NPN is a little odd, owing to being a three-step process. The client
offers NPN, then the server accepts NPN and offers a list of protocols,
then the client picks a protocol.
The server is permitted to accept NPN but then offer zero supported
protocols. This worked, but was not tested or clearly documented.
In the last step, the client *must* pick a protocol, but it is permitted
to pick the empty string. The semantics of this are not explicitly
stated in the draft, but one can infer it means we aren't picking a
protocol. This also worked but was not tested or clearly documented.
Change-Id: I26d7089f4902834ff68a97467fc826e957d5fdf8
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/69027
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
Update-Note: By default, BoringSSL will now disable (D)TLS 1.0 and 1.1,
aligning with RFC 8996 and the behavior of modern browsers. Callers
which leave the minimum version at the default will now change to a
higher minimum version.
Callers which use SSL_CTX_set_proto_version and SSL_set_proto_version
are not affected. If this breaks something and the application needed
the legacy versions after all, call one of those functions with
TLS1_VERSION (or DTLS1_VERSION for DTLS) to restore the old behavior.
Bug: 599
Change-Id: Ic6e2d7a2ac7c190e81bd715b736c20e942f580b5
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/68487
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
This adds a notion of "credentials" to BoringSSL's API, to support
certificate selection by key type (typically ECDSA vs RSA), though the
aim is for it to be generalizable to other certificate types and other
kinds of selection criteria, such as Trust Expressions, or Merkle Tree
Certificates. Since we already had some nascent delegated credentials
I've reworked that feature with SSL_CREDENTIALs as well.
The model is that you create an SSL_CREDENTIAL object containing all the
configuration for what you are authenticating as. An X.509
SSL_CREDENTIAL has a certificate chain, private key, optionally an OCSP
response and SCT list. Delegated credentials are similar. In the future,
we might use this for raw public keys, other certificate types, etc.
Once you set those up, you configure those on the SSL or SSL_CTX in
preference order, and BoringSSL will internally pick the first one that
is usable.
The current implementation ends up redundantly selecting the signature
algorithm a couple of times. This works but is a little goofy. A
follow-up change will remove this redundancy. The protocol between the
runner and shim for tests is also a little weird, but it was the easiest
way I could think of for injecting that. Long-term, I think we should
just replace that protocol with a JSON structure. (See
https://crbug.com/boringssl/704.)
As split handshakes are in the process of being replaced with handshake
hints, this won't work with split handshakes. It works with handshake
hints without any extra work.
Update-Note: The delegated credentials API has been revamped.
Previously, it worked by configuring an optional delegated credential
and key with your normal certificate chain. This has the side effect of
forcing your DC issuer and your fallback certificate to be the same. The
SSL_CREDENTIAL API lifts this restriction.
A delegated credential is now just a different kind of credential. It
may use the same certificate chain as an X.509 credential or be
completely separate. All the SSL_CREDENTIAL APIs take CRYPTO_BUFFERs,
so, if common, the buffers may be shared to reduce memory.
The SSL_delegated_credential_used API is also removed, in favor of the
more general SSL_get0_selected_credential API. Callers can use ex_data
or pointer equality to identify the credential.
Bug: 249
Change-Id: Ibc290df3b7b95f148df12625e41cf55c50566602
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/66690
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
In TLS 1.2 and below, the supported_curves list simultaneously contrains
ECDH and ECDSA. Since BoringSSL, previously, did not handle ECDSA
certificate selection in the library, we ignored the latter and left it
to the callers. If configured with an ECDSA certificate that didn't
match the peer's curve list, we proceeded anyway, and left it to the
client to reject the connection.
This contradicts RFC 8422, which says:
The server constructs an appropriate certificate chain and conveys it
to the client in the Certificate message. If the client has used a
Supported Elliptic Curves Extension, the public key in the server's
certificate MUST respect the client's choice of elliptic curves. A
server that cannot satisfy this requirement MUST NOT choose an ECC
cipher suite in its ServerHello message.)
As with the previous client certificate change, once we move certificate
selection into the library, we'll need to evaluate this ourselves. A
natural implementation of it will, as a side effect, cause us to enforce
this match, even when only a single certificate is configured. This CL
lands that behavior change ahead of time and, in case there are
compatibility impats, leaves a flag, SSL_set_check_ecdsa_curve, to
restore the old behavior. If the change goes through fine, we can retire
the flag after a few months.
