On Mon, Jul 06, 2009 at 12:56:10AM +0200, Enrico Zini wrote: > What seems needed here is to have the supervisor to detect that the > child has been killed with signal 15, and in that case just quit. > > It should be easy enough to do, and quite ok because if I kill, for > example, matchbox, nodm/xsession won't exit as killed with signal 15 and > so the nodm/supervisor will properly restart it as it should.
Unfortunately, when the session ends xinit does not return the exit code of the session command, but it just returns 0. I don't see a way for the monitor part of nodm to detect that the session part of nodm has been killed with signal 15, unless I trap it in the session part and write a note on a file, which is something that would create more trouble than it would solve, because then there needs to be extra machinery to detect stale files. Alternatively, I can just ignore signals in the session part, so that if you do "killall nodm" you will not activate the session but only the monitor, as you would expect. The current session code is inherited from su and does several things with signals, like forwarding of SIGSTOP and SIGCONT, but the more I look into it, the more I find it unnecessary and I can just block everything. However, I still need one unblocked signal that the monitor can use to tell the session to exit: SIGTERM would have been the obvious choice, but if "killall nodm" is the use case, then we are back to square one. Ciao, Enrico -- GPG key: 4096R/E7AD5568 2009-05-08 Enrico Zini <enr...@enricozini.org>
signature.asc
Description: Digital signature
