If psa_key_derivation_internal() fails, it's up to the caller to clean
up. Do this, and add a note at the top of
psa_key_derivation_internal() and its auxiliary functions.
There is no non-regression test because at the moment the only way to
trigger an error is a borderline low-memory condition and we don't
have the means to trigger this.
Add missing checks for defined(MBEDTLS_MD_C) around types and
functions that require it (HMAC, HKDF, TLS12_PRF).
Add missing checks for defined(MBEDTLS_ECDSA_DETERMINISTIC) around
code that calls mbedtls_ecdsa_sign_det().
Add missing checks for defined(MBEDTLS_ECDH_C) around ECDH-specific
functions.
The standard prohibits calling memcpy() with NULL pointer
arguments, even if the size argument is 0.
The TLS-1.2 PRF generator setup function previously called
memcpy() with the label and salt as the source, even if
they were of length 0, as exercised by the derive_key_policy
test case in the PSA crypto test suite.
This commit adds guards around the memcpy() calls so that they
are only executed of salt or label have positive length, respectively.
In psa_key_agreement_ecdh, check that the public key is on the same
curve as the private key. The underlying mbedtls API doesn't check.
If the curves don't match, psa_key_agreement_ecdh is practically
guaranteed to return INVALID_ARGUMENT anyway, because way the code is
written, the public point is interpreted on the curve of the private
point, and it is rejected because the point is not on the curve. This
is why the test case "PSA key agreement setup: ECDH, raw: public key
on different curve" passed even before adding this check.
In ECDH key agreement, allow a public key with the OID id-ECDH, not
just a public key with the OID id-ecPublicKey.
Public keys with the OID id-ECDH are not permitted by psa_import_key,
at least for now. There would be no way to use the key for a key
agreement operation anyway in the current API.
psa_key_derivation requires the caller to specify a maximum capacity.
This commit adds a special value that indicates that the maximum
capacity should be the maximum supported by the algorithm. This is
currently meant only for selection algorithms used on the shared
secret produced by a key agreement.
On key import and key generation, for RSA, reject key sizes that are
not a multiple of 8. Such keys are not well-supported in Mbed TLS and
are hardly ever used in practice.
The previous commit removed support for non-byte-aligned keys at the
PSA level. This commit actively rejects such keys and adds
corresponding tests (test keys generated with "openssl genrsa").
Remove the need for an extra function mbedtls_rsa_get_bitlen. Use
mbedtls_rsa_get_len, which is only correct for keys whose size is a
multiple of 8. Key sizes that aren't a multiple of 8 are extremely
rarely used, so in practice this is not a problematic limitation.
Deprecate the module-specific XXX_HW_ACCEL_FAILED and
XXX_FEATURE_UNAVAILABLE errors, as alternative implementations should now
return `MBEDTLS_ERR_PLATFORM_HW_FAILED` and
`MBEDTLS_ERR_PLATFORM_FEATURE_UNSUPPORTED`.
ssl_write_handshake_msg() includes the assertion that
`ssl->handshake != NULL` when handling a record which is
(a) a handshake message, and NOT
(b) a HelloRequest.
However, it later calls `ssl_append_flight()` for any
record different from a HelloRequest handshake record,
that is, records satisfying !(a) || !(b), instead of
(a) && !(b) as covered by the assertion (specifically,
CCS or Alert records).
Since `ssl_append_flight()` assumes that `ssl->handshake != NULL`,
this rightfully triggers static analyzer warnings.
This commit expands the scope of the assertion to check
that `ssl->handshake != NULL` for any record which is not
a HelloRequest.
Revert changes for checking whether `MBEDTLS_ECP_RESTARTABLE`
is defined, since it broke the CI. The context is used whether the
restartable feature is defined or not.
1. Checge to check for `MBEDTLS_ECP_RESTARTABLE` for all definitions
of `rs_ctx`.
2. Remove checks for `_ALT` when using `rs_ctx` as they cannot coexist
with the Restartable configuration.
Previously, when checking whether a CRT was revoked through
one of the configured CRLs, the library would only consider
those CRLs whose `issuer` field binary-matches the `subject`
field of the CA that has issued the CRT in question. If those
fields were not binary equivalent, the corresponding CRL was
discarded.
This is not in line with RFC 5280, which demands that the
comparison should be format- and case-insensitive. For example:
- If the same string is once encoded as a `PrintableString` and
another time as a `UTF8String`, they should compare equal.
