This is a minor regression from
https://boringssl-review.googlesource.com/5235.
If the client, for whatever reason, had an ID-based session but also
supports tickets, it will send non-empty ID + empty ticket extension.
If the ticket extension is non-empty, then the ID is not an ID but a
dummy signaling value, so 5235 avoided looking it up. But if it is
present and empty, the ID is still an ID and should be looked up.
This shouldn't have any practical consequences, except if a server
switched from not supporting tickets and then started supporting it,
while keeping the session cache fixed.
Add a test for this case, and tighten up existing ID vs ticket tests so
they fail if we resume with the wrong type.
Change-Id: Id4d08cd809af00af30a2b67fe3a971078e404c75
Reviewed-on: https://boringssl-review.googlesource.com/6554
Reviewed-by: Adam Langley <alangley@gmail.com>
We don't actually have an API to let you know if the value is legal to
interpret as a curve ID. (This was kind of a poor API. Oh well.) Also add tests
for key_exchange_info. I've intentionally left server-side plain RSA missing
for now because the SSL_PRIVATE_KEY_METHOD abstraction only gives you bytes and
it's probably better to tweak this API instead.
(key_exchange_info also wasn't populated on the server, though due to a
rebasing error, that fix ended up in the parent CL. Oh well.)
Change-Id: I74a322c8ad03f25b02059da7568c9e1a78419069
Reviewed-on: https://boringssl-review.googlesource.com/6783
Reviewed-by: Adam Langley <agl@google.com>
The new curve is not enabled by default.
As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an
SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It
also tidies up some of the curve code which kept converting back and
force between NIDs and curve IDs. Now everything transits as curve IDs
except for API entry points (SSL_set1_curves) which take NIDs. Those
convert immediately and act on curve IDs from then on.
Note that, like the Go implementation, this slightly tweaks the order of
operations. The client sees the server public key before sending its
own. To keep the abstraction simple, SSL_ECDH_METHOD expects to
generate a keypair before consuming the peer's public key. Instead, the
client handshake stashes the serialized peer public value and defers
parsing it until it comes time to send ClientKeyExchange. (This is
analogous to what it was doing before where it stashed the parsed peer
public value instead.)
It still uses TLS 1.2 terminology everywhere, but this abstraction should also
be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges.
(Accordingly, this abstraction intentionally does not handle parsing the
ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous
plain RSA or the authentication bits.)
BUG=571231
Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932
Reviewed-on: https://boringssl-review.googlesource.com/6780
Reviewed-by: Adam Langley <agl@google.com>
This injects an interface to abstract between elliptic.Curve and a
byte-oriented curve25519. The C implementation will follow a similar
strategy.
Note that this slightly tweaks the order of operations. The client sees
the server public key before sending its own. To keep the abstraction
simple, ecdhCurve expects to generate a keypair before consuming the
peer's public key. Instead, the client handshake stashes the serialized
peer public value and defers parsing it until it comes time to send
ClientKeyExchange. (This is analogous to what it was doing before where
it stashed the parsed peer public value instead.)
BUG=571231
Change-Id: I771bb9aee0dd6903d395c84ec4f2dd7b3e366c75
Reviewed-on: https://boringssl-review.googlesource.com/6777
Reviewed-by: Adam Langley <agl@google.com>
Hopefully this can be replaced with a standard library version later.
BUG=571231
Change-Id: I61ae1d9d057c6d9e1b92128042109758beccc7ff
Reviewed-on: https://boringssl-review.googlesource.com/6776
Reviewed-by: Adam Langley <agl@google.com>
We don't live in a workspace, but relative import paths exist, so we
don't have to modify the modules we bundle to avoid naming collisions.
Change-Id: Ie7c70dbc4bb0485421814d40b6a6bd5f140e1d29
Reviewed-on: https://boringssl-review.googlesource.com/6781
Reviewed-by: Adam Langley <agl@google.com>
It already wasn't in the default list and no one enables it. Remove it
altogether. (It's also gone from the current TLS 1.3 draft.)
