Motivation:
To reduce latency and RTTs we should use TLS False Start when jdkCompatibilityMode is not required and its supported
Modifications:
Use SSL_MODE_ENABLE_FALSE_START when jdkCompatibilityMode is false
Result:
Less RTTs and so lower latency when TLS False Start is supported
Motivation:
We need limit the number of queries done to resolve unresolved nameserver as otherwise we may never fail the resolve when there is a missconfigured dns server that lead to a "resolve loop".
Modifications:
- When resolve unresolved nameservers ensure we use the allowedQueries as max upper limit for queries so it will eventually fail
- Add unit tests
Result:
No more possibility to fail in a loop while resolve. This is related to https://github.com/netty/netty/issues/10420
Motivation:
AbstractDiskHttpData#getChunk opens and closes fileChannel every time when it is invoked,
as a result the uploaded file is corrupted. This is a regression caused by #10270.
Modifications:
- Close the fileChannel only if everything was read or an exception is thrown
- Add unit test
Result:
AbstractDiskHttpData#getChunk closes fileChannel only if everything was read or an exception is thrown
Motivation:
We should include TLSv1.3 ciphers as well as recommented ciphers these days for HTTP/2. That is especially true as Java supports TLSv1.3 these days out of the box
Modifications:
- Add TLSv1.3 ciphers that are recommended by mozilla as was for HTTP/2
- Add unit test
Result:
Include TLSv1.3 ciphers as well
Motivation:
DnsAddressResolverGroup allows to be constructed with a DnsNameResolverBuilder and so should respect its configured EventLoop.
Modifications:
- Correctly respect the configured EventLoop
- Ensure there are no thread-issues by calling copy()
- Add unit tests
Result:
Fixes https://github.com/netty/netty/issues/10460
Motivation:
When we were using the netty http protocol, OOM occurred, this problem has been in 4.1.51.Final Fix [# 10424](https://github.com/netty/netty/issues/10424), even if OOM is up, the service will still receive new connection events, will occur again OOM and eventually cause the connection not to be released.
code `byteBuf = allocHandle.allocate(allocator);`
Modification:
I fail to create buffer when I try to receive new data, i determine if it is OOM then the close read event releases the connection.
```java
if (close || cause instanceof OutOfMemoryError || cause instanceof IOException) {
closeOnRead(pipeline);
}
```
Result:
Fixes # [10434](https://github.com/netty/netty/issues/10434).
Motivation:
Setting a dependency on the connection is normal and permitted; streams
actually default to depending on the connection. Using a PRIORITY frame
with a dependency on the connection could reset a previous PRIORITY,
change the relative weight, or make all streams dependent on one stream.
The previous code was disallowing these usages as it considered
depending on the connection to be a validation failure.
Modifications:
Loosen validation check to also allow depending on the connection. Fix
error message when the validation check fails.
Result:
Setting a dependency on connection would be permitted. Fixes#10416
Motivation:
How we init our static fields in Conscrypt was kind of error-prone and may even lead to NPE later on when methods were invoked out of order.
Modifications:
- Move all the init code to a static block
- Remove static field which is not needed anymore
Result:
Cleanup and also fixes https://github.com/netty/netty/issues/10413
Motivation:
The executor chooser should pluck executors in round-robin, but at the 32-bit
overflow boundary, the round-robin sequence was disrupted when the number of
executors are not a power of 2.
Modification:
Changed the index counter from a 32-bit to a 64-bit long. The overflow bug is
still technically there, but it now takes so long to reach that it will never
happen in practice. For example, 2^63 nanoseconds is almost 300 years.
Result:
The round-robin behaviour for all EventExecutorChoosers is now preserved in
practice.
This fixes#10423.
Motivation:
Recent Intellij versions are starting to anticipate
future versions of Java that include a
`java.lang.Record` class, and the Intellij compiler
gets confused by the `Record` class in our
`ResorceLeakDetector`.
Modification:
Rename our `Record` class to `TracerRecord`.
