventSink // 端口、证书指纹或对外地址变了，Probe 需要新的列表 AgentRecord? after = await _agents.GetAsync(id, ct); - _nodes.ObserveAgentInBackground(id.Value, after?.PublicHost, meta.Platform); + _nodes.ObserveAgentInBackground(id.Value, GeoHostOf(after, remoteIp), meta.Platform); await _releases.OfferToNodeAsync(id, ct); await _schedule.SendToNodeAsync(id, ct); if (before is not null && after is not null diff --git a/backend/src/MoShiro.Backend.Application/Nodes/NodeInfoService.cs b/backend/src/MoShiro.Backend.Application/Nodes/NodeInfoService.cs index 4270f83..70372b2 100644 --- a/backend/src/MoShiro.Backend.Application/Nodes/NodeInfoService.cs +++ b/backend/src/MoShiro.Backend.Application/Nodes/NodeInfoService.cs @@ -26,6 +26,7 @@ public sealed class NodeInfoService private readonly TimeProvider _time; private readonly ConcurrentDictionary _cache = new(); private readonly ConcurrentDictionary _inflight = new(); + private readonly ConcurrentDictionary _lastTry = new(); private Task? _load; public NodeInfoService(INodeMetaStore store, IGeoLookup geo, TimeProvider time) @@ -119,6 +120,7 @@ public sealed class NodeInfoService public async ValueTask ObserveAddressAsync(string nodeId, string kind, string? host, CancellationToken ct) { host = host?.Trim(); + _lastTry[nodeId] = Now(); if (string.IsNullOrEmpty(host) || !IsPublic(host) || !_inflight.TryAdd(nodeId, 0)) { return; @@ -162,6 +164,17 @@ public sealed class NodeInfoService } } + /// 国家还没查到、并且离上一次尝试已经过了重试间隔：调用方可以再查一次（心跳里调，不会每次都真去查）。 + public bool ShouldRetryGeo(string nodeId) + { + if (_cache.TryGetValue(nodeId, out NodeMeta? meta) && (meta.Info.Country is not null || meta.Info.IspManual)) + { + return false; + } + + return !_lastTry.TryGetValue(nodeId, out long at) || Now() - at >= s_retryAfter.TotalMilliseconds; + } + /// Agent 报告某个 Probe 连着它：在后台用 Probe 的来源地址查国家和运营商（系统信息是 Probe 自己连上面板时报的）。出错只吞掉。 public void ObserveProbeInBackground(string probeId, string? remote) => _ = Task.Run(async () => diff --git a/backend/tests/MoShiro.Backend.Tests/ClientAddressTests.cs b/backend/tests/MoShiro.Backend.Tests/ClientAddressTests.cs index 6e785d8..4c3847e 100644 --- a/backend/tests/MoShiro.Backend.Tests/ClientAddressTests.cs +++ b/backend/tests/MoShiro.Backend.Tests/ClientAddressTests.cs @@ -48,4 +48,12 @@ public sealed class ClientAddressTests [Fact] public void AMalformedTrustedProxyEntry_FailsAtStartup() => Assert.Throws(() => new ClientAddressResolver(new BackendOptions { TrustedProxies = "banana" })); + + [Fact] + public void TheFleetLinkAsksTheSameResolver_SoANodeBehindAProxyKeepsItsRealAddress() + { + Fleet.Server.IClientAddressSource source = new ClientAddressResolver(new BackendOptions { TrustedProxies = "127.0.0.1" }); + Assert.Equal("198.51.100.7", source.Resolve(IPAddress.Parse("127.0.0.1"), _ => ["198.51.100.7"])); + Assert.Equal("203.0.113.9", source.Resolve(IPAddress.Parse("203.0.113.9"), _ => ["198.51.100.7"])); + } } diff --git a/backend/tests/MoShiro.Backend.Tests/NodeInfoTests.cs b/backend/tests/MoShiro.Backend.Tests/NodeInfoTests.cs index 63d40aa..2c06e9e 100644 --- a/backend/tests/MoShiro.Backend.Tests/NodeInfoTests.cs +++ b/backend/tests/MoShiro.Backend.Tests/NodeInfoTests.cs @@ -102,6 +102,24 @@ public sealed partial class NodeInfoTests Assert.Null(Assert.Single(await service.ListAsync(CancellationToken.None)).Country); } + [Fact] + public async Task Service_AnUnlocatedNode_IsRetriedFromTheHeartbeat_UntilItHasACountry() + { + var time = new FakeTimeProvider(DateTimeOffset.Parse("2026-10-06T00:00:00Z")); + var service = new NodeInfoService(new MemoryStore(), new FakeGeo(null), time); + + Assert.True(service.ShouldRetryGeo("agt_1")); + await service.ObserveAddressAsync("agt_1", NodeKinds.Agent, "", CancellationToken.None); + Assert.False(service.ShouldRetryGeo("agt_1")); + time.Advance(TimeSpan.FromMinutes(11)); + Assert.True(service.ShouldRetryGeo("agt_1")); + + var found = new NodeInfoService(new MemoryStore(), new FakeGeo(new GeoInfo("jp", 1, "Example", false)), time); + await found.ObserveAddressAsync("agt_2", NodeKinds.Agent, "8.8.8.8", CancellationToken.None); + time.Advance(TimeSpan.FromMinutes(11)); + Assert.False(found.ShouldRetryGeo("agt_2")); + } + [Fact] public async Task Service_PlatformIsSavedAndSurvivesARestart() { diff --git a/docs/ops/deployment.md b/docs/ops/deployment.md index 9fead2e..2220e54 100644 --- a/docs/ops/deployment.md +++ b/docs/ops/deployment.md @@ -199,7 +199,7 @@ OpenWrt 默认不带 OpenSSL：`opkg install libopenssl3`（较新的快照和 2 没有公网入站的家宽机器上跑 Backend 时，用隧道（Cloudflare Tunnel 之类）把一个公网域名转到 Backend 的端口；Backend 只写一个监听地址（同时写 `0.0.0.0` 和 `[::]` 会报地址已被占用）；Probe 的 `fleet.server` 就是这个面板地址。小 VPS 上把 `limits.max_total_mbps` 调低。家宽的 IPv6 前缀会在路由器重新拨号（比如停电）后变化，用 IPv6 直连时地址要跟着换；走隧道的域名不受影响。 -**国家和运营商**：Backend 用 Probe、Agent 的来源地址向 IP 归属服务查一次，展示在国旗和运营商图标上（`MoShiro__GeoLookupUrl`，缺省 `https://ip.007995.xyz`；留空就不查；这个服务只供公益使用，署名和使用条款由部署的人自己遵守）。Probe、Agent 自己不访问外部服务，所以客户的防火墙审计里不会多出出站连接；Backend 发出去的只有公网 IP（内网地址不发）。 +**国家和运营商**：Backend 用 Probe、Agent 的来源地址向 IP 归属服务查一次，展示在国旗和运营商图标上（`MoShiro__GeoLookupUrl`，缺省 `https://ip.007995.xyz`；留空就不查；这个服务只供公益使用，署名和使用条款由部署的人自己遵守）。Probe、Agent 自己不访问外部服务，所以客户的防火墙审计里不会多出出站连接；Backend 发出去的只有公网 IP（内网地址不发）。Agent 的地址优先用它的对外地址（`public_host`），没有就用 IPv6 地址，再没有（NAT 后面、没有上报公网地址）才用它连面板的来源地址；面板前面有反向代理时，要配置 `MoShiro__TrustedProxies`，否则来源地址是代理的内网地址，这类节点查不到国家。没查到的节点每隔十分钟在心跳里重试一次。 **从别处看这个面板**：本机跑前端，在页面里把数据源指向面板的地址，用 API 令牌访问，Backend 设 `MoShiro__CorsOrigins` 放行前端的来源（见 [web/README.md](../../web/README.md)）。