Get rid of the usage of an sprintf and use snprintf instead. In this
case the buffer is 128, so big enough to not cause problems, though.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Fix a potential buffer overflow bug in the creation of filenames
that were using sprintf() rather than snprintf(). The buffer overflow
could occurr if the buffer is longer than 4096 bytes. The state path
may alone be 4096 bytes and could possibly trigger the overflow.
Swtpm for example is not affected from this since it uses the callbacks
that are invoked before the faulty function is called.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Check the size of the available data before hashing them. A minimum
of 20 bytes needs to be passed into the function so that we can hash
the data 'before' it.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Reported-by: Yi Ren <yunye.ry@alibaba-inc.com>
Fix a Coverity complaint by using iv.t.buffer rather than the
1-byte synonym (due to union) iv.b.buffer.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Implement key creation with OpenSSL if rand == NULL, meaning
that we create a non-KDF-derived key, so a purely random key.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Use the CRT (Chinese Remainder Theorem) method for doing the RSA
operations. Unfortunately it is not sufficient to just pass the
precalculated dP, dQ, and qInv parameters to OpenSSL private key
crypto functions but it also needs D, which is a bit more involved
to calculate.
We are not caching D (private exponent) as part of the OBJECT since
the OBJECT lives in size-limited NVRAM and we would need to add
'D' to the privateExponent_t, which would make it bigger and which
would allow less keys to fit into memory, which in turn could be-
come an issue if we wanted to resume a TPM that filled up the NVRAM
with keys and now less keys would fit into the NVRAM.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Implement helper functions to create an RSA public and private
key. Create the latter with the CRT parameters.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Use OpenSSL function to create TDES keys if rand == NULL,
which indicates that a truely random key needs to be generated
rather than one derived from a KDF.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Use OpenSSL functions to create EC keys only for the case that
rand == NULL in which case no KDF is being used and where we can
create a truly random key. This doesn't break the upgrade path.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Switch the ECDSA signature verification to OpenSSL. Do the signature
creation in the next step so we can verify the creation / verification
against the original TPM 2 code.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
OpenSSL's crypto library does not support all crypto functionality
we need in all versions. Elliptic curve support via EVP seems to
have been added much later than for example symmetric crypto support.
So, we move the USE_OPENSSL_FUNCTIONS out of Implementation.h
into configure.ac and let the build system detect what functionality
is available in the crypto library. In this patch we now also rename
USE_OPENSSL_FUNCTIONS to USE_OPENSSL_FUNCTIONS_SYMMETRIC to indicate
that we can use the symmetric crypto functions of the crypto lib.
Using the OpenSSL crypto support is enabled by default, so one has
to use --disable-use-openssl-functions, which we do for Travis now.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Convert functions for symmetric encryption via AES and TDES to
use OpenSSL EVP functions where necessary. Leave the old code around
and use the #if USE_OPENSSL_FUNCTIONS to activate the new parts.
OpenSSL does not provide an implementation for TDES in CTR mode, so
we reuse the original TPM 2 code for this mode.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Implement helper for getting OpenSSL crypto functions from
crytpo algorithm IDs, encryption modes, and key sizes.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Add a script that uses openssl to create the test data we are using
in the TPM 2 code already. Extend this script with TDES 128 and 192 bit
key sizes and tests.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Add some missing #define TPM_HAVE_TPM2_DECLARATIONS before the include
of "tpm_library_intern.h" in TPM 2 code so we don't run into compile
errors on RHEL 6 when data types are redefined in TPM 1.2 code.
Previous patch 73cad883ba seems to have missed those.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Remove some cruft that was useful when TPM 2 code was still in the
works but now that things have settled don't need it anymore.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Introduce TPM_NOMAINTENANCE_COMMANDS #define to build the TPM 1.2 code
without maintenance commands support. The state for the maintenance
commands has been written out so far, so we have to leave this part
alive even though nothing can be done with the maintenance key anymore.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
There are several functions in the code that do not have a caller.
Deactivate all of them to get higher coverage of the code.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
This script will permit integration with Google OSS-FUZZ
https://github.com/google/oss-fuzz
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
When no callbacks are set and the plain NVChip file is used it wasn't
possible to save the state of the TPM 2 after a TPMLIB_Terminate()
TPMLIB_MainInit() sequence with a cached state (TPMLIB_SetState) since
in this case the NVRAM was not initialized. We now initialize the NV
in case no callback for restoring the state is set.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Split off the part that was handling the NVChip file in _plat__NVEnable
and call this new function _plat__NVEnable_ChipFile so we can call it
separately in case the caller works with the plain NVChip file.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Some OpenSSL EC API calls are deprecated starting in OpenSSL 1.2
and may fail the build due to the #warning they produce.
The affected functions have been renamed and consolidated with
other functions inside OpenSSL:
- EC_POINT_set_affine_coorindates_GFp
https://github.com/openssl/openssl/blob/OpenSSL_1_1_1-stable/crypto/ec/ec_lib.c#L730
- EC_POINT_get_affine_coordinates_GFp
https://github.com/openssl/openssl/blob/OpenSSL_1_1_1-stable/crypto/ec/ec_lib.c#L768
On both functions the suffix _GFp has been removed by the renaming.
We need to call the new functions if OPENSSL_API_COMPAT is defined
and is >= 0x10200000L.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Add support for --enable-test-coverage that sets additional CFLAGS
and LDFLAGS.
gcov creates files with suffixes .gcov, .gcno, and .gcda that we
need clean up in a few directories.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
CryptRandStartup() did not propagate initialization errors to the caller
but always returned TRUE. We now return the initialization errors and
display an error message and some state variables.
This error should almost never occur, but the fuzzer was able to trigger
it with rev 146 when the random number generated returned 32 bit numbers only
and the same 32 bit number was returned as the one before it. With rev 150
the random number generating function now returns as many random numbers as
the caller requested and this error is not getting triggered anymore.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
