osquery-1/include/osquery/logger.h

374 lines
12 KiB
C++

/*
* Copyright (c) 2014-present, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under the BSD-style license found in the
* LICENSE file in the root directory of this source tree. An additional grant
* of patent rights can be found in the PATENTS file in the same directory.
*
*/
#pragma once
#include <string>
#include <vector>
#ifdef WIN32
#define GLOG_NO_ABBREVIATED_SEVERITIES
#define GOOGLE_GLOG_DLL_DECL
#endif
#include <glog/logging.h>
#include <boost/noncopyable.hpp>
#include <osquery/core.h>
#include <osquery/flags.h>
#include <osquery/query.h>
#include <osquery/registry.h>
namespace osquery {
/**
* @brief An internal severity set mapping to Glog's LogSeverity levels.
*/
enum StatusLogSeverity {
O_INFO = 0,
O_WARNING = 1,
O_ERROR = 2,
O_FATAL = 3,
};
/// An intermediate status log line.
struct StatusLogLine {
public:
/// An integer severity level mimicking Glog's.
StatusLogSeverity severity;
/// The name of the file emitting the status log.
std::string filename;
/// The line of the file emitting the status log.
size_t line;
/// The string-formatted status message.
std::string message;
/// The ASCII time stamp for when the status message was emitted
std::string calendar_time;
/// The UNIX time for when the status message was emitted
size_t time;
/**
* @brief The host identifier at the time when logs are flushed.
*
* There is occasionally a delay between logging a status and decorating
* with the host identifier. In most cases the identifier is static so this
* does not matter. In some cases the host identifier causes database lookups.
*/
std::string identifier;
};
/**
* @brief Logger plugin feature bits for complicated loggers.
*
* Logger plugins may opt-in to additional features like explicitly handling
* Glog status events or requesting event subscribers to forward each event
* directly to the logger. This enumeration tracks, and corresponds to, each
* of the feature methods defined in a logger plugin.
*
* A specific registry call action can be used to retrieve an overloaded Status
* object containing all of the opt-in features.
*/
enum LoggerFeatures {
LOGGER_FEATURE_BLANK = 0,
LOGGER_FEATURE_LOGSTATUS = 1,
LOGGER_FEATURE_LOGEVENT = 2,
};
/**
* @brief Helper logging macro for table-generated verbose log lines.
*
* Since logging in tables does not always mean a critical warning or error
* but more likely a parsing or expected edge-case, we provide a TLOG.
*
* The tool user can set within config or via the CLI what level of logging
* to tolerate. It's the table developer's job to assume consistency in logging.
*/
#define TLOG VLOG(1)
/**
* @brief Prepend a reference number to the log line.
*
* A reference number is an external-search helper for somewhat confusing or
* seeminly-critical log lines.
*/
#define RLOG(n) "[Ref #" #n "] "
/**
* @brief Superclass for the pluggable logging facilities.
*
* In order to make the logging of osquery results and inline debug, warning,
* error status easy to integrate into your environment, we take advantage of
* a plugin interface which allows you to integrate osquery with your internal
* large-scale logging infrastructure.
*
* You may use flume, splunk, syslog, scribe, etc. In order to use your
* specific upstream logging systems, one simply needs to create a custom
* subclass of LoggerPlugin. That subclass should at least implement the
* LoggerPlugin::logString method.
*
* Consider the following example:
*
* @code{.cpp}
* class TestLoggerPlugin : public LoggerPlugin {
* public:
* osquery::Status logString(const std::string& s) {
* int i = 0;
* internal::logStringToFlume(s, i);
* std::string message;
* if (i == 0) {
* message = "OK";
* } else {
* message = "Failed";
* }
* return osquery::Status(i, message);
* }
* };
*
* REGISTER(TestLoggerPlugin, "logger", "test");
* @endcode
*/
class LoggerPlugin : public Plugin {
public:
/// The LoggerPlugin PluginRequest action router.
Status call(const PluginRequest& request, PluginResponse& response) override;
/**
* @brief A feature method to decide if Glog should stop handling statuses.
*
* Return true if this logger plugin's #logStatus method should handle Glog
* statuses exclusively. If true then Glog will stop writing status lines
* to the configured log path.
*
* @return false if this logger plugin should NOT handle Glog statuses.
*/
virtual bool usesLogStatus() {
return false;
}
/**
* @brief A feature method to decide if events should be forwarded.
*
* See the optional logEvent method.
*
* @return false if this logger plugin should NOT handle events directly.
*/
virtual bool usesLogEvent() {
return false;
}
/**
* @brief Set the process name.
*/
void setProcessName(const std::string& name) {
process_name_ = name;
}
/**
* @brief Get the process name.
*/
const std::string& name() const {
return process_name_;
}
/** @brief Virtual method which should implement custom logging.
*
* LoggerPlugin::logString should be implemented by a subclass of
* LoggerPlugin which needs to log a string in a custom way.
*
* @return an instance of osquery::Status which indicates the success or
* failure of the operation.
*/
virtual Status logString(const std::string& s) = 0;
/**
* @brief See the usesLogStatus method, log a Glog status.
*
* @param log A vector of parsed Glog log lines.
