//.........这里部分代码省略.........
            snprintf(hoobab, sizeof hoobab, "%s/messages/base%3.3d/fa%5.5d", conference()->conf.CONF_PATH, current_msgbase->MSGBASE_NUMBER, header.MSG_NUMBER);
            mkdir(hoobab, 0777);
            setperm(hoobab, 0777);

            while (fgetsnolf(hoobab, 1024, fd)) {
                char sr[1024];
                char de[1024];
                snprintf(sr, sizeof sr, "%s/%s", currnode->MULTI_TEMPORARY, hoobab);
                snprintf(de, sizeof de, "%s/messages/base%3.3d/fa%5.5d/%s", conference()->conf.CONF_PATH, current_msgbase->MSGBASE_NUMBER, header.MSG_NUMBER, hoobab);
                newrename(sr, de);
            }
            fclose(fd);
            snprintf(hoobab, sizeof hoobab, "%s/messages/base%3.3d/msf%5.5d", conference()->conf.CONF_PATH, current_msgbase->MSGBASE_NUMBER, header.MSG_NUMBER);
            newrename(fabuf, hoobab);
        } else {
            *header.MSG_ATTACH = 0;
        }
    }

    if (toupper(current_msgbase->MSGBASE_FN_FLAGS) != 'L') {
        header.MSG_FN_ORIG_ZONE = current_msgbase->MSGBASE_FN_ZONE;
        header.MSG_FN_ORIG_NET = current_msgbase->MSGBASE_FN_NET;
        header.MSG_FN_ORIG_NODE = current_msgbase->MSGBASE_FN_NODE;
        header.MSG_FN_ORIG_POINT = current_msgbase->MSGBASE_FN_POINT;
        header.MSG_FLAGS |= (1L << 2);
    }
    if ((msgfd = ddmsg_open_base(conference()->conf.CONF_PATH, current_msgbase->MSGBASE_NUMBER, O_RDWR | O_CREAT, 0666)) == -1) {
        DDPut(sd[emwriteerrstr]);
        free(lineedmem);
        return 0;
    }
    fsetperm(msgfd, 0666);
    lseek(msgfd, 0, SEEK_END);
    safe_write(msgfd, &header, sizeof(struct DayDream_Message));
    ddmsg_close_base(msgfd);

    if ((msgfd = ddmsg_open_msg(conference()->conf.CONF_PATH, current_msgbase->MSGBASE_NUMBER, header.MSG_NUMBER, O_RDWR | O_CREAT | O_TRUNC, 0666)) == -1) {
        DDPut(sd[emwriteerrstr]);
        free(lineedmem);
        return 0;
    }
    fsetperm(msgfd, 0666);
    if (toupper(current_msgbase->MSGBASE_FN_FLAGS) == 'E') {
        char ub[128];
        int uq;

        strlcpy(ub, current_msgbase->MSGBASE_FN_TAG, sizeof ub);
        strupr(ub);
        snprintf(ebuf, sizeof ebuf, "AREA:%s\n", ub);
        safe_write(msgfd, ebuf, strlen(ebuf));
        if ((uq = getfidounique())) {
            snprintf(ebuf, sizeof ebuf, "\001MSGID: %d:%d/%d.%d %8.8x\n", current_msgbase->MSGBASE_FN_ZONE, current_msgbase->MSGBASE_FN_NET, current_msgbase->MSGBASE_FN_NODE, current_msgbase->MSGBASE_FN_POINT, uq);
            safe_write(msgfd, ebuf, strlen(ebuf));
            if (header.MSG_ORIGINAL) {
                if (getreplyid(header.MSG_ORIGINAL, ebuf, sizeof ebuf))
                    safe_write(msgfd, ebuf, strlen(ebuf));
            }
        }
    } else if (toupper(current_msgbase->MSGBASE_FN_FLAGS) == 'N') {
        snprintf(ebuf, sizeof ebuf, "\001INTL %d:%d/%d %d:%d/%d\n",
                 header.MSG_FN_DEST_ZONE, header.MSG_FN_DEST_NET,
                 header.MSG_FN_DEST_NODE, header.MSG_FN_ORIG_ZONE,
                 header.MSG_FN_ORIG_NET, header.MSG_FN_ORIG_NODE);
        safe_write(msgfd, ebuf, strlen(ebuf));

