We actively use only the latest DCID received. And retire only
DCIDs requested by the peer to be retired.
Also changed the active_conn_id_limit to 2 as the minimum value allowed.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20892)
This allows QUIC_STREAM objects to be deleted when they are no longer
needed.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
We need to get acknowledgement notifications for our STOP_SENDING and
STREAM_RESET frames as this information is needed to know when we can
delete a QUIC_STREAM object.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
Also use accept queue popping by the application as the retirement
event, i.e., as the cue to increase the limit.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
Though the RXFC was designed for stream flow control, its logic
is generic enough to use to control MAX_STREAMS generation.
Control of when _we_ can open streams is already done in a bespoke
fashion and doesn't use a TXFC, however (see
ossl_quic_stream_map_update_state).
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
QUIC in single-stream mode could be used with a protocol where the
server writes first or the client writes first. This determines
whether the single stream would be client or server initiated,
which affects the stream ID allocated to the stream. We should support
both client-sends-first and server-sends-first application protocols.
Thus, defer default XSO creation until the point in time at which
we know whether a client-first or server-first application protocol
is being used. We do this by taking whether SSL_read() or SSL_write()
is called first as a cue.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
We switch to instantiating the QUIC_CHANNEL up front at QCSO
instantiation time. This creates the QUIC_STREAM_MAP early and makes it
easy for us to allocate streams prior to connection initiation. The role
(client or server) is determined at QCSO allocation time and cannot be
changed.
SSL_set_connect/accept_state() are still modelled but their usage must
be consistent with the chosen SSL_METHOD which dictates which role is
being used.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
In QUIC, we have an architectural need (in future, when we implement
0-RTT, etc.) to be able to create streams before we start connecting.
This requires we allocate a stream, including a stream ID, after
creating a QCSO but prior to connecting. However stream IDs are
dependent on whether the endpoint is in the client or server role,
therefore we must know whether we are going to be a client or server
before any pre-connection streams are created. Moreover, the originally
defined QUIC_client_method() and QUIC_server_method() functions heavily
implied the original plan was to have different SSL_METHODs for clients
and servers. Up until now we had been relying on
SSL_set_connect/accept_state() instead.
Solve these problems by basing client/server identity on whether
QUIC_server_method() is used (in future, when we support servers). This
ensures that once a QCSO is created its client/server identity are fixed
and cannot change, allowing pre-connection stream IDs, etc. to be
allocated.
Client/server uncertainty was the primary reason why QUIC_CHANNEL
creation was deferred until connection time up until now, so this
enables further refactoring to facilitate eager allocation of the
QUIC_CHANNEL at QCSO allocation time. This is important as allocating a
stream including its write buffers is hard without having the
QUIC_CHANNEL (which owns the QUIC_STREAM_MAP) in existence.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
This amends the dispatch code from ssl_lib.c to the QUIC API Personality
Layer to use the new approach of dispatching using SSL object pointers
rather than raw QUIC_CONNECTION pointers. This completes the said
refactor.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
We now refactor the interface between ssl_lib.c frontend functions and
the QUIC API Personality Layer so that the respective functions
comprising the interface use SSL object pointers rather than raw
QUIC_CONNECTION pointers. This is in preparation for stream support
since once streams are supported, calls to e.g. ossl_quic_write() may be
made on a QUIC_CONNECTION or a QUIC_XSO (that is, a stream object). Thus
we take a uniform approach across all functions comprising the interface
between the ssl_lib.c frontend and the QUIC API Personality Layer of
using SSL pointers always. This provides a uniform approach and
ensures that any function of the API personality layer can be easily
adapted to support being called on a stream object in the future.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
The QUIC_XSO (external stream object) is to a QUIC stream what a
QUIC_CONNECTION is to a QUIC connection. Both are SSL objects. The
QUIC_CONNECTION type is the internal representation of a QUIC connection
SSL object (QCSO) and the QUIC_XSO type is the internal representation
of a QUIC stream SSL object (QSSO) type. The name QUIC_XSO has been
chosen to be distinct from the existing QUIC_STREAM type which is our
existing internal stream type. QUIC_XSO is to a QUIC_STREAM what
QUIC_CONNECTION is to a QUIC_CHANNEL; in other words, QUIC_CONNECTION
and QUIC_XSO objects form part of the API personality layer, whereas
QUIC_CHANNEL and QUIC_STREAM objects form part of the QUIC core and are
distinct from the API personality layer.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
Uniform changes to all dispatch functions to use the new dispatch
functionality follows this commit. Separated into a core commit
and a commit containing the uniform pattern (monotonous) changes
for ease of review.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20765)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20848)
CLA: trivial
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20930)
We accepted more version numbers as valid DTLS then we really should do.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20830)
There is no point in sending that when min_proto_version is >= TLS1_3_VERSION.
