【go/方法记录】cgo静态库编译以及使用dlv定位cgo崩溃问题

说在前面

• 测试环境：WSL2
• go版本：go version go1.23.1 linux/amd64
• gcc版本：gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
• cmake版本：3.22.1

文件树

├── buffer # go package
│   ├── buffer.go
│   ├── buffer_go.h
│   └── libbuffer.a
├── c # c/c++源代码
│   ├── CMakeLists.txt # cmake
│   ├── buffer # c/c++源代码
│   │   ├── buffer.h
│   │   ├── buffer_go.cpp
│   │   └── buffer_go.h
│   ├── lib # 静态库目录
│   │   └── libbuffer.a
│   └── main.cpp
└── main.go

静态库编译

• 这部分和go无关，按照正常的静态库编译走就行，这里我使用cmake进行编译
• buffer.h

  #ifndef __BUFFER_H__
  #define __BUFFER_H__
  #include <string>struct Buffer {std::string* s_;Buffer(int size) {this->s_ = new std::string(size, char('\0'));}~Buffer() {delete this->s_;}int Size() const {return this->s_->size();}char* Data() {return (char*)this->s_->data();}
  };#endif
  

• 然后我们需要对其进行封装，这部分可以参考这里
• buffer_go.h

  #define DLLIMPORT#ifdef __cplusplus
  extern "C" {
  #endiftypedef struct Buffer_T Buffer_T;Buffer_T* NewBuffer(int size);
  void DeleteBuffer(Buffer_T* p);
  int BufferSize(Buffer_T* p);#ifdef __cplusplus
  }
  #endif
  

• buffer_go.cpp

  #include "buffer.h"
  #include "buffer_go.h"#ifdef __cplusplus
  extern "C" {
  #endifstruct Buffer_T: Buffer {Buffer_T(int size): Buffer(size) {}~Buffer_T() {}
  };DLLIMPORT Buffer_T* NewBuffer(int size) {auto p = new Buffer_T(size);return p;
  }DLLIMPORT void DeleteBuffer(Buffer_T* p) {delete p;
  }DLLIMPORT int BufferSize(Buffer_T* p) {return p->Size();
  }#ifdef __cplusplus
  }
  #endif
  

• 然后就是cmake

  cmake_minimum_required(VERSION 3.2)
  project(Buffer)
  set(CMAKE_BUILD_TYPE "RELEASE")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")# 设置编译后库文件目录
  set(LIBRARY_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/lib)include_directories(${PROJECT_SOURCE_DIR}/buffer)
  # 添加编译可执行文件
  aux_source_directory(${PROJECT_SOURCE_DIR}/buffer SRC)add_library(buffer STATIC ${SRC})
  add_executable(buffer_exemain.cpp)# 链接主程序
  target_link_libraries(buffer_exePRIVATEbuffer)
  

• 执行编译

  cd ./c
  cmake .
  make
  

  动态库将生成在目录./c/lib

cgo使用

• 首先将buffer_go.h、libbuffer.a拷贝到./buffer目录下，并创建buffer.go文件。注意，以下注释中的#include "buffer_go.h"和import "C"不要有空行
• buffer.go

  package buffer// #cgo CFLAGS: -I.
  // #cgo CXXFLAGS: -std=c++11
  // #cgo LDFLAGS: -L./ -lbuffer -lstdc++
  //
  // #include "buffer_go.h"
  import "C"type BufferT C.Buffer_Tfunc NewBuffer(size int) *BufferT {p := C.NewBuffer(C.int(size))return (*BufferT)(p)
  }func DeleteBuffer(p *BufferT) {C.DeleteBuffer((*C.Buffer_T)(p))
  }func BufferSize(p *BufferT) C.int {return C.BufferSize((*C.Buffer_T)(p))
  }
  

• 然后是main.go

  package mainimport ("fmt"
  )func main() {b := buffer.NewBuffer(2)fmt.Println(buffer.BufferSize(b))buffer.DeleteBuffer(b)
  }
  