        if (header.MSG_FN_DEST_POINT) {

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

	safe_read(fd, &signal, sizeof(int));
	safe_read(fd, &ltaskid, sizeof(int));
	debug("_handle_signal_task_local for step=%u.%u uid=%d signal=%d",
	      job->jobid, job->stepid, (int) uid, signal);

	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
	 */
	slurm_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = ESLURMD_STEP_SUSPENDED;
		slurm_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);
	}
	slurm_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, stepd_step_rec_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;
	}

	slurm_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = ESLURMD_STEP_SUSPENDED;
		slurm_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);
	}

	slurm_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_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;
	}

	_wait_for_job_init(job);

	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 (!xstrcmp(launch_type, "launch/poe"))
			launch_poe = 1;
		else
			launch_poe = 0;
		xfree(launch_type);
	}

	/*
	 * Signal the container
	 */
	slurm_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = -1;
		errnum = ESLURMD_STEP_SUSPENDED;
		slurm_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");

	slurm_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_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;
	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, &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
	 */
	slurm_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));
	slurm_cond_signal(&step_complete.cond);
	slurm_mutex_unlock(&step_complete.lock);

	return SLURM_SUCCESS;


//.........这里部分代码省略.........

//.........这里部分代码省略.........
	debug3("  uid = %d", uid);
	if (uid != job->uid && !_slurm_authorized_user(uid)) {
		debug("kill container req from uid %ld for job %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 ((job->nodeid == 0) && (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_DEBUG_WAKE) {
		int i;
		for (i = 0; i < job->node_tasks; i++)
			pdebug_wake_process(job, job->task[i]->pid);
		goto done;
	}
	if (sig == SIG_ABORT) {
		sig = SIGKILL;
		job->aborted = true;
	}

	pthread_mutex_lock(&suspend_mutex);
	if (suspended && (sig != SIGKILL)) {
		rc = -1;
		errnum = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	}

	/*
	 * Signal the container
	 */
	if (slurm_container_signal(job->cont_id, sig) < 0) {
		rc = -1;
		errnum = errno;
		verbose("Error sending signal %d to %u.%u: %m",
			sig, job->jobid, job->stepid);
	} else {
		verbose("Sent signal %d to %u.%u",
			sig, job->jobid, job->stepid);
	}
	pthread_mutex_unlock(&suspend_mutex);

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

static int
_handle_terminate(int fd, stepd_step_rec_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;
	stepd_step_task_info_t *task;
	uint32_t i;

	debug("_handle_terminate for step=%u.%u uid=%d",
	      job->jobid, job->stepid, uid);
	step_terminate_monitor_start(job->jobid, job->stepid);

	if (uid != job->uid && !_slurm_authorized_user(uid)) {
		debug("terminate req from uid %ld for job %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;
	}

	/* 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;
	}

	/*
	 * Signal the container with SIGKILL
	 */
	slurm_mutex_lock(&suspend_mutex);
	if (suspended) {
		debug("Terminating suspended job step %u.%u",
		      job->jobid, job->stepid);
		suspended = false;
	}

	if (proctrack_g_signal(job->cont_id, SIGKILL) < 0) {
		if (errno != ESRCH) {	/* No error if process already gone */
			rc = -1;
			errnum = errno;
		}
		verbose("Error sending SIGKILL signal to %u.%u: %m",
			job->jobid, job->stepid);
	} else {
		verbose("Sent SIGKILL signal to %u.%u",
			job->jobid, job->stepid);
	}
	slurm_mutex_unlock(&suspend_mutex);

