So let's start at the beginning with a question that might help some unfamiliar people understand all of this: what radare2 actually is. According to Wikipedia,
"Radare2 is a complete framework for reverse-engineering and analyzing binaries; composed of a set of small utilities that can be used together or independently from the command line.
— Wikipedia - Radare2
So we can use radare to take apart binaries, but using all the tools included, we can do much more as you'll see next.
radare2 commands tend to be not so descriptive. If you don't know what a command does, you can append a question mark to the command to get some help. If you know what you want to do, but don't know the command that might be able to do what you want to do, you can use this alias to search through all radare commands interactively:
Over the last few days, I’ve played around with r2wars, a competion typically between two programs that try to survive as much time as possible in a shared memory space. A python implementation of r2wars can be found on github as well as a C# implementation.
There exist similar forms of games such as Core Wars, but what makes r2wars different is that the bots can be built in any architecture supported by ESIL Evaluable Strings Intermediate Language, more than 2 programs can run at the same time and cyclic execution cost matters for the turns.
The "players" are bots. A bot is a piece of assembly, written in either x86, arm or mips using either 8, 16, 32, or 64 bit registers. A super simple bot doing nothing but locating itself in memory might look like this: (x86, 32 bits)
Assembling such a bot can be done using rasm2, the radare2 assembler and disassembler tool, as displayed in the listing below.
; rasm2 -a x86 -b 32 -f bot.asm
The bot created can be inspected, by disassembling it again using rasm2 like below.
The arena is now set up, the next step is to insert the bots into the arena. Selecting where to insert the bots is kind of crucial, because the bots should not be inserted into each other and not to close to the end of the arena (0x400 in this case).
In order to generate a random offsets where the bots can be placed, multiple addresses should be generated in the following way:
genspace = [0x000, 0x3c0)
maxbotspace = [0x3c0, 0x400)
0x000 0x340 0x400
+ -------------------------- + -----------------+
| gen space | max bot space |
The space in which the bots should be generated is defined as "gen space". This means we can generate a random address in the range [0, 0x340) in which we can (in theory) place the first bot.
After placing the first bot at, for example, 0x40, the address space in which the address for the second bot is chosen from is shrunk to [0x40 + maxbotsize, 0x340] as displayed below.
0x000 0x040 0x340 0x400
+ ---------+---------------- + -----------------+
| reserved | gen space | max bot space |
This might not work the first time, for example when using the default example of two bots in a memory space 1024 bytes big, each bot has (in theory) 512 bytes of memory to position itself in. Doing this for n bots in x bytes of memory results in n / x bytes per bot. This gets more problematic with a greater amount of bots in a limited memory space.
r2wars is a round based game. This means that we need to store the state of each "player" (bot) each round, so that the others can execute their operation. When it's the players turn again, the state has to be restored, so that the player can continue execution as if nothing had happened. r2 can dump all ESIL registers using the aer command and can even print a command to set the registers, aerR.
aer eax = 0x00000000
aer esp = 0x00000000
aer ebp = 0x00000000
aer eip = 0x00000000
By dumping these registers, we can easily restore the state of the bot, by replacing newline chars (\n) by semicolons (;) and executing the result with r2.
After having created an "arena" and inserted a bot into the arena, we can execute an instruction, but before doing so, we still need to set up the"Progam Counter" (PC) and the "Stack Pointer" (SP) for the bot. We can do this by using the aer command, that can be used to manipulate the ESIL registers:
[0x00000000]> aer PC = 0x100
[0x00000000]> aer SP = SP + 0x100
After having done this, the VM is setup and the instruction pointer (Program Counter in ESIL slang) is pointing to the first instruction of our bot. In order to step into, we can use the aes command:
We haven't seen much of our bot yet, so let's look at what's happening. r2 can print the disassembly of the instructions at a specific offset using the pd command, so let's look at what is happening at the offset 0x100, that's where our bot is located.
[0x00000105]> pd 0x4 @ 0x100
0x00000100 e800000000 call 0x105
0x00000105 58 pop eax
0x00000106 0000 add byte [eax], al
0x00000108 0000 add byte [eax], al
What we've done above is we've printed the 0x4 instructions at the offset 0x100. As you can see, the instruction at 0x100 contains a call to 0x105, the pop eax instruction. Radare also displays the current location of the Instruction Pointer (eip) that is currently pointing to 0x105.
Well, We're at the point at which you should have understood the basics, if not, DO NOT PANIC! You can read the "original" description here.
If you have the desire to play around with this, you can clone my implementation from here. It is ready to go with two example bots that you can adjust to your needs.
So from here on, you're on your own. Good luck.