This directory contains a demonstration of applying transparency patterns and principles to the problem of firmware updates. In particular it will focus on making firmware updates for a small compute platform discoverable.
See below for implementation status.
Firmware is ubiquitous; it's in your phone, watch, TV, alarm clock, baby monitor, WiFi devices, possibly even your lightbulbs if you have LED lamps. In any given desktop PC there's the BIOS/UEFI type firmware that many people are familiar with, but there are also scores of other hidden firmware blobs running on small controllers which power things like management engines, keyboards, network cards, hard-disks/SSDs. The list goes on.
Firmware is powerful, it runs at the highest privilege level possible and is often the bedrock on which the security story of the devices it powers depends.
It is also often almost entirely invisible and inscrutable, and in many cases has been shown to be insecure and vulnerable.
Today, the best-in-class vendors who supply the firmware also provide an update framework which verifies the integrity and authenticity of firmware updates before allowing them to be installed.
Even in this best case, how do we know that the signed firmware is not faulty, or even malicious? What if the signing identity used to assert authenticity of the firmware is somehow used to sign unintended updates (whether through outright compromise as in the Realtek identity used to sign the Stuxnet worm, or, perhaps, more subtly via some form of insider risk - be it malicious or otherwise)?
How will the publisher themselves even know this has happened? If they have been compromised, can they trust their key protection or audit logging?
How will the consumer of the update know whether they're being given the same update as all the other devices, or one especially crafted for just a small subset of folks?
Firmware Transparency is a mechanism to ensure that all firmware is discoverable. This means that the same list of published firmware is visible to the publisher, the devices which will be updated, and to folks like security researchers who can use state-of-the-art static analysis and inspection tooling to analyse them.
The goal is to have a system where firmware updates can not be {installed/booted} unless they have been made discoverable via a verifiable log.
- Define a claimant model description: FT Claimant Model
- Specify/document a system architecture for that model.
- Come up with some metadata format: FirmwareMetadata
- Build a simple personality around that format.
- Extend personality to also store firmware images.
- Build a simple tool to create metadata given a "boot" image (e.g. Linux Kernel, WASM binary, etc.), and log it via the personality.
- Figure out a way to package the metadata with the bootable image.
- Build a noddy "device" emulator which enforces logging requirements, and refuses to boot
an image unless all of the following are true:
- the metadata is present.
- the metadata has a "valid" signature
- the boot-image hash matches the one committed do in the metadata.
- a valid STH and inclusion proof for the metatdata is available and verifies correctly.
- Build a simple "flash" tool which refuses to flash an image to a device unless all of the boot-time requirements above are satisfied, in addition to requesting and validating a valid consistency proof between the previously seen STH and the new STH.
- Flash stores "proof bundle" on device for validation at boot time.
- Build simple monitor to tail the log and dump info from meta-data in realtime.
- Monitor is extended to validate firmware images hash
- Integrated STH Witness support, as an optional feature
Planned future enhancements:
- Add support for emulated and real hardware, e.g. via QEmu.
In running this demo you will take on the role of several different actors within the ecosystem to see how making firmware discoverable works in practice. You will then take the role of an attacker trying to install malicious code onto a device, and see how the application of transparency has made this attack much more expensive.
During the demo flow you will open several different terminals. It's far easier to use a terminal that supports tabs in order to do this. The terminals simply make your life easier as different logical actors have their own terminals in which to perform their actions.
Prerequisites:
- Install Docker and docker-compose
- Install Go (1.15+)
- Checkout:
- Open terminal in a directory that you're happy to check out code to:
git checkout https://github.com/google/trillian
git checkout https://github.com/google/trillian-examples
docker-compose -f trillian/examples/deployment/docker-compose.yml -f trillian-examples/binary_transparency/firmware/docker-compose.override.yml up -d personalityTrillian and the FT server will start in the background and provision a new log.
The monitor "tails" the log, fetching each of the added entries and checking for inconsistencies in the structure and unexpected or malicious entries.
