Blob Blame History Raw
From 0d89bdef1e7f698787967bffed5c413ef0dee761 Mon Sep 17 00:00:00 2001
From: Phil Sutter <psutter@redhat.com>
Date: Fri, 15 Mar 2019 17:51:28 +0100
Subject: [PATCH] libxtables: move some code to avoid cautions in vfork man
 page

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1525980
Upstream Status: iptables commit 78683093cf4f0

commit 78683093cf4f059531e5f929a4884ffaecb8411c
Author: Dan Wilder <dan.wilder@watchguard.com>
Date:   Sat Oct 25 00:51:59 2014 +0200

    libxtables: move some code to avoid cautions in vfork man page

    Running iptables-restore on an embedded platform containing no modprobe program, the following lines in xtables.c lead to corrupted stack frame:

     357     switch (vfork()) {
     358     case 0:
     359         argv[0] = (char *)modprobe;
     360         argv[1] = (char *)modname;
     361         if (quiet) {
     362             argv[2] = "-q";
     363             argv[3] = NULL;
     364         } else {
     365             argv[2] = NULL;
     366             argv[3] = NULL;
     367         }
     368         execv(argv[0], argv);
     369
     370         /* not usually reached */
     371         exit(1);

    modprobe pointed to a non-existant program /sbin/modprobe, so execv()
    always failed.  Not a problem in itself on our platform, as the kernel
    modules are pre-loaded before iptables-restore is run, but it took a
    bit of headscratching to track this down, as a stack frame was
    corrupted, leading to failures quite a while after the function
    containing this code had returned!

    Relevant caution in man 2 vfork:

        "The vfork() function has the same effect as fork(2), except that
        the behavior is undefined if the process created by vfork() either
        modifies any data ... or calls any other function before
        successfully calling _exit(2) or one of the exec(3) family of
        functions."

    Apparently this has not been a problem for us in earlier versions of
    glibc, maybe because vfork was more like fork, maybe because the
    stack corruption was innocuous.  Ours is a corner case anyway, as
    it might not have been a problem had modprobe existed or had
    modprobe been a symlink to /bin/true.  But it seems odd to disregard
    man page cautions, and our problem goes away if they are heeded.

    Signed-off-by: Florian Westphal <fw@strlen.de>

Signed-off-by: Phil Sutter <psutter@redhat.com>
---
 libxtables/xtables.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/libxtables/xtables.c b/libxtables/xtables.c
index cf9a59d5ec095..bca9863acc566 100644
--- a/libxtables/xtables.c
+++ b/libxtables/xtables.c
@@ -352,6 +352,11 @@ int xtables_insmod(const char *modname, const char *modprobe, bool quiet)
 		modprobe = buf;
 	}
 
+	argv[0] = (char *)modprobe;
+	argv[1] = (char *)modname;
+	argv[2] = quiet ? "-q" : NULL;
+	argv[3] = NULL;
+
 	/*
 	 * Need to flush the buffer, or the child may output it again
 	 * when switching the program thru execv.
@@ -360,19 +365,10 @@ int xtables_insmod(const char *modname, const char *modprobe, bool quiet)
 
 	switch (vfork()) {
 	case 0:
-		argv[0] = (char *)modprobe;
-		argv[1] = (char *)modname;
-		if (quiet) {
-			argv[2] = "-q";
-			argv[3] = NULL;
-		} else {
-			argv[2] = NULL;
-			argv[3] = NULL;
-		}
 		execv(argv[0], argv);
 
 		/* not usually reached */
-		exit(1);
+		_exit(1);
 	case -1:
 		free(buf);
 		return -1;
-- 
2.21.0