Heap Overflows
==============

* what's a heap?
  memory that dynamically allocated by application
    (un-)initialized globals, dynamically allocated, ...
  sampe code:    
    int i = 3;
    int j;
    
    void f ()
    {
      int x = 4;
      int *p = malloc (sizeof(*p));
    }
  where is i, j, f, x, and p located?
  heap grows upwards
  
Part I: data and bss

* what does a heap overflow look like?
  read the code on page 3
    try it, and see what the output is on your machine
  what's the heap layout?
    what's in the 0x18-0x10 bytes?
  how to apply this to a BSS overflow?
  
* what does a BSS overflow look like?
  read the code on page 4
    try it, and see what the output is on your machine
  and then what about there is a file pointer (file name) in the heap
    e.g., read the vulnerable code on page 5
      how to overflow "buf"?
      what to be overflowed in to "tmpfile"?
        an address
      how to find this address (for a string)?
        user input (say argv[2], where is it? how does on locate it?)
        or env. vars.
      the final attack string argv[1] looks like:
        "A...A0xdeadbeef"
      code on page 6 achieve this goal
  
* note that one does not need an executable heap for now
  but the heap is executable indeed!

* of course, file pointer is not the only source for heap overflows, 
  let's turn to function pointers, what's the vulnerability in 
  code on top of page 9?
    what should be written into "funcptr"?
      one can "call-into-libc"
        recall how we locate the function "system" (page 10)
      one can "call-into-argv" (page 11)
        then shellcode needs an executable stack
      or anywhere that contains some code

* another common vulnerability comes from the jmp_bufs, what are they?
  external variables holding machine context
    "setjmp" saves machine state (stack pointers, eip, callee-saved, etc.)
    "longjmp" restores machine state later
    poor-man's exception handling in C
  they are standard C library functions (<stdjmp.h>),
  and read the sample code on page 12

* summary so far:
  file pointers
  external function pointers
  jump buffers
  but these are not exhausted
    C++ virtual function, and others on bottom on page 15
    
Part II: dynamic heap

* what's a dynamic heap?
  applications need to allocate memory at runtime
    OSes offer coarse utilities
      brk, sbrk, or mmap, etc.
    (C) libraries implement fine-grained interefaces via "malloc" families
      malloc(), free(), calloc(), etc.
      organizing the memory region returned by brk or mmap, as some
      kind of data structures

* what's under the hood of "malloc" or "free"?
  typical C library implementation (say glibc) are too complex to be 
  discussed in a class hour
  so, we start with a dirty simple allocator
    read the question 
  this allocator is loosely based on section 8.7 of K&R
    free list + first fit
  what happened on "malloc(0)"?
  how to hack this memeory allocator?
  what about changing the field "size"?
  what about "free()" a pointer twice?
  can you cook some code which achieves "write-anything-anywhere"?
  
* this algorithm is slow, so can not be used in production code
    both malloc and free are O(n)
  so some advanced algorithms are deployed (but with same key ideas)
    on page 2, the paper describes a splay-tree-based algorithm in system V
  from page 7, the paper discusses an algorithm (called Douge Lee allocator) 
  used in (an older version of) Linux GNU glibc malloc
  with the same theme of hacking the (meta) memory management information