The bit length of m is leaked through through timing in ecp_mul_mxz().
Initially found by Manuel Pégourié-Gonnard on ecp_mul_edxyz(), which has
been inspired from ecp_mul_mxz(), during initial review of the EdDSA PR.
See: https://github.com/Mbed-TLS/mbedtls/pull/3245#discussion_r490827996
Fix that by using grp->nbits + 1 instead, which anyway is very close to
the length of m, which means there is no significant performance impact.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
These issues were flagged by Coverity as instances where a local
variable may be used prior to being initialized. Please note that
none of these changes fixes any particular bug, this is just an attempt
to add more robustness.
Signed-off-by: Leonid Rozenboim <leonid.rozenboim@oracle.com>
Previously, ecp_add_mixed(), commputing say P+Q, would allow for the
Q parameter to have an unset Z coordinate as a shortcut for Z == 1.
This was leveraged during computation and usage of the T-table
(storing low multiples of the to-be-multiplied point on the curve).
It is a potentially error-prone corner case, though, since an MPIs
with unset data pointer coordinate and limb size 0 is also a valid
representation of the number 0.
As a first step towards removing ECP points with unset Z coordinate,
the constant time T-array getter ecp_select_comb() has previously
been modified to return 'full' mbedtls_ecp_point structures,
including a 1-initialized Z-coordinate.
Similarly, this commit ...
- Modifies ecp_normalize_jac_many() to set the Z coordinates
of the points it operates on to 1 instead of freeing them.
- Frees the Z-coordinates of the T[]-array explicitly
once the computation and normalization of the T-table has finished.
As a minimal functional difference between old and new code,
the new code also frees the Z-coordinate of T[0]=P, which the
old code did not.
- Modifies ecp_add_mixed() to no longer allow unset Z coordinates.
Except for the post-precomputation storage form of the T[] array,
the code does therefore no longer use EC points with unset Z coordinate.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
ecp_select_comb() did previously not set the Z coordinate of the target point.
Instead, callers would either set it explicitly or leave it uninitialized,
relying on the (only partly upheld) convention that sometimes an uninitialized
Z value represents 1.
This commit modifies ecp_select_comb() to always set the Z coordinate to 1.
This comes at the cost of memory for a single coordinate, which seems worth
it for the increased robustness.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
This improves readibility and prepares for further changes
like the introduction of a single double-width temporary for
ECP arithmetic.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
`ecp_add_mixed()` and `ecp_double_jac()` are the core subroutines
for elliptic curve arithmetic, and as such crucial for the performance
of ECP primitives like ECDHE and ECDSA.
This commit provides a very slight simplification and performance and
memory usage improvement to `ecp_add_mixed()` by removing the use of
three temporary MPIs used for coordinate calculations.
Where those variables were used, the code now writes directly to the
coordinate MPIs of the target elliptic curve point.
This is a valid change even if there is aliasing between input and
output, since at the time any of the coordinate MPIs in question is
written, the corresponding coordinates of both inputs are no longer
read.
(The analogous change in `ecp_double_jac()` can not be made since
this property does not hold for `ecp_double_jac()`.)
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
minor changes, such as improving the documentation for the byte reading
macros, and using MBEDTLS_PUT_UINT16_xy in place of byte reading
macro combinations
Signed-off-by: Joe Subbiani <joe.subbiani@arm.com>
byte shifting opertations throughout library/ were only replaced with
the byte reading macros when an 0xff mask was being used.
The byte reading macros are now more widley used, however they have not
been used in all cases of a byte shift operation, as it detracted from
the immediate readability or otherwise did not seem appropriate.
Signed-off-by: Joe Subbiani <joe.subbiani@arm.com>
mbedtls_ecp_read_key and mbedtls_ecp_write_key are updated to include
support for Curve448 as prescribed by RFC 7748 §5.
Test suites have been updated to validate curve448 under Montgomery
curves.
Signed-off-by: Archana <archana.madhavan@silabs.com>
These macros were moved into a header and now check-names.sh is failing.
Add an MBEDTL_ prefix to the macro names to make it pass.
Signed-off-by: Janos Follath <janos.follath@arm.com>
We were already rejecting them at the end, due to the fact that with the
usual (x, z) formulas they lead to the result (0, 0) so when we want to
normalize at the end, trying to compute the modular inverse of z will
give an error.
If we wanted to support those points, we'd a special case in
ecp_normalize_mxz(). But it's actually permitted by all sources (RFC
7748 say we MAY reject 0 as a result) and recommended by some to reject
those points (either to ensure contributory behaviour, or to protect
against timing attack when the underlying field arithmetic is not
constant-time).
