We never ended up using these, or making them work with TLS 1.3 (which
has KeyUpdates and NewSessionTickets). It'd still be nice to have an
in-place API, but for now unwind these ones until we have time to give
it another go. Supporting TLS 1.3's post-handshake messages will
probably require a slightly more involved design.
(I suspect some of the seal_scatter bits in tls_record.cc can also be
simplified with these removed, but I've left them alone here.)
Update-Note: Removed some unused, experimental APIs.
Change-Id: Iad1245fa467cc6e599d20561f5db44d236219e06
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/54527
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Among many many problems with renegotiation is it makes every API
ambiguous. Do we return the pending handshake's properties, or the most
recently completed handshake? Neither answer is unambiguously correct:
On the one hand, OpenSSL's API makes renegotiation transparent, so the
pending handshake breaks invariants. E.g., currently,
SSL_get_current_cipher and other functions can return NULL mid
renegotiation. See https://crbug.com/1010748.
On the other hand, OpenSSL's API is callback-heavy. During a handshake
callback, the application most likely wants to check the pending
parameters. Most notably, cert verify callbacks calling
SSL_get_peer_certificate.
Historically, only the pending state was available to return anyway.
We've since changed this
(https://boringssl-review.googlesource.com/8612), but we kept the public
APIs as-is. I was particularly worried about cert verify callbacks.
As of https://boringssl-review.googlesource.com/c/boringssl/+/14028/ and
https://boringssl-review.googlesource.com/c/boringssl/+/19665/, cert
verify is moot. We implement the 3-SHAKE mitigation in library, so the
peer cert cannot change, and we don't reverify the certificate at all.
With that, I think we should switch to returning the established
parameters. Chromium is the main consumer that enables renegotiation,
and it would be better off with this behavior. (Maybe we should try to
forbid other properties, like the cipher suite, from changing on
renegotiation. Unchangeable properties make this issue moot.)
This CL would break if the handshake internally used SSL_get_session,
but this is no longer true as of
https://boringssl-review.googlesource.com/c/boringssl/+/41865.
Update-Note: Some APIs will now behave differently mid-renegotation. I
think this is the safer option, but it is possible code was relying on
the other one.
Fixed: chromium:1010748
Change-Id: I42157ccd9704cde3eebf947136d47cda6754c36e
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/54165
Auto-Submit: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
Most SSL_ERROR_* values are tracked directly with rwstate. SSL_get_error
is just reading the extra return value out from the previous call.
However, SSL_ERROR_ZERO_RETURN infers close_notify from the SSL's
shutdown state and a zero return value (EOF).
This works, but if we implement SSL_read_ex and SSL_write_ex, a zero
return value is no longer as carefully correlated with EOF. Moreover,
it's already possible to get a non-EOF zero return post-close_notify if
BIO_write returns an (arguably incorrect) return value. Instead, track
SSL_ERROR_ZERO_RETURN in rwstate explicitly.
Since rwstate is exposed as SSL_want and SSL_ERROR_ZERO_RETURN was
previously never returned there, I've made it map SSL_ERROR_ZERO_RETURN
back to SSL_ERROR_NONE. I've also added a test for BIO_write returning
zero, though the real purpose is for a subsequent SSL_write_ex
implementation to retain all the other tests we've added in here.
Update-Note: This is intended to be safe, but if anything breaks around
EOFs, this change is a likely culprit.
Bug: 507
Change-Id: Ide0807665f2e02ee695c4976dc5e99fb10502cf0
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53946
Auto-Submit: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
A later CL will tighten up SSL_ERROR_ZERO_RETURN handling. In
preparation for this, test that SSL_CTX_set_quiet_shutdown can trigger
SSL_ERROR_ZERO_RETURN.
Bug: 507
Change-Id: Ib50a02c514673ad4b73540934480d54b372d9505
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53945
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
https://boringssl-review.googlesource.com/c/boringssl/+/53007
inadvertently changed the semantics of SSL_load_client_CA_file slightly.
The original implementation, by delaying allocating ret, would fail
rather than return an empty list.
