如何从子进程写入共享内存

Question

我正在试验子进程，因为这似乎是使无法在线程中使用的旧 C 库并行运行的唯一方法。

我的最小示例将三个整数值写入映射内存。在子进程中，我想更改这些值。更改最后一个(第三个)值有效，但是当仅更改第二个值时，我收到以下输出的错误：

initial mapped memory in parent process:                      000000000000000000000000000000010000000000000000000000000000001000000000000000000000000000000011
mapped memory in child process before writing:                000000000000000000000000000000010000000000000000000000000000001000000000000000000000000000000011
mapped memory in child process after writing:                 0000000000000000000000000000000100000000000000000000000000011011
mapped memory in parent process after child process finished: 0000000000000000000000000000000100000000000000000000000000011011
1
27
terminate called after throwing an instance of 'std::invalid_argument'
what():  bitset::_M_copy_from_ptr
Process finished with exit code 134 (interrupted by signal 6: SIGABRT)

最小的代码示例如下：

#include <iostream>
#include <vector>
#include <unistd.h>
#include <sys/wait.h>
#include <sys/shm.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <bitset>
#include <semaphore.h>
using namespace std;
int main(int /*argc*/, char* /*argv*/[])
{
const char * shm_name = "/a_shm_name";
const int SIZE = 4096;
std::vector<int> v{1, 2, 3};
int shm_fd = shm_open(shm_name, O_CREAT | O_RDWR, 0666);
ftruncate(shm_fd, 32 * v.size());
void * ptr0 = mmap(0, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0);
//write into the memory segment
void * ptr = ptr0;
for(int i = 0; i < v.size(); i++)
{
std::string s = std::bitset<32>(v[i]).to_string();
int count = sprintf((char*)ptr, "%s", s.c_str());
ptr = ptr + count;
}
cout << "initial mapped memory in parent process:                      " << (char*)ptr0 << endl << flush;
//fork
int n1 = fork();
if(n1 == 0)
{
//client process
shm_fd = shm_open(shm_name, O_CREAT | O_RDWR, 0666);
ptr0 = mmap(0, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0);
cout << "mapped memory in child process before writing:                " << (char*)ptr0 << endl << flush;
//write into the memory segment
//sem_wait(semId);
ptr = ptr0 + 32 * 1;
sprintf((char*)ptr, "%s", std::bitset<32>(27).to_string().c_str());
cout << "mapped memory in child process after writing:                 " << (char*)ptr0 << endl << flush;
exit(0);
}
else
{
//parent process
wait(nullptr);
}
//child process finished, back in the parent process
//client process
shm_fd = shm_open(shm_name, O_RDONLY, 0666);
ptr = mmap(0, SIZE, PROT_READ, MAP_SHARED, shm_fd, 0);
cout << "mapped memory in parent process after child process finished: " << (char*)ptr0 << endl << flush;
for(int i = 0; i < v.size(); i++)
{
cout << std::bitset<32>(std::string((char*)ptr, 32)).to_ulong() << endl;
ptr = ptr + 32;
}
munmap(ptr0, SIZE);
shm_unlink(shm_name);
return 0;
}

我有两个问题：

• 为什么我不能只更改第二个整数值？
• 是否有更好的技术在单独的子进程之间共享数据？

Answer 1

这与流程或mmap完全无关：具有相似结果的更简单的程序。

#include <vector>
#include <bitset>
#include <iostream>
int main() {
using std::cout, std::endl, std::flush;
const int SIZE = 4096;
std::vector<int> v{1, 2, 3};
char buffer[SIZE];
char * ptr0 = buffer;
//write into the memory segment
char * ptr = ptr0;
for(unsigned int i = 0; i < v.size(); i++)
{
std::string s = std::bitset<32>(v[i]).to_string();
int count = sprintf(ptr, "%s", s.c_str());
ptr = ptr + count;
}
cout << "initial memory:                                " << ptr0 << endl << flush;
//write into the memory segment
ptr = ptr0 + 32 * 1;
sprintf(ptr, "%s", std::bitset<32>(27).to_string().c_str());
cout << "mapped memory in child process after writing:  " << ptr0 << endl << flush;
}

(我将所有void*更改为char*，因为向void*添加数字不是标准C++，尽管某些编译器有一个扩展将其视为添加到char*。

sprintf在实际输出结束后写入一个空终止符。因此，当您使用它在子项中写入"第二个值"时，它会在"第三个值"的第一个字节中放入一个''。此外，您实际上一直依靠"第三个值"之后((char*)ptr0)[3*32]处的''，以便std::cout知道何时停止打印。

我建议使用仅复制字节的函数，而不是使用const char*的sprintf和operator<<，它们都适用于以 null 结尾的字符串：

std::string data = std::bitset<32>(value).to_string();
std::copy(data.begin(), data.end(), (char*) ptr);
cout.write((char*) ptr, 32);