If this does cause a problem, we can opt to turn it off for the default
certificate, or only enable it when multiple certificates are
configured, but these all result in some slightly suboptimal behavior,
so I think we should treat them as contingency plans.
To help debugging, I gave this a dedicated error, though doing so is a
little tricky because of the PSK fallback. (See the
CheckECDSACurve-PSK-TLS12 test.)
Update-Note: A TLS 1.2 (or below) server, using an ECDSA certificate,
connecting to a client which doesn't advertise its ECDSA curve will now
fail the connection slightly earlier, rather than sending the
certificate and waiting for the client to reject it. The connection
should fail either way, but now it will fail earlier with
SSL_R_WRONG_CURVE. If the client was buggy and did not correctly
advertise its own capabilities, this may cause a connection to fail
despite previously succeeding. We have included a temporary API,
SSL_set_check_ecdsa_curve, to disable this behavior in the event this
has any impact, but please contact the BoringSSL team if you need it,
as it will interfere with improvements down the line.
TLS 1.3 is not impacted by this change, neither are clients, or RSA
certificiates. Additionally, if your server was already looking at the
curve list before configuring an ECDSA certificate in TLS 1.2, this
will also have no impact.
Bug: 249
Change-Id: I2f1d4e2627641319556847cbbbcdddf347bbc8a9
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/66688
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
TLS <= 1.2 servers indicate supported client certificate key types with
a certificate_types field in the CertificateRequest. Historically, we've
just ignored this field, because we've always outsourced certificate
selection to the caller anyway. This meant that, if you configured an
RSA client certificate in response to a server that requested only ECDSA
certificates, we would happily send the certificate and leave it to the
server to decide if it was happy.
Strictly speaking, this was in violation of RFC 5246:
- The end-entity certificate provided by the client MUST contain a
key that is compatible with certificate_types. [...]
Although prior TLS versions didn't say anything useful about this either
way.
Once we move certificate selection into the library, we'll want to start
evaluating supported algorithms ourselves. A natural implementation of
it will, as a side effect, cause us to enforce this match, even when
only a single certificate is configured. Since this is unlikely to have
any real compatibility impact (every TLS server I've seen just hardcodes
this list), let's just try turning it on. On the off chance it does
break someone, I've left a flag, SSL_set_check_client_certificate_type,
for folks to turn this check off. The flag will most likely be
unnecessary, in which case we can retire it after a few months.
If this does cause a problem, we can opt to turn it off for the default
certificate, or only enable it when multiple certificates are
configured, or lean on the sigalgs list (doesn't work for 1.0/1.1), but
these all result in some slightly suboptimal behavior, so I think we
should treat them as contingency plans.
Update-Note: A TLS 1.2 (or below) client, using client certificates,
connecting to a TLS server which doesn't support its certificate type
will now fail the connection slightly earlier, rather than sending the
certificate and waiting for the server to reject it. The connection
should fail either way, but now it will fail earlier with
SSL_R_UNKNOWN_CERTIFICATE_TYPE. If the server was buggy and did not
correctly advertise its own capabilities (very very unlikely), this may
cause a connection to fail despite previously succeeding. We have
included a temporary API, SSL_set_check_client_certificate_type, to
disable this behavior in the unlikely event this has any impact, but
please contact the BoringSSL team if you need it, as it will interfere
with improvements down the line.
This change does not affect servers requesting client certificates, only
clients sending them.
Bug: 249
Change-Id: I159bc444c4ee79fbe5c476d4253b48d58d2538be
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/66687
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Trying to handle malloc failures here is a bit tedious. Just suppress
malloc failures because nothing useful can progress when we can't even
allocate ex_data.
Change-Id: Ieaf417bcf9285783a76097319782282ce74e4734
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/66648
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Rather than using the pre-generated certificates, generate them on the
fly. This allows TLS stacks for which certificate validation and
verification are coupled to work as expected. Certificates and keys are
written to temporary files which are then passed to the shim, and
cleaned up on exit. This requires reworking how testCase passes
certs/keys by adding a new field, sendCertificate, rather than manually
setting the -cert-file and -key-file flags. Incidentally the
rsaChainCertificate is removed, since it was essentially unused, and all
tests that used it also work with rsaCertificate. Finally, include a
single SAN ("test") in all certificates, which fixes some TLS stacks
which require this to operate (such as rustls, which currently
regenerates all the certificates currently in the tree to add a SAN).