This matches what the class is doing, while avoiding
any future name clashes.
Result:
Intellij can now compile the project again, even when
configured to use a future (snapshot or early access)
version of Java.
Motivation:
The feature test macros did not work as expected on MacOS, so we ended up compiling for the GNU
variant which resulted in compilation errors.
Modification:
Add `__APPLE__` as another indicator to use the XSI variant of `strerror_r`.
Result:
The project now once again compiles on MacOS.
Motivation:
We previously relied on `strerror`, but this function is unfortunately not thread-safe.
Modification:
The use of `strerror` has been changed to `strerror_r`, which is thread-safe.
This function has a more complicated API, and has portability concerns that needs to be handled.
This accounts for the relatively large increase in lines of code.
Result:
Error messages from JNI are now always generated in a thread-safe way.
Motivation:
Even if the system does not support ipv6 we should try to use it if the user explicit pass an Inet6Address. This way we ensure we fail and not try to convert this to an ipv4 address internally.
This incorrect behavior was introduced by 70731bfa7e
Modifications:
If the user explicit passed an Inet6Address we force the usage of ipv6
Result:
Fixes https://github.com/netty/netty/issues/10402
Motiviation:
When TLSv1.3 was introduced almost 2 years ago, it was decided to disable it by default, even when it's supported by the underlying TLS implementation.
TLSv13 is pretty stable now in Java (out of the box in Java 11, OpenJSSE for Java 8, BoringSSL and OpenSSL) and may be enabled by default.
Modifications:
Ensure TLSv13 is enabled by default when the underyling JDK SSLEngine implementation enables it as well
Result:
TLSv1.3 is now enabled by default, so users don't have to explicitly enable it.
Co-authored-by: Stephane Landelle <slandelle@gatling.io>
Motivation:
AlgorithmId.sha256WithRSAEncryption_oid was removed in JDK15 and later so we should not depend on it as otherwise we will see compilation errors
Modifications:
Replace AlgorithmId.sha256WithRSAEncryption_oid usage with specify the OID directly
Result:
Compiles on JDK15+
Motivation:
JCTools 3.1.0 is out and includes several fixes, see https://github.com/JCTools/JCTools/releases/tag/v3.1.0
Modification:
Upgrade jctools-core version in pom.xml
Result:
Netty ships latest version of jctools.
Motivation:
SSLEngineImpl.unwrap(...) may call FileInputStream.read(...) internally when TLS1.3 is used. This will cause an BlockingOperationError when BlockHound is enabled.
For more details see https://mail.openjdk.java.net/pipermail/security-dev/2020-August/022271.html
Modifications:
- Add whitelist entry to BlockHound config for now
- Update NettyBlockHoundIntegrationTest to include testing for this workaround
Result:
No BlockingOperationError when TLS1.3 is used with JDK SSL implementation and BlockHound is enabled
Motivation:
Regression appeared after making changes in fix#10360 .
The main problem here that `buffer.getBytes(buffer.readerIndex(), fileChannel, fileChannel.position(), localsize)`
doesn't change channel position after writes.
Modification:
Manually set position according to the written bytes.
Result:
Fixes#10449 .
Motivation:
If a request to open a new h2 stream was made from outside of the
EventLoop it will be scheduled for future execution on the EventLoop.
However, during the time before the `open0` task will be executed the
parent channel may already be closed. As the result,
`Http2MultiplexHandler#newOutboundStream()` will throw an
`IllegalStateException` with the message that is hard to
interpret correctly for this use-case: "Http2FrameCodec not found. Has
the handler been added to a pipeline?".
Modifications:
- Check that the parent h2 `Channel` is still active before creating a
new stream when `open0` task is picked up by EventLoop;
Result:
Users see a correct `ClosedChannelException` in case the parent h2
`Channel` was closed concurrently with a request for a new stream.