- If two strings differ only in their choice of upper and lower case
letters, they should compare equal.
This commit fixes this by using the dedicated x509_name_cmp()
function to compare the CRL issuer with the CA subject.
Fixes#1784.
library/certs.c provides some hardcoded certificates that
are used e.g. by the test applications ssl_server2, ssl_client2
in case no certificates are provided on the command line.
The certificates used are from the tests/data_files folder
and have been updated in the latest commits. This commit
updates their copies in certs.c. It also adds comments
indicating the files from which the data is taken, in
order to ease update in the future.
Previous commits have added or modified build instructions for
server1*, server2*, server5*, test-ca*, cli-rsa* in the Makefile
tests/data_files/Makefile, or the apps they invoke have been changed.
This commit regenerates those files to make sure they are in match with
the build instructions.
Previously, when checking whether a CRT was revoked through
one of the configured CRLs, the library would only consider
those CRLs whose `issuer` field binary-matches the `subject`
field of the CA that has issued the CRT in question. If those
fields were not binary equivalent, the corresponding CRL was
discarded.
This is not in line with RFC 5280, which demands that the
comparison should be format- and case-insensitive. For example:
- If the same string is once encoded as a `PrintableString` and
another time as a `UTF8String`, they should compare equal.
- If two strings differ only in their choice of upper and lower case
letters, they should compare equal.
This commit fixes this by using the dedicated x509_name_cmp()
function to compare the CRL issuer with the CA subject.
Fixes#1784.
Return the condition compilation flags surrounding
`mbedtls_ecdh_compute_shared()`, `mbedtls_ecdh_gen_public()`,
`mbedtls_ecdsa_sign()` and `mbedtls_ecdsa_verify()` that were accidentally
removed in a previous merge.
Resolves#2163
Skip all writing to the target buffer if its size is 0, since in this
case the pointer might be invalid and this would cause the calls to
memcpy and memset to have undefined behavior.
Change the import/export format of private elliptic curve keys from
RFC 5915 to the raw secret value. This commit updates the format
specification and the import code, but not the export code.
This commit modifies a bounds check in `mbedtls_ecp_check_budget()` to
be correct even if the requested number of ECC operations would overflow
the operation counter.
Use `( x >> y ) & z` instead of `x >> y & z`. Both are equivalent
by operator precedence, but the former is more readable and the
commonly used idiom in the library.
Correct a typo in an AES XTS implementation comment where the relevant
NIST standard was incorrectly referred to as NIST 80-38E instead of NIST
800-38E.
It is inaccurate to call a data unit a "sector". A disk sector is a
common use case for the data unit, but there exist other types of data
units that are not sectors.
This commit fixes issue #1212 related to platform-specific entropy
polling in an syscall-emulated environment.
Previously, the implementation of the entropy gathering function
`mbedtls_platform_entropy_poll()` for linux machines used the
following logic to determine how to obtain entropy from the kernel:
1. If the getrandom() system call identifier SYS_getrandom is present and
the kernel version is 3.17 or higher, use syscall( SYS_getrandom, ... )
2. Otherwise, fall back to reading from /dev/random.
There are two issues with this:
1. Portability:
When cross-compiling the code for a different
architecture and running it through system call
emulation in qemu, qemu reports the host kernel
version through uname but, as of v.2.5.0,
doesn't support emulating the getrandom() syscall.
This leads to `mbedtls_platform_entropy_poll()`
failing even though reading from /dev/random would
have worked.
2. Style:
Extracting the linux kernel version from
the output of `uname` is slightly tedious.
This commit fixes both by implementing the suggestion in #1212:
- It removes the kernel-version detection through uname().
- Instead, it checks whether `syscall( SYS_getrandom, ... )`
fails with errno set to ENOSYS indicating an unknown system call.
If so, it falls through to trying to read from /dev/random.
Fixes#1212.
This commit changes the behavior of the record decryption routine
`ssl_decrypt_buf()` in the following situation:
1. A CBC ciphersuite with Encrypt-then-MAC is used.
2. A record with valid MAC but invalid CBC padding is received.
In this situation, the previous code would not raise and error but
instead forward the decrypted packet, including the wrong padding,
to the user.
This commit changes this behavior to return the error
MBEDTLS_ERR_SSL_INVALID_MAC instead.
While erroneous, the previous behavior does not constitute a
security flaw since it can only happen for properly authenticated
records, that is, if the peer makes a mistake while preparing the
padded plaintext.