Additionally, add a new flag, -trust-cert, which tells the the shim
which certificates it should trust. Shims for TLS stacks which
can completely decouple validation and verification of X509 certificates
(like BoringSSL) can ignore this flag, but for stacks where this
functionality is somewhat more intertwined (like Go), this allows the
shim to properly process the sent certificates.
Change-Id: Ic5c63e18fb2b852cc693aacb3b06cfe7993bc90c
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/62565
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Old version Chrome with the existing ALPS codepoint can potentially cause network error due to an arithmetic overflow bug in Chrome ALPS decoder (We already fixed the issues starting from M100 in Chrome).
This CL add a new codepoint for ALPS extension in a way that can be enabled on individual connections., To support multiple versions of Chrome, we need to support both codepoints in BoringSSL.
For details: https://docs.google.com/document/d/16pysbV_ym_qAau_DBYnrw2A4h5ve2212wfcoYASt52U
Change-Id: Iea7822e757d23009648febc8eaff1c91b0f06e18
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/61125
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
We've required C++14 for a while now. As we're mostly C with a little
C++, this is less helpful, but may as well avoid bare new where
possible.
Change-Id: Icf3386e3f3b6f2092bb0089ed874cc50985f1a40
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/61429
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
The shim is now passed a shim ID which allows us to dispatch.
Previously, we dispatched with a distinct TCP server port per test, but
this seems to exhaust local ports.
As with https://boringssl-review.googlesource.com/c/boringssl/+/60885,
this changes the runner/shim protocol, but not the shim/handshaker
protocol. See that CL for discussion.
Change-Id: Ia22d1bc0bcda1b1869bce4630138538c30f02668
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/60888
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
The runner currently tries to listen on IPv6 and then falls back to IPv4
on error. The shim does the same. If they pick different ones, this
breaks down.
Normally, fallback happens because the system doesn't have IPv6, and
both sides will make the same decision. But if binding to IPv6 fails for
other reasons, they may mismatch. We're observing them fail due to what
seems to port exhaustion. When this happens, shim and runner don't
agree on the same address family.
Instead, just tell the shim which address to connect to. This doesn't
fix the underlying port exhaustion problem, but it does seem to fix the
flakes. Although given we are still exhausting ports and falling back to
IPv4, it doesn't truly fix it. Later CLs will address port exhaustion by
using a single server port.
This changes the runner <-> shim protocol, but hopefully in a fairly
obvious way that others using BoGo can easily follow. It shouldn't break
our cross-version tests because we keep runner and shim at the same
versio there.
To avoid needing to make an incompatible change to the shim <->
handshaker protocol, which would impact our cross-version tests, this
introduces a mechanism for the shim omit flags when talking to the
handshaker. The handshaker doesn't need to know how to connect to the
runner.
Also print the error string on Windows. Sadly this is a bit tedious.
Change-Id: Ic8bda9a854a115c206c05a659a2e34f544b844a6
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/60885
Auto-Submit: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Right now we use NIDs to configure the group list, but group IDs (the
TLS codepoints) to return the negotiated group. The NIDs come from
OpenSSL, while the group ID was original our API. OpenSSL has since
added SSL_get_negotiated_group, but we don't implement it.
To add Kyber to QUIC, we'll need to add an API for configuring groups to
QUICHE. Carrying over our inconsistency into QUICHE's public API would
be unfortunate, so let's use this as the time to align things.
We could either align with OpenSSL and say NIDs are now the group
representation at the public API, or we could add a parallel group ID
API. (Or we could make a whole new SSL_NAMED_GROUP object to pattern
after SSL_CIPHER, which isn't wrong, but is even more new APIs.)
Aligning with OpenSSL would be fewer APIs, but NIDs aren't a great
representation. The numbers are ad-hoc and even diverge a bit between
OpenSSL and BoringSSL. The TLS codepoints are better to export out to
callers. Also QUICHE has exported the negotiated group using the
codepoints, so the natural solution would be to use codepoints on input
too.
Thus, this CL adds SSL_CTX_set1_group_ids and SSL_set1_group_ids. It
also rearranges the API docs slightly to put the group ID ones first,
and leaves a little note about the NID representation before introducing
those.
While I'm here, I've added SSL_get_negotiated_group. NGINX seems to use
it when available, so we may as well fill in that unnecessary
compatibility hole.