Motivation:
Replace Class.getClassLoader with io.netty.util.internal.PlatformDependent.getClassLoader in Openssl so it also works when a SecurityManager is in place
Modification:
Replace Class.getClassLoader with io.netty.util.internal.PlatformDependent.getClassLoader in Openssl
Result:
No issues when a SecurityManager is in place
Motivation:
`Http2StreamChannelBootstrap#open0` invokes
`Http2MultiplexHandler#newOutboundStream()` which may throw an
`IllegalStateException`. In this case, it will never complete
the passed promise.
Modifications:
- `try-catch` all invocations of `newOutboundStream()` and fail
promise in case of any exception;
Result:
New H2 stream promise always completes.
Motivation:
GraalVM's native images built with native-image tool do not support Unsafe.staticFieldOffset() method (at least, currently). If an application using Netty (and causing initialization of io.netty.util.internal.PlatformDependent0 class) is built into a native image and run, this results in the following error thrown during initialization:
Exception in thread "main" com.oracle.svm.core.jdk.UnsupportedFeatureError: Unsupported method of Unsafe
at com.oracle.svm.core.util.VMError.unsupportedFeature(VMError.java:86)
at jdk.internal.misc.Unsafe.staticFieldOffset(Unsafe.java:230)
at sun.misc.Unsafe.staticFieldOffset(Unsafe.java:662)
at io.netty.util.internal.PlatformDependent0$5.run(PlatformDependent0.java:294)
at java.security.AccessController.doPrivileged(AccessController.java:83)
at io.netty.util.internal.PlatformDependent0.<clinit>(PlatformDependent0.java:279)
This seems to be the reason of the behavior described in #10051.
Modification:
The idea of this commit is to only invoke Unsafe.staticFieldOffset() is we are not in a native image; if we are, behave like if we could not find the field at all.
GraalDetector is borrowed from Spring framework.
Result:
Fixes#10051
Motivation:
If we cancel the returned future of resolve query, we may get LEAK. Try to release the ByteBuf if netty can't pass the DnsRawRecord to the caller.
Modification:
Using debug mode I saw there are two places that don't handle trySuccess with release. Try to release there.
Result:
Fixes#10447.
Motivation:
`DefaultHttp2FrameStream.stream` is not used outside of its class and
therefore can be private.
Modifications:
- Make `DefaultHttp2FrameStream.stream` private;
Result:
Correct visibility scoping for `DefaultHttp2FrameStream.stream`.
Motivation:
This request only has headers frame, it should set endOfStream flag, or
it will never get a response.
Modifications:
Set endOfStream=true in header frame.
Result:
Http2FrameClient can get a response now.
Motivation:
IntelliJ IDEA may generate a local folder `.shelf/` for version control.
For more information, see
https://www.jetbrains.com/help/idea/shelving-and-unshelving-changes.html
Modifications:
- Add `.shelf/` folder to the `.gitignore` file;
Result:
IntelliJ IDEA's `.shelf/` folder is ignored by git.
Motivation:
From the recent benchmark using gRPC-Java based on Netty's HTTP2, it appears that it prefers `ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256` over `ECDHE_ECDSA_WITH_AES_128_GCM_SHA256` since it uses the Netty HTTPS Cipher list as is. Both are considered safe but `ECDHE_ECDSA_WITH_AES_128_GCM_SHA256` has a good chance to get more optimized implementation. (e.g. AES-NI) When running both on GCP Intel Haswell VM, `TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256` spent 3x CPU time than `TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256`. (Note that this VM supports AES-NI)
From the cipher suites listed on `Intermediate compatibility (recommended)` of [Security/Server Side TLS](https://wiki.mozilla.org/Security/Server_Side_TLS#Modern_compatibility), they have a cipher preference which is aligned with this PR.
