/* Perform a raw tape operation on remote tape connection HANDLE.
   Return the results of the ioctl, or -1 on error.  */
int
rmt_ioctl__ (int handle, int operation, char *argument)
{
  switch (operation)
    {
    default:
      errno = EOPNOTSUPP;
      return -1;

#ifdef MTIOCTOP
    case MTIOCTOP:
      {
	char command_buffer[COMMAND_BUFFER_SIZE];
	char operand_buffer[UINTMAX_STRSIZE_BOUND];
	uintmax_t u = (((struct mtop *) argument)->mt_count < 0
		       ? - (uintmax_t) ((struct mtop *) argument)->mt_count
		       : (uintmax_t) ((struct mtop *) argument)->mt_count);
	char *p = operand_buffer + sizeof operand_buffer;

        *--p = 0;
	do
	  *--p = '0' + (int) (u % 10);
	while ((u /= 10) != 0);
	if (((struct mtop *) argument)->mt_count < 0)
	  *--p = '-';

	/* MTIOCTOP is the easy one.  Nothing is transferred in binary.  */

	sprintf (command_buffer, "I%d\n%s\n",
		 ((struct mtop *) argument)->mt_op, p);
	if (do_command (handle, command_buffer) == -1)
	  return -1;

	return get_status (handle);
      }
#endif /* MTIOCTOP */

#ifdef MTIOCGET
    case MTIOCGET:
      {
	ssize_t status;
	size_t counter;

	/* Grab the status and read it directly into the structure.  This
	   assumes that the status buffer is not padded and that 2 shorts
	   fit in a long without any word alignment problems; i.e., the
	   whole struct is contiguous.  NOTE - this is probably NOT a good
	   assumption.  */

	if (do_command (handle, "S") == -1
	    || (status = get_status (handle), status == -1))
	  return -1;

	if (status > sizeof (struct mtop))
	  {
	    errno = EOVERFLOW;
	    return -1;
	  }

	for (; status > 0; status -= counter, argument += counter)
	  {
	    counter = safe_read (READ_SIDE (handle), argument, status);
	    if (counter == SAFE_READ_ERROR || counter == 0)
	      {
		_rmt_shutdown (handle, EIO);
		return -1;
	      }
	  }

	/* Check for byte position.  mt_type (or mt_model) is a small integer
	   field (normally) so we will check its magnitude.  If it is larger
	   than 256, we will assume that the bytes are swapped and go through
	   and reverse all the bytes.  */

	if (((struct mtget *) argument)->MTIO_CHECK_FIELD < 256)
	  return 0;

	for (counter = 0; counter < status; counter += 2)
	  {
	    char copy = argument[counter];

	    argument[counter] = argument[counter + 1];
	    argument[counter + 1] = copy;
	  }

	return 0;
      }
#endif /* MTIOCGET */

    }
}

static int
_handle_signal_task_local(int fd, slurmd_job_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int signal;
	int ltaskid; /* local task index */

	debug("_handle_signal_task_local for job %u.%u",
	      job->jobid, job->stepid);

	safe_read(fd, &signal, sizeof(int));
	safe_read(fd, &ltaskid, sizeof(int));

	debug3("  uid = %d", uid);
	if (uid != job->uid && !_slurm_authorized_user(uid)) {
		debug("kill req from uid %ld for job %u.%u owned by uid %ld",
		      (long)uid, job->jobid, job->stepid, (long)job->uid);
		rc = EPERM;
		goto done;
	}

	/*
	 * Sanity checks
	 */
	if (ltaskid < 0 || ltaskid >= job->node_tasks) {
		debug("step %u.%u invalid local task id %d",
		      job->jobid, job->stepid, ltaskid);
		rc = SLURM_ERROR;
		goto done;
	}
	if (!job->task
	    || !job->task[ltaskid]) {
		debug("step %u.%u no task info for task id %d",
		      job->jobid, job->stepid, ltaskid);
		rc = SLURM_ERROR;
		goto done;
	}
	if (job->task[ltaskid]->pid <= 1) {
		debug("step %u.%u invalid pid %d for task %d",
		      job->jobid, job->stepid,
		      job->task[ltaskid]->pid, ltaskid);
		rc = SLURM_ERROR;
		goto done;
	}

	/*
	 * Signal the task
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	}

	if (kill(job->task[ltaskid]->pid, signal) == -1) {
		rc = -1;
		verbose("Error sending signal %d to %u.%u, pid %d: %m",
			signal, job->jobid, job->stepid,
			job->task[ltaskid]->pid);
	} else {
		verbose("Sent signal %d to %u.%u, pid %d",
			signal, job->jobid, job->stepid,
			job->task[ltaskid]->pid);
	}
	pthread_mutex_unlock(&suspend_mutex);

done:
	/* Send the return code */
	safe_write(fd, &rc, sizeof(int));
	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}

static int
_handle_checkpoint_tasks(int fd, slurmd_job_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	time_t timestamp;
	int len;
	char *image_dir = NULL;

	debug3("_handle_checkpoint_tasks for job %u.%u",
	       job->jobid, job->stepid);

	safe_read(fd, &timestamp, sizeof(time_t));
	safe_read(fd, &len, sizeof(int));
	if (len) {
		image_dir = xmalloc (len);
		safe_read(fd, image_dir, len); /* '\0' terminated */
	}

	debug3("  uid = %d", uid);
	if (uid != job->uid && !_slurm_authorized_user(uid)) {
		debug("checkpoint req from uid %ld for job %u.%u "
		      "owned by uid %ld",
		      (long)uid, job->jobid, job->stepid, (long)job->uid);
		rc = EPERM;
		goto done;
	}

	if (job->ckpt_timestamp &&
	    timestamp == job->ckpt_timestamp) {
		debug("duplicate checkpoint req for job %u.%u, "
		      "timestamp %ld. discarded.",
		      job->jobid, job->stepid, (long)timestamp);
		rc = ESLURM_ALREADY_DONE; /* EINPROGRESS? */
		goto done;
	}

