zeus911/freewaf

High-performance WAF built on the OpenResty stack

##Name

FreeWAF - High-performance WAF built on the OpenResty stack

##Status

Development of new features is currently on hiatus. New bugs and questions opened in the issue tracker will be answered within a day or two, and performance impacting / security related issues will be continue to be patched.

##Description

FreeWAF is a reverse proxy WAF built using the OpenResty stack. It uses the Nginx Lua API to analyze HTTP request information and process against a flexible rule structure. FreeWAF is distributed with a ruleset that mimics the ModSecurity CRS, as well as a few custom rules built during initial development and testing.

FreeWAF was initially developed by Robert Paprocki for his Master's thesis at Western Governor's University.

##Requirements

FreeWAF requires several third-party resty lua modules, though these are all packaged with FreeWAF, and thus do not need to be installed separately. It is recommended to install FreeWAF on a system running the OpenResty software bundle; FreeWAF has not been tested on platforms built using separate Nginx source and Nginx Lua module packages.

For optimal regex compilation performance, it is recommended to build Nginx/OpenResty with a version of PCRE that supports JIT compilation. If your OS does not provide this, you can build JIT-capable PCRE directly into your Nginx/OpenResty build. To do this, reference the path to the PCRE source in the --with-pcre configure flag. For example:

	# ./configure --with-pcre=/path/to/pcre/source --with-pcre-jit

You can download the PCRE source from the PCRE website. See also my blog post for a step-by-step walkthrough on building OpenResty with a JIT-enabled PCRE library.

##Performance

FreeWAF was designed with efficiency and scalability in mind. It leverages Nginx's asynchronous processing model and an efficient design to process each transaction as quickly as possible. Early testing has show that deployments implementing all provided rulesets, which are designed to mimic the logic behind the ModSecurity CRS, process transactions in roughly 300-500 microseconds per request; this equals the performance advertised by Cloudflare's WAF. Tests were run on a reasonable hardware stack (E3-1230 CPU, 32 GB RAM, 2 x 840 EVO in RAID 0), maxing at roughly 15,000 requests per second. See this blog post for more information.

##Installation

Clone the FreeWAF repo into Nginx/OpenResty's Lua package path. Module setup and configuration is detailed in the synopsis.

Note that by default FreeWAF runs in SIMULATE mode, to prevent immediately affecting an application; users who wish to enable rule actions must explicitly set the operational mode to ACTIVE.

##Synopsis

	http {
		-- include FreeWAF in the lua_package_path
		lua_package_path '/usr/local/openresty/lualib/FreeWAF/?.lua;;';
	}

	server {
		location / {
			access_by_lua '
				FreeWAF = require "FreeWAF.fw"

				-- instantiate a new instance of the module
				local fw = FreeWAF:new()

				-- setup FreeWAF to deny requests that match a rule
				fw:set_option("mode", "ACTIVE")

				-- run the firewall
				fw:exec()
			';
		}
	}

##Options

Module options can be configured using the set_option function. Details for available options are provided below.

###mode

Default: SIMULATE

Sets the operational mode of the module. Options are ACTIVE, INACTIVE, and SIMULATE. In ACTIVE mode, rule matches are logged and actions are run. In SIMULATE mode, FreeWAF loops through each enabled rule and logs rule matches, but does not complete the action specified in a given run. INACTIVE mode prevents the module from running.

By default, SIMULATE is selected if a mode is not explicitly set; this requires new users to actively implement blocking by setting the mode to ACTIVE.

Example:

	location / {
		access_by_lua '
			fw:set_option("mode", "ACTIVE")
		';
	}

###whitelist

Default: none

Adds an address to the module whitelist. Whitelisted addresses will not have any rules applied to their requests, and will be immediately passed through the module.

Example:

	location / {
		access_by_lua '
			fw:set_option("whitlist", "127.0.0.1")
		';
	}

Multiple addresses can be whitelisted by passing a table of addresses to set_option.

###blacklist

Default: none

Adds an address to the module blacklist. Blacklisted addresses will not have any rules appled to their requests, and will be immediately rejected by the module (Nginx will return a 403 to the client)

Example:

	location / {
		access_by_lua '
			fw:set_option("blacklist", "5.6.7.8")
		';
	}

Multiple addresses can be whitelisted by passing a table of addresses to set_option. Note that blacklists are processed after whitelists, so an address that is whitelisted and blacklisted will always be processed as a whitelisted address.

