When we linked a new entry on to the global request queue, we forgot
to set its next pointer to NULL, so that when it was removed again,
s->globreq_head could end up pointing to nonsense.
In addition, even if the next pointer happened to be NULL by luck, we
also did not notice that s->globreq_head had become NULL and respond
by nulling out s->globreq_tail, which would leave s->globreq_tail as a
stale pointer to the just-freed list element, causing a memory access
error on the next attempt to link something on to the list.
This could come up in the situation where you open Change Settings and
configure a remote port forwarding, close it (so that the global
request is sent, queued, replied to, and unqueued again), and then
reopen Change Settings and configure a second one (so that the linked
list in the confused state actually gets used).
03e71efcc5