If lsof is not available, wait_server_start uses a fixed timeout,
which can trigger a race condition if the timeout turns out to be too
short. Emit a warning so that we know this is going on from the test
logs.
- Some of the CI machines don't have lsof installed yet, so rely on an sleeping
an arbitrary number of seconds while the server starts. We're seeing
occasional failures with the current delay because the CI machines are highly
loaded, which seems to indicate the current delay is not quite enough, but
hopefully not to far either, so double it.
- While at it, also double the watchdog delay: while I don't remember seeing
much failures due to client timeout, this change doesn't impact normal
running time of the script, so better err on the safe side.
These changes don't affect the test and should only affect the false positive
rate coming from the test framework in those scripts.
In wait_server_start, fork less. When lsof is present, call it on the
expected process. This saves a few percent of execution time on a
lightly loaded machine. Also, sleep for a short duration rather than
using a tight loop.
Previously, MAC validation for an incoming record proceeded as follows:
1) Make a copy of the MAC contained in the record;
2) Compute the expected MAC in place, overwriting the presented one;
3) Compare both.
This resulted in a record buffer overflow if truncated MAC was used, as in this
case the record buffer only reserved 10 bytes for the MAC, but the MAC
computation routine in 2) always wrote a full digest.
For specially crafted records, this could be used to perform a controlled write of
up to 6 bytes past the boundary of the heap buffer holding the record, thereby
corrupting the heap structures and potentially leading to a crash or remote code
execution.
This commit fixes this by making the following change:
1) Compute the expected MAC in a temporary buffer that has the size of the
underlying message digest.
2) Compare to this to the MAC contained in the record, potentially
restricting to the first 10 bytes if truncated HMAC is used.
A similar fix is applied to the encryption routine `ssl_encrypt_buf`.
Some tests in ssl-opt.sh require MBEDTLS_SSL_MAX_CONTENT_LEN to be set to its
default value of 16384 to succeed. While ideally such a dependency should not
exist, as a short-term remedy this commit adds a small check that will at least
lead to graceful exit if that assumption is violated.
This commit adds four tests to ssl-opt.sh testing the library's behavior when
`mbedtls_ssl_write` is called with messages beyond 16384 bytes. The combinations
tested are TLS vs. DTLS and MBEDTLS_SSL_MAX_FRAGMENT_LENGTH enabled vs. disabled.
This commit adds regression tests for the bug when we didn't parse the
Signature Algorithm extension when renegotiating. (By nature, this bug
affected only the server)
The tests check for the fallback hash (SHA1) in the server log to detect
that the Signature Algorithm extension hasn't been parsed at least in
one of the handshakes.
A more direct way of testing is not possible with the current test
framework, since the Signature Algorithm extension is parsed in the
first handshake and any corresponding debug message is present in the
logs.
The check uses grep, not config.pl, on the x509 headers - not where it should
be configured - config.h. grep syntax isn't very portable. Without config.pl
it's quite hard to do this check properly so removing this check.
Some tests in ssl-opt.sh assumes the value 8 for the maximal number
MBEDTLS_X509_MAX_INTERMEDIATE_CA of intermediate CA's. This commit adds a check
before conducting the respective tests.
This commit adds four tests to tests/ssl-opt.sh:
(1) & (2): Check behaviour of optional/required verification when the
trusted CA chain is empty.
(3) & (4): Check behaviour of optional/required verification when the
client receives a server certificate with an unsupported curve.
By default, keep allowing SHA-1 in key exchange signatures. Disabling
it causes compatibility issues, especially with clients that use
TLS1.2 but don't send the signature_algorithms extension.
SHA-1 is forbidden in certificates by default, since it's vulnerable
to offline collision-based attacks.
Fixed a bug in ssl_srv.c when parsing TLS_FALLBACK_SCSV in the
ciphersuite list that caused it to miss it sometimes. Reported by Hugo
Leisink as issue #810. Fix initially by @andreasag01; this commit
isolates the bug fix and adds a non-regression test.
Add a test to ssl-opt.sh to ensure that in DTLS a 6 byte record counter
is compared in ssl_check_ctr_renegotiate() instead of a 8 byte one as in
the TLS case. Because currently there are no testing facilities to check
that renegotiation routines are triggered after X number of input/output
messages, the test consists on setting a renegotiation period that
cannot be represented in 6 bytes, but whose least-significant byte is 2.
If the library behaves correctly, the renegotiation routines will be
executed after two exchanged.
In a USENIX WOOT '16 paper the authors exploit implementation
mistakes that cause Initialisation Vectors (IV) to repeat. This
did not happen in mbed TLS, and this test makes sure that this
won't happen in the future either.
A new test option is introduced to ssl-opt.sh that checks the server
and client logs for a pattern and fails in case there are any
duplicates in the lines following the matching ones. (This is
necessary because of the structure of the logging)
Added a test case as well to utilise the new option. This test forces
the TLS-ECDHE-ECDSA-WITH-AES-256-GCM-SHA384 ciphersuite to make the
client and the server use an AEAD cipher.
Hanno Böck, Aaron Zauner, Sean Devlin, Juraj Somorovsky and Philipp
Jovanovic, "Nonce-Disrespecting Adversaries: Practical Forgery Attacks
on GCM in TLS", USENIX WOOT '16
Running valgrind on: "DTLS client reconnect from same port: reconnect,
nbio" results in timeouts.
New version added that runs only under valgrind. Original only runs when
valgrind is not used
Let the client retry longer, to make sure the server will time out before the
client gives up. Make it really longer to get a deterministic client exit
status (make sure it has time to reconnect after the server timeout).
This is not very useful for TLS as mbedtls_ssl_write() will automatically
fragment and return the length used, and the application should check for that
anyway, but this is useful for DTLS where mbedtls_ssl_write() returns an
error, and the application needs to be able to query the maximum length
instead of just guessing.
Tends to cause spurious failures on buildbots due to peer timing out.
Anyway, those tests are mainly for interop, any memory error is most likely
catched by some earlier self-op test. (Also, we'll run these tests with ASan
anyway.)
Apparently openssl s_server does not flush stdout, anyway sometimes the client
receives the reply and exits, thus terminating the test, before is request is
visible on the server's stdout. So, just don't check that, checking the
client's output and exit code is already enough.
Retry one time in case we have a client timeout. These should be fairly rare
but still happen from time to time with udp_proxy tests which is annoying, and
until now has never indicated an actual issue.
Rather than flat-out die when we can't see the server started with lsof, just
stop waiting and try to go ahead with the test. Maybe it'll work if there was
a problem with lsof, most probably it will fail, but at least we'll have the
log, and the results of the following tests.
Note: date +%s isn't POSIX, but it works at least on Linux, Darwin/FreeBSD and
OpenBSD, which should be good enough for a test script.
This is not required nor recommended by the protocol, and it's a layering
violation, but it's a know flaw in the protocol that you can't detect a PSK
auth error in any other way, so it is probably the right thing to do.
closes#227