【腾讯犀牛鸟开源课题实战】wireshark协议解析 (#36)

* feat: add wireshark support for trpc

* docs: add wireshark usage

---------

Co-authored-by: weimch <[email protected]>

Author: RunoobACMer

README.zh_CN.md

@@ -49,6 +49,10 @@ tRPC是基于插件化理念设计的一款支持多语言、高性能的RPC框
     - [用户指南](https://github.com/trpc-group/trpc-go/tree/main/docs/README.zh_CN.md)
     - [代码示例](https://github.com/trpc-group/trpc-go/tree/main/examples)
 
+## 如何用wireshark分析tRPC协议
+
+参考 [docs/zh/wireshark_trpc.md](docs/zh/wireshark_trpc.md)。
+
 ## 如何参与贡献
 
 非常欢迎大家给tRPC做贡献!

docs/images/wireshark/pic1.png

@@ -0,0 +1,82 @@
+# tRPC Wireshark解析器
+
+## 前言
+
+`tool/wireshark_trpc.lua` 是 tRPC 协议的 Wireshark 解析器，提供了对 tRPC 协议头、业务 pb 的解析能力。
+
+当前使用有如下限制：
+
+1. 暂不支持解析tRPC流式协议；
+2. 暂不支持attachment；
+3. 暂不支持UDP；
+4. 不支持解析开启压缩后的pb数据（业务pb数据会被压缩）；
+
+## 用法
+
+### tcpdump抓包
+
+执行下面的tcpdump抓包
+
+```bash
+# 10001 替换为服务端端口
+tcpdump -iany port 10001 -w trpc_packet.pcap
+```
+
+**注意：** 有时可能因为机器环境问题，导致tcpdump抓包被截断，比如只抓了请求包的部分，wireshark使用lua脚本解析会失败，这时可以抓包时使用 `-s xxx` 防止截断，代表调整低于xxx字节的packet不截断。
+
+### Wireshark配置
+
+1. 从选项卡 `About Wireshark` -> `Floders` 查看 `Personal Lua Plugins` 的目录。
+
+<img src="../images/wireshark/pic1.png" width="600" />
+
+2. 进入 `Personal Lua Plugins` 目录，放入 [wireshark_trpc.lua](../../tool/wireshark_trpc.lua)。
+
+<img src="../images/wireshark/pic2.png" width="400" />
+
+3. 设置proto扫描文件目录，并开启基于protobuf的解析。
+
+- 此文件目录下需要放置 [trpc.proto](../../trpc/trpc.proto)以及业务pb，比如此处我们使用 tRPC-Cpp 示例的 [helloworld.proto](https://github.com/trpc-group/trpc-cpp/blob/main/examples/helloworld/helloworld.proto)。
+- 需要勾选 `Load .proto files on startup.`、`Dissect Protobuf fields as Wireshark fields.`、`Show details of message, fields and enums.`、`Show all fields of bytes type as string.` 以通过protobuf解析器来解析字段。
+
+  <img src="../images/wireshark/pic3.png" width="400" />
+
+  <img src="../images/wireshark/pic4.png" width="600" />
+
+  <img src="../images/wireshark/pic5.png" width="400" />
+
+### Wireshark加载tcpdump抓包
+
+加载包之后，如果没有看到 `Protocol` 显示 `tRPC`，需要强制加载下lua脚本。
+
+<img src="../images/wireshark/pic6.png" width="200" />
+
+如果都成功，能看如下图所示tRPC协议以及业务pb都被正常解析了。
+
+<img src="../images/wireshark/pic7.png" width="600" />
+
+<img src="../images/wireshark/pic8.png" width="600" />
+
+### 筛选指定请求和响应
+
+有时会发现并发很多个请求，可能有某几个请求会调用失败，为了进一步排查，需要更多tRPC协议/业务pb字段等信息，这时候可以通过 protobuf 字段来筛选出指定请求/响应。
+
+Wireshark 支持通过 protobuf 字段做筛选，业务可以选择使用trpc协议头的 request_id 或者业务 pb 某个字段做匹配来查看指定请求和响应的交互情况。
+
+下面是根据trpc协议头的 `requet_id=3` 筛选指定请求和响应的示例，大家可根据自身情况指定筛选条件。
+
+```text
+protobuf.field.name == "request_id" and protobuf.field.value == 3
+```
+
+<img src="../images/wireshark/pic9.png" width="600" />
+
+### 忽略tcp的控制帧
+
+我们更多的会关注协议包的交互情况，而不关注tcp控制帧（tcp握手/挥手等），这时可以通过设置下面的筛选条件清除tcp控制帧的显示。
+
+```text
+protobuf
+```
+
+<img src="../images/wireshark/pic10.png" width="600" />

