Use the attached test script for checking the case where the global limit applies. Make sure it runs on a system with `hw.ncpu=1` and `net.inet.tcp.reass.maxsegments=83`, the minimal value. This script reproduces the issue reported by jhb@.
This is the problem based on my analysis: When TCP receives out of order segments it limits the resources being used. There is a global limit on the number of queue entries. If the limit is reached, no new queue entries are allocated. This applies currently also to segments with `SEG.SEQ==RCV.NXT`, which results in stalled TCP connections.
{F4076240}