LittleDemon WebShell

#!/opt/imunify360/venv/bin/python3
import argparse
import asyncio
import logging
import sys
import os
from pathlib import Path
from contextlib import contextmanager, suppress
import gc
import time
import signal
from concurrent.futures import ThreadPoolExecutor
from functools import partial
from typing import Tuple

import daemon
from lockfile import AlreadyLocked
import daemon.pidfile
import sentry_sdk

from defence360agent.internals.global_scope import g
from defence360agent.contracts.plugins import MessageSink, MessageSource
from defence360agent.internals.the_sink import TheSink
from defence360agent import files
from imav.internals.lazy_load import AVSource
from defence360agent.subsys import systemd_notifier
import defence360agent.internals.logger
from defence360agent.api import health
from defence360agent.contracts.config import Model
from defence360agent.contracts.license import LicenseCLN
from defence360agent.contracts.plugins import BasePlugin
from defence360agent.internals.cln import subscribe_to_license_changes
from defence360agent.internals.iaid import IndependentAgentIDAPI
from defence360agent.internals.lazy_load import CoreSource
from defence360agent.contracts.hook_events import HookEvent
from defence360agent.model import instance, simplification, tls_check
from defence360agent.utils import (
    create_task_and_log_exceptions,
    is_root_user,
    importer,
    Scope,
    is_systemd_boot,
    Task,
)
from defence360agent.contracts.config import (
    Core,
    Merger,
    ConfigsValidator,
)
from defence360agent.utils.common import DAY, rate_limit
from defence360agent.utils.cli import EXITCODE_GENERAL_ERROR
from defence360agent.utils.check_db import is_db_corrupted
from im360.application.settings import configure
from im360.internals.core import ip_versions
from im360.internals.lazy_load import IM360Source
from im360.subsys import features


# Increase recursion depth to allow malware scanner into deeply nested
# directories with absolute path length up to 4096 symbols
_MAX_RECURSION_DEPTH = 2100
_DB_IS_CORRUPTED_FLAG = Path("%s.is_corrupted" % Model.RESIDENT_PATH)
_DB_IS_CORRUPTED_MSG = (
    "Imunify360 database is corrupt. "
    "Application cannot run with corrupt database. "
    "Please, contact Imunify360 support team at "
    "https://cloudlinux.zendesk.com"
)


IM360_RESIDENT_PLUGINS_PACKAGES = (
    "defence360agent.plugins.accumulate",
    "im360.plugins.resident",
    "im360.plugins.protector",
    "im360.plugins.sensor",
)

logger = logging.getLogger("resident")
throttled_log_error = rate_limit(period=DAY)(logger.error)


class TaskFactory:
    def __init__(self):
        self.pool = set()

    def __call__(self, loop, coro):
        task = Task(coro, loop=loop)
        self.pool.add(task)
        task.add_done_callback(self.pool.discard)
        return task


def get_plugins() -> list:
    """Return plugins in unspecified order."""
    importer.load_packages(
        CoreSource.MESSAGES + AVSource.MESSAGES + IM360Source.MESSAGES
    )
    importer.load_packages(IM360_RESIDENT_PLUGINS_PACKAGES)
    # use lexicographical order (but don't rely on it in code)
    return sorted(
        [
            plugin
            for plugin in BasePlugin.get_active_plugins()
            if plugin.SCOPE is not Scope.AV
        ],
        key=lambda item: f"{item.__module__}.{item.__name__}",
    )


@contextmanager
def log_and_suppress_error(message):
    """Log *message* on any error & suppress it."""
    try:
        yield
    except Exception as e:
        sentry_sdk.capture_exception(e)
        logger.error(message)


async def update_health_sensor():
    if LicenseCLN.is_valid():
        health.sensor.registered()
    else:
        health.sensor.unregistered()


async def init_actions():
    # Any uncaught exceptions here prevents the agent from
    # starting. Non-critical (agent can continue) functionality should
    # either be moved to a plugin or caught&log *all* its exceptions.