###ignore_rule

Default: none

Instructs the module to ignore a specified rule ID. Note that ignoring rules in a chain will likely cause many headaches, so it's best to ignore all rules in a given chain if one rule is causing problems.

Example:

	location / {
		access_by_lua '
			fw:set_option("ignore_rule", 40294)
		';
	}

###ignore_ruleset

Default: none

Instructs the module to ignore an entire ruleset. This can be useful when some rulesets (such as the SQLi or XSS CRS rulesets) are too prone to false positives, or aren't applicable to your web app.

Example:

	location / {
		access_by_lua '
			fw:set_option("ignore_ruleset", 40000)
		';
	}

###score_threshold

Default: 5

Sets the threshold for anomaly scoring. When the threshold is reached, FreeWAF will deny the request.

Example:

	location / {
		access_by_lua '
			fw:set_option("score_threshold", 10)
		';
	}

###allowed_content_types

Default: none

Defines one or more Content-Type headers that will be allowed. Requests that do not contain a Content-Type matching one of these values, or application/x-www-form-urlencoded or multipart/form-data, will be rejected.

Example:

	location / {
		access_by_lua '
			-- define a single allowed Content-Type value
			fw:set_option("allowed_content_types", "text/xml")

			-- defines multiple allowed Content-Type values
			fw:set_option("allowed_content_types", { "text/html", "text/json", "application/json" })
		';
	}

Note that mutiple set_option calls with a parameter of allowed_content_types will simply override the existing options table, so if you want to define multiple allowed content types, you must define them as a Lua table as shown above.

###debug

Default: false

Disables/enables debug logging. Debug log statements are printed to the error_log. Note that debug logging is very expensive and should not be used in production environments.

Example:

	location / {
		access_by_lua '
			fw:set_option("debug", true)
		';
	}

###debug_log_level

Default: ngx.INFO

Sets the nginx log level constant used for debug logging.

Example:

	location / {
		access_by_lua '
			fw:set_option("debug_log_level", ngx.DEBUG)
		';
	}

###event_log_level

Default: ngx.INFO

Sets the nginx log level constant used for event logging.

Example:

	location / {
		access_by_lua '
			fw:set_option("event_log_level", ngx.WARN)
		';
	}

###event_log_verbosity

Default: 1

Sets the verbosity used in writing event log notification. The higher the verbosity, the more information will be included in the JSON blob generated for each notification.

Example:

	location / {
		access_by_lua '
			-- default verbosity. the client IP, request URI, rule match data, and rule ID will be logged
			fw:set_option("event_log_verbosity", 1)

			-- the rule description will be written in addition to existing data
			fw:set_option("event_log_verbosity", 2)

			-- the rule description, options and action will be written in addition to existing data
			fw:set_option("event_log_verbosity", 3)

			-- the entire rule definition, including the match pattern, will be written in addition to existing data
			-- note that for some rule definitions, such as the XSS and SQLi rulesets, this pattern can be large
			fw:set_option("event_log_verbosity", 4)
		';
	}

###event_log_target

Default: error

Defines the destination for event logs. FreeWAF currently supports logging to the error log, a separate file on the local file system, or a remote UDP server. In the latter two cases, event logs are buffered and flushed when a defined threshold is reached (see below for further options regarding event logging options).

Example:

	location / {
		access_by_lua '
			-- send event logs to the server's error_log location (default)
			fw:set_option("event_log_target", "error")

			-- send event logs to a local file on disk
			fw:set_option("event_log_target", "file")

			-- send event logs to a remote UDP server
			fw:set_option("event_log_target", "socket")
		';
	}

Note that, due to a limition in the logging library used, only a single target socket (and separate file target) can be defined. This is to say, you may elect to use both socket and file logging in different locations, but you may only configure one socket target with a specific host/port combination; if you configure a second host/port combination, data will not be properly logged. Similarly, you may only define one file path if using a file logging target; writes to a second path location will be lost.

###event_log_target_host

Default: none

Defines the target server for event logs that target a UDP server.

Example:

	location / {
		access_by_lua '
			fw:set_option("event_log_target_host", "10.10.10.10")
		';
	}

###event_log_target_port

Default: none

Defines the target port for event logs that target a UDP server.

