APDU Communication between Device and Host – Hardware Wallet Research #6

After we reverse engineered some parts of
the ledger firmware, AND found source code
of an older ledger firmware version, our next
goal will be to find, where the ledger handles
the APDU messages coming from USB.
And that is actually pretty easy.
We know from the debug output that the ledger
uses APDU messages to communicate.
application protocol data unit ( APDU ) is
the communication unit between a smart card
reader and a smart card.
It’s just one simple standard that you can
use to implement messages for communication.
When we reverse engineered the button_press
handler, we found a function that we called
Because this is where it continued to, after
we released the button.
And we also compared the assembly, with the
old source code from Ledger’s github.
And we saw that what we called bootloader_continue
appears to be the function bootloader_apdu_interp.
And the name already says a lot.
In there we see a io_usb_hid_init, followed
by a while(true) loop.
In case you didn’t know, having a for-loop
without any conditions, it is an endless loop,
like while(true).
And in this loop we see io_exchange (input/output
exchange) and a lot of references to APDU.
CHANNEL_APDU, no apdu received, G_io_apdu_buffer.
And we can also see switch/case statements,
that appear to uses one byte of the APDU buffer,
specifically the second byte – APDU Offset
for the Instruction.
And for example this case statement – Instruction
Get_version shows, that the byte 0x01 is the
instruction GET_VERSION.
From the other offset values we can also see
that the first byte of the APDU commands is
reffered to as CLA.
And we can see that before the switch/case
statement it checks that first byte.
It compares it to a constant CLA, which has
to be 0xE0!
And now remembe the APDU commands we saw in
the chrome app’s debug output.
Each APDU command started with 0xe0.
Which means each second byte here refers to
the Instruction, or the meaning, of this particular
apdu message.
Cool, right?
Now let’s have a quick look at this function
in IDA.
Zooming out we can directly spot the big loop.
Around everything.
And right at the start of the function we
see a compare to 0xe0.
So this part here is this this!
And when we follow the call graph we can also
find that IDA identified this as switch-case
This one has 5 cases.
If you leave the graph view it’s easier
to see how this is implemented.
First there is a compare to R0, to check if
it’s less or same.
And then branch, or jump, to the switch jump
Which means there are 5 switch cases.
0,1,2,3 and 4.
And this switch jump is implemented as a small
Here we can find a major difference between
ARM and other architectures like Intel x86.
ARM has a so called Link Register.
A link register is a special-purpose register
which holds the address to return to when
a function call completes.
On ARM the return address is not automatically
pushed onto the stack like with a call on
Instead it just writes the return address
into the link register LR.
If a function has multiple hierarchies down,
it has to remember the LR on the stack though,
otherwise the next branch will overwrite the
register again.
Thus many functions will Push it to the stack,
along a lot of other registers, very similar
to x86.
And also pop the pushed LR value back into
the program counter PC to return.
But you don’t need it if you don’t branch
And this small function doesn’t call anything
else, so it can just branch back to LR when
it’s done.
However if you pay attention, it actually
modifies LR before it jumps there.
LR would point into the data you see here.
This is not assembly.
But IDA already recognized this as a jump
table for the switch statement.
So first it moves LR, which points at the
start of the jump table into R1.
Then it performs a shift right, and a shift
It does that to clear the last bit that is
set, because we have thump code.
If that’s new to you, please checkout earlier
videos in the series.
So R1 now really points to here at the start.
And then we load a byte at the offset R0.
And the offset r0 is the switch value.
So it will load one of these bytes depending
on if R0 was 0, 1,2,3 or 4.
Then it will shift that byte to the left by
Which sounds weird but it’s actually super
Like we know, ARM code is always aligned.
And the last bit 0 or 1 decides if the instruction
is a thumb instruction or not.
We have here only thumb code, so the LR register,
which points to the next code to return to,
will have the last bit set.
And with the next instruction we see that
it adds the offset, it got in R1, onto the
link register.
The link register already has the last bit
set, which means this jump offset will never
have a bit 1 at the end.
It will always be 0.
This means you can safe one bit for your jump
For example if you want to jump to offset
That would be 1100.
The last bit will always be 0.
So you can just store 110.
And then shift left by one.
That doesn;t make much sense for a small value
like that.
But what if you want to jump to 0x14C.
That doesn’t fit into a byte?
Or does it?
Shifting it right by one, will make it 0xA6.
And then it fits into a byte.
And when we want to use it, we load that single
Shift it by left by one, and we get the last
zero back.
This means this 0xA6 here actually represents
a jump to the offset 0x14C.
So 0x14c PLUS 0x806 is 0x952.
So this case would juuuuump here!
Because my cursor is still placed on the jump
table function, IDA highlights the switch-case