```
0x13,0x01 - TLS_AES_128_GCM_SHA256 TLSv1.3 Kx=any Au=any Enc=AESGCM(128) Mac=AEAD
0x13,0x02 - TLS_AES_256_GCM_SHA384 TLSv1.3 Kx=any Au=any Enc=AESGCM(256) Mac=AEAD
0x13,0x03 - TLS_CHACHA20_POLY1305_SHA256 TLSv1.3 Kx=any Au=any Enc=CHACHA20/POLY1305(256) Mac=AEAD
0xC0,0x2B - ECDHE-ECDSA-AES128-GCM-SHA256 TLSv1.2 Kx=ECDH Au=ECDSA Enc=AESGCM(128) Mac=AEAD
0xC0,0x2F - ECDHE-RSA-AES128-GCM-SHA256 TLSv1.2 Kx=ECDH Au=RSA Enc=AESGCM(128) Mac=AEAD
0xC0,0x2C - ECDHE-ECDSA-AES256-GCM-SHA384 TLSv1.2 Kx=ECDH Au=ECDSA Enc=AESGCM(256) Mac=AEAD
0xC0,0x30 - ECDHE-RSA-AES256-GCM-SHA384 TLSv1.2 Kx=ECDH Au=RSA Enc=AESGCM(256) Mac=AEAD
0xCC,0xA9 - ECDHE-ECDSA-CHACHA20-POLY1305 TLSv1.2 Kx=ECDH Au=ECDSA Enc=CHACHA20/POLY1305(256) Mac=AEAD
0xCC,0xA8 - ECDHE-RSA-CHACHA20-POLY1305 TLSv1.2 Kx=ECDH Au=RSA Enc=CHACHA20/POLY1305(256) Mac=AEAD
0x00,0x9E - DHE-RSA-AES128-GCM-SHA256 TLSv1.2 Kx=DH Au=RSA Enc=AESGCM(128) Mac=AEAD
0x00,0x9F - DHE-RSA-AES256-GCM-SHA384 TLSv1.2 Kx=DH Au=RSA Enc=AESGCM(256) Mac=AEAD
```
Modification:
Moving up `AES_128_GCM_SHA256` in the CIPHERS of HTTPS so that it gets priority.
Result:
When connecting to the server supporting both `ECDHE_ECDSA_WITH_AES_128_GCM_SHA256` and `ECDSA_WITH_CHACHA20_POLY1305_SHA256` and respecting the client priority of cipher suites, it will be able to save significant cpu time when running it on machines with AES-NI support.
Motivation:
TLSv1.3 is not strictly limited to Java11+ anymore as different vendors backported TLSv1.3 to Java8 as well. We should ensure we make the detection of if TLSv1.3 is supported not depend on the Java version that is used.
Modifications:
- Add SslProvider.isTlsv13Supported(...) and use it in tests to detect if we should run tests against TLSv1.3 as well
- Adjust testcase to work on latest JDK 8 release as well
Result:
Correct detection of TLSv1.3 support even if Java version < 11.
Motivation:
We shouldn't call incSmallAllocation() in a synchronized block as its backed by a concurrent datastructure
Modifications:
Move call of incSmallAllocation() out of synchronized block
Result:
Minimize scope of synchronized block
Motivation:
We should provide details about why an IOOBE was thrown
Modification:
Add IndexOutOfBoundsException error information in io.netty.util.internal.AppendableCharSequence and io.netty.handler.codec.CodecOutputList class
Result:
Easier to debug
Motivation:
When a switch statement is used we should always define a `default:` so we don't introduce bugs due fall-through.
Modification:
Add missing `default:`s
Result:
Less error-prone code
Motivation:
For size from 512 bytes to chunkSize, we use a buddy algorithm. The
drawback is that it has a large internal fragmentation.
Modifications:
1. add SizeClassesMetric and SizeClasses
2. remove tiny size, now we have small, normal and huge size
3. rewrite the structure of PoolChunk
4. rewrite pooled allocate algorithm in PoolChunk
5. when allocate subpage, using lowest common multiple of pageSize and
elemSize instead of pageSize.
6. add more tests in PooledByteBufAllocatorTest and PoolArenaTest
Result:
Reduce internal fragmentation and closes#3910
Motivation:
At the moment we don't support session caching for client side when using native SSLEngine implementation and our implementation of SSLSessionContext is incomplete.
