Comment # 15 on bug 1107197 from Martin Wilck

I've filed a bug upstream, as we have only very few very harmless patches for
NM in factory.

https://bugzilla.gnome.org/show_bug.cgi?id=797136

On comment 13:

All but the last trace shows libcurl4-internal callbacks accessing data freed
in 
nm_connectivity_check_cancel(). The last one also occurs in a libcurl callback
related to connectivity checking, but accesses a block freed elsewhere.

That kind-of confirms my suspicion from comment 8 - the new asynchronous
connectivity checking code seems to be susceptible to race conditions, in
particular if nm_connectivity_check_cancel() is called. Unfortunately I've no
time to dig into that deeper now.