Bug: chromium:1442377
Change-Id: I47ca8ae52c274133f28da9893aed7fc70f942bf8
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/60208
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This adds "group" versions of our codepoint-based APIs. This is mostly
because it looks weird to switch terminology randomly in the
implementation. See I7a356793d36419fc668364c912ca7b4f5c6c23a2 for
additional context.
I've not bothered renaming the bssl_shim flags. That seems a waste of
time.
Change-Id: I566ad132f5a33d99efd8cb2bb8b76d9a6038c825
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/60207
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
This TLS 1.2 algorithm is substantially inferior to AES-GCM and should
never be used. It will not be available unless configured by name.
However, in can be used to provide backwards-compatibility with devices
that cannot be updated if so needed.
Change-Id: I1fd78efeb33aceca76ec2e7cb76b70f761ed1af8
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/59585
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Auto-Submit: Adam Langley <agl@google.com>
The initial codepoint is called X25519Kyber786Draft00 in the draft, so
align with that name for this version. Also remove the placeholder bits
for the other combinations, which haven't gotten that far yet.
Update-Note: Update references to NID_X25519Kyber768 to
NID_X25519Kyber768Draft00. For now, the old name is available as an
alias.
Change-Id: I2e531947f41e589cec61607944dca844722f0947
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/59605
Auto-Submit: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
It's unwise for organisations to try and define TLS profiles. As in this
case, they usually make security worse. However, since this is already
established and supported by Android, this change raises it to the level
of a supported policy.
Change-Id: Ic66d5eaa33d884e57fc6d8eb922d86882b621e9e
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/58626
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
This relands
https://boringssl-review.googlesource.com/c/boringssl/+/54606, which was
temporarily reverted.
Update-Note: By default, clients will now require RSA server
certificates used in TLS 1.2 and earlier to include the keyEncipherment
or digitalSignature bit. keyEncipherment is required if using RSA key
exchange. digitalSignature is required if using ECDHE_RSA key exchange.
If unsure, TLS RSA server signatures should include both, but some
deployments may wish to include only one if separating keys, or simply
disabling RSA key exchange. The latter is useful to mitigate either the
Bleichenbacher attack (from 1998, most recently resurfaced in 2017 as
ROBOT), or to strengthen TLS 1.3 downgrade protections, which is
particularly important for enterprise environments using client
certificates (aka "mTLS") because, prior to TLS 1.3, the TLS client
certificate flow was insufficiently encrypted or authenticated. Without
reflecting an RSA key exchange disable into key usage, and then the
client checking it, an attacker can spoof a CertificateRequest as coming
from some server.
This aligns with standard security requirements for using X.509
certificates, specified in RFC 5280, section 4.2.1.3, and reiterated in
TLS as early as TLS 1.0, RFC 2246, section 7.4.2, published 24 years ago
on January 1999. Constraints on usage of keys are important to mitigate
cross-protocol attacks, a class of cryptographic attacks that is
well-studied in the literature.
We already checked this for each of ECDSA, TLS 1.3, and servers
verifying client certificates, so this just fills in the remaining hole.
As a result, this change is also important for avoiding some weird
behaviors when configuration changes transition a server in or out of
this hole. (We've seen connection failures get misattributed to TLS 1.3
when it was really a certificate misconfiguration.)
Chrome has also enforced this for some time with publicly-trusted
certificates. As a temporary measure for callers that need more time,
the SSL_set_enforce_rsa_key_usage API, added to BoringSSL in January
2019, still exists where we need to turn this off.
Fixed: 519
Change-Id: I91bf2cfb04c92aec7875e640f90ba6f837146dc1
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/58805
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
HRSS itself remains in libcrypto because there are some direct users of
it. But this will let it be dropped by the linker in many cases.
Change-Id: I870eda30c9ed1d08693c770e9e7df45a2711b7df
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/58645
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Also unexport PEM_proc_type and PEM_dek_info. They're never called
externally, just private functions within one file. Also, while I'm
here, fix the include guard on asn1/internal.h.
Bug: 516
Change-Id: I6961a65f638e7b464a8c349663898a954d7826b4
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/58548
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Rather than trying to override the actual malloc symbol, just intercept
OPENSSL_malloc and gate it on a build flag. (When we first wrote these,
OPENSSL_malloc was just an alias for malloc.)
This has several benefits:
- This is cross platform. We don't interfere with sanitizers or the
libc, or have to mess with global symbols.
- This removes the reason bssl_shim and handshaker linked
test_support_lib, so we can fix the tes_support_lib / gtest
dependency.
