With --enable-dtlscid, a client sending a Client Hello to a DLTS1.2
server that supports CID, the server provides the appropriate CID and
assumes that CID has been negotiated.
However, in the case of MbedTLS, it then rejects packets that do not
match its expected CID from the client - as wolfSSL no longer sends
the CID as it is not DTLS1.2.
https://datatracker.ietf.org/doc/html/rfc9147#section-4
If a Connection ID is negotiated, then it MUST be contained in all datagrams.
This fix drops the CID if a Hello Verify Request is received, so the
second Client Hello does not include the CID.
https://datatracker.ietf.org/doc/html/rfc6347#section-4.2.1
When responding to a HelloVerifyRequest, the client MUST use the same
parameter values (version, random, session_id, cipher_suites,
compression_method) as it did in the original ClientHello.
Dropping the CID extension does not violate this.
If the server sends Server Hello, Server Key Exchange and Server Hello Done
in a single DTLS packet, but for DTLS1.2 in response to a client DTLS1.3
request, then FIRST_REPLAY state does not occur until the server re-sends
the packet. At this point wolfSSLconnect() gets used and all bursts into
life.
When processing handshakes in wolfSSL_connect_TLSv13() for case
HELLO_AGAIN_REPLY, downgrade to using wolfSSLconnect() to continue
processing the remaining handshakes in the packet.
Found when using Mbed TLS for the server.
ASN macro simplification
merged with github CI tests failing due to unrelated upstream changes (same tests all previously succeeded on this PR, with only 25d14f1937 added in the meantime).
supplementary testing with `wolfssl-multi-test.sh ... super-quick-check` after rebase on then-current `master` 15e99c8eff.
ChaCha20:
scalar and vector implementations
vector implementations doing 6, 4, 2, 1 block at a time.
scalar implemetations using roriw and pack
vector implementations using VROR_VI and roriw.
RISC-V SHA-256: avoid using s0 if it can be helped.
In the wc_GeneratePreTBS() method (used for WOLFSSL_DUAL_ALG_CERTS
support), there was a workaround for alt names in certificates, as the
CopyDecodedToX509() method wasn't properly copying them. As a proper
copy mechanism is implemented now, we have to remove the workaround as
it now causes a memory leak of the copied values.
Signed-off-by: Tobias Frauenschläger
<tobias.frauenschlaeger@oth-regensburg.de>
Added tests to cover ssl_load.c functions.
Fixes from testing.
pk.c: renamed wolfssl_dh_load_key to wolfssl_dh_load_params as it
doesn't handle keys - just parameters.
* refactor WOLFSSL_BIO.num and WOLFSSL_BIO.ptr as unions, for clarity and bug resistance (no functional changes).
* in wolfSSL_BIO_free(), add WOLFSSL_BIO_DGRAM to the test for closing bio->num.fd, fixing a descriptor leak.
* use SOCKET_INVALID consistently as the invalid value for WOLFSSL_BIO.num.fd, and use SOCKET_T consistently as the internal type for file descriptors.
* move the definitions for SOCKET_T and SOCKET_INVALID from wolfio.h to the filesystem section of wc_port.h, and allow override definitions of SOCKET_T.
detected and tested with wolfssl-multi-test.sh ... pq-hybrid-all-rpk-valgrind-unittest. also tested with wolfssl-multi-test.sh ... super-quick-check.
