代码分析

Code analysis is a common technique used to extract information from assembly code.

Radare2 has different code analysis techniques implemented in the core and available in different commands.

As long as the whole functionalities of r2 are available with the API as well as using commands. This gives you the ability to implement your own analysis loops using any programming language, even with r2 oneliners, shellscripts, or analysis or core native plugins.

The analysis will show up the internal data structures to identify basic blocks, function trees and to extract opcode-level information.

The most common radare2 analysis command sequence is aa, which stands for "analyze all". That all is referring to all symbols and entry-points. If your binary is stripped you will need to use other commands like aaa, aab, aar, aac or so.

Take some time to understand what each command does and the results after running them to find the best one for your needs.

[0x08048440]> aa
[0x08048440]> pdf @ main
           ; DATA XREF from 0x08048457 (entry0)
/ (fcn) fcn.08048648 141
|     ;-- main:
|     0x08048648    8d4c2404     lea ecx, [esp+0x4]
|     0x0804864c    83e4f0       and esp, 0xfffffff0
|     0x0804864f    ff71fc       push dword [ecx-0x4]
|     0x08048652    55           push ebp
|     ; CODE (CALL) XREF from 0x08048734 (fcn.080486e5)
|     0x08048653    89e5         mov ebp, esp
|     0x08048655    83ec28       sub esp, 0x28
|     0x08048658    894df4       mov [ebp-0xc], ecx
|     0x0804865b    895df8       mov [ebp-0x8], ebx
|     0x0804865e    8975fc       mov [ebp-0x4], esi
|     0x08048661    8b19         mov ebx, [ecx]
|     0x08048663    8b7104       mov esi, [ecx+0x4]
|     0x08048666    c744240c000. mov dword [esp+0xc], 0x0
|     0x0804866e    c7442408010. mov dword [esp+0x8], 0x1 ;  0x00000001
|     0x08048676    c7442404000. mov dword [esp+0x4], 0x0
|     0x0804867e    c7042400000. mov dword [esp], 0x0
|     0x08048685    e852fdffff   call sym..imp.ptrace
|        sym..imp.ptrace(unk, unk)
|     0x0804868a    85c0         test eax, eax
| ,=< 0x0804868c    7911         jns 0x804869f
| |   0x0804868e    c70424cf870. mov dword [esp], str.Don_tuseadebuguer_ ;  0x080487cf
| |   0x08048695    e882fdffff   call sym..imp.puts
| |      sym..imp.puts()
| |   0x0804869a    e80dfdffff   call sym..imp.abort
| |      sym..imp.abort()
| `-> 0x0804869f    83fb02       cmp ebx, 0x2
|,==< 0x080486a2    7411         je 0x80486b5
||    0x080486a4    c704240c880. mov dword [esp], str.Youmustgiveapasswordforusethisprogram_ ;  0x0804880c
||    0x080486ab    e86cfdffff   call sym..imp.puts
||       sym..imp.puts()
||    0x080486b0    e8f7fcffff   call sym..imp.abort
||       sym..imp.abort()
|`--> 0x080486b5    8b4604       mov eax, [esi+0x4]
|     0x080486b8    890424       mov [esp], eax
|     0x080486bb    e8e5feffff   call fcn.080485a5
|        fcn.080485a5() ; fcn.080484c6+223
|     0x080486c0    b800000000   mov eax, 0x0
|     0x080486c5    8b4df4       mov ecx, [ebp-0xc]
|     0x080486c8    8b5df8       mov ebx, [ebp-0x8]
|     0x080486cb    8b75fc       mov esi, [ebp-0x4]
|     0x080486ce    89ec         mov esp, ebp
|     0x080486d0    5d           pop ebp
|     0x080486d1    8d61fc       lea esp, [ecx-0x4]
\     0x080486d4    c3           ret

In this example, we analyze the whole file (aa) and then print disassembly of the main() function (pdf). The aa command belongs to the family of auto analysis commands and performs only the most basic auto analysis steps. In radare2 there are many different types of the auto analysis commands with a different analysis depth, including partial emulation: aa, aaa, aab, aaaa, ... There is also a mapping of those commands to the r2 CLI options: r2 -A, r2 -AA, and so on.

It is a common sense that completely automated analysis can produce non sequitur results, thus radare2 provides separate commands for the particular stages of the analysis allowing fine-grained control of the analysis process. Moreover, there is a treasure trove of configuration variables for controlling the analysis outcomes. You can find them in anal.* and emu.* cfg variables' namespaces.

One of the most important "basic" analysis commands is the set of af subcommands. af means "analyze function". Using this command you can either allow automatic analysis of the particular function or perform completely manual one.

Some of the most challenging tasks while performing a function analysis are merge, crop and resize. As with other analysis commands you have two modes: semi-automatic and manual. For the semi-automatic, you can use afm <function name> to merge the current function with the one specified by name as an argument, aff to readjust the function after analysis changes or function edits, afu <address> to do the resize and analysis of the current function until the specified address.

Apart from those semi-automatic ways to edit/analyze the function, you can hand craft it in the manual mode with af+ command and edit basic blocks of it using afb commands. Before changing the basic blocks of the function it is recommended to check the already presented ones:

There are two very important commands for this: afc and afB. The latter is a must-know command for some platforms like ARM. It provides a way to change the "bitness" of the particular function. Basically, allowing to select between ARM and Thumb modes.

afc on the other side, allows to manually specify function calling convention. You can find more information on its usage in calling_conventions.