Wipe the whole MAC intermediate buffer, not just the requested MAC
size. With truncated MAC algorithms, the requested MAC size may be
smaller than what is written to the intermediate buffer.
This commit duplicates the public function mbedtls_asn1_find_named_data()
defined in library/asn1parse.c within library/asn1write.c in order to
avoid a dependency of the ASN.1 writing module on the ASN.1 parsing module.
The duplication is unproblematic from a semantic and an efficiency
perspective becasue it is just a short list traversal that doesn't
actually do any ASN.1 parsing.
Previously, mbedtls_pkcs5_pbes2() was unconditionally declared
in `pkcs5.h` but defined as a stub returning
`MBEDTLS_ERR_PKCS5_FEATURE_UNAVAILABLE` in case
MBEDTLS_ASN1_PARSE_C was not defined.
In line with the previous commits, this commit removes declaration
and definition from both `pkcs5.h` and `pkcs5.c` in case
MBEDTLS_ASN1_PARSE_C is not defined.
Our API makes no guarantee that you can use a context after free()ing it
without re-init()ing it first, so better not give the wrong impression that we
do, while it's not policy and the rest of the code might not allow it.
Rename the PLATFORM HW error, to avoid ABI breakage with Mbed OS.
The value changed as well, as previous value was not in the range of
Mbed TLS low level error codes.
* development:
ssl-opt.sh: change expected output for large srv packet test with SSLv3
Adapt ChangeLog
Fix bug in SSL ticket implementation removing keys of age < 1s
ssl-opt.sh: Add DTLS session resumption tests
Add ChangeLog entry
Fix typo
Fix hmac_drbg failure in benchmark, with threading
Remove trailing whitespace
Remove trailing whitespace
ssl_server2: add buffer overhead for a termination character
Add missing large and small packet tests for ssl_server2
Added buffer_size and response_size options for ssl-server2. Added appropriate tests.
Solving a conflict in tests/ssl-opt.sh: two set of tests were added at the
same place (just after large packets):
- restartable ECC tests (in this branch)
- server-side large packets (in development)
Resolution was to move the ECC tests after the newly added server large packet
ones.
The code assumed that `int x = - (unsigned) u` with 0 <= u < INT_MAX
sets `x` to the negative of u, but actually this calculates
(UINT_MAX - u) and then converts this value to int, which overflows.
Cast to int before applying the unary minus operator to guarantee the
desired behavior.
The code was making two unsequenced reads from volatile locations.
This is undefined behavior. It was probably harmless because we didn't
care in what order the reads happened and the reads were from ordinary
memory, but UB is UB and IAR8 complained.
This commit ensures that buffers holding fragmented or
future handshake messages get zeroized before they are
freed when the respective handshake message is no longer
needed. Previously, the handshake message content would
leak on the heap.
Context: This commit makes a change to mbedtls_pk_parse_key() which
is responsible for parsing of private keys. The function doesn't know
the key format in advance (PEM vs. DER, encrypted vs. unencrypted) and
tries them one by one, resetting the PK context in between.
Issue: The previous code resets the PK context through a call to
mbedtls_pk_free() along, lacking the accompanying mbedtls_pk_init()
call. Practically, this is not an issue because functionally
mbedtls_pk_free() + mbedtls_pk_init() is equivalent to mbedtls_pk_free()
with the current implementation of these functions, but strictly
speaking it's nonetheless a violation of the API semantics according
to which xxx_free() functions leave a context in uninitialized state.
(yet not entirely random, because xxx_free() functions must be idempotent,
so they cannot just fill the context they operate on with garbage).
Change: The commit adds calls to mbedtls_pk_init() after those calls
to mbedtls_pk_free() within mbedtls_pk_parse_key() after which the
PK context might still be used.
This commit removes the definition of the API function
`mbedtls_platform_set_calloc_free()`
from `library/platform.c` in case the macros
`MBEDTLS_PLATFORM_CALLOC_MACRO`
`MBEDTLS_PLATFORM_FREE_MACRO`
for compile time configuration of calloc/free are set.
This is in line with the corresponding header `mbedtls/platform.h`
which declares `mbedtls_platform_set_calloc_free()` only if
`MBEDTLS_PLATFORM_{CALLOC/FREE}_MACRO` are not defined.
Fixes#1642.
The previous code triggered a compiler warning because of a comparison
of a signed and an unsigned integer.
The conversion is safe because `len` is representable by 16-bits,
hence smaller than the maximum integer.