Example:

	location / {
		access_by_lua '
			fw:set_option("event_log_target_port", 9001)
		';
	}

###event_log_target_path

Default: none

Defines the target path for event logs that target a local file system location.

Example:

	location / {
		access_by_lua '
			fw:set_option("event_log_target_path", "/var/log/freewaf/event.log")
		';
	}

This path must be in a location writeable by the nginx user. Note that, by nature, on-disk logging can cause significant performance degredation in high-concurrency environments.

###event_log_buffer_size

Default: 4096

Defines the threshold size, in bytes, of the buffer to be used to hold event logs. The buffer will be flushed when this threshold is met.

Example:

	location / {
		access_by_lua '
			-- 8 KB event log message buffer
			fw:set_option("event_log_buffer_size", 8192)
		';
	}

###event_log_periodic_flush

Default: none

Defines an interval, in seconds, at which the event log buffer will periodically flush. If no value is configured, the buffer will not flush periodically, and will only flush when the event_log_buffer_size threshold is reached. Configure this option for very low traffic sites that may not receive any event log data in a long period of time, to prevent stale data from sitting in the buffer.

	location / {
		access_by_lua '
			-- flush the event log buffer every 30 seconds
			fw:set_option("event_log_periodic_flush", 30)
		';
	}

###storage_zone

Default: none

Defines the lua_shared_dict that will be used to hold persistent storage data. This zone must be defined in the http{} block of the configuration.

Example:

	http {
		-- define a 64M shared memory zone to hold persistent storage data
		lua_shared_dict persistent_storage 64m;
	}

	location / {
		access_by_lua '
			fw:set_option("storage_zone", "persistent_storage")
		';
	}

Multiple shared zones can be defined and used, though only one zone can be defined per configuration location. If a zone becomes full and the shared dictionary interface cannot add additional keys, the following will be entered into the error log:

Could not add key to persistent storage, increase the size of the lua_shared_dict

###disable_pcre_optimization

Default: false

Removes the oj flags from all ngx.re.match, ngx.re.find, and ngx.re.sub calls. This may be useful in some cases where older PCRE libraries are used, but will cause severe performance degradation, so its use is strongly discouraged; users are instead encouraged to build OpenResty with a modern, JIT-capable PCRE library.

Example:

	location / {
		access_by_lua '
			fw:set_option("disable_pcre_optimization", 30)
		';
	}

##Included Rulesets

FreeWAF is distributed with a number of rulesets that are designed to mimic the functionality of the ModSecurity CRS. For reference, these rulesets are listed here:

• 10000: Whitelist/blacklis handling
• 11000: Local policty whitelisting
• 20000: HTTP protocol violation
• 21000: HTTP protocol anomalies
• 35000: Malicious/suspect user agents
• 40000: Generic attacks
• 41000: SQLi
• 42000: XSS
• 90000: Custom rules/virtual patching
• 99000: Anomaly score handling

##Rule Definitions

FreeWAF uses Lua tables to define its rules. Rules are grouped based on purpose and severity, defined as a ruleset. The included rulesets were created to mimic the functionality of the ModSecurity CRS. Each rule requires the following elements:

###id

A unique integer use to define each rule. By convention, the first two digits in a rule match those of its parent ruleset.

###description

A string that describes the purpose of the rule. This is purely descriptive.

###action

An enum (currently implemented as a string) that defines how the rule processor will act if a rule is a positive match. See the section on rule actions for available options.

###opts

A table that defines options specific to rule. The following options are currently supported:

• chainchild: Defines a rule that is part of a rule chain.
• chainend: Defines the last rule in the rule chain.
• nolog: Do not create a log entry if a rule match occurs. This is most commonly used in rule chains, with rules that have the CHAIN action (to avoid unnecessarily large quantities of log entries).
• parsepattern: Activate dynamic string parsing of the rule's var.pattern field; see the section on dynamic string parsing for more details.
• score: Defines the score for a rule with the SCORE action. Must be a numeric value.
• setvar: Defines persistent storage data key, value and optional expiry time.
• skipend: Ends a skip chain. Note that the rule containing this option will be included as part of the skip chain, e.g. it will not be processed.
• transform: Defines how collection data is altered as an anti-evasion technique. Multiple transforms for a single collection can be specified by defining the transform option value as a table itself. See the section on data transformation for more detail.

