kung-fu Chambly/Marieville

Kung-fu Chambly

Interators are usually implemented using signed integers like the typical "for (int i=0; ..." and in fact is the type used indexing "cstr[i]", most of methods use the signed int, int by default is signed.
Nevertheless, the "std::string::operator[]" index is size_t which is unsigned, and so does size(), and same happens with vectors.
Besides the operator[] lack of negative index control, I will explain this later.

Do the compilers doesn't warn about this?

If his code got a large input it would index a negative numer, let see g++ and clang++ warnings:

No warnings so many bugs out there...

In order to reproduce the crash we can load a big string or vector from file, for example:

I've implemented a loading function, getting the file size with tellg() and malloc to allocate the buffer, then in this case used as a string.
Let see how the compiler write asm code based on this c++ code.

So the string constructor, getting size and adding -2 is clear. Then come the operator<< to concat the strings.
Then we see the operator[] when it will crash with the negative index.
In assembly is more clear, it will call operator[] to get the value, and there will hapen the magic dereference happens. The operator[] will end up returning an invalid address that will crash at [RAX]

In gdb the operator[] is a allq 0x555555555180 <_znst7__cxx1112basic_stringicst11char_traitsicesaiceeixem plt="">

(gdb) i r rsi
rsi 0xfffffffffffefffe -65538

The implmementation of operator ins in those functions below:

(gdb) bt
#0 0x00007ffff7feebf3 in strcmp () from /lib64/ld-linux-x86-64.so.2
#1 0x00007ffff7fdc9a5 in check_match () from /lib64/ld-linux-x86-64.so.2
#2 0x00007ffff7fdce7b in do_lookup_x () from /lib64/ld-linux-x86-64.so.2
#3 0x00007ffff7fdd739 in _dl_lookup_symbol_x () from /lib64/ld-linux-x86-64.so.2
#4 0x00007ffff7fe1eb7 in _dl_fixup () from /lib64/ld-linux-x86-64.so.2
#5 0x00007ffff7fe88ee in _dl_runtime_resolve_xsavec () from /lib64/ld-linux-x86-64.so.2
#6 0x00005555555554b3 in main (argc=2, argv=0x7fffffffe118) at main.cpp:29

Then crashes on the MOVZX EAX, byte ptr [RAX]

Program received signal SIGSEGV, Segmentation fault.

0x00005555555554b3 in main (argc=2, argv=0x7fffffffe118) at main.cpp:29
29 cout << "penultimate byte is " << hex << s[i] << endl;
(gdb)

What about negative indexing in std::string::operator[] ?
It's exploitable!

In a C char array is known that having control of the index, we can address memory.
Let's see what happens with C++ strings:

The operator[] function call returns the address of string plus 10, and yes, we can do abitrary writes.

Note that gdb displays by default with at&t asm format wich the operands are in oposite order:

And having a string that is in the stack, controlling the index we can perform a write on the stack.

To make sure we are writing outside the string, I'm gonna do 3 writes:

See below the command "i r rax" to view the address where the write will be performed.

The beginning of the std::string object is 0x7fffffffde50.
Write -10 writes before the string 0x7fffffffde46.
And write -100 segfaults because is writting in non paged address.

So, C++ std::string probably is not vulnerable to buffer overflow based in concatenation, but the std::string::operator[] lack of negative indexing control and this could create vulnerable and exploitable situations, some times caused by a signed used of the unsigned std::string.size()

More articles

Kung-fu Chambly

During a CTF game, we came across very-very old systems. Turns out, it is not that easy to hack those dinosaur old systems, because modern tools like Metasploit do not have sploits for those old boxes and of course our "133t h4cking skillz" are useless without Metasploit... :)

But I had an idea: This can be a pretty good small research for fun.

The rules for the hack are the following:

Only publicly available tools can be used for this hack, so no tool development. This is a CTF for script bunniez, and we can't haz code!
Only hacks without user interaction are allowed (IE based sploits are out of scope).
I need instant remote code execution. For example, if I can drop a malware to the c: drive, and change autoexec.bat, I'm still not done, because no one will reboot the CTF machine in a real CTF for me. If I can reboot the machine, that's OK.
I don't have physical access.

I have chosen Windows 95 for this task. First, I had to get a genuine Windows 95 installer, so I visited the Microsoft online shop and downloaded it from their official site.

