Kmyth provides a simple mechanism for interacting with the TPM. The three core components are:
- kmyth-seal: A tool that encrypts a file and protects the encryption key by sealing it to the TPM
- kmyth-unseal: A tool that reads a file (.ski) that has been kmyth-sealed and produces the original content
- kmyth-getkey: A tool that demonstrates the programatic api of kmyth-seal and kmyth-unseal by protecting a certificate private key used in a TLS connection
In addition Kmyth provides a simple API which allows developers to use the TPM without having to become experts in the underlying TPM libraries.
Table of Contents
For build and installation instructions see the INSTALL file.
This tool will kmyth-seal a file using the TPM 2.0. In TPM parlance, 'sealing' typically refers to encrypting the file using a key known only to that TPM (i.e., binding that data to a specific TPM) and imposing system state criteria (i.e., Platform Configuration Register or PCR constraints) on the decryption of that data. While kmyth-seal utilizes the TPM's capability to 'seal' and/or 'bind' under the hood, it references the entire process used to create a .ski file result. This includes:
- generation and use of a wrapping key to symmetrically encrypt the input data
- use of the TPM to derive the Kmyth SRK as a primary key
- use of the TPM to create a Kmyth SK that is sealed to the SRK with an authorization policy
- use of the TPM to seal the symmetric wrapping key to the SK and an authorization policy
- compilation of the encrypted secret being protected (e.g., CAPK), TPM sealed storage key, TPM sealed symmetric key, and the symmetric cipher specification into a .ski file to facilitate the ability to later kmyth-unseal it
Because the TPM is used to encrypt the symmetric key that encrypts your data, the encrypted data can only be decrypted using the same machine (and the same TPM) with which it was encrypted. If the data is moved to a different machine, if the PCRs used to encrypt it change, or if the TPM is reset, the data will be permanently lost.
usage: ./bin/kmyth-seal [options]
: ./bin/kmyth-reseal [options]
options are:
-a or --auth_string String used to create 'authVal' digest. Defaults to empty string (all-zero digest).
-i or --input Path to file containing the data to be sealed.
-o or --output Destination path for the sealed file. Defaults to <filename>.ski in the CWD.
-f or --force Force the overwrite of an existing .ski file when using default output.
-p or --pcrs_list List of TPM platform configuration registers (PCRs) to apply to authorization policy.
Defaults to no PCRs specified. Encapsulate in quotes (e.g. "0, 1, 2").
-c or --cipher Specifies the cipher type to use. Defaults to 'AES/GCM/NoPadding/256'
-g or --get_exp_policy Retrieves the PolicyPCR digest associated with the current value of pcr registers
-e or --expected_policy Specifies an alternative digest value that can satisfy the authorization policy.
-l or --list_ciphers Lists all valid ciphers and exits.
-w or --owner_auth TPM 2.0 storage (owner) hierarchy authorization. Defaults to emptyAuth to match TPM default.
-v or --verbose Enable detailed logging.
-h or --help Help (displays this usage).
*kmyth-reseal is a specialization of kmyth-seal where the -g / --get_exp_policy option is forced but otherwise all the options are available. One may think of 'kmyth-reseal ..' as identical functionality as 'kmyth-seal -g ..'. Use of the -g flag within kmyth-reseal while allowed, is superfluous.
This tool will kmyth-unseal a file using the TPM 2.0. In TPM parlance, 'unsealing' typically refers to decrypting 'sealed' input data using a key known only to that TPM (i.e., the encrypted input is 'bound' to a specific TPM) while imposing system state criteria (i.e., Platform Configuration Register or PCR constraints) on the ability to decrypt. While kmyth-unseal utilizes the TPM's capability to 'unseal' under the hood, it references the entire process used to recover Kmyth protected secret data (e.g., a CAPK) from a .ski file input. This includes:
- recovery of data contained in the input .ski file
- use of the TPM and its Kmyth SRK to recover the Kmyth SK
- loading the recovered Kmyth SK into the TPM
- use of the TPM and loaded Kmyth SK to recover the symmetric wrapping key
- use of the symmetric wrapping key and the cipher specification to recover the 'kmyth-sealed' secret
- providing the recovered result to the user in the required format (e.g., a file)
usage: ./bin/kmyth-unseal [options]
options are:
-a or --auth_string String used to create 'authVal' digest. Defaults to empty string (all-zero digest).
