Re: crash on GCJ_PROPERTIES="foo=" program

[Paul Bolle <pebolle at tiscali dot nl>]

0) This is more of a bug report then a patch. (I do hope this is the
correct list to report it. Otherwise I'll gladly forward this message to
gcc-patches@gcc.gnu.org or somewhere else.)

1) I've been playing with gcj recently (as included in gcc-3.4.2, in the
version shipped with Fedora Core 3).

2) I noticed that programs compiled by gcj would segfault when given a
GCJ_PROPERTIES environment variable with an empty value, for example:

GCJ_PROPERTIES="foo=" program

3) Looking at the code of libjava/prims.cc I noticed next_property_value
doesn't seem to handle the length of an empty (zero length) value
correctly. In that case it seems to reuse the length of the key part of
that key/value pair. I guess that this might cause a crash (when null
terminating the key/value pairs later). Please find attached a (trivial)
patch, written on that idea.

4) However, I didn't manage to test this patch as recompiling etc. gcc
and friends ended in a "/usr/bin/ld: Argument list too long" (after
which I gave up my attempt at correcting this myself). But maybe someone
else might like to review my analysis and maybe even test my patch.

5) Well, at least you have a bug report now ...

Kind regards,


Paul Bolle

--- libjava/prims.cc.oud	2005-02-26 00:27:53.282039752 +0100
+++ libjava/prims.cc	2005-02-26 00:42:50.913579016 +0100
@@ -779,6 +779,7 @@
 next_property_value (char *s, size_t *length)
 {
   size_t l = 0;
+  *length = 0;
 
   JvAssert (s);