[noPoll] Deadlock

[ondrej zaruba]

Hi,

I am using the function nopoll_thread_handlers to make the library thread safe

************************************
noPollPtr WebSocket::mutexCreate(void)
{
    std::mutex * mutex = new std::mutex;
    std::cout<<".................create"<<std::endl;
    return static_cast<noPollPtr>(mutex);
}

void WebSocket::mutexDestroy(noPollPtr _mutex)
{
    std::cout<<".................destroy"<<std::endl;
    std::mutex * mutex = static_cast<std::mutex *>(_mutex);
    delete mutex;
}

void WebSocket::mutexLock(noPollPtr _mutex)
{
    std::cout<<".................lock"<<std::endl;
    std::mutex * mutex = static_cast<std::mutex *>(_mutex);
    mutex->lock();
}

void WebSocket::mutexUnlock(noPollPtr _mutex)
{
    std::cout<<".................unlock"<<std::endl;
    std::mutex * mutex = static_cast<std::mutex *>(_mutex);
    mutex->unlock();
}

nopoll_thread_handlers(&WebSocket::mutexCreate,
&WebSocket::mutexDestroy, &WebSocket::mutexLock,
&WebSocket::mutexUnlock);

***********************************

When the socket closes (timeout on TCP socket) the library dead-locks
DEBUG PRINT:

***********************************
(proc 9668): (debug)nopoll_conn.c:3008 nopoll_conn_send_text:
Attempting to send 16 bytes

(proc 9668): (debug)nopoll_conn.c:3691 Copying into the buffer 6 bytes
of header (total memory allocated: 23)

(proc 9668): (debug)nopoll_conn.c:3705 Mask used for this delivery:
-604747364 (about to send 22 bytes)

(proc 9668): (debug)nopoll_conn.c:3743 Bytes written to the wire 22
(masked? 1, mask: 1133566964, header size: 6, length: 16)

(proc 9668): (debug)nopoll_conn.c:3781 Write operation finished with
with last result=16, bytes_written=16, requested=16, remaining=0
(conn-id=2)

.................lock

.................unlock

.................lock

.................unlock

.................lock

(proc 9668): (debug)nopoll_conn.c:2518 === START: conn-id=2 (errno=11,
session: 6, conn->handshake_ok: 1, conn->pending_ssl_accept: 0) ===

(proc 9668): (critical) nopoll_conn.c:1876 unable to readn=2, error
code was: 110 (Connection timed out) (shutting down connection)

(proc 9668): (debug)nopoll_conn.c:1436 shutting down connection id=2
(session: 6, role: client)

(proc 9668): (critical) nopoll_conn.c:2695 Received connection close,
finishing connection session

(proc 9668): (debug)nopoll_conn.c:1436 shutting down connection id=2
(session: -1, role: client)

.................unlock

.................lock

.................lock
**************************************

The problem happens because the function nopoll_ctx_foreach_conn locks
the mutex and function nopoll_ctx_unregister_conn tries to lock the
mutex again.