Modification:
- Consume netty-tcnative changes to be able to cache session in an external cache
- Add and adjust unit tests to test session caching
- Add an in memory session cache that is hooked into native SSLEngine
Result:
Support session caching on the client and server side
Motivation:
The result of `header.size()` is already cached in `headerSize`. There is no need to call it again actually.
Modification:
Replace the second `header.size()` with `headerSize` directly.
Result:
Improve performance slightly.
Motivation:
jdk.tls.client.enableSessionTicketExtension property must be respect by OPENSSL and OPENSSL_REFCNT SslProvider to ensure a consistent behavior. Due a bug this was not the case and it only worked for OPENSSL_REFCNT but not for OPENSSL.
Modifications:
Move the property check into static method that is used by both
Result:
Correctly respect jdk.tls.client.enableSessionTicketExtension
Motivation:
At the moment we may report BUFFER_OVERFLOW when wrap(...) fails to consume data but still prodce some. This is not correct and we should better report NEED_WRAP as we already have produced some data (for example tickets). This way the user will try again without increasing the buffer size which is more correct and may reduce the number of allocations
Modifications:
Return NEED_WRAP when we produced some data but not consumed any.
Result:
Fix ReferenceCountedOpenSslEngine.wrap(...) state machine
**Motivation:**
We are interested in building Netty libraries with Ahead-of-time compilation with GraalVM. We saw there was [prior work done on this](https://github.com/netty/netty/search?q=graalvm&unscoped_q=graalvm). We want to introduce a change which will unblock GraalVM support for applications using netty and `netty-tcnative`.
This solves the error [that some others encounter](https://github.com/oracle/graal/search?q=UnsatisfiedLinkError+sslOpCipherServerPreference&type=Issues):
```
Exception in thread "main" java.lang.UnsatisfiedLinkError: io.grpc.netty.shaded.io.netty.internal.tcnative.NativeStaticallyReferencedJniMethods.sslOpCipherServerPreference()I [symbol: Java_io_grpc_netty_shaded_io_netty_internal_tcnative_NativeStaticallyReferencedJniMethods_sslOpCipherServerPreference or Java_io_grpc_netty_shaded_io_netty_internal_tcnative_NativeStaticallyReferencedJniMethods_sslOpCipherServerPreference__]
```
**Modification:**
The root cause of the issue is that in the tcnative libraries, the [`SSL.java` class](783a8b6b69/openssl-dynamic/src/main/java/io/netty/internal/tcnative/SSL.java (L67)) references a native call in the static initialization of the class - the method `sslOpCipherServerPreference()` on line 67 is used to initialize the static variable. But you see that `OpenSsl` also uses[ `SSL.class` to check if `netty-tcnative` is on the classpath before loading the native library](cbe238a95b/handler/src/main/java/io/netty/handler/ssl/OpenSsl.java (L123)).
So this is the problem because in ahead-of-time compilation, when it references the SSL class, it will try to load those static initializers and make the native library call, but it cannot do so because the native library has not been loaded yet since the `SSL` class is being referenced to check if the library should be loaded in the first place.
**Solution:** So the proposed solution here is to just choose a different class in the `tcnative` package which does not make a native library call during static initialization. I just chose `SSLContext` if this is OK.
This should have no negative effects other than unblocking the GraalVM use-case.
**Result:**
It fixes the unsatisfied link error. It will fix error for future users; it is a common error people encounter.
Motivation:
There was a new netty-tcnative release which we should use. Beside this the SSLErrorTest was quite fragile and so should be adjusted.