-i or --input Path to file containing data the to be unsealed
-o or --output Destination path for unsealed file. This or -s must be specified. Will not overwrite any
existing files unless the 'force' option is selected.
-s or --stdout Output unencrypted result to stdout instead of file.
-w or --owner_auth TPM 2.0 storage (owner) hierarchy authorization. Defaults to emptyAuth to match TPM default.
-v or --verbose Enable detailed logging.
-h or --help Help (displays this usage).
This tool is used specifically for obtaining a key from a remote server.
In order to use kmyth-getkey some preliminary setup is required.
-
You must have a key server listening for connections.
-
The client running kmyth-getkey must have a private key sealed using kmyth-seal along with a corresponding certificate.
-
The key server must be able to authenticate the client's certificate.
usage: ./bin/kmyth-getkey [options]
options are:
Client Information --
-i or --input Path to file containing the kmyth-sealed client's certificate private key.
-l or --client Path to file containing the client's certificate.
Server Information --
-t or --type Type of key server backend (e.g., 'kmip', 'simple').
-s or --server Path to file containing the certificate
for the CA that issued the server cert.
-c or --conn_addr The ip_address:port for the TLS connection.
-m or --message An optional message to send the key server.
Output Parameters --
-o or --output Output file path to write the key. If none is selected, key will be sent to stdout.
Sealed Key Parameters --
-a or --auth_string String used to create 'authVal' digest. Defaults to empty string (all-zero digest)
-w or --owner_auth TPM 2.0 storage (owner) hierarchy authorization. Defaults to emptyAuth to match TPM default.
Misc --
-v or --verbose Detailed logging mode to help with debugging.
-h or --help Help (displays this usage).
-
TPM 2.0 Software Stack (TSS2)
-
The Feature API (FAPI) provides the highest-level of abstraction and was the last layer implemented (tss2-fapi).
-
The Enhanced System API (ESAPI) is the next layer down. While less abstract than the FAPI, it hides much of the session management overhead and provides cryptographic function support. Our development efforts preceded the initial release of an implementation of this layer (tss2-esys).
-
The System API (SAPI) provides the developer functionality that maps to the TPM 2.0 commands. The Kmyth TPM 2.0 code is currently based primarily on this layer (tss2-sys) of the TSS2 library.
-
The TPM Command Transmission Interface (TCTI) represents the layer where the actual TPM commands and responses are exchanged. In the TSS2 library, this layer is implemented by the tss2-tcti component.
-
TSS 2.0 Marshalling/Unmarshalling functionality is used to convert back and forth between C structures in the SAPI layer and the canonicalized commands/responses found in the TCTI layer. In the TSS2 library, these features are implemented by the tss2-mu component.
-
The TPM Access Broker (TAB) layer controls multi-process synchronization for the TPM. The Resource Manager (RM) swaps TPM sessions and objects in and out of very limited TPM memory, as needed. An open source implementation of these two layers is provided by the TPM2 Access Broker & Resource Manager Daemon (tss2-abrmd).
-
At the bottom of the software stack, the TPM device driver handles the physical transmission of data to and from the TPM. This is totally abstracted from our Kmyth code, however.
-
-
IBM's Software TPM 2.0 is "an implementation of the TCG TPM 2.0 specification. It is based on the TPM specification Parts 3 and 4 source code donated by Microsoft, with additional files to complete the implementation."
-
Microsoft provides the official TCG reference implementation of the TPM 2.0 Specification that includes a TPM 2.0 Simulator. The project provides Windows (Visual Studio) and Linux (autotools) build scripts.
-
hostname : 127.0.0.1 (localhost)
-
port : 2321 (TPM command)
-
the tpm2-abrmd binary is used to start the TPM Access Broker (TAB) and Resource Manager (RM) daemon.
-
The tpm2_takeownership binary can be used to set (change) the TPM's owner (storage) and endorsement hierarchy passwords (empty string by default), as well as the lockout authorization value.
-
Can be run on either emulator or device.
-
Must be run in root if running on device.
-
Must be in white list group or root to run.