• 执行程序输出

  ~/cgotest$ go run main.go
  2
  

崩溃模拟

• 正常go中的崩溃，例如空指针调用等是可以通过recover处理的，例如

  func main() {go func() {defer func() {if err := recover(); err != nil {fmt.Println("err: ", string(debug.Stack()))}}()time.Sleep(5 * time.Second)var b uint32var a uint32_ = b / a}()var i intfor {time.Sleep(time.Second)i++fmt.Println(i)}
  }
  

  可以看到进程可以继续运行

  ~/cgotest$ go run main.go
  1
  2
  3
  4
  err:  goroutine 18 [running]:
  runtime/debug.Stack()/usr/local/go/src/runtime/debug/stack.go:26 +0x5e
  main.main.func1.1()/home/xx/cgotest/main.go:14 +0x2a
  panic({0x49cb60?, 0x54d840?})/usr/local/go/src/runtime/panic.go:785 +0x132
  main.main.func1()/home/xx/cgotest/main.go:23 +0x3f
  created by main.main in goroutine 1/home/xx/cgotest/main.go:11 +0x1a5
  6
  

• 而对于cgo中的崩溃，目前go是没法进行recover的，会导致进程直接崩溃

  package mainimport ("cgotest/cgotest/buffer""fmt""runtime/debug""time"
  )func main() {go func() {defer func() {if err := recover(); err != nil {fmt.Println("err: ", string(debug.Stack()))}}()time.Sleep(5 * time.Second)b := buffer.NewBuffer(2)buffer.DeleteBuffer(b)b = nilfmt.Println(buffer.BufferSize(b))}()var i intfor {time.Sleep(time.Second)i++fmt.Println(i)}
  }
  

  ~/cgotest$ go run main.go
  1
  2
  3
  4
  SIGSEGV: segmentation violation
  PC=0x4930b4 m=5 sigcode=1 addr=0x0
  signal arrived during cgo executiongoroutine 6 gp=0xc000007dc0 m=5 mp=0xc000100008 [syscall]:
  runtime.cgocall(0x492550, 0xc000076750)goroutine 6 gp=0xc000007dc0 m=5 mp=0xc000100008 [syscall]:
  runtime.cgocall(0x492550, 0xc000076750)/usr/local/go/src/runtime/cgocall.go:167 +0x4b fp=0xc000076728 sp=0xc0000766f0 pc=0x462f8b
  cgotest/cgotest/buffer._Cfunc_BufferSize(0x0)_cgo_gotypes.go:52 +0x47 fp=0xc000076750 sp=0xc000076728 pc=0x475b87
  main.main.func1.BufferSize.3(0x0)/home/xx/cgotest/buffer/buffer.go:22 +0x3b fp=0xc000076788 sp=0xc000076750 pc=0x4923fb
  cgotest/cgotest/buffer.BufferSize(...)/home/xx/cgotest/buffer/buffer.go:22
  main.main.func1()/home/xx/cgotest/main.go:29 +0x6c fp=0xc0000767e0 sp=0xc000076788 pc=0x49234c
  runtime.goexit({})/usr/local/go/src/runtime/asm_amd64.s:1700 +0x1 fp=0xc0000767e8 sp=0xc0000767e0 pc=0x46e601
  created by main.main in goroutine 1/home/xx/cgotest/main.go:12 +0x1a
  

• 可以看到进程退出并输出了堆栈信息

使用dlv定位崩溃

• 虽然崩溃时的堆栈信息有标准输出，但是有时标准输出中的堆栈信息可能会被覆盖掉，少了最前面的信息，这个时候可以开启coredump并使用dlv来定位
• 启用coredump

  ulimit -c unlimited
  GOTRACEBACK=crash go run main.go
  

  在进程崩溃后会生成coredump文件
• 查看coredump文件目录

  ~/cgotest$ cat /proc/sys/kernel/core_pattern
  /mnt/wslg/dumps/core.%e
  

  可以看到文件生成在/mnt/wslg/dumps/

  ~/cgotest$ ll /mnt/wslg/dumps/
  total 6684
  -rw------- 1 le   le   83562496 Sep 21 16:04 core.main
  