Fix this and add a test. We don't have much support for testing
filesystem-related things yet, so I've just used /dev/null and gated it
to Linux + macOS for now. If we need it later, we can add temporary file
support to the test_support library.
Change-Id: If77dd493a433819a65378d76cf400cce48c0abaa
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53785
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
This works correctly, but part of implementing SSL_write_ex will, if not
done correctly, regress this. Specifically, if the read_shutdown check
in SSL_get_error were not conditioned on ret == 0, the last
SSL_get_error in the test would mistakenly classify the write error as
SSL_ERROR_ZERO_RETURN.
Add a regression test in advance.
Bug: 507
Change-Id: I8ddb4606e291977506ee81f4ed11427e5b1636d8
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53626
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Writing application data goes through three steps:
1. Encrypt the data into the write buffer.
2. Flush the write buffer to the network.
3. Report to SSL_write's caller that the write succeeded.
In principle, steps 2 and 3 are done together, but it is possible that
BoringSSL needs to write something, but we are not in the middle of
servicing an SSL_write call. Then we must perform (2) but cannot perform
(3).
TLS 1.3 0-RTT on a client introduces a case like this. Suppose we write
some 0-RTT data, but it is blocked on the network. Meanwhile, the
application tries to read from the socket (protocols like HTTP/2 read
and write concurrently). We discover ServerHello..Finished and must then
respond with EndOfEarlyData..Finished. But to write, we must flush the
current write buffer.
To fix this, https://boringssl-review.googlesource.com/14164 split (2)
and (3) more explicitly. The write buffer may be flushed to the network
at any point, but the wpend_* book-keeping is separate. It represents
whether (3) is done. As part of that, we introduced a wpend_pending
boolean to track whether there was pending data.
This introduces an interesting corner case. We now keep NewSessionTicket
messages buffered until the next SSL_write. (KeyUpdate ACKs are
implemented similarly.) Suppose the caller calls SSL_write(nullptr, 0)
to flush the NewSessionTicket and this hits EWOULDBLOCK. We'll track a
zero-length pending write in wpend_*! A future attempt to write non-zero
data would then violate the moving buffer check. This is strange because
we don't build records for zero-length application writes in the first
place.
Instead, wpend_pending should have been wpend_tot > 0. Remove that and
rearrange the code to check that properly. Also remove wpend_ret as it
has the same data as wpend_tot.
Change-Id: I58c23842cd55e8a8dfbb1854b61278b108b5c7ea
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53546
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
This is split out from
https://boringssl-review.googlesource.com/c/boringssl/+/47544 just to
get the bugfixes and tests out of the way of the refactor.
If we trip the SSL_R_BAD_LENGTH check in tls_write_app_data, wnum is set
to zero. But wnum should only be cleared on a successful write. It
tracks the number of input bytes that have been written to the transport
but not yet reported to the caller. Instead, move it to the success
return in that function. All the other error paths already set it to
something else.
Change-Id: Ib22f9cf04454ecdb0062077f183be5070ab7d791
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53545
Reviewed-by: Bob Beck <bbe@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
CPython uses this function. Our implementation is slightly weird since
it leaks the clamping behavior, but probably not a big deal.
Update-Note: When this is merged into the internal repository, we can
simplify the CPython patches.
Change-Id: I291ddf852fb463bf02998fe04d0d0e8cb358dc55
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53485
Commit-Queue: Bob Beck <bbe@google.com>
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
Our TLS 1.3 stack predates OpenSSL's. We chose TLS1_CK_* to align with
the existing names. OpenSSL made a new convention, TLS1_3_CK_*. Match
them.
This means that, in the likely event that TLS 1.4 uses the same
constants, they'll have weird names, just as several of our constants
still say SSL3_* but it doesn't particularly matter.
Change-Id: I97f29b224d0d282e946344e4b907f2df2be39ce1
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53425
Auto-Submit: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Bob Beck <bbe@google.com>
Reviewed-by: Bob Beck <bbe@google.com>
SSL_load_client_CA_file can just call
SSL_add_file_cert_subjects_to_stack.