Change-Id: I143d07d390d186252204df6bdb8ffd22649f80e3
Reviewed-on: https://boringssl-review.googlesource.com/6775
Reviewed-by: Adam Langley <agl@google.com>
Apple these days ships lldb without gdb. Teach runner how to launch it
too.
Change-Id: I25f845f84f1c87872a9e3bc4b7fe3e7344e8c1f7
Reviewed-on: https://boringssl-review.googlesource.com/6769
Reviewed-by: Adam Langley <agl@google.com>
Only ECDHE-based ciphers are implemented. To ease the transition, the
pre-standard cipher shares a name with the standard one. The cipher rule parser
is hacked up to match the name to both ciphers. From the perspective of the
cipher suite configuration language, there is only one cipher.
This does mean it is impossible to disable the old variant without a code
change, but this situation will be very short-lived, so this is fine.
Also take this opportunity to make the CK and TXT names align with convention.
Change-Id: Ie819819c55bce8ff58e533f1dbc8bef5af955c21
Reviewed-on: https://boringssl-review.googlesource.com/6686
Reviewed-by: Adam Langley <agl@google.com>
This will be used to test the C implementation against.
Change-Id: I2d396d27630937ea610144e381518eae76f78dab
Reviewed-on: https://boringssl-review.googlesource.com/6685
Reviewed-by: Adam Langley <agl@google.com>
In preparation for a Go implementation of the new TLS ciphers to test
against, implement the AEAD primitive.
Change-Id: I69b5b51257c3de16bdd36912ed2bc9d91ac853c8
Reviewed-on: https://boringssl-review.googlesource.com/6684
Reviewed-by: Adam Langley <agl@google.com>
In preparation for implementing the RFC 7539 variant to test against.
Change-Id: I0ce5e856906e00925ad1d849017f9e7fda087a8e
Reviewed-on: https://boringssl-review.googlesource.com/6683
Reviewed-by: Adam Langley <agl@google.com>
This uses ssl3_read_bytes for now. We still need to dismantle that
function and then invert the handshake state machine, but this gets
things closer to the right shape as an intermediate step and is a large
chunk in itself. It simplifies a lot of the CCS/handshake
synchronization as a lot of the invariants much more clearly follow from
the handshake itself.
Tests need to be adjusted since this changes some error codes. Now all
the CCS/Handshake checks fall through to the usual
SSL_R_UNEXPECTED_RECORD codepath. Most of what used to be a special-case
falls out naturally. (If half of Finished was in the same record as the
pre-CCS message, that part of the handshake record would have been left
unconsumed, so read_change_cipher_spec would have noticed, just like
read_app_data would have noticed.)
Change-Id: I15c7501afe523d5062f0e24a3b65f053008d87be
Reviewed-on: https://boringssl-review.googlesource.com/6642
Reviewed-by: Adam Langley <agl@google.com>
Sometimes BadRSAClientKeyExchange-1 fails with DATA_TOO_LARGE_FOR_MODULUS if
the corruption brings the ciphertext above the RSA modulus. Ensure this does
not happen.
Change-Id: I0d8ea6887dfcab946fdf5d38f5b196f5a927c4a9
Reviewed-on: https://boringssl-review.googlesource.com/6731
Reviewed-by: Adam Langley <agl@google.com>
Cover not just the wrong version, but also other mistakes.
Change-Id: I46f05a9a37b7e325adc19084d315a415777d3a46
Reviewed-on: https://boringssl-review.googlesource.com/6610
Reviewed-by: Adam Langley <agl@google.com>
I don't think we're ever going to manage to enforce this, and it doesn't
seem worth the trouble. We don't support application protocols which use
renegotiation outside of the HTTP/1.1 mid-stream client auth hack.
There, it's on the server to reject legacy renegotiations.
This removes the last of SSL_OP_ALL.
Change-Id: I996fdeaabf175b6facb4f687436549c0d3bb0042
Reviewed-on: https://boringssl-review.googlesource.com/6580
Reviewed-by: Adam Langley <agl@google.com>
RFC 5746 forbids a server from downgrading or upgrading
renegotiation_info support. Even with SSL_OP_LEGACY_SERVER_CONNECT set
(the default), we can still enforce a few things.