	/*
	 * Sanity checks
	 */
	if (job->pgid <= (pid_t)1) {
		debug ("step %u.%u invalid [jmgr_pid:%d pgid:%u]",
		       job->jobid, job->stepid, job->jmgr_pid, job->pgid);
		rc = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	/*
	 * Signal the process group
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	}

	/* set timestamp in case another request comes */
	job->ckpt_timestamp = timestamp;

	/* TODO: do we need job->ckpt_dir any more,
	 *	except for checkpoint/xlch? */
/*	if (! image_dir) { */
/*		image_dir = xstrdup(job->ckpt_dir); */
/*	} */

	/* call the plugin to send the request */
	if (checkpoint_signal_tasks(job, image_dir) != SLURM_SUCCESS) {
		rc = -1;
		verbose("Error sending checkpoint request to %u.%u: %s",
			job->jobid, job->stepid, slurm_strerror(rc));
	} else {
		verbose("Sent checkpoint request to %u.%u",
			job->jobid, job->stepid);
	}

	pthread_mutex_unlock(&suspend_mutex);

done:
	/* Send the return code */
	safe_write(fd, &rc, sizeof(int));
	xfree(image_dir);
	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}

static int
_handle_attach(int fd, slurmd_job_t *job, uid_t uid)
{
	srun_info_t *srun;
	int rc = SLURM_SUCCESS;

	debug("_handle_attach for job %u.%u", job->jobid, job->stepid);

	srun       = xmalloc(sizeof(srun_info_t));
	srun->key  = (srun_key_t *)xmalloc(SLURM_IO_KEY_SIZE);

	debug("sizeof(srun_info_t) = %d, sizeof(slurm_addr_t) = %d",
	      (int) sizeof(srun_info_t), (int) sizeof(slurm_addr_t));
	safe_read(fd, &srun->ioaddr, sizeof(slurm_addr_t));
	safe_read(fd, &srun->resp_addr, sizeof(slurm_addr_t));
	safe_read(fd, srun->key, SLURM_IO_KEY_SIZE);

	/*
	 * Check if jobstep is actually running.
	 */
	if (job->state != SLURMSTEPD_STEP_RUNNING) {
		rc = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	/*
	 * At the moment, it only makes sense for the slurmd to make this
	 * call, so only _slurm_authorized_user is allowed.
	 */
	if (!_slurm_authorized_user(uid)) {
		error("uid %ld attempt to attach to job %u.%u owned by %ld",
		      (long) uid, job->jobid, job->stepid, (long)job->uid);
		rc = EPERM;
		goto done;
	}

	list_prepend(job->sruns, (void *) srun);
	rc = io_client_connect(srun, job);
	debug("  back from io_client_connect, rc = %d", rc);
done:
	/* Send the return code */
	safe_write(fd, &rc, sizeof(int));

	debug("  in _handle_attach rc = %d", rc);
	if (rc == SLURM_SUCCESS) {
		/* Send response info */
		uint32_t *pids, *gtids;
		int len, i;

		debug("  in _handle_attach sending response info");
		len = job->node_tasks * sizeof(uint32_t);
		pids = xmalloc(len);
		gtids = xmalloc(len);

		if (job->task != NULL) {
			for (i = 0; i < job->node_tasks; i++) {
				if (job->task[i] == NULL)
					continue;
				pids[i] = (uint32_t)job->task[i]->pid;
				gtids[i] = job->task[i]->gtid;
			}
		}

		safe_write(fd, &job->node_tasks, sizeof(uint32_t));
		safe_write(fd, pids, len);
		safe_write(fd, gtids, len);
		xfree(pids);
		xfree(gtids);

		for (i = 0; i < job->node_tasks; i++) {
			len = strlen(job->task[i]->argv[0]) + 1;
			safe_write(fd, &len, sizeof(int));
			safe_write(fd, job->task[i]->argv[0], len);
		}
	}

	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}

static void *
_handle_accept(void *arg)
{
	/*struct request_params *param = (struct request_params *)arg;*/
	int fd = ((struct request_params *)arg)->fd;
	slurmd_job_t *job = ((struct request_params *)arg)->job;
	int req;
	int len;
	Buf buffer;
	void *auth_cred;
	int rc;
	uid_t uid;
	gid_t gid;

	debug3("Entering _handle_accept (new thread)");
	xfree(arg);

	safe_read(fd, &req, sizeof(int));
	if (req != REQUEST_CONNECT) {
		error("First message must be REQUEST_CONNECT");
		goto fail;
	}

	safe_read(fd, &len, sizeof(int));
	buffer = init_buf(len);
	safe_read(fd, get_buf_data(buffer), len);

	/* Unpack and verify the auth credential */
	auth_cred = g_slurm_auth_unpack(buffer);
	if (auth_cred == NULL) {
		error("Unpacking authentication credential: %s",
		      g_slurm_auth_errstr(g_slurm_auth_errno(NULL)));
		free_buf(buffer);
		goto fail;
	}
	rc = g_slurm_auth_verify(auth_cred, NULL, 2, NULL);
	if (rc != SLURM_SUCCESS) {
		error("Verifying authentication credential: %s",
		      g_slurm_auth_errstr(g_slurm_auth_errno(auth_cred)));
		(void) g_slurm_auth_destroy(auth_cred);
		free_buf(buffer);
		goto fail;
	}

