Motivation:
ByteToMessageDecoder requires using an intermediate List to put results into. This intermediate list adds overhead (memory/CPU) which grows as the number of objects increases. This overhead can be avoided by directly propagating events through the ChannelPipeline via ctx.fireChannelRead(...). This also makes the semantics more clear and allows us to keep track if we need to call ctx.read() in all cases.
Modifications:
- Remove List from the method signature of ByteToMessageDecoder.decode(...) and decodeLast(...)
- Adjust all sub-classes
- Adjust unit tests
- Fix javadocs.
Result:
Adjust ByteToMessageDecoder as noted in https://github.com/netty/netty/issues/8525.
# Motivation:
`DefaultByteBufHolder.equals()` considers another object equal if it's an instance of `ByteBufferHolder` and if the contents of two objects are equal. However, the behavior of `equals` method is not a part of the `ByteBufHolder` contract so `DefaultByteBufHolder`'s version may be causing violation of the symmetric property if other classes have different logic.
There are already a few classes that are affected by this: `DefaultHttp2GoAwayFrame`, `DefaultHttp2UnknownFrame`, and `SctpMessage` are all overriding `equals` method breaking the symmetric property.
Another effect of this behavior is that all instances with empty data are considered equal. That may not be desireable in the situations when instances are created for predefined constants, e.g. `FullBulkStringRedisMessage.NULL_INSTANCE` and `FullBulkStringRedisMessage.EMPTY_INSTANCE` in `codec-redis`.
# Modification:
Make `DefaultByteBufHolder.equals()` implementation only work for the objects of the same class.
# Result:
- The symmetric property of the `equals` method is restored for the classes in question.
- Instances of different classes are not considered equal even if the content of the data they hold are the same.
Motivation:
We should use Objects.requireNonNull(...) as we require java8
Modifications:
Replace ObjectUtil.checkNonNull(...) with Objects.requireNonNull(...)
Result:
Code cleanup
Motivation:
The current implementation delegates to writeHeaders(ChannelHandlerContext ctx, int streamId, Http2Headers headers, int streamDependency, short weight, boolean exclusive, int padding, boolean endStream, ChannelPromise promise) that will send an header frame with the priority flag set and the default priority values even if the user didnt want too.
Modifications:
- Change DefaultHttp2ConnectionEncoder to call the correct Http2FrameWriter method depending on if the user wants to use priorities or not
- Adjust tests
Result:
Fixes https://github.com/netty/netty/issues/9842
Motivation:
We use the onStreamClosed(...) callback to return unconsumed bytes back to the window of the connection when needed. When this happens we will write a window update frame but not automatically call ctx.flush(). As the user has no insight into this it could in the worst case result in a "deadlock" as the frame is never written out ot the socket.
Modifications:
- If onStreamClosed(...) produces a window update frame call ctx.flush()
- Add unit test
Result:
No stales possible due unflushed window update frames produced by onStreamClosed(...) when not all bytes were consumed before the stream was closed
Motivation:
At the moment we send a window update frame for the connection + stream when a stream is closed and there are unconsumed bytes left. While we need to do this for the connection it makes no sense to write a window update frame for the stream itself as it is already closed
Modifications:
- Don't write the window update frame for the stream when the stream is closed
- Add unit test
Result:
Don't write the window frame for closed streams
Motivation:
Http2ConnectionHandler tries to guard against sending multiple RST frames for the same stream. Unfortunally the code is not 100 % correct as it only updates the state after it calls write. This may lead to the situation of have an extra RST frame slip through if the second write for the RST frame is done from a listener that is attached to the promise.
Modifications:
- Update state before calling write
- Add unit test
Result:
Only ever send one RST frame per stream
Motivation:
There is an intrinsic race between a local session resetting a stream
and the peer no longer sending any frames. This can result in the
session receiving frames for a stream that the local peer no longer
tracks. This results in a StreamException being thrown which triggers a
RST_STREAM frame, which is a good thing, but also logging at level WARN,
which is noisy for an expected and benign condition.
Modification:
Change the log level to DEBUG when logging stream errors with code
STREAM_CLOSED. All others are more interesting and will continue to be
logged at level WARN.
