Feature: Exclude human Web viewing from MCP remote control sessions

Complete the bidirectional mutual exclusion between MCP remote control
and human Web viewing (§8.2). Direction 1 (remote_open rejected while a
human session holds the screen sub-connection) already existed; this adds
direction 2: a human Web viewer is now rejected while an MCP session owns
the device.

Add a m_McpTriggeredDevices marker (mirroring m_MfcTriggeredDevices) that
is set when an MCP session is created (BeginScreenCtrlOpen) and cleared at
every session-erasure site (CloseScreenCtrlSession, SweepIdleScreenCtrl,
OnScreenControlClosed, EndScreenCtrlAction), so the marker cannot go stale
and permanently block humans. HandleConnect checks IsMcpTriggered before
mutating client state or starting the remote desktop.

Co-Authored-By: deepseek-v4-pro
This commit was merged in pull request #4.
This commit is contained in:
yuanyuanxiang
2026-08-25 14:14:26 +02:00
parent 25a6e2d07a
commit d55d40e7a2
3 changed files with 45 additions and 1 deletions

View File

@@ -3851,6 +3851,10 @@ bool CMcpServer::BeginScreenCtrlOpen(uint64_t device_id, const std::string& sess
s.sessionId = sessionId;
s.lastActiveAt = time(nullptr);
m_ScreenCtrlSessions[device_id] = std::move(s);
// 双向互斥(方向二):标记该设备屏幕子连接归 MCP 会话独占,人类 Web 观看期间被拒绝
// HandleConnect 查 IsMcpTriggered。会话擦除点CloseScreenCtrlSession /
// SweepIdleScreenCtrl / OnScreenControlClosed / EndScreenCtrlAction同步 ClearMcpTriggered。
WebService().SetMcpTriggered(device_id);
return true;
}
@@ -3881,6 +3885,7 @@ int CMcpServer::CloseScreenCtrlSession(uint64_t device_id, const std::string& se
}
if (it->second.subCtx) m_ScreenCtrlContextToDevice.erase(it->second.subCtx);
m_ScreenCtrlSessions.erase(it);
WebService().ClearMcpTriggered(device_id);
return 0;
}
@@ -3893,8 +3898,10 @@ int CMcpServer::SweepIdleScreenCtrl(time_t idleTimeoutSec) {
ScreenCtrlSession& s = it->second;
if (!s.busy && difftime(now, s.lastActiveAt) > (double)idleTimeoutSec) {
if (s.subCtx) m_ScreenCtrlContextToDevice.erase(s.subCtx);
toClose.push_back(it->first);
uint64_t devId = it->first;
toClose.push_back(devId);
it = m_ScreenCtrlSessions.erase(it);
WebService().ClearMcpTriggered(devId); // 锁内清除,避免与并发 remote_open 竞态
} else {
++it;
}
@@ -3921,8 +3928,10 @@ void CMcpServer::OnScreenControlClosed(context* subCtx) {
// 注入在飞:不擦会话(注入线程仍持 subCtx仅置 closed由 EndScreenCtrlAction 收尾。
sit->second.closed = true;
} else {
uint64_t devId = it->second;
m_ScreenCtrlSessions.erase(sit);
m_ScreenCtrlContextToDevice.erase(it);
WebService().ClearMcpTriggered(devId);
}
}
@@ -3954,6 +3963,7 @@ void CMcpServer::EndScreenCtrlAction(uint64_t device_id, const std::string& sess
if (it->second.closed) { // 注入期间子连接已断:擦会话+路由
if (it->second.subCtx) m_ScreenCtrlContextToDevice.erase(it->second.subCtx);
m_ScreenCtrlSessions.erase(it);
WebService().ClearMcpTriggered(device_id);
}
}

View File

@@ -643,6 +643,15 @@ void CWebService::HandleConnect(void* ws_ptr, const std::string& token, uint64_t
}
}
// Bidirectional mutex (direction 2, §8.2): reject human Web viewing while an MCP
// remote control session owns this host's screen sub-connection (SetMcpTriggered on
// remote_open, cleared on session teardown), so AI input cannot be injected into a
// picture a human is watching/controlling.
if (IsMcpTriggered(device_id)) {
SendText(ws_ptr, BuildJsonResponse("connect_result", false, "Device busy: remote control session active"));
return;
}
// Check max clients per device
int current_count = GetWebClientCount(device_id);
if (current_count >= m_nMaxClientsPerDevice) {
@@ -2150,6 +2159,21 @@ void CWebService::ClearMfcTriggered(uint64_t device_id) {
m_MfcTriggeredDevices.erase(device_id);
}
void CWebService::SetMcpTriggered(uint64_t device_id) {
std::lock_guard<std::mutex> lock(m_McpTriggeredMutex);
m_McpTriggeredDevices.insert(device_id);
}
bool CWebService::IsMcpTriggered(uint64_t device_id) {
std::lock_guard<std::mutex> lock(m_McpTriggeredMutex);
return m_McpTriggeredDevices.find(device_id) != m_McpTriggeredDevices.end();
}
void CWebService::ClearMcpTriggered(uint64_t device_id) {
std::lock_guard<std::mutex> lock(m_McpTriggeredMutex);
m_McpTriggeredDevices.erase(device_id);
}
bool CWebService::HasActiveSession(uint64_t device_id) {
std::lock_guard<std::mutex> lock(m_ScreenContextsMutex);
return m_ScreenContexts.find(device_id) != m_ScreenContexts.end();

View File

@@ -299,6 +299,12 @@ public:
bool IsMfcTriggered(uint64_t device_id);
void ClearMfcTriggered(uint64_t device_id);
// MCP trigger management - an MCP remote control session owns the screen
// sub-connection and blocks human Web viewing (bidirectional mutex, §8.2).
void SetMcpTriggered(uint64_t device_id);
bool IsMcpTriggered(uint64_t device_id);
void ClearMcpTriggered(uint64_t device_id);
// Check if a remote desktop session already exists for device
bool HasActiveSession(uint64_t device_id);
@@ -344,6 +350,10 @@ private:
std::set<uint64_t> m_MfcTriggeredDevices;
std::mutex m_MfcTriggeredMutex;
// MCP triggered devices: screen sub-connection owned by an MCP remote control session
std::set<uint64_t> m_McpTriggeredDevices;
std::mutex m_McpTriggeredMutex;
// Web 终端会话状态
struct WebTermSession {
void* ws_ptr; // browser WebSocket