fix: prevent double-free of controlChannelRequest on connection timeout

[bdasnevesKP]

Background

I'm not a C developer. I came across issue Ylianst/MeshCentral#7407 and the related issues #110, #151, and #281 about the agent leaking memory during connection retries. I used GitHub Copilot to help me trace through the code and identify the root cause.

Root Cause

In \meshcore/agentcore.c, \MeshServer_ConnectEx_NetworkError\ frees the \j\ pointer (which is the same as \agent->controlChannelRequest) but does not set \agent->controlChannelRequest = NULL\ afterward.

The issue is that \ILibWebClient_CancelRequest\ — called immediately after — executes synchronously on the chain thread (see the \ILibChain_RunOnMicrostackThread\ macro in \ILibParsers.h):

#define ILibChain_RunOnMicrostackThread(chain, handler, user) \
    if(ILibIsRunningOnChainThread(chain)==0){ \
        ILibChain_RunOnMicrostackThreadEx(chain, handler, user); \
    } else { \
        handler(chain,user);  // <-- synchronous when already on chain thread \
    }

This synchronous cancel internally calls \MeshServer_OnResponse\ with \ReceiveStatus_Complete. That function checks \agent->controlChannelRequest != NULL\ and calls \ILibMemory_Free(agent->controlChannelRequest)\ — on a pointer that was already freed moments earlier. This is a double-free, which corrupts the heap allocator's internal state and causes memory usage to grow continuously with every retry cycle (~every 20 seconds when the server is unreachable).

The Fix

One line added in \MeshServer_ConnectEx_NetworkError, right after \ILibMemory_Free(j):

agent->controlChannelRequest = NULL;

This ensures \MeshServer_OnResponse\ safely skips the cleanup block (it checks for != NULL\ first), preventing the double-free.

Caveats

• I have not been able to compile or test this change — I don't have the build environment set up.
• The logic analysis was done with AI assistance (GitHub Copilot). The reasoning is sound to the best of my understanding, but I'd strongly appreciate a review from someone familiar with the codebase before merging.
• I'm happy to update or close this PR if the analysis is wrong.

[si458]

this doesnt appear to fix the issue 😢
i build a test agent with this patch,
set 0.0.0.0 my.domain.com in the /etc/hosts file so i knew the machine would never connect
left it for 24 hours, the memory usage showed 300MB and was still increasing :(