	/* Get the uid & gid from the credential, then destroy it. */
	uid = g_slurm_auth_get_uid(auth_cred, NULL);
	gid = g_slurm_auth_get_gid(auth_cred, NULL);
	debug3("  Identity: uid=%d, gid=%d", uid, gid);
	g_slurm_auth_destroy(auth_cred);
	free_buf(buffer);

	rc = SLURM_SUCCESS;
	safe_write(fd, &rc, sizeof(int));

	while (1) {
		rc = _handle_request(fd, job, uid, gid);
		if (rc != SLURM_SUCCESS)
			break;
	}

	if (close(fd) == -1)
		error("Closing accepted fd: %m");

	pthread_mutex_lock(&message_lock);
	message_connections--;
	pthread_cond_signal(&message_cond);
	pthread_mutex_unlock(&message_lock);

	debug3("Leaving  _handle_accept");
	return NULL;

fail:
	rc = SLURM_FAILURE;
	safe_write(fd, &rc, sizeof(int));
rwfail:
	if (close(fd) == -1)
		error("Closing accepted fd after error: %m");
	debug("Leaving  _handle_accept on an error");
	return NULL;
}

//.........这里部分代码省略.........
		/* '[','1',';','3','B' |0x80,KEYCODE_ALT_DOWN  , - unugsed */
		'[','1',';','3','C' |0x80,KEYCODE_ALT_RIGHT,
		'[','1',';','3','D' |0x80,KEYCODE_ALT_LEFT ,
		/* '[','3',';','3','~' |0x80,KEYCODE_ALT_DELETE, - unugsed */
		0
	};

	pfd.fd = fd;
	pfd.events = POLLIN;

	buffer++; /* saved chars counter is in buffer[-1] now */

 start_over:
	errno = 0;
	n = (unsigned char)buffer[-1];
	if (n == 0) {
		/* If no data, wait for input.
		 * If requested, wait TIMEOUT ms. TIMEOUT = -1 is useful
		 * if fd can be in non-blocking mode.
		 */
		if (timeout >= -1) {
			if (safe_poll(&pfd, 1, timeout) == 0) {
				/* Timed out */
				errno = EAGAIN;
				return -1;
			}
		}
		/* It is tempting to read more than one byte here,
		 * but it breaks pasting. Example: at shell prompt,
		 * user presses "c","a","t" and then pastes "\nline\n".
		 * When we were reading 3 bytes here, we were eating
		 * "li" too, and cat was getting wrong input.
		 */
		n = safe_read(fd, buffer, 1);
		if (n <= 0)
			return -1;
	}

	{
		unsigned char c = buffer[0];
		n--;
		if (n)
			memmove(buffer, buffer + 1, n);
		/* Only ESC starts ESC sequences */
		if (c != 27) {
			buffer[-1] = n;
			return c;
		}
	}

	/* Loop through known ESC sequences */
	seq = esccmds;
	while (*seq != '\0') {
		/* n - position in sequence we did not read yet */
		int i = 0; /* position in sequence to compare */

		/* Loop through chars in this sequence */
		while (1) {
			/* So far escape sequence matched up to [i-1] */
			if (n <= i) {
				/* Need more chars, read another one if it wouldn't block.
				 * Note that escape sequences come in as a unit,
				 * so if we block for long it's not really an escape sequence.
				 * Timeout is needed to reconnect escape sequences
				 * split up by transmission over a serial console. */
				if (safe_poll(&pfd, 1, 50) == 0) {

/*
 * This function is very close to the cache_read function of the caching
 * library, which is used in the caching daemon. It reads cached data with the
 * specified key from the cache entry with entry_name.
 */
int
__cached_read(struct cached_connection_ *connection, const char *entry_name,
    const char *key, size_t key_size, char *data, size_t *data_size)
{
	size_t name_size, result_size;
	int error_code, rec_error_code;
	int result;

	assert(connection != NULL);
	result = 0;
	error_code = -1;

	result = send_credentials(connection, CET_READ_REQUEST);
	if (result != 0)
		goto fin;

	name_size = strlen(entry_name);
	result = safe_write(connection, &name_size, sizeof(size_t));
	if (result != 0)
		goto fin;

	result = safe_write(connection, &key_size, sizeof(size_t));
	if (result != 0)
		goto fin;

	result = safe_write(connection, entry_name, name_size);
	if (result != 0)
		goto fin;

	result = safe_write(connection, key, key_size);
	if (result != 0)
		goto fin;

	result = safe_read(connection, &rec_error_code, sizeof(int));
	if (result != 0)
		goto fin;

	if (rec_error_code != 0) {
		error_code = rec_error_code;
		goto fin;
	}

	result = safe_read(connection, &result_size, sizeof(size_t));
	if (result != 0)
		goto fin;

	 if (result_size > *data_size) {
		 *data_size = result_size;
		 error_code = -2;
		 goto fin;
	 }

	result = safe_read(connection, data, result_size);
	if (result != 0)
		goto fin;

	*data_size = result_size;
	error_code = 0;

fin:
	return (error_code);
}

int rtcwake_main(int argc UNUSED_PARAM, char **argv)
{
	unsigned opt;
	const char *rtcname = NULL;
	const char *suspend = "standby";
	const char *opt_seconds;
	const char *opt_time;

	time_t rtc_time;
	time_t sys_time;
	time_t alarm_time = alarm_time;
	unsigned seconds = seconds; /* for compiler */
	int utc = -1;
	int fd;

#if ENABLE_LONG_OPTS
	static const char rtcwake_longopts[] ALIGN1 =
		"auto\0"    No_argument "a"
		"local\0"   No_argument "l"
		"utc\0"     No_argument "u"
		"device\0"  Required_argument "d"
		"mode\0"    Required_argument "m"
		"seconds\0" Required_argument "s"
		"time\0"    Required_argument "t"
		;
#endif
	opt = getopt32long(argv,
			/* Must have -s or -t, exclusive */
			"^alud:m:s:t:" "\0" "s:t:s--t:t--s", rtcwake_longopts,
			&rtcname, &suspend, &opt_seconds, &opt_time);

