Fix a buffer overflow in mbedtls_mpi_sub_abs() when calculating
|A| - |B| where |B| is larger than |A| and has more limbs (so the
function should return MBEDTLS_ERR_MPI_NEGATIVE_VALUE).
Fix#4042
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
T1 is set to a 2-limb value. The first operation that takes it as
input is mbedtls_mpi_mul_int, which makes it grow to 3 limbs. Later it
is shifted left, which causes it to grow again. Set its size to the
final size from the start. This saves two calls to calloc(), at the
expense of a slowdown in some operations involving T1 as input since
it now has more leading zeros.
Setting T1 to 3 limbs initially instead of 2 saves about 6% of the
calloc() calls in test_suite_ecp and does not incur a performance
penalty. Setting T1 to A->n + 2 limbs instead of 2 saves about 20% of
the calloc calls and does not cause a measurable performance
difference on my Linux PC.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Note a possible microoptimization in mbedtls_mpi_mul_hlp that I tried
in the hope of reducing the number of allocations, but turned out to
be counterproductive.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Rewrite mbedtls_mpi_mul_int to call mpi_mul_hlp directly rather than
create a temporary mpi object. This has the benefit of not performing
an allocation when the multiplication is in place (mpi operand aliased
with the result) and the result mpi is large enough.
This saves about 40% of the calloc() calls in test_suite_ecp. There is
no measurable performance difference on my Linux PC.
The cost is a few bytes in bignum.o.
When there is no aliasing, or when there is aliasing but the mpi
object needs to be enlarged, the performance difference is negligible.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
mbedtls_mpi_sub_abs systematically allocated a new mpi when the result
was aliased with the right operand (i.e. X = A - X). This aliasing
very commonly happens during ECP operations. Rewrite the function to
allocate only if the result might not fit otherwise.
This costs a few bytes of code size in bignum.o, and might make
mbedtls_mpi_sub_abs very very slightly slower when no reallocation is
done. However, there is a substantial performance gain in ECP
operations with Montgomery curves (10-20% on my PC).
test_suite_ecp drops from 1422794 to 1271506 calls to calloc().
This commit also fixes a bug whereby mbedtls_mpi_sub_abs would leak
memory when X == B (so TB was in use) and the result was negative.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Fix a memory leak in mbedtls_mpi_sub_abs when the output parameter is
aliased to the second operand (X = A - X) and the result is negative.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
* development: (488 commits)
Fix removal of deprecated PSA constants
Use GitHub-compatible table formatting
Remove psa header files in uninstall part
Change function casting in `ssl_calc_finished_tls_sha384`
Fix GCC warning in `ssl_calc_finished_tls_sha384`
Add changelog entry file to `ChangeLog.d`
Fix GCC warning in `ssl_calc_finished_tls_sha384`
Fix GCC warning about `test_snprintf`
Fix mismatched function parameters (prototype/definition)
Fix build failure on gcc-11
Copyediting
Clarifications around key import
Fix copypasta
A variable is unused in some configurations
Rename test_driver_keygen to test_driver_key_management
Move "internal use" sentence attached to the wrong function
Added changelog
Plug in the entry point for public key export through driver
tests: psa: Reset key attributes where needed
Improve/fix documentation
...
Check that the exponent and modulus is below `MBEDTLS_MPI_MAX_BITS` before
performing a time expensive operation (modular exponentiation). This prevents
a potential DoS from Diffie-Hellman computations with extremely
large key sizes.
Signed-off-by: Chris Jones <christopher.jones@arm.com>
Probably the `W[2 << MBEDTLS_MPI_WINDOW_SIZE]` notation is based on a transcription of 2**MBEDTLS_MPI_WINDOW_SIZE.
Signed-off-by: Daniel Otte <d.otte@wut.de>
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
$!N
/This file is part of Mbed TLS/Id
P
D
'
# ========================
Signed-off-by: Bence Szépkúti <bence.szepkuti@arm.com>
In library source files, include "common.h", which takes care of
including "mbedtls/config.h" (or the alternative MBEDTLS_CONFIG_FILE)
and other things that are used throughout the library.
