Closed Bug 164695 Opened 22 years ago Closed 22 years ago

Heap corruption in libjar using manifest length -1


(Core :: Security, defect)

Not set





(Reporter: security-bugs, Assigned: dveditz)



(Whiteboard: [sg:blocker])


(2 files)

Submitter name:                zen-parse
   Submitter email address:
   Acknowledgement checkbox:      on
   Product:                       Netscape 6.x
   Operating system:              Unix: x86 Linux
   OS version:                    Redhat 7.0
   Issue summary:                 nsJAR::LoadEntry - heap corruption - code

Issue details:
By supplying a malformed jar file, where the 
manifest file is described as being (unsigned 
int)-1 bytes long, it is possible to cause 
exploitable heap corruption.

(This is not the same as the other jar 

[root@clarity secjars]# unzip -v jar.jar 
Archive:  jar.jar
 Length   Method    Size  Ratio   Date   Time 
  CRC-32    Name
--------  ------  ------- -----   ----   ---- 
  ------    ----
     268  Defl:X      199  26%  08-27-02 
04:21  18dc0dd8  page.html
      56  Stored 4294967295 -7669483%  
08-27-02 06:22  8cb14d9d  META-INF/
--------          -------  ---                
     324              198  39%                
            2 files


R::LoadEntry(const char* aFilename, char** 
aBuf, PRUint32* aBufLen)
  //-- Read the manifest file into memory
  char* buf;
  PRUint32 len;
  rv = manifestStream->Available(&len);
length now contains 4294967295 (== (int)-1)
  if (NS_FAILED(rv)) return rv;
  buf = (char*)PR_MALLOC(len+1);
allocated (-1+1)==0 bytes. 
(PR_MALLOC converts that into 1 byte)
  if (!buf) return NS_ERROR_OUT_OF_MEMORY;
  PRUint32 bytesRead;
  rv = manifestStream->Read(buf, len, 
and now it reads the rest of the file into 
that buffer.

There is now a heap based overflow

It is exploitable in a similar manner to the 
other jar problem I submitted.

(There is a similar problem in 
mozilla/security/nss/lib/jar/jarfile.c, but I 
am not sure how (if it is possible) to 
trigger it.)

This form was submitted from
with Mozilla/4.0 (compatible; MSIE 5.0; Linux) Opera 6.03  [en].
I can see how you've corrupted the heap, but I don't see how you turned that
into an exploit.
Attached file jar-exploit-2.tar.gz
Demo from zen-parse
This is easy to fix and it must be fixed for 1.2 final.

Dan, heap corruption can be turned into an exploit in some cases. It's a ton of
work but we can't discount it.
Whiteboard: [sg:blocker]
Anyone working on this?  This is a 1.2 blocker.
Comment on attachment 104404 [details] [diff] [review]
prevent attack (plus fix unrelated nearby memory leak)

Dan, is -1 the only possible value that can be exploited? Is it ever legitimate
for this value to be negative? We might want to be extra safe and check for any
nonsensical value, not just -1.

If that's the only value that could possibly cause  heap corruption, then
Attachment #104404 - Flags: review+
The size value is not actually -1, it's unsigned. When you add 1 to it (to
account for the null) it overflows; any other large value is OK.

Would be nice to re-write this to keep track of the file size rather than rely
on null termination.
Comment on attachment 104404 [details] [diff] [review]
prevent attack (plus fix unrelated nearby memory leak)

>Index: nsJAR.cpp

>+  if (len == 0xFFFFFFFF)
>+    return NS_ERROR_FILE_CORRUPTED; // bug 164695

nit: how about PRUint32(-1) instead?  that way no one has to count digits :-/

either way, sr=darin.
Attachment #104404 - Flags: superreview+
This should go on the branch.
Comment on attachment 104404 [details] [diff] [review]
prevent attack (plus fix unrelated nearby memory leak)

a=roc+moz for trunk
Attachment #104404 - Flags: approval+
Discussed in bBird team meeting.  Plussing for adt only.
Keywords: adt1.0.2adt1.0.2+
Checked into trunk
Closed: 22 years ago
Resolution: --- → FIXED
a=chofmann 1.0.2
Checked into branch
Verified on 2002-11-20-branch build on Linux. Loaded the attached test case and
the crash does not happen.The page shows up with the line streaks.
Group: security
You need to log in before you can comment on or make changes to this bug.