Protection against Physical Attacks
Physical attacks require adversaries to have direct access to a user's computer and cannot be conducted remotely. This section should be read in conjunction with the Full Disk Encryption and Encrypted Images chapters.
The Basic Input/Output System (BIOS) is non-volatile firmware which performs hardware initialization during the computer's booting process after it is powered on. It also provides runtime services for operating systems and progams. BIOS in modern PCs initialize and test system hardware components, as well as loading a boot loader or operating system from a mass memory device. The Unified Extensible Firmware Interface (UEFI) is the successor to BIOS that was released in 2011. 
All local settings are stored in BIOS, including power options, boot options and memory information. The BIOS menu allows the user to set and change a boot password for the computer upon startup. An administrator password can also be set to prevent others from changing BIOS settings. To set a BIOS boot password:  
- Turn on / restart the computer.
- Press the relevant key to access the BIOS menu. It is usually one of:
- Navigate to the Security or Password section using the arrow keys.
- Search for an entry named "Password on boot" or similar.
- Enter the new, strong password.
- Save the changes made to BIOS settings. On most PCs, this is done by pressing
Save and Exit. Check the bottom of the BIOS screen to be sure.
- Reboot the computer and confirm a password prompt now appears.
For greater security, a password should be set to access the BIOS menu itself. Search the Security or Password BIOS menu for "Set supervisor password", "User password", "System password", or something similar.  Also, users may prefer to configure BIOS to only allow booting from HDD/SSD so the computer cannot be booted from CD-ROM or USB flash drives.
Cold Boot Attacks
Modern computer architecture poses a significant risk to Whonix ™ users. Adversaries with physical access to a computer running Whonix ™ may be able to recover all session activities, even if FDE is enabled.
Even when a computer is powered off, the data in RAM does not immediately disappear. Depending on the circumstances, data can survive for up to several minutes. For example, this occurs when a computer loses power abruptly and does not go through the normal shutdown cycle.  If an adversary has immediate physical access to a computer, a cold boot attack can be mounted.
Forensic experts have two main methods of extracting data from RAM: 
- The running computer is cold-booted and a lightweight operating system is booted from a removable disk. A tool is used to dump pre-boot physical memory contents to a file.
- The memory modules are quickly removed from the original system and placed in another computer under the adversary's control. The machine is then booted to access the memory contents.
In both cases, the RAM contents can be analyzed in a computer forensics laboratory. Depending on what is found, the user may be in serious peril. Notably, cold boot attacks have proven effective against Trusted Platform Modules (TPMs), as well as full disk encryption regardless of the vendor or operating system. For certain memory modules, the time window for an attack can be extended to several hours by cooling them with a refrigerant. 
Cold boot attacks are thought to be a very uncommon method of recovering data, but high-risk users should be prepared for such a contingency to stay on the safe side. So long as a cold boot attack is not mounted directly after shutdown, then contents of RAM should be emptied within minutes. 
Whonix ™ does not yet provide an analogous feature to Tails, which wipes RAM on shutdown by overwriting it with random data. Possible interim solutions include:
- Configuring a computer to automatically shut down after a set period of inactivity.
- Dismounting encrypted disks.
- Not leaving the computer unattended immediately after shutdown.
- Soldering memory modules onto the motherboard.
- Using a kexec script to wipe RAM on shutdown.   
- Using TCG-compliant computer hardware.
- Using the TRESOR Linux kernel patch so CPU registers store encryption keys. 
- Waiting for full memory encryption.
Cold boot attacks are a clear and present danger for high-risk users due to the limited countermeasures available. In the purely hypothetical situation where an adversary is knocking earnestly on the door, safest would be pressing the panic button on the host, leading to the contents of RAM being quickly wiped. Failing that, the computer should be immediately shut down and access to the computer delayed as long as possible.
Evil Maid Attack
If an encrypted computer is left unattended, it is possible for an adversary ("evil maid") to secretly access it and install a hacked bootloader. The process has two steps: 
- After accessing the shutdown computer, the attacker boots it from a separate volume. A hacked bootloader is written to the system, then shut down.
- Later on the owner boots the computer and enters their encryption key. Once the disk is unlocked, the malicious bootloader can capture the key and send it over the internet, store it in a secret location and so on. 
In 2009, security researcher Joanna Rutkowska (who coined the term "evil maid attack") demonstrated this technique against a TrueCrypt encrypted system using a small bootable USB stick image; see here for further details and a discussion about possible solutions.
If the user has a TPM chip,  anti-evil maid measures are available for:
There are a number of computer interfaces that pose the risk of a direct memory access (DMA) attack. Potentially exploitable interfaces include ExpressCard, PCMCIA, FireWire, PCI, PCI Express or Thunderbolt.
In practice, attached devices are permitted to read and write directly to memory, often without supervision of the operating system. This is in contrast to user-mode applications that are usually prevented from accessing memory locations that are not explicitly authorized by virtual memory controllers. 
- Access sensitive cryptographic material in memory.
- Circumvent FDE.
- Inject executable code.
- Partially or fully read the memory address space.
- Read documents, files or other digital traces present in memory.
