This is going to be a low-level article, but I guess you already knew that since you landed here, right?I wanted to talk about this mysterious term Mach-O…

What is it?

How does it work?

Really, what is it!?

To answer all of that, we’ll have to dig deep and get our hands dirty…When we build an application in Xcode, a lot of things happen at the same time. One of them is converting all the source code into an executable. This executable contains the byte code that will run on the CPU, the ARM processor on an iOS device, or the intel processor on Mac.This executable is called Mach-O.Well, this was easy and fun, Goodbye!Unless you want to stick around to know about the internals 😈

Mach-O format

Mach-O is a binary stream of bytes grouped into meaningful chunks of data. These chunks contain information about the meta like byte order, CPU type, size of the chunk and so on.There are various Mach-O types, most typically you’d have seen these

• Executable — Main app binary, like Example.app/Example

• Dylib — Dynamic library, like libSwiftCore.dylib

Yeah, I know right, lurking right under our nose!So, Mach-O files are divided into several segments and it looks something like thisbut before diving into the segments let’s look into something else

Mach-O Header

Every Mach-O file begins with a header structure that defines the structure of the file. It also contains information about file type and target architecture (armv7, armv7s, i386, etc).

Just below the header structure are a bunch of load commands which help in the layout and linking of the file. Also, load commands can specifyThe initial layout of the file in the virtual memory (we’ll come back to this)

• Section names and addresses
• dylibs to be loaded
• “main” function address
• Code Signature

And, this is how the complete header looks like!As you can see, Mach-O header consists of a bunch of load commands which are defining the addresses of the sections, main function, and the dependent binaries to be loaded.

The addresses mentioned above are actually offset from the memory address where your Mach-O is loaded. This is done because the starting memory address is randomised every time your app launches using a nifty technique called Automatic Space Layout Randomisation or as we lovingly call it ASLR.

What this means is when your app process starts, you do not know from which address will it start beforehand.Let’s imagine its implications, assume you have a global variable which occupies some memory address in your RAM, but since you don’t know where your process started from, you cannot possibly determine the memory address of this global variable!As you might have guessed, this is done for security purposes, otherwise, it would become very easy to hack the binary if everything has the same address on every launch!

Segments

Let’s look at the individual segments of the Mach-O file

__PAGEZERO

This is the first segment of an executable file and it has no data inside so it takes up no space in the file. This segment is full of zeroes to catch NULL pointer dereferences. You might have faced a 

EXC_BAD_ACCESS

 crash, that is precisely because something in your code tried to access data from here, which is not allowed.

As an aside, this segment can be a good place to hide malicious code 😉

__TEXT

This segment contains executable code and read-only data. It is made read-only to allow the sharing of the segment when it is mapped into the memory. This is primarily used with frameworks, bundles and shared libraries.

And, since the __TEXT segment is read-only, there are no changes that need to be saved back to the disk. If the kernel needs to free up memory, it will simply remove the __TEXT page and re-read them when needed.

This is the reason how iOS and OSX cache their dynamic libraries so aggressively.

__DATA

This segment contains writable data (e.g. globals, static variables, etc), and because it is writable, the __DATA segment of a framework or other shared library is logically copied for each process linking with the library.

If you have any experience with Swift, you must be familiar with copy-on-write, this essentially means do not create a copy until the thing being referenced is edited. Similarly, when the __DATA segment is copied, it isn’t really until some process modifies it, that process then receives its own private copy of the page.

__OBJC

This is an optional segment and contains data used by the Objective-C language runtime support library.

__IMPORT

This is also an optional segment and contains symbol stubs and non-lazy pointers to symbols not defined in the executable. This segment is generated only for executables targeted for the IA-32 architecture.

__LINKEDIT

This segment contains raw data for the linker (link editor) like symbols and string tables, compressed dynamic linking info, code signing info, and the indirect symbol table — all of which occupy regions as specified by the load commands.

So, now that we have an understanding of the individual segments, let’s try to look at the bigger picture and see how it all fits together.

The Big Picture — DYLD

Till now we know

1. How a Mach-O file is generated and its load commands used to link dependencies in various ways.
2. Load Commands are used to map the segments in the memory commands.
3. Execution of the file begins from 

   LC_MAIN

Well, this is only information and this information requires a brain to process it.And this brain is called Dyld!

Let me tell you a secret! 🤫
Well, it’s not really a secret

When you launch your app by tapping the app icon, instead of launching your app the kernel launches 

dyld

 !

I know right! This guy is a big deal around here.

The kernel will actually load dyld at some random address space and it will itself has its own __TEXT segment, __DATA segment… well, you get the idea.

It's the job of dyld to basically load and setup all the dependent dylibs for us.

Load dylibs

This is where the dyld reads Mach-O header to find out about the dependent dylibs. It then finds that library file on the file system and parses them.

This process is done recursively because a dylib A can be dependent on dylib B which can be dependent on dylib C, so it has to resolve this whole graph of dependencies and finally memory map all of these dylib’s segments to the original Mach-O header.

And, this whole transaction might look something like this.

Now, remember we talked about ASLR and how you cannot know which address will be assigned to all the variables in your app. This is something that dyld has to fix using the below techniques.

Rebasing

__LINKEDIT section contains locations of all the pointers that need to be shifted. Dyld will go through all these pointers and shift them based on your application’s start address.

Notice that to do this, we have to read and write to the data pages, causing those pages to become dirty, and would need a copy on write. This is why Rebasing is expensive in IO.

Binding

References to other dylib functions are fixed using binding, like NSLog, malloc, etc.

Once dyld loads the dependent libraries, it needs to search the symbol tables and find the implementation of these symbols. So, there’s actually a string named 

_NSLog

 inside your binary that is unresolved and what dyld will do is look up the symbol table and fill it up with the addresses of these functions from the dependent libraries.

This is computationally complex and is expensive

Objc Runtime

All Objc class definitions need to be registered, why? because you can construct an Objc class from a string calling

NSClassFromString(_:) 

method.

So, dyld has to build this table before the app can launch.Adding categories to method lists — what this means is, if you have created a category over UIView and added a bunch of new functions, those new functions will be added to the method list of UIView.It also ensures 

Run Initialisers

Objc 

+load

 methods are called at this pointC++ static initialisers

This happens in a bottom-up fashion so basically the dependent libraries will be initialised first.

Whew!! After all of this is done, finally, your 

main()

 will execute.

And, this is the story behind the elusive Mach-O file.

Summary

1. Binaries use Mach-O format with __TEXT, __DATA and __LINKEDIT  segments.
2. Dyld needs to parse and load all dynamic library dependencies.
3. Dyld needs to fix all pointers both internal and external (rebase, bind, setup runtime).
4. Run static initialisers and 

   +load

    methods.
5. AND THEN 

   main()

Where to go from here?

1. To the dark side — 
2. If you want to dive even deeper and extract some code from dylibs, check out this  from objc.io

Previously published at