docs/images/wireshark/pic10.png

@@ -0,0 +1,236 @@
+--
+--
+-- Tencent is pleased to support the open source community by making tRPC available.
+--
+-- Copyright (C) 2024 THL A29 Limited, a Tencent company.
+-- All rights reserved.
+--
+-- If you have downloaded a copy of the tRPC source code from Tencent,
+-- please note that tRPC source code is licensed under the  Apache 2.0 License,
+-- A copy of the Apache 2.0 License is included in this file.
+--
+--
+-- tRPC is licensed under the Apache 2.0 License, and includes source codes from
+-- the following components:
+-- 1. incubator-brpc
+-- Copyright (C) 2019 The Apache Software Foundation
+-- incubator-brpc is licensed under the Apache 2.0 License.
+--
+--
+
+local protocol_name    = "trpc" 
+local trpc_proto       = Proto(protocol_name, "tRPC Protocol Dissector")
+
+local field_magic      = ProtoField.uint16(protocol_name .. ".magic",       "Magic",       base.HEX)
+local field_type       = ProtoField.uint8(protocol_name .. ".type",         "Packet Type", base.DEC)
+local field_stream     = ProtoField.uint8(protocol_name .. ".stream",       "Stream Type", base.DEC)
+local field_total_size = ProtoField.uint32(protocol_name .. ".total_size",  "Total Size",  base.DEC)
+local field_header_size= ProtoField.uint16(protocol_name .. ".header_size", "Header Size", base.DEC)
+local field_unique_id  = ProtoField.uint32(protocol_name .. ".unique_id",   "Unique ID",   base.DEC)
+local field_version    = ProtoField.uint8(protocol_name .. ".version",      "Version",     base.DEC)
+local field_reserved   = ProtoField.uint8(protocol_name .. ".reserved",     "Reserved",    base.DEC)
+trpc_proto.fields      = {field_magic, field_type, field_stream, field_total_size, field_header_size, field_unique_id, field_version, field_reserved}
+
+local MAGIC_CODE_TRPC = "0930"
+local PROTO_HEADER_LENGTH = 16
+
+local tcp_src_port     = Field.new("tcp.srcport")
+local tcp_dst_port     = Field.new("tcp.dstport")
+local tcp_stream       = Field.new("tcp.stream")
+
+local proto_f_protobuf_field_name  = Field.new("protobuf.field.name")
+local proto_f_protobuf_field_value = Field.new("protobuf.field.value")
+
+local data_dissector     = Dissector.get("data")
+local protobuf_dissector = Dissector.get("protobuf")
+
+----------------------------------------
+-- declare functions
+local check_length  = function() end
+local dissect_proto = function() end
+
+----------------------------------------
+-- main dissector
+function trpc_proto.dissector(tvbuf, pktinfo, root)
+    local pktlen = tvbuf:len()
+
+    local bytes_consumed = 0
+
+    while bytes_consumed < pktlen do
+        local result = dissect_proto(tvbuf, pktinfo, root, bytes_consumed)
+
+        if result > 0 then
+            bytes_consumed = bytes_consumed + result
+        elseif result == 0 then
+            -- hit an error
+            return 0
+        else
+            pktinfo.desegment_offset = bytes_consumed
+            -- require more bytes
+            pktinfo.desegment_len = -result
+
+            return pktlen
+        end
+    end
+
+    return bytes_consumed
+end
+
+--------------------------------------------------------------------------------
+-- heuristic
+-- tcp_stream_id <-> {client_port, server_port, {request_id<->method_name}}
+local stream_map = {}
+local function heur_dissect_proto(tvbuf, pktinfo, root)
+    -- dynmaic decide client or server data
+    -- by first tcp syn frame
+    local f_src_port = tcp_src_port()()
+    local f_dst_port = tcp_dst_port()()
+    local stream_n = tcp_stream().value
+    if stream_map[stream_n] == nil then
+        stream_map[stream_n] = {f_src_port, f_dst_port, {}}
+    end
+
+    if (tvbuf:len() < PROTO_HEADER_LENGTH) then
+        return false
+    end
+
+    local magic = tvbuf:range(0, 2):bytes():tohex()
+    -- for range dissectors
+    if magic ~= MAGIC_CODE_TRPC then
+        return false
+    end
+
+    trpc_proto.dissector(tvbuf, pktinfo, root)
+
+    pktinfo.conversation = trpc_proto
+
+    return true
+end
+
+trpc_proto:register_heuristic("tcp", heur_dissect_proto)
+
+--------------------------------------------------------------------------------
+
+-- check packet length, return length of packet if valid
+check_length = function(tvbuf, offset)
+    local msglen = tvbuf:len() - offset
+
+    if msglen ~= tvbuf:reported_length_remaining(offset) then
+        -- captured packets are being sliced/cut-off, so don't try to desegment/reassemble
