Quelle  pidfd_test.c   Sprache: C

/* SPDX-License-Identifier: GPL-2.0 */

#define _GNU_SOURCE
#include <errno.h>
#include <fcntl.h>
#include <linux/types.h>
#include <pthread.h>
#include <sched.h>
#include <signal.h>
#include <stdio.h>
#include <stdbool.h>
#include <stdlib.h>
#include <string.h>
#include <syscall.h>
#include <sys/epoll.h>
#include <sys/mman.h>
#include <sys/mount.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>

#include "pidfd.h"
#include "../kselftest.h"

#define str(s) _str(s)
#define _str(s) #s
#define CHILD_THREAD_MIN_WAIT 3 /* seconds */

#define MAX_EVENTS 5

static bool have_pidfd_send_signal;

static pid_t pidfd_clone(int flags, int *pidfd, int (*fn)(void *))
{
 size_t stack_size = 1024;
 char *stack[1024] = { 0 };

#ifdef __ia64__
 return __clone2(fn, stack, stack_size, flags | SIGCHLD, NULL, pidfd);
#else
 return clone(fn, stack + stack_size, flags | SIGCHLD, NULL, pidfd);
#endif
}

static pthread_t signal_received;

static void set_signal_received_on_sigusr1(int sig)
{
 if (sig == SIGUSR1)
  signal_received = pthread_self();
}

static int send_signal(int pidfd)
{
 int ret = 0;

 if (sys_pidfd_send_signal(pidfd, SIGUSR1, NULL, 0) < 0) {
  ret = -EINVAL;
  goto exit;
 }

 if (signal_received != pthread_self()) {
  ret = -EINVAL;
  goto exit;
 }

exit:
 signal_received = 0;
 return ret;
}

static void *send_signal_worker(void *arg)
{
 int pidfd = (int)(intptr_t)arg;
 int ret;

 /* We forward any errors for the caller to handle. */
 ret = send_signal(pidfd);
 return (void *)(intptr_t)ret;
}

/*
 * Straightforward test to see whether pidfd_send_signal() works is to send
 * a signal to ourself.
 */
static int test_pidfd_send_signal_simple_success(void)
{
 int pidfd;
 const char *test_name = "pidfd_send_signal send SIGUSR1";
 pthread_t thread;
 void *thread_res;
 int err;

 if (!have_pidfd_send_signal) {
  ksft_test_result_skip(
   "%s test: pidfd_send_signal() syscall not supported\n",
   test_name);
  return 0;
 }

 signal(SIGUSR1, set_signal_received_on_sigusr1);

 /* Try sending a signal to ourselves via /proc/self. */
 pidfd = open("/proc/self", O_DIRECTORY | O_CLOEXEC);
 if (pidfd < 0)
  ksft_exit_fail_msg(
   "%s test: Failed to open process file descriptor\n",
   test_name);
 err = send_signal(pidfd);
 if (err)
  ksft_exit_fail_msg(
   "%s test: Error %d on sending pidfd signal\n",
   test_name, err);
 close(pidfd);

 /* Now try the same thing only using PIDFD_SELF_THREAD_GROUP. */
 err = send_signal(PIDFD_SELF_THREAD_GROUP);
 if (err)
  ksft_exit_fail_msg(
   "%s test: Error %d on PIDFD_SELF_THREAD_GROUP signal\n",
   test_name, err);

 /*
 * Now try the same thing in a thread and assert thread ID is equal to
 * worker thread ID.
 */
 if (pthread_create(&thread, NULL, send_signal_worker,
      (void *)(intptr_t)PIDFD_SELF_THREAD))
  ksft_exit_fail_msg("%s test: Failed to create thread\n",
       test_name);
 if (pthread_join(thread, &thread_res))
  ksft_exit_fail_msg("%s test: Failed to join thread\n",
       test_name);
 err = (int)(intptr_t)thread_res;
 if (err)
  ksft_exit_fail_msg(
   "%s test: Error %d on PIDFD_SELF_THREAD signal\n",
   test_name, err);