- Take control of the entire system, for example via the network.
- Unlock screensavers without a passphrase.
DMA attack software tools which mimic the abilities of state-level adversaries are even available on GitHub!  Mitigating the threat of a DMA attack requires mostly physical security countermeasures; it is recommended to:
- Consider blocking or removing them completely.
- Disable them in BIOS or UEFI.
- Never allow unknown and potentially malicious devices to be inserted into these ports. 
- Securely configure these interfaces.
- Use IOMMU technology if available, along with software which supports it, like Qubes. 
- Use Linux kernel options to disable DMA by Firewire devices.
Locking the screen on the host prevents others from viewing or using the device. It is advisable to set the screen to lock after a certain period of inactivity, and a strong password is recommended. Note that screen lockers provide notoriously weak protection, so do not overestimate their effectiveness. 
Open Start Menu→
Click User Icon→
Select Lock; or
Select Lock; or
Apple menu button→
Lock Screen; or
- Shortcuts are specific to the desktop environment in use, for example, GNOME, KDE, Xfce and so on.
Side Channel Attacks
Side-channel attacks are made possible by physical effects caused by cryptosystem operations (on the side) which provide extra information about system secrets like cryptographic keys, state information, or full/partial plaintexts. Wikipedia defines side-channel attacks as: 
...any attack based on information gained from the physical implementation of a cryptosystem, rather than brute force or theoretical weaknesses in the algorithms (compare cryptanalysis). For example, timing information, power consumption, electromagnetic leaks or even sound can provide an extra source of information, which can be exploited to break the system.
Side-channels emerge because computation takes place on a non-ideal system, composed of transistors, wires, power supplies, memory, and peripherals. Component characteristics vary with the instructions and data that are processed, allowing measurable variance to be used by attackers. 
Table: Primary Side-channel Attack Classes 
|Acoustic Cryptanalysis||Sound produced during computation is used for attacks.|
|Cache Attacks||Attackers monitor cache accesses made by the user in shared physical systems like virtualized environments or cloud services.|
|Data Remanence||Sensitive data are read after supposedly being deleted.|
|Differential Fault Analysis||Secrets are discovered by introducing faults in a computation.|
|Electromagnetic Attacks||Leaked electromagnetic radiation allows attacks that can provide plaintexts and other information. Cryptographic keys can be inferred via this method; for example, see TEMPEST.|
|Optical||Secrets and sensitive data are read by visual recordings with a high resolution camera, or other devices.|
|Power-monitoring Attacks||Attacks use measurements of varying hardware power consumption during computation.|
|Software-initiated Fault Attacks||Row hammer is an example of this attack, whereby off-limits memory is changed by rapidly accessing adjacent memory, leading to state retention loss.|
|Timing Attacks||Attacks are based on measuring how long various computations take to perform, such as the attacker's password compared to the user's unknown one.|
While Whonix ™ has some limited countermeasures to side-channel attacks, in general it cannot provide protection against most classes, nor hardware keyloggers, TEMPEST, miniature cameras and so on. Full disk encryption is also helpless against these attacks.
- If the system has both a supervisor password and a user password, then set passwords for both.
- Unfortunately, an upstream script does not yet exist to implement this feature, so Whonix ™ is currently unable to provide a solution for this attack.
- Tails and Liberte Linux have partially solved this problem.
- Instead of waiting for an upstream solution, see the Dev#Wipe RAM panic script. The user would need to implement a panic button which will wipe RAM. Please contribute by coding this feature.
- An advanced adversary is likely to return again and remove any evidence of the hack.
TPM is is a microcontroller that stores keys, passwords, and digital certificates.
- By installing the anti-evil-maid package in dom0 and following the configuration steps.
- This is not an endorsement for the use of hacking tools.
- This is another reason why high-risk users should never leave their devices unattended.
- IOMMU maps device-visible virtual addresses to physical addresses. The security benefit is that operating systems that are run in guest virtualized machines -- AppVMs in Qubes -- do not know the physical memory addresses on the host that are being accessed. This makes DMA attacks very difficult and can lead to memory corruption if attempted.
- Attacks that have bypassed screen lockers on most platforms can easily be found online.
- Some macOS systems instead use
Whonix ™ Protection Against Physical Attacks wiki page Copyright (C) Amnesia <amnesia at boum dot org>
Whonix ™ Protection Against Physical Attacks wiki page Copyright (C) 2012 - 2019 ENCRYPTED SUPPORT LP <email@example.com>
This program comes with ABSOLUTELY NO WARRANTY; for details see the wiki source code.
This is free software, and you are welcome to redistribute it under certain conditions; see the wiki source code for details.
This is a wiki. Want to improve this page? Help is welcome and volunteer contributions are happily considered! Read, understand and agree to Conditions for Contributions to Whonix ™, then Edit! Edits are held for moderation.
Copyright (C) 2012 - 2019 ENCRYPTED SUPPORT LP. Whonix ™ is a trademark. Whonix ™ is a licensee of the Open Invention Network. Unless otherwise noted, the content of this page is copyrighted and licensed under the same Freedom Software license as Whonix ™ itself. (Why?)