Paxtest v0.9.13 README file
Copyright (c)2004 by Peter Busser <peter@adamantix.org>
Additional architecture/target support, tests, & fixes
by Brad Spengler <spender@grsecurity.net>
This file has been released under the GNU GPL version 2 or later.
History:
-------
When I started the Adamantix project, one of the first things I did was to add
PaX functionality to the kernel. PaX is a process memory protection patch.
Anything that happens outside the kernel on a UNIX system happens inside a
process. There are many attacks on the Internet that try to corrupt the process
memory, in order to make it do something for which it was not intended. One
example of such an attack is the so called buffer overflow attack. This kind of
attack is one of the most popular at this moment.
PaX protects against such attacks. Or so the author claims. When I started to
add PaX to Adamantix, almost nothing happened. A few libraries broke, but that
was easy to fix, and that was it. I expected many programs to break. So I
started to wonder: ``Does this patch really do anything?'' Instead of
speculating, I decided to write a test suite. After some time, the first
version of paxtest was ready. More functionality was added. With the addition
of every test, it proved that PaX was working just fine. I decided to publish
paxtest, because it can be useful for other people to test the functionality
of the memory protection of their system(s).
Compiling paxtest:
-----------------
Compiling paxtest should be quite straightforward. First unpack paxtest. Then
cd into the directory. And then run one of the following:
make linux (to create Linux binaries)
make openbsd (to create OpenBSD binaries)
(Type make to get the above list.)
Running paxtest:
---------------
In order to run paxtest, type the following:
./paxtest kiddie (to run paxtest in script kiddie mode)
or
./paxtest blackhat (to run paxtest in blackhat mode)
This will create a file paxtest.log in the user's home directory.
Kiddie and blackhat mode:
------------------------
Paxtest provides two modes. The first mode is the so called kiddie mode, which
makes paxtest behave rather nicely. This simulates the level of sophistication
displayed by script kiddies. Good results in this mode do not necessarily mean
that the security level of your system is high.
The second mode, the blackhat mode, makes paxtest use of a few tricks to try
to get around the protection of the kernel. The attacks done by paxtest in this
mode can be performed in real life by attackers too. But the level of
sophistication is somewhat higher than those of the kiddie mode and the chance
of success is somewhat lower. But there is still a realistic chance of success.
People who are concerned about the security of their systems should look at
the blackhat mode results.
The tests:
---------
There are two types of tests in paxtest. First there are tests which try
different ways to write and then run exploit code. The more different exploits
are stopped, the better the protection provided by the system. Note that the
tests simulate what an exploit could (and would) do, not what an application
does normally. This is especially important in the mprotect tests because
some people claim that normal applicatons do not do this - indeed, but that
is not what these tests are about, they simulate exploits, not applications.
Second, there is a number of tests that measure the randomization of the
system. Randomisation does not provide any security, but only makes it harder
for the attacker to guess where to attack (it is in fact security through
obscurity). It is like rearranging the furniture in a pitch dark room every
time someone enters. The more random the furniture is placed, the more likely
it is that someone stumbles over it. The same applies to randomization in this
context, randomization makes it more likely that an attack does not succeed.
In short: More random bits is better. Zero bits means: No randomization.
Executable anonymous mapping
Executable bss
Executable data
Executable heap
Executable stack
These tests try to write data to memory and then try to execute it as
if it was executable code. This is exactly what most buffer exploits
do. There are five tests, each test tests one specific memory area.
Executable anonymous mapping (mprotect)
Executable bss (mprotect)
Executable data (mprotect)
Executable heap (mprotect)
Executable stack (mprotect)
These tests do the same as the previous tests. But now the test also
tries to disable the memory protection in the kernel using mprotect().
It changes the protection status of memory. An attacker can use it to
try to switch off memory protection and execute shellcode as usual.
Executable shared library bss (mprotect)
Executable shared library data (mprotect)
These tests try to overwrite two different data areas in a shared
library. After that, it tries to execute the data as if it was code.
Again, these tests try to switch off the memory protection using
mprotect().
Anonymous mapping randomization test
Anonymous mappings are used for dynamically allocating memory in a
program. This test tries to figure out the number of random bits used
when such an anonymous mapping is requested. More bits is better.
Heap randomization test (ET_EXEC)
Heap randomization test (ET_DYN)
The heap is used for allocating small chunks of memory in a program.
There there are two different types of executables, ET_EXEC (normal
executables) and ET_DYN (basically executable shared libraries). The
randomization can depend on the kind of executable, therefore there
are two tests.
Most Linux distributions ship with ET_EXEC executables only. So for
those the ET_EXEC randomization is the most relevant. Adamantix ships
with mostly ET_DYN executables, so the ET_EXEC randomization is not
very relevant, but the ET_DYN randomization is.
Normally ET_DYN randomization is bigger than the ET_EXEC randomisation.
Main executable randomization (ET_EXEC)
Main executable randomization (ET_DYN)
Same here, two tests, because the two different types of executables
can have different randomization.
Shared library randomization test
Shared libraries can be located at random addresses too, which is what
this test tries to find out.
Stack randomization test (SEGMEXEC)
Stack randomization test (PAGEEXEC)
The stack is used for storing intermediate data. It also contains
addresses that influence the way programs run. That combination makes
the stack a popular memory area for attacks. Randomisation makes it
harder though, which is what these test try to find out.
Return to function (strcpy)
Return to function (strcpy, RANDEXEC)
Return to function (memcpy)
Return to function (memcpy, RANDEXEC)
Return to function attacks are very nasty. These tests are hard to
stop by kernel patches, but they show that there you should not expect
perfect protection from this kind of security patches.
Executable shared library bss
Executable shared library data
These tests try to overwrite two different data areas in a shared
library. After that, it tries to execute the data as if it was code.
Writable text segments
When an attacker can overwrite code, he can basically alter the
program while it runs. This test tries to do that.
Perfect protection:
------------------
Perfect protection is not possible. That is also the reason why there are
so called return to function tests in paxtest. PaX does not prevent return to
function attacks. Neither do any of the other memory protection patches. But it
is important that people who use kernel patches like PaX do not get a false
sense of security. As the PaX documentation points out: There are three
different classes of attacks, and at this moment PaX can only guarantee
protection against one of them.
Further reading:
---------------
PaX home page:
http://pax.grsecurity.net/
PaX mailing list:
http://lists.adamantix.org/
PaX documentation (also a thorough introduction into memory protection):
http://pax.grsecurity.net/docs/
An article I wrote about PaX has been published in Linux Magazine. The whole
article is available on-line as PDF. It is not very technical, but describes
what memory protection is and why it is important. It can be found here:
http://www.linux-magazine.com/issue/40
Generated by dwww version 1.15 on Sun Jun 16 21:08:29 CEST 2024.