Skip to content

Latest commit

 

History

History
531 lines (383 loc) · 24.3 KB

DESIGN.md

File metadata and controls

531 lines (383 loc) · 24.3 KB
Raw

Nuclei Architecture Document

A brief overview of Nuclei Engine architecture. This document will be kept updated as the engine progresses.

pkg/templates

Template

Template is the basic unit of input to the engine which describes the requests to be made, matching to be done, data to extract, etc.

The template structure is described here. Template level attributes are defined here as well as convenience methods to validate, parse and compile templates creating executers.

Any attributes etc. required for the template, engine or requests to function are also set here.

Workflows are also compiled, their templates are loaded and compiled as well. Any validations etc. on the paths provided are also done here.

Parse function is the main entry point which returns a template for a filePath and executorOptions. It compiles all the requests for the templates, all the workflows, as well as any self-contained request etc. It also caches the templates in an in-memory cache.

Preprocessors

Preprocessors are also applied here which can do things at template level. They get data of the template which they can alter at will on runtime. This is used in the engine to do random string generation.

Custom processor can be used if they satisfy the following interface.

type Preprocessor interface {
	Process(data []byte) []byte
}

pkg/model

Model package implements Information structure for Nuclei Templates. Info contains all major metadata information for the template. Classification structure can also be used to provide additional context to vulnerability data.

It also specifies a WorkflowLoader interface that is used during workflow loading in template compilation stage.

type WorkflowLoader interface {
	GetTemplatePathsByTags(tags []string) []string
	GetTemplatePaths(templatesList []string, noValidate bool) []string
}

pkg/protocols

Protocols package implements all the request protocols supported by Nuclei. This includes http, dns, network, headless and file requests as of now.

Request

It exposes a Request interface that is implemented by all the request protocols supported.

// Request is an interface implemented any protocol based request generator.
type Request interface {
	Compile(options *ExecuterOptions) error
	Requests() int
	GetID() string
	Match(data map[string]interface{}, matcher *matchers.Matcher) (bool, []string)
	Extract(data map[string]interface{}, matcher *extractors.Extractor) map[string]struct{}
	ExecuteWithResults(input string, dynamicValues, previous output.InternalEvent, callback OutputEventCallback) error
	MakeResultEventItem(wrapped *output.InternalWrappedEvent) *output.ResultEvent
	MakeResultEvent(wrapped *output.InternalWrappedEvent) []*output.ResultEvent
	GetCompiledOperators() []*operators.Operators
}

Many of these methods are similar across protocols while some are very protocol specific.

A brief overview of the methods is provided below -

  • Compile - Compiles the request with provided options.
  • Requests - Returns total requests made.
  • GetID - Returns any ID for the request
  • Match - Used to perform matching for patterns using matchers
  • Extract - Used to perform extraction for patterns using extractors
  • ExecuteWithResults - Request execution function for input.
  • MakeResultEventItem - Creates a single result event for the intermediate InternalWrappedEvent output structure.
  • MakeResultEvent - Returns a slice of results based on an InternalWrappedEvent internal output event.
  • GetCompiledOperators - Returns the compiled operators for the request.

MakeDefaultResultEvent function can be used as a default for MakeResultEvent function when no protocol-specific features need to be implemented for result generation.

For reference protocol requests implementations, one can look at the below packages -

  1. pkg/protocols/http
  2. pkg/protocols/dns
  3. pkg/protocols/network

Executer

All these different requests interfaces are converted to an Executer which is also an interface defined in pkg/protocols which is used during final execution of the template.

// Executer is an interface implemented any protocol based request executer.
type Executer interface {
	Compile() error
	Requests() int
	Execute(input string) (bool, error)
	ExecuteWithResults(input string, callback OutputEventCallback) error
}

The ExecuteWithResults function accepts a callback, which gets provided with results during execution in form of *output.InternalWrappedEvent structure.

The default executer is provided in pkg/protocols/common/executer . It takes a list of Requests and relevant ExecuterOptions and implements the Executer interface required for template execution. The executer during Template compilation process is created from this package and used as-is.

A different executer is the Clustered Requests executer which implements the Nuclei Request clustering functionality in pkg/templates We have a single HTTP request in cases where multiple templates can be clustered and multiple operator lists to match/extract. The first HTTP request is executed while all the template matcher/extractor are evaluated separately.

For Workflow execution, a separate RunWorkflow function is used which executes the workflow independently of the template execution.

