If a command we run times out, the machinery in cmd() will arrange
for the ts-* script to spot the timeout, and stop waiting for it.
However it is also necessary for the command we ran to die. It has a
copy of the owner daemon fd, so if it doesn't, our resources won't get
freed. In sufficiently exciting bugs, our allocation might continue
indefinitely, while a subprocess of ours hangs on after we are long
gone.
timeout(1) does not print a message when the process times out (!) So
we can't do away with the logic in cmd(). We set the timeout(1)
timeout to 30s more than our own timeout, so that cmd() will time
out first and print a message.
We could use alarm(1) as we do in Osstest/Serial/sympathy.pm but that
program isn't packaged and its unsophisticated approach is not really
appropriate for arbitrary nonconsenting programs.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
sub tcmdex {
my ($timeout,$stdout,$cmd,$optsref,@args) = @_;
logm("executing $cmd ... @args");
- my $r= cmd($timeout,$stdout, $cmd,@$optsref,@args);
+ # We use timeout(1) as a backstop, in case $cmd doesn't die. We
+ # need $cmd to die because we won't release the resources we own
+ # until all of our children are dead.
+ my $r= cmd($timeout,$stdout, 'timeout',$timeout+30, $cmd,@$optsref,@args);
$r and die "status $r";
}