	/* this is the default
	if (opt & RTCWAKE_OPT_AUTO)
		utc = -1;
	*/
	if (opt & (RTCWAKE_OPT_UTC | RTCWAKE_OPT_LOCAL))
		utc = opt & RTCWAKE_OPT_UTC;
	if (opt & RTCWAKE_OPT_SECONDS) {
		/* alarm time, seconds-to-sleep (relative) */
		seconds = xatou(opt_seconds);
	} else {
		/* RTCWAKE_OPT_TIME */
		/* alarm time, time_t (absolute, seconds since 1/1 1970 UTC) */
		if (sizeof(alarm_time) <= sizeof(long))
			alarm_time = xatol(opt_time);
		else
			alarm_time = xatoll(opt_time);
	}

	if (utc == -1)
		utc = rtc_adjtime_is_utc();

	/* the rtcname is relative to /dev */
	xchdir("/dev");

	/* this RTC must exist and (if we'll sleep) be wakeup-enabled */
	fd = rtc_xopen(&rtcname, O_RDONLY);

	if (strcmp(suspend, "on") != 0)
		if (!may_wakeup(rtcname))
			bb_error_msg_and_die("%s not enabled for wakeup events", rtcname);

	/* relative or absolute alarm time, normalized to time_t */
	sys_time = time(NULL);
	{
		struct tm tm_time;
		rtc_read_tm(&tm_time, fd);
		rtc_time = rtc_tm2time(&tm_time, utc);
	}

	if (opt & RTCWAKE_OPT_TIME) {
		/* Correct for RTC<->system clock difference */
		alarm_time += rtc_time - sys_time;
		if (alarm_time < rtc_time)
			/*
			 * Compat message text.
			 * I'd say "RTC time is already ahead of ..." instead.
			 */
			bb_error_msg_and_die("time doesn't go backward to %s", ctime(&alarm_time));
	} else
		alarm_time = rtc_time + seconds + 1;

	setup_alarm(fd, &alarm_time, rtc_time);
	sync();
#if 0 /*debug*/
	printf("sys_time: %s", ctime(&sys_time));
	printf("rtc_time: %s", ctime(&rtc_time));
#endif
	printf("wakeup from \"%s\" at %s", suspend, ctime(&alarm_time));
	fflush_all();
	usleep(10 * 1000);

	if (strcmp(suspend, "on") != 0)
		xopen_xwrite_close(SYS_POWER_PATH, suspend);
	else {
		/* "fake" suspend ... we'll do the delay ourselves */
		unsigned long data;

		do {
			ssize_t ret = safe_read(fd, &data, sizeof(data));
//.........这里部分代码省略.........

int last_main(int argc, char **argv)
{
	struct utmp ut;
	int n, file = STDIN_FILENO;
	time_t t_tmp;

	if (argc > 1) {
		bb_show_usage();
	}
	file = xopen(bb_path_wtmp_file, O_RDONLY);

	printf("%-10s %-14s %-18s %-12.12s %s\n", "USER", "TTY", "HOST", "LOGIN", "TIME");
	while ((n = safe_read(file, (void*)&ut, sizeof(struct utmp))) != 0) {

		if (n != sizeof(struct utmp)) {
			bb_perror_msg_and_die("short read");
		}

		if (ut.ut_line[0] == '~') {
			if (strncmp(ut.ut_user, "shutdown", 8) == 0)
				ut.ut_type = SHUTDOWN_TIME;
			else if (strncmp(ut.ut_user, "reboot", 6) == 0)
				ut.ut_type = BOOT_TIME;
			else if (strncmp(ut.ut_user, "runlevel", 7) == 0)
				ut.ut_type = RUN_LVL;
		} else {
			if (!ut.ut_name[0] || strcmp(ut.ut_name, "LOGIN") == 0 ||
					ut.ut_name[0] == 0)
			{
				/* Don't bother.  This means we can't find how long
				 * someone was logged in for.  Oh well. */
				continue;
			}
			if (ut.ut_type != DEAD_PROCESS &&
					ut.ut_name[0] && ut.ut_line[0])
			{
				ut.ut_type = USER_PROCESS;
			}
			if (strcmp(ut.ut_name, "date") == 0) {
				if (ut.ut_line[0] == '|') ut.ut_type = OLD_TIME;
				if (ut.ut_line[0] == '{') ut.ut_type = NEW_TIME;
			}
		}

		if (ut.ut_type!=USER_PROCESS) {
			switch (ut.ut_type) {
				case OLD_TIME:
				case NEW_TIME:
				case RUN_LVL:
				case SHUTDOWN_TIME:
					continue;
				case BOOT_TIME:
					strcpy(ut.ut_line, "system boot");
					break;
			}
		}
		t_tmp = (time_t)ut.ut_tv.tv_sec;
		printf("%-10s %-14s %-18s %-12.12s\n", ut.ut_user, ut.ut_line, ut.ut_host,
				ctime(&t_tmp) + 4);
	}

	fflush_stdout_and_exit(EXIT_SUCCESS);
}

static int
_handle_completion(int fd, stepd_step_rec_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;
	int first;
	int last;
	jobacctinfo_t *jobacct = NULL;
	int step_rc;
	char* buf;
	int len;
	Buf buffer;
	int version;	/* For future use */
	bool lock_set = false;

	debug("_handle_completion for job %u.%u",
	      job->jobid, job->stepid);

	debug3("  uid = %d", uid);
	if (!_slurm_authorized_user(uid)) {
		debug("step completion message from uid %ld for job %u.%u ",
		      (long)uid, job->jobid, job->stepid);
		rc = -1;
		errnum = EPERM;
		/* Send the return code and errno */
		safe_write(fd, &rc, sizeof(int));
		safe_write(fd, &errnum, sizeof(int));
		return SLURM_SUCCESS;
	}

	safe_read(fd, &version, sizeof(int));
	safe_read(fd, &first, sizeof(int));
	safe_read(fd, &last, sizeof(int));
	safe_read(fd, &step_rc, sizeof(int));