Since our field arithmetic is indeed not constant-time, let's reject
those points before they get mixed with sensitive data (in
ecp_mul_mxz()), in order to avoid exploitable leaks caused by the
special cases they would trigger. (See the "May the Fourth" paper
https://eprint.iacr.org/2017/806.pdf)
Signed-off-by: Janos Follath <janos.follath@arm.com>
Clang was complaining and check-names.sh too
This only duplicates macros, so no impact on code size. In 3.0 we can
probably avoid the duplication by using an internal header under
library/ but this won't work for 2.16.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
We were already rejecting them at the end, due to the fact that with the
usual (x, z) formulas they lead to the result (0, 0) so when we want to
normalize at the end, trying to compute the modular inverse of z will
give an error.
If we wanted to support those points, we'd a special case in
ecp_normalize_mxz(). But it's actually permitted by all sources
(RFC 7748 say we MAY reject 0 as a result) and recommended by some to
reject those points (either to ensure contributory behaviour, or to
protect against timing attack when the underlying field arithmetic is
not constant-time).
Since our field arithmetic is indeed not constant-time, let's reject
those points before they get mixed with sensitive data (in
ecp_mul_mxz()), in order to avoid exploitable leaks caused by the
special cases they would trigger. (See the "May the Fourth" paper
https://eprint.iacr.org/2017/806.pdf)
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Upgrade the default list of hashes and curves allowed for TLS. The list is
now aligned with X.509 certificate verification: hashes and curves with at
least 255 bits (Curve25519 included), and RSA 2048 and above.
Remove MBEDTLS_TLS_DEFAULT_ALLOW_SHA1_IN_KEY_EXCHANGE which would no
longer do anything.
Document more precisely what is allowed by default.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
We know that Montgomery multiplication will never be called without an
RNG, so make that clear from the beginning of the function.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
It was indirectly calling ecp_mul() without an RNG. That's actually the
rare case where this should be allowed, as ecp_muladd() is typically
used on non-secret data (to verify signatures or ZKPs) and documented as
not being constant-time.
Refactor a bit in order to keep the ability to call ecp_mul() without a
RNG, but not exposed publicly (except though muladd).
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Fix trivial faulty calls in ECP test suite and ECP/ECJPAKE self-tests (by
adding a dummy RNG).
Several tests suites are not passing yet, as a couple of library
function do call ecp_mul() with a NULL RNG. The complexity of the fixes
range from "simple refactoring" to "requires API changes", so these will
be addressed in separate commits.
This makes the option MBEDTLS_ECP_NO_INTERNAL_RNG, as well as the whole
"internal RNG" code, obsolete. This will be addressed in a future
commit, after getting the test suites to pass again.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
library/ecp_alt.h (declaring individual functions of the ECP module that can
be substituted, included when building the library with
MBEDTLS_ECP_INTERNAL_ALT enabled) clashes with ecp_alt.h (not provided,
declaring types of the ECP module when the whole implementation is
substituted, included when building the library with MBEDTLS_ECP_ALT enabled).
Depending on the search path during build, this can make MBEDTLS_ECP_ALT
unusable.
Rename library/ecp_alt.h to follow the naming convention of other alt headers:
MBEDTLS_XXX_ALT corresponds to xxx_alt.h.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Instead of generating blinding values in a not-quite-uniform way
(https://github.com/ARMmbed/mbedtls/issues/4245) with copy-pasted
code, use mbedtls_mpi_random().
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Since mbedtls_mpi_random() is not specific to ECC code, move it from
the ECP module to the bignum module.
This increases the code size in builds without short Weierstrass
curves (including builds without ECC at all) that do not optimize out
unused functions.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Rename mbedtls_ecp_gen_privkey_sw to mbedtls_mpi_random since it has
no particular connection to elliptic curves beyond the fact that its
operation is defined by the deterministic ECDSA specification. This is
a generic function that generates a random MPI between 1 inclusive and
N exclusive.
Slightly generalize the function to accept a different lower bound,
which adds a negligible amount of complexity.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
mbedtls_ecp_gen_privkey_mx generates a random number with a certain
top bit set. Depending on the size, it would either generate a number
with that top bit being random, then forcibly set the top bit to
1 (when high_bit is not a multiple of 8); or generate a number with
that top bit being 0, then set the top bit to 1 (when high_bit is a
multiple of 8). Change it to always generate the top bit randomly
first.