• 安装dlv

  go install github.com/go-delve/delve/cmd/dlv@latest
  

  可以在gopath文件夹下找到

  ~/cgotest$ /home/xx/go/bin/
  dlv          gopls        staticcheck
  

• 使用

  # dlv core 二进制程序 coredump文件
  ~/go/bin/dlv core cgotest /mnt/wslg/dumps/core.main 
  

  结果如下

  ~/cgotest$ ~/go/bin/dlv core cgotest /mnt/wslg/dumps/core.main 
  Type 'help' for list of commands.
  (dlv) bt0  0x000000000046fde1 in runtime.raiseat /usr/local/go/src/runtime/sys_linux_amd64.s:1541  0x000000000044b625 in runtime.dieFromSignalat /usr/local/go/src/runtime/signal_unix.go:9422  0x000000000044bc86 in runtime.sigfwdgoat /usr/local/go/src/runtime/signal_unix.go:11543  0x000000000044a625 in runtime.sigtrampgoat /usr/local/go/src/runtime/signal_unix.go:4324  0x000000000046fde1 in runtime.raiseat /usr/local/go/src/runtime/sys_linux_amd64.s:1535  0x000000000044b625 in runtime.dieFromSignalat /usr/local/go/src/runtime/signal_unix.go:9426  0x000000000044b1a6 in runtime.crashat /usr/local/go/src/runtime/signal_unix.go:1031
  Sending output to pager...0  0x000000000046fde1 in runtime.raiseat /usr/local/go/src/runtime/sys_linux_amd64.s:1541  0x000000000044b625 in runtime.dieFromSignalat /usr/local/go/src/runtime/signal_unix.go:9422  0x000000000044bc86 in runtime.sigfwdgoat /usr/local/go/src/runtime/signal_unix.go:11543  0x000000000044a625 in runtime.sigtrampgoat /usr/local/go/src/runtime/signal_unix.go:4324  0x000000000046fde1 in runtime.raiseat /usr/local/go/src/runtime/sys_linux_amd64.s:1535  0x000000000044b625 in runtime.dieFromSignalat /usr/local/go/src/runtime/signal_unix.go:9426  0x000000000044b1a6 in runtime.crashat /usr/local/go/src/runtime/signal_unix.go:10317  0x000000000044b1a6 in runtime.sighandlerat /usr/local/go/src/runtime/signal_unix.go:8068  0x000000000044a726 in runtime.sigtrampgoat /usr/local/go/src/runtime/signal_unix.go:4909  0x00000000004930b4 in ???at ?:-1
  10  0x000000000049256b in C._cgo_a4c0d8419d5e_Cfunc_BufferSizeat /tmp/go-build/buffer.cgo2.c:55
  11  0x000000000046e284 in runtime.asmcgocallat /usr/local/go/src/runtime/asm_amd64.s:923
  12  0x0000000000000000 in ???at ?:-1
  13  0x0000000000462fb5 in runtime.cgocallat /usr/local/go/src/runtime/cgocall.go:185
  14  0x0000000000475b87 in cgotest/cgotest/buffer._Cfunc_BufferSizeat _cgo_gotypes.go:52
  15  0x00000000004923fb in main.main.func1.BufferSize.3at ./buffer/buffer.go:22
  16  0x0000000000492345 in cgotest/cgotest/buffer.BufferSizeat ./buffer/buffer.go:22
  17  0x0000000000492345 in main.main.func1at ./main.go:29
  18  0x000000000046e601 in runtime.goexitat /usr/local/go/src/runtime/asm_amd64.s:1700
  

参考

• dlv安装