Refactor Mizu, define an extension API and add new protocols: AMQP, Kafka (#224)

* Separate HTTP related code into `extensions/http` as a Go plugin

* Move `extensions` folder into `tap` folder

* Move HTTP files into `tap/extensions/lib` for now

* Replace `orcaman/concurrent-map` with `sync.Map`

* Remove `grpc_assembler.go`

* Remove `github.com/up9inc/mizu/tap/extensions/http/lib`

* Add a build script to automatically build extensions from a known path and load them

* Start to define the extension API

* Implement the `run()` function for the TCP stream

* Add support of defining multiple ports to the extension API

* Set the extension name inside the extension

* Declare the `Dissect` function in the extension API

* Dissect HTTP request from inside the HTTP extension

* Make the distinction of outbound and inbound ports

* Dissect HTTP response from inside the HTTP extension

* Bring back the HTTP request-response pair matcher

* Return a `*api.RequestResponsePair` from the dissection

* Bring back the gRPC-HTTP/2 parser

* Fix the issues in `handleHTTP1ClientStream` and `handleHTTP1ServerStream`

* Call a function pointer to emit dissected data back to the `tap` package

* roee changes -
trying to fix agent to work with the "api" object) - ***still not working***

* small mistake in the conflicts

* Fix the issues that are introduced by the merge conflict

* Add `Emitter` interface to the API and send `OutputChannelItem`(s) to `OutputChannel`

* Fix the `HTTP1` handlers

* Set `ConnectionInfo` in HTTP handlers

* Fix the `Dockerfile` to build the extensions

* remove some unwanted code

* no message

* Re-enable `getStreamProps` function

* Migrate back from `gopacket/tcpassembly` to `gopacket/reassembly`

* Introduce `HTTPPayload` struct and `HTTPPayloader` interface to `MarshalJSON()` all the data structures that are returned by the HTTP protocol

* Read `socketHarOutChannel` instead of `filteredHarChannel`

* Connect `OutputChannelItem` to the last WebSocket means that finally the web UI started to work again

* Add `.env.example` to React app

* Marshal and unmarshal `*http.Request`, `*http.Response` pairs

* Move `loadExtensions` into `main.go` and map extensions into `extensionsMap`

* Add `Summarize()` method to the `Dissector` interface

* Add `Analyze` method to the `Dissector` interface and `MizuEntry` to the extension API

* Add `Protocol` struct and make it effect the UI

* Refactor `BaseEntryDetails` struct and display the source and destination ports in the UI

* Display the protocol name inside the details layout

* Add `Represent` method to the `Dissector` interface and manipulate the UI through this method

* Make the protocol color affect the details layout color and write protocol abbreviation vertically

* Remove everything HTTP related from the `tap` package and make the extension system fully functional

* Fix the TypeScript warnings

* Bring in the files related AMQP into `amqp` directory

* Add `--nodefrag` flag to the tapper and bring in the main AMQP code

* Implement the AMQP `BasicPublish` and fix some issues in the UI when the response payload is missing

* Implement `representBasicPublish` method

* Fix several minor issues

* Implement the AMQP `BasicDeliver`

* Implement the AMQP `QueueDeclare`

* Implement the AMQP `ExchangeDeclare`

* Implement the AMQP `ConnectionStart`

* Implement the AMQP `ConnectionClose`

* Implement the AMQP `QueueBind`

* Implement the AMQP `BasicConsume`

* Fix an issue in `ConnectionStart`

* Fix a linter error

* Bring in the files related Kafka into `kafka` directory

* Fix the build errors in Kafka Go files

* Implement `Dissect` method of Kafka and adapt request-response pair matcher to asynchronous client-server stream

* Do the "Is reversed?" checked inside `getStreamProps` and fix an issue in Kafka `Dissect` method

* Implement `Analyze`, `Summarize` methods of Kafka

* Implement the representations for Kafka `Metadata`, `RequestHeader` and `ResponseHeader`

* Refactor the AMQP and Kafka implementations to create the summary string only inside the `Analyze` method