FROM=$'#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>
The function mbedtls_mpi_sub_abs first checked that A >= B and then
performed the subtraction, relying on the fact that A >= B to
guarantee that the carry propagation would stop, and not taking
advantage of the fact that the carry when subtracting two numbers can
only be 0 or 1. This made the carry propagation code a little hard to
follow.
Write an ad hoc loop for the carry propagation, checking the size of
the result. This makes termination obvious.
The initial check that A >= B is no longer needed, since the function
now checks that the carry propagation terminates, which is equivalent.
This is a slight performance gain.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
There was some confusion during review about when A->p[n] could be
nonzero. In fact, there is no need to set A->p[n]: only the
intermediate result d might need to extend to n+1 limbs, not the final
result A. So never access A->p[n]. Rework the explanation of the
calculation in a way that should be easier to follow.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
The function mpi_sub_hlp had confusing semantics: although it took a
size parameter, it accessed the limb array d beyond this size, to
propagate the carry. This made the function difficult to understand
and analyze, with a potential buffer overflow if misused (not enough
room to propagate the carry).
Change the function so that it only performs the subtraction within
the specified number of limbs, and returns the carry.
Move the carry propagation out of mpi_sub_hlp and into its caller
mbedtls_mpi_sub_abs. This makes the code of subtraction very slightly
less neat, but not significantly different.
In the one other place where mpi_sub_hlp is used, namely mpi_montmul,
this is a net win because the carry is potentially sensitive data and
the function carefully arranges to not have to propagate it.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
mpi_sub_hlp performs a subtraction A - B, but took parameters in the
order (B, A). Swap the parameters so that they match the usual
mathematical syntax.
This has the additional benefit of putting the output parameter (A)
first, which is the normal convention in this module.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Let code analyzers know that this is deliberate. For example MSVC
warns about the conversion if it's implicit.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
In mpi_montmul, an auxiliary function for modular
exponentiation (mbedtls_mpi_mod_exp) that performs Montgomery
multiplication, the last step is a conditional subtraction to force
the result into the correct range. The current implementation uses a
branch and therefore may leak information about secret data to an
adversary who can observe what branch is taken through a side channel.
Avoid this potential leak by always doing the same subtraction and
doing a contant-trace conditional assignment to set the result.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Separate out a version of mpi_safe_cond_assign that works on
equal-sized limb arrays, without worrying about allocation sizes or
signs.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
This reverts commit 2cc69fffcf.
A check was added in mpi_montmul because clang-analyzer warned about a
possibly null pointer. However this was a false positive. Recent
versions of clang-analyzer no longer emit a warning (3.6 does, 6
doesn't).
Incidentally, the size check was wrong: mpi_montmul needs
T->n >= 2 * (N->n + 1), not just T->n >= N->n + 1.
Given that this is an internal function which is only used from one
public function and in a tightly controlled way, remove both the null
check (which is of low value to begin with) and the size check (which
would be slightly more valuable, but was wrong anyway). This allows
the function not to need to return an error, which makes the source
code a little easier to read and makes the object code a little
smaller.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
If Y was constructed through functions in this module, then Y->n == 0
iff Y->p == NULL. However we do not prevent filling mpi structures
manually, and zero may be represented with n=0 and p a valid pointer.
Most of the code can cope with such a representation, but for the
source of mbedtls_mpi_copy, this would cause an integer underflow.
Changing the test for zero from Y->p==NULL to Y->n==0 causes this case
to work at no extra cost.
In the case of *ret we might need to preserve a 0 value throughout the
loop and therefore we need an extra condition to protect it from being
overwritten.
The value of done is always 1 after *ret has been set and does not need
to be protected from overwriting. Therefore in this case the extra
condition can be removed.
The code relied on the assumptions that CHAR_BIT is 8 and that unsigned
does not have padding bits.
In the Bignum module we already assume that the sign of an MPI is either
-1 or 1. Using this, we eliminate the above mentioned dependency.
The signature of mbedtls_mpi_cmp_mpi_ct() meant to support using it in
place of mbedtls_mpi_cmp_mpi(). This meant full comparison functionality
and a signed result.
To make the function more universal and friendly to constant time
coding, we change the result type to unsigned. Theoretically, we could
encode the comparison result in an unsigned value, but it would be less
intuitive.