    # Also nothing should block here indefinitely (any operation that
    # may block should have a timeout).

    ip_versions.init()
    subscribe_to_license_changes(features.update_repos)
    subscribe_to_license_changes(update_health_sensor)
    instance.db.execute_sql(
        "ATTACH '{}' AS proactive".format(Model.PROACTIVE_PATH)
    )


def parse_cli():
    parser = argparse.ArgumentParser(description="Run imunify agent")
    parser.add_argument(
        "-v",
        dest="verbose",
        action="count",
        default=0,
        help=(
            "Level of logging. Each value corresponds to:"
            "1 - console only log level,"
            "2 - previous plus add network log,"
            "3 - all previous plus add process message log,"
            "4 - all previous plus add debug log"
        ),
    )
    parser.add_argument("--daemon", action="store_true", help="run as daemon")
    parser.add_argument(
        "--pidfile",
        default="/var/run/imunify360.pid",
        help="use with --daemon",
    )
    parser.add_argument("--log-config", help="logging config filename")
    return parser.parse_args(sys.argv[1:])


def _check_able_to_start(pidfile):
    if is_db_corrupted(db_path=Model.RESIDENT_PATH):
        if not _DB_IS_CORRUPTED_FLAG.exists():
            logger.error(_DB_IS_CORRUPTED_MSG)
            _DB_IS_CORRUPTED_FLAG.touch()
        else:
            logger.warning(_DB_IS_CORRUPTED_MSG)
        sys.exit(EXITCODE_GENERAL_ERROR)
    else:
        with suppress(FileNotFoundError):
            _DB_IS_CORRUPTED_FLAG.unlink()


def _daemonize(pidfilepath):
    logger.info("Run as daemon [pidfile = %s]", pidfilepath)

    dc = daemon.DaemonContext()
    dc.pidfile = daemon.pidfile.PIDLockFile(pidfilepath)

    dc.prevent_core = False
    dc.umask = Core.FILE_UMASK
    if is_systemd_boot():
        dc.detach_process = False
    else:
        dc.detach_process = True
    dc.files_preserve = defence360agent.internals.logger.get_fds()
    try:
        dc.open()
    except AlreadyLocked:
        logger.error("PID file already locked by another process")
        sys.exit(EXITCODE_GENERAL_ERROR)
    gc.collect()

    # quirk: somehow this is needed for root logger messages to do not
    #        propagate to specialized loggers, e.g. 'perf', 'nework'
    defence360agent.internals.logger.reconfigure()


def _tls_check_reset(loop):
    # init thread id for simplification.run_in_executor() worker thread
    loop.run_until_complete(
        simplification.run_in_executor(loop, tls_check.reset)
    )


def validate_configs_on_start(loop):
    try:
        ConfigsValidator.validate_config_layers()
    except Exception as e:
        from defence360agent.hooks.execute import execute_hooks

        agent_misconfig = HookEvent.AgentMisconfig(error=repr(e))
        loop.run_until_complete(execute_hooks(agent_misconfig))
        logger.warning(str(e))
        sys.exit(EXITCODE_GENERAL_ERROR)


def plugin_instances(objs, pclass):
    return [p for p in objs if isinstance(p, pclass)]


def _start_plugins(loop, plugin_classes) -> Tuple[TheSink, list, list]:
    plugins = [plugin_class() for plugin_class in plugin_classes]

    # instantiate sinks
    sinks = plugin_instances(plugins, MessageSink)
    for s in sinks:
        logger.info("Creating sink %r", s)
        loop.run_until_complete(s.create_sink(loop))

    # instantiate sources
    the_sink = TheSink(sinks, loop)
    sources = plugin_instances(plugins, MessageSource)
    for s in sources:
        logger.info("Creating source %r", s)
        loop.run_until_complete(s.create_source(loop, the_sink))

    the_sink.start()

    return the_sink, sinks, sources


def _setup_signal_handlers(loop, shutdowntask):
    called = False  # whether the signal handler was called already

    def _sighandler(loop, sig):
        nonlocal called
        if not called:
            called = True
            logger.info("Caught %s", sig)
            # note: store ref, to keep the task alive, just in case
            called = create_task_and_log_exceptions(loop, shutdowntask)
        else:
            logger.info(
                "Caught %s. Shutdown task is already running, please wait.",
                sig,
            )

    for sig in (signal.SIGINT, signal.SIGTERM, signal.SIGUSR1, signal.SIGUSR2):
        loop.add_signal_handler(sig, _sighandler, loop, sig)


async def _shutdown_task(loop, the_sink, plugin_list):
    with log_and_suppress_error("marking the start of the shutdown process"):
        # (there is SHUTDOWN_TIMEOUT)
        health.sensor.shutting_down(time.time())