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

static int
decompress_fd_xz(int fdi, int fdo)
{
#ifdef HAVE_LZMA
    uint8_t buf_in[BUFSIZ];
    uint8_t buf_out[BUFSIZ];

    lzma_stream strm = LZMA_STREAM_INIT;
    lzma_ret ret = lzma_stream_decoder(&strm, UINT64_MAX, 0);
    if (ret != LZMA_OK)
    {
        close(fdi);
        close(fdo);
        log_error("Failed to initialize XZ decoder: code %d", ret);
        return -ENOMEM;
    }

    lzma_action action = LZMA_RUN;

    strm.next_out = buf_out;
    strm.avail_out = sizeof(buf_out);

    for (;;)
    {
        if (strm.avail_in == 0 && action == LZMA_RUN)
        {
            strm.next_in = buf_in;
            strm.avail_in = safe_read(fdi, buf_in, sizeof(buf_in));

            if (strm.avail_in < 0)
            {
                perror_msg("Failed to read source core file");
                close(fdi);
                close(fdo);
                lzma_end(&strm);
                return -1;
            }

            if (strm.avail_in == 0)
                action = LZMA_FINISH;
        }

        ret = lzma_code(&strm, action);

        if (strm.avail_out == 0 || ret == LZMA_STREAM_END)
        {
            const ssize_t n = sizeof(buf_out) - strm.avail_out;
            if (n != safe_write(fdo, buf_out, n))
            {
                perror_msg("Failed to write decompressed data");
                close(fdi);
                close(fdo);
                lzma_end(&strm);
                return -1;
            }

            if (ret == LZMA_STREAM_END)
            {
                log_debug("Successfully decompressed coredump.");
                break;
            }

            strm.next_out = buf_out;
            strm.avail_out = sizeof(buf_out);
        }
    }

    return 0;
#else /*HAVE_LZMA*/
    const char *cmd[] = { "xzcat", "-d", "-", NULL };
    return decompress_using_fork_execvp(cmd, fdi, fdo);
#endif /*HAVE_LZMA*/
}

static int
decompress_fd_lz4(int fdi, int fdo)
{
#ifdef HAVE_LZ4
    enum { LZ4_DEC_BUF_SIZE = 64*1024u };

    LZ4F_decompressionContext_t ctx = NULL;
    LZ4F_errorCode_t c;
    char *buf = NULL;
    char *src = NULL;
    int r = 0;
    struct stat fdist;

    c = LZ4F_createDecompressionContext(&ctx, LZ4F_VERSION);
    if (LZ4F_isError(c))
    {
        log_debug("Failed to initialized LZ4: %s", LZ4F_getErrorName(c));
        r = -ENOMEM;
        goto cleanup;
    }

    buf = malloc(LZ4_DEC_BUF_SIZE);
    if (!buf)
    {
        r = -errno;
        goto cleanup;
    }

    if (fstat(fdi, &fdist) < 0)
    {
        r = -errno;
        log_debug("Failed to stat the input fd");
        goto cleanup;
    }

    src = mmap(NULL, fdist.st_size, PROT_READ, MAP_PRIVATE, fdi, 0);
    if (!src)
    {
        r = -errno;
        log_debug("Failed to mmap the input fd");
        goto cleanup;
    }

    off_t total_in = 0;
    while (fdist.st_size != total_in)
    {
        size_t used = fdist.st_size - total_in;
        size_t produced = LZ4_DEC_BUF_SIZE;

        c = LZ4F_decompress(ctx, buf, &produced, src + total_in, &used, NULL);
        if (LZ4F_isError(c))
        {
            log_debug("Failed to decode LZ4 block: %s", LZ4F_getErrorName(c));
            r = -EBADMSG;
            goto cleanup;
        }

        r = safe_write(fdo, buf, produced);
        if (r < 0)
        {
            log_debug("Failed to write decoded block");
            goto cleanup;
        }

        total_in += used;
    }
    r = 0;

cleanup:
    if (ctx != NULL)
        LZ4F_freeDecompressionContext(ctx);

    if (buf != NULL)
        free(buf);

    if (src != NULL)
        munmap(src, fdist.st_size);

    return r;
#else /*HAVE_LZ4*/
    const char *cmd[] = { "lz4", "-cd", "-", NULL};
    return decompress_using_fork_execvp(cmd, fdi, fdo);
#endif /*HAVE_LZ4*/
}

