workaround for race between waitpid(-1) and waitpid(pid)

Sometimes waitpid(-1) in cb_check_data reaps a child that has just terminated,
before waitpid(pid) in wait_process can reap it.  Linux waitpid(pid,,WNOHANG)
may not return an error if the child has already terminated, which results in
wait_process looping waiting for the child which has already terminated.  The
symptom is a spurious "timeout waiting for filter command to exit".

1 files changed, 5 insertions, 0 deletions

diff --git a/src/proxsmtpd.c b/src/proxsmtpd.c
index 278ecee..8f50d96 100644
--- a/src/proxsmtpd.c
+++ b/src/proxsmtpd.c
@@ -925,6 +925,11 @@ static int wait_process(spctx_t* sp, pid_t pid, int* status)
         switch(waitpid(pid, status, WNOHANG))
         {
         case 0:
+	    /* Linux may return 0 if the task has already terminated and was
+	     * caught by waitpid(-1) above, double check it still exists.
+	     */
+            if (kill(pid, 0) < 0 && errno == ESRCH)
+                return 0;
             break;
         case -1:
             if(errno != ECHILD && errno != ESRCH)