- If we ever reduce the scope of fallible mallocs, we'll want to
constrain the tests to only the ones that are fallible. An
interception strategy like this can do it. Hopefully that will also
take less time to run in the future.
Also fix the ssl malloc failure tests, as they haven't been working for
a while. (Malloc failure tests still take far too long to run to the end
though. My immediate motivation is less malloc failure and more to tidy
up the build.)
Bug: 563
Change-Id: I32165b8ecbebfdcfde26964e06a404762edd28e3
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/56925
Commit-Queue: Bob Beck <bbe@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
and isxdigit.
All of these can be affected by locale, and although we weren't using
them directly (except for isxdigit) we instead had manual versions inline
everywhere.
While I am here add OPENSSL_fromxdigit and deduplicate a bunch of code
in hex decoders pulling out a hex value.
Change-Id: Ie75a4fba0f043208c50b0bb14174516462c89673
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/56648
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
While information is contradictory on this subject, investigation
of several implementaions and Posix appears to indicate that it
is possible to change the behaviour of isdigit() with locale.
Change-Id: I6ba9ecbb5563d04d41c54dd071e86b2354483f77
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/56625
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
We've been setting /Wall on MSVC for a while, but /Wall in MSVC is
really "all". I.e., it's -Weverything in Clang and GCC, and includes
many warnings that are simply diagnostics. MSVC's own headers do not
promise to be clean under /Wall.
Rather, the equivalent of Clang and GCC's -Wall is /W4, which MSVC does
promise to pass. Under /Wall, every new MSVC release we've had to update
an ever-growing list of warning exceptions, to disable the
off-by-default warnings that were off for a reason. Switch to MSVC's
recommendations and just enable /W4.
From there, I've trimmed the exception list, now that we don't need to
re-disable disabled warnings. A few non-disabled warnings also no longer
need exceptions.
This should fix the build with VS2022, which was failing due to C5264.
C5264 flags a couple of instances in the library, but tons and tons in
MSVC headers and googletest. (The instances in the library are a little
goofy and reflect things that should have been 'inline constexpr', but
we can't rely on C++17 yet. Though I may go add a compat macro for
that.)
Fixed: 552
Change-Id: I9031c0d5bd4c7a4df1dc3040b38af9cfbcffc06e
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/56045
Auto-Submit: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
This function reports when security-critical checks on the X.509 key
usage extension would have failed, but were skipped due to the temporary
exception in SSL_set_enforce_rsa_key_usage. This function is meant to
aid deployments as they work through enabling this.
Change-Id: Ice0359879c0a6cbe55bf0cb81a63685506883123
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/55465
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
This reverts commit 64393b57e8. We'll
reland this change in January. Projects that rely on this revert should
use SSL_set_enforce_rsa_key_usage, available since 2019, to control the
security check without being reliant on the defaults.
Bug: 519
Change-Id: Icf53eae8c29f316c7df4ec1a7c16626ac3af8560
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/55005
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Bug: 516
Change-Id: Iba2014da414658c08e42e0993912fa73848832d3
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/54945
Reviewed-by: Bob Beck <bbe@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Update-Note: Clients will now require RSA server certificates used in
TLS 1.2 and earlier to include the keyEncipherment or digitalSignature
bit. keyEncipherment is required if using RSA key exchange.
digitalSignature is required if using ECDHE_RSA key exchange.
We already required this for each of ECDSA, TLS 1.3, and servers
verifying client certificates, so this just fills in the remaining hole.
Chrome has also enforced this for some time with publicly-trusted
certificates. For now, the SSL_set_enforce_rsa_key_usage API still
exists where we need to turn this off.
Fixed: 519
Change-Id: Ia440b00b60a224fa608702439aa120d633d81ddc
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/54606
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
These functions aid in meeting specific compliance goals and allows
configuration of things like TLS 1.3 cipher suites, which are otherwise
not configurable.
Change-Id: I668afc734a19ecd4b996eaa23be73ce259b13fa2
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/52625
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
In testing out the ECH bits on the Chromium side, it is much harder to
tell what's going on without some indication that we sent a
ClientHelloInner. This CL routes it into the callback. A corresponding
CL in Chromium will add it to NetLog.
Bug: 275
Change-Id: I945ab2679614583e875a0ba90d6cf1481ed315d9
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/51205
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
This silences a pile of -Wformat-signedness warnings. We still need
casts in a few places where the API gives int but really wanted
uint16_t. There I cast to unsigned instead of uint16_t for the sake of
not losing information.