 ksft_test_result_pass("%s test: Sent signal\n", test_name);
 return 0;
}

static int test_pidfd_send_signal_exited_fail(void)
{
 int pidfd, ret, saved_errno;
 char buf[256];
 pid_t pid;
 const char *test_name = "pidfd_send_signal signal exited process";

 if (!have_pidfd_send_signal) {
  ksft_test_result_skip(
   "%s test: pidfd_send_signal() syscall not supported\n",
   test_name);
  return 0;
 }

 pid = fork();
 if (pid < 0)
  ksft_exit_fail_msg("%s test: Failed to create new process\n",
       test_name);

 if (pid == 0)
  _exit(EXIT_SUCCESS);

 snprintf(buf, sizeof(buf), "/proc/%d", pid);

 pidfd = open(buf, O_DIRECTORY | O_CLOEXEC);

 ret = wait_for_pid(pid);
 ksft_print_msg("waitpid WEXITSTATUS=%d\n", ret);

 if (pidfd < 0)
  ksft_exit_fail_msg(
   "%s test: Failed to open process file descriptor\n",
   test_name);

 ret = sys_pidfd_send_signal(pidfd, 0, NULL, 0);
 saved_errno = errno;
 close(pidfd);
 if (ret == 0)
  ksft_exit_fail_msg(
   "%s test: Managed to send signal to process even though it should have failed\n",
   test_name);

 if (saved_errno != ESRCH)
  ksft_exit_fail_msg(
   "%s test: Expected to receive ESRCH as errno value but received %d instead\n",
   test_name, saved_errno);

 ksft_test_result_pass("%s test: Failed to send signal as expected\n",
         test_name);
 return 0;
}

/*
 * Maximum number of cycles we allow. This is equivalent to PID_MAX_DEFAULT.
 * If users set a higher limit or we have cycled PIDFD_MAX_DEFAULT number of
 * times then we skip the test to not go into an infinite loop or block for a
 * long time.
 */
#define PIDFD_MAX_DEFAULT 0x8000

static int test_pidfd_send_signal_recycled_pid_fail(void)
{
 int i, ret;
 pid_t pid1;
 const char *test_name = "pidfd_send_signal signal recycled pid";

 if (!have_pidfd_send_signal) {
  ksft_test_result_skip(
   "%s test: pidfd_send_signal() syscall not supported\n",
   test_name);
  return 0;
 }

 ret = unshare(CLONE_NEWPID);
 if (ret < 0) {
  if (errno == EPERM) {
   ksft_test_result_skip("%s test: Unsharing pid namespace not permitted\n",
           test_name);
   return 0;
  }
  ksft_exit_fail_msg("%s test: Failed to unshare pid namespace\n",
       test_name);
 }

 ret = unshare(CLONE_NEWNS);
 if (ret < 0) {
  if (errno == EPERM) {
   ksft_test_result_skip("%s test: Unsharing mount namespace not permitted\n",
           test_name);
   return 0;
  }
  ksft_exit_fail_msg("%s test: Failed to unshare mount namespace\n",
       test_name);
 }

 ret = mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0);
 if (ret < 0)
  ksft_exit_fail_msg("%s test: Failed to remount / private\n",
       test_name);

 /* pid 1 in new pid namespace */
 pid1 = fork();
 if (pid1 < 0)
  ksft_exit_fail_msg("%s test: Failed to create new process\n",
       test_name);

 if (pid1 == 0) {
  char buf[256];
  pid_t pid2;
  int pidfd = -1;

  (void)umount2("/proc", MNT_DETACH);
  ret = mount("proc", "/proc", "proc", 0, NULL);
  if (ret < 0)
   _exit(PIDFD_ERROR);