For our demo, we'll scan each firmware binary for the word H4x0r3d
and consider that any binary containing that string is a bad one.
- Open a terminal in the
binary_transparency/firmwaredirectory, and run the command below to start a monitor:
go run ./cmd/ft_monitor/ --logtostderr --keyword="H4x0r3d" --state_file=/tmp/ftmon.stateThe vendor is going to publish a new, legitimate, firmware now.
- cd to the root of
binary_transparency/firmwarefor the following steps: - We're going to log a new "firmware" build, as the vendor would:
go run cmd/publisher/publish.go --logtostderr --v=2 --timestamp="2020-10-10T15:30:20.10Z" --binary_path=./testdata/firmware/dummy_device/example.wasm --output_path=/tmp/update.ota --device=dummyThis creates and submits a new firmware manifest to the log, waits for it to be included, and then builds a firmware update package ("OTA") and writes it out to local disk.
🔎 Very shortly you should see that the new firmware entry has been spotted by the
FT monitorabove.This is important! If the
Firmware Vendoris paying attention to the contents of the log, they can check that every piece of firmware they see logged there is expected and corresponds to a legitimate and known-good build. If they spot something unexpected then they're now aware that there is a problem which needs investigation...
Through the power of scripted narrative, the owner of the target device now
has a firmware update to install (we'll re-use the /tmp/update.ota file created
in the last step).
-
cd to the root of
binary_transparency/firmwarefor the following steps:Now that we have an update package for our new firmware, we can try flashing it to a device.
🐸 The repo contains a "dummy device" which uses the local disk to store the device's state. You'll need to choose and create a directory where this dummy device state will live - the instructions below assume that is `/tmp/dummy_device', change the path if you're using something different.
mkdir /tmp/dummy_device
We'll use the
cmd/flash_toolto do this flashing.⚠️ Note that the first time you do this the "dummy device" will have no state and the flashing process will fail. It will also fail if you've previously flashed firmware onto the device from a different log. In both of these cases, you can use the--forceflag on theflash_tool.go run ./cmd/flash_tool/ --logtostderr --update_file=/tmp/update.ota --device_storage=/tmp/dummy_device --device=dummy # --force if it's the first time -
Boot the device.
We'll boot the device emulator to check that everything is working ok. The "ROM" on the dummy device verifies the integrity of the firmware and proofs stored on the device.
go run ./cmd/emulator/dummy --logtostderr --dummy_storage_dir=/tmp/dummy_device
🐸 Because both the
flash_tooland the device itself verifies the correctness of the inclusion proofs, they are convinced that the firmware is now discoverable - anybody looking at the contents of the log also knows about its existence: this doesn't guarantee that the firmware is "good", but we know at least that it can't be a covert targeted attack, and we can assume that theFirmware vendoris aware of it too.
💻"Nice system you've got there. Let's test it..."
- cd to the root of
binary_transparency/firmwarefor the following steps.
🧨 The hacker has a malicious firmware they want to install on our
device. It's in testdata/firmware/dummy_device/hacked.wasm.
- Write malicious firmware directly onto the device.
Let's imagine the hacker has access to our device, they're going to write their malicious firmware directly over the top of our device's firmware:
cp testdata/firmware/dummy_device/hacked.wasm /tmp/dummy_device/firmware.bin
echo "mwuhahahaha :eyes:"Let's watch as the device owner turns on their device in the next step...
The device owner wants to use their device, however, unbeknownst to them it's been HACKED!
Start up the device:
go run ./cmd/emulator/dummy --logtostderr --dummy_storage_dir=/tmp/dummy_deviceWe should see that the device refuses to boot, with an error similar to this:
dummy_emu.go:41] ROM: "failed to verify bundle: firmware measurement does not match metadata (0xefb19feba9ea0e0d5de73ac16d8aa9c4ceb092ecd13eab5548f49a61e85c367a2f2c8ce1eb36b67e1407148406705e67663dc5b6d3f05a45475f6e4a2b69e285 != 0xbf2f21936b66a0665883716ea4b1ceda609304ad76dd48f6423128bc36d4cb0fb5effaa9c1f2e328a5cfc25d2cb89a337d4285a8bc3e22dbb99bddbed19e7095)"
🐸 This happened because the device measured the firmware and compared that with what the firmware manifest claimed the expected measurement should be. Since the device firmware had been overwritten, they didn't match and the device refused to boot.
