Blame SOURCES/make-4.3-cloexec.patch

e896df
From d79fe162c009788888faaf0317253b6f0cac7092 Mon Sep 17 00:00:00 2001
e896df
From: Kevin Buettner <kevinb@redhat.com>
e896df
Date: Thu, 23 Apr 2020 17:05:34 -0400
e896df
Subject: [SV 58232] Disable inheritance of jobserver FDs for recursive make
e896df
e896df
A parent make will invoke a sub-make with close-on-exec disabled for
e896df
the jobserver pipe FDs.  Force close-on-exec to be to be enabled in
e896df
the sub-make so the pipe is not always passed to child jobs.
e896df
e896df
I have a test case which, when invoked with a suitable -j switch,
e896df
will hang if the recipe inherits the jobserver pipe.  This test case
e896df
was inspired by a real world case in which testing GDB on Fedora
e896df
would hang due to some poorly written test GDB cases having been
e896df
passed the jobserver file descriptors.
e896df
e896df
* src/posixos.c (jobserver_parse_auth): Call fd_noinherit() for
e896df
jobserver pipe descriptors.
e896df
e896df
Copyright-paperwork-exempt: yes
e896df
e896df
diff --git a/src/posixos.c b/src/posixos.c
e896df
index 525f292c..eab175a4 100644
e896df
--- a/src/posixos.c
e896df
+++ b/src/posixos.c
e896df
@@ -145,6 +145,11 @@ jobserver_parse_auth (const char *auth)
e896df
   /* When using pselect() we want the read to be non-blocking.  */
e896df
   set_blocking (job_fds[0], 0);
e896df
 
e896df
+  /* By default we don't send the job pipe FDs to our children.
e896df
+     See jobserver_pre_child() and jobserver_post_child().  */
e896df
+  fd_noinherit (job_fds[0]);
e896df
+  fd_noinherit (job_fds[1]);
e896df
+
e896df
   return 1;
e896df
 }
e896df