Additionally, it was found that DATA frames for streams that could not
have existed only resulted in a StreamException when the spec is clear
that such a situation should be fatal to the connection, resulting in a
GOAWAY(PROTOCOL_ERROR).
Fixes#8025.
Motivation:
To avoid regression regarding connection-specific headers[1], we should add a test.
[1] https://tools.ietf.org/html/rfc7540#section-8.1.2.2
Modification:
Add test that checks the following headers are removed.
- Connection
- Host
- Keep-Alive
- Proxy-Connection
- Transfer-Encoding
- Upgrade
Result:
There's no functional change.
Motivation:
Padding was removed from CONTINUATION frame in http2-spec, as showed in [PR](https://github.com/http2/http2-spec/pull/510). We should follow it.
Modifications:
- Remove padding when writing CONTINUATION frame in DefaultHttp2FrameWriter
- Add a unit test for writing large header with padding
Result:
More spec-compliant
Motivation
Per javadoc in 4.1.x SimpleChannelInboundHandler:
"Please keep in mind that channelRead0(ChannelHandlerContext, I) will be
renamed to messageReceived(ChannelHandlerContext, I) in 5.0."
Modifications
Rename aforementioned method and all references/overrides.
Result
Method is renamed.
Motivation:
Netty HTTP/2 implementation is not 100% compliant to the spec. This
commit improves the compliance regarding headers validation,
in particular pseudo-headers and connection ones.
According to the spec:
All HTTP/2 requests MUST include exactly one valid value for the
":method", ":scheme", and ":path" pseudo-header fields, unless it is
a CONNECT request (Section 8.3). An HTTP request that omits
mandatory pseudo-header fields is malformed (Section 8.1.2.6).
Modifications:
- Introduce Http2HeadersValidator class capable of validating HTTP/2
headers
- Invoke validation from DefaultHttp2ConnectionDecoder#onHeadersRead
- Modify tests to use valid headers when required
- Modify HttpConversionUtil#toHttp2Headers to not add :scheme and
:path header on CONNECT method in order to conform to the spec
Result:
- Initial requests without :method, :path, :scheme will fail
- Initial requests with multiple values for :method, :path, :scheme
will fail
- Initial requests with an empty :path fail
- Requests with connection-specific header field will fail
- Requests with TE header different than "trailers" will fail
-
- Fixes 8.1.2.2 tests from h2spec #5761
- Fixes 8.1.2.3 tests from h2spec #5761
Motivation:
Constant time comparison functions are used to compare HTTP/2 header
values, even if they are not sensitive.
Modification:
After checking for sensitivity, use fast comparison.
Result: Faster HPACK table reads/writes
Motivation:
In https://github.com/netty/netty/issues/8692, `Http2FrameCodec` was
updated to keep track of all "being initialized" streams, allocating
memory before initialization begins, and releasing memory after
initialization completes successfully.
In some instances where stream initialization fails (e.g. because this
connection has received a GOAWAY frame), this memory is never released.
Modifications:
This change updates the `Http2FrameCodec` to use a separate promise
for monitoring the success of sending HTTP2 headers. When sending of
headers fails, we now make sure to release memory allocated for stream
initialization.
Result:
After this change, failures in writing HTTP2 Headers (e.g. because this
connection has received a GOAWAY frame) will no longer leak memory.
Motiviation:
EmbeddedChannel currently is quite differently in terms of semantics to other Channel implementations. We should better change it to be more closely aligned and so have the testing code be more robust.
Modifications:
- Change EmbeddedEventLoop.inEventLoop() to only return true if we currenlty run pending / scheduled tasks
- Change EmbeddedEventLoop.execute(...) to automatically process pending tasks if not already doing so
- Adjust a few tests for the new semantics (which is closer to other Channel implementations)
Result:
EmbeddedChannel works more like other Channel implementations
Motivation:
At the moment it is quite easy to hit reentrance issues when you have multiple handlers in the pipeline and each of the handlers does not correctly protect against these. To make it easier for the user we should try to protect from these. The issue is usually if and inbound event will trigger and outbound event and this outbound event then against triggeres an inbound event. This may result in having methods in a ChannelHandler re-enter some method and so state can be corrupted or messages be re-ordered.
Modifications:
- Keep track of inbound / outbound operations in DefaultChannelHandlerContext and if reentrancy is detected break it by scheduling the action on the EventLoop. This will then be picked up once the method returns and so the reentrancy is broken up.
