Section 4.2.8 of RFC 6347 describes how to handle the case of a DTLS client
establishing a new connection using the same UDP quartet as an already active
connection, which we implement under the compile option
MBEDTLS_SSL_DLTS_CLIENT_PORT_REUSE. Relevant excerpts:
[the server] MUST NOT destroy the existing
association until the client has demonstrated reachability either by
completing a cookie exchange or by completing a complete handshake
including delivering a verifiable Finished message.
[...]
The reachability requirement prevents
off-path/blind attackers from destroying associations merely by
sending forged ClientHellos.
Our code chooses to use a cookie exchange for establishing reachability, but
unfortunately that check was effectively removed in a recent refactoring,
which changed what value ssl_handle_possible_reconnect() needs to return in
order for ssl_get_next_record() (introduced in that refactoring) to take the
proper action. Unfortunately, in addition to changing the value, the
refactoring also changed a return statement to an assignment to the ret
variable, causing the function to reach the code for a valid cookie, which
immediately destroys the existing association, effectively bypassing the
cookie verification.
This commit fixes that by immediately returning after sending a
HelloVerifyRequest when a ClientHello without a valid cookie is found. It also
updates the description of the function to reflect the new return value
convention (the refactoring updated the code but not the documentation).
The commit that changed the return value convention (and introduced the bug)
is 2fddd3765e, whose commit message explains the
change.
Note: this bug also indirectly caused the ssl-opt.sh test case "DTLS client
reconnect from same port: reconnect" to occasionally fail due to a race
condition between the reception of the ClientHello carrying a valid cookie and
the closure of the connection by the server after noticing the ClientHello
didn't carry a valid cookie after it incorrectly destroyed the previous
connection, that could cause that ClientHello to be invisible to the server
(if that message reaches the server just before it does `net_close()`). A
welcome side effect of this commit is to remove that race condition, as the
new connection will immediately start with a ClientHello carrying a valid
cookie in the SSL input buffer, so the server will not call `net_close()` and
not risk discarding a better ClientHello that arrived in the meantime.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Conflicts:
mbedtls.doxyfile - PROJECT_NAME - mbed TLS v2.16.6 chosen.
doc_mainpage.h - mbed TLS v2.16.6 version chosen.
hmac_drbg.h - line 260, extended description chosen.
- line 313, extended description chosen.
- line 338, extended description chosen.
version.h - 2.16.6 chosen.
CMakeLists.txt - 2.16.6 chosen.
test_suite_version.data - 2.16.6 chosen.
Makefile - 141 - manual correction - baremetal version of C_SOURCE_FILES
with variables for directories plus 2.16.6 CTAGS addition.
pkparse.c - lines 846 onwards - the asn1_get_nonzero_mpi implementation chosen.
ssl_tls.c - line 5269 - edited manually, left the ret=0, because baremetal has
a different behaviour since commit 87b5626, but added a debug
message that's new in 2.16.6.
all.sh:
- component_build_deprecated - chosen the refactored version from 2.16.6,
but with extra flags from baremetal.
- rest of the _no_xxx tests - merged make options to have PTHREAD=1 and
other changes from 2.16.6 (like -O1 instead of -O0).
- component_build_arm_none_eabi_gcc_no_64bit_multiplication - added
TINYCRYPT_BUILD=0 to the 2.16.6 version of make.
x509/req_app.c - left baremetal log but with mbedtls_exit( 0 ) call.
x509/crl_app.c - left baremetal log but with mbedtls_exit( 0 ) call.
x509/cert_app.c - left baremetal log but with mbedtls_exit( 0 ) call.
ssl/ssl_mail_client.c - left baremetal log but with mbedtls_exit( 0 ) call.
ssl/ssl_pthread_server.c - left baremetal log but with mbedtls_exit( 0 ) call.
ssl/ssl_fork_server.c - left baremetal log but with mbedtls_exit( 0 ) call.
ssl_client1.c - line 54 - left baremetal log but with mbedtls_exit( 0 ) call.
ssl_client2.c - line 54 - left baremetal log but with mbedtls_exit( 0 ) call.
- line 132 - new options of both branches added.
- skip close notify handled as in 2.16.6, but with `ssl` instead of `&ssl`.
- Merged the 2.16.6 usage split with additional baremetal usages.
- Merged options from baremetal and 2.16.6.
ssl_server.c - left baremetal log but with mbedtls_exit( 0 ) call.
ssl_server2.c - Merged the 2.16.6 usage split with additional baremetal usages.
config.pl - fixed missing defines from the documentation, removed duplicates,
and reorganised so that the documentation and excluded list
are ordered in the same way.
test_suite_x509parse.data - only added the two new pathlen tests.
x509_crt.c - change the return code by removing
MBEDTLS_ERR_X509_INVALID_EXTENSIONS, since it's added by
x509_crt_frame_parse_ext not by an "or", but by "+=".
Changelog - Assigned all entries to appropriate sections.
ssl-opt.sh - line 8263 - merged options.
- removed lines 1165 - 1176 - there was a duplicate test, probably
an artifact of previous merges.
check-files.py - sticked to old formatting.
Signed-off-by: Andrzej Kurek <andrzej.kurek@arm.com>
Goals:
* Build with common compilers with common options, so that we don't
miss a (potentially useful) warning only triggered with certain
build options.
* A previous commit removed -O0 test jobs, leaving only the one with
-m32. We have inline assembly that is disabled with -O0, falling
back to generic C code. This commit restores a test that runs the
generic C code on a 64-bit platform.
