[X2Go-Dev] debian jessie general protection trap in libNX_X11.so.6.2 caused by some apps

[Jean-Francois Dagenais]

Hi Mike,

Thanks for the reply. I was able to get a bit further since my mail. I did the 'apt-get source x2goagent' and 'debuild' to get a stack trace.

Tried the deb repo you pointed, same behaviour. Although "deb http://packages.arctica-project.org/debian-nightly/ jessie main" gave me versions "3.5.99.0-0~git20150928.1021+jessie.nightly.1", not 3.6.x. I have built 3.6.x (commit 81d92cd0fa293e44bbb063d373d9289ddc358c99) using "DEB_BUILD_OPTIONS=nocheck debuild -b -uc -us -j8", I will try the generated .debs later.

In the mean time, here's what i got from my debug effort on the version from x2go.org. The fault comes from:

Program received signal SIGSEGV, Segmentation fault.
_XData32 (dpy=dpy at entry=0x2b319b0, data=<optimized out>, data at entry=0x34580d4, len=14588, len at entry=18652) at XlibInt.c:3792
3792		    *buf++ = *data++;
(gdb) l
3787		    i = len;
3788		dpy->bufptr = (char *)buf + i;
3789		len -= i;
3790		i >>= 2;
3791		while (--i >= 0)
3792		    *buf++ = *data++;
3793	    }
3794	    return 0;
3795	}
3796	#endif /* LONG64 */

Then I did several tests. To get better debug resolution, I compiled the lib again, but with '-Og' (or -O0, I tried both). Oddly, the problem goes away! I think there's some alignment problems here: See also: https://www.mail-archive.com/misc@openbsd.org/msg129941.html

I'm really confused, If i change the "data" member to an int (32bits) then it doesn't work. I get no errors, my x11 clients just don't show up (I use xterm as a test client app). I did manage to fix it by doing:
diff --git a/nx-X11/lib/X11/XlibInt.c b/nx-X11/lib/X11/XlibInt.c
index 47e2ea1..bfcab8d 100644
--- a/nx-X11/lib/X11/XlibInt.c
+++ b/nx-X11/lib/X11/XlibInt.c
@@ -3770,11 +3770,11 @@ void Data(
 int
 _XData32(
     Display *dpy,
-    register long *data,
+    volatile long *data,
     unsigned len)
 {
-    register int *buf;
-    register long i;
+    volatile int *buf;
+    volatile long i;
 
     while (len) {
        buf = (int *)dpy->bufptr;
@@ -3785,6 +3785,7 @@ _XData32(
        }
        if (len < i)
            i = len;
+       i &= ~0x3;
        dpy->bufptr = (char *)buf + i;
        len -= i;
        i >>= 2;


I also tried to rewrite the method using memcpy but could not get it to work. It compiles and looks sound, but at execution, my windows (xterm) never appear in the server. Lack of proper debugging environment and further time to invest (too much already) means I will need to drop this for a while. Especially since now my system runs all of the gnome3 apps I've tried, so I'm no longer stalled.

P.S. I was unable to use the nxagent :1 strategy. I do get a Xephyr style black window, but when I start a client I get this:
dagenaisj at jfddesk:~$ export DISPLAY=:1
dagenaisj at jfddesk:~$ xeyes
No protocol specified
Error: Can't open display: :1

Thanks for your help, and let me know if I can further help!
/jfd