Learning Linux Binary Analysis: Learning Linux Binary Analysis

GNU Debugger (GDB) is not only good to debug buggy applications. It can also be used to learn about a program's control flow, change a program's control flow, and modify the code, registers, and data structures. These tasks are common for a hacker who is working to exploit a software vulnerability or is unraveling the inner workings of a sophisticated virus. GDB works on ELF binaries and Linux processes. It is an essential tool for Linux hackers and will be used in various examples throughout this book.

Object dump (objdump) is a simple and clean solution for a quick disassembly of code. It is great for disassembling simple and untampered binaries, but will show its limitations quickly when attempting to use it for any real challenging reverse engineering tasks, especially against hostile software. Its primary weakness is that it relies on the ELF section headers and doesn't perform control flow analysis, which are both limitations that greatly reduce its robustness. This results in not being able to correctly disassemble the code within a binary, or even open the binary at all if there are no section headers. For many conventional tasks, however, it should suffice, such as when disassembling common binaries that are not fortified, stripped, or obfuscated in any way. It can read all common ELF types. Here are some common examples of how to use objdump:

We will be exploring objdump and other tools in great depth during our introduction to the ELF format in Chapter 2, The ELF Binary Format.

Object copy (Objcopy) is an incredibly powerful little tool that we cannot summarize with a simple synopsis. I recommend that you read the manual pages for a complete description. Objcopy can be used to analyze and modify ELF objects of any kind, although some of its features are specific to certain types of ELF objects. Objcopy is often times used to modify or copy an ELF section to or from an ELF binary.

To copy the .data section from an ELF object to a file, use this line:

objcopy –only-section=.data <infile> <outfile>

The objcopy tool will be demonstrated as needed throughout the rest of this book. Just remember that it exists and can be a very useful tool for the Linux binary hacker.

System call trace (strace) is a tool that is based on the ptrace(2) system call, and it utilizes the PTRACE_SYSCALL request in a loop to show information about the system call (also known as syscalls) activity in a running program as well as signals that are caught during execution. This program can be highly useful for debugging, or just to collect information about what syscalls are being called during runtime.

This is the strace command used to trace a basic program:

strace /bin/ls -o ls.out

The strace command used to attach to an existing process is as follows:

strace -p <pid> -o daemon.out

The initial output will show you the file descriptor number of each system call that takes a file descriptor as an argument, such as this:

SYS_read(3, buf, sizeof(buf));

If you want to see all of the data that was being read into file descriptor 3, you can run the following command:

strace -e read=3 /bin/ls

You may also use -e write=fd to see written data. The strace tool is a great little tool, and you will undoubtedly find many reasons to use it.

library trace (ltrace) is another neat little tool, and it is very similar to strace. It works similarly, but it actually parses the shared library-linking information of a program and prints the library functions being used.

You may see system calls in addition to library function calls with the -S flag. The ltrace command is designed to give more granular information, since it parses the dynamic segment of the executable and prints actual symbols/functions from shared and static libraries:

ltrace <program> -o program.out

Function trace (ftrace) is a tool designed by me. It is similar to ltrace, but it also shows calls to functions within the binary itself. There was no other tool I could find publicly available that could do this in Linux, so I decided to code one. This tool can be found at https://github.com/elfmaster/ftrace. A demonstration of this tool is given in the next chapter.

The readelf command is one of the most useful tools around for dissecting ELF binaries. It provides every bit of the data specific to ELF necessary for gathering information about an object before reverse engineering it. This tool will be used often throughout the book to gather information about symbols, segments, sections, relocation entries, dynamic linking of data, and more. The readelf command is the Swiss Army knife of ELF. We will be covering it in depth as needed, during Chapter 2, The ELF Binary Format, but here are a few of its most commonly used flags:

ERESI project (http://www.eresi-project.org) contains a suite of many tools that are a Linux binary hacker's dream. Unfortunately, many of them are not kept up to date and aren't fully compatible with 64-bit Linux. They do exist for a variety of architectures, however, and are undoubtedly the most innovative single collection of tools for the purpose of hacking ELF binaries that exist today. Because I personally am not really familiar with using the ERESI project's tools, and because they are no longer kept up to date, I will not be exploring their capabilities within this book. However, be aware that there are two Phrack articles that demonstrate the innovation and powerful features of the ERESI tools:

/proc/<pid>/maps file contains the layout of a process image by showing each memory mapping. This includes the executable, shared libraries, stack, heap, VDSO, and more. This file is critical for being able to quickly parse the layout of a process address space and is used more than once throughout this book.