* Implement the representations for Kafka `ApiVersions`

* Implement the representations for Kafka `Produce`

* Implement the representations for Kafka `Fetch`

* Implement the representations for Kafka `ListOffsets`, `CreateTopics` and `DeleteTopics`

* Fix the encoding of AMQP `BasicPublish` and `BasicDeliver` body

* Remove the unnecessary logging

* Remove more logging

* Introduce `Version` field to `Protocol` struct for dynamically switching the HTTP protocol to HTTP/2

* Fix the issues in analysis and representation of HTTP/2 (gRPC) protocol

* Fix the issues in summary section of details layout for HTTP/2 (gRPC) protocol

* Fix the read errors that freezes the sniffer in HTTP and Kafka

* Fix the issues in HTTP POST data

* Fix one more issue in HTTP POST data

* Fix an infinite loop in Kafka

* Fix another freezing issue in Kafka

* Revert "UI Infra - Support multiple entry types + refactoring (#211)"

This reverts commit f74a52d4dc.

* Fix more issues that are introduced by the merge

* Fix the status code in the summary section

* adding the cleaner again (why we removed it?).
add TODO: on the extension loop .

* fix dockerfile (remove deleting .env file) - it is found in dockerignore and fails to build if the file not exists

* fix GetEntrties ("/entries" endpoint) - working with "tapApi.BaseEntryDetail" (moved from shared)

* Fix an issue in the UI summary section

* Refactor the protocol payload structs

* Fix a log message in the passive tapper

* Adapt `APP_PORTS` environment variable to the new extension system and change its format to `APP_PORTS='{"http": ["8001"]}' `

* Revert "fix dockerfile (remove deleting .env file) - it is found in dockerignore and fails to build if the file not exists"

This reverts commit 4f514ae1f4.

* Bring in the necessary changes from f74a52d4dc

* Open the API server URL in the web browser as soon as Mizu is ready

* Make the TCP reader consists of a single Go routine (instead of two) and try to dissect in both client and server mode by rewinding

* Swap `TcpID` without overwriting it

* Sort extension by priority

* Try to dissect with looping through all the extensions

* fix getStreamProps function.
(it should be passed from CLI as it was before).

* Turn TCP reader back into two Goroutines (client and server)

* typo

* Learn `isClient` from the TCP stream

* Set `viewer` style `overflow: "auto"`

* Fix the memory leaks in AMQP and Kafka dissectors

* Revert some of the changes in be7c65eb6d

* Remove `allExtensionPorts` since it's no longer needed

* Remove `APP_PORTS` since it's no longer needed

* Fix all of the minor issues in the React code

* Check Kafka header size and fail-fast

* Break the dissectors loop upon a successful dissection

* Don't break the dissector loop. Protocols might collide

* Improve the HTTP request-response counter (still not perfect)

* Make the HTTP request-response counter perfect

* Revert "Revert some of the changes in be7c65eb6d3fb657a059707da3ca559937e59739"

This reverts commit 08e7d786d8.

* Bring back `filterItems` and `isHealthCheckByUserAgent` functions

* Remove some development artifacts

* remove unused and commented lines that are not relevant

* Fix the performance in TCP stream factory. Make it create two `tcpReader`(s) per extension

* Change a log to debug

* Make `*api.CounterPair` a field of `tcpReader`

* Set `isTapTarget` to always `true` again since `filterAuthorities` implementation has problems

* Remove a variable that's only used for logging even though not introduced by this branch

* Bring back the `NumberOfRules` field of `ApplicableRules` struct

* Remove the unused `NewEntry` function

* Move `k8sResolver == nil` check to a more appropriate place

* default healthChecksUserAgentHeaders should be empty array (like the default config value)

* remove spam console.log

* Rules button cause app to crash (access the service via incorrect property)

* Ignore all .env* files in docker build.

* Better caching in dockerfile: only copy go.mod before go mod download.

