Re: Linux 2.6.9-rc2 : oops

[Andrew Morton <akpm@osdl.org>]

Linus Torvalds <torvalds@osdl.org> wrote:
>
>  Looks like somebody is trying to free an invalid address. Sadly, your 
>  traceback doesn't show _who_, because it's hidden in the buffering.

yes, I doubt if we get synchronous notification of a double-free without
CONFIG_DEBUG_SLAB.

There's a known double-free in the isofs filesystem.  Christian, were you
using CDROMs at the time?

>  The only changes to slab itself have been by Christoph lately, I don't 
>  think that matters. Can you enable slab debugging? That should catch it 
>  much earlier..

Yup.  Enable CONFIG_DEBUG_SLAB.



Invalidate this pointer so it doesn't get freed twice.

Signed-off-by: Andrew Morton <akpm@osdl.org>
---

 25-akpm/fs/isofs/rock.c |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)

diff -puN fs/isofs/rock.c~rock-fix fs/isofs/rock.c
--- 25/fs/isofs/rock.c~rock-fix	2004-09-10 01:47:00.135392480 -0700
+++ 25-akpm/fs/isofs/rock.c	2004-09-10 01:47:00.139391872 -0700
@@ -62,7 +62,7 @@
 }                                     
 
 #define MAYBE_CONTINUE(LABEL,DEV) \
-  {if (buffer) kfree(buffer); \
+  {if (buffer) { kfree(buffer); buffer = NULL; } \
   if (cont_extent){ \
     int block, offset, offset1; \
     struct buffer_head * pbh; \
_