	/*
	 * We must not use getinfo over a pipe with slurmd here
	 * Indeed, slurmstepd does a large use of setinfo over a pipe
	 * with slurmd and doing the reverse can result in a deadlock
	 * scenario with slurmd :
	 * slurmd(lockforread,write)/slurmstepd(write,lockforread)
	 * Do pack/unpack instead to be sure of independances of
	 * slurmd and slurmstepd
	 */
	safe_read(fd, &len, sizeof(int));
	buf = xmalloc(len);
	safe_read(fd, buf, len);
	buffer = create_buf(buf, len);
	jobacctinfo_unpack(&jobacct, SLURM_PROTOCOL_VERSION,
			   PROTOCOL_TYPE_SLURM, buffer, 1);
	free_buf(buffer);

	/*
	 * Record the completed nodes
	 */
	pthread_mutex_lock(&step_complete.lock);
	lock_set = true;
	if (! step_complete.wait_children) {
		rc = -1;
		errnum = ETIMEDOUT; /* not used anyway */
		goto timeout;
	}

	/* SlurmUser or root can craft a launch without a valid credential
	 * ("srun --no-alloc ...") and no tree information can be built
	 *  without the hostlist from the credential. */
	if (step_complete.rank >= 0) {
#if 0
		char bits_string[128];
		debug2("Setting range %d (bit %d) through %d(bit %d)",
		       first, first-(step_complete.rank+1),
		       last, last-(step_complete.rank+1));
		bit_fmt(bits_string, sizeof(bits_string), step_complete.bits);
		debug2("  before bits: %s", bits_string);
#endif
		bit_nset(step_complete.bits,
			 first - (step_complete.rank+1),
			 last - (step_complete.rank+1));
#if 0
		bit_fmt(bits_string, sizeof(bits_string), step_complete.bits);
		debug2("  after bits: %s", bits_string);
#endif
	}
	step_complete.step_rc = MAX(step_complete.step_rc, step_rc);

	/************* acct stuff ********************/
	jobacctinfo_aggregate(step_complete.jobacct, jobacct);
timeout:
	jobacctinfo_destroy(jobacct);
	/*********************************************/

	/* Send the return code and errno, we do this within the locked
	 * region to ensure that the stepd doesn't exit before we can
	 * perform this send. */
	safe_write(fd, &rc, sizeof(int));
	safe_write(fd, &errnum, sizeof(int));
	pthread_cond_signal(&step_complete.cond);
	pthread_mutex_unlock(&step_complete.lock);

	return SLURM_SUCCESS;
//.........这里部分代码省略.........

static int
_handle_signal_container(int fd, stepd_step_rec_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;
	int sig;
	static int msg_sent = 0;
	char *ptr = NULL;
	int target_node_id = 0;
	stepd_step_task_info_t *task;
	uint32_t i;

	safe_read(fd, &sig, sizeof(int));
	debug("_handle_signal_container for step=%u.%u uid=%d signal=%d",
	      job->jobid, job->stepid, (int) uid, sig);
	if ((uid != job->uid) && !_slurm_authorized_user(uid)) {
		error("signal container req from uid %ld for step=%u.%u "
		      "owned by uid %ld",
		      (long)uid, job->jobid, job->stepid, (long)job->uid);
		rc = -1;
		errnum = EPERM;
		goto done;
	}

	/*
	 * Sanity checks
	 */
	if (job->cont_id == 0) {
		debug ("step %u.%u invalid container [cont_id:%"PRIu64"]",
			job->jobid, job->stepid, job->cont_id);
		rc = -1;
		errnum = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	if ((sig == SIGTERM) || (sig == SIGKILL)) {
		/* cycle thru the tasks and mark those that have not
		 * called abort and/or terminated as killed_by_cmd
		 */
		for (i = 0; i < job->node_tasks; i++) {
			if (NULL == (task = job->task[i])) {
				continue;
			}
			if (task->aborted || task->exited) {
				continue;
			}
			/* mark that this task is going to be killed by
			 * cmd so we ignore its exit status - otherwise,
			 * we will probably report the final exit status
			 * as SIGKILL
			 */
			task->killed_by_cmd = true;
		}
	}

	ptr = getenvp(job->env, "SLURM_STEP_KILLED_MSG_NODE_ID");
	if (ptr)
		target_node_id = atoi(ptr);
	if ((job->nodeid == target_node_id) && (msg_sent == 0) &&
	    (job->state < SLURMSTEPD_STEP_ENDING)) {
		time_t now = time(NULL);
		char entity[24], time_str[24];
		if (job->stepid == SLURM_BATCH_SCRIPT) {
			snprintf(entity, sizeof(entity), "JOB %u", job->jobid);
		} else {
			snprintf(entity, sizeof(entity), "STEP %u.%u",
				 job->jobid, job->stepid);
		}
		slurm_make_time_str(&now, time_str, sizeof(time_str));

		/* Not really errors,
		 * but we want messages displayed by default */
		if (sig == SIG_TIME_LIMIT) {
			error("*** %s CANCELLED AT %s DUE TO TIME LIMIT ***",
			      entity, time_str);
			msg_sent = 1;
		} else if (sig == SIG_PREEMPTED) {
			error("*** %s CANCELLED AT %s DUE TO PREEMPTION ***",
			      entity, time_str);
			msg_sent = 1;
		} else if (sig == SIG_NODE_FAIL) {
			error("*** %s CANCELLED AT %s DUE TO NODE FAILURE ***",
			      entity, time_str);
			msg_sent = 1;
		} else if (sig == SIG_FAILURE) {
			error("*** %s FAILED (non-zero exit code or other "
			      "failure mode) ***", entity);
			msg_sent = 1;
		} else if ((sig == SIGTERM) || (sig == SIGKILL)) {
			error("*** %s CANCELLED AT %s ***", entity, time_str);
			msg_sent = 1;
		}
	}
	if ((sig == SIG_TIME_LIMIT) || (sig == SIG_NODE_FAIL) ||
	    (sig == SIG_PREEMPTED)  || (sig == SIG_FAILURE))
		goto done;

	if (sig == SIG_ABORT) {
		sig = SIGKILL;
		job->aborted = true;
//.........这里部分代码省略.........