  /* grab pid PID_RECYCLE */
  for (i = 0; i <= PIDFD_MAX_DEFAULT; i++) {
   pid2 = fork();
   if (pid2 < 0)
    _exit(PIDFD_ERROR);

   if (pid2 == 0)
    _exit(PIDFD_PASS);

   if (pid2 == PID_RECYCLE) {
    snprintf(buf, sizeof(buf), "/proc/%d", pid2);
    ksft_print_msg("pid to recycle is %d\n", pid2);
    pidfd = open(buf, O_DIRECTORY | O_CLOEXEC);
   }

   if (wait_for_pid(pid2))
    _exit(PIDFD_ERROR);

   if (pid2 >= PID_RECYCLE)
    break;
  }

  /*
 * We want to be as predictable as we can so if we haven't been
 * able to grab pid PID_RECYCLE skip the test.
 */
  if (pid2 != PID_RECYCLE) {
   /* skip test */
   close(pidfd);
   _exit(PIDFD_SKIP);
  }

  if (pidfd < 0)
   _exit(PIDFD_ERROR);

  for (i = 0; i <= PIDFD_MAX_DEFAULT; i++) {
   char c;
   int pipe_fds[2];
   pid_t recycled_pid;
   int child_ret = PIDFD_PASS;

   ret = pipe2(pipe_fds, O_CLOEXEC);
   if (ret < 0)
    _exit(PIDFD_ERROR);

   recycled_pid = fork();
   if (recycled_pid < 0)
    _exit(PIDFD_ERROR);

   if (recycled_pid == 0) {
    close(pipe_fds[1]);
    (void)read(pipe_fds[0], &c, 1);
    close(pipe_fds[0]);

    _exit(PIDFD_PASS);
   }

   /*
 * Stop the child so we can inspect whether we have
 * recycled pid PID_RECYCLE.
 */
   close(pipe_fds[0]);
   ret = kill(recycled_pid, SIGSTOP);
   close(pipe_fds[1]);
   if (ret) {
    (void)wait_for_pid(recycled_pid);
    _exit(PIDFD_ERROR);
   }

   /*
 * We have recycled the pid. Try to signal it. This
 * needs to fail since this is a different process than
 * the one the pidfd refers to.
 */
   if (recycled_pid == PID_RECYCLE) {
    ret = sys_pidfd_send_signal(pidfd, SIGCONT,
           NULL, 0);
    if (ret && errno == ESRCH)
     child_ret = PIDFD_XFAIL;
    else
     child_ret = PIDFD_FAIL;
   }

   /* let the process move on */
   ret = kill(recycled_pid, SIGCONT);
   if (ret)
    (void)kill(recycled_pid, SIGKILL);

   if (wait_for_pid(recycled_pid))
    _exit(PIDFD_ERROR);

   switch (child_ret) {
   case PIDFD_FAIL:
    /* fallthrough */
   case PIDFD_XFAIL:
    _exit(child_ret);
   case PIDFD_PASS:
    break;
   default:
    /* not reached */
    _exit(PIDFD_ERROR);
   }

   /*
 * If the user set a custom pid_max limit we could be
 * in the millions.
 * Skip the test in this case.
 */
   if (recycled_pid > PIDFD_MAX_DEFAULT)
    _exit(PIDFD_SKIP);
  }

  /* failed to recycle pid */
  _exit(PIDFD_SKIP);
 }

 ret = wait_for_pid(pid1);
 switch (ret) {
 case PIDFD_FAIL:
  ksft_exit_fail_msg(
   "%s test: Managed to signal recycled pid %d\n",
   test_name, PID_RECYCLE);
 case PIDFD_PASS:
  ksft_exit_fail_msg("%s test: Failed to recycle pid %d\n",
       test_name, PID_RECYCLE);
 case PIDFD_SKIP:
  ksft_test_result_skip("%s test: Skipping test\n", test_name);
  ret = 0;
  break;
 case PIDFD_XFAIL:
  ksft_test_result_pass(
   "%s test: Failed to signal recycled pid as expected\n",
   test_name);
  ret = 0;
  break;
 default /* PIDFD_ERROR */:
  ksft_exit_fail_msg("%s test: Error while running tests\n",
       test_name);
 }

 return ret;
}

static int test_pidfd_send_signal_syscall_support(void)
{
 int pidfd, ret;
 const char *test_name = "pidfd_send_signal check for support";

