Today I had a look at the OpenSSH sources again. It seems to me that they overdo the whole PrivSep thing.
Its quite complicated to follow all the calls across
processes and privileges. Do you think they are bug-free?
I made a call-graph which honours all the different
UIDs and PIDs. I used the graphviz package which
core-dumped once in a while due to the complexity. I know there are
better tools for graphing. You can read a paper
describing my graph-tool using graphviz here or download the
complete OpenSSH 4.6 daemon graph here.
I just substituted the PDF with a new one. Some
nodes were located within the wrong cluster. It turned
out that one pass (of data recording) is not enough to properly cluster the graph. Its also quite hard to
verify that the nodes are linked correctly. I hope
it is this time! :-) I want to encourage you to submit comments if you feel something is wrong or you (dis-)like