static int
_handle_resume(int fd, stepd_step_rec_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;
	uint16_t job_core_spec = (uint16_t) NO_VAL;

	safe_read(fd, &job_core_spec, sizeof(uint16_t));

	debug("_handle_resume for step:%u.%u uid:%ld core_spec:%u",
	      job->jobid, job->stepid, (long)uid, job_core_spec);

	if (!_slurm_authorized_user(uid)) {
		debug("job step resume request from uid %ld for job %u.%u ",
		      (long)uid, job->jobid, job->stepid);
		rc = -1;
		errnum = EPERM;
		goto done;
	}

	if (job->cont_id == 0) {
		debug ("step %u.%u invalid container [cont_id:%"PRIu64"]",
			job->jobid, job->stepid, job->cont_id);
		rc = -1;
		errnum = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	acct_gather_resume_poll();
	/*
	 * Signal the container
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (!suspended) {
		rc = -1;
		errnum = ESLURMD_STEP_NOTSUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	} else {
		if (!job->batch && switch_g_job_step_pre_resume(job))
			error("switch_g_job_step_pre_resume: %m");
		if (!job->batch && core_spec_g_resume(job->cont_id,
						      job_core_spec))
			error("core_spec_g_resume: %m");
		if (proctrack_g_signal(job->cont_id, SIGCONT) < 0) {
			verbose("Error resuming %u.%u: %m",
				job->jobid, job->stepid);
		} else {
			verbose("Resumed %u.%u", job->jobid, job->stepid);
		}
		suspended = false;
	}
	if (!job->batch && switch_g_job_step_post_resume(job))
		error("switch_g_job_step_post_resume: %m");
	/* set the cpu frequencies if cpu_freq option used */
	if (job->cpu_freq != NO_VAL)
		cpu_freq_set(job);

	pthread_mutex_unlock(&suspend_mutex);

done:
	/* Send the return code and errno */
	safe_write(fd, &rc, sizeof(int));
	safe_write(fd, &errnum, sizeof(int));
	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}

static int
_handle_suspend(int fd, stepd_step_rec_t *job, uid_t uid)
{
	static int launch_poe = -1;
	int rc = SLURM_SUCCESS;
	int errnum = 0;
	uint16_t job_core_spec = (uint16_t) NO_VAL;

	safe_read(fd, &job_core_spec, sizeof(uint16_t));

	debug("_handle_suspend for step:%u.%u uid:%ld core_spec:%u",
	      job->jobid, job->stepid, (long)uid, job_core_spec);

	if (!_slurm_authorized_user(uid)) {
		debug("job step suspend request from uid %ld for job %u.%u ",
		      (long)uid, job->jobid, job->stepid);
		rc = -1;
		errnum = EPERM;
		goto done;
	}

	if (job->cont_id == 0) {
		debug ("step %u.%u invalid container [cont_id:%"PRIu64"]",
			job->jobid, job->stepid, job->cont_id);
		rc = -1;
		errnum = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	acct_gather_suspend_poll();
	if (launch_poe == -1) {
		char *launch_type = slurm_get_launch_type();
		if (!strcmp(launch_type, "launch/poe"))
			launch_poe = 1;
		else
			launch_poe = 0;
		xfree(launch_type);
	}

	/*
	 * Signal the container
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = -1;
		errnum = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	} else {
		if (!job->batch && switch_g_job_step_pre_suspend(job))
			error("switch_g_job_step_pre_suspend: %m");

		/* SIGTSTP is sent first to let MPI daemons stop their tasks,
		 * then wait 2 seconds, then send SIGSTOP to the spawned
		 * process's container to stop everything else.
		 *
		 * In some cases, 1 second has proven insufficient. Longer
		 * delays may help insure that all MPI tasks have been stopped
		 * (that depends upon the MPI implementaiton used), but will
		 * also permit longer time periods when more than one job can
		 * be running on each resource (not good). */
		if (launch_poe == 0) {
			/* IBM MPI seens to periodically hang upon receipt
			 * of SIGTSTP. */
			if (proctrack_g_signal(job->cont_id, SIGTSTP) < 0) {
				verbose("Error suspending %u.%u (SIGTSTP): %m",
					job->jobid, job->stepid);
			} else
				sleep(2);
		}

		if (proctrack_g_signal(job->cont_id, SIGSTOP) < 0) {
			verbose("Error suspending %u.%u (SIGSTOP): %m",
				job->jobid, job->stepid);
		} else {
			verbose("Suspended %u.%u", job->jobid, job->stepid);
		}
		suspended = true;
	}
	if (!job->batch && switch_g_job_step_post_suspend(job))
		error("switch_g_job_step_post_suspend: %m");
	if (!job->batch && core_spec_g_suspend(job->cont_id, job_core_spec))
		error("core_spec_g_suspend: %m");

	pthread_mutex_unlock(&suspend_mutex);

done:
	/* Send the return code and errno */
	safe_write(fd, &rc, sizeof(int));
	safe_write(fd, &errnum, sizeof(int));
	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}

extern int jobacctinfo_getinfo(
	jobacctinfo_t *jobacct, enum jobacct_data_type type, void *data,
	uint16_t protocol_version)
{
	int rc = SLURM_SUCCESS;
	int *fd = (int *)data;
	uint32_t *uint32 = (uint32_t *) data;
	uint64_t *uint64 = (uint64_t *) data;
	double *dub = (double *) data;
	jobacct_id_t *jobacct_id = (jobacct_id_t *) data;
	struct rusage *rusage = (struct rusage *)data;
	struct jobacctinfo *send = (struct jobacctinfo *) data;

	if (!plugin_polling)
		return SLURM_SUCCESS;

