When we made setReadTrafficSecret send an alert when there are pending
handshake messages, we introduced a deadlock when the client sends
multiple key update messages that request a response, as handleKeyUpdate
will lock the mutex, and defer the unlocking until the end of the
function, but setReadTrafficSecret called sendAlert in the failure case,
which also tries to lock the mutex.
Add an argument to setReadTrafficSecret which lets the caller indicate
if the mutex is already locked, and if so, call sendAlertLocked instead
of sendAlert.
Thanks to Jakub Ciolek for reporting this issue.
Fixes#78334
Fixes CVE-2026-32283
Change-Id: Id8e56974233c910e0d66ba96eafbd2ea57832610
Reviewed-on: https://go-internal-review.googlesource.com/c/go/+/3881
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-by: Nicholas Husin <husin@google.com>
Reviewed-on: https://go-review.googlesource.com/c/go/+/763767
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: David Chase <drchase@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Jakub Ciolek <jakub@ciolek.dev>
Go 1.21 introduced bytes.Buffer.AvailableBuffer. This mechanism lets us
write directly into a Buffer's backing memory while also implementing
custom logic without going through interfaces.
atLeastReader caused an allocation on every read. On workloads with very
small reads, these allocations can add up to substantial overhead. By
using Buffer.AvailableBytes(), we can avoid these allocations and an
interface indirection in the Read fast path.
Fixes#58249.
Change-Id: Icf26ec1dd7ef88154c47356ef9c26a516a6a6964
Reviewed-on: https://go-review.googlesource.com/c/go/+/739980
Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Junyang Shao <shaojunyang@google.com>
Reviewed-by: Daniel McCarney <daniel@binaryparadox.net>
Methods on QUICConn are synchronous:
The connection state is expected to change only in reaction
to a user calling a QUICConn method, and the state change
should finish completely before the method returns.
The connection context provided to QUICConn.Start violates
this model, because canceling the context causes an
asynchronous state change.
Prior to CL 719040, this caused no problems because canceling
the context did not cause any user-visible state changes.
In particular, canceling the context did not cause any new
events to be immediately returned by QUICConn.NextEvent.
CL 719040 introduced a new error event. Now, canceling a
QUICConn's context causes a new connection event to be
generated.
Receiving this event causes a data race visible to the
race detector, but the core problem is not the data race
itself: It's that an asynchronous event (canceling the
connection context) causes an change to the connection
events.
Fix this race by reworking the handling of QUICConn
context cancellation a bit. We no longer react to
cancellation while control of the connection lies
with the user. We only process cancellation as
part of a user call, such as QUICConn.Close
or QUICConn.HandleData.
Fixes#77274
Change-Id: If2e0f73618c4852114e0931b6bd0cb0b6a6a6964
Reviewed-on: https://go-review.googlesource.com/c/go/+/742561
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Damien Neil <dneil@google.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
For TLS 1.3, after procesesing the server/client hello, if there isn't a
CCS message, reject the trailing messages which were appended to the
hello messages. This prevents an on-path attacker from injecting
plaintext messages into the handshake.
Additionally, check that we don't have any buffered messages before we
switch the read traffic secret regardless, since any buffered messages
would have been under an old key which is no longer appropriate.
We also invert the ordering of setting the read/write secrets so that if
we fail when changing the read secret we send the alert using the
correct write secret.
Fixes#76443
Fixes CVE-2025-61730
Change-Id: If6ba8ad16f48d5cd5db5574824062ad4244a5b52
Reviewed-on: https://go-review.googlesource.com/c/go/+/724120
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Daniel McCarney <daniel@binaryparadox.net>
Reviewed-by: Coia Prant <coiaprant@gmail.com>
This commit adds fields to the ClientHelloInfo and ConnectionState
structures to represent hello retry request state information.
ClientHelloInfo gains a new HelloRetryRequest bool field that indicates
if the client hello was sent in response to a TLS 1.3 hello retry
request message previously emitted by the server.
ConnectionState gains a new HelloRetryRequest bool field that indicates
(depending on the connection role) whether the client received a TLS 1.3
hello retry request message from the server, or whether the server sent
such a message to a client.
