abstract_tcp_server2: avoid deadlock waiting for send queue to drain

If we reach the send queue size limit, we need to release the lock,
or we will deadlock and it will never drain.

If we reach that limit, it's likely there's another problem in the
first place though, so it will probably not drain in practice either,
unless some kind of transient network timeout.
This commit is contained in:
moneromooo-monero 2016-04-17 11:45:38 +01:00
parent 7450656bf3
commit 6bca9a8ef4
No known key found for this signature in database
GPG key ID: 686F07454D6CEFC3

View file

@ -490,7 +490,9 @@ PRAGMA_WARNING_DISABLE_VS(4355)
sleep_before_packet(cb, 1, 1);
}
epee::critical_region_t<decltype(m_send_que_lock)> send_guard(m_send_que_lock); // *** critical ***
m_send_que_lock.lock(); // *** critical ***
epee::misc_utils::auto_scope_leave_caller scope_exit_handler = epee::misc_utils::create_scope_leave_handler([&](){m_send_que_lock.unlock();});
long int retry=0;
const long int retry_limit = 5*4;
while (m_send_que.size() > ABSTRACT_SERVER_SEND_QUE_MAX_COUNT)
@ -504,11 +506,12 @@ PRAGMA_WARNING_DISABLE_VS(4355)
long int ms = 250 + (rand()%50);
_info_c("net/sleep", "Sleeping because QUEUE is FULL, in " << __FUNCTION__ << " for " << ms << " ms before packet_size="<<cb); // XXX debug sleep
m_send_que_lock.unlock();
boost::this_thread::sleep(boost::posix_time::milliseconds( ms ) );
m_send_que_lock.lock();
_dbg1("sleep for queue: " << ms);
if (retry > retry_limit) {
send_guard.unlock();
_erro("send que size is more than ABSTRACT_SERVER_SEND_QUE_MAX_COUNT(" << ABSTRACT_SERVER_SEND_QUE_MAX_COUNT << "), shutting down connection");
// _dbg1_c("net/sleep", "send que size is more than ABSTRACT_SERVER_SEND_QUE_MAX_COUNT(" << ABSTRACT_SERVER_SEND_QUE_MAX_COUNT << "), shutting down connection");
close();