jump targets with a black arrow.
So we can verify, yep, this is correct.
It jumps here.
IDA also added an automatic comment that this
is one of the cases for the jumptable from
So this is how you can reverse engineer and
work through that code.
Now let’s move away from IDA and the code
for a moment and look at the communication
from the HOST side.
So your computer.
If you do your research into ledger and the
provided software well, you will stumble over
a repository from ledger that contains “Python
tools for Ledger Blue and Nano S”.
For example you could find this code when
you look through ledger’s support article
describing how to check the hardware integrity.
And there they install a python module called
ledgerblue and use the checkGenuine function,
to verify the software on the chip.
And the source code for it is linked here.
When you look into the main function of this
module you can see how it talks to the ledger
There is a communication module that exposes
the functio getDongle, that returns a dongle
This is then passed to getDeployedSecretV2.
There we immediatly can see that it builds
an APDU message from raw bytes.
0xe0, is the CLA.
And the 0x4 is the instruction.
Looking into the firmware source code we can
check what instruction nr 0x4 would be, and
so that’s Instruction VALIDATE_TARGET_ID.
And it’s simple what it does.
It takes a received ID and compares it to
the constant device’s ID.
It’s just a function that you can use to
check that your device you are talking to
is in fact the device you meant to talk to.
Here in the python code you can see how the
ID you can specify is added to the APDU message.
Once you have that APDU data, you can then
call donge.exchange() to send that message
to the device!
From other instances of this exchange function
we can also see that it can return data.
So let’s test this.
Let’s get a python virtual environment…
Then let’s pip install ledgerblue.
Aaaaand now we can use it.
To test I open just a python interpreter.
from ledgerblue.comm import getDongle.
By the way, python pro-tip.
There is a builtin function help() which shows
you information about python functions and
For some stuff the help can be super detailed.
It’s like a linux man page.
So we can call help(getDongle) and we get
some info.
In this case it’s not amazing, but we will
see that the first parameter turns Debug on
and off.
So let’s get a dongle with getDongle and
set debug to true.
Next we can call dongle.exchange with a bytearray.
The bytearray starts with 0xe0, and then the
instruction, in this case 0x01, which stands
for Get_version.
Now before we hit enter we have to make sure
that the ledger is connected in bootloader
So pressing the button and connecting it.
Then let’s issue the command!
And it worked!
We see an outgoing APDU message which was
our get_version.
And we see an incoming message.
Now what these values exactly mean, I don’t
I guess somehow they contain the current version
of the device firmware.
And we can probably look through more of the
code to see how ledger parses these value
in their own software, but it’s not that
important for us right now.
Before we finish this video.
One last small thing.
Of course we can also debug this now easily
with gdb.
We know that here it checks the first byte
of the received APDU command to check if it’s
So we could use our GDB setup to set a breakpoint
But I wanted to show you something else.
IDA also can attach to a remote GDB server.
So we can select the debugger to be Remote
GDB debugger, and then go into the debugger
settings and set it to the gdb server port
exposed by the st-link utility.
We just have to make sure the utility is running,
and then we can select “attach to process”
in IDA. “attach to the process started on
target”, and IDA switches to the debugger
And we are at reset now.
And debugging like this is maybe a bit nicer,
because we can also have the graph view.
So let’s quickly navigate to the place where
it compares the APDU message and set a breakpoint
In this case a regular software breakpoint
didn’t seem to work for me.
So I went into the breakpoint list and edited
the breakpoint.
Making it a hardware breakpoint.
This is a feature of the Chip itself.
The st-utility even has an output telling
us that it found 4 hw breakpoint registers.
So hardware breakpoints allow us to tell the
arm chip to break on certain conditions.
In this case we say: “when you access this
address with the intention to execute the
instruction there, please stop!”.
Let’s try it out.
We do the same thing again, we press the button
of the ledger and hit continue, so the ledger
Okay, we are in bootloader mode and IDA shows
the process is running.
Now let’s issue our APDU command again,
and IDA immediately gets focus because we
hit the breakpoint.
So the byte of the APDU command is copied
from this address, which we know is a RAM
So we can use the hex view to navigate to
this address and look into the RAM of the
Here is our APDU command.
So this is cool.
And maybe looks better having this graph view
and so forth.
But like I already mentioned with the breakpoint
behaviour, this is also super janky.
For example if I try to perform a single step,
it just breaks.
Maybe there is somewhere a setting to configure
single stepping to also use hardware breakpoints
or something, but I don’t know.
Also I’m kinda used to normal GDB at this
Anyway… now we have everything in place
to look into the firmware update next video.


Add a Comment

Your email address will not be published. Required fields are marked *