SSL_add_file_cert_subjects_to_stack itself is rewritten to use scopers
and also give subquadratic running time. Sorting after every insertion
does not actually help. (It would have been faster to do a linear
search.) Instead, gather the names first, then sort and deduplicate.
Finally, add a SSL_add_bio_cert_subjects_to_stack. This is both to
simplify testing and because Envoy code copied from
SSL_add_file_cert_subjects_to_stack, complete with the quadratic
behavior. It is the only external project that depends on the
STACK_OF(T) comparison function. To simplify making that const-correct,
just export the function they needed anyway.
Bug: 498
Change-Id: I00d13c949a535c0d60873fe4ba2e5604bb585cca
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/53007
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
CPython and wpa_supplicant are using this nowadays. To avoid needing to
tweak the ticket nonce derivation, I've just internally capped the
number of tickets at 16, which should be plenty.
Change-Id: Ie84c15b81a2abe8ec729992e515e0bd4cc351037
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/52465
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
The check finds implicit conversions of integer literals to bools:
bool b1 = 1;
bool b2 = static_cast<bool>(1);
and transforms them to:
bool b1 = true;
bool b2 = true;
Bug: chromium:1290142
Change-Id: I15579e28f544d07b331a230b70a8278e0651150d
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/51085
Reviewed-by: David Benjamin <davidben@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>
Always enable X509_CHECK_FLAG_NO_PARTIAL_WILDCARDS and never enable
X509_CHECK_FLAG_MULTI_LABEL_WILDCARDS.
Update-Note: BoringSSL will no longer accept wildcard patterns like
*www.example.com or www*.example.com. (It already did not accept
ww*w.example.com.) X509_CHECK_FLAG_MULTI_LABEL_WILDCARDS will also be
ignored and can no longer be used to allow foo.bar.example.com to match
*.example.com.
Fixes: 462
Change-Id: I004e087bf70f4c3f249235cd864d9e19cc9a5102
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/50705
Reviewed-by: Adam Langley <agl@google.com>
This was added in OpenSSL 1.1.x. It is slightly different from
SSL_pending in that it also reports buffered transport data.
Change-Id: I81e217aad1ceb6f4c31c36634a546e12b6dc8dfc
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/50445
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
There is a long outdated comment that TLS 1.3 is disabled by default,
which is no longer true. While I'm here, run through all TLS and DTLS
versions, now that we have that table.
Change-Id: I7b813111ad3be295cc5a7e0eb0c7088e40df2a35
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/49905
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Later CLs will clean up the ClientHello construction a bit (draft-12
avoids computing ClientHelloOuter twice). I suspect the transcript
handling on the client can also be simpler, but I'll see what's
convenient after I've changed how ClientHelloOuter is constructed.
Changes of note between draft-10 and draft-13:
- There is now an ECH confirmation signal in both HRR and SH. We don't
actually make much use of this in our client right now, but it
resolves a bunch of weird issues around HRR, including edge cases if
HRR applies to one ClientHello but not the other.
- The confirmation signal no longer depends on key_share and PSK, so we
don't have to work around a weird ordering issue.
- ech_is_inner is now folded into the main encrypted_client_hello code
point. This works better with some stuff around HRR.
- Padding is moved from the padding extension, computed with
ClientHelloInner, to something we fill in afterwards. This makes it
easier to pad up the whole thing to a multiple of 32. I've accordingly
updated to the latest recommended padding construction, and updated
the GREASE logic to match.
- ech_outer_extensions is much easier to process because the order is
required to be consistent. We were doing that anyway, and now a simple
linear scan works.
- ClientHelloOuterAAD now uses an all zero placeholder payload of the
same length. This lets us simplify the server code, but, for now, I've
kept the client code the same. I'll follow this up with a CL to avoid
computing ClientHelloOuter twice.
- ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't
filled that in and will do it in a follow-up CL.
Bug: 275
Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912
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>
ssl_update_cache takes the cache lock to add to the session cache,
releases it, and then immediately takes and releases the lock to
increment handshakes_since_cache_flush. Then, in 1/255 connections, does
the same thing again to flush stale sessions.