When a random number is generated for the Miller-Rabin primality test,
if the bit length of the random number is larger than the number being
tested, the random number is shifted right to have the same bit length.
This introduces bias, as the random number is now guaranteed to be
larger than 2^(bit length-1).
Changing this to instead zero all bits higher than the tested numbers
bit length will remove this bias and keep the random number being
uniformly generated.
When using a primality testing function the tolerable error rate depends
on the scheme in question, the required security strength and wether it
is used for key generation or parameter validation. To support all use
cases we need more flexibility than what the old API provides.
The input distribution to primality testing functions is completely
different when used for generating primes and when for validating
primes. The constants used in the library are geared towards the prime
generation use case and are weak when used for validation. (Maliciously
constructed composite numbers can pass the test with high probability)
The mbedtls_mpi_is_prime() function is in the public API and although it
is not documented, it is reasonable to assume that the primary use case
is validating primes. The RSA module too uses it for validating key
material.
The FIPS 186-4 RSA key generation prescribes lower failure probability
in primality testing and this makes key generation slower. We enable the
caller to decide between compliance/security and performance.
This python script calculates the base two logarithm of the formulas in
HAC Fact 4.48 and was used to determine the breakpoints and number of
rounds:
def mrpkt_log_2(k, t):
if t <= k/9.0:
return 3*math.log(k,2)/2+t-math.log(t,2)/2+4-2*math.sqrt(t*k)
elif t <= k/4.0:
c1 = math.log(7.0*k/20,2)-5*t
c2 = math.log(1/7.0,2)+15*math.log(k,2)/4.0-k/2.0-2*t
c3 = math.log(12*k,2)-k/4.0-3*t
return max(c1, c2, c3)
else:
return math.log(1/7.0)+15*math.log(k,2)/4.0-k/2.0-2*t
There was a lot of repetition between psa_aead_encrypt and
psa_aead_decrypt. Refactor the code into a new function psa_aead_setup.
The new code should behave identically except that in some cases where
multiple error conditions apply, the code may now return a different
error code.
Internally, I rearranged some of the code:
* I removed a check that the key type was in CATEGORY_SYMMETRIC because
it's redundant with mbedtls_cipher_info_from_psa which enumerates
supported key types explicitly.
* The order of some validations is different to allow the split between
setup and data processing. The code now calls a more robust function
psa_aead_abort in case of any error after the early stage of the setup.
In the previous bounds check `(*p) > end - len`, the computation
of `end - len` might underflow if `end` is within the first 64KB
of the address space (note that the length `len` is controlled by
the peer). In this case, the bounds check will be bypassed, leading
to `*p` exceed the message bounds by up to 64KB when leaving
`ssl_parse_server_psk_hint()`. In a pure PSK-based handshake,
this doesn't seem to have any consequences, as `*p*` is not accessed
afterwards. In a PSK-(EC)DHE handshake, however, `*p` is read from
in `ssl_parse_server_ecdh_params()` and `ssl_parse_server_dh_params()`
which might lead to an application crash of information leakage.
Get rid of the variable p. This makes it more apparent where the code
accesses the buffer at an offset whose value is sensitive.
No intended behavior change in this commit.
Rather than doing the quadratic-time constant-memory-trace on the
whole working buffer, do it on the section of the buffer where the
data to copy has to lie, which can be significantly smaller if the
output buffer is significantly smaller than the working buffer, e.g.
for TLS RSA ciphersuites (48 bytes vs MBEDTLS_MPI_MAX_SIZE).
In mbedtls_rsa_rsaes_pkcs1_v15_decrypt, use size_greater_than (which
is based on bitwise operations) instead of the < operator to compare
sizes when the values being compared must not leak. Some compilers
compile < to a branch at least under some circumstances (observed with
gcc 5.4 for arm-gnueabi -O9 on a toy program).
Replace memmove(to, to + offset, length) by a functionally equivalent
function that strives to make the same memory access patterns
regardless of the value of length. This fixes an information leak
through timing (especially timing of memory accesses via cache probes)
that leads to a Bleichenbacher-style attack on PKCS#1 v1.5 decryption
using the plaintext length as the observable.
mbedtls_rsa_rsaes_pkcs1_v15_decrypt takes care not to reveal whether
the padding is valid or not, even through timing or memory access
patterns. This is a defense against an attack published by
Bleichenbacher. The attacker can also obtain the same information by
observing the length of the plaintext. The current implementation
leaks the length of the plaintext through timing and memory access
patterns.