With this basic premise set, we can now start exploring the current runner implementation which will also walk us through the architecture of nuclei.

internal/runner

Template loading

The first process after all CLI specific initialisation is the loading of template/workflow paths that the user wants to run. This is done by the packages described below.

pkg/catalog

This package is used to get paths using mixed syntax. It takes a template directory and performs resolving for template paths both from provided template and current user directory.

The syntax is very versatile and can include filenames, glob patterns, directories, absolute paths, and relative-paths.

Next step is the initialisation of the reporting modules which is handled in pkg/reporting.

pkg/reporting

Reporting module contains exporters and trackers as well as a module for deduplication and a module for result formatting.

Exporters and Trackers are interfaces defined in pkg/reporting.

// Tracker is an interface implemented by an issue tracker
type Tracker interface {
	CreateIssue(event *output.ResultEvent) error
}

// Exporter is an interface implemented by an issue exporter
type Exporter interface {
	Close() error
	Export(event *output.ResultEvent) error
}

Exporters include Elasticsearch, markdown, sarif . Trackers include GitHub , GitLab and Jira.

Each exporter and trackers implement their own configuration in YAML format and are very modular in nature, so adding new ones is easy.

After reading all the inputs from various sources and initialisation other miscellaneous options, the next bit is the output writing which is done using pkg/output module.

pkg/output

Output package implements the output writing functionality for Nuclei.

Output Writer implements the Writer interface which is called each time a result is found for nuclei.

// Writer is an interface which writes output to somewhere for nuclei events.
type Writer interface {
	Close()
	Colorizer() aurora.Aurora
	Write(*ResultEvent) error
	Request(templateID, url, requestType string, err error)
}

ResultEvent structure is passed to the Nuclei Output Writer which contains the entire detail of a found result. Various intermediary types like InternalWrappedEvent and InternalEvent are used throughout nuclei protocols and matchers to describe results in various stages of execution.

Interactsh is also initialised if it is not explicitly disabled.

pkg/protocols/common/interactsh

Interactsh module is used to provide automatic Out-of-Band vulnerability identification in Nuclei.

It uses two LRU caches, one for storing interactions for request URLs and one for storing requests for interaction URL. These both caches are used to correlated requests received to the Interactsh OOB server and Nuclei Instance. Interactsh Client package does most of the heavy lifting of this module.

Polling for interactions and server registration only starts when a template uses the interactsh module and is executed by nuclei. After that no registration is required for the entire run.

RunEnumeration

Next we arrive in the RunEnumeration function of the runner.

HostErrorsCache is initialised which is used throughout the run of Nuclei enumeration to keep track of errors per host and skip further requests if the errors are greater than the provided threshold. The functionality for the error tracking cache is defined in hosterrorscache.go and is pretty simplistic in nature.

Next the WorkflowLoader is initialised which used to load workflows. It exists in pkg/parsers/workflow_loader.go

The loader is initialised moving forward which is responsible for Using Catalog, Passed Tags, Filters, Paths, etc. to return compiled Templates and Workflows.

pkg/catalog/loader

First the input passed by the user as paths is normalised to absolute paths which is done by the pkg/catalog module. Next the path filter module is used to remove the excluded template/workflows paths.

pkg/parsers module's LoadTemplate,LoadWorkflow functions are used to check if the templates pass the validation + are not excluded via tags/severity/etc. filters. If all checks are passed, then the template/workflow is parsed and returned in a compiled form by the pkg/templates's Parse function.

Parse function performs compilation of all the requests in a template + creates Executers from them returning a runnable Template/Workflow structure.

Clustering module comes in next whose job is to cluster identical HTTP GET requests together (as a lot of the templates perform the same get requests many times, it's a good way to save many requests on large scans with lots of templates).

pkg/operators

Operators package implements all the matching and extracting logic of Nuclei.

// Operators contain the operators that can be applied on protocols
type Operators struct {
	Matchers []*matchers.Matcher
	Extractors []*extractors.Extractor
	MatchersCondition string
}

A protocol only needs to embed the operators.Operators type shown above, and it can utilise all the matching/extracting functionality of nuclei.