 pidfd = open("/proc/self", O_DIRECTORY | O_CLOEXEC);
 if (pidfd < 0)
  ksft_exit_fail_msg(
   "%s test: Failed to open process file descriptor\n",
   test_name);

 ret = sys_pidfd_send_signal(pidfd, 0, NULL, 0);
 if (ret < 0) {
  if (errno == ENOSYS) {
   ksft_test_result_skip(
    "%s test: pidfd_send_signal() syscall not supported\n",
    test_name);
   return 0;
  }
  ksft_exit_fail_msg("%s test: Failed to send signal\n",
       test_name);
 }

 have_pidfd_send_signal = true;
 close(pidfd);
 ksft_test_result_pass(
  "%s test: pidfd_send_signal() syscall is supported. Tests can be executed\n",
  test_name);
 return 0;
}

static void *test_pidfd_poll_exec_thread(void *priv)
{
 ksft_print_msg("Child Thread: starting. pid %d tid %ld ; and sleeping\n",
   getpid(), syscall(SYS_gettid));
 ksft_print_msg("Child Thread: doing exec of sleep\n");

 execl("/bin/sleep", "sleep", str(CHILD_THREAD_MIN_WAIT), (char *)NULL);

 ksft_print_msg("Child Thread: DONE. pid %d tid %ld\n",
   getpid(), syscall(SYS_gettid));
 return NULL;
}

static void poll_pidfd(const char *test_name, int pidfd)
{
 int c;
 int epoll_fd = epoll_create1(EPOLL_CLOEXEC);
 struct epoll_event event, events[MAX_EVENTS];

 if (epoll_fd == -1)
  ksft_exit_fail_msg("%s test: Failed to create epoll file descriptor "
       "(errno %d)\n",
       test_name, errno);

 event.events = EPOLLIN;
 event.data.fd = pidfd;

 if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, pidfd, &event)) {
  ksft_exit_fail_msg("%s test: Failed to add epoll file descriptor "
       "(errno %d)\n",
       test_name, errno);
 }

 c = epoll_wait(epoll_fd, events, MAX_EVENTS, 5000);
 if (c != 1 || !(events[0].events & EPOLLIN))
  ksft_exit_fail_msg("%s test: Unexpected epoll_wait result (c=%d, events=%x) "
       "(errno %d)\n",
       test_name, c, events[0].events, errno);

 close(epoll_fd);
 return;

}

static int child_poll_exec_test(void *args)
{
 pthread_t t1;

 ksft_print_msg("Child (pidfd): starting. pid %d tid %ld\n", getpid(),
   syscall(SYS_gettid));
 pthread_create(&t1, NULL, test_pidfd_poll_exec_thread, NULL);
 /*
 * Exec in the non-leader thread will destroy the leader immediately.
 * If the wait in the parent returns too soon, the test fails.
 */
 while (1)
  sleep(1);

 return 0;
}

static void test_pidfd_poll_exec(int use_waitpid)
{
 int pid, pidfd = 0;
 int status, ret;
 time_t prog_start = time(NULL);
 const char *test_name = "pidfd_poll check for premature notification on child thread exec";

 ksft_print_msg("Parent: pid: %d\n", getpid());
 pid = pidfd_clone(CLONE_PIDFD, &pidfd, child_poll_exec_test);
 if (pid < 0)
  ksft_exit_fail_msg("%s test: pidfd_clone failed (ret %d, errno %d)\n",
       test_name, pid, errno);

 ksft_print_msg("Parent: Waiting for Child (%d) to complete.\n", pid);

 if (use_waitpid) {
  ret = waitpid(pid, &status, 0);
  if (ret == -1)
   ksft_print_msg("Parent: error\n");

  if (ret == pid)
   ksft_print_msg("Parent: Child process waited for.\n");
 } else {
  poll_pidfd(test_name, pidfd);
 }

 time_t prog_time = time(NULL) - prog_start;

 ksft_print_msg("Time waited for child: %lu\n", prog_time);

 close(pidfd);