The splitting of this test into two versions depending on whether SHA-1 was
allowed by the server was a mistake in
5d2511c4d4 - the test has nothing to do with
SHA-1 in the first place, as the server doesn't request a certificate from
the client so it doesn't matter if the server accepts SHA-1 or not.
While the whole script makes (often implicit) assumptions about the version of
GnuTLS used, generally speaking it should work out of the box with the version
packaged on our reference testing platform, which is Ubuntu 16.04 so far.
With the update from Jan 8 2020 (3.4.10-4ubuntu1.6), the patches for rejecting
SHA-1 in certificate signatures were backported, so we should avoid presenting
SHA-1 signed certificates to a GnuTLS peer in ssl-opt.sh.
None of the test cases in tests_suite_memory_buffer_alloc actually
need MBEDTLS_MEMORY_DEBUG. Some have additional checks when
MBEDTLS_MEMORY_DEBUG but all are useful even without it. So enable
them all and #ifdef out the parts that require DEBUG.
The test case "Memory buffer small buffer" emits a message
"FATAL: verification of first header failed". In this test case, it's
actually expected, but it looks weird to see this message from a
passing test. Add a comment that states this explicitly, and modify
the test description to indicate that the failure is expected, and
change the test function name to be more accurate.
Fix#309
The corner case tests were designed for 32 and 64 bit limbs
independently and performed only on the target platform. On the other
platform they are not corner cases anymore, but we can still exercise
them.
The corner case tests were designed for 64 bit limbs and failed on 32
bit platforms because the numbers in the test ended up being stored in a
different number of limbs and the function (correctly) returnd an error
upon receiving them.
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.
mbedtls_ctr_drbg_seed() always set the entropy length to the default,
so a call to mbedtls_ctr_drbg_set_entropy_len() before seed() had no
effect. Change this to the more intuitive behavior that
set_entropy_len() sets the entropy length and seed() respects that and
only uses the default entropy length if there was no call to
set_entropy_len().
The former test-only function mbedtls_ctr_drbg_seed_entropy_len() is
no longer used, but keep it for strict ABI compatibility.
When running 'make test' with GNU make, if a test suite program
displays "PASSED", this was automatically counted as a pass. This
would in particular count as passing:
* A test suite with the substring "PASSED" in a test description.
* A test suite where all the test cases succeeded, but the final
cleanup failed, in particular if a sanitizer reported a memory leak.
Use the test executable's return status instead to determine whether
the test suite passed. It's always 0 on PASSED unless the executable's
cleanup code fails, and it's never 0 on any failure.
FixARMmbed/mbed-crypto#303
Some sanitizers default to displaying an error message and recovering.
This could result in a test being recorded as passing despite a
complaint from the sanitizer. Turn off sanitizer recovery to avoid
this risk.
Exercise the library functions with calloc returning NULL for a size
of 0. Make this a separate job with UBSan (and ASan) to detect
places where we try to dereference the result of calloc(0) or to do
things like
buf = calloc(size, 1);
if (buf == NULL && size != 0) return INSUFFICIENT_MEMORY;
memcpy(buf, source, size);
which has undefined behavior when buf is NULL at the memcpy call even
if size is 0.
This is needed because other test components jobs either use the system
malloc which returns non-NULL on Linux and FreeBSD, or the
memory_buffer_alloc malloc which returns NULL but does not give as
useful feedback with ASan (because the whole heap is a single C
object).
- Add configuration for AES_SCA_COUNTERMEASURES to config.h. By
default the feature is disabled.
- Add AES_SCA_COUNTERMEASURES configuration check to check_config.h
- Add AES_SCA_COUNTERMEASURES test to all.sh
This commit first changes the return convention of EccPoint_mult_safer() so
that it properly reports when faults are detected. Then all functions that
call it need to be changed to (1) follow the same return convention and (2)
properly propagate UECC_FAULT_DETECTED when it occurs.
Here's the reverse call graph from EccPoint_mult_safer() to the rest of the
library (where return values are translated to the MBEDTLS_ERR_ space) and test
functions (where expected return values are asserted explicitly).
EccPoint_mult_safer()
EccPoint_compute_public_key()
uECC_compute_public_key()
pkparse.c
tests/suites/test_suite_pkparse.function
uECC_make_key_with_d()
uECC_make_key()
ssl_cli.c
ssl_srv.c
tests/suites/test_suite_pk.function
tests/suites/test_suite_tinycrypt.function
uECC_shared_secret()
ssl_tls.c
tests/suites/test_suite_tinycrypt.function
uECC_sign_with_k()
uECC_sign()
pk.c
tests/suites/test_suite_tinycrypt.function
Note: in uECC_sign_with_k() a test for uECC_vli_isZero(p) is suppressed
because it is redundant with a more thorough test (point validity) done at the
end of EccPoint_mult_safer(). This redundancy was introduced in a previous
commit but not noticed earlier.
Added an additional Makefile option of 'TINYCRYPT_BUILD' to exclude the
TinyCrypt source files from the build. This allows some tests to exclude those
files as and when necessary.
Specifically this includes in all.sh the test
'component_build_arm_none_eabi_gcc_no_64bit_multiplication' which was failing as
64bit cannot be disabled in TinyCrypt, and check-names.sh as TinyCrypt obviously
does not conform to Mbed TLS naming conventions.
This is a first step in protecting against fault injection attacks: the
attacker can no longer change failure into success by flipping a single bit.
Additional steps are needed to prevent other attacks (instruction skip etc)
and will be the object of future commits.
The return value of uECC_vli_equal() should be protected as well, which will
be done in a future commit as well.