//.........这里部分代码省略.........
    len = (hdr.contentLengthB1 << 8) | hdr.contentLengthB0;
    padding = hdr.paddingLength;

    while (hdr.type == FCGI_STDIN && len == 0) {
        if (safe_read(req, &hdr, sizeof(fcgi_header)) != sizeof(fcgi_header) ||
                hdr.version < FCGI_VERSION_1) {
            return 0;
        }

        len = (hdr.contentLengthB1 << 8) | hdr.contentLengthB0;
        padding = hdr.paddingLength;
    }

    req->id = (hdr.requestIdB1 << 8) + hdr.requestIdB0;

    if (hdr.type == FCGI_BEGIN_REQUEST && len == sizeof(fcgi_begin_request)) {
        char *val;

        if (safe_read(req, buf, len+padding) != len+padding) {
            return 0;
        }

        req->keep = (((fcgi_begin_request*)buf)->flags & FCGI_KEEP_CONN);
        switch ((((fcgi_begin_request*)buf)->roleB1 << 8) + ((fcgi_begin_request*)buf)->roleB0) {
        case FCGI_RESPONDER:
            val = strdup("RESPONDER");
            zend_hash_update(&req->env, "FCGI_ROLE", sizeof("FCGI_ROLE"), &val, sizeof(char*), NULL);
            break;
        case FCGI_AUTHORIZER:
            val = strdup("AUTHORIZER");
            zend_hash_update(&req->env, "FCGI_ROLE", sizeof("FCGI_ROLE"), &val, sizeof(char*), NULL);
            break;
        case FCGI_FILTER:
            val = strdup("FILTER");
            zend_hash_update(&req->env, "FCGI_ROLE", sizeof("FCGI_ROLE"), &val, sizeof(char*), NULL);
            break;
        default:
            return 0;
        }

        if (safe_read(req, &hdr, sizeof(fcgi_header)) != sizeof(fcgi_header) ||
                hdr.version < FCGI_VERSION_1) {
            return 0;
        }

        len = (hdr.contentLengthB1 << 8) | hdr.contentLengthB0;
        padding = hdr.paddingLength;

        while (hdr.type == FCGI_PARAMS && len > 0) {
            if (safe_read(req, buf, len+padding) != len+padding) {
                req->keep = 0;
                return 0;
            }

            if (!fcgi_get_params(req, buf, buf+len)) {
                req->keep = 0;
                return 0;
            }

            if (safe_read(req, &hdr, sizeof(fcgi_header)) != sizeof(fcgi_header) ||
                    hdr.version < FCGI_VERSION_1) {
                req->keep = 0;
                return 0;
            }
            len = (hdr.contentLengthB1 << 8) | hdr.contentLengthB0;
            padding = hdr.paddingLength;
        }
    } else if (hdr.type == FCGI_GET_VALUES) {
        int j;
        unsigned char *p = buf + sizeof(fcgi_header);

        if (safe_read(req, buf, len+padding) != len+padding) {
            req->keep = 0;
            return 0;
        }

        if (!fcgi_get_params(req, buf, buf+len)) {
            req->keep = 0;
            return 0;
        }

        for (j = 0; j < sizeof(fcgi_mgmt_vars)/sizeof(fcgi_mgmt_vars[0]); j++) {
            if (zend_hash_exists(&req->env, fcgi_mgmt_vars[j].name, fcgi_mgmt_vars[j].name_len+1) == 0) {
                sprintf((char*)p, "%c%c%s%c", fcgi_mgmt_vars[j].name_len, 1, fcgi_mgmt_vars[j].name, fcgi_mgmt_vars[j].val);
                p += fcgi_mgmt_vars[j].name_len + 3;
            }
        }
        len = p - buf - sizeof(fcgi_header);
        len += fcgi_make_header((fcgi_header*)buf, FCGI_GET_VALUES_RESULT, 0, len);
        if (safe_write(req, buf, sizeof(fcgi_header)+len) != (int)sizeof(fcgi_header)+len) {
            req->keep = 0;
            return 0;
        }
        return 0;
    } else {
        return 0;
    }