Fixes#74425
Change-Id: Ic1a5290b8a4ba1568da1d2c2cf9f148150955fa5
Reviewed-on: https://go-review.googlesource.com/c/go/+/717440
Reviewed-by: Roland Shoemaker <roland@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Auto-Submit: Daniel McCarney <daniel@binaryparadox.net>
Previously if instances of the handshakeMessage interface returned false
from unmarshal(), indicating an umarshalling error, the crypto/tls
package would emit an unexpected_message alert. This commit changes to
use a decode_error alert for this condition instead.
The usage-pattern of the handshakeMessage interface is that we switch on
the message type, invoke a specific concrete handshakeMessage type's
unmarshal function, and then return it to the caller on success. At this
point the caller looks at the message type and can determine if the
message was unexpected or not. If it was unexpected, the call-sites emit
the correct error for that case. Only the caller knows the current
protocol state and allowed message types, not the generic handshake
decoding logic.
With the above in mind, if we find that within the unmarshal logic for
a specific message type that the data we have in hand doesn't match the
protocol syntax we should emit a decode_error. An unexpected_message
error isn't appropriate because we don't yet know if the message is
unexpected or not, only that the message can't be decoded based on the
spec's syntax for the type the message claimed to be.
Notably one unit test, TestQUICPostHandshakeKeyUpdate, had to have its
test data adjusted because it was previously not testing the right
thing: it was double-encoding the type & length prefix data for a key
update message and expecting the QUIC logic to reject it as an
inappropriate post-handshake message. In reality it was being rejected
sooner as an invalid key update message from the double-encoding and
this was masked by the previous alert for this condition matching the
expected alert.
Finally, changing our alert allows enabling a handful of BoGo tests
related to duplicate extensions of the form
"DuplicateExtension[Server|Client]-TLS-[TLS1|TLS11|TLS12|TLS13]". One
test remains skipped (DuplicateExtensionClient-TLS-TLS13), as it
requires additional follow-up.
Updates #72006
Change-Id: I27a2cd231e4b8762b0d9e2dbd3d8ddd5b87fd5d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/673738
Reviewed-by: Roland Shoemaker <roland@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Auto-Submit: Daniel McCarney <daniel@binaryparadox.net>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Uses the new weak package to replace the existing custom intern cache
with a map of weak.Pointers instead. This simplifies the cache, and
means we don't need to store a slice of handles on the Conn anymore.
Change-Id: I5c2bf6ef35fac4255e140e184f4e48574b34174c
Reviewed-on: https://go-review.googlesource.com/c/go/+/644176
TryBot-Bypass: Roland Shoemaker <roland@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Roland Shoemaker <roland@golang.org>
This required adding a new field to SessionState for TLS 1.0–1.2, since
the key exchange is not repeated on resumption. The additional field is
unfortunately not backwards compatible because current Go versions check
that the encoding has no extra data at the end, but will cause
cross-version tickets to be ignored. Relaxed that so we can add fields
in a backwards compatible way the next time.
For the cipher suite, we check that the session's is still acceptable
per the Config. That would arguably make sense here, too: if a Config
for example requires PQ, we should reject resumptions of connections
that didn't use PQ. However, that only applies to pre-TLS 1.3
connections, since in TLS 1.3 we always do a fresh key exchange on
resumption. Since PQ is the only main differentiator between key
exchanges (aside from off-by-default non-PFS RSA, which are controlled
by the cipher suite in TLS 1.0–1.2) and it's PQ-only, we can skip that
check.
Fixes#67516
Change-Id: I6a6a465681a6292edf66c7b8df8f4aba4171a76b
Reviewed-on: https://go-review.googlesource.com/c/go/+/653315
Reviewed-by: David Chase <drchase@google.com>
Auto-Submit: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Daniel McCarney <daniel@binaryparadox.net>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
When encountering alertUserCanceled in a TLS 1.3 handshake, ignore the
alert and retry reading a record. This matches existing logic for how
TLS 1.2 alertLevelWarning alerts are handled.
For broader context, TLS 1.3 removed warning-level alerts except for
alertUserCanceled (RFC 8446, § 6.1). Since at least one major
implementation (https://bugs.openjdk.org/browse/JDK-8323517)
misuses this alert, many TLS stacks now ignore it outright when seen in
a TLS 1.3 handshake (e.g. BoringSSL, NSS, Rustls).
With the crypto/tls behaviour changed to match peer implementations we
can now enable the "SendUserCanceledAlerts-TLS13" BoGo test.
"SendUserCanceledAlerts-TooMany-TLS13" remains ignored, because like
"SendWarningAlerts*" fixing the test requires some general spam
protocol message enhancements be done first.