 if (prog_time < CHILD_THREAD_MIN_WAIT || prog_time > CHILD_THREAD_MIN_WAIT + 2)
  ksft_exit_fail_msg("%s test: Failed\n", test_name);
 else
  ksft_test_result_pass("%s test: Passed\n", test_name);
}

static void *test_pidfd_poll_leader_exit_thread(void *priv)
{
 ksft_print_msg("Child Thread: starting. pid %d tid %ld ; and sleeping\n",
   getpid(), syscall(SYS_gettid));
 sleep(CHILD_THREAD_MIN_WAIT);
 ksft_print_msg("Child Thread: DONE. pid %d tid %ld\n", getpid(), syscall(SYS_gettid));
 return NULL;
}

static time_t *child_exit_secs;
static int child_poll_leader_exit_test(void *args)
{
 pthread_t t1, t2;

 ksft_print_msg("Child: starting. pid %d tid %ld\n", getpid(), syscall(SYS_gettid));
 pthread_create(&t1, NULL, test_pidfd_poll_leader_exit_thread, NULL);
 pthread_create(&t2, NULL, test_pidfd_poll_leader_exit_thread, NULL);

 /*
 * glibc exit calls exit_group syscall, so explicitly call exit only
 * so that only the group leader exits, leaving the threads alone.
 */
 *child_exit_secs = time(NULL);
 syscall(SYS_exit, 0);
 /* Never reached, but appeases compiler thinking we should return. */
 exit(0);
}

static void test_pidfd_poll_leader_exit(int use_waitpid)
{
 int pid, pidfd = 0;
 int status, ret = 0;
 const char *test_name = "pidfd_poll check for premature notification on non-empty"
    "group leader exit";

 child_exit_secs = mmap(NULL, sizeof *child_exit_secs, PROT_READ | PROT_WRITE,
   MAP_SHARED | MAP_ANONYMOUS, -1, 0);

 if (child_exit_secs == MAP_FAILED)
  ksft_exit_fail_msg("%s test: mmap failed (errno %d)\n",
       test_name, errno);

 ksft_print_msg("Parent: pid: %d\n", getpid());
 pid = pidfd_clone(CLONE_PIDFD, &pidfd, child_poll_leader_exit_test);
 if (pid < 0)
  ksft_exit_fail_msg("%s test: pidfd_clone failed (ret %d, errno %d)\n",
       test_name, pid, errno);

 ksft_print_msg("Parent: Waiting for Child (%d) to complete.\n", pid);

 if (use_waitpid) {
  ret = waitpid(pid, &status, 0);
  if (ret == -1)
   ksft_print_msg("Parent: error\n");
 } else {
  /*
 * This sleep tests for the case where if the child exits, and is in
 * EXIT_ZOMBIE, but the thread group leader is non-empty, then the poll
 * doesn't prematurely return even though there are active threads
 */
  sleep(1);
  poll_pidfd(test_name, pidfd);
 }

 if (ret == pid)
  ksft_print_msg("Parent: Child process waited for.\n");

 time_t since_child_exit = time(NULL) - *child_exit_secs;

 ksft_print_msg("Time since child exit: %lu\n", since_child_exit);

 close(pidfd);

 if (since_child_exit < CHILD_THREAD_MIN_WAIT ||
   since_child_exit > CHILD_THREAD_MIN_WAIT + 2)
  ksft_exit_fail_msg("%s test: Failed\n", test_name);
 else
  ksft_test_result_pass("%s test: Passed\n", test_name);
}

int main(int argc, char **argv)
{
 ksft_print_header();
 ksft_set_plan(8);

 test_pidfd_poll_exec(0);
 test_pidfd_poll_exec(1);
 test_pidfd_poll_leader_exit(0);
 test_pidfd_poll_leader_exit(1);
 test_pidfd_send_signal_syscall_support();
 test_pidfd_send_signal_simple_success();
 test_pidfd_send_signal_exited_fail();
 test_pidfd_send_signal_recycled_pid_fail();

 ksft_exit_pass();
}