	/* jobacct needs to be allocated before this is called.	*/
	xassert(jobacct);

	switch (type) {
	case JOBACCT_DATA_TOTAL:
		memcpy(send, jobacct, sizeof(struct jobacctinfo));
		break;
	case JOBACCT_DATA_PIPE:
		if (protocol_version >= SLURM_MIN_PROTOCOL_VERSION) {
			char* buf;
			int len;
			Buf buffer;

			safe_read(*fd, &len, sizeof(int));
			buf = xmalloc(len);
			safe_read(*fd, buf, len);
			buffer = create_buf(buf, len);
			jobacctinfo_unpack(&jobacct, protocol_version,
					   PROTOCOL_TYPE_SLURM, buffer, 0);
			free_buf(buffer);
		}

		break;
	case JOBACCT_DATA_RUSAGE:
		memset(rusage, 0, sizeof(struct rusage));
		rusage->ru_utime.tv_sec = jobacct->user_cpu_sec;
		rusage->ru_utime.tv_usec = jobacct->user_cpu_usec;
		rusage->ru_stime.tv_sec = jobacct->sys_cpu_sec;
		rusage->ru_stime.tv_usec = jobacct->sys_cpu_usec;
		break;
	case JOBACCT_DATA_MAX_RSS:
		*uint64 = jobacct->max_rss;
		break;
	case JOBACCT_DATA_MAX_RSS_ID:
		*jobacct_id = jobacct->max_rss_id;
		break;
	case JOBACCT_DATA_TOT_RSS:
		*uint64 = jobacct->tot_rss;
		break;
	case JOBACCT_DATA_MAX_VSIZE:
		*uint64 = jobacct->max_vsize;
		break;
	case JOBACCT_DATA_MAX_VSIZE_ID:
		*jobacct_id = jobacct->max_vsize_id;
		break;
	case JOBACCT_DATA_TOT_VSIZE:
		*uint64 = jobacct->tot_vsize;
		break;
	case JOBACCT_DATA_MAX_PAGES:
		*uint64 = jobacct->max_pages;
		break;
	case JOBACCT_DATA_MAX_PAGES_ID:
		*jobacct_id = jobacct->max_pages_id;
		break;
	case JOBACCT_DATA_TOT_PAGES:
		*uint64 = jobacct->tot_pages;
		break;
	case JOBACCT_DATA_MIN_CPU:
		*uint32 = jobacct->min_cpu;
		break;
	case JOBACCT_DATA_MIN_CPU_ID:
		*jobacct_id = jobacct->min_cpu_id;
		break;
	case JOBACCT_DATA_TOT_CPU:
		*dub = jobacct->tot_cpu;
		break;
	case JOBACCT_DATA_ACT_CPUFREQ:
		*uint32 = jobacct->act_cpufreq;
		break;
	case JOBACCT_DATA_CONSUMED_ENERGY:
		*uint64 = jobacct->energy.consumed_energy;
		break;
	case JOBACCT_DATA_MAX_DISK_READ:
		*dub = jobacct->max_disk_read;
		break;
	case JOBACCT_DATA_MAX_DISK_READ_ID:
		*jobacct_id = jobacct->max_disk_read_id;
		break;
	case JOBACCT_DATA_TOT_DISK_READ:
		*dub = jobacct->tot_disk_read;
		break;
	case JOBACCT_DATA_MAX_DISK_WRITE:
		*dub = jobacct->max_disk_write;
		break;
//.........这里部分代码省略.........

int main(int ac, char **av)
{
	int lc;			/* loop counter */
	char *msg;		/* message returned from parse_opts */

	int i, red, wtstatus;
	int pipefd[2];		/* fds for pipe read/write */
	char rebuf[BUFSIZ];
	int Acnt = 0, Bcnt = 0;	/* count 'A' and 'B' */
	int fork_1, fork_2;	/* ret values in parent */

	/* parse standard options */
	if ((msg = parse_opts(ac, av, NULL, NULL)) != NULL)
		tst_brkm(TBROK, NULL, "OPTION PARSING ERROR - %s", msg);

	setup();

	for (lc = 0; TEST_LOOPING(lc); lc++) {

		/* reset Tst_count in case we are looping */
		Tst_count = 0;

		TEST(pipe(pipefd));

		if (TEST_RETURN == -1) {
			tst_resm(TFAIL, "pipe() call failed");
			continue;
		}

		if (!STD_FUNCTIONAL_TEST) {
			tst_resm(TWARN, "-f option should not be used");
			tst_resm(TPASS, "call succeeded");
			continue;
		}

		if ((fork_1 = FORK_OR_VFORK()) == -1) {
			tst_brkm(TBROK, cleanup, "fork() #1 failed");
		}

		if (fork_1 == 0) {	/* 1st child */
			if (close(pipefd[0]) != 0) {
				tst_resm(TWARN, "pipefd[0] close failed, "
					 "errno = %d", errno);
				exit(1);
			}

			for (i = 0; i < PIPEWRTCNT / 2; ++i) {
				if (write(pipefd[1], "A", 1) != 1) {
					tst_resm(TWARN, "write to pipe failed");
					exit(1);
				}
			}
			exit(0);
		}