- Adjust tests which made strange assumptions about execution order
Result:
No more reentrancy of handlers possible.
Motivation:
In the release (4.1.37) we introduced Http2MultiplexHandler as a
replacement of Http2MultiplexCodec. This did split the frame parsing from
the multiplexing to allow a more flexible way to handle frames and to make
the code cleaner. Unfortunally we did miss to special handle this in
Http2ClientUpgradeCodec and so did not correctly add Http2MultiplexHandler
to the pipeline before calling Http2FrameCodec.onHttpClientUpgrade(...).
This did lead to the situation that we did not correctly receive the event
on the Http2MultiplexHandler and so did not correctly created the
Http2StreamChannel for the upgrade stream. Because of this we ended up
with an NPE if a frame was dispatched to the upgrade stream later on.
Modifications:
- Correctly add Http2MultiplexHandler to the pipeline before calling Http2FrameCodec.onHttpClientUpgrade(...)
Result:
Fixes#9495.
Motivation:
As we decorate the Http2FrameListener under the covers we should ensure the user can still access the original Http2FrameListener.
Modifications:
- Unwrap the Http2FrameListener in frameListener()
- Add unit test
Result:
Less suprises for users.
Motivation:
We recently introduced Http2ControlFrameLimitEncoderTest which did not correctly notify the goAway promises and so leaked buffers.
Modifications:
Correctly notify all promises and so release the debug data.
Result:
Fixes leak in HTTP2 test
Motivation:
It is possible for a remote peer to flood the server / client with empty DATA frames (without end_of_stream flag) set and so cause high CPU usage without the possibility to ever hit a limit. We need to guard against this.
See CVE-2019-9518
Modifications:
- Add a new config option to AbstractHttp2ConnectionBuilder and sub-classes which allows to set the max number of consecutive empty DATA frames (without end_of_stream flag). After this limit is hit we will close the connection. A limit of 10 is used by default.
- Add unit tests
Result:
Guards against CVE-2019-9518
Motivation:
Due how http2 spec is defined it is possible by a remote peer to flood us with frames that will trigger control frames as response, the problem here is that the remote peer can also just stop reading these (while still produce more of these) and so may drive us to the pointer where we either run out of memory or burn all CPU. To protect against this we need to implement some kind of limit that will tear down connections that cause the above mentioned situation.
See CVE-2019-9512 / CVE-2019-9514 / CVE-2019-9515
Modifications:
- Add Http2ControlFrameLimitEncoder which limits the number of queued control frames that were caused because of the remote peer.
- Allow to insert ths Http2ControlFrameLimitEncoder by setting AbstractHttp2ConnectionBuilder.encoderEnforceMaxQueuedControlFrames(...) to a number higher then 0. The default is 10000 which provides some protection by default but will hopefully not cause too many false-positives.
- Add unit tests
Result:
Protect against DDOS due control frames. Fixes CVE-2019-9512 / CVE-2019-9514 / CVE-2019-9515 .
Motivation:
306299323c introduced some code change to move the responsibility of creating the stream for the upgrade to Http2FrameCodec. Unfortunaly this lead to the situation of having newStream().setStreamAndProperty(...) be called twice. Because of this we only ever saw the channelActive(...) on Http2StreamChannel but no other events as the mapping was replaced on the second newStream().setStreamAndProperty(...) call.
Beside this we also did not use the correct handler for the upgrade stream in some cases
Modifications:
- Just remove the Http2FrameCodec.onHttpClientUpgrade() method and so let the base class handle all of it. The stream is created correctly as part of the ConnectionListener implementation of Http2FrameCodec already.
- Consolidate logic of creating stream channels
- Adjust unit test to capture the bug
Result:
Fixes https://github.com/netty/netty/issues/9395
Motivation:
When using the HTTP/2 multiplex implementation we need to ensure we correctly drain the buffered inbound data even if the RecvByteBufallocator.Handle tells us to stop reading in between.
Modifications:
Correctly loop through the buffered inbound data until the user does stop to request from it.
Result:
Fixes https://github.com/netty/netty/issues/9387.