So we set that during SSL_CTX initialization and skip adding the SCSV.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
(Merged from https://github.com/openssl/openssl/pull/20764)
If we are processing a hanshake fragment and we end up with a
zero length record, then we still need to release it to avoid an
infinite loop.
Fixes#20821
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20824)
We make these APIs work more like the TLS versions do.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20827)
We initialise the record layer rstate variable to ensure the
SSL_rstate_string*() APIs return values that are consistent with
previous versions.
Fixes#20808
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20827)
Instead of using stat() to check if a file is a directory
we just skip . and .. as a workaround.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/20786)
- factorize BIO_CTRL_DGRAM_SET_NEXT_TIMEOUT calls.
- simplify a return type
- style nits
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20602)
Fixes#20710
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20745)
CLA: trivial
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Tom Cosgrove <tom.cosgrove@arm.com>
(Merged from https://github.com/openssl/openssl/pull/20703)
This enables the cleansing of plaintext to occur in the record layer and
avoids the need to cast away const above the record layer.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20404)
Improves consistency with the QUIC rstream implementation - and improves
the abstraction between the TLS implementation and the abstract record
layer. We should not expect that the TLS implementation should be able to
change the underlying buffer. Future record layers may not expect that.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20404)
The QUIC TLS layer was taking an internal copy of rstream data while
reading. The QUIC rstream code has recently been extended to enable a
get/release model which avoids the need for this internal copy, so we use
that instead.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20404)
We have the existing functions SSL_version(), SSL_get_version() and
SSL_is_dtls(). We extend the first two to return something sensible when
using QUIC. We additionally provide the new functions SSL_is_tls() and
SSL_is_quic() to provide a mechanism to figure out what protocol we are
using.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20650)
It is possible that DTLS records are received out of order such that
records from the next epoch arrive before we have finished processing the
current epoch. We are supposed to buffer such records but for some reason
we only did that for handshake and alert records. This is incorrect since
it is perfectly possible for app data records to arrive early too.
Fixes#20597
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20628)
Add support for the RFC7250 certificate-type extensions.
Alows the use of only private keys for connection (i.e. certs not needed).
Add APIs
Add unit tests
Add documentation
Add s_client/s_server support
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18185)
The documentation didn't mention the development where EVP_PKEY_get_id()
returns a negative value for provider-only implementations, and the
migration guide didn't mention how to cope with that.
Fixes#20497
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20501)
calling SSL_CTX_set1_groups_list() twice on one SSL_CTX* caused a memory
leak visible in valgrind:
4 bytes in 1 blocks are definitely lost in loss record 1 of 1
at 0x4841888: malloc (vg_replace_malloc.c:381)
by 0x4B1EE96: CRYPTO_memdup (in libcrypto.so.3)
by 0x48993A0: tls1_set_groups_list (in libssl.so.3)
by 0x487AA7E: ssl3_ctx_ctrl (in libssl.so.3)
by 0x1091EA: main (mem_leak.c:10)
LEAK SUMMARY:
definitely lost: 4 bytes in 1 blocks
Freeing *pext to fix it.