This commit is a first step towards fixing this leak. It reduces the
leak to a single memmove call inside the working buffer.
Make the function more robust by taking an arbitrary zero/nonzero
argument instead of insisting on zero/all-bits-one. Update and fix its
documentation.
stdio.h was being included both conditionally if MBEDTLS_FS_IO was
defined, and also unconditionally, which made at least one of them
redundant.
This change removes the unconditional inclusion of stdio.h and makes it
conditional on MBEDTLS_PLATFORM_C.
mbedtls_rsa_rsaes_pkcs1_v15_decrypt took care of calculating the
padding length without leaking the amount of padding or the validity
of the padding. However it then skipped the copying of the data if the
padding was invalid, which could allow an adversary to find out
whether the padding was valid through precise timing measurements,
especially if for a local attacker who could observe memory access via
cache timings.
Avoid this leak by always copying from the decryption buffer to the
output buffer, even when the padding is invalid. With invalid padding,
copy the same amount of data as what is expected on valid padding: the
minimum valid padding size if this fits in the output buffer,
otherwise the output buffer size. To avoid leaking payload data from
an unsuccessful decryption, zero the decryption buffer before copying
if the padding was invalid.
It should be valid to RSASSA-PSS sign a SHA-512 hash with a 1024-bit or
1032-bit RSA key, but with the salt size being always equal to the hash
size, this isn't possible: the key is too small.
To enable use of hashes that are relatively large compared to the key
size, allow reducing the salt size to no less than the hash size minus 2
bytes. We don't allow salt sizes smaller than the hash size minus 2
bytes because that too significantly changes the security guarantees the
library provides compared to the previous implementation which always
used a salt size equal to the hash size. The new calculated salt size
remains compliant with FIPS 186-4.
We also need to update the "hash too large" test, since we now reduce
the salt size when certain key sizes are used. We used to not support
1024-bit keys with SHA-512, but now we support this by reducing the salt
size to 62. Update the "hash too large" test to use a 1016-bit RSA key
with SHA-512, which still has too large of a hash because we will not
reduce the salt size further than 2 bytes shorter than the hash size.
The RSA private key used for the test was generated using "openssl
genrsa 1016" using OpenSSL 1.1.1-pre8.
$ openssl genrsa 1016
Generating RSA private key, 1016 bit long modulus (2 primes)
..............++++++
....++++++
e is 65537 (0x010001)
-----BEGIN RSA PRIVATE KEY-----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-----END RSA PRIVATE KEY-----
Setting the dh_flag to 1 used to indicate that the caller requests safe
primes from mbedtls_mpi_gen_prime. We generalize the functionality to
make room for more flags in that parameter.
OFB and CFB are streaming modes. XTS is a not a cipher mode but it
doesn't use a separate padding step. This leaves only CBC as a block
cipher mode that needs a padding step.
Since CBC is the only mode that uses a separate padding step, and is
likely to remain the only mode in the future, encode the padding mode
directly in the algorithm constant, rather than building up an
algorithm value from a chaining mode and a padding mode. This greatly
simplifies the interface as well as some parts of the implementation.
Mbed TLS distinguishes "invalid padding" from "valid padding but the
rest of the signature is invalid". This has little use in practice and
PSA doesn't report this distinction. We just report "invalid
signature".
There were only 5 categories (now 4). Reduce the category mask from 7
bits to 3.
Combine unformatted, not-necessarily-uniform keys (HMAC, derivation)
with raw data.
Reintroduce a KEY_TYPE_IS_UNSTRUCTURED macro (which used to exist
under the name KEY_TYPE_IS_RAW_DATA macro) for key types that don't
have any structure, including both should-be-uniform keys (such as
block cipher and stream cipher keys) and not-necessarily-uniform
keys (such as HMAC keys and secrets for key derivation).
The last slot in the array was not freed due to an off-by-one error.
Amend the fill_slots test to serve as a non-regression test for this
issue: without this bug fix, it would cause a memory leak.
MBEDTLS_PK_WRITE_C only requires either MBEDTLS_RSA_C or MBEDTLS_ECP_C to be defined.
Added wrappers to handle the cases where only one has been defined.
Moved mbedtls_pk_init to be within the ifdefs, so it's only called if appropriate.
In psa_generator_import_key, if generating a DES or 3DES key, set the
parity bits.