* Check for errors while loading an extension

* Add a comment about why `Protocol` is not a pointer

* Bring back the call to `deleteOlderThan`

* Remove the `nil` check

* Reduce the maximum allowed AMQP message from 128MB to 1MB

* Fix an error that only occurs when a Kafka broker is initiating

* Revert the change in b2abd7b990

* Fix the service name resolution in all protocols

* Remove the `anydirection` flag and fix the issue in `filterAuthorities`

* Pass `sync.Map` by reference to `deleteOlderThan` method

* Fix the packet capture issue in standalone mode that's introduced by the removal of `anydirection`

* Temporarily resolve the memory exhaustion in AMQP

* Fix a nil pointer dereference error

* Fix the CLI build error

* Fix a memory leak that's identified by `pprof`

Co-authored-by: Roee Gadot <roee.gadot@up9.com>
Co-authored-by: Nimrod Gilboa Markevich <nimrod@up9.com>

tap/tcp_reader.go

@@ -0,0 +1,103 @@
+package tap
+
+import (
+	"bufio"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"sync"
+	"time"
+
+	"github.com/bradleyfalzon/tlsx"
+	"github.com/romana/rlog"
+	"github.com/up9inc/mizu/tap/api"
+)
+
+const checkTLSPacketAmount = 100
+
+type tcpReaderDataMsg struct {
+	bytes     []byte
+	timestamp time.Time
+}
+
+type tcpID struct {
+	srcIP   string
+	dstIP   string
+	srcPort string
+	dstPort string
+}
+
+type ConnectionInfo struct {
+	ClientIP   string
+	ClientPort string
+	ServerIP   string
+	ServerPort string
+	IsOutgoing bool
+}
+
+func (tid *tcpID) String() string {
+	return fmt.Sprintf("%s->%s %s->%s", tid.srcIP, tid.dstIP, tid.srcPort, tid.dstPort)
+}
+
+/* tcpReader gets reads from a channel of bytes of tcp payload, and parses it into requests and responses.
+ * The payload is written to the channel by a tcpStream object that is dedicated to one tcp connection.
+ * An tcpReader object is unidirectional: it parses either a client stream or a server stream.
+ * Implements io.Reader interface (Read)
+ */
+type tcpReader struct {
+	ident              string
+	tcpID              *api.TcpID
+	isClient           bool
+	isOutgoing         bool
+	msgQueue           chan tcpReaderDataMsg // Channel of captured reassembled tcp payload
+	data               []byte
+	captureTime        time.Time
+	parent             *tcpStream
+	messageCount       uint
+	packetsSeen        uint
+	outboundLinkWriter *OutboundLinkWriter
+	extension          *api.Extension
+	emitter            api.Emitter
+	counterPair        *api.CounterPair
+}
+
+func (h *tcpReader) Read(p []byte) (int, error) {
+	var msg tcpReaderDataMsg
+
+	ok := true
+	for ok && len(h.data) == 0 {
+		msg, ok = <-h.msgQueue
+		h.data = msg.bytes
+
+		h.captureTime = msg.timestamp
+		if len(h.data) > 0 {
+			h.packetsSeen += 1
+		}
+		if h.packetsSeen < checkTLSPacketAmount && len(msg.bytes) > 5 { // packets with less than 5 bytes cause tlsx to panic
+			clientHello := tlsx.ClientHello{}
+			err := clientHello.Unmarshall(msg.bytes)
+			if err == nil {
+				rlog.Debugf("Detected TLS client hello with SNI %s\n", clientHello.SNI)
+				// TODO: Throws `panic: runtime error: invalid memory address or nil pointer dereference` error.
+				// numericPort, _ := strconv.Atoi(h.tcpID.DstPort)
+				// h.outboundLinkWriter.WriteOutboundLink(h.tcpID.SrcIP, h.tcpID.DstIP, numericPort, clientHello.SNI, TLSProtocol)
+			}
+		}
+	}
+	if !ok || len(h.data) == 0 {
+		return 0, io.EOF
+	}
+
+	l := copy(p, h.data)
+	h.data = h.data[l:]
+	return l, nil
+}
+
+func (h *tcpReader) run(wg *sync.WaitGroup) {
+	defer wg.Done()
+	b := bufio.NewReader(h)
+	err := h.extension.Dissector.Dissect(b, h.isClient, h.tcpID, h.counterPair, h.emitter)
+	if err != nil {
+		io.Copy(ioutil.Discard, b)
+	}
+}