I do not believe this has practical consequences. The attack variant
where the server half is prefixed does not involve a renegotiation on
the client. The converse where the client sees the renegotiation and
prefix does, but we only support renego for the mid-stream HTTP/1.1
client auth hack, which doesn't do this. (And with triple-handshake,
HTTPS clients should be requiring the certificate be unchanged across
renego which makes this moot.)
Ultimately, an application which makes the mistake of using
renegotiation needs to be aware of what exactly that means and how to
handle connection state changing mid-stream. We make renego opt-in now,
so this is a tenable requirement.
(Also the legacy -> secure direction would have been caught by the
server anyway since we send a non-empty RI extension.)
Change-Id: I915965c342f8a9cf3a4b6b32f0a87a00c3df3559
Reviewed-on: https://boringssl-review.googlesource.com/6559
Reviewed-by: Adam Langley <agl@google.com>
This dates to SSLeay 0.8.0 (or earlier). The use counter sees virtually
no hits.
Change-Id: Iff4c8899d5cb0ba4afca113c66d15f1d980ffe41
Reviewed-on: https://boringssl-review.googlesource.com/6558
Reviewed-by: Adam Langley <agl@google.com>
This dates to SSLeay 0.9.0. The Internet seems to have completely
forgotten what "D5" is. (I can't find reference to it beyond
documentation of this quirk.) The use counter we added sees virtually no
hits.
Change-Id: I9781d401acb98ce3790b1b165fc257a6f5e9b155
Reviewed-on: https://boringssl-review.googlesource.com/6557
Reviewed-by: Adam Langley <agl@google.com>
yaSSL has a couple of bugs in their DH client implementation. This
change works around the worst of the two.
Firstly, they expect the the DH public value to be the same length as
the prime. This change pads the public value as needed to ensure this.
Secondly, although they handle the first byte of the shared key being
zero, they don't handle the case of the second, third, etc bytes being
zero. So whenever that happens the handshake fails. I don't think that
there's anything that we can do about that one.
Change-Id: I789c9e5739f19449473305d59fe5c3fb9b4a6167
Reviewed-on: https://boringssl-review.googlesource.com/6578
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
dh.c had a 10k-bit limit but it wasn't quite correctly enforced. However,
that's still 1.12s of jank on the IO thread, which is too long. Since the SSL
code consumes DHE groups from the network, it should be responsible for
enforcing what sanity it needs on them.
Costs of various bit lengths on 2013 Macbook Air:
1024 - 1.4ms
2048 - 14ms
3072 - 24ms
4096 - 55ms
5000 - 160ms
10000 - 1.12s
UMA says that DHE groups are 0.2% 4096-bit and otherwise are 5.5% 2048-bit and
94% 1024-bit and some noise. Set the limit to 4096-bit to be conservative,
although that's already quite a lot of jank.
BUG=554295
Change-Id: I8e167748a67e4e1adfb62d73dfff094abfa7d215
Reviewed-on: https://boringssl-review.googlesource.com/6464
Reviewed-by: Adam Langley <agl@google.com>
The current check has two problems:
- It only runs on the server, where there isn't a curve list at all. This was a
mistake in https://boringssl-review.googlesource.com/1843 which flipped it
from client-only to server-only.
- It only runs in TLS 1.2, so one could bypass it by just negotiating TLS 1.1.
Upstream added it as part of their Suite B mode, which requires 1.2.
Move it elsewhere. Though we do not check the entire chain, leaving that to the
certificate verifier, signatures made by the leaf certificate are made by the
SSL/TLS stack, so it's reasonable to check the curve as part of checking
suitability of a leaf.
Change-Id: I7c12f2a32ba946a20e9ba6c70eff23bebcb60bb2
Reviewed-on: https://boringssl-review.googlesource.com/6414
Reviewed-by: Adam Langley <agl@google.com>
This exposes the ServerKeyExchange signature hash type used in the most recent
handshake, for histogramming on the client.
BUG=549662
Change-Id: I8a4e00ac735b1ecd2c2df824112c3a0bc62332a7
Reviewed-on: https://boringssl-review.googlesource.com/6413
Reviewed-by: Adam Langley <agl@google.com>
This fixes a number of bugs with the original logic:
- If handshake messages are fragmented and writes need to be retried, frag_off
gets completely confused.