###var

A table that defines the rule's signature. Each var table must contain the following keys:

• type: Defines which collection of request data to parse; see the collections description for available options.
• opts: Defines options specific to the rule's signature. This value may be nil, or a table with a specific key/value definition. See the collections description for more detail regarding request data parsing.
• pattern: Defines the target match. This value can be a string, numeric value, table (for PM operators), or a regular expression. All regexes are case-insensitive.
• operator: Defines how to match the request against the pattern. See the section on operators for currently supported options.

##Actions

The following rule actions are currently supported:

• ACCEPT: Explicitly accepts the request, stopping all further rule processing and passing the request to the next phase handler.
• CHAIN: Sets a flag in the rule processor to proceed to the next rule in the rule chain. Rule chaining allows the rule processor to mimic logical AND operations; multiple rules can be chained together to define very specific signatures. If a rule in a rule chain does not match, all further rules in the chain are skipped.
• DENY: Explictly denies the request, stopping all further rule processing and exiting the phase handler with a 403 response (ngx.HTTP_FORBIDDEN).
• IGNORE: No action is taken, rule processing continues.
• LOG: A placeholder, as all rule matches that do not have the nolog opton set will be logged.
• SCORE: Increments the running request score by the score defined in the rule's option table.
• SETVAR: Set a persistent variable, using the setvar rule options table.
• SKIP: Skips processing of all further rules until a rule with the skipend flag is specified.

##Operators

The following pattern operators are currently supported:

• EQUALS: Matches using the == operator; comparison values can be any Lua primitive that can be compared directly (most commonly this is strings or integers).
• EXISTS: Searches for the existence of a given key in a table.
• GREATER: Matches using the > operator. Returns true if the collection data is greater than the pattern. Most commonly this is used for comparing running counters stored in persistent storage.
• PM: Performs an efficient pattern match using Aho-Corasick searching.
• REGEX: Matches using Perl compatible regular expressions.

All operators have a corresponding negated option, e.g., NOT_EQUALS, NOT_EXISTS, etc.

##Collections

FreeWAF's rule processor works on a basic principle of matching a pattern against a given collection. The following collections are currently supported:

• BLACKLIST: A table containing user-defined blacklisted IPs.
• COOKIES: A table containing the values of the cookies sent in the request.
• HEADERS: A table containing the request headers. Note that cookies are not included in this collection.
• HEADER_NAMES: A table containing the keys of the HEADERS table. Note that header names are automatically converted to a lowercase form.
• HTTP_VERSION: An integer representation of the HTTP version used in the request.
• IP: The IP address of client.
• METHOD: The HTTP method specified in the request.
• REQUEST_ARGS: A table containing the keys and values of all the arguments in the request, including query string arguments, POST arguments, and request cookies.
• REQUEST_BODY: A table containing the request body. This typically contains POST arguments.
• SCORE: An integer representing the currently anomaly score for the request.
• SCORE_THRESHOLD: An integer representing the user-defined score threshold.
• URI: The request URI.
• URI_ARGS: A table containing the request query strings.
• USER_AGENT: The value of the User-Agent header.
• VAR: The persistent storage variable collection. Specific values are obtained by defining the value key of the rule's var.opts table (see below).
• WHITELIST: A table containing user-defined whitelisted IPs.

Collections can be parsed based on the contents of a rule's var.opts table. This table must contain two keys: key, which defines how to parse the collection, and value, which determines what to parse out of the collection. The following values are supported for key:

• all: Retrieves both the keys and values of the collection. Note that this key does not require a value counterpart.
• ignore: Returns the collection minus the key (and its associated value) specified.
• keys: Retrieves the keys in the given collection. For example, the HEADER_NAMES collection is just a shortcut for the HEADERS collecton parsed by { key = "keys" }. Note that this key does not require a value counterpart.
• specific: Retrieves a specific value from the collection. For example, the USER_AGENT collection is just a shortcut for the HEADERS collections parsed by { key = "specific", value = "user-agent" }.
• values: Retrieves the values in the given collection. Note that this key does not require a value counterpart.