CLA: trivial
Signed-off-by: Peter Kaestle <peter.kaestle@nokia.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20317)
(cherry picked from commit fcf3a9f7c6)
- Set `CERTIFICATE_VERIFY_MAX_LENGTH` to 65539
(2 bytes for the algorithm identifier + 2 bytes of signature length
+ 65535 bytes of signature)
- Changed `SSL3_RT_MAX_PLAIN_LENGTH` to `CERTIFICATE_VERIFY_MAX_LENGTH`
in `statem_srvr.c` and `statem_clnt.c`
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20486)
An incorrect macro name was being used for Windows detection which meant
we were going down a codepath not intended for Windows and thus failing.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20514)
We create "real" sockets for blocking mode so that we can block on them.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20514)
If a libctx has been associated with the channel, it must be passed down
to the QRX and QTX.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20514)
For example if would be helpful if we got more useful information if the
caller forgot to set the peer address.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20514)
If the QUIC oject receives a ctrl it does not recognise, the chances are
it is intended for the underlying SSL object - so we forward unknown ctrls
there.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20514)
Reported by Marc Schönefeld.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19794)
Add API calls to avoid copying data when reading
These are ossl_quic_rstream_get_record() and
ossl_quic_rstream_release_record().
Add side storage for the stream frame data.
When there are too many packets referenced by the
receiving stream the function ossl_quic_rstream_move_to_rbuf()
can be called to move the data to a ring buffer.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19794)
Early data decryption is expected to fail sometimes. If it does we should
not leave spurious error entries on the queue.
Fixes#20377
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20401)
The ticket_age/age_add values use ms granualarity. We were incorrectly
treating them as seconds and so the ticket was always being rejected for
early data. We also clarify a comment which could have been the source of
the confusion.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20387)
QUIC TLS was sending some ciphersuites twice in the ClientHello. This
was due to us declaring some TLSv1.3 ciphersuites in the list intended to
describe the TLSv1.2 ciphersuites supported by the SSL_METHOD.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20148)
It causes problems with some compilation options
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
If we complete the TLS handshake but transport params were not received
then this is a protcol error and we should fail.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
Check that we fail if the server has failed to provide transport params.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
We add callbacks so that TLS handshake messages can be modified by the test
framework before they are passed to the handshake hash, possibly encrypted
and written to the network. This enables us to simulate badly behaving
endpoints.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
We enable querying of the termination reason which is useful for tests.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
From RFC9000, section 19.21 "An extension to QUIC that wishes to use a new
type of frame MUST first ensure that a peer is able to understand the
frame". So if we receive an unknown frame type from a peer we should treat
it as a protocol violation. In fact we ignore it, and ignore all the
contents of the rest of the packet and continue on regardless.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
Also includes helper support to create a QUIC connection inside a test.
We wil use quicfaultstest to deliberately inject faulty datagrams/packets
to test how we handle them.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
We add callbacks so that QUIC packets can be modified by the test
framework before they are encrypted and written to the network. This
enables us to simulate badly behaving endpoints.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20030)
The two places that call `ossl_ssl_init()` assume that no additional
memory has been allocated when this fails; they subsequently free
the QUIC_CONNECTION/SSL_CONNECTION via OPENSSL_free() without freeing
any other resources.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20316)
Avoid including QUIC related stuff in the FIPS sources.
Also avoid including libssl headers in ssl3_cbc.c.
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19658)
So that CVE-2021-3449 can be found through fuzzing
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/20128)
The user_ssl field in an SSL_CONNECTION is no longer used - so remove it.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
Now that we have a real TLS handshake we no longer need the dummy handshake
implementation and it can be removed.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
Having support for the msg_callback will improve debug capabilities.
For record headers we "manufacture" dummy ones so that as far as the
callback is concerned we are doing "normal" TLS.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
We start using the QUIC TLS implementation rather than the dummy one.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
Previously we could only do this at the SSL_CTX level. We add the ability
to also do this on an SSL - but only for internal code.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
Add the ability to pass the main secret and length, as well as the
digest used for the KDF.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
This is just an internal API for now. Something like this will be made
public API at some point - but it is likely to be based on the provider
interface rather that a direct setting of a METHOD like we do for now.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
The reset() function was never called so it can be removed.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
The "user" SSL object which represents the QUIC connection should have an
"inner" SSL object to represent the TLS connection.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19748)
Some ciphers/protocol versions have an explicit IV. We need to make sure we
have sufficient room for it in the underlying buffer.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20085)
CONNECTION_CLOSE frames can be generated on multiple ELs, so the TX
packetiser was generating it on multiple ELs simultaneously. This fixes
the CONNECTION_CLOSE generation logic so that the lowest non-dropped EL
is always used.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19734)