+        LM_WARN("Captured packet was shorter than original, can't reassemble")
+        return 0
+    end
+
+    if msglen < PROTO_HEADER_LENGTH then
+        -- we need more bytes, so tell the main dissector function that we
+        -- didn't dissect anything, and we need an unknown number of more
+        -- bytes (which is what "DESEGMENT_ONE_MORE_SEGMENT" is used for)
+        return -DESEGMENT_ONE_MORE_SEGMENT
+    end
+
+    -- if we got here, then we know we have enough bytes in the Tvb buffer
+    -- to at least figure out whether this is valid trpc packet
+
+    local magic = tvbuf:range(offset, 2):bytes():tohex()
+    if magic ~= MAGIC_CODE_TRPC then
+        return 0
+    end
+
+    local packet_size = tvbuf:range(offset+4, 4):uint()
+    if msglen < packet_size then
+        -- Need more bytes to desegment full trpc packet
+        return -(packet_size - msglen)
+    end
+
+    return packet_size
+end
+
+--------------------------------------------------------------------------------
+
+dissect_proto = function(tvbuf, pktinfo, root, offset)
+    local len = check_length(tvbuf, offset)
+    if len <= 0 then
+        return len
+    end
+
+    -- update 'Protocol' field
+    if offset == 0 then
+        pktinfo.cols.protocol:set("tRPC")
+    end
+
+    local f_src_port = tcp_src_port()()
+    local f_dst_port = tcp_dst_port()()
+
+    local direction
+    local stream_n = tcp_stream().value
+    if f_src_port == stream_map[stream_n][1] then
+        pktinfo.private["pb_msg_type"] = "message,trpc.RequestProtocol"
+        direction = "request"
+    end
+    if f_src_port == stream_map[stream_n][2] then
+        pktinfo.private["pb_msg_type"] = "message,trpc.ResponseProtocol"
+        direction = "response"
+    end
+
+    -- check packet length, 
+    local magic_value       = tvbuf(offset, 2)
+    local type_value        = tvbuf(offset+2, 1)
+    local stream_value      = tvbuf(offset+3, 1)
+    local total_size_value  = tvbuf(offset+4, 4)
+    local header_size_value = tvbuf(offset+8, 2)
+    local unique_id_value   = tvbuf(offset+10, 4)
+    local version_value     = tvbuf(offset+14, 1)
+    local reserved_value    = tvbuf(offset+15, 1)
+
+    local header_length   = header_size_value:uint()
+    local total_length    = total_size_value:uint()
+    local tree            = root:add(trpc_proto, tvbuf:range(offset, len), "tRPC Protocol Data")
+
+    data_dissector:call(tvbuf, pktinfo, tree)
+
+    local t = tree:add(trpc_proto, tvbuf)
+    t:add(field_magic, magic_value)
+    t:add(field_type, type_value)
+    t:add(field_stream, stream_value)
+    t:add(field_total_size, total_size_value)
+    t:add(field_header_size, header_size_value)
+    t:add(field_unique_id, unique_id_value)
+    t:add(field_version, version_value)
+    t:add(field_reserved, reserved_value)
+
+    -- solve the problem of parsing errors when multiple RPCs are included in a packet
+    local protobuf_field_names = { proto_f_protobuf_field_name() }
+    local pre_field_nums = #protobuf_field_names
+    pcall(Dissector.call, protobuf_dissector, tvbuf(offset+16, header_length):tvb(), pktinfo, tree)
+
+    -- Add bussiness rpc pb
+    -- Get invoke rpc method name from trpc.RequestProtocol
+    protobuf_field_names = { proto_f_protobuf_field_name() }
+    local cur_field_nums = #protobuf_field_names
+    local protobuf_field_values = { proto_f_protobuf_field_value() }
+    local method
+    -- default request id
+    local request_id = 0
+    for k = pre_field_nums + 1, cur_field_nums do
+        local v = protobuf_field_names[k]
+        if v.value == "func" then
+            method = protobuf_field_values[k].range:string(ENC_UTF8)
+        elseif v.value == "request_id" then
+            request_id = protobuf_field_values[k].range:uint()
+        end
+    end
+
+    local tvb_body = tvbuf:range(offset + 16 + header_length, total_length - header_length - 16):tvb()
+    if method ~= nil then
+        -- only req contains method, correlate it with request id so that response protocol can use.
+        stream_map[stream_n][3][request_id] = method
+        pktinfo.private["pb_msg_type"] = "application/trpc," .. method .. "," .. direction
+    else
+        -- get method for the same request id
+        method = stream_map[stream_n][3][request_id]
+        pktinfo.private["pb_msg_type"] = "application/trpc," .. method .. "," .. direction
+    end
+    pcall(Dissector.call, protobuf_dissector, tvb_body, pktinfo, tree)
+
+    return total_length
+end