[BusyBox] patch: awk segfaults on export of empty env vars

[Keith Smith]

Hi

Using 1.00-pre2, I have found that awk segfaults when empty environment variables are exported.

# export x=''
# awk
SIGSEGV

This was independently discovered elsewhere, and resulted in the following change to awk in CVS:
http://www.busybox.net/cgi-bin/cvsweb/busybox/editors/awk.c?r1=1.9&r2=1.10

However, I believe that the core problem lies with ash.  It maintains the environment as an array of strings of the form 'name=value', but this format is not respected by setvar() as invoked by the export builtin.  To demonstrate, note the different results for x and y:

~ # env
USER=root
HOME=/
LOGNAME=root
TERM=vt102
PATH=/usr/sbin:/bin:/usr/bin:/sbin
SHELL=/bin/sh
PWD=/

# y=''
# export y
# env
USER=root
HOME=/
LOGNAME=root
TERM=vt102
PATH=/usr/sbin:/bin:/usr/bin:/sbin
SHELL=/bin/sh
y=
PWD=/

# export x=''
# env
USER=root
HOME=/
x
LOGNAME=root
TERM=vt102
PATH=/usr/sbin:/bin:/usr/bin:/sbin
SHELL=/bin/sh
y=
PWD=/


So while awk _should_ be checking its inputs, it's not unreasonable to expect that _all_ entries in the environment will conform to the 'name=value' convention.  I propose the following patch, which works fine in my testing, to resolve the issue.

Index: busybox-1.00-pre2/shell/ash.c
===================================================================
--- busybox-1.00-pre2/shell/ash.c	(revision 1189)
+++ busybox-1.00-pre2/shell/ash.c	(working copy)
@@ -11843,9 +11843,8 @@
 	}
 	INTOFF;
 	p = mempcpy(nameeq = ckmalloc(namelen + vallen + 2), name, namelen);
-	*p++ = '\0';
+	*p++ = '=';
 	if (vallen) {
-		p[-1] = '=';
 		p = mempcpy(p, val, vallen);
 	}
 	*p = '\0';


Can anybody think of a reason why this would be a bad idea?

Regards
Keith Smith