Re: another make check failure on PPC

[Patrick Smith]

Andreas Jaeger <address@hidden> wrote:

> Ethan <address@hidden> writes:
> > make check (this time on a PPC-750 UP) fails with:
> >
> > /tmp/glibc-build/io/test-lfs  > /tmp/glibc-build/io/test-lfs.out
> > Timed out: killed the child process but it exited 0
> >
> > This system is also running gcc-3.2, binutils 2.13.90.0.4, kernel
> > ~2.4.19.  I assume lfs is `large files' ?
>
> Yes.
>
> it seems that the underlying filesystem is too slow or broken,

I had this error; in my case the filesystem was just too slow for this 
test.  (Reiserfs on a PPC G4 @ 450 Mhz or thereabouts).  Attached is a 
small patch that increases the timeout used in the test.
--
address@hidden

--- login-utils/simpleinit.c.orig       2001-09-29 11:09:10.000000000 -0400
+++ login-utils/simpleinit.c    2002-05-23 22:16:07.000000000 -0400
@@ -203,6 +203,18 @@
                if ( ( initctl_fd = open (initctl_name, O_RDWR, 0) ) < 0 )
                        err ( _("error opening fifo\n") );
        }
+        if ( initctl_fd >= 0 )
+                if ( fcntl (initctl_fd, F_SETFD, FD_CLOEXEC) != 0 ) {
+                        err ( _("error setting close-on-exec on /dev/initctl") 
);
+                        /* Can the fcntl ever fail?  If it does, and we leave
+                           the descriptor open in child processes, then any
+                           process on the system will be able to write to
+                           /dev/initctl and have us execute arbitrary commands
+                           as root. So let's refuse to use the fifo in this
+                           case. */
+                        close(initctl_fd);
+                        initctl_fd = -1;
+                }
 
        if ( want_single || (access (_PATH_SINGLE, R_OK) == 0) ) do_single ();