- The BIO_flush call didn't set rwstate, so it wasn't resumable at that point.
- The msg_callback call gets garbage because the fragment header would get
scribbled over the handshake buffer.
The original logic was also extremely confusing with how it handles init_off.
(init_off gets rewound to make room for the fragment header. Depending on
where you pause, resuming may or may not have already been rewound.)
For simplicity, just allocate a new buffer to assemble the fragment in and
avoid clobbering the old one. I don't think it's worth the complexity to
optimize that. If we want to optimize this sort of thing, not clobbering seems
better anyway because the message may need to be retransmitted. We could avoid
doing a copy when buffering the outgoing message for retransmission later.
We do still need to track how far we are in sending the current message via
init_off, so I haven't opted to disconnect this function from
init_{buf,off,num} yet.
Test the fix to the retry + fragment case by having the splitHandshake option
to the state machine tests, in DTLS, also clamp the MTU to force handshake
fragmentation.
Change-Id: I66f634d6c752ea63649db8ed2f898f9cc2b13908
Reviewed-on: https://boringssl-review.googlesource.com/6421
Reviewed-by: Adam Langley <agl@google.com>
This option causes clients to ignore HelloRequest messages completely.
This can be suitable in cases where a server tries to perform concurrent
application data and handshake flow, e.g. because they are trying to
“renew” symmetric keys.
Change-Id: I2779f7eff30d82163f2c34a625ec91dc34fab548
Reviewed-on: https://boringssl-review.googlesource.com/6431
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
Although the DTLS transport layer logic drops failed writes on the floor, it is
actually set up to work correctly. If an SSL_write fails at the transport,
dropping the buffer is fine. Arguably it works better than in TLS because we
don't have the weird "half-committed to data" behavior. Likewise, the handshake
keeps track of how far its gotten and resumes the message at the right point.
This broke when the buffering logic was rewritten because I didn't understand
what the DTLS code was doing. The one thing that doesn't work as one might
expect is non-fatal write errors during rexmit are not recoverable. The next
timeout must fire before we try again.
This code is quite badly sprinkled in here, so add tests to guard it against
future turbulence. Because of the rexmit issues, the tests need some hacks
around calls which may trigger them. It also changes the Go DTLS implementation
from being completely strict about sequence numbers to only requiring they be
monotonic.
The tests also revealed another bug. This one seems to be upstream's fault, not
mine. The logic to reset the handshake hash on the second ClientHello (in the
HelloVerifyRequest case) was a little overenthusiastic and breaks if the
ClientHello took multiple tries to send.
Change-Id: I9b38b93fff7ae62faf8e36c4beaf848850b3f4b9
Reviewed-on: https://boringssl-review.googlesource.com/6417
Reviewed-by: Adam Langley <agl@google.com>
It's somewhat annoying to have to parse out the packetAdaptor mini-language.
Actually seeing those is only useful when debugging the adaptor itself, rather
than DTLS. Switch the order of the two middleware bits and add an escape hatch
to log the funny opcodes.
Change-Id: I249c45928a76b747d69f3ab972ea4d31e0680a62
Reviewed-on: https://boringssl-review.googlesource.com/6416
Reviewed-by: Adam Langley <agl@google.com>
Right whether NPN is advertised can only be configured globally on the SSL_CTX.
Rather than adding two pointers to each SSL*, add an options bit to disable it
so we may plumb in a field trial to disable NPN.
Chromium wants to be able to route a bit in to disable NPN, but it uses SSL_CTX
incorrectly and has a global one, so it can't disconnect the callback. (That
really needs to get fixed. Although it's not clear this necessarily wants to be
lifted up to SSL_CTX as far as Chromium's SSLClientSocket is concerned since
NPN doesn't interact with the session cache.)