Back to the drawing board, hacker!
We need to update the manifest on the device with the correct measurement for our hacked firmware. Run the following command:
go run ./cmd/hacker/modify_bundle \
--device dummy \
--binary ./testdata/firmware/dummy_device/hacked.wasm \
--input /tmp/dummy_device/bundle.json \
--output /tmp/dummy_device/bundle.json \
--sign=falseLet's watch as the device owner turns on their device in the next step...
Start the device:
go run ./cmd/emulator/dummy --logtostderr --dummy_storage_dir=/tmp/dummy_deviceand again, even though the device now gets past the firmware measurement check we should see that the device still refuses to boot, with an error similar to:
F1215 17:47:06.449540 1724707 dummy_emu.go:41] ROM: "failed to verify bundle: failed to verify signature on SignedStatement: failed to verify signature crypto/rsa: verification error"
exit status 1
Ok, this time we'll "steal" the key and use it to sign our malicious firmware. This is equivalent to the status-quo today.
Run the following command:
go run ./cmd/hacker/modify_bundle \
--device dummy \
--binary ./testdata/firmware/dummy_device/hacked.wasm \
--input /tmp/dummy_device/bundle.json \
--output /tmp/dummy_device/bundle.json \
--sign=trueLet's watch as the device owner turns on their device in the next step...
Start the device:
go run ./cmd/emulator/dummy --logtostderr --dummy_storage_dir=/tmp/dummy_deviceWe've got past the signature check, but now there's another error:
dummy_emu.go:41] ROM: "failed to verify bundle: invalid inclusion proof in bundle: calculated root:\n[202 202 214 35 92 129 74 43 92 63 27 232 69 79 93 26 187 86 24 174 32 49 53 19 122 252 252 241 139 226 122 79]\n does not match expected root:\n[186 61 229 40 73 60 245 168 87 2 6 107 225 25 186 169 85 12 74 158 126 168 255 168 27 149 245 138 27 211 67 234]"
exit status 1
🐸 This message indicates that the device was unable to verify that the firmware manifest was included in the log using the inclusion proof and checkpoint which are present in the
bundle.json. This makes sense - the modified firmware manifest hasn't been included!Think again, hacker!
The only way we're going to get the hacked firmware onto the device is to have a valid inclusion proof and checkpoint from the log which covers the new firmware manifest.
Let's have the hacker break into the firmware vendor's offices and add their modified manifest+firmware to the log...
go run cmd/publisher/publish.go --logtostderr --v=2 --timestamp="2020-10-10T23:00:00.00Z" --binary_path=./testdata/firmware/dummy_device/hacked.wasm --output_path=/tmp/bad_update.ota --device=dummy🐸 However, notice that the
FT Monitorhas spotted the firmware! Now the firmware vendor knows they have been compromised and can take action. :police:If the firmware vendor is complicit in the attack, then all is still not lost... Notice that the FT monitor has detected the malware in the new firmware too:
Malware detected matched pattern H4x0r3dAnybody else running a monitor also knows that malicious firmware has been logged and can raise the alarm.
We have seen how to publish firmware and its metadata to a log, and how this can be used by firmware readers: clients flashing the firmware, and monitors scanning the firmware. Having all parties seeing the same view of the available data sets the foundation for a secure system, but making clients aware of bad firmware cannot be done only with the tools we have above.
The ftmap README outlines a verifiable solution to this problem. There are 2 key changes from the solution above:
- In addition to firmware & metadata, the log also accepts annotations about firmware already in the log
- A Verifiable Map is built from the log, which aggregates each piece of firmware with the annotations for it
Clients can then use this verifiable map to efficiently discover all annotations about a piece of firmware before they rely on it.