Co-authored-by: Bryce Anderson <banderson@twitter.com>
Motivation:
If a read triggers a AbstractHttp2StreamChannel to close we can
get an NPE in the read loop.
Modifications:
Make sure that the inboundBuffer isn't null before attempting to
continue the loop.
Result:
No NPE.
Fixes#9337
Motivation:
There are situations where the user may want to be more flexible when to send the PING acks for various reasons or be able to attach a listener to the future that is used for the ping ack. To be able to do so we should allow to manage the acks manually.
Modifications:
- Add constructor to DefaultHttp2ConnectionDecoder that allows to disable the automatically sending of ping acks (default is to send automatically to not break users)
- Add methods to AbstractHttp2ConnectionHandlerBuilder (and sub-classes) to either enable ot disable auto acks for pings
- Make DefaultHttp2PingFrame constructor public that allows to write acks.
- Add unit test
Result:
More flexible way of handling acks.
Motivation:
Http2ConnectionHandler (and sub-classes) allow to configure a graceful shutdown timeout but only apply it if there is at least one active stream. We should always apply the timeout. This is also true when we try to send a GO_AWAY and close the connection because of an connection error.
Modifications:
- Always apply the timeout if one is configured
- Add unit test
Result:
Always respect gracefulShutdownTimeoutMillis
Motivation:
b3dba317d7 introduced the concept of Http2SettingsReceivedConsumer but did not correctly inplement DecoratingHttp2ConnectionEncoder.consumeRemoteSettings(...).
Modifications:
- Add missing `else` around the throws
- Add unit tests
Result:
Correctly implement DecoratingHttp2ConnectionEncoder.consumeRemoteSettings(...)
Motivation:
ff0045e3e1 changed HpackHuffmanDecoder to use a lookup-table which greatly improved performance. We can squeeze out another 3% win by using an ByteProcessor which will reduce the number of bound-checks / reference-count-checks needed by processing byte-by-byte.
Modifications:
Implement logic with ByteProcessor
Result:
Another ~3% perf improvement which shows up when using h2load to simulate load.
`h2load -c 100 -m 100 --duration 60 --warm-up-time 10 http://127.0.0.1:8080`
Before:
```
finished in 70.02s, 620051.67 req/s, 20.70MB/s
requests: 37203100 total, 37203100 started, 37203100 done, 37203100 succeeded, 0 failed, 0 errored, 0 timeout
status codes: 37203100 2xx, 0 3xx, 0 4xx, 0 5xx
traffic: 1.21GB (1302108500) total, 41.84MB (43872600) headers (space savings 90.00%), 460.24MB (482598600) data
min max mean sd +/- sd
time for request: 404us 24.52ms 15.93ms 1.45ms 87.90%
time for connect: 0us 0us 0us 0us 0.00%
time to 1st byte: 0us 0us 0us 0us 0.00%
req/s : 6186.64 6211.60 6199.00 5.18 65.00%
```
With this change:
```
finished in 70.02s, 642103.33 req/s, 21.43MB/s
requests: 38526200 total, 38526200 started, 38526200 done, 38526200 succeeded, 0 failed, 0 errored, 0 timeout
status codes: 38526200 2xx, 0 3xx, 0 4xx, 0 5xx
traffic: 1.26GB (1348417000) total, 42.39MB (44444900) headers (space savings 90.00%), 466.25MB (488893900) data
min max mean sd +/- sd
time for request: 370us 24.89ms 15.52ms 1.35ms 88.02%
time for connect: 0us 0us 0us 0us 0.00%
time to 1st byte: 0us 0us 0us 0us 0.00%
req/s : 6407.06 6435.19 6419.74 5.62 67.00%
```
Motivation:
In the latest release we introduced Http2MultiplexHandler as a replacement of Http2MultiplexCodec. This did split the frame parsing from the multiplexing to allow a more flexible way to handle frames and to make the code cleaner. Unfortunally we did miss to special handle this in Http2ServerUpgradeCodec and so did not correctly add Http2MultiplexHandler to the pipeline before calling Http2FrameCodec.onHttpServerUpgrade(...). This did lead to the situation that we did not correctly receive the event on the Http2MultiplexHandler and so did not correctly created the Http2StreamChannel for the upgrade stream. Because of this we ended up with an NPE if a frame was dispatched to the upgrade stream later on.
