[distcc] Problem running from init

[Martin Pool]

On 27 May 2003, George Garvey <tmwg-distcc at inxservices.com> wrote:
>    Just switched from inetd to init on Linux 2.4.20, glibc 2.3.2.
> setpgid fails under these conditions. Removed it, and everything's
> happy. Perhaps I'm not using the right set of options.

Thanks for reporting it.  The options look reasonable, but there was a
bug in distcc that made it fail when run by init.

> Index: inX.16/src/dparent.c
> +++ inX.17/src/dparent.c Tue, 27 May 2003 03:38:11 -0700 tmwg (distcc/35_dparent.c 1.13 640)
> @@ -126,8 +126,10 @@ int dcc_standalone_server(void)
>      } else {
>          /* Still create a new process group, even if not detached */
>          rs_trace("not detaching");
> +/*
>          if ((ret = dcc_setpgid(0, 0)) != 0)
>              return ret;
> +*/
>          dcc_save_pid(getpid());
>      }

That fixes init but would break other uses of --no-detach, such as
running from a debugger..  It's better to just insert a check beforehand:

--- exec.c	27 May 2003 08:16:34 -0000	1.90
+++ exec.c	28 May 2003 04:42:45 -0000	1.91
@@ -187,14 +187,21 @@ static int dcc_inside_child(char **argv,
 }
 
 
-int dcc_setpgid(pid_t pid, pid_t pgid)
+int dcc_new_pgrp(void)
 {
-    if (setpgid(pid, pgid) == 0) {
-        rs_trace("setpgid(%ld, %ld)", (long) pid, (long) pgid);
+    /* If we're a session group leader, then we are not able to call
+     * setpgid().  However, setsid will implicitly have put us into a new
+     * process group, so we don't have to do anything. */
+    if (getpgid(0) == getpid()) {
+        rs_trace("already a process group leader");
+        return 0;
+    }
+    
+    if (setpgid(0, 0) == 0) {
+        rs_trace("entered process group");
         return 0;
     } else {
-        rs_trace("setpgid(%ld, %ld) failed: %s",
-                 (long) pid, (long) pgid, strerror(errno));
+        rs_trace("setpgid(0, 0) failed: %s", strerror(errno));
         return EXIT_DISTCC_FAILED;
     }
 }

-- 
Martin