Updates #72006
Change-Id: I570c1fa674b5a4760836c514d35ee17f746fe28d
Reviewed-on: https://go-review.googlesource.com/c/go/+/650716
Reviewed-by: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
This CL adds a (very opinionated) client-side ECH implementation.
In particular, if a user configures a ECHConfigList, by setting the
Config.EncryptedClientHelloConfigList, but we determine that none of
the configs are appropriate, we will not fallback to plaintext SNI, and
will instead return an error. It is then up to the user to decide if
they wish to fallback to plaintext themselves (by removing the config
list).
Additionally if Config.EncryptedClientHelloConfigList is provided, we
will not offer TLS support lower than 1.3, since negotiating any other
version, while offering ECH, is a hard error anyway. Similarly, if a
user wishes to fallback to plaintext SNI by using 1.2, they may do so
by removing the config list.
With regard to PSK GREASE, we match the boringssl behavior, which does
not include PSK identities/binders in the outer hello when doing ECH.
If the server rejects ECH, we will return a ECHRejectionError error,
which, if provided by the server, will contain a ECHConfigList in the
RetryConfigList field containing configs that should be used if the user
wishes to retry. It is up to the user to replace their existing
Config.EncryptedClientHelloConfigList with the retry config list.
Fixes#63369
Cq-Include-Trybots: luci.golang.try:gotip-linux-amd64-longtest
Change-Id: I9bc373c044064221a647a388ac61624efd6bbdbf
Reviewed-on: https://go-review.googlesource.com/c/go/+/578575
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Auto-Submit: Roland Shoemaker <roland@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
I initially thought the logic was broken, but writing the test I
realized it was actually very clever (derogative). It was relying on the
outer loop continuing after a supported match without a key share,
allowing a later key share to override it (but not a later supported
match because of the "if selectedGroup != 0 { continue }").
Replaced the clever loop with two hopefully more understandable loops,
and added a test (which was already passing).
We were however not checking that the selected group is in the supported
list if we found it in key shares first. (This was only a MAY.) Fixed.
Fixes#65686
Change-Id: I09ea44f90167ffa36809deb78255ed039a217b6d
Reviewed-on: https://go-review.googlesource.com/c/go/+/586655
Reviewed-by: Roland Shoemaker <roland@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Filippo Valsorda <filippo@golang.org>
All OpenSSL tests now test operation with EMS. To test a handshake
*without* EMS we need to pass -Options=-ExtendedMasterSecret which is
only available in OpenSSL 3.1, which breaks a number of other tests.
Updates #43922
Change-Id: Ib9ac79a1d03fab6bfba5fe9cd66689cff661cda7
Reviewed-on: https://go-review.googlesource.com/c/go/+/497376
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
Auto-Submit: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Message marshalling makes use of BytesOrPanic a lot, under the
assumption that it will never panic. This assumption was incorrect, and
specifically crafted handshakes could trigger panics. Rather than just
surgically replacing the usages of BytesOrPanic in paths that could
panic, replace all usages of it with proper error returns in case there
are other ways of triggering panics which we didn't find.
In one specific case, the tree routed by expandLabel, we replace the
usage of BytesOrPanic, but retain a panic. This function already
explicitly panicked elsewhere, and returning an error from it becomes
rather painful because it requires changing a large number of APIs.
The marshalling is unlikely to ever panic, as the inputs are all either
fixed length, or already limited to the sizes required. If it were to
panic, it'd likely only be during development. A close inspection shows
no paths for a user to cause a panic currently.
This patches ends up being rather large, since it requires routing
errors back through functions which previously had no error returns.
Where possible I've tried to use helpers that reduce the verbosity
of frequently repeated stanzas, and to make the diffs as minimal as
possible.
Thanks to Marten Seemann for reporting this issue.
Fixes#58001
Fixes CVE-2022-41724
Change-Id: Ieb55867ef0a3e1e867b33f09421932510cb58851
Reviewed-on: https://team-review.git.corp.google.com/c/golang/go-private/+/1679436
Reviewed-by: Julie Qiu <julieqiu@google.com>
TryBot-Result: Security TryBots <security-trybots@go-security-trybots.iam.gserviceaccount.com>
Run-TryBot: Roland Shoemaker <bracewell@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-on: https://go-review.googlesource.com/c/go/+/468125
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Michael Pratt <mpratt@google.com>
yuhan6665