Modifications:
- Correctly add Http2MultiplexHandler to the pipeline before calling Http2FrameCodec.onHttpServerUpgrade(...)
- Add unit test
Result:
Fixes https://github.com/netty/netty/issues/9314.
Motivation:
We should not propage Http2WindowUpdateFrames to the child channels at all as these are not really use-ful and should not be flow-controlled via `read()` anyway. In the other hand Http2ResetFrame is very useful but should be propagated via an user event so the user is aware of it directly even if the user stops reading.
Modifications:
- Dont propagate Http2WindowUpdateFrames when using Http2MultiplexHandler
- Use user event for Http2ResetFrame when using Http2MultiplexHandler
- Adjust javadoc of Http2MultiplexHandler
- Add unit tests
Result:
Fixes https://github.com/netty/netty/pull/8889 and https://github.com/netty/netty/pull/7635
Motivation:
In the past we had the following class hierarchy:
Http2ConnectionHandler --- Http2FrameCodec -- Http2MultiplexCodec
This hierarchy makes it impossible to plug in any code that would like to act on Http2Frame and Http2StreamFrame which can be quite useful for various situations (like metrics, logging etc). Beside this it also made the implementtion very hacky. To allow easier maintainance and also allow more flexible costumizations we should split Http2MultiplexCodec and Http2FrameCode.
Modifications:
- Introduce Http2MultiplexHandler (which is a replacement for Http2MultiplexCodec when used together with Http2FrameCodec)
- Mark Http2MultiplexCodecBuilder and Http2MultiplexCodec as deprecated. People should use Http2FrameCodecBuilder / Http2FrameCodec together with Http2MultiplexHandlder in the future
- Adjust / Add tests
- Adjust examples
Result:
More flexible usage possible and less hacky / coupled implementation for http2 multiplexing
Motivation:
For HTTP/2 messages with multiple cookies HttpConversionUtil.addHttp2ToHttpHeaders spends a good portion of time creating throwaway StringBuilders.
Modification:
Handle cookies lazily by using a ThreadLocal StringBuilder and then converting it to the H1 header at the end.
Result:
Less allocations.
Motivation:
f945a071db decoupled the writability state from the flow controller but could lead to the situation of a lot of writability updates events were propagated to the child channels. This change ensure we only take into account if the parent channel becomes writable again before we try to set the child channels to writable.
Modifications:
Only listen for channel writability changes for if the parent channel becomes writable again.
Result:
Less writability updates.
Motivation:
We should decouple the writability state of the http2 child channels from the flow-controller and just tie it to its own pending bytes counter that is decremented by the parent Channel once the bytes were written.
Modifications:
- Decouple writability state of child channels from flow-contoller
- Update tests
Result:
Less coupling and more correct behavior. Fixes https://github.com/netty/netty/issues/8148.
Motivation:
b4e3c12b8e introduced code to avoid coupling
close() to graceful close. It also added some code which attempted to infer when
a graceful close was being done in writing of a GOAWAY to preserve the
"connection is closed when all streams are closed behavior" for the child
channel API. However the implementation was too overzealous and may preemptively
close the connection if there are not currently any open streams (and close if
there are any frames which create streams in flight).
Modifications:
- Decouple writing a GOAWAY from trying to infer if a graceful close is being
done and closing the connection. Even if we could enhance this logic (e.g.
wait to close until the second GOAWAY with no error) it is possible the user
doesn't want the connection to be closed yet. We can add a means for the codec
to orchestrate the graceful close in the future (e.g. write some special "close
the connection when all streams are closed") but for now we can just let the
application handle this.
Result:
Fixes https://github.com/netty/netty/issues/9207
Motivation:
Http2ConnectionHandler#close(..) always runs the GOAWAY and graceful close
logic. This coupling means that a user would have to override
Http2ConnectionHandler#close(..) to modify the behavior, and the
Http2FrameCodec and Http2MultiplexCodec are not extendable so you cannot
override at this layer. Ideally we can totally decouple the close(..) of the
transport and the GOAWAY graceful closure process completely, but to preserve
backwards compatibility we can add an opt-out option to decouple where the
application is responsible for sending a GOAWAY with error code equal to
NO_ERROR as described in https://tools.ietf.org/html/rfc7540#section-6.8 in
order to initiate graceful close.