This doesn't make any difference in practice: the probability
distribution is the same either way, and no supported or plausible
curve has a size of the form 8n+1 anyway. But it slightly simplifies
reasoning about the behavior of this function.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Don't calculate the bit-size of the initially generated random number.
This is not necessary to reach the desired distribution of private
keys, and creates a (tiny) side channel opportunity.
This changes the way the result is derived from the random number, but
does not affect the resulting distribution.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
The library rejected an RNG input of all-bits-zero, which led to the
key 2^{254} (for Curve25519) having a 31/32 chance of being generated
compared to other keys. This had no practical impact because the
probability of non-compliance was 2^{-256}, but needlessly
complicated the code.
The exception was added in 98e28a74e3 to
avoid the case where b - 1 wraps because b is 0. Instead, change the
comparison code to avoid calculating b - 1.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
For Montgomery keys, n_bits is actually the position of the highest
bit and not the number of bits, which would be 1 more (fence vs
posts). Rename the variable accordingly to lessen the confusion.
No semantic change.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Put the Montgomery and short Weierstrass implementations of
mbedtls_ecp_gen_privkey into their own function which can be tested
independently, but will not be part of the public ABI/API.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Prepare to isolate the Montgomery and short Weierstrass
implementations of mbedtls_ecp_gen_privkey into their own function.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
MBEDTLS_ECP_FIXED_POINT_OPTIM aims to speed up ecc multiplication performance.
We compute the comb table in runtime now. It is a costly operation.
This patch add a pre-computed table to initialize well-known curves. It speed up ECDSA signature verify process in runtime by using more ROM size.
Signed-off-by: kXuan <kxuanobj@gmail.com>
This header file will contain declarations of functions that are not
part of the public ABI/API, and must not be called from other modules,
but can be called from unit tests.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
This gives it a more descriptive name and indicates to alt developers
that the definitions inside are not fully internal and are available
to alt developers for use.
Signed-off-by: Chris Jones <christopher.jones@arm.com>
Simple find and replace using `#include (<|")mbedtls/(.*)_internal.h(>|")`
and `#include $1$2_internal.h$3`.
Also re-generated visualc files by running
`scripts/generate_visualc_files.pl`.
Signed-off-by: Chris Jones <christopher.jones@arm.com>
Fix function mbedtls_ecp_mul_shortcuts() to skip multiplication when m
is 0 and simply assignt 0 to R. Additionally fix ecjpake_zkp_read() to
return MBEDTLS_ERR_ECP_INVALID_KEY when the above condintion is met.
Fix#1792
Signed-off-by: TRodziewicz <rodziewicz@gmail.com>
mbedtls_ecp_curve_list() now lists Curve25519 and Curve448 under the names
"x25519" and "x448". These curves support ECDH but not ECDSA.
This was meant ever since the introduction of mbedtls_ecdsa_can_do()
in 0082f9df6f, but
2c69d10bac had removed the claim
that Montgomery curves support ECDH except through Everest.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
As a result, the copyright of contributors other than Arm is now
acknowledged, and the years of publishing are no longer tracked in the
source files.
Also remove the now-redundant lines declaring that the files are part of
MbedTLS.
This commit was generated using the following script:
# ========================
#!/bin/sh
# Find files
find '(' -path './.git' -o -path './3rdparty' ')' -prune -o -type f -print | xargs sed -bi '
# Replace copyright attribution line
s/Copyright.*Arm.*/Copyright The Mbed TLS Contributors/I
# Remove redundant declaration and the preceding line
$!N
/This file is part of Mbed TLS/Id
P
D
'
# ========================
Signed-off-by: Bence Szépkúti <bence.szepkuti@arm.com>
Run some self-test both for a short Weierstrass curve and for a
Montgomery curve, if the build-time configuration includes a curve of
both types. Run both because there are significant differences in the
implementation.
The test data is suitable for Curve25519.
Signed-off-by: Gilles Peskine <gilles.peskine@arm.com>
The constants used in the test worked with every supported curve
except secp192k1. For secp192k1, the "N-1" exponent was too large.
Signed-off-by: Gilles Peskine <gilles.peskine@arm.com>
Replace the now-redundant internal curve type macros ECP_xxx by the
macros MBEDTLS_ECP__xxx_ENABLED which are declared in ecp.h.
Signed-off-by: Gilles Peskine <gilles.peskine@arm.com>
Document that mbedtls_ecp_muladd and mbedtls_ecp_muladd_restartable
are only implemented on short Weierstrass curves.
Exclude these functions at build time if no short Weierstrass curve
is included in the build. Before, these functions failed to compile in
such a configuration.
Signed-off-by: Gilles Peskine <gilles.peskine@arm.com>
mbedtls_ecp_write_key is a mirror function to mbedtls_ecp_read_key, which
writes a private key back into a byte buffer in the correct format.
This is a helpful convenience function, since the byte order is defined
differently between Montgomery and Weierstrass curves. Since this difference
is accounted for in mbedtls_ecp_read_key, it made sense to add
mbedtls_ecp_write_key for the purpose of abstracting this away such that
psa_export_key doesn't need to take byte order into account.
Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>
In library source files, include "common.h", which takes care of
including "mbedtls/config.h" (or the alternative MBEDTLS_CONFIG_FILE)
and other things that are used throughout the library.
FROM=$'#if !defined(MBEDTLS_CONFIG_FILE)\n#include "mbedtls/config.h"\n#else\n#include MBEDTLS_CONFIG_FILE\n#endif' perl -i -0777 -pe 's~\Q$ENV{FROM}~#include "common.h"~' library/*.c 3rdparty/*/library/*.c scripts/data_files/error.fmt scripts/data_files/version_features.fmt
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
While this is a static function, so right now we know we don't need the check,
things may change in the future, so better be on the safe side.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
CTR-DRBG and HMAC-DRBG may used the seed differently depending on its length.
To avoid leaks, pass them a constant-length seed.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Checking the budget only after the randomization is done means sometimes we
were randomizing first, then noticing we ran out of budget, return, come back
and randomize again before we finally normalize.
While this is fine from a correctness and security perspective, it's a minor
inefficiency, and can also be disconcerting while debugging, so we might as
well avoid it.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
It results in smaller code than using CTR_DRBG (64 bytes smaller on ARMv6-M
with arm-none-eabi-gcc 7.3.1), so let's use this by default when both are
available.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Unless MBEDTLS_ECP_NO_INTERNAL_RNG is defined, it's no longer possible for
f_rng to be NULL at the places that randomize coordinates.
Eliminate the NULL check in this case:
- it makes it clearer to reviewers that randomization always happens (unless
the user opted out at compile time)
- a NULL check in a place where it's easy to prove the value is never NULL
might upset or confuse static analyzers (including humans)
- removing the check saves a bit of code size
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Currently we draw pseudo-random numbers at the beginning and end of the main
loop. With ECP_RESTARTABLE, it's possible that between those two occasions we
returned from the multiplication function, hence lost our internal DRBG
context that lives in this function's stack frame. This would result in the
same pseudo-random numbers being used for blinding in multiple places. While
it's not immediately clear that this would give rise to an attack, it's also
absolutely not clear that it doesn't. So let's avoid that by using a DRBG
context that lives inside the restart context and persists across
return/resume cycles. That way the RESTARTABLE case uses exactly the
same pseudo-random numbers as the non-restartable case.
Testing and compile-time options:
- The case ECP_RESTARTABLE && !ECP_NO_INTERNAL_RNG is already tested by
component_test_no_use_psa_crypto_full_cmake_asan.
- The case ECP_RESTARTABLE && ECP_NO_INTERNAL_RNG didn't have a pre-existing
test so a component is added.
Testing and runtime options: when ECP_RESTARTABLE is enabled, the test suites
already contain cases where restart happens and cases where it doesn't
(because the operation is short enough or because restart is disabled (NULL
restart context)).
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
While it seems cleaner and more convenient to set it in the top-level
mbedtls_ecp_mul() function, the existence of the restartable option changes
things - when it's enabled the drbg context needs to be saved in the restart
context (more precisely in the restart_mul sub-context), which can only be
done when it's allocated, which is in the curve-specific mul function.
This commit only internal drbg management from mbedtls_ecp_mul() to
ecp_mul_mxz() and ecp_mul_comb(), without modifying behaviour (even internal),
and a future commit will modify the ecp_mul_comb() version to handle restart
properly.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
The case of MBEDTLS_ECP_RESTARTABLE isn't handled correctly yet: in that case
the DRBG instance should persist when resuming the operation. This will be
addressed in the next commit.
When both CTR_DRBG and HMAC_DRBG are available, CTR_DRBG is preferred since
both are suitable but CTR_DRBG tends to be faster and I needed a tie-breaker.
There are currently three possible cases to test:
- NO_INTERNAL_RNG is set -> tested in test_ecp_no_internal_rng
- it's unset and CTR_DRBG is available -> tested in the default config
- it's unset and CTR_DRBG is disabled -> tested in
test_ecp_internal_rng_no_ctr_drbg
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