Add tests for deriving a DES key. Also test deriving an AES key while
I'm at it.
In psa_generator_hkdf_read, return BAD_STATE if we're trying to
construct more output than the algorithm allows. This can't happen
through the API due to the capacity limit, but it could potentially
happen in an internal call.
Also add a test case that verifies that we can set up HKDF with its
maximum capacity and read up to the maximum capacity.
New key type PSA_KEY_TYPE_DERIVE. New usage flag PSA_KEY_USAGE_DERIVE.
New function psa_key_derivation.
No key derivation algorithm is implemented yet. The code may not
compile with -Wunused.
Write some unit test code for psa_key_derivation. Most of it cannot be
used yet due to the lack of a key derivation algorithm.
Add an API for byte generators: psa_crypto_generator_t,
PSA_CRYPTO_GENERATOR_INIT, psa_crypto_generator_init,
psa_get_generator_capacity, psa_generator_read,
psa_generator_import_key, psa_generator_abort.
This commit does not yet implement any generator algorithm, it only
provides the framework. This code may not compile with -Wunused.
In psa_mac_setup and psa_hmac_setup_internal, perform a sanity check
on the hash size and the hash block size respectively. These sanity
checks should only trigger on an incompletely or incorrectly
implemented hash function.
Remove the check on the block size in psa_hmac_finish_internal
because at this point it has already been checked and used.
In the common case (key no longer than the block size), psa_hash_setup
was being called twice in succession. With current implementations
this is just a small performance loss, but potentially with
alternative implementations this could have lead to a memory leak.
Call psa_hash_setup in psa_hmac_setup_internal rather than
psa_mac_init. This makes it easier to use psa_hmac_setup_internal on
its own (for the sake of using HMAC internally inside the library).
Create internal functions for HMAC operations. This prepares for two
things: separating crypto-sensitive code from argument decoding and
validation, and using HMAC for other purposes than a MAC inside the
library (e.g. HMAC_DRBG, HKDF).
No intended observable behavior change in this commit.
Although RSASSA-PSS defines its input as a message to be hashed, we
implement a sign-the-hash function. This function can take an input
which isn't a hash, so don't restrict the size of the input, any more
than Mbed TLS does.
Remove a redundant check that hash_length fits in unsigned int for the
sake of Mbed TLS RSA functions.
Test that PSS accepts inputs of various lengths. For PKCS#1 v1.5
signature in raw mode, test the maximum input length.
No common signature algorithm uses a salt (RSA-PKCS#1v1.5, RSA-PSS,
DSA, ECDSA, EdDSA). We don't even take an IV for MAC whereas MAC
algorithms with IV are uncommon but heard of. So remove the salt
parameter from psa_asymmetric_sign and psa_asymmetric_verify.
We failed check-names.sh due to using a define which wasn't described or
defined anywhere. Even though we won't realistically enable
MBEDTLS_PSA_CRYPTO_SPM via the configuration system (and will do it from
PSA Crypto SPM tooling instead), add a description of the configuration to
config.h as good practice. Exclude MBEDTLS_PSA_CRYPTO_SPM from the "full"
configuration as well.
We failed all.sh on the "test: doxygen markup outside doxygen blocks" due
to doxygen markup being outside a Doxygen block. Add an extra `*` to the
psa_get_key_from_slot comment to denote the comment as a Doxygen comment.
Make function names for multipart operations more consistent (cipher
edition).
Rename symmetric cipher multipart operation functions so that they all
start with psa_cipher_:
* psa_encrypt_setup -> psa_cipher_encrypt_setup
* psa_decrypt_setup -> psa_cipher_decrypt_setup
* psa_encrypt_set_iv -> psa_cipher_set_iv
* psa_encrypt_generate_iv -> psa_cipher_generate_iv
Use if-else-if chains rather than switch because many blocks apply to
a class of algoritmhs rather than a single algorithm or a fixed set
of algorithms.
Call abort on more error paths that were missed earlier.
Reorganize error handling code in psa_mac_finish_internal,
psa_mac_sign_finish and psa_mac_verify finish to ensure that:
* psa_mac_abort() is always called, on all success and error paths.
* psa_mac_finish places a safe value in the output parameters on
all error paths, even if abort fails.
Make function names for multipart operations more consistent (MAC
setup edition).
Split psa_mac_setup into two functions psa_mac_sign_setup and
psa_mac_verify_setup. These functions behave identically except that
they require different usage flags on the key. The goal of the split
is to enforce the key policy during setup rather than at the end of
the operation (which was a bit of a hack).