Merge the first two into one lock. In doing so, move ssl_update_cache to
ssl_session.cc, so it can access a newly-extracted add_session_lock.
Also remove the mode parameter (the SSL knows if it's a client or
server), and move the established_session != session check to the
caller, which more directly knows whether there was a new session.
Also add some TSan coverage for this path in the tests. In an earlier
iteration of this patch, I managed to introduce a double-locking bug
because we weren't testing it at all. Confirmed this test catches both
double-locking and insufficient locking. (It doesn't seem able to catch
using a read lock instead of a write lock in SSL_CTX_flush_sessions,
however. I suspect the hash table is distributing the cells each thread
touches.)
Update-Note: This reshuffles some locks around the session cache.
(Hopefully for the better.)
Change-Id: I78dca53fda74e036b90110cca7fbcc306a5c8ebe
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48133
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This was added in draft-11, which I'll update to more broadly in a
follow-up CL. This is an easily separable component: we don't want to
allow the DNS to arbitrarily insert strings in the ClientHello, so
invalid public names are rejected.
Unfortunately, we have a bit of a mess because DNS syntax does not
exclude IPv4 literals, yet everyone sticks DNS and IP literals in the
same string. The RFC3986 rules are alright, but don't match reality.
Reality is (probably?) the WHATWG rules, which are a mess.
The load-bearing bit of the spec is that, at certificate verification,
you should reject whatever strings your application refuses to represent
as a DNS name. I'll have Chromium call into its URL parser.
https://www.ietf.org/archive/id/draft-ietf-tls-esni-11.html#section-6.1.4.3-3
But there's still a bit at the validation step where clients "SHOULD"
run the IPv4 parser. In case downstream logic forgets, I've gone ahead
and implemented the WHATWG IPv4 parser.
https://www.ietf.org/archive/id/draft-ietf-tls-esni-11.html#section-4-6.6.1
Bug: 275
Change-Id: I15aa1ac0391df9c3859c44b8a259296e1907b7d4
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48085
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This CL fixes a couple of things. First, we never tested that SSL_write
refuses to write application data after a fatal alert, so add some tests
here. With those tests, we can revise some of this logic:
Next, this removes the write_shutdown check in SSL_write and instead
relies on the lower-level versions of the check in the write_app_data,
etc., hooks. This improves error-reporting on handshake errors:
We generally try to make SSL_do_handshake errors sticky, analogous to
handshakeErr in the Go implementation. SSL_write and SSL_read both
implicitly call SSL_do_handshake. Callers driving the two in parallel
will naturally call SSL_do_handshake twice. Since the error effectively
applies to both operations, we save and replay handshake errors
(hs->error).
Handshake errors typically come with sending alerts, which also sets
write_shutdown so we don't try to send more data over the channel.
Checking this early in SSL_write means we don't get a chance to replay
the handshake error. So this CL defers it, and the test ensures we still
ultimately get it right.
Finally, https://crbug.com/1078515 is a particular incarnation of this.
If the server enables 0-RTT and then reverts to TLS 1.2, clients need
to catch the error and retry. There, deferring the SSL_write check
isn't sufficient, because the can_early_write bit removes the write
path's dependency on the handshake, so we don't call into
SSL_do_handshake at all.
For now, I've made this error path clear can_early_write. I suspect
we want it to apply to all handshake errors, though it's weird because
the handshake error is effectively a read error in 0-RTT. We don't
currently replay record decryption failures at SSL_write, even though
those also send a fatal alert and thus break all subsequent writes.
Bug: chromium:1078515
Change-Id: Icdfae6a8f2e7c1b1c921068dca244795a670807f
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48065
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>
We'll probably need to make this more complex later, but this should be
a start. I had hoped this would also simplify tests, MakeECHConfig() was
still needed to generate weird inputs for tests. I've instead tidied
that up a bit with a params structure. Now the only hard-coded ECHConfig
in tests is to check the output of the new API.