* @return Status non-op indicating success or failure.
*/
virtual Status logStatus(const std::vector<StatusLogLine>& log) {
(void)log;
return Status(1, "Not enabled");
}
/**
* @brief Optionally handle snapshot query results separately from events.
*
* If a logger plugin wants to write snapshot query results (potentially
* large amounts of data) to a specific sink it should implement logSnapshot.
* Otherwise the serialized log item data will be forwarded to logString.
*
* @param s A special log item will complete results from a query.
* @return log status
*/
virtual Status logSnapshot(const std::string& s) {
return logString(s);
}
/**
* @brief Optionally handle each published event via the logger.
*
* It is possible to skip the database representation of event subscribers
* and instead forward each added event to the active logger plugin.
*/
virtual Status logEvent(const std::string& /*s*/) {
return Status(1, "Not enabled");
}
protected:
/**
* @brief Initialize the logger with the name of the binary and any status
* logs generated between program launch and logger start.
*
* The logger initialization is called once CLI flags have been parsed, the
* registry items are constructed, extension routes broadcasted and extension
* plugins discovered (as a logger may be an extension plugin) and the config
* has been loaded (which may include additional CLI flag-options).
*
* All of these actions may have generated VERBOSE, INFO, WARNING, or ERROR
* logs. The internal logging facility, Glog, collects these intermediate
* status logs and a customized log sink buffers them until the active
* osquery logger's `init` method is called.
*
* @param binary_name The string name of the process (argv[0]).
* @param log The set of status (INFO, WARNING, ERROR) logs generated before
* the logger's `init` method was called.
*/
virtual void init(const std::string& binary_name,
const std::vector<StatusLogLine>& log) = 0;
private:
std::string process_name_;
};
/// Set the verbose mode, changes Glog's sinking logic and will affect plugins.
void setVerboseLevel();
/**
* @brief Start status logging to a buffer until the logger plugin is online.
*
* This will also call google::InitGoogleLogging. Use the default init_glog
* to control this in tests to protect against calling the API twice.
*/
void initStatusLogger(const std::string& name, bool init_glog = true);
/**
* @brief Initialize the osquery Logger facility by dumping the buffered status
* logs and configurating status log forwarding.
*
* initLogger will disable the `BufferedLogSink` facility, dump any status logs
* emitted between process start and this init call, then configure the new
* logger facility to receive status logs.
*
* The `forward_all` control is used when buffering logs in extensions.
* It is fine if the logger facility in the core app does not want to receive
* status logs, but this is NOT an option in extensions/modules. All status
* logs must be forwarded to the core.
*
* @param name The process name.
*/
void initLogger(const std::string& name);
/**
* @brief Log a string using the default logger receiver.
*
* Note that this method should only be used to log results. If you'd like to
* log normal osquery operations, use Google Logging.
*
* @param message the string to log
* @param category a category/metadata key
*
* @return Status indicating the success or failure of the operation
*/
Status logString(const std::string& message, const std::string& category);
/**
* @brief Log a string using a specific logger receiver.
*
* Note that this method should only be used to log results. If you'd like to
* log normal osquery operations, use Google Logging.
*
* @param message the string to log
* @param category a category/metadata key
* @param receiver a string representing the log receiver to use
*
* @return Status indicating the success or failure of the operation
*/
Status logString(const std::string& message,
const std::string& category,
const std::string& receiver);
/**
* @brief Log results of scheduled queries to the default receiver
*
* @param item a struct representing the results of a scheduled query
*
* @return Status indicating the success or failure of the operation
*/
Status logQueryLogItem(const QueryLogItem& item);
/**
* @brief Log results of scheduled queries to a specified receiver
*
* @param item a struct representing the results of a scheduled query
* @param receiver a string representing the log receiver to use
*
* @return Status indicating the success or failure of the operation
*/
Status logQueryLogItem(const QueryLogItem& item, const std::string& receiver);
/**
* @brief Log raw results from a query (or a snapshot scheduled query).
*
* @param item the unmangled results from the query planner.
*
* @return Status indicating the success or failure of the operation
*/
Status logSnapshotQuery(const QueryLogItem& item);
/**
* @brief Sink a set of buffered status logs.
*
* When the osquery daemon uses a watcher/worker set, the watcher's status logs
* are accumulated in a buffered log sink. Well-performing workers should have
* the set of watcher status logs relayed and sent to the configured logger
* plugin.
*
* Status logs from extensions will be forwarded to the extension manager (core)
* normally, but the watcher does not receive or send registry requests.
* Extensions, the registry, configuration, and optional config/logger plugins
* are all protected as a monitored worker.
*/
void relayStatusLogs(bool async = false);
/// Inspect the number of internal-buffered status log lines.
size_t queuedStatuses();
/// Inspect the number of active internal status log sender threads.
size_t queuedSenders();
/**
* @brief Write a log line to the OS system log.
*
* There are occasional needs to log independently of the osquery logging
* facilities. This allows a feature (not a table) to bypass all osquery
* configuration and log to the OS system log.
*
* Linux/Darwin: this uses syslog's LOG_NOTICE.
*/
void systemLog(const std::string& line);
} // namespace osquery