// we need this so sched.h exports unshare and CLONE_*
#define _GNU_SOURCE
#include <sched.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>
#include <signal.h>

pid_t pid_child;

void fatal(const char* str, int errcode)
{
  printf("%s (%d)\n", str, errcode);
  exit(errcode);
}

void drop_root(void)
{
  /// Drop root privileges
  // First group then user because we might not
  // be able to drop group once we dropped user
  gid_t gid = getgid();
  if (setresgid(-1,gid,gid) == -1)
    fatal("Failed to drop root privileges with setresgid", errno);

  uid_t uid = getuid();
  if (setresuid(-1,uid,uid) == -1)
    fatal("Failed to drop root privileges with setresuid", errno);

  // sanity check
  if (seteuid(0) != -1)
    fatal("Sanity check failed. Able to regain root", 42);
}

struct sigaction forward_signal_descriptor;
void forward_signal(int sig)
{
  if (kill(pid_child, sig) == -1)
    {
      if (sig == SIGTERM)
        exit(1);
    }
}

char** argdup(int argc, const char** argv)
{
  char** newargs = malloc(sizeof(char*) * (argc+1));
  for (size_t i = 0; i < argc; i++)
    {
      newargs[i] = strdup(argv[i]);
    }
  newargs[argc] = NULL;
  return newargs;
}

int main(int argc, const char** argv)
{
  int err;

  forward_signal_descriptor.sa_flags = SA_RESTART;
  forward_signal_descriptor.sa_handler = &forward_signal;

  if (argc == 1)
    {
      fprintf(stderr,"Usage: %s PROGRAM ARGUMENTS...\n"
              "Run command within its own pid namespace. Integrated init process.\n",
              argv[0]);
      return 0;
    }

  // next fork shall be in a new pid namespace
  if (unshare(CLONE_NEWPID) != 0)
    {
      fatal("Failed to unshare pid namespace", errno);
    }

  // Drop root privileges, we only needed those for the unshare call.
  drop_root();

  pid_t pid = fork();

  if (pid == -1)
    {
      fatal("Failed to fork", errno);
    }

  if (pid != 0)
    {
      /// Head process
      // Wait for the init process in the PID namespace to terminate and forward its exit code.
      // Also forward SIGTERM signals towards that init process.

      // Setup signal handler to forward SIGTERM
      pid_child = pid;
      if (sigaction(SIGTERM, &forward_signal_descriptor, NULL) == -1)
        {
          printf("Unable to setup signal handler in head\n");
        }
      // parent waits for child then exits
      // Could be interrupt due to a signal. Retry in that case.
      int status;
      if (waitpid(pid, &status, 0) == -1)
        {
          fatal("Failed to wait for init process", errno);
        }

      return WEXITSTATUS(status);
    }
  else
    {
      // Child should be in new pid namespace and
      // functions as the the init process
      // it needs to fork again then wait for any child.
      // if the forked child exits then exit.

      pid = fork();
      if (pid == -1)
        {
          fatal("Failed to fork in init process", errno);
        }

      if (pid != 0)
        {
          /// Init process
          // This part of the program runs as first process in the pid namespace
          // When this terminates then Linux terminates all remaining processes
          // in the PID namespace. As we want this to happen when our first child
          // terminates, we wait for our first child to terminate before terminating
          // ourselves.
          // As first process in the PID namespace, this also functions as adopting parent
          // for orphaned processes in the PID namespace and therefore has to wait for
          // any child process and then check if the a child process that has terminated
          // is the one we were waiting for.

          pid_t first_child = pid;
          pid_t exited_child;
          int child_status;

          // Setup forward for SIGTERM
          pid_child = first_child;
          if (sigaction(SIGTERM, &forward_signal_descriptor, NULL) == -1)
            {
              fatal("Unable to setup signal forward in init", 1);
            }

          // wait could be interrupt due to a signal. In that case just call wait again.
          do {
            exited_child = wait(&child_status);
            err = errno;
          } while (!(exited_child == first_child || (exited_child == -1 && err == ECHILD)));

          if (exited_child == -1)
            {
              return err;
            }
          else
            {
              int exit_code = WEXITSTATUS(child_status);
              return exit_code;
            }
        }
      else
        {
          // First child of init process. do exec here
          // use cli arguments for subprocess. skip 0 as it's our programs name.

          char** newargs = argdup(argc-1, &argv[1]);

          if (execvp(newargs[0], newargs) == -1)
            {
              fatal("Failed to exec", errno);
            }
        }
    }
}