In psa_mac_sign_finish and psa_mac_verify_finish, if the operation is
of the wrong type, abort the operation before returning BAD_STATE.
In mbedtls_rsa_rsaes_oaep_encrypt and
mbedtls_rsa_rsaes_pkcs1_v15_encrypt, if the input length is 0 (which
is unusual and mostly useless, but permitted) then it is fine for the
input pointer to be NULL. Don't return an error in this case.
When `input` is NULL, `memcpy( p, input, ilen )` has undefined
behavior even if `ilen` is zero. So skip the `memcpy` call in this
case. Likewise, in `mbedtls_rsa_rsaes_oaep_decrypt` and
`mbedtls_rsa_rsaes_pkcs1_v15_decrypt`, skip the `memcpy` call if
`*olen` is zero.
Isolate the code of psa_get_key_information that calculates the bit
size of a key into its own function which can be called by functions
that have a key slot pointer.
Add required includes in tests and psa_crypto.c file in order to be able to compilef for the SPM solution.
Some functions needed to be deprecated from psa_crypto.c since they already implemented in the SPM.
New functions psa_get_key_slot(), psa_get_empty_key_slot(),
psa_get_key_from_slot() to access a key slot object from a key slot
number. These functions perform all requisite validations:
* psa_get_key_slot() verifies that the key slot number is in range.
* psa_get_empty_key_slot() verifies that the slot is empty.
* psa_get_key_from_slot() verifies that the slot contains a key with
a suitable policy.
Always use these functions so as to make sure that the requisite
validations are always performed.
In psa_hash_finish and psa_mac_finish_internal, set the fallback
output length (which is reported on error) to the output buffer size,
not to the _expected_ buffer size which could be larger.
When the size of a buffer is 0, the corresponding pointer argument may
be null. In such cases, library functions must not perform arithmetic
on the pointer or call standard library functions such as memset and
memcpy, since that would be undefined behavior in C. Protect such
cases.
Refactor the storage of a 0-sized raw data object to make it store a
null pointer, rather than depending on the behavior of calloc(1,0).
The RSA module uses unsigned int for hash_length. The PSA Crypto API
uses size_t for hash_length. Cast hash_length to unsigned int when
passed to the hash module.
The GCM, CCM, RSA, and cipher modules inconsistently use int or unsigned
int for a count of bits. The PSA Crypto API uses size_t for counting
things. This causes issues on LLP64 systems where a size_t can hold more
than an unsigned int. Add casts for where key_bits and bits are passed to
mbedtls_* APIs.
Use size_t for block_size in psa_mac_abort() because
psa_get_hash_block_size() returns a size_t. This also helps to avoid
compiler warnings on LLP64 systems.
To avoid a possible loss of precision, and to be semantically correct,
use psa_key_slot_t (which is 16 bits) instead of size_t (which is 32 or
64 bits on common platforms) in mbedtls_psa_crypto_free().
Previously, the psa_set_key_lifetime() implementation did not match the
function declaration in psa/crypto.h. Value types don't need const,
since they are passed by value. Fix psa_set_key_lifetime()
implementation by making it match its declaration in the header.
This requires defining a maximum RSA key size, since the RSA key size
is the signature size. Enforce the maximum RSA key size when importing
or generating a key.
Fill the unused part of the output buffer with '!', for consistency
with hash and mac.
On error, set the output length to the output buffer size and fill the
output buffer with '!', again for consistency with hash and mac. This
way an invalid output is more visible in a memory dump.
Restructure the error paths so that there is a single place where the
unused part of the output buffer is filled.
Also remove a redundant initialization of *signature_length to 0.
Change the representation of an ECDSA signature from the ASN.1 DER
encoding used in TLS and X.509, to the concatenation of r and s
in big-endian order with a fixed size. A fixed size helps memory and
buffer management and this representation is generally easier to use
for anything that doesn't require the ASN.1 representation. This is
the same representation as PKCS#11 (Cryptoki) except that PKCS#11
allows r and s to be truncated (both to the same length), which
complicates the implementation and negates the advantage of a
fixed-size representation.
* Distinguish randomized ECDSA from deterministic ECDSA.
* Deterministic ECDSA needs to be parametrized by a hash.
* Randomized ECDSA only uses the hash for the initial hash step,
but add ECDSA(hash) algorithms anyway so that all the signature
algorithms encode the initial hashing step.