With that, we should be -Wformat-signedness-clean on GCC, so enable the
warning.
Bug: 450
Change-Id: I3ab10348bb47d398b8b9b39acf360284a8ab04d7
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/50771
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Whether the order makes sense is another matter, but keep them aligned
so future flags have an easier time with it.
Change-Id: I3c3912039b593a55af86078b2e9768c76ee2ee14
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/50770
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
The command-line parser is slightly showing its age: first, it is hard
to add new integral types, such as uint16_t, which is getting in the way
of fixing some of the -Wformat-signedness errors. Second, the parameter
extraction logic and skipping logic is duplicated in every type.
While I'm here, use a binary search to look up the flag, since we have
rather a lot of them. With more C++ template tricks, we could avoid the
std::function, but that seemed more trouble than was worth it,
especially since, prior to C++17, it's a little hard to convince
template argument deduction to infer one of the parameters.
Change-Id: I208f89d46371b31fc8b44487725296bcd9d7c8e7
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/50769
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
GCC has a warning that complains about even more type mismatches in
printf. Some of these are a bit messy and will be fixed in separate CLs.
This covers the easy ones.
The .*s stuff is unfortunate, but printf has no size_t-clean string
printer. ALPN protocol lengths are bound by uint8_t, so it doesn't
really matter.
The IPv6 printing one is obnoxious and arguably a false positive. It's
really a C language flaw: all types smaller than int get converted to
int when you do arithmetic. So something like this first doesn't
overflow the shift because it computes over int, but then the result
overall is stored as an int.
uint8_t a, b;
(a << 8) | b
On the one hand, this fixes a "missing" cast to uint16_t before the
shift. At the same time, the incorrect final type means passing it to
%x, which expects unsigned int. The compiler has forgotten this value
actually fits in uint16_t and flags a warning. Mitigate this by storing
in a uint16_t first.
The story doesn't quite end here. Arguments passed to variadic functions
go through integer promotion[0], so the argument is still passed to
snprintf as an int! But then va_arg allows for a signedness mismatch[1],
provided the value is representable in both types. The combination means
that %x, though actually paired with unsigned, also accept uint8_t and
uint16_t, because those are guaranteed to promote to an int that meets
[1]. GCC recognizes [1] applies here.
(There's also PRI16x, but that's a bit tedious to use and, in glibc, is
defined as plain "x" anyway.)
[0] https://en.cppreference.com/w/c/language/conversion#Default_argument_promotions
[1] https://en.cppreference.com/w/c/variadic/va_arg
Bug: 450
Change-Id: Ic1d41356755a18ab922956dd2e07b560470341f4
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/50765
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
This unexports X509, X509_CINF, X509_NAME_ENTRY, X509_NAME, X509_OBJECT,
X509_LOOKUP_METHOD, X509_STORE, X509_LOOKUP, and X509_STORE_CTX.
Note this means X509_STORE_CTX can no longer be stack-allocated.
Update-Note: Patch cl/390055173 into the roll that includes this. This
unexports most of the X.509 structs, aligning with OpenSSL. Use the
accessor APIs instead.
Bug: 425
Change-Id: I53e915bfae3b8dc4b67642279d0e54dc606f2297
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48985
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
If a client offers ECH, but the server rejects it, the client completes
the handshake with ClientHelloOuter in order to authenticate retry keys.
Implement this flow. This is largely allowing the existing handshake to
proceed, but with some changes:
- Certificate verification uses the other name. This CL routes this up to
the built-in verifier and adds SSL_get0_ech_name_override for the
callback.
- We need to disable False Start to pick up server Finished in TLS 1.2.
- Client certificates, notably in TLS 1.3 where they're encrypted,
should only be revealed to the true server. Fortunately, not sending
client certs is always an option, so do that.
Channel ID has a similar issue. I've just omitted the extension in
ClientHelloOuter because it's deprecated and is unlikely to be used
with ECH at this point. ALPS may be worth some pondering but, the way
it's currently used, is not sensitive.
(Possibly we should change the draft to terminate the handshake before
even sending that flight...)
- The session is never offered in ClientHelloOuter, but our internal
book-keeping doesn't quite notice.
I had to replace ech_accept with a tri-state ech_status to correctly
handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT +
reverify_on_resume are all enabled, the first certificate verification
is for the 0-RTT session and should be against the true name, yet we
have selected_ech_config && !ech_accept. A tri-state tracks when ECH is
actually rejected. I've maintained this on the server as well, though
the server never actually cares.