		/* parent */

		if (waitpid(fork_1, &wtstatus, 0) == -1)
			tst_brkm(TBROK, cleanup, "waitpid failed");
		if (WIFEXITED(wtstatus) && WEXITSTATUS(wtstatus) != 0) {
			tst_brkm(TBROK, cleanup, "child exited abnormally");
		}

		if ((fork_2 = FORK_OR_VFORK()) == -1) {
			tst_brkm(TBROK, cleanup, "fork() #2 failed");
		 }

		if (fork_2 == 0) {	/* 2nd child */
			if (close(pipefd[0]) != 0) {
				perror("pipefd[0] close failed");
				exit(1);
			}

			for (i = 0; i < PIPEWRTCNT / 2; ++i) {
				if (write(pipefd[1], "B", 1) != 1) {
					perror("write to pipe failed");
					exit(1);
				}
			}
			exit(0);
		}

		/* parent */

		if (waitpid(fork_2, &wtstatus, 0) == -1)
			tst_brkm(TBROK, cleanup, "waitpid failed");
		if (WEXITSTATUS(wtstatus) != 0) {
			tst_brkm(TBROK, cleanup, "problem detected in child, "
				 "wait status %d, errno = %d", wtstatus, errno);
		}

		if (close(pipefd[1]) != 0) {
			tst_brkm(TBROK|TERRNO, cleanup,
			    "pipefd[1] close failed");
		}

		while ((red = safe_read(pipefd[0], rebuf, 100)) > 0) {
			for (i = 0; i < red; i++) {
				if (rebuf[i] == 'A') {
					Acnt++;
//.........这里部分代码省略.........

//.........这里部分代码省略.........
			break; // case 0 (timeout)

		// packets arriving, or link went down
		case 1:
			// We need to adjust the timeout in case we didn't receive
			// a conflicting packet.
			if (timeout_ms > 0) {
				unsigned diff = deadline_us - MONOTONIC_US();
				if ((int)(diff) < 0) {
					// Current time is greater than the expected timeout time.
					// Should never happen.
					VDBG("missed an expected timeout\n");
					timeout_ms = 0;
				} else {
					VDBG("adjusting timeout\n");
					timeout_ms = (diff / 1000) | 1; /* never 0 */
				}
			}

			if ((fds[0].revents & POLLIN) == 0) {
				if (fds[0].revents & POLLERR) {
					// FIXME: links routinely go down;
					// this shouldn't necessarily exit.
					bb_error_msg("iface %s is down", argv_intf);
					if (ready) {
						run(argv, "deconfig", &ip);
					}
					return EXIT_FAILURE;
				}
				continue;
			}

			// read ARP packet
			if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
				bb_perror_msg_and_die(bb_msg_read_error);
			}
			if (p.eth.ether_type != htons(ETHERTYPE_ARP))
				continue;
#ifdef DEBUG
			{
				struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
				struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
				struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
				struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
				VDBG("%s recv arp type=%d, op=%d,\n",
					argv_intf, ntohs(p.eth.ether_type),
					ntohs(p.arp.arp_op));
				VDBG("\tsource=%s %s\n",
					ether_ntoa(sha),
					inet_ntoa(*spa));
				VDBG("\ttarget=%s %s\n",
					ether_ntoa(tha),
					inet_ntoa(*tpa));
			}
#endif
			if (p.arp.arp_op != htons(ARPOP_REQUEST)
			 && p.arp.arp_op != htons(ARPOP_REPLY)
			) {
				continue;
			}

			source_ip_conflict = 0;
			target_ip_conflict = 0;

			if (memcmp(&p.arp.arp_sha, &eth_addr, ETH_ALEN) != 0) {
				if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0) {

//.........这里部分代码省略.........
		log_printf (LOG_NORMAL,"\rErasing blocks: %d/%d (100%%)\n",blocks,blocks);
	 }
   else
	 {
		/* if not, erase the whole chunk in one shot */
		if (ioctl (dev_fd,MEMERASE,&erase) < 0)
		  {
				  log_printf (LOG_ERROR,
						   "While erasing blocks from 0x%.8x-0x%.8x on %s: %m\n",
						   (unsigned int) erase.start,(unsigned int) (erase.start + erase.length),device);
			 exit (EXIT_FAILURE);
		  }
	 }
   DEBUG("Erased %u / %luk bytes\n",erase.length,filestat.st_size);

   /**********************************
	* write the entire file to flash *
	**********************************/

   if (flags & FLAG_VERBOSE) log_printf (LOG_NORMAL,"Writing data: 0k/%luk (0%%)",KB (filestat.st_size));
   size = filestat.st_size;
   i = BUFSIZE;
   written = 0;
   while (size)
	 {
		if (size < BUFSIZE) i = size;
		if (flags & FLAG_VERBOSE)
		  log_printf (LOG_NORMAL,"\rWriting data: %dk/%luk (%lu%%)",
				  KB (written + i),
				  KB (filestat.st_size),
				  PERCENTAGE (written + i,filestat.st_size));

		/* read from filename */
		safe_read (fil_fd,filename,src,i,flags & FLAG_VERBOSE);

		/* write to device */
		result = write (dev_fd,src,i);
		if (i != result)
		  {
			 if (flags & FLAG_VERBOSE) log_printf (LOG_NORMAL,"\n");
			 if (result < 0)
			   {
				  log_printf (LOG_ERROR,
						   "While writing data to 0x%.8x-0x%.8x on %s: %m\n",
						   written,written + i,device);
				  exit (EXIT_FAILURE);
			   }
			 log_printf (LOG_ERROR,
					  "Short write count returned while writing to x%.8x-0x%.8x on %s: %d/%lu bytes written to flash\n",
					  written,written + i,device,written + result,filestat.st_size);
			 exit (EXIT_FAILURE);
		  }

		written += i;
		size -= i;
	 }
   if (flags & FLAG_VERBOSE)
	 log_printf (LOG_NORMAL,
				 "\rWriting data: %luk/%luk (100%%)\n",
				 KB (filestat.st_size),
				 KB (filestat.st_size));
   DEBUG("Wrote %d / %luk bytes\n",written,filestat.st_size);

   /**********************************
	* verify that flash == file data *
	**********************************/