I installed it in a virtualized environment (remember, you need a boot floppy to install from the CD), and it hit me with a serious nostalgia bomb after watching the installer screens. "Easier to use", "faster and more efficient", "high-powered performance", "friendly", "intuitive interface". Who does not want that? :)

Now that I have a working Windows 95 box, setting up the TCP/IP is easy, let's try to hack it!

My first tool is always nmap. Let's scan the box! Below I'm showing the interesting parts from the result:

PORT      STATE           SERVICE       VERSION 139/tcp   open            netbios-ssn 137/udp   open|filtered   netbios-ns 138/udp   open|filtered   netbios-dgm Running: Microsoft Windows 3.X|95 OS details: Microsoft Windows for Workgroups 3.11 or Windows 95 TCP Sequence Prediction: Difficulty=25 (Good luck!) IP ID Sequence Generation: Broken little-endian incremental

The first exciting thing to note is that there is no port 445! Port 445 is only since NT 4.0. If you check all the famous windows sploits (e.g., MS03-026, MS08-067), all of them use port 445 and named pipes. But there are no named pipes on Windows 95!

Because I'm a Nessus monkey, let's run a free Nessus scan on it!

Only one critical vulnerability found:

Microsoft Windows NT 4.0 Unsupported Installation Detection

Thanks for nothing, Nessus! But at least it was for free.

Next, I tried GFI Languard, nothing. It detected the machine as Win95, the opened TCP port, and some UDP ports as open (false-positive), and that's all...

Let's try another free vulnerability scanner tool, Nexpose. The results are much better:

CIFS NULL Session Permitted
Weak LAN Manager hashing permitted
SMB signing not required
Windows 95/98/ME Share Level Password Bypass
TCP Sequence Number Approximation Vulnerability
ICMP netmask response
CIFS Share Readable By Everyone

I think the following vulnerabilities are useless for me at the moment:

Weak LAN Manager hashing permitted - without user interaction or services looking at the network, useless (I might be wrong here, will check this later)
TCP Sequence Number Approximation Vulnerability - not interesting
ICMP netmask response - not interesting
CIFS Share Readable By Everyone - unless there is a password in a text file, useless

But we have two interesting vulns:

CIFS NULL Session Permitted - this could be interesting, I will check this later ...
Windows 95/98/ME Share Level Password Bypass - BINGO!

Let me quote Nexpose here:

"3.2.3 Windows 95/98/ME Share Level Password Bypass (CIFS-win9x-onebyte-password)

A flaw in the Windows 95/98/ME File and Print Sharing service allows unauthorized users to access file and print shares by sending the first character of the password. Due to the limited number of attempts required to guess the password, brute force attacks can be performed in just a few seconds.

Established connection to share TEST with password P."

The vulnerability description at MS side:

http://technet.microsoft.com/en-us/security/bulletin/ms00-072

For example if the password is "Password" (without quotes) and the client sends the password "P" (without quotes) and the length of 1, the client is authenticated. To find the rest of the password, the attacker increments the length to 2 and starts guessing the second letter until he reaches "PA" and gets authenticated again. As share passwords in Windows 95 are not case sensitive, "Pa" and "PA" will also be accepted. The attacker can continue to increment the length and guessing the next letter one-by-one until he gets the full "PASSWORD" (as the maximum length is 8 characters).

I believe all characters between ALT+033 and ALT+255 can be used in the share password in Windows 95, but as it is case insensitive, we have 196 characters to use, and a maximum length of 8 characters. In worst case this means that we can guess the full password in 1568 requests. The funny thing is that the share password is not connected to (by default) any username/account, and it cannot be locked via brute force.

Luckily there is a great tool which can exploit this vulnerability:

Share Password Checker

Let's check this tool in action:

W00t w00t, it brute forced the password in less then 2 seconds!

Looking at a wireshark dump we can see how it is done:

As you can see, in the middle of the dump we can see that it already guessed the part "PASS" and it is brute-forcing the fifth character, it founds that "W" is the correct fifth character, and starts brute-forcing the sixth character.

If we are lucky with the CTF, the whole C:\ drive is shared with full read-write access, and we can write our team identifier into the c:\flag.txt. But what if we want remote code execution? Stay tuned, this is going to be the topic of the next part of this post.

kung-fu Chambly/Marieville

mercredi 31 mai 2023

C++ Std::String Buffer Overflow And Integer Overflow

More articles

Hacking Windows 95, Part 1

Related articles

mardi 30 mai 2023

3220 hash passwords

Archives du blogue

Qui suis-je