Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/27264)
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/27091)
It is no longer static.
Also add it to libssl only with quic enabled.
Reviewed-by: Paul Dale <ppzgs1@gmail.com>
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26882)
We've not implemented it yet, and don't need it for MVP, so move the
TODO's to QUIC FUTURE and remove the docs for it.
Fixesopenssl/project#1074
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26593)
I cloned a copy of fnv1a_hash from hashtable.c. Deduplicate that so we
have common source code.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
@sashan and I were discussing the usefulness of the public facing api
for NEW_TOKEN support, and he has concerns over its usefulness and our
being stuck with it if we need to make changes later. Given that it is
a convience api for using multiple CTX-es to share a cache, its fine if
we remove it for now, as that seems like a less common use case.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
Replace it with SSL_TOKEN_STORE and make the structure opaque in the
public api
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
we use this struct internally to track computed tokens, we may as well
use it when fetching those tokens, as it allows the removeal of the QTOK
type
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
we use get0 to get a token store, but set to set it. Since the latter
takes a refcount, change that to set1. Also rename the interal quic
functions to match.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
Forgot to free the CRYPTO_REF when freeing a token
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
closer reading of RFC 9000 indicates that a NEW_TOKEN token can be
(re)used repeatedly.
so instead of creating a use once and discard pattern in the token api.
Let the tokens stick around until they are replaced with a new token
from the server. To do this, we need to ref count the tokens so that we
don't accidentally free them while a given client is waiting to send an
initial frame making use of them.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
This will make it easier to refcount them in a moment
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
We don't want to schedule the NEW_TOKEN frame until such time as the
handshake is complete, otherwise we risk giving a token to validate a
future connection to a peer we haven't decided to trust yet
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
Instead of copying the token thats store, return a pointer to it
along with a pointer to the token struct to free should we need to
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
the SSL_new_from_listener api creates a client SSL from a server
SSL_CTX context. Normally server contexts need no token cache, but once
we start using it as a client, that changes. Allocate one here when
needed
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
Start storing new tokens in our new cache
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26517)
These are either already implemented or not relevant for
the QUIC server MVP.
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26544)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
Not strictly needed
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
As per @sashan suggestion, try pre-creating user ssls with a NULL
listener
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
We have a chicken and egg problem.
Normally when we create a connection object in quic, we associate it
with a listener, and up the ref on the parent listener, which is fine.
However, now that we are pre-allocating user_ssl objects for incomming
connections we have a situation in which:
1) The pre-alocated connection object holds a ref on the listener
2) The application has no awareness of the quic connection object (and
so can't free it)
3) The freeing of the listener object never calls into the quic stack,
because its reference count may hold references from connections that
haven't been accepted yet
We could require that applications register a function for the
new_pending_conn callback, and track/free these pending connections, but
that seems like alot of extra unneeded work to place on the application
Instead:
a) add a quic_conn_st flag named accepted
b) When pre-allocating connections, clear the flag in (a) and _dont_
hold a reference to the parent listener
c) in SSL_accept_connection, set the accepted flag and reference the
listener
d) in ossl_quic_free drop the listener reference only if the accepted
flag is set
c) expressly free all user_ssl objects in ossl_quic_port_drop_incoming
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26361)
SSL_new_from_listner() creates QUIC connection object (QCSO)
from listener. Caller can use the object retuned from
SSL_new_from_listener() to connect to remote QUIC server.
The QCSO created here shares engine/port with listener.
the change is covered by `test_ssl_new_from_listener()` in
test/quicapitest.c
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26138)
I had experimented with starting the ssl handshake during accept, and
forgot to remove it
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26178)
Now that we have the infrastructure to skip address validation, add a
public flag to SSL_new_listener and SSL_new_listener_from to allow the
skipping of address validation on selected quic listener SSL objects
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/26114)
This got introduced by #595288251bb (QUIC APL: Ensure APL
functions use correct prologue)
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25659)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25416)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25416)
The ossl_quic_get_net_write_desired() and
ossl_quic_reactor_net_read_desired() implementations can be used by
listeners. But in that case there is no ctx.qc object present. Instead we
should use the reactor from ctx.obj which will work also for a listener.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25642)
Just because one connection has not started yet, it does not mean that
we should not tick the QUIC_ENGINE. There may be other connections that do
need ticking.
Reviewed-by: Saša Nedvědický <sashan@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25452)
The comment is no longer accurance so it can be removed
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25457)
Centralise the storage of the override in the QUIC_ENGINE rather than in
the QUIC_CONNECTION. We can now set the override on any type of QUIC SSL
object as needed.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25457)
Now that we also QUIC server (listener) handles, we may have a NULL quic
connection (ctx.qc), and so need to either return early or handle the
NULL `qc` gracefully.
Reviewed-by: Sasa Nedvedicky <sashan@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25432)
Samson S. Kolge
Matt Caswell
openssl-machine
Andrew Dinh
Tomas Mraz
Neil Horman
Alexandr Nedvedicky
sashan
Hugo Landau
Viktor Dukhovni