Modifications:
- Http2ConnectionHandler supports an additional boolean constructor argument to
opt out of close(..) going through the graceful close path.
- Http2FrameCodecBuilder and Http2MultiplexCodec expose
gracefulShutdownTimeoutMillis but do not hook them up properly. Since these
are already exposed we should hook them up and make sure the timeout is applied
properly.
- Http2ConnectionHandler's goAway(..) method from Http2LifecycleManager should
initiate the graceful closure process after writing a GOAWAY frame if the error
code is NO_ERROR. This means that writing a Http2GoAwayFrame from
Http2FrameCodec will initiate graceful close.
Result:
Http2ConnectionHandler#close(..) can now be decoupled from the graceful close
process, and immediately close the underlying transport if desired.
Motivation:
The HTTP/2 codec will synchronously respond to a SETTINGS frame with a SETTINGS
ACK before the application sees the SETTINGS frame. The application may need to
adjust its state depending upon what is in the SETTINGS frame before applying
the remote settings and responding with an ACK (e.g. to adjust for max
concurrent streams). In order to accomplish this the HTTP/2 codec should allow
for the application to opt-in to sending the SETTINGS ACK.
Modifications:
- DefaultHttp2ConnectionDecoder should support a mode where SETTINGS frames can
be queued instead of immediately applying and ACKing.
- DefaultHttp2ConnectionEncoder should attempt to poll from the queue (if it
exists) to apply the earliest received but not yet ACKed SETTINGS frame.
- AbstractHttp2ConnectionHandlerBuilder (and sub classes) should support a new
option to enable the application to opt-in to managing SETTINGS ACK.
Result:
HTTP/2 allows for asynchronous SETTINGS ACK managed by the application.
Motivation:
Http2FrameCodec currently fails the write promise associated with creating a
stream with a Http2NoMoreStreamIdsException. However this means the user code
will have to listen to all write futures in order to catch this scenario which
is the same as receiving a GOAWAY frame. We can also simulate receiving a GOAWAY
frame from our remote peer and that allows users to consolidate graceful close
logic in the GOAWAY processing.
Modifications:
- Http2FrameCodec should simulate a DefaultHttp2GoAwayFrame when trying to
create a stream but the stream IDs have been exhausted.
Result:
Applications can rely upon GOAWAY for graceful close processing instead of also
processing write futures.
Motivation:
We should not throw check exceptions when the user calls sync*() but should better wrap it in a CompletionException to make it easier for people to reason about what happens.
Modifications:
- Change sync*() to throw CompletionException
- Adjust tests
- Add some more tests
Result:
Fixes https://github.com/netty/netty/issues/8521.
Motivation:
DefaultPromise requires an EventExecutor which provides the thread to notify listeners on and this EventExecutor can never change. We can remove the code that supported the possibility of a changing the executor as this is not possible anymore.
Modifications:
- Remove constructor which allowed to construct a *Promise without an EventExecutor
- Remove extra state
- Adjusted SslHandler and ProxyHandler for new code
Result:
Fixes https://github.com/netty/netty/issues/8517.
Motivation:
In 42742e233f we already added default methods to Channel*Handler and deprecated the Adapter classes to simplify the class hierarchy. With this change we go even further and merge everything into just ChannelHandler. This simplifies things even more in terms of class-hierarchy.
Modifications:
- Merge ChannelInboundHandler | ChannelOutboundHandler into ChannelHandler
- Adjust code to just use ChannelHandler
- Deprecate old interfaces.
Result:
Cleaner and simpler code in terms of class-hierarchy.
Motivation:
As we now us java8 as minimum java version we can deprecate ChannelInboundHandlerAdapter / ChannelOutboundHandlerAdapter and just move the default implementations into the interfaces. This makes things a bit more flexible for the end-user and also simplifies the class-hierarchy.
Modifications:
- Mark ChannelInboundHandlerAdapter and ChannelOutboundHandlerAdapter as deprecated
- Add default implementations to ChannelInboundHandler / ChannelOutboundHandler
- Refactor our code to not use ChannelInboundHandlerAdapter / ChannelOutboundHandlerAdapter anymore
Result:
Cleanup class-hierarchy and make things a bit more flexible.