Modifications:
Update netty-tcnative and adjust test
Result:
Use latest netty-tcnative release
Motivation:
Since https://github.com/netty/netty/pull/9865 (Netty 4.1.44) the
default behavior of the HttpObjectDecoder has been to reject any HTTP
message that is found to have multiple Content-Length headers when
decoding. This behavior is well-justified as per the risks outlined in
https://github.com/netty/netty/issues/9861, however, we can see from the
cited RFC section that there are multiple possible options offered for
responding to this scenario:
> If a message is received that has multiple Content-Length header
> fields with field-values consisting of the same decimal value, or a
> single Content-Length header field with a field value containing a
> list of identical decimal values (e.g., "Content-Length: 42, 42"),
> indicating that duplicate Content-Length header fields have been
> generated or combined by an upstream message processor, then the
> recipient MUST either reject the message as invalid or replace the
> duplicated field-values with a single valid Content-Length field
> containing that decimal value prior to determining the message body
> length or forwarding the message.
https://tools.ietf.org/html/rfc7230#section-3.3.2
Netty opted for the first option (rejecting as invalid), which seems
like the safest, but the second option (replacing duplicate values with
a single value) is also valid behavior.
Modifications:
* Introduce "allowDuplicateContentLengths" parameter to
HttpObjectDecoder (defaulting to false).
* When set to true, will allow multiple Content-Length headers only if
they are all the same value. The duplicated field-values will be
replaced with a single valid Content-Length field.
* Add new parameterized test class for testing different variations of
multiple Content-Length headers.
Result:
This is a backwards-compatible change with no functional change to the
existing behavior.
Note that the existing logic would result in NumberFormatExceptions
for header values like "Content-Length: 42, 42". The new logic correctly
reports these as IllegalArgumentException with the proper error message.
Additionally note that this behavior is only applied to HTTP/1.1, but I
suspect that we may want to expand that to include HTTP/1.0 as well...
That behavior is not modified here to minimize the scope of this change.
Motivation:
`getEntry(...)` may throw an IndexOutOfBoundsException without any error messages.
Modification:
Add detailed error message corresponding to the IndexOutOfBoundsException while calling `getEntry(...)` in HpackDynamicTable.java.
Result:
Easier to debug
Motivation:
When BlockHound is installed,
ReferenceCountedOpenSslClientContext$ExtendedTrustManagerVerifyCallback.verify
is reported as blocking call.
Modifications:
Add allowBlockingCallsInside configuration for
ReferenceCountedOpenSslClientContext$ExtendedTrustManagerVerifyCallback.verify
Result:
Fixes#10384
Motivation:
`HpackDynamicTable` needs some test cases to ensure bug-free.
Modification:
Add unit test for `HpackDynamicTable`.
Result:
Improve test coverage slightly.
Motivation:
BoringSSL behaves differently then OpenSSL and not include any TLS1.3 ciphers in the returned array when calling SSL_get_ciphers(...). This is due the fact that it also not allow to explicit configure which are supported and which not for TLS1.3. To mimic the behaviour expected by the SSLEngine API we should workaround this.
Modifications:
- Add a unit test that verifies enabled protocols / ciphers
- Add special handling for BoringSSL and tls1.3
Result:
Make behaviour consistent
Motivation:
newResourceLeakDetector(...) did not correctly pass the samplingInterval parameter and so it was ignored.
Modification:
ResourceLeakDetectorFactory.newResourceLeakDetector(Class, int) use the second parameter as the sampling interval of the newly created ResourceLeakDetector.
Result:
Fixes#10378
Motivation:
We did not correctly preserve the original cause of the SSLException when all the following are true:
* SSL_read failed
* handshake was finished
* some outbound data was produced durigin SSL_read (for example ssl alert) that needs to be picked up by wrap(...)
Modifications:
Ensure we correctly preserve the original cause of the SSLException and rethrow it once we produced all data via wrap(...)
Result:
Be able to understand why we see an error in all cases
Motivation:
Unsafe users are getting MpscChunkedArrayQueue while no Unsafe ones MpscGrowableAtomicArrayQueue
Modifications:
MpscChunkedAtomicArrayQueue should be used for no Unsafe users (matching MpscChunkedArrayQueue behaviour)
Result:
no Unsafe users uses MpscChunkedAtomicArrayQueue while allocating bounded Mpsc Queues
