Add shutdown tracing for nwserv, nwbind and ncpserv

src/ncpserv.c

@@ -428,6 +428,7 @@ static void ncp_response(int type, int sequence,
 
 static void close_all(void)
 {
+  XDPRINTF((1,0, "close_all: entering ncpserv shutdown"));
   int k=0;
   while (k++ < count_connections) clear_connection(k);
   kill_connections();
@@ -437,7 +438,7 @@ static void close_all(void)
   if (ncp_fd > -1) {
     t_unbind(ncp_fd);
     t_close(ncp_fd);
-    XDPRINTF((2,0, "LEAVE ncpserv"));
+    XDPRINTF((1,0, "LEAVE ncpserv"));
     ncp_fd = -1;
     if (ipx_out_fd > -1) {
       t_unbind(ipx_out_fd);
@@ -450,6 +451,7 @@ static void close_all(void)
 static int server_is_down=0;
 static void sig_quit(int isig)
 {
+  XDPRINTF((1, 0, "sig_quit: ncpserv got signal=%d", isig));
   server_is_down++;
 }
 
@@ -514,8 +516,10 @@ static int handle_ctrl(void)
 
       case 0xffff : /* server down */
         data_len = xread(ipxd, &offs, (uint8*)&conn, sizeof(int));
-        if (sizeof(int) == data_len && conn == what)
+        if (sizeof(int) == data_len && conn == what) {
+          XDPRINTF((1, 0, "handle_ctrl: ncpserv got CTRL 0xffff, conn=0x%x", conn));
           server_goes_down++;
+        }
         break;
 
       default : break;

src/nwbind.c

@@ -2067,8 +2067,10 @@ static void handle_ctrl()
 
       case 0xffff : /* server down */
         data_len = xread(&ipxd, &offs, (char*)&conn, sizeof(int));
-        if (sizeof(int) == data_len && conn == what)
+        if (sizeof(int) == data_len && conn == what) {
+           XDPRINTF((1, 0, "handle_ctrl: nwbind got CTRL 0xffff, conn=0x%x", conn));
            sent_down_message();
+        }
         break;
       default : break;
     } /* switch */
@@ -2081,6 +2083,7 @@ static void handle_ctrl()
 static int got_sig=0;
 static void sig_handler(int isig)
 {
+  XDPRINTF((1, 0, "sig_handler: nwbind got signal=%d", isig));
   got_sig=isig;
   signal(isig, sig_handler);
 }
@@ -2227,11 +2230,15 @@ int main(int argc, char *argv[])
     }
     if (got_sig == SIGHUP) {
      /* here I update some Bindery stuff from nwserv.conf */
+      XDPRINTF((1, 0, "main: nwbind handling SIGHUP"));
       reinit_nwbind();
       got_sig = 0;
     } else if (got_sig == SIGUSR2) {
+      XDPRINTF((1, 0, "main: nwbind handling SIGUSR2"));
       handle_usr2();  /* mst:25-Apr-00 */
       got_sig = 0;
+    } else if (got_sig == SIGQUIT) {
+      XDPRINTF((1, 0, "main: nwbind leaving main loop on SIGQUIT"));
     }
   }
   
@@ -2251,6 +2258,6 @@ int main(int argc, char *argv[])
   internal_act=1;
   nw_exit_dbm();
   xfree(connections);
-  XDPRINTF((2,0, "LEAVE nwbind"));
+  XDPRINTF((1,0, "LEAVE nwbind"));
   return(0);
 }

src/nwserv.c

@@ -24,6 +24,8 @@
 #ifdef LINUX
 #  include <netdb.h>
 #endif
+#include <sys/wait.h>
+#include <unistd.h>
 
 uint32     internal_net  = 0x0L;     /* NETWORKNUMMER INTERN (SERVER) */
 int        no_internal   =   0;      /* no use of internal net        */
@@ -1088,24 +1090,61 @@ static void close_all(void)
 
   if (pid_ncpserv > 0) {
     int status;
+    int tries = 20;
     if (fd_ncpserv_in  > -1)  {
       close(fd_ncpserv_in);
       fd_ncpserv_in = -1;
     }
+    XDPRINTF((1, 0, "close_all: sending SIGQUIT to ncpserv pid=%d", pid_ncpserv));
     kill(pid_ncpserv, SIGQUIT);  /* terminate ncpserv */
-    waitpid(pid_ncpserv, &status, 0);
-    kill(pid_ncpserv, SIGKILL);  /* kill ncpserv */
+    while (tries-- > 0) {
+      pid_t r = waitpid(pid_ncpserv, &status, WNOHANG);
+      if (r == pid_ncpserv) {
+        XDPRINTF((1, 0, "close_all: ncpserv pid=%d exited after SIGQUIT", pid_ncpserv));
+        pid_ncpserv = -1;
+        break;
+      }
+      if (r < 0) {
+        XDPRINTF((1, 0, "close_all: waitpid for ncpserv pid=%d failed", pid_ncpserv));
+        pid_ncpserv = -1;
+        break;
+      }
+      usleep(100000);
+    }
+    if (pid_ncpserv > 0) {
+      XDPRINTF((1, 0, "close_all: ncpserv pid=%d did not exit, sending SIGKILL", pid_ncpserv));
+      kill(pid_ncpserv, SIGKILL);  /* kill ncpserv */
+      waitpid(pid_ncpserv, &status, 0);
+    }
   }
 
   if (pid_nwbind > 0) {
     int status;
+    int tries = 20;
     if (fd_nwbind_in  > -1)  {
       close(fd_nwbind_in);
       fd_nwbind_in = -1;
     }
     kill(pid_nwbind, SIGQUIT);  /* terminate nwbind */
-    waitpid(pid_nwbind, &status, 0);
-    kill(pid_nwbind, SIGKILL);  /* kill nwbind */
+    while (tries-- > 0) {
+      pid_t r = waitpid(pid_nwbind, &status, WNOHANG);
+      if (r == pid_nwbind) {
+        XDPRINTF((1, 0, "close_all: nwbind pid=%d exited after SIGQUIT", pid_nwbind));
+        pid_nwbind = -1;
+        break;
+      }
+      if (r < 0) {
+        XDPRINTF((1, 0, "close_all: waitpid for nwbind pid=%d failed", pid_nwbind));
+        pid_nwbind = -1;
+        break;
+      }
+      usleep(100000);
+    }
+    if (pid_nwbind > 0) {
+      XDPRINTF((1, 0, "close_all: nwbind pid=%d did not exit, sending SIGKILL", pid_nwbind));
+      kill(pid_nwbind, SIGKILL);  /* kill nwbind */
+      waitpid(pid_nwbind, &status, 0);
+    }
   }
 
 #ifdef LINUX
@@ -1142,6 +1181,7 @@ static void down_server(void)
   if (!server_down_stamp) {
     signal(SIGHUP,   SIG_IGN);
     signal(SIGPIPE,  SIG_IGN);
+    XDPRINTF((1, 0, "down_server: sending CTRL 0xffff to ncpserv and nwbind"));
     write_to_ncpserv(0xffff, 0, NULL, 0);
     write_to_nwbind( 0xffff, 0, NULL, 0);
     fprintf(stderr, "\007");
@@ -1170,7 +1210,7 @@ static void sig_quit(int rsig)
 {
   signal(rsig,   SIG_IGN);
   signal(SIGHUP, SIG_IGN);  /* don't want it anymore */
-  XDPRINTF((2, 0, "Got Signal=%d", rsig));
+  XDPRINTF((1, 0, "sig_quit: nwserv got signal=%d", rsig));
   fl_get_int|=2;
 }