BUG=526713
Change-Id: I49c86828b963eb341c6ea6a442557b7dfa190ed3
Reviewed-on: https://boringssl-review.googlesource.com/6351
Reviewed-by: Adam Langley <alangley@gmail.com>
RFC 5077 explicitly allows the server to change its mind and send no
ticket by sending an empty NewSessionTicket. See also upstream's
21b538d616b388fa0ce64ef54da3504253895cf8.
CBS_stow handles this case somewhat, so we won't get confused about
malloc(0) as upstream did. But we'll still fill in a bogus SHA-256
session ID, cache the session, and send a ClientHello with bogus session
ID but empty ticket extension. (The session ID field changes meaning
significantly when the ticket is or isn't empty. Non-empty means "ignore
the session ID, but echo if it resuming" while empty means "I support
tickets, but am offering this session ID".
The other behavior change is that a server which changes its mind on a
resumption handshake will no longer override the client's session cache
with a ticket-less session.
(This is kind of silly. Given that we don't get completely confused due
to CBS_stow, it might not be worth bothering with the rest. Mostly it
bugged me that we send an indicator session ID with no ticket.)
Change-Id: Id6b5bde1fe51aa3e1f453a948e59bfd1e2502db6
Reviewed-on: https://boringssl-review.googlesource.com/6340
Reviewed-by: Adam Langley <alangley@gmail.com>
Add a slightly richer API. Notably, one can configure ssl_renegotiate_once to
only accept the first renego.
Also, this API doesn't repeat the mistake I made with
SSL_set_reject_peer_renegotiations which is super-confusing with the negation.
Change-Id: I7eb5d534e3e6c553b641793f4677fe5a56451c71
Reviewed-on: https://boringssl-review.googlesource.com/6221
Reviewed-by: Adam Langley <agl@google.com>
SSL 3.0 used to have a nice and simple rule around extensions. They don't
exist. And then RFC 5746 came along and made this all extremely confusing.
In an SSL 3.0 server, rather than blocking ServerHello extension
emission when renegotiation_info is missing, ignore all ClientHello
extensions but renegotiation_info. This avoids a mismatch between local
state and the extensions with emit.
Notably if, for some reason, a ClientHello includes the session_ticket
extension, does NOT include renegotiation_info or the SCSV, and yet the
client or server are decrepit enough to negotiate SSL 3.0, the
connection will fail due to unexpected NewSessionTicket message.
See https://crbug.com/425979#c9 for a discussion of something similar
that came up in diagnosing https://poodle.io/'s buggy POODLE check.
This is analogous to upstream's
5a3d8eebb7667b32af0ccc3f12f314df6809d32d.
(Not supporting renego as a server in any form anyway, we may as well
completely ignore extensions, but then our extensions callbacks can't
assume the parse hooks are always called. This way the various NULL
handlers still function.)
Change-Id: Ie689a0e9ffb0369ef7a20ab4231005e87f32d5f8
Reviewed-on: https://boringssl-review.googlesource.com/6180
Reviewed-by: Adam Langley <agl@google.com>
This change makes the runner tests (in ssl/test/runner) act like a
normal Go test rather than being a Go binary. This better aligns with
some internal tools.
Thus, from this point onwards, one has to run the runner tests with `go
test` rather than `go run` or `go build && ./runner`.
This will break the bots.
Change-Id: Idd72c31e8e0c2b7ed9939dacd3b801dbd31710dd
Reviewed-on: https://boringssl-review.googlesource.com/6009
Reviewed-by: Matt Braithwaite <mab@google.com>
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
Allow configuring digest preferences for the private key. Some
smartcards have limited support for signing digests, notably Windows
CAPI keys and old Estonian smartcards. Chromium used the supports_digest
hook in SSL_PRIVATE_KEY_METHOD to limit such keys to SHA1. However,
detecting those keys was a heuristic, so some SHA256-capable keys
authenticating to SHA256-only servers regressed in the switch to
BoringSSL. Replace this mechanism with an API to configure digest
preference order. This way heuristically-detected SHA1-only keys may be
configured by Chromium as SHA1-preferring rather than SHA1-requiring.
In doing so, clean up the shared_sigalgs machinery somewhat.
