connectionmanager: avoid deadlock on cinfo->mtx_

cinfo->mtx_ MUST be locked before socketsMutex, so it SHOULD never
be locked in a callback. Retrigger the lock in its own thread avoid
a deadlock there.

GitLab: #20
Change-Id: If17af7db81d8ca2b990629446b5d78dc11a785b1

src/connectionmanager.cpp

@@ -1062,10 +1062,13 @@ ConnectionManager::Impl::sendChannelRequest(const std::weak_ptr<DeviceInfo>& din
         [dinfow, cinfow, wSock = std::weak_ptr(channelSock), name, vid](bool accepted) {
             if (auto dinfo = dinfow.lock()) {
                 dinfo->executePendingOperations(vid, accepted ? wSock.lock() : nullptr, accepted);
+                // Always lock top-down cinfo->mutex
+                dht::ThreadPool::io().run([cinfow, vid]() {
                     if (auto cinfo = cinfow.lock()) {
                         std::lock_guard<std::mutex> lk(cinfo->mutex_);
                         cinfo->cbIds_.erase(vid);
                     }
+                });
             }
         });