    return 1;
}

int fcgi_write(fcgi_request *req, fcgi_request_type type, const char *str, int len)
{
    int limit, rest;

    if (len <= 0) {
        return 0;
    }

    if (req->out_hdr && req->out_hdr->type != type) {
        close_packet(req);
    }
#if 0
    /* Unoptimized, but clear version */
    rest = len;
    while (rest > 0) {
        limit = sizeof(req->out_buf) - (req->out_pos - req->out_buf);

        if (!req->out_hdr) {
            if (limit < sizeof(fcgi_header)) {
                if (!fcgi_flush(req, 0)) {
                    return -1;
                }
            }
            open_packet(req, type);
        }
        limit = sizeof(req->out_buf) - (req->out_pos - req->out_buf);
        if (rest < limit) {
            memcpy(req->out_pos, str, rest);
            req->out_pos += rest;
            return len;
        } else {
            memcpy(req->out_pos, str, limit);
            req->out_pos += limit;
            rest -= limit;
            str += limit;
            if (!fcgi_flush(req, 0)) {
                return -1;
            }
        }
    }
#else
    /* Optimized version */
    limit = sizeof(req->out_buf) - (req->out_pos - req->out_buf);
    if (!req->out_hdr) {
        limit -= sizeof(fcgi_header);
        if (limit < 0) limit = 0;
    }

    if (len < limit) {
        if (!req->out_hdr) {
            open_packet(req, type);
        }
        memcpy(req->out_pos, str, len);
        req->out_pos += len;
    } else if (len - limit < sizeof(req->out_buf) - sizeof(fcgi_header)) {
        if (!req->out_hdr) {
            open_packet(req, type);
        }
        if (limit > 0) {
            memcpy(req->out_pos, str, limit);
            req->out_pos += limit;
        }
        if (!fcgi_flush(req, 0)) {
            return -1;
        }
        if (len > limit) {
            open_packet(req, type);
            memcpy(req->out_pos, str + limit, len - limit);
            req->out_pos += len - limit;
        }
    } else {
        int pos = 0;
        int pad;

        close_packet(req);
        while ((len - pos) > 0xffff) {
            open_packet(req, type);
            fcgi_make_header(req->out_hdr, type, req->id, 0xfff8);
            req->out_hdr = NULL;
            if (!fcgi_flush(req, 0)) {
                return -1;
            }
            if (safe_write(req, str + pos, 0xfff8) != 0xfff8) {
                req->keep = 0;
                return -1;
            }
            pos += 0xfff8;
        }

        pad = (((len - pos) + 7) & ~7) - (len - pos);
        rest = pad ? 8 - pad : 0;

        open_packet(req, type);
        fcgi_make_header(req->out_hdr, type, req->id, (len - pos) - rest);
        req->out_hdr = NULL;
        if (!fcgi_flush(req, 0)) {
            return -1;
        }
        if (safe_write(req, str + pos, (len - pos) - rest) != (len - pos) - rest) {
            req->keep = 0;
//.........这里部分代码省略.........