// MatchFunc performs matching operation for a matcher on model and returns true or false.
type MatchFunc func(data map[string]interface{}, matcher *matchers.Matcher) (bool, []string)

// ExtractFunc performs extracting operation for an extractor on model and returns true or false.
type ExtractFunc func(data map[string]interface{}, matcher *extractors.Extractor) map[string]struct{}

// Execute executes the operators on data and returns a result structure
func (operators *Operators) Execute(data map[string]interface{}, match MatchFunc, extract ExtractFunc, isDebug bool) (*Result, bool)

The core of this process is the Execute function which takes an input dictionary as well as a Match and Extract function and return a Result structure which is used later during nuclei execution to check for results.

// Result is a result structure created from operators running on data.
type Result struct {
	Matched bool
	Extracted bool
	Matches map[string][]string
	Extracts map[string][]string
	OutputExtracts []string
	DynamicValues map[string]interface{}
	PayloadValues map[string]interface{}
}

The internal logics for matching and extracting for things like words, regexes, jq, paths, etc. is specified in pkg/operators/matchers, pkg/operators/extractors. Those packages should be investigated for further look into the topic.

Template Execution

pkg/core provides the engine mechanism which runs the templates/workflows on inputs. It exposes an Execute function which does the task of execution while also doing template clustering. The clustering can also be disabled optionally by the user.

An example of using the core engine is provided below.

engine := core.New(r.options)
engine.SetExecuterOptions(executerOpts)
results := engine.ExecuteWithOpts(finalTemplates, r.hmapInputProvider, true)

Adding a New Protocol

Protocols form the core of Nuclei Engine. All the request types like http, dns, etc. are implemented in form of protocol requests.

A protocol must implement the Protocol and Request interfaces described above in pkg/protocols. We'll take the example of an existing protocol implementation - websocket for this short reference around Nuclei internals.

The code for the websocket protocol is contained in pkg/protocols/others/websocket.

Below a high level skeleton of the websocket implementation is provided with all the important parts present.

package websocket

// Request is a request for the Websocket protocol
type Request struct {
	// Operators for the current request go here.
	operators.Operators `yaml:",inline,omitempty"`
	CompiledOperators   *operators.Operators `yaml:"-"`

	// description: |
	//   Address contains address for the request
	Address string `yaml:"address,omitempty" jsonschema:"title=address for the websocket request,description=Address contains address for the request"`

    // declarations here
}

// Compile compiles the request generators preparing any requests possible.
func (r *Request) Compile(options *protocols.ExecuterOptions) error {
	r.options = options

    // request compilation here as well as client creation
 
	if len(r.Matchers) > 0 || len(r.Extractors) > 0 {
		compiled := &r.Operators
		if err := compiled.Compile(); err != nil {
			return errors.Wrap(err, "could not compile operators")
		}
		r.CompiledOperators = compiled
	}
	return nil
}

// Requests returns the total number of requests the rule will perform
func (r *Request) Requests() int {
	if r.generator != nil {
		return r.generator.NewIterator().Total()
	}
	return 1
}

// GetID returns the ID for the request if any.
func (r *Request) GetID() string {
	return ""
}

// ExecuteWithResults executes the protocol requests and returns results instead of writing them.
func (r *Request) ExecuteWithResults(input string, dynamicValues, previous output.InternalEvent, callback protocols.OutputEventCallback) error {
    // payloads init here
	if err := r.executeRequestWithPayloads(input, hostname, value, previous, callback); err != nil {
		return err
	}
	return nil
}

// ExecuteWithResults executes the protocol requests and returns results instead of writing them.
func (r *Request) executeRequestWithPayloads(input, hostname string, dynamicValues, previous output.InternalEvent, callback protocols.OutputEventCallback) error {
	header := http.Header{}

    // make the actual request here after setting all options

	event := eventcreator.CreateEventWithAdditionalOptions(r, data, r.options.Options.Debug || r.options.Options.DebugResponse, func(internalWrappedEvent *output.InternalWrappedEvent) {
		internalWrappedEvent.OperatorsResult.PayloadValues = payloadValues
	})
	if r.options.Options.Debug || r.options.Options.DebugResponse {
		responseOutput := responseBuilder.String()
		gologger.Debug().Msgf("[%s] Dumped Websocket response for %s", r.options.TemplateID, input)
		gologger.Print().Msgf("%s", responsehighlighter.Highlight(event.OperatorsResult, responseOutput, r.options.Options.NoColor))
	}

	callback(event)
	return nil
}

func (r *Request) MakeResultEventItem(wrapped *output.InternalWrappedEvent) *output.ResultEvent {
	data := &output.ResultEvent{
		TemplateID:       types.ToString(r.options.TemplateID),
		TemplatePath:     types.ToString(r.options.TemplatePath),
		// ... setting more values for result event
	}
	return data
}