##Data Transformation

FreeWAF has the ability to modify request data, similar to ModSecurity's transformation pipeline, as an anti-evasion tactic. Request data is not permanently modified before being sent upstream; local copies of data collections are used as the basis for transformation. The following data transforms are available:

• base64_decode: Decode a Base64-encoded value.
• base64_encode: Encode data into a Base64 representation.
• compress_whitespace: Globally replace all sequential whitespace characters with a single ' ' space character.
• html_decode: Decode an HTML-encoded string.
• lowercase: Convert all uppercase alphabetic characters to their lowercase varients.
• remove_comments: Globally remove all C-style comment characters and their enclosed data. For example, the string UNI/*xxx*/ON SELECT would be transformed to UNION SELECT.
• remove_whitespace: Globally remove all whitespace characters.
• replace_comments: Globally replace all C-style comment characters and their enclosed data with a single ' ' space character. For example, the string UNION/*xxxx*/SELECT would be transformed to UNION SELECT.
• uri_decode: Decode a string based on URI encoding rules.

##Dynamic Parsing in Rule Definitions

Certain parts of a rule definition may be dynamically defined at runtime via a special syntax %{VAL}, where VAL is a key in the collections table. This allows FreeWAF to take advantage of changing values throughout the life of one or multiple requests, greatly increasing flexibility. For example, including the string %{IP} in a dynamically parsed rule definition will translate to the IP address of the client. Other useful collections are the WHITELIST an and BLACKLIST collections, as well as SCORE and SCORE_THRESHOLD.

Currently, both persistent storage keys and values can be dynamically defined, as well as the rule's var.pattern if a separate option was set to explicitly parse the rule pattern definition. See the included 99000 ruleset for an example of dynamic parsing rule patterns and persistent storage data.

##Persistent Storage

FreeWAF supports storage of long-term (inter-request) data via the lua_shared_dict interface. Under the hood this uses Nginx's shared memory zone pattern, which uses a red-black tree. This means that persistent data storage operations, including search, insertion, and deletion, run in O(log n) time, so be wary of performance degredation if the size of the memory zone grows to tens or hundreds of thousands of keys. This data will persist over the lifetime of the Nginx master process, meaning that data will survive a server reload, e.g. a HUP, but will not survive a restart.

Persistent data is set with the SETVAR action. This requires the associated rule to return a positive match. Variable data is defined via the setvar rule option:

• key: String value to define the variable key. Portions of the key value can be dynamically defined using the syntax %{VAL}, where VAL is a key in the collections table.
• value: String, boolean, or integer value. If a key already exists, the value of the key will be overwritten with the given value. Integer values can have arithmetic operations performed on them by prepending an arithmetic operator (any of +-*/).
• expire: Optional integer to determine how long, in seconds, the key will live in persistent storage.

Storage keys can be dynamically defined using dynamic parse syntax; this mimics the functionality of ModSecurity's initcol and setvar options. For example, a rule group to set a storage variable designed to track requests to a specific resource might look like this:

{   
	id = 12345,
	var = { 
		type = "URI",
		opts = nil,
		pattern = '/wp-login.php',
		operator = 'EQUALS'
	},
	opts = { nolog = true },
	action = "CHAIN",
	description = "WP-Login brute force detection"
},
{
	id = 12346,
	var = {
		type = "METHOD",
		opts = nil,
		pattern = "POST",
		operator = "EQUALS"
	},
	opts = { setvar = { key = '%{IP}.%{URI}.hitcount', value = '+1', expire = 60 }, chainchild = true, chainend = true, nolog = true },
	action = "SETVAR",
	description = "WP-Login brute force detection"
},
{
	id = 12347,
	var = {
		type = "VAR",
		opts = { value = '%{IP}.%{URI}.hitcount' },
		pattern = 5,
		operator = "GREATER"
	},
	opts = {},
	action = "DENY",
	description = "Deny more than 5 POST requests to wp-login.php in 60 seconds"
}

##Limitations

FreeWAF is undergoing continual development and improvement, and as such, may be limited in its functionality and performance. Currently known limitations can be found within the GitHub issue tracker for this repo.

##License

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/

##Bugs

Please report bugs by creating a ticket with the GitHub issue tracker.

##See Also

• The OpenResty project http://openresty.org/
• My personal blog for updates and notes on FreeWAF development http://www.cryptobells.com/