[Linux] [C / C ++] Operation check memo of fork, thread and thread local variables

A note on the relationship between forks, threads and thread-local variables

fork and thread_local

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/wait.h>
#include <pthread.h>
#include <thread>

static __thread int val = 0;

int main()
{
	int p_id;
	int status;
	int return_code = 0;

	printf("%d :Parent process start val= %d\n", getpid(), val);

	val = 1000;

	auto th = std::thread([]{
			printf("%d :thread start val= %d\n", getpid(), val);
			val = 100;
			printf("%d :thread end val= %d\n", getpid(), val);
			});

	if ((p_id = fork()) == 0) {
		/*Child process*/
		printf("%d :Child process start val= %d\n", getpid(), val);
		sleep(1);
		val = 200;
		printf("%d :End child process val= %d\n", getpid(), val);
	} else {
		/*Parent process*/
		if (p_id == -1) {
			exit(EXIT_FAILURE);
		}

		wait(&status);
		printf("%d :Parent process termination val= %d\n", getpid(), val);
	}

	th.join();

	return 0;
}

Build and execution results

$ g++ -std=c++11 fork_and_threadlocal.cpp -pthread
$ ./a.out
11217 :Parent process start val= 0
11217 :thread start val= 0       //The initial value is 0 instead of 1000
11217 :thread end val= 100
11219 :Child process start val= 1000  //Copy the value in the parent process as it is when forking(Or still shared by COW?)Has been
11219 :End child process val= 200
11217 :Parent process termination val= 1000  //Unaffected by changes in child processes

The thread_local variable did not go to its initial state (0 this time), which is not surprising due to the specifications of fork. However, it seems that exec * is usually called after fork, so it may not be relevant when actually using it.

reference

Man page of FORK fork - Wikipedia [Technical chat-Process creation and cleanup (Linux C ++)-Tsubasa's HomePage](http://www.himajin2001.com/fswiki/wiki.cgi?page=%B5%BB%BD%D1%C5% AA% BB% A8% C3% CC-% A5% D7% A5% ED% A5% BB% A5% B9% A4% CE% C0% B8% C0% AE% A4% C8% B8% E5% BB% CF % CB% F6 (Linux% 20C% 2B% 2B% CA% D4)) Cost associated with multithreaded context switching --naoya's Hatena Diary