int recv_sideband(const char *me, int in_stream, int out, int err)
{
	unsigned pf = strlen(PREFIX);
	unsigned sf;
	char buf[LARGE_PACKET_MAX + 2*FIX_SIZE];
	char *suffix, *term;
	int skip_pf = 0;

	memcpy(buf, PREFIX, pf);
	term = getenv("TERM");
	if (term && strcmp(term, "dumb"))
		suffix = ANSI_SUFFIX;
	else
		suffix = DUMB_SUFFIX;
	sf = strlen(suffix);

	while (1) {
		int band, len;
		len = packet_read_line(in_stream, buf + pf, LARGE_PACKET_MAX);
		if (len == 0)
			break;
		if (len < 1) {
			len = sprintf(buf, "%s: protocol error: no band designator\n", me);
			safe_write(err, buf, len);
			return SIDEBAND_PROTOCOL_ERROR;
		}
		band = buf[pf] & 0xff;
		len--;
		switch (band) {
		case 3:
			buf[pf] = ' ';
			buf[pf+1+len] = '\n';
			safe_write(err, buf, pf+1+len+1);
			return SIDEBAND_REMOTE_ERROR;
		case 2:
			buf[pf] = ' ';
			do {
				char *b = buf;
				int brk = 0;

				/*
				 * If the last buffer didn't end with a line
				 * break then we should not print a prefix
				 * this time around.
				 */
				if (skip_pf) {
					b += pf+1;
				} else {
					len += pf+1;
					brk += pf+1;
				}

				/* Look for a line break. */
				for (;;) {
					brk++;
					if (brk > len) {
						brk = 0;
						break;
					}
					if (b[brk-1] == '\n' ||
					    b[brk-1] == '\r')
						break;
				}

				/*
				 * Let's insert a suffix to clear the end
				 * of the screen line if a line break was
				 * found.  Also, if we don't skip the
				 * prefix, then a non-empty string must be
				 * present too.
				 */
				if (brk > (skip_pf ? 0 : (pf+1 + 1))) {
					char save[FIX_SIZE];
					memcpy(save, b + brk, sf);
					b[brk + sf - 1] = b[brk - 1];
					memcpy(b + brk - 1, suffix, sf);
					safe_write(err, b, brk + sf);
					memcpy(b + brk, save, sf);
					len -= brk;
				} else {
					int l = brk ? brk : len;
					safe_write(err, b, l);
					len -= l;
				}

				skip_pf = !brk;
				memmove(buf + pf+1, b + brk, len);
			} while (len);
			continue;
		case 1:
			safe_write(out, buf + pf+1, len);
			continue;
		default:
			len = sprintf(buf,
				      "%s: protocol error: bad band #%d\n",
				      me, band);
			safe_write(err, buf, len);
			return SIDEBAND_PROTOCOL_ERROR;
		}
	}
//.........这里部分代码省略.........

//.........这里部分代码省略.........
			udp_packet tmp;
			struct sockaddr_in mon_side;
			
			/* ricezione dei dal monitor */
			general_recvfrom((int32_t *) &(tmp), sizeof(udp_packet), udp_gate, &mon_side);

			if (check_port(port_list, ntohs(mon_side.sin_port)) == FALSE) /* inserimento di una, forse, nuova porta monitor */
				insert_port(&port_list, ntohs(mon_side.sin_port));
			switch (tmp.type_packet) {
				case STANDARD: {
				#ifdef O_DEBUG
					printf("id pacchetto ricevuto: %d\n", tmp.id_packet);
				#endif
					up_packets[BUFF_PACK_SIZE - 1].udp_data = tmp;	/* speranza che sia vuoto per la legge di Gigio */
					gettimeofday(&(up_packets[BUFF_PACK_SIZE - 1].delivered), NULL);
					if ((actual_zero.tv_usec == 0) && (actual_zero.tv_sec == 0))	/* nuovo zero dell'opposto balancer */
						i_2_s(s_2_i(&(up_packets[BUFF_PACK_SIZE - 1].delivered)) - (INTER_PACKET * tmp.id_packet), &actual_zero);
					mighty_f_revenge(up_packets, BUFF_PACK_SIZE - 1, &actual_zero, 0);
					if (tmp.id_packet == (last_uploaded + 1)) {	/* e' il successivo */
						clear_pack(up_packets, BUFF_PACK_SIZE - 1);
						to_app_packets[index_udpfree_pack(to_app_packets)] = tmp;
						last_uploaded = tmp.id_packet;	/* aggiorno ultimo upload */
						check_inline_pack(up_packets, to_app_packets, &last_uploaded, tmp.id_packet);
						FD_SET(new_app_fd, &writefds);
					}
					else {
						if (tmp.id_packet > last_uploaded) {
							sort_buffer(up_packets);
						}
						else	/* in ritado abissale, drop */
							clear_pack(up_packets, BUFF_PACK_SIZE - 1);
					}
					break;
				}
				default : {		/* pacchetto di notifica */
					port_monitoring *ptmp = NULL;
					stat_notify *trick = (stat_notify*) &tmp;
				#ifdef O_DEBUG_N
					printf("pacchetto id: %d\n", tmp.id_packet);
					printf("pacchetto di notifica con porta id: %d\n", trick->id_port);
				#endif
				