Therefore we won't be able to represent the result as unsigned anymore
and the functionality will be constrained to checking if the first
operand is less than the second. This is sufficient to support the
current use case and to check any relationship between MPIs.
The only drawback is that we need to call the function twice when
checking for equality, but this can be optimised later if an when it is
needed.
Multiplication is known to have measurable timing variations based on
the operands. For example it typically is much faster if one of the
operands is zero. Remove them from constant time code.
1. variable name accoriding to the Mbed TLS coding style;
2. add a comment explaining safety of the optimization;
3. safer T2 initialization and memory zeroing on the function exit;
* restricted/pr/551:
ECP: Clarify test descriptions
ECP: remove extra whitespaces
Fix ECDH secret export for Mongomery curves
Improve ECP test names
Make ecp_get_type public
Add more tests for ecp_read_key
ECP: Catch unsupported import/export
Improve documentation of mbedtls_ecp_read_key
Fix typo in ECP module
Remove unnecessary cast from ECP test
Improve mbedtls_ecp_point_read_binary tests
Add Montgomery points to ecp_point_write_binary
ECDH: Add test vectors for Curve25519
Add little endian export to Bignum
Add mbedtls_ecp_read_key
Add Montgomery points to ecp_point_read_binary
Add little endian import to Bignum
The function `mbedtls_mpi_write_binary()` writes big endian byte order,
but we need to be able to write little endian in some caseses. (For
example when handling keys corresponding to Montgomery curves.)
Used `echo xx | tac -rs ..` to transform the test data to little endian.
The private keys used in ECDH differ in the case of Weierstrass and
Montgomery curves. They have different constraints, the former is based
on big endian, the latter little endian byte order. The fundamental
approach is different too:
- Weierstrass keys have to be in the right interval, otherwise they are
rejected.
- Any byte array of the right size is a valid Montgomery key and it
needs to be masked before interpreting it as a number.
Historically it was sufficient to use mbedtls_mpi_read_binary() to read
private keys, but as a preparation to improve support for Montgomery
curves we add mbedtls_ecp_read_key() to enable uniform treatment of EC
keys.
For the masking the `mbedtls_mpi_set_bit()` function is used. This is
suboptimal but seems to provide the best trade-off at this time.
Alternatives considered:
- Making a copy of the input buffer (less efficient)
- removing the `const` constraint from the input buffer (breaks the api
and makes it less user friendly)
- applying the mask directly to the limbs (violates the api between the
modules and creates and unwanted dependency)
The function `mbedtls_mpi_read_binary()` expects big endian byte order,
but we need to be able to read from little endian in some caseses. (For
example when handling keys corresponding to Montgomery curves.)
Used `echo xx | tac -rs .. | tr [a-z] [A-Z]` to transform the test data
to little endian and `echo "ibase=16;xx" | bc` to convert to decimal.
In mbedtls_mpi_exp_mod(), the limit check on wsize is never true when
MBEDTLS_MPI_WINDOW_SIZE is at least 6. Wrap in a preprocessor guard
to remove the dead code and resolve a Coverity finding from the
DEADCODE checker.
Change-Id: Ice7739031a9e8249283a04de11150565b613ae89
Fixes memory leak in mpi_miller_rabin() that occurs when the function has
failed to obtain a usable random 'A' 30 turns in a row.
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>
mbedtls_mpi_read_binary() calls memcpy() with the source pointer being
the source pointer passed to mbedtls_mpi_read_binary(), the latter may
be NULL if the buffer length is 0 (and this happens e.g. in the ECJPAKE
test suite). The behavior of memcpy(), in contrast, is undefined when
called with NULL source buffer, even if the length of the copy operation
is 0.
This commit fixes this by explicitly checking that the source pointer is
not NULL before calling memcpy(), and skipping the call otherwise.
Context: The function `mbedtls_mpi_fill_random()` uses a temporary stack
buffer to hold the random data before reading it into the target MPI.
Problem: This is inefficient both computationally and memory-wise.
Memory-wise, it may lead to a stack overflow on constrained devices with
limited stack.
