According to https://www.bearssl.org/ctmul.html even single-precision
multiplication is not constant-time on some older platforms.
An added benefit of the new code is that it removes the somewhat mysterious
constant 0x1ff - which was selected because at that point the maximum value of
padlen was 256. The new code is perhaps a bit more readable for that reason.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
The previous code used comparison operators >= and == that are quite likely to
be compiled to branches by some compilers on some architectures (with some
optimisation levels).
For example, take the following function:
void old_update( size_t data_len, size_t *padlen )
{
*padlen *= ( data_len >= *padlen + 1 );
}
With Clang 3.8, let's compile it for the Arm v6-M architecture:
% clang --target=arm-none-eabi -march=armv6-m -Os foo.c -S -o - |
sed -n '/^old_update:$/,/\.size/p'
old_update:
.fnstart
@ BB#0:
.save {r4, lr}
push {r4, lr}
ldr r2, [r1]
adds r4, r2, #1
movs r3, #0
cmp r4, r0
bls .LBB0_2
@ BB#1:
mov r2, r3
.LBB0_2:
str r2, [r1]
pop {r4, pc}
.Lfunc_end0:
.size old_update, .Lfunc_end0-old_update
We can see an unbalanced secret-dependant branch, resulting in a total
execution time depends on the value of the secret (here padlen) in a
straightforward way.
The new version, based on bit operations, doesn't have this issue:
new_update:
.fnstart
@ BB#0:
ldr r2, [r1]
subs r0, r0, #1
subs r0, r0, r2
asrs r0, r0, #31
bics r2, r0
str r2, [r1]
bx lr
.Lfunc_end1:
.size new_update, .Lfunc_end1-new_update
(As a bonus, it's smaller and uses less stack.)
While there's no formal guarantee that the version based on bit operations in
C won't be translated using branches by the compiler, experiments tend to show
that's the case [1], and it is commonly accepted knowledge in the practical
crypto community that if we want to sick to C, bit operations are the safest
bet [2].
[1] https://github.com/mpg/ct/blob/master/results
[2] https://github.com/veorq/cryptocoding
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Return a name that more clearly returns nonzero=true=good, 0=bad. We'd
normally expect check_xxx to return 0=pass, nonzero=fail so
check_parity was a bad name.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
An early draft of the PSA crypto specification required multipart
operations to keep working after destroying the key. This is no longer
the case: instead, now, operations are guaranteed to fail. Mbed TLS
does not comply yet, and still allows the operation to keep going.
Stop testing Mbed TLS's non-compliant behavior.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Rely on Asan to detect a potential buffer overflow, instead of doing a
manual check. This makes the code simpler and Asan can detect
underflows as well as overflows.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
In the cleanup code for persistent_key_load_key_from_storage(), we
only attempt to reopen the key so that it will be deleted if it exists
at that point. It's intentional that we do nothing if psa_open_key()
fails here.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
If any of the TEST_ASSERT()s that are before the call to
mbedtls_pk_warp_as_opaque() failed, when reaching the exit label
psa_destroy_key() would be called with an uninitialized argument.
Found by Clang.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
I might be wrong, but lcc's optimizer is curious about this,
and I am too: shouldn't we free allocated stuff correctly
before exiting `dh_genprime` in this certain point of code?
Signed-off-by: makise-homura <akemi_homura@kurisa.ch>
Replace server2.crt with server2-sha256.crt which, as the name implies, is
just the SHA-256 version of the same certificate.
Replace server1.crt with cert_sha256.crt which, as the name doesn't imply, is
associated with the same key and just have a slightly different Subject Name,
which doesn't matter in this instance.
The other certificates used in this script (server5.crt and server6.crt) are
already signed with SHA-256.
This change is motivated by the fact that recent versions of GnuTLS (or older
versions with the Debian patches) reject SHA-1 in certificates by default, as
they should. There are options to still accept it (%VERIFY_ALLOW_BROKEN and
%VERIFY_ALLOW_SIGN_WITH_SHA1) but:
- they're not available in all versions that reject SHA-1-signed certs;
- moving to SHA-2 just seems cleaner anyway.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Recent GnuTLS packages on Ubuntu 16.04 have them disabled.
From /usr/share/doc/libgnutls30/changelog.Debian.gz:
gnutls28 (3.4.10-4ubuntu1.5) xenial-security; urgency=medium
* SECURITY UPDATE: Lucky-13 issues
[...]
- debian/patches/CVE-2018-1084x-4.patch: hmac-sha384 and sha256
ciphersuites were removed from defaults in lib/gnutls_priority.c,
tests/priorities.c.
Since we do want to test the ciphersuites, explicitly re-enable them in the
server's priority string. (This is a no-op with versions of GnuTLS where those
are already enabled by default.)
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Update copyright notices to newly added files since merge of original
PR #3546 "Update copyright notices to use Linux Foundation guidance".
Generated using the same script.
Signed-off-by: Dan Handley <dan.handley@arm.com>
