Programs that set a SIGCHLD handler before calling ssh wreak havoc on the
waitpid() calls within entropy.c and ssh-rand-helper. Entropy.c should set the
SIGCHLD handler to SIG_DFL before calling ssh-rand-helper to avoid SIGCHLD
conflicts.

This was noticed when our ssh version was upgraded to 3.1.  Our local rsync
processes were failing with the following output from SSH:

Couldn't wait for child 'ls -alni /usr/adm' completion: No child processes
Couldn't wait for child 'ls -alni /usr/mail' completion: No child processes
Couldn't wait for child 'netstat -an' completion: No child processes
Couldn't wait for child 'netstat -in' completion: No child processes
Couldn't wait for child 'netstat -rn' completion: No child processes
Couldn't wait for child 'netstat -ia' completion: No child processes
Couldn't wait for child 'netstat -s' completion: No child processes
Couldn't wait for child 'netstat -is' completion: No child processes
Couldn't wait for child 'arp -a -n' completion: No child processes
Couldn't wait for child 'ifconfig -a' completion: No child processes
Couldn't wait for child 'ps -al' completion: No child processes
Couldn't wait for child 'ps -efl' completion: No child processes
Couldn't wait for child 'last' completion: No child processes
Couldn't wait for child 'sar -d' completion: No child processes
Couldn't wait for child 'uptime' completion: No child processes
Couldn't wait for child 'ipcs -a' completion: No child processes
Not enough entropy in RNG
ssh-rand-helper child produced insufficient data
unexpected EOF in read_timeout

It was discovered that rsync had set SIGCHLD to SIG_IGN prior to calling ssh.
Setting SIGCHLD to SIG_DFL prior to forking ssh-rand-helper (seed_rng() in
entropy.c) solved this problem.