The /proc/kcore is an entry in the proc filesystem that acts as a dynamic core file of the Linux kernel. That is, it is a raw dump of memory that is presented in the form of an ELF core file that can be used by GDB to debug and analyze the kernel. We will explore /proc/kcore in depth in Chapter 9, Linux /proc/kcore Analysis.

This file is available on almost all Linux distributions and is very useful for kernel hackers. It contains every symbol for the entire kernel.

The kallsyms is very similar to System.map, except that it is a /proc entry that means that it is maintained by the kernel and is dynamically updated. Therefore, if any new LKMs are installed, the symbols will be added to /proc/kallsyms on the fly. The /proc/kallsyms contains at least most of the symbols in the kernel and will contain all of them if specified in the CONFIG_KALLSYMS_ALL kernel config.

The iomem is a useful proc entry as it is very similar to /proc/<pid>/maps, but for all of the system memory. If, for instance, you want to know where the kernel's text segment is mapped in the physical memory, you can search for the Kernel string and you will see the code/text segment, the data segment, and the bss segment:

  $ grep Kernel /proc/iomem
  01000000-016d9b27 : Kernel code
  016d9b28-01ceeebf : Kernel data
  01df0000-01f26fff : Kernel bss

Extended core file snapshot (ECFS) is a special core dump technology that was specifically designed for advanced forensic analysis of a process image. The code for this software can be found at https://github.com/elfmaster/ecfs. Also, Chapter 8, ECFS – Extended Core File Snapshot Technology, is solely devoted to explaining what ECFS is and how to use it. For those of you who are into advanced memory forensics, you will want to pay close attention to this.

The LD_PRELOAD environment variable can be set to specify a library path that should be dynamically linked before any other libraries. This has the effect of allowing functions and symbols from the preloaded library to override the ones from the other libraries that are linked afterwards. This essentially allows you to perform runtime patching by redirecting shared library functions. As we will see in later chapters, this technique can be used to bypass anti-debugging code and for userland rootkits.

This environment variable tells the program loader to display the program's auxiliary vector during runtime. The auxiliary vector is information that is placed on the program's stack (by the kernel's ELF loading routine), with information that is passed to the dynamic linker with certain information about the program. We will examine this much more closely in Chapter 3, Linux Process Tracing, but the information might be useful for reversing and debugging. If, for instance, you want to get the memory address of the VDSO page in the process image (which can also be obtained from the maps file, as shown earlier) you have to look for AT_SYSINFO.

Here is an example of the auxiliary vector with LD_SHOW_AUXV:

$ LD_SHOW_AUXV=1 whoami
AT_SYSINFO: 0xb7779414
AT_SYSINFO_EHDR: 0xb7779000
AT_HWCAP: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2
AT_PAGESZ: 4096
AT_CLKTCK: 100
AT_PHDR:  0x8048034
AT_PHENT: 32
AT_PHNUM: 9
AT_BASE:  0xb777a000
AT_FLAGS: 0x0
AT_ENTRY: 0x8048eb8
AT_UID:  1000
AT_EUID: 1000
AT_GID:  1000
AT_EGID: 1000
AT_SECURE: 0
AT_RANDOM: 0xbfb4ca2b
AT_EXECFN: /usr/bin/whoami
AT_PLATFORM: i686
elfmaster

The auxiliary vector will be covered in more depth in Chapter 2, The ELF Binary Format.

Linker scripts are a point of interest to us because they are interpreted by the linker and help shape a program's layout with regard to sections, memory, and symbols. The default linker script can be viewed with ld -verbose.

The ld linker program has a complete language that it interprets when it is taking input files (such as relocatable object files, shared libraries, and header files), and it uses this language to determine how the output file, such as an executable program, will be organized. For instance, if the output is an ELF executable, the linker script will help determine what the layout will be and what sections will exist in which segments. Here is another instance: the .bss section is always at the end of the data segment; this is determined by the linker script. You might be wondering how this is interesting to us. Well! For one, it is important to have some insights into the linking process during compile time. The gcc relies on the linker and other programs to perform this task, and in some instances, it is important to be able to have control over the layout of the executable file. The ld command language is quite an in-depth language and is beyond the scope of this book, but it is worth checking out. And while reverse engineering executables, remember that common segment addresses may sometimes be modified, and so can other portions of the layout. This indicates that a custom linker script is involved. A linker script can be specified with gcc using the -T flag. We will look at a specific example of using a linker script in Chapter 5, Linux Binary Protection.