    logger.info("shutdown task starting, pid=%s", os.getpid())
    with log_and_suppress_error(
        "preventing new messages (if any) processing to start"
    ):
        g.sensor_server.close()
        # note: first exception is propagated; tasks are no canceled
        await asyncio.gather(
            g.sensor_server.wait_closed(), the_sink.shutdown()
        )

    for plugin in plugin_list:
        with log_and_suppress_error(
            "This happened while shutting down a plugin!!"
        ):
            logger.info(
                "Shutting down %s.%s...",
                plugin.__class__.__module__,
                plugin.__class__.__name__,
            )
            # make shutting down running task be a responsibility
            # of a particular plugin but not of a universal shotgun
            await plugin.shutdown()

    with log_and_suppress_error("shutting down IAID API"):
        await IndependentAgentIDAPI.shutdown()

    # Wait for graceful web-server restart (if it was started before shutdown)
    if (restart_task := g.get("web_server_restart_task")) is not None:
        with log_and_suppress_error("waiting for web server restart"):
            await asyncio.wait_for(restart_task)

    with log_and_suppress_error("stopping loop"):
        loop.stop()

    logger.info("shutdown task finished, pid=%s", os.getpid())


def start(plugin_classes: list, init_actions) -> None:
    """Common function for agent service startup.

    plugin_classes is a list of classes implementing message processing
    plugins. init_actions is a coroutine that will be called prior to starting
    RPC and message processing."""
    if not is_root_user():
        logger.info("Imunify agent could be started by the root user only!")
        sys.exit(EXITCODE_GENERAL_ERROR)

    args = parse_cli()

    defence360agent.internals.logger.setLogLevel(args.verbose)
    if args.log_config or os.environ.get("IMUNIFY360_LOGGING_CONFIG_FILE"):
        defence360agent.internals.logger.update_logging_config_from_file(
            args.log_config or os.environ.get("IMUNIFY360_LOGGING_CONFIG_FILE")
        )

    sys.setrecursionlimit(_MAX_RECURSION_DEPTH)

    _check_able_to_start(args.pidfile)

    if args.daemon:
        _daemonize(args.pidfile)
        systemd_notifier.notify(systemd_notifier.AgentState.DAEMONIZED)

    health.sensor.starting(time.time())
    if not LicenseCLN.is_registered():
        health.sensor.unregistered()

    loop = asyncio.get_event_loop()
    _cpu = os.cpu_count()
    # https://docs.python.org/3/library/concurrent.futures.html#concurrent.futures.ThreadPoolExecutor
    # default's in Python 3.8
    loop.set_default_executor(
        ThreadPoolExecutor(max_workers=min(32, _cpu + 4 if _cpu else 5))
    )
    loop.set_task_factory(TaskFactory())
    try:
        _tls_check_reset(loop)
        instance.db.init(Model.PATH)
        instance.db.execute_sql(
            "ATTACH '{}' AS resident".format(Model.RESIDENT_PATH)
        )

        validate_configs_on_start(loop)
        Merger.update_merged_config()

        loop.run_until_complete(init_actions())

        # If this is first agent run - we expect that *only* non-resident
        # process will download all of the static files
        # If it isn't first agent run - essential files already downloaded
        # and will be updated asynchronously
        if not loop.run_until_complete(files.essential_files_exist()):
            logger.info("Essential files are missing. Try again later.")
            # wait a while to avoid restarting the service too often
            loop.run_until_complete(asyncio.sleep(5))
            sys.exit(EXITCODE_GENERAL_ERROR)

        the_sink, sinks, sources = _start_plugins(loop, plugin_classes)
        logger.info("Message Bus started")
        agent_started = HookEvent.AgentStarted(
            version=Core.VERSION, resident=True
        )
        create_task_and_log_exceptions(
            loop, the_sink.process_message, agent_started
        )

        # note: plugins are started before the shutdown task has been setup
        #  therefore plugin.shutdown() won't be called before create_source()
        _setup_signal_handlers(
            loop, partial(_shutdown_task, loop, the_sink, sinks + sources)
        )
        loop.run_forever()
    finally:
        # closing the loop after loop.stop() cuts off pending tasks stacktraces
        loop.close()


def run():
    configure(resident=True)
    plugins = get_plugins()
    start(plugins, init_actions)


if __name__ == "__main__":
    run()