CTF-practice-evening:2014-08-04

CTF-practice-evening:2014-08-04
Date	2014/08/04
Time
Location	ACTA
Type	Workshop
Contact	Melanie

Capture The Flag evening - Part 23

• 4 August, 2014 - 7 PM
• Please bring along a laptop with you!!!

General CTF Info

• See the page for the Ctf-evenings
• Link to the Tech Inc Challenge Website Scoreboard

Binary Exploitation

• Brainsmoke is talking about binary exploitation today

Examining the challenge

• objdump -d: see the disassembly, sometimes you can see symbols
  • This example has mangled C++ symbols
• From running it, the program appears to be a daemon of some kind - a Socks proxy
  • This is a proxy for TCP - we can look at the protocol details w/ Google
  • netstat -uplanet (we can see which ports are used)
• What was added between Socks4 and Socks5? (there might be a bug)
  • Authentication and connecting directly to a domain
  • Most of the fields are fixed length
  • But the domain name is a string - it could have a buffer overflow
  • There's a 1 byte name length - if you use a 1 byte length, you might end up w/ a negative number
  • If you try to read a negative number, you will try to read a lot of bytes

• We want to find out what happens when you tell the program to read and send 255 bytes
  • We want to establish a connection
  • We need to specify Socks5

Verifying the buffer overflow

• Use xxd to construct the command (converts hex code to binary, and vice-versa)
  • Example: xxd -r -p <<< 41414141410a = AAA
  • xxd -r -p <<< 050100 | nc localhost 1080 | dd bs=1 > file (authentication)
  • hexdump -C file
  • xxd -r -p <<< 05010005010003ff:python<<<'print "A"*2000') | nc localhost 1080 | dd bs=1 > file (connection request)
  • We're following along w/ the protocol here
  • It crashed - a child exited with Signal 6 (SIGABT)
  • You can use gdb to find the segfault
  • Now we send the payload - nothing crashed, so the stopped child must have the connection
  • We continue it, and we see the segfault
  • You can use 'info reg' to see the register contents
  • eax now has the value of 0x41414141!
  • Also the instruction point (eip), we can look at what it's pointing at
  • You can use: x/i $eip = call *0x8(%eax)
  • We have control over eax. So we can get it to call code that we enter!
  • x/1000i Starting from the right instruction thread
  • x/32x 0x8058b80-32 (you can see 16 bytes of crap, and then our data, and then the address that we need)
  • (You can also oftentimes see the input you provide in the dmesg output that happens during the segfault)
  • xxd -r -p <<< 05010005010003ff:python<<<'print "A"*16 + "ABCD" + "A"*2000 ') | nc localhost 1080 | dd bs=1 > file (connection request)
  • We can use Metasploit cyclic patterns for this
• He also used his own emulator w/ an elf loader
  • mimenu
  • It produces a taint tracking dump
  • We can see that there's a second area where the tainted data is stored
  • We have data at a known location - that's where we are going to want to put out code
  • We should note down the address
  • ebx is a heap pointer - if you use ASLR, then this will wind up in a different place every time

Crafting the exploit

• He's changing the commandline script into a proper python program
• He's reusing socket routines from an old exploit
• He's also finding old previously used shellcode - the sample he's using works on lots of architectures - everything afterwards will be executed as a command
• He's opening a shell - he's trying to reuse a variety of file descriptors w/ bash