Bug: 275
Change-Id: I640a224fb4b7a7d20e8a2cd7a1e75d1e3fe69936
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48003
Commit-Queue: David Benjamin <davidben@google.com>
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>
When decrypting a ticket we would copy the client's session ID into the
session and then copy the session's ID into the ServerHello (if
resuming). That seems icky. Instead install the same placeholder on the
server as we do on the client.
Change-Id: Icb50a3be2f05e6428f1b286c8c09015f7bb4af16
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47784
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
This introduces an EVP_HPKE_KEM, to capture the KEM choice, and
EVP_HPKE_KEY, to capture the key import (and thus avoids asking
receivers to pass in the full keypair). It is a bit more wordy now, but
we'll be in a better place when some non-TLS user inevitably asks for a
P-256 version.
Bug: 410
Change-Id: Icb9cc8b028e6d1f86e6d8adb31ebf1f975181675
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47329
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Callers using private key callbacks may retain non-trivial state with a
private key. In many cases, the private key is no longer necessary
immediately after the first round-trip (e.g. non-HRR TLS 1.3
connections). Add a function that callers can query to drop the state a
hair earlier.
This is tested in two ways. First, the asserts in front of using the
key, combined with existing tests, ensure we don't start reporting it
too early. Second, I've added tests in ssl_test.cc to assert we report
it as early as we expect to.
In doing so, the number of parameters on ConnectClientAndServer()
started getting tedious, so I've split that into a
CreateClientAndServer() and CompleteHandshakes(). Callers that need to
configure weird things or drive the handshake manually can call
CreateClientAndServer() (which takes care of the BIO pair business) and
continue from there.
Bug: b/183734559
Change-Id: I05e1edb6d269c8468ba7cde7dc90e0856694a0ca
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47344
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This replaces the ID-based API with one that is more static linker
friendly. For ECH, it doesn't make a difference because we currently
pull in all the options we've implemented. But this means other HPKE
uses need not pull in everything ECH needs and vice versa.
Along the way, fix an inconsistency: we prefixed all the AEAD constants
with "AEAD", but not the others. Since the rest of the name already
determines everything, go with the shorter version.
Bug: 410
Change-Id: I56e46c13b43c97e15eeb45204cde7019dd21e250
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47327
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
We can add them if we need them, but we're only using HKDF-SHA256 in
ECH. Keep the set small to encourage a common set of parameters.
Bug: 410
Change-Id: I5b9ddf3daa1d0c7f35df473470998369e9882553
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47324
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Bug: 275
Change-Id: I8096070386af7d2b5020875ea09bcc0c04ebc8cd
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/47245
Reviewed-by: Steven Valdez <svaldez@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: Steven Valdez <svaldez@google.com>
We have a few too many X.509 parsers.
Bug: chromium:1199744
Change-Id: Ib6f6b7bf6059ed542c334a5ca5a2d3928aae3bef
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/46904
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
See also 86a90dc749af91f8a7b8da6628c9ffca2bae3009 from upstream. This
differs from upstream's which treats {NULL, 2} as a valid way to spell
the empty list. (I think this is a mistake and have asked them about
it.)
Upstream's CL also, for them, newly makes the empty list disable ALPN,
when previously they'd disable it but misread it as a malloc failure.
For us, we'd already fixed the misreading due to our switch to
bssl::Array and bssl::Span, but the documentation was odd. This CL
preserves that behavior, but updates the documentation and writes a
test.
Update-Note: SSL_CTX_set_alpn_protos and SSL_set_alpn_protos will now
reject invalud inputs. Previously, they would accept them, but silently
send an invalid ALPN extension which the server would almost certainly
error on.
Change-Id: Id5830b2d8c3a5cee4712878fe92ee350c4914367
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/46804
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>
QUICHE currently does not know to call
SSL_set_quic_use_legacy_codepoint, picking up the current default of the
legacy code point. It then assumes that the
TLSEXT_TYPE_quic_transport_parameters constant may be used to extract
transport parameters, so after
https://boringssl-review.googlesource.com/c/boringssl/+/44704, it
breaks.
