Gentoo PaX Utilities

1.  What is this guide about?

Introduction

The security of a system goes beyond setting up a decent firewall and good service configurations. The binaries you run, the libraries you load, might also be vulnerable against attacks. Although the exact vulnerabilities are not known until they are discovered, there are ways to prevent them from happening.

One possible attack vector is to make advantage of writable and executable segments in a program or library, allowing malicious users to run their own code using the vulnerable application or library.

This guide will inform you how to use the pax-utils package to find and identify problematic binaries. We will also cover the use of pspax (a tool to view PaX-specific capabilities) and dumpelf (a tool that prints out a C structure containing a workable copy of a given object).

But before we start with that, some information on objects is in place. Users familiar with segments and dynamic linking will not learn anything from this and can immediately continue with Extracting ELF Information from Binaries.

ELF objects

Every executable binary on your system is structured in a specific way, allowing the Linux kernel to load and execute the file. Actually, this goes beyond plain executable binaries: this also holds for shared objects; more about those later.

The structure of such a binary is defined in the ELF standard. ELF stands for Executable and Linkable Format. If you are really interested in the gory details, check out the Generic ELF spec or the elf(5) man page.

An executable ELF file has the following parts:

• The ELF header contains information on the type of file (is it an executable, a shared library, ...), the target architecture, the location of the Program Header, Section Header and String Header in the file and the location of the first executable instruction
• The Program Header informs the system how to create a process from the binary file. It is actually a table consisting of entries for each segment in the program. Each entry contains the type, addresses (physical and virtual), size, access rights, ...
• The Section Header is a table consisting of entries for each section in the program. Each entry contains the name, type, size, ... and what information the section holds.
• Data, containing the sections and segments mentioned previously.

A section is a small unit consisting of specific data: instructions, variable data, symbol table, relocation information, and so on. A segment is a collection of sections; segments are the units that are actually transferred to memory.

Shared Objects

Way back when, every application binary contained everything it needed to operate correctly. Such binaries are called statically linked binaries. They are, however, space consuming since different applications use the same functions over and over again.

A shared object contains the definition and instructions for such functions. Every application that wants can dynamically link against such a shared object so that it can benefit from the already existing functionality.

An application that is dynamically linked to a shared object contains symbols, references for the real functionality. When such an application is loaded in memory, it will first ask the runtime linker to resolve each and every symbol it has. The runtime linker will load the appropriate shared objects in memory and resolve the symbolic references between them.

Segments and Sections

How the ELF file is looked upon depends on the view we have: when we are dealing with a binary file in Execution View, the ELF file contains segments. When the file is seen in Linking View, the ELF file contains sections. One segment spans just one or more (continuous) sections.

2.  Extracting ELF Information from Binaries

The scanelf Application

The scanelf application is part of the app-misc/pax-utils package. With this application you can print out information specific to the ELF structure of a binary. The following table sums up the various options.

The format specifiers for the -F option are given in the following table. Prefix each specifier with % (verbose) or # (silent) accordingly.

Using scanelf for Text Relocations

As an example, we will use scanelf to find binaries containing text relocations.

A relocation is an operation that rewrites an address in a loaded segment. Such an address rewrite can happen when a segment has references to a shared object and that shared object is loaded in memory. In this case, the references are substituted with the real address values. Similar events can occur inside the shared object itself.

A text relocation is a relocation in the text segment. Since text segments contain executable code, system administrators might prefer not to have these segments writable. This is perfectly possible, but since text relocations actually write in the text segment, it is not always feasible.

If you want to eliminate text relocations, you will need to make sure that the application and shared object is built with Position Independent Code (PIC), making references obsolete. This not only increases security, but also increases the performance in case of shared objects (allowing writes in the text segment requires a swap space reservation and a private copy of the shared object for each application that uses it).

The following example will search your library paths recursively, without leaving the mounted file system and ignoring symbolic links, for any ELF binary containing a text relocation:

# scanelf -lqtmyR

If you want to scan your entire system for any file containing text relocations:

# scanelf -qtmyR /

Using scanelf for Specific Header

The scanelf util can be used to quickly identify files that contain a given section header using the -k .section option.

In this example we are looking for all files in /usr/lib/debug recursively using a format modifier with quiet mode enabled that have been stripped. A stripped elf will lack a .symtab entry, so we use the '!' to invert the matching logic.

# scanelf -k '!.symtab' /usr/lib/debug -Rq -F%F#k

Using scanelf for Specific Segment Markings

Each segment has specific flags assigned to it in the Program Header of the binary. One of those flags is the type of the segment. Interesting values are PT_LOAD (the segment must be loaded in memory from file), PT_DYNAMIC (the segment contains dynamic linking information), PT_INTERP (the segment contains the name of the program interpreter), PT_GNU_STACK (a GNU extension for the ELF format, used by some stack protection mechanisms), and PT_PAX_FLAGS (a PaX extension for the ELF format, used by the security-minded PaX Project.

If we want to scan all executables in the current working directory, PATH environment and library paths and report those who have a writable and executable PT_LOAD or PT_GNU_STACK marking, you could use the following command:

# scanelf -lpqe .

Using scanelf's Format Modifier Handler

A useful feature of the scanelf utility is the format modifier handler. With this option you can control the output of scanelf, thereby simplifying parsing the output with scripts.

As an example, we will use scanelf to print the file names that contain text relocations:

# scanelf -l -p -R -q -F "%F #t"

3.  Listing PaX Flags and Capabilities

About PaX

PaX is a project hosted by the grsecurity project. Quoting the PaX documentation, its main goal is "to research various defense mechanisms against the exploitation of software bugs that give an attacker arbitrary read/write access to the attacked task's address space. This class of bugs contains among others various forms of buffer overflow bugs (be they stack or heap based), user supplied format string bugs, etc."

To be able to benefit from these defense mechanisms, you need to run a Linux kernel patched with the latest PaX code. The Hardened Gentoo project supports PaX and its parent project, grsecurity. Supported kernel packages are sys-kernel/hardened-sources-2.(4|6).

The Gentoo/Hardened project has a Gentoo PaX Quickstart Guide for your reading pleasure.

Flags and Capabilities

If your tool chain supports it, your binaries can have additional PaX flags in their Program Header. The following flags are supported:

The default Linux kernel also supports certain capabilities, grouped in the so-called POSIX.1e Capabilities. You can find a listing of those capabilities in our POSIX Capabilities document.

Using pspax

The pspax application, part of the pax-utils package, displays the run-time capabilities of all programs you have permission for. On Linux kernels with additional support for extended attributes (such as SELinux) those attributes are shown as well.

When ran, pspax shows the following information:

By default, pspax does not show any kernel processes. If you want those to be taken as well, use the -a switch.

4.  Programming with ELF files

The dumpelf Utility

With the dumpelf utility you can convert a ELF file into human readable C code that defines a structure with the same image as the original ELF file.

$ dumpelf /bin/hostname
#include <elf.h>

/*
 * ELF dump of '/bin/hostname'
 *     10276 (0x2824) bytes
 */

struct {
        Elf32_Ehdr ehdr;
        Elf32_Phdr phdrs[8];
        Elf32_Shdr shdrs[26];
} dumpedelf_0 = {

.ehdr = {
(... Output stripped ...)

The contents of this document are licensed under the Creative Commons - Attribution / Share Alike license.