Bug: 275
Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Although not permitted by the TLS specification, systems sometimes
ossify TLS extension order, or byte offsets of various fields. To
keep the ecosystem healthy, add an API to reorder ClientHello
extensions.
Since ECH, HelloRetryRequest, and HelloVerifyRequest are sensitive to
extension order, I've implemented this by per-connection permutation of
the indices in the kExtensions structure. This ensures that all
ClientHellos within a connection are consistently ordered. As follow-up
work, permuting the other messages would also be nice, though any server
messages would need to be incorporated in handshake hints.
Change-Id: I18ce39b4df5ee376c654943f07ec26a50e0923a9
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48045
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Based on an initial implementation by Dan McArdle at
https://boringssl-review.googlesource.com/c/boringssl/+/46784
This CL contains most of a client implementation for
draft-ietf-tls-esni-10. The pieces missing so far, which will be done in
follow-up CLs are:
1. While the ClientHelloInner is padded, the server Certificate message
is not. I'll add that once we resolve the spec discussions on how to
do that. (We were originally going to use TLS record-level padding,
but that doesn't work well with QUIC.)
2. The client should check the public name is a valid DNS name before
copying it into ClientHelloOuter.server_name.
3. The ClientHelloOuter handshake flow is not yet implemented. This CL
can detect when the server selects ClientHelloOuter, but for now the
handshake immediately fails. A follow-up CL will remove that logic
and instead add the APIs and extra checks needed.
Otherwise, this should be complete, including padding and compression.
The main interesting point design-wise is that we run through
ClientHello construction multiple times. We need to construct
ClientHelloInner and ClientHelloOuter. Then each of those has slight
variants: EncodedClientHelloInner is the compressed form, and
ClientHelloOuterAAD just has the ECH extension erased to avoid a
circular dependency.
I've computed ClientHelloInner and EncodedClientHelloInner concurrently
because the compression scheme requires shifting the extensions around
to be contiguous. However, I've computed ClientHelloOuterAAD and
ClientHelloOuter by running through the logic twice. This probably can
be done better, but the next draft revises the construction anyway, so
I'm thinking I'll rework it then. (In the next draft, we use a
placeholder payload of the same length, so we can construct the
ClientHello once and fill in the payload.)
Additionally, now that we have a client available in ssl_test, this adds
a threading test to confirm that SSL_CTX_set1_ech_keys is properly
synchronized. (Confirmed that, if I drop the lock in
SSL_CTX_set1_ech_keys, TSan notices.)
Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004
Reviewed-by: Adam Langley <agl@google.com>
Previously we would extract the KEM ID from the ECHConfig and then parse
the private key using the corresponding KEM type. This CL makes it take
a pre-pared EVP_HPKE_KEY and checks it matches. This does require the
caller pass the key type through externally, which is probably prudent?
(On the other hand we are still inferring config from the rest of the
ECHConfig... maybe we can add an API to extract the EVP_HPKE_KEM from a
serialized ECHConfig if it becomes a problem. I could see runner or tool
wanting that out of convenience.)
The immediate motivation is to add APIs to programmatically construct
ECHConfigs. I'm thinking we can pass a const EVP_HPKE_KEY * to specify
the key, at which point it's weird for SSL_ECH_KEYS_add to look
different.
Bug: 275
Change-Id: I2d424323885103d3fe0a99a9012c160baa8653bd
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48002
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
While decompression is deterministic, compression is not. New revisions
of the compression algorithm may start using different (hopefully
smaller!) compressions. So this doesn't cause hint mismatches, add a
certificate compression hint. If the shim's Certificate message matches
the handshaker, we'll reuse the already compressed message.
This also adds what appears to be a missing test for when the server
cannot find compression algorithms in common.
Change-Id: Idbedaceb20208463d8f61581ee27971c17fcd126
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48005
Reviewed-by: Adam Langley <agl@google.com>
The first thing any deployment will want to monitor is whether ECH was
actually used. Also it's useful if the command-line tool can output
this. (The alert is how the client signals it discarded the connection
due to ECH reject.)
This also disables ECH with the handoff mechanism for now. (The
immediate cause being that ech_accept isn't serialized.) We'll probably
need to make some decisions around the ordering here, since ECH affects
where the true ClientHello is available.