To smooth over the transition, we now define three constants:
TLSEXT_TYPE_quic_transport_parameters_legacy,
TLSEXT_TYPE_quic_transport_parameters_standard, and the old constant.
The old constant will match whatever the default is (for now, legacy) so
the default is self-consistent. Then plan is then:
1. BoringSSL switches to the state in this CL: the default code point
and constant are the legacy one, but there are APIs to specify the
code point. This will not affect QUICHE, which only uses the
defaults.
2. QUICHE calls SSL_set_quic_use_legacy_codepoint and uses the
corresponding _legacy or _standard constant. It should *not* use the
unsuffixed constant at this point.
3. BoringSSL switches the default setting and the constant to the
standard code point. This will not affect QUICHE, which explicitly
configures the code point it wants.
4. Optional: BoringSSL won't switch the default back to legacy, so
QUICHE can switch _standard to unsuffixed and BoringSSL
can remove the _standard alias (but not the function) early.
5. When QUICHE no longer needs both code points, it unwinds the
SSL_set_quic_use_legacy_codepoint code and switches back to the
unsuffixed constant.
6. BoringSSL removes all this scaffolding now that it's no longer
needed.
Update-Note: This this fixes a compatibility issue with
https://boringssl-review.googlesource.com/c/boringssl/+/44704.
Change-Id: I9f75845aba58ba93e9665cd6f05bcd080eb5f139
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/45124
Reviewed-by: David Schinazi <dschinazi@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
IETF QUIC draft 33 is replacing the TLS extension
codepoint for QUIC transport parameters from 0xffa5
to 57. To support multiple versions of Chrome, we
need to support both codepoints in BoringSSL. This
CL adds support for the new codepoint in a way that
can be enabled on individual connections.
Note that when BoringSSL is not in QUIC mode, it
will error if it sees the new codepoint as a server
but it will ignore the legacy codepoint as that could
be a different private usage of that codepoint.
Change-Id: I314f8f0b169cedd96eeccc42b44153e97044388c
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44704
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
IETF QUIC draft 33 is replacing the TLS extension
codepoint for QUIC transport parameters from 0xffa5
to 57. To support multiple versions of Chrome, we
need to support both codepoints in BoringSSL. This
CL adds support for the new codepoint in a way that
can be enabled on individual connections.
Change-Id: I3bf06ea0710702c0dc45bb3ff2e3d772e9f87f9b
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44585
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
(Original CL by svaldez, reworked by davidben.)
Change-Id: I8570808fa5e96a1c9e6e03c4877039a22e73254f
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/42404
Reviewed-by: Steven Valdez <svaldez@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
OpenSSL synchronizes bio->next_bio and ssl->rbio with a variety of
callbacks, so BIO_copy_next_retry worked. We do not, so attempting to
flush the BIO crashed.
The SSL BIO is a compatibility hack and intentionally much more limited,
so start by just copying things from the right BIO directly. Add a basic
unit test for SSL BIOs. If we need to, we can implement a more complex
synchronization later.
Additionally reject reconfiguring an SSL BIO because that will leak the
object right now.
Change-Id: I724c95ab6f1a3a1aa1889b0483c81ce3bdc534ae
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/43424
Reviewed-by: Adam Langley <agl@google.com>
The quic_early_data_context should always be saved in the SSL_SESSION.
This change fixes a bug where it was only saved in the SSL_SESSION on
full handshakes (but not resumption handshakes).
Bug: 376
Change-Id: I3659d3398e85ac4263760b504d7ea8458fc7e1e2
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/43264
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Update-Note: If an SSL_QUIC_METHOD is set, connections will now fail if
ALPN is not negotiated. This new behavior can be detected by checking
if the value of BORINGSSL_API_VERSION is greater than 10.
Bug: 294
Change-Id: I42fb80aa09268e77cec4a51e49cdad79bd72fa58
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/42304
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
David Benjamin
Anton Bikineev
Adam Langley
Steven Valdez
Daniel McArdle
David Schinazi
Nick Harper