					if ((last_notify < trick->id_notify) || (last_seq == trick->id_sequence)) { /* non e' in ritardo abissale */
						if ((trick->id_sequence < last_seq) || (trick->id_sequence > last_seq)) { /* nuova sequenza di notifiche */
							m_id_ports.message = trick->type_packet - NOTIFY_F;
							last_seq = trick->id_sequence;
						}
						last_notify = trick->id_notify;
						m_id_ports.message--; /* nuova notifica della stessa sequenza, decrementa contatore */
						m_id_ports.ports[m_id_ports.message] = trick->id_port + 1000;
						ptmp = find_port(port_list, trick->id_port + 1000); /* abbiamo spedito ID lato mobile, vogliamo quello fixed */
						if (ptmp == NULL) { /* porta di cui il fixed non era a conoscenza */
							if (m_id_ports.message == 0) {
							/* non ci sono piu' pacchetti di notifica, costruzione della lista delle porte aperte
							 * e reset-default per la prossima volta */
								m_id_ports.message = trick->type_packet - NOTIFY_F; /* lunghezza lista porte del mobile */
								build_act_ports(&port_list, &m_id_ports);
								m_id_ports.message = 0; /* reset per la prossima notifica */
								break;
							}
							insert_port(&port_list, trick->id_port + 1000);
							break; /* esce forzatamente dallo switch */
						}
						ptmp->not_to_fixed = *trick;	/* aggiorna la notifica */
						ptmp->last_tot_packets = ptmp->tot_packets - ptmp->medium_tot_packets;
						good_percentual = partial_pp_lost(port_list);
						ptmp->ack_packets += ptmp->not_to_fixed.n_received;
						ptmp->nack_packets = ptmp->tot_packets - ptmp->ack_packets;
						ptmp->medium_tot_packets = ptmp->tot_packets;
						mighty_f_ports(&port_list); /* calcolo performance delle porte */
						if (m_id_ports.message == 0) {
						/* non ci sono piu' pacchetti di notifica, costruzione della lista delle porte aperte
						* e reset-default per la prossima volta */
							m_id_ports.message = trick->type_packet - NOTIFY_F; /* lunghezza lista porte del mobile */
							build_act_ports(&port_list, &m_id_ports);
							m_id_ports.message = 0; /* reset per la prossima notifica */
						}
					}
				}
			}
		}
		if (FD_ISSET(new_app_fd, &s_writefds)) {
			memcpy(data_app_fixed, &(to_app_packets[0].id_packet), sizeof(uint32_t));
			memcpy(&(data_app_fixed[4]), &(to_app_packets[0].data), DATA_BUFF_SIZE - sizeof(uint32_t));
			safe_write(new_app_fd, DATA_BUFF_SIZE, data_app_fixed);
			udp_shift_pack(to_app_packets, 0);	/* clear e shift */
			if (to_app_packets[0].type_packet == FALSE)	/* non ce ne sono piu' */
				FD_CLR(new_app_fd, &writefds);
		}
		if (nready == 0) {	/* timeout espirato, nessun fd pronto */
			if (check_inline_pack(up_packets, to_app_packets, &last_uploaded, up_packets[0].udp_data.id_packet - 1))
				FD_SET(new_app_fd, &writefds);
		}
		index_to = BUFF_PACK_SIZE;
		timeout.tv_sec = timeout.tv_usec = 0;
		index_to = manage_timeout(up_packets, &timeout);
	}

	return 0;
}

static int
_handle_suspend(int fd, slurmd_job_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;