Fix: This commit introduces the following changes to get rid of the
temporary stack buffer entirely:
1. It modifies the call to the PRNG to output the random data directly
into the target MPI's data buffer.
This alone, however, constitutes a change of observable behaviour:
The previous implementation guaranteed to interpret the bytes emitted by
the PRNG in a big-endian fashion, while rerouting the PRNG output into the
target MPI's limb array leads to an interpretation that depends on the
endianness of the host machine.
As a remedy, the following change is applied, too:
2. Reorder the bytes emitted from the PRNG within the target MPI's
data buffer to ensure big-endian semantics.
Luckily, the byte reordering was already implemented as part of
`mbedtls_mpi_read_binary()`, so:
3. Extract bigendian-to-host byte reordering from
`mbedtls_mpi_read_binary()` to a separate internal function
`mpi_bigendian_to_host()` to be used by `mbedtls_mpi_read_binary()`
and `mbedtls_mpi_fill_random()`.
The MPI_VALIDATE_RET() macro cannot be used for parameter
validation of mbedtls_mpi_lsb() because this function returns
a size_t.
Use the underlying MBEDTLS_INTERNAL_VALIDATE_RET() insteaed,
returning 0 on failure.
Also, add a test for this behaviour.
Refactor `mpi_write_hlp()` to not be recursive, to fix stack overflows.
Iterate over the `mbedtls_mpi` division of the radix requested,
until it is zero. Each iteration, put the residue in the next LSB
of the output buffer. Fixes#2190
In mbedtls_mpi_write_binary, avoid leaking the size of the number
through timing or branches, if possible. More precisely, if the number
fits in the output buffer based on its allocated size, the new code's
trace doesn't depend on the value of the number.
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
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.
The specification requires that numbers are the raw entropy (except for odd/
even) and at least 2^(nbits-0.5). If not, new random bits need to be used for
the next number. Similarly, if the number is not prime new random bits need to
be used.
This commit modifies mpi_read_binary to always allocate the minimum number of
limbs required to hold the entire buffer provided to the function, regardless of
its content. Previously, leading zero bytes in the input data were detected and
used to reduce memory footprint and time, but this non-constant behavior turned
out to be non-tolerable for the cryptographic applications this function is used
for.
When provided with an empty line, mpi_read_file causes a numeric
underflow resulting in a stack underflow. This commit fixes this and
adds some documentation to mpi_read_file.
The modular inversion function hangs when provided with the modulus 1. This commit refuses this modulus with a BAD_INPUT error code. It also adds a test for this case.
Fix a buffer overflow when writting a string representation of an MPI
number to a buffer in hexadecimal. The problem occurs because hex
digits are written in pairs and this is not accounted for in the
calculation of the required buffer size when the number of digits is
odd.
The function appears to be safe, since grow() is called with sensible
arguments in previous functions. Ideally Clang would be clever enough to
realise this. Even if N has size MBEDTLS_MPI_MAX_LIMBS, which will
cause the grow to fail, the affected lines in montmul won't be reached.
Having this sanity check can hardly hurt though.
* yanesca/iss309:
Improved on the previous fix and added a test case to cover both types of carries.
Removed recursion from fix#309.
Improved on the fix of #309 and extended the test to cover subroutines.
Tests and fix added for #309 (inplace mpi doubling).
Found by Guido Vranken.
Two possible integer overflows (during << 2 or addition in BITS_TO_LIMB())
could result in far too few memory to be allocated, then overflowing the
buffer in the subsequent for loop.
Both integer overflows happen when slen is close to or greater than
SIZE_T_MAX >> 2 (ie 2^30 on a 32 bit system).
Note: one could also avoid those overflows by changing BITS_TO_LIMB(s << 2) to
CHARS_TO_LIMB(s >> 1) but the solution implemented looks more robust with
respect to future code changes.
* commit 'ce60fbe':
Fix potential timing difference with RSA PMS
Update Changelog for recent merge
Added more constant-time code and removed biases in the prime number generation routines.
Conflicts:
library/bignum.c
library/ssl_srv.c
Also fix bug/limitation in mpi_copy: would segfault if src just initialised
and not set to a value yet. (This case occurs when copying a context which
contains only the public part of the key, eg.)