Recursive analysis

There are 4 important program wide half-automated analysis commands:

  • aab - perform basic-block analysis ("Nucleus" algorithm)

  • aac - analyze function calls from one (selected or current function)

  • aaf - analyze all function calls

  • aar - analyze data references

  • aad - analyze pointers to pointers references

Those are only generic semi-automated reference searching algorithms. Radare2 provides a wide choice of manual references' creation of any kind. For this fine-grained control you can use ax commands.

The most commonly used ax commands are axt and axf, especially as a part of various r2pipe scripts. Lets say we see the string in the data or a code section and want to find all places it was referenced from, we should use axt:

Apart from predefined algorithms to identify functions there is a way to specify a function prelude with a configuration option anal.prelude. For example, like e anal.prelude = 0x554889e5 which means

on x86_64 platform. It should be specified before any analysis commands.

Configuration

Radare2 allows to change the behavior of almost any analysis stages or commands. There are different kinds of the configuration options:

  • Flow control

  • Basic blocks control

  • References control

  • IO/Ranges

  • Jump tables analysis control

  • Platform/target specific options

Control flow configuration

Two most commonly used options for changing the behavior of control flow analysis in radare2 are anal.hasnext and anal.afterjump. The first one allows forcing radare2 to continue the analysis after the end of the function, even if the next chunk of the code wasn't called anywhere, thus analyzing all of the available functions. The latter one allows forcing radare2 to continue the analysis even after unconditional jumps.

In addition to those we can also set anal.ijmp to follow the indirect jumps, continuing analysis; anal.pushret to analyze push ...; ret sequence as a jump; anal.nopskip to skip the NOP sequences at a function beginning.

For now, radare2 also allows you to change the maximum basic block size with anal.bb.maxsize option . The default value just works in most use cases, but it's useful to increase that for example when dealing with obfuscated code. Beware that some of basic blocks control options may disappear in the future in favor of more automated ways to set those.

For some unusual binaries or targets, there is an option anal.noncode. Radare2 doesn't try to analyze data sections as a code by default. But in some cases - malware, packed binaries, binaries for embedded systems, it is often a case. Thus - this option.

Reference control

The most crucial options that change the analysis results drastically. Sometimes some can be disabled to save the time and memory when analyzing big binaries.

  • anal.jmpref - to allow references creation for unconditional jumps

  • anal.cjmpref - same, but for conditional jumps

  • anal.datarefs - to follow the data references in code

  • anal.refstr - search for strings in data references

  • anal.strings - search for strings and creating references

Note that strings references control is disabled by default because it increases the analysis time.

Analysis ranges

There are a few options for this:

  • anal.limits - enables the range limits for analysis operations

  • anal.from - starting address of the limit range

  • anal.to - the corresponding end of the limit range

  • anal.in - specify search boundaries for analysis (io.maps, io.sections.exec, dbg.maps and many

    more - see e anal.in=? for the complete list)

Jump tables

Jump tables are one of the trickiest targets in binary reverse engineering. There are hundreds of different types, the end result depending on the compiler/linker and LTO stages of optimization. Thus radare2 allows enabling some experimental jump tables detection algorithms using anal.jmptbl option. Eventually, algorithms moved into the default analysis loops once they start to work on every supported platform/target/testcase. Two more options can affect the jump tables analysis results too:

  • anal.ijmp - follow the indirect jumps, some jump tables rely on them

  • anal.datarefs - follow the data references, some jump tables use those

Platform specific controls

There are two common problems when analyzing embedded targets: ARM/Thumb detection and MIPS GP value. In case of ARM binaries radare2 supports some auto-detection of ARM/Thumb mode switches, but beware that it uses partial ESIL emulation, thus slowing the analysis process. If you will not like the results, particular functions' mode can be overridden with afB command.

The MIPS GP problem is even trickier. It is a basic knowledge that GP value can be different not only for the whole program, but also for some functions. To partially solve that there are options anal.gp and anal.gp2. The first one sets the GP value for the whole program or particular function. The latter allows to "constantify" the GP value if some code is willing to change its value, always resetting it if the case. Those are heavily experimental and might be changed in the future in favor of more automated analysis.

Visuals

One of the easiest way to see and check the changes of the analysis commands and variables is to perform a scrolling in a Vv special visual mode, allowing functions preview:

vv

When we want to check how analysis changes affect the result in the case of big functions, we can use minimap instead, allowing to see a bigger flow graph on the same screen size. To get into the minimap mode type VV then press p twice:

vv2

This mode allows you to see the disassembly of each node separately, just navigate between them using Tab key.

Analysis hints

It is not an uncommon case that analysis results are not perfect even after you tried every single configuration option. This is where the "analysis hints" radare2 mechanism comes in. It allows to override some basic opcode or meta-information properties, or even to rewrite the whole opcode string. These commands are located under ah namespace:

One of the most common cases is to set a particular numeric base for immediates:

It is notable that some analysis stages or commands add the internal analysis hints, which can be checked with ah command:

Sometimes we need to override jump or call address, for example in case of tricky relocation, which is unknown for radare2, thus we can change the value manually. The current analysis information about a particular opcode can be checked with ao command. We can use ahc command for performing such a change:

As you can see, despite the unchanged disassembly view the jump address in opcode was changed (jump option).

If anything of the previously described didn't help, you can simply override shown disassembly with anything you like:

上一页分析下一页变量

最后更新于

这有帮助吗？