BUG=468076
Change-Id: I996a2df213ae4d8b4062f0ab85b15262ca26f3c6
Reviewed-on: https://boringssl-review.googlesource.com/5755
Reviewed-by: Adam Langley <agl@google.com>
SCT and OCSP are part of the session data and as such shouldn't be sent
again to the client when resuming.
Change-Id: Iaee3a3c4c167ea34b91504929e38aadee37da572
Reviewed-on: https://boringssl-review.googlesource.com/5900
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
Last set of changes didn't do that.
Change-Id: Iae24e75103529ce4d50099c5cbfbcef0e10ba663
Reviewed-on: https://boringssl-review.googlesource.com/5871
Reviewed-by: Adam Langley <agl@google.com>
Change-Id: Ifa44fef160fc9d67771eed165f8fc277f28a0222
Reviewed-on: https://boringssl-review.googlesource.com/5840
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
Applications may require the stapled OCSP response in order to verify
the certificate within the verification callback.
Change-Id: I8002e527f90c3ce7b6a66e3203c0a68371aac5ec
Reviewed-on: https://boringssl-review.googlesource.com/5730
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
This change adds the ability to configure ciphers specifically for
TLS ≥ 1.0. This compliments the existing ability to specify ciphers
for TLS ≥ 1.1.
This is useful because TLS 1.0 is the first version not to suffer from
POODLE. (Assuming that it's implemented correctly[1].) Thus one might
wish to reserve RC4 solely for SSLv3.
[1] https://www.imperialviolet.org/2014/12/08/poodleagain.html
Change-Id: I774d5336fead48f03d8a0a3cf80c369692ee60df
Reviewed-on: https://boringssl-review.googlesource.com/5793
Reviewed-by: Adam Langley <agl@google.com>
setup_key_block is called when the first CCS resolves, but for resumptions this
is the incoming CCS (see ssl3_do_change_cipher_spec). Rather than set
need_record_splitting there, it should be set in the write case of
tls1_change_cipher_state.
This fixes a crash from the new record layer code in resumption when
record-splitting is enabled. Tweak the record-splitting tests to cover this
case.
This also fixes a bug where renego from a cipher which does require record
splitting to one which doesn't continues splitting. Since version switches are
not allowed, this can only happen after a renego from CBC to RC4.
Change-Id: Ie4e1b91282b10f13887b51d1199f76be4fbf09ad
Reviewed-on: https://boringssl-review.googlesource.com/5787
Reviewed-by: Adam Langley <agl@google.com>
Note that DTLS treats oversized ciphertexts different from everything else.
Change-Id: I71cba69ebce0debdfc96a7fdeb2666252e8d28ed
Reviewed-on: https://boringssl-review.googlesource.com/5786
Reviewed-by: Adam Langley <agl@google.com>
If the two extensions select different next protocols (quite possible since one
is server-selected and the other is client-selected), things will break. This
matches the behavior of NSS (Firefox) and Go.
Change-Id: Ie1da97bf062b91a370c85c12bc61423220a22f36
Reviewed-on: https://boringssl-review.googlesource.com/5780
Reviewed-by: Adam Langley <agl@google.com>
The maximum buffer size computation wasn't quite done right in
ssl_buffer.c, so we were failing with BUFFER_TOO_SMALL for sufficiently
large records. Fix this and, as penance, add 103 tests.
(Test that we can receive maximum-size records in all cipher suites.
Also test SSL_OP_MICROSOFT_BIG_SSLV3_BUFFER while I'm here.)
BUG=526998
Change-Id: I714c16dda2ed13f49d8e6cd1b48adc5a8491f43c
Reviewed-on: https://boringssl-review.googlesource.com/5785
Reviewed-by: Adam Langley <agl@google.com>
Now that it even works at all (type = 0 bug aside), add tests for it.
Test both close_notify being received before and after SSL_shutdown is
called. In the latter case, have the peer send some junk to be ignored
to test that works.
Also test that SSL_shutdown fails on unclean shutdown and that quiet
shutdowns ignore it.
BUG=526437
Change-Id: Iff13b08feb03e82f21ecab0c66d5f85aec256137
Reviewed-on: https://boringssl-review.googlesource.com/5769
Reviewed-by: Adam Langley <agl@google.com>