Bug: 275
Change-Id: Ie4559733290e653a514fcd94431090bf86bc3172
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47911
Reviewed-by: Adam Langley <agl@google.com>
The remaining remnants of Channel ID all configure the private key ahead
of time. Unwind the callback machinery, which cuts down on async points
and the cases we need to test.
This also unwinds some odd interaction between the callback and
SSL_set_tls_channel_id_enabled: If a client uses
SSL_set_tls_channel_id_enabled but doesn't set a callback, the handshake
would still pause at SSL_ERROR_WANT_CHANNEL_ID_LOOKUP. This is now
removed, so SSL_set_tls_channel_id_enabled only affects the server and
SSL_CTX_set1_tls_channel_id only affects the client.
Update-Note: SSL_CTX_set_channel_id_cb is removed.
SSL_set_tls_channel_id_enabled no longer enables Channel ID as a client,
only as a server.
Change-Id: I89ded99ca65e1c61b1bc4e009ca0bdca0b807359
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47907
Reviewed-by: Adam Langley <agl@google.com>
We didn't end up deploying this. We also never implemented the final
RFC, so what we do have isn't useful for someone who wishes to deploy
it anyway.
Update-Note: Token binding APIs are removed.
Change-Id: Iecea7c3dcf9d3e2644a3b7afaf61511310b45d5f
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47584
Reviewed-by: Adam Langley <agl@google.com>
ServerHello/EncryptedExtensions/Finished is logically one atomic flight
that exits the early data state, we have process each message
sequentially. Until we've processed Finished, we are still in the early
data state and must support writing data. Individual messages *are*
processed atomically, so the interesting points are before ServerHello
(already tested), after ServerHello, and after EncryptedExtensions.
The TLS 1.3 handshake internally clears ssl->session when processing
ServerHello, so getting the early data information from ssl->session
does not work. Instead, use hs->early_session, which is what other
codepaths use.
I've tested this with runner rather than ssl_test, so we can test both
post-SH and post-EE states. ssl_test would be more self-contained, since
we can directly control the API calls, but it cannot test the post-EE
state. To reduce record overhead, our production implementation packs EE
and Finished into the same record, which means the handshake will
process the two atomically. Instead, I've tested this in runner, with a
flag to partially drive the handshake before reading early data.
I've also tweaked the logic to hopefully be a little clearer.
Bug: chromium:1208784
Change-Id: Ia4901042419c5324054f97743bd1aac59ebf8f24
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47485
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
See go/handshake-hints (internal).
CL originally by Bin Wu <wub@google.com>. I just reworked the tests and
tidied it up a bit. This is the start of a replacement for the split
handshakes API. For now, only TLS 1.3 is supported. It starts with an
initial set of hints, but we can add more later. (In particular, we
should probably apply the remote handshaker's extension order to avoid
needing to capability protect such changes.)
Change-Id: I7b6a6dfaa84c6c6e3436d2a4026c3652b8a79f0f
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/46535
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This CL adds an initial implementation of the ECH server, with pieces of
the client in BoGo as necessary for testing. In particular, the server
supports ClientHelloInner compression with ech_outer_extensions. When
ECH decryption fails, it can send retry_configs back to the client.
This server passes the "ech-accept" and "ech-reject" test cases in
tls-interop-runner[0] when tested against both the cloudflare-go and nss
clients. For reproducibility, I started with the main branch at commit
707604c262d8bcf3e944ed1d5a675077304732ce and updated the endpoint's
script to pass the server's ECHConfig and private key to the boringssl
tool.
Follow-up CLs will update HPKE to the latest draft and catch us up to
draft-10.
[0]: https://github.com/xvzcf/tls-interop-runner
Bug: 275
Change-Id: I49be35af46d1fd5dd9c62252f07d0bae179381ab
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/45285
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
We don't support renegotiation on the server anymore. Even if we did, we
wouldn't want to rerun ALPN anyway, and we don't do resumption on
renegotiation.
Change-Id: I43438d084bfe5fbe9b011ae0f53349df1baf6c97
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/46533
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This wasn't being used and wasn't even set correctly in split handshake
tests.
Change-Id: I03000db8dd3c227ea44e7bacaf3d1341259fae44
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/46384
Reviewed-by: Adam Langley <agl@google.com>
David Benjamin
Roland Shoemaker
Victor Tan
Adam Langley
Adam Langley
Bob Beck
Piotr Sikora
Adam Langley
Daniel McArdle