* Add brief documentation for the ECDSA signature mechanisms.
* Also define DSA signature mechanisms while I'm at it. There were
already key types for DSA.
* PSS needs to be parametrized by a hash.
* Don't use `_MGF1` in the names of macros for OAEP and PSS. No one
ever uses anything else.
* Add brief documentation for the RSA signature mechanisms.
Make psa_export_key() always set a valid data_length when exporting,
even when there are errors. This makes the API easier to use for buggy
programs (like our test code).
Our test code previously used exported_length uninitialized when
checking to see that the buffer returned was all zero in import_export()
in the case where an error was returned from psa_export_key().
Initialize exported_length to an invalid length, and check that it gets
set properly by psa_export_key(), to avoid this using export_length
uninitialized. Note that the mem_is_zero() check is still valid when
psa_export_key() returns an error, e.g. where exported_length is 0, as
we want to check that nothing was written to the buffer on error.
Out test code also previous passed NULL for the data_length parameter of
psa_export_key() when it expected a failure (in key_policy_fail()).
However, data_length is not allowed to be NULL, especially now that we
write to data_length from psa_export_key() even when there are errors.
Update the test code to not pass in a NULL data_length.
psa_hash_abort, psa_mac_abort and psa_cipher_abort now return
PSA_ERROR_BAD_STATE if operation->alg is obviously not valid, which
can only happen due to a programming error in the caller or in the
library. We can't detect all cases of calling abort on uninitialized
memory but this is dirt cheap and better than nothing.
It isn't used to define other macros and it doesn't seem that useful
for users. Remove it, we can reintroduce it if needed.
Define a similar function key_type_is_raw_bytes in the implementation
with a clear semantics: it's a key that's represented as a struct
raw_data.
Also add what was missing in the test suite to support block ciphers
with a block size that isn't 16.
Fix some buggy test data that passed only due to problems with DES
support in the product.
In psa_hash_start, psa_mac_start and psa_cipher_setup, return
PSA_ERROR_INVALID_ARGUMENT rather than PSA_ERROR_NOT_SUPPORTED when
the algorithm parameter is not the right category.
When psa_mac_start(), psa_encrypt_setup() or psa_cipher_setup()
failed, depending on when the failure happened, it was possible that
psa_mac_abort() or psa_cipher_abort() would crash because it would try
to call a free() function uninitialized data in the operation
structure. Refactor the functions so that they initialize the
operation structure before doing anything else.
Add non-regression tests and a few more positive and negative unit
tests for psa_mac_start() and psa_cipher_setup() (the latter via
psa_encrypt_setip()).
In psa_export_key, ensure that each byte of the output buffer either
contains its original value, is zero, or is part of the actual output.
Specifically, don't risk having partial output on error, and don't
leave extra data at the end of the buffer when exporting an asymmetric
key.
Test that exporting to a previously zeroed buffer leaves the buffer
zeroed outside the actual output if any.
Exporting an asymmetric key only worked if the target buffer had
exactly the right size, because psa_export_key uses
mbedtls_pk_write_key_der or mbedtls_pk_write_pubkey_der and these
functions write to the end of the buffer, which psa_export_key did not
correct for. Fix this by moving the data to the beginning of the
buffer if necessary.
Add non-regression tests.
psa_import_key must check that the imported key data matches the
expected key type. Implement the missing check for EC keys that the
curve is the expected one.
Avoid lines longer than 80 columns.
Remove some redundant parentheses, e.g. change
if( ( a == b ) && ( c == d ) )
to
if( a == b && c == d )
which makes lines less long and makes the remaining parentheses more
relevant.
Add missing parentheses around return statements.
There should be no semantic change in this commit.
Store the temporary key in the long-key case (where the key is first
hashed) directly into ipad. This reduces the stack usage a little, at
a slight cost in complexity.
In psa_mac_start, the hash of the key and ipad contain material that
can be used to make HMAC calculations with the key, therefore they
must be wiped.
In psa_mac_finish_internal, tmp contains an intermediate value which
could reveal the HMAC. This is definitely sensitive in the verify case,
and marginally sensitive in the finish case (it isn't if the hash
function is ideal, but it could make things worse if the hash function
is partially broken).
Split algorithm-specific code out of psa_mac_start. This makes the
function easier to read.
The behavior is mostly unchanged. In a few cases, errors before
setting a key trigger a context wipe where they didn't. This is a
marginal performance loss but only cases that are an error in caller
code.