-
To white list a group do the following:
- In /etc/dbus-1/system.d/tpm2-abrmd.conf add:
<policy group="[insert name of group here]"> <allow own="com.intel.tss2.Tabrmd"> </policy>- Then reset connection (e.g., dzdo pkill -HUP dbus-daemon)
-
Valgrind gives glib errors when connecting to resource manager. These are known errors present in the glib code. To supress only these errors do the following:
- Create a file, say libgio.supp, with the following text:
{ ignore_libglib_stuff Memcheck:Leak ... obj:*/libglib-2.0.so.* } { ignore_lib_gobject_stuff Memcheck:Leak ... obj:*/libgobject-2.0.so.* }- Then run valgrind with option --suppressions=libgio.supp
-
On TPM2_Shutdown() it indicates that the TPM should prepare for loss of power and save state required for an orderly startup TPM Reset.
-
On TPM2_Startup() it indicates that the TPM should perform Reset or Restart.
-
On TPM2_Shutdown(), it indicates that the TPM should prepare for loss of power and save state required for an orderly startup (TPM Restart or Resume).
-
On TPM2_Startup(), it indicates that the TPM should restore the state saved by the last TPM2_Shutdown().
-
0x80XXXXXX - Transient Objects that will not stay loaded between boots.
-
0x81XXXXXX - Persistent Objects that should stay loaded if you reboot. the machine.
-
A key hierarchy is created in TPM 2.0 by deriving a primary key using the TPM2_CreatePrimary() command.
-
A primary key has a "type" (e.g., must use one of the algorithms supported by the TPM.
-
The unique parameter in the inPublic struct passed to TPM2_CreatePrimary() provides a mechanism for the caller to generate (derive) different primary keys for the same algorithm within the same TPM hierarchy.
-
The outsideInfo parameter passed to TPM2_CreatePrimary() is a label for the primary key's creationData.
-
"TPM2.0 in Context" (Proudler, Chen, and Dalton) quote:
-
"Other input fields in TPM2_CreatePrimary are there to enable production of a 'creation' credential that is returned by the TPM2 with the SRK. The creation credential describes the circumstances in which the SRK was created: the current values of the selected PCRs ("creationPCR"), the name of the parent key (the reserved handle of the SPS), plus an arbitrary label ("outsideInfo") describing the SRK. Just like the SRK itself, this credential can be recreated at any time by repeating the TPM2_CreatePrimary command with the same parameters, as long as the SPS hasn't changed."
-
A key hierarchy is populated (e.g., Kmyth creates storage keys and sealed data objects) by using the TPM2_Create() command. Non-primary key and data objects have the following objectAttributes:
-
fixedTPM: SET prevents the key from being duplicated
-
stClear: SET clears object if platform properly switched off
-
fixedParent: SET only allows duplication with the parent (parent must also be duplicable).
-
sensitiveDataOrigin: SET indicates privKey generated by TPM
-
userWithAuth: CLEAR forces policy authorization for USER role
-
adminWithPolicy: SET forces policy authorization for ADMIN role
-
noDA: CLEAR enables dictionary attack protections
-
encryptedDuplication: SET forces encryption of sensitive portion of an object when duplicated.
-
restricted: SET means that key will refuse to operate on external data that mimics TPM-generated data. Storage keys (keys that are parents of other keys) must be restricted.
-
decrypt: SET means key can be used for decryption
-
sign: SET means key can be used for signatures
The most reliable and available source of random on a commercial machine is the kernel entropy accumulator. On modern UNIX/Linux machine this takes the form of /dev/random and /dev/urandom. These random sources have the advantage of being able to collect entropy from a variety of sources only available to the kernel. If correctly configured, they can carry entropy across a reboot cycle.
We use OpenSSL's RAND_bytes to generate our wrapping key. It seeds from /dev/urandom.
The user should be aware of the slow start issue present in kernel entropy accumulators. When a random number generator (RNG) is first configured, it has very little information available which cannot be guessed. Various mitigations for this problem are available, among them, writing a fresh random seed from some known good source to the entropy accumulator by writing to /dev/random. As the RNG continues to run, and more system input is fed in (interrupt timings, packet data, keyboard activity,...), anyone who is not in a position to see all of the input will eventually lose track of the internal state of the RNG.
For some applications, sensitive material must be available through the operating system's file IO interface; for example, one may need operational key accesible to processes that expect to read key material from a file. One option for limiting the exposure of sensitive material is to store it in a file in the ramfs RAM file system. We do not recommend the tmpfs RAM file system, because tmpfs may use swap space.