(from git.bootmii.org)
Go to file
bushing 4b65f3b500 initial commit 2011-06-20 16:11:10 -07:00
Makefile initial commit 2011-06-20 16:11:10 -07:00
README initial commit 2011-06-20 16:11:10 -07:00
babelfish.c initial commit 2011-06-20 16:11:10 -07:00
babelfish.ld initial commit 2011-06-20 16:11:10 -07:00
common.mk initial commit 2011-06-20 16:11:10 -07:00
elf.h initial commit 2011-06-20 16:11:10 -07:00
gecko.c initial commit 2011-06-20 16:11:10 -07:00
gecko.h initial commit 2011-06-20 16:11:10 -07:00
hollywood.h initial commit 2011-06-20 16:11:10 -07:00
memory.c initial commit 2011-06-20 16:11:10 -07:00
memory.h initial commit 2011-06-20 16:11:10 -07:00
memory_asm.S initial commit 2011-06-20 16:11:10 -07:00
starlet.mk initial commit 2011-06-20 16:11:10 -07:00
start.S initial commit 2011-06-20 16:11:10 -07:00
types.h initial commit 2011-06-20 16:11:10 -07:00
utils.c initial commit 2011-06-20 16:11:10 -07:00
utils.h initial commit 2011-06-20 16:11:10 -07:00
vectors.s initial commit 2011-06-20 16:11:10 -07:00

README

This is babelfish, a just-in-time self-propagting IOS patcher for the Nintendo Wii.

Background:

The Wii poses a peculiar architectural challenge when it comes to
modifying firmware.  Instead of having one firmware binary to patch,
the Wii keeps 20+ slightly incompatible versions of IOS around in
NAND; it reloads the firmware entirely (often to a different version)
when a new title is launched.  This makes it difficult to permanently
modify the behavior (e.g. preventing a bootloader update) because
firmware is frequently reloaded.

The most naive approach to solving this is to patch every single
version of the firmware in NAND; this is done by e.g. CIOSCORP.  This
approach is unacceptable for two reasons:

1) Any failure in patching could either leave a system unbootable (and
bricked), or could fail to apply the desired protection

2) Any update will wipe out some of the patched firmwares.  It's
possible to disable firmware updates by various methods, but it's
still ugly.

A better solution was necessary, so I started working on a piece of
code that could stay resident between IOS reloads -- it would have to
transfer itself along with each new version of IOS.  "IOS Virus" was
the first term that came to mind, but that has a negative connotation;
I chose "Babelfish" to express the benevolent intent of this software. (See
http://en.wikipedia.org/wiki/Babel_fish_(The_Hitchhiker%27s_Guide_to_the_Galaxy))

Theory of operation:
In order for this to work, we need two things:
1) An ability to consistently patch any IOS
2) An ability to patch any IOS when it gets loaded

To simplify this task, a few assumptions have been made:

1) All versions of IOS have an ELF loader in the beginning of the
   image, which we may safely replace with our own loader

2) All versions of IOS have a predictable syscall structure, which
   allows us to hook the right events

3) There is one safe, unused area of memory that we can "hide" a copy
   of ourself without getting overwritten when IOS is reloaded

4) We can make patches that are generic enough that they work on all
   versions of IOS.

As long as these are all true, then we can implement our code as a
replacement ELF loader; our loader also has the capability to patch
binaries as it loads them.  One of the patches it does is ... to hook
something* in the IOS reload path to replace the new IOS's loader with
a stashed copy of ourself before jumping to it.

The other patches it makes are just there to make life more
interesting; the most useful one is a patch to SVC call #4 to redirect
the normally /dev/nulled IOS debug output to USBGecko.

Usage:

After compiling Babelfish, you will end up with a babelfish.bin, which
is a replacement ELF loader that can be used with any version of IOS,
including boot2.  One easy way to use it is to take boot2v4, replace
the ELF loader with babelfish and use BootMii to load it as
armboot.bin.  After that, if all works well, it will stay resident in
memory until the system is powered down.

Known limitations and bugs:

1) This code is somewhat old, and was written before we found
   HW_AHBPROT; some of this may no longer be necessary.

2) There are several fragile points highlighted in the code, such as
   the maximum code size / stack size, reload_ios/xchange_osvers,
   syscall numbering, etc.

3) PPC patching "almost" works.

4) It won't survive a reboot to GameCube mode or through BootMii.

5) You probably have to have a USBGecko to get any use out of this.

6) There are several different patching strategies employed; it might
   be nice to clean up and/or refactor that code.

Safety notes:

Despite the Rube-Goldberg patching mechanism involved here, Babelfish
should pratically always be safer for your Wii than patching firmware
in flash.  The only dangerous scenario I can think of is that any
patches you make to the FFS driver may cause NAND to be corrupted in a
way that "vanilla" IOS doesn't undestand, which will cause IOS to
break if Babelfish fails to load.  Just be extra careful when touching
any code that deals with NAND unless you have BootMii/boot2 installed
with a recent NAND flash backup.bin.