Theory behind relocation

[mariuszp]

OK, so relocation is basically "moving" addresses inside an executable so that it can be loaded at a position different from the one it wants to be loaded it, right? But how exactly does relocation happen? I mean, do I have to have some really advanced loader (in my kernel) that can update whole instructions to change the addresses based on what the executable says? Or is the instruction 'broken up' so I can insert the address in?

[Combuster]

Relocating is just the process of changing addresses when segments of the executable is loaded at a different location as they were originally linked to. So if you link a binary to 1M, and you load it to 3G, all absolute addresses will need to have the difference (3G-1M) added to them.

Relocation entries consist of just the section and offset of the address to be patch, the segment of the target, and the type of relocation. So absolute relocations will be updated when the offset changes, and relative relocations will be update when the distance between section starts change. In both cases you simply add the difference to the original location. And those are the only relocations you will find in a simple 32-bits executable.

The process is relatively simple: my own relocator does the job in about 50 lines of assembly.

[mariuszp]

You mean, it's something like:

The relocation table contains the address of the address to relocate (target?) and I just change the 'target' to the address needed (add the difference?)

[Owen]

For simple executables, yes. For more complex ones (e.g. if doing runtime linking against a shared object or DLL) and dealing with things like position independent code and/or IP-relative addressing, things will be more involved.

[JamesM]

You mean, it's something like:

The relocation table contains the address of the address to relocate (target?) and I just change the 'target' to the address needed (add the difference?)

Pretty much - although how you change the target may be slightly more involved. Most 32-bit x86 instructions take a 32-bit absolute/relative offset, so yes, just adding the difference will work.

Some instructions, and especially in other architectures, are formatted differently so you may need to do more involved calculations.