On Fri, Oct 21, 2016 at 01:26:36PM +0200, Jeremie Courreges-Anglas wrote: > Rafael Zalamena <rzalam...@gmail.com> writes: > > On Fri, Oct 14, 2016 at 06:47:09PM +0200, Rafael Zalamena wrote: > >> On Mon, Sep 26, 2016 at 03:45:59PM +0200, Rafael Zalamena wrote: > >> ---snip--- > > > > I got feedback from jca@ that the trap handler wasn't working, so after > > trying to reproduce the problem myself I found one 'env' global variable > > that was not being set and the child process was dying silently. > > (thanks jca@ !) > > > > Instead of depending on snmpe.c:snmpe env initialization (p_init), I'm > > now calling smi_setenv() to do that in the main() function so all children > > get the same behaviour. Also note that we don't have an 'extern' env in > > smi.c anymore. > > > > ok? > > Works fine here, but then I don't understand the relationship between > static struct snmpd *env in smi.c and struct snmpd *env in snmpe.c.
smi.c had a "extern struct snmpd *env" and that variable was only being set during the snmpe initialzation (p_init). Since with fork+exec the child process runs entirely from scratch (no memory / socket sharing with the parent process), we need to set it somewhere else. It is a known problem that everyone that used to set things in the p_init and expected it to work for everyother process was wrong. sunil@ found this the hard-way when he found out that p_env wasn't being set for his process and he noticed that now p_init is ran in the child process already. Before fork+exec the p_init() functions were run by the parent process. To fix the current problem I made the 'env' for smi.c to be a local file global variable and set it in the main() process for every child.