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

	debug3("  uid = %d", uid);
	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;
	}

	jobacct_gather_g_suspend_poll();

	/*
	 * Signal the container
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = -1;
		errnum = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	} else {
		/* 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 (slurm_container_signal(job->cont_id, SIGTSTP) < 0) {
			verbose("Error suspending %u.%u (SIGTSTP): %m",
				job->jobid, job->stepid);
		} else
			sleep(2);

		if (slurm_container_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;
	}
	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;
}

/*
 * Process TaskProlog output
 * "export NAME=value"	adds environment variables
 * "unset  NAME"	clears an environment variable
 * "print  <whatever>"	writes that to the job's stdout
 */
static void _proc_stdout(char *buf, stepd_step_rec_t *job)
{
	bool end_buf = false;
	int len;
	char *buf_ptr, *name_ptr, *val_ptr;
	char *end_line, *equal_ptr;
	char ***env = &job->env;

	buf_ptr = buf;
	while (buf_ptr[0]) {
		end_line = strchr(buf_ptr, '\n');
		if (!end_line) {
			end_line = buf_ptr + strlen(buf_ptr);
			end_buf = true;
		}
		if (!strncmp(buf_ptr, "print ", 6)) {
			buf_ptr += 6;
			while (isspace(buf_ptr[0]))
				buf_ptr++;
			len = end_line - buf_ptr + 1;
			safe_write(1, buf_ptr, len);
		} else if (!strncmp(buf_ptr, "export ",7)) {
			name_ptr = buf_ptr + 7;
			while (isspace(name_ptr[0]))
				name_ptr++;
			equal_ptr = strchr(name_ptr, '=');
			if (!equal_ptr || (equal_ptr > end_line))
				goto rwfail;
			val_ptr = equal_ptr + 1;
			while (isspace(equal_ptr[-1]))
				equal_ptr--;
			equal_ptr[0] = '\0';
			end_line[0] = '\0';
			if (!strcmp(name_ptr, "SLURM_PROLOG_CPU_MASK")) {
				job->cpu_bind_type = CPU_BIND_MASK;
				xfree(job->cpu_bind);
				job->cpu_bind = xstrdup(val_ptr);
				if (task_g_pre_launch(job)) {
					error("Failed SLURM_PROLOG_CPU_MASK "
					      "setup");
					exit(1);
				}
			}
			debug("export name:%s:val:%s:", name_ptr, val_ptr);
			if (setenvf(env, name_ptr, "%s", val_ptr)) {
				error("Unable to set %s environment variable",
				      buf_ptr);
			}
			equal_ptr[0] = '=';
			if (end_buf)
				end_line[0] = '\0';
			else
				end_line[0] = '\n';
		} else if (!strncmp(buf_ptr, "unset ", 6)) {
			name_ptr = buf_ptr + 6;
			while (isspace(name_ptr[0]))
				name_ptr++;
			if ((name_ptr[0] == '\n') || (name_ptr[0] == '\0'))
				goto rwfail;
			while (isspace(end_line[-1]))
				end_line--;
			end_line[0] = '\0';
			debug(" unset name:%s:", name_ptr);
			unsetenvp(*env, name_ptr);
			if (end_buf)
				end_line[0] = '\0';
			else
				end_line[0] = '\n';
		}

rwfail:		 /* process rest of script output */
		if (end_buf)
			break;
		buf_ptr = end_line + 1;
	}
	return;
}

static int
_handle_signal_process_group(int fd, slurmd_job_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int signal;

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

	safe_read(fd, &signal, 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 (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 && (signal != SIGKILL)) {
		rc = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto d