Random failures starting clamav-milter [patch]

Describe the bug

Seeming random failures for clamav-milter to start; hangs in startup.

clamav 0.103.7, but this is a long-standing bug.

How to reproduce the problem

Unfortunately, that's tough.

I happened to move clamav-milter to a machine where the timing made it 100% reproducible; previously the failures were very rare.

The issues are dependent on timing and environment.

• daemonize_parent_wait() forks, then sets up a signal handler so the child can signal initialization complete. If the child sends the signal first, the parent (usually) exits due to the unhandled signal, not the planned wait.
• Worse, daemonize_signal_parent() drops privileges before sending the complete signal. Now the unpriv'd child tries to signal the still priv'd parent, and the kill() fails silently. In this case, the parent hangs forevever.
• Finally, if the child doesn't exit but dies from a signal while the parent waits, the caller in the parent doesn't get an error return, and things get weird.

The attached patch resolves these issues.

• The child now signals the parent just before dropping privileges, so the kill() will succeed.
• The parent sets up its signal handler before fork(), resolving the race with the child.
• The proper error error code is returned if the child dies unexpectedly while the parent is waiting for it to initialize.

Attachments

milter-hang.patch