// Match performs matching operation for a matcher on model and returns:
// true and a list of matched snippets if the matcher type is supports it
// otherwise false and an empty string slice
func (r *Request) Match(data map[string]interface{}, matcher *matchers.Matcher) (bool, []string) {
	return protocols.MakeDefaultMatchFunc(data, matcher)
}

// Extract performs extracting operation for an extractor on model and returns true or false.
func (r *Request) Extract(data map[string]interface{}, matcher *extractors.Extractor) map[string]struct{} {
	return protocols.MakeDefaultExtractFunc(data, matcher)
}

// MakeResultEvent creates a result event from internal wrapped event
func (r *Request) MakeResultEvent(wrapped *output.InternalWrappedEvent) []*output.ResultEvent {
	return protocols.MakeDefaultResultEvent(r, wrapped)
}

// GetCompiledOperators returns a list of the compiled operators
func (r *Request) GetCompiledOperators() []*operators.Operators {
	return []*operators.Operators{r.CompiledOperators}
}

// Type returns the type of the protocol request
func (r *Request) Type() templateTypes.ProtocolType {
	return templateTypes.WebsocketProtocol
}

Almost all of these protocols have boilerplate functions for which default implementations have been provided in the providers package. Examples are the implementation of Match, Extract, MakeResultEvent, GetCompiledOperators`, etc. which are almost same throughout Nuclei protocols code. It is enough to copy-paste them unless customization is required.

eventcreator package offers CreateEventWithAdditionalOptions function which can be used to create result events after doing request execution.

Step by step description of how to add a new protocol to Nuclei -

  1. Add the protocol implementation in pkg/protocols directory. If it's a small protocol with fewer options, considering adding it to the pkg/protocols/others directory. Add the enum for the new protocol to pkg/templates/types/types.go.

  2. Add the protocol request structure to the Template structure fields. This is done in pkg/templates/templates.go with the corresponding import line.

import (
	...
	"github.com/projectdiscovery/nuclei/v3/pkg/protocols/others/websocket"
)

// Template is a YAML input file which defines all the requests and
// other metadata for a template.
type Template struct {
	...
	// description: |
	//   Websocket contains the Websocket request to make in the template.
	RequestsWebsocket []*websocket.Request `yaml:"websocket,omitempty" json:"websocket,omitempty" jsonschema:"title=websocket requests to make,description=Websocket requests to make for the template"`
	...
}

Also add the protocol case to the Type function as well as the TemplateTypes array in the same templates.go file.

// TemplateTypes is a list of accepted template types
var TemplateTypes = []string{
	...
	"websocket",
}

// Type returns the type of the template
func (t *Template) Type() templateTypes.ProtocolType {
	...
	case len(t.RequestsWebsocket) > 0:
		return templateTypes.WebsocketProtocol
	default:
		return ""
	}
}
  1. Add the protocol request to the Requests function and compileProtocolRequests function in the compile.go file in same directory.
// Requests return the total request count for the template
func (template *Template) Requests() int {
	return len(template.RequestsDNS) +
		...
		len(template.RequestsSSL) +
		len(template.RequestsWebsocket)
}


// compileProtocolRequests compiles all the protocol requests for the template
func (template *Template) compileProtocolRequests(options protocols.ExecuterOptions) error {
	...

	case len(template.RequestsWebsocket) > 0:
		requests = template.convertRequestToProtocolsRequest(template.RequestsWebsocket)
	}
	template.Executer = executer.NewExecuter(requests, &options)
	return nil
}

That's it, you've added a new protocol to Nuclei. The next good step would be to write integration tests which are described in integration-tests and cmd/integration-tests directories.

Profiling Instructions

To enable dumping of Memory profiling data, -profile-mem flag can be used along with path to a file. This writes a pprof formatted file which can be used for investigate resource usage with pprof tool.

$ nuclei -t nuclei-templates/ -u https://example.com -profile-mem mem.pprof

To view profile data in pprof, first install pprof. Then run the below command -

$ go tool pprof mem.pprof

To open a web UI on a port to visualize debug data, the below command can be used.

$ go tool pprof -http=:8081 mem.pprof

Project Structure

Notes

  1. The matching as well as interim output functionality is a bit complex, we should simplify it a bit as well.