htcap is a web application scanner able to crawl single page application (SPA) in a recursive
manner by intercepting ajax calls and DOM changes.
Htcap is not just another vulnerability scanner since it's focused mainly on the crawling process
and uses external tools to discover vulnerabilities. It's designed to be a tool for both manual and
automated penetration test of modern web applications.
The scan process is divided in two parts, first htcap crawls the target and collects as many
requests as possible (urls, forms, ajax ecc..) and saves them to a sql-lite database. When the
crawling is done it is possible to launch several security scanners against the saved requests and
save the scan results to the same database.
When the database is populated (at least with crawing data), it's possible to explore it with
ready-available tools such as sqlite3 or DBEaver or export the results in various formats using the
built-in scripts.
Let's assume that we have to perform a penetration test against target.local, first we crawl the
site:
$ htcap/htcap.py crawl target.local target.db
Once the crawl is done, the database (target.db) will contain all the requests discovered by the
crawler. To explore/export the database we can use the built-in scripts or ready available tools.
For example, to list all discovered ajax calls we can use a single shell command:
$ echo "SELECT method,url,data FROM request WHERE type = 'xhr';" | sqlite3 target.db
Now that the site is crawled it's possible to launch several vulnerability scanners against the
requests saved to the database. A scanner is an external program that can fuzz requests to spot
security flaws.
Htcap uses a modular architecture to handle different scanners and execute them in a multi-threaded
environment. For example we can run ten parallel instances of sqlmap against saved ajax requests
with the following command:
$ htcap/htcap.py scan -r xhr -n 10 sqlmap target.db
Htcap comes with sqlmap and arachni modules built-in.
Sqlmap is used to discover SQL-Injection vulnerabilities and arachni is used to discover XSS, XXE,
Code Executions, File Inclusions ecc.
Since scanner modules extend the BaseScanner class, they can be easly created or modified (see the
section "Writing Scanner Modules" of this manual).
Htcap comes with several standalone scripts to export the crawl and scan results.
For example we can generate an interactive report containing the relevant informations about
website/webapp with the command below.
Relevant informations will include, for example, the list of all pages that trigger ajax calls or
websockets and the ones that contain vulnerabilities.
$ htcap/scripts/htmlreport.py target.db target.html
To scan a target with a single command use/modify the quickscan.sh script.
$ htcap/scripts/quickscan.sh https://target.local
- Python 2.7
- PhantomJS v2
- Sqlmap (for sqlmap scanner module)
- Arachni (for arachni scanner module)
$ git clone https://github.com/segment-srl/htcap.git htcap
$ htcap/htcap.py
Alternatively you can download the latest zip here.
PhantomJs can be downloaded here. It comes as a self-contained executable with all libraries linked
statically, so there is no need to install or compile anything else.
Htcap will search for phantomjs executable in the locations listed below and in the paths listed in
$PATH environment varailbe:
- ./
- /usr/bin/
- /usr/local/bin/
- /usr/share/bin/
To install htcap system-wide:
# mv htcap /usr/share/
# ln -s /usr/share/htcap/htcap.py /usr/local/bin/htcap
# ln -s /usr/share/htcap/scripts/htmlreport.py /usr/local/bin/htcap_report
# ln -s /usr/share/htcap/scripts/quickscan.sh /usr/local/bin/htcapquick
You can find an online demo of the html report here and a screenshot of the database view here
You can also explore the test pages here to see from what the report has been generated. They also
include a page to test ajax recursion.
In order to read the database it's possible to use the built-in scripts or any ready-available
sqlite3 client.
Generate the html report. (demo report available here)
$ htcap/scripts/htmlreport.py target.db target.html
List all pages that trigger ajax requests:
$ htcap/scripts/ajax.py target.db
Request ID: 6
Page URL: http://target.local/dashboard
Referer: http://target.local/
Ajax requests:
[BUTTON txt=Statistics].click() -> GET http://target.local/api/get_stats
List all discovered SQL-Injection vulnerabilities:
$ htcap/scripts/vulns.py target.db "type='sqli'"
C O M M A N D
python /usr/local/bin/sqlmap --batch -u http://target.local/api/[...]
D E T A I L S
Parameter: name (POST)
Type: error-based
Title: PostgreSQL AND error-based - WHERE or HAVING clause
Payload: id=1' AND 4163=CAST [...]
[...]
Search for login forms
SELECT referer, method, url, data FROM request WHERE type='form' AND (url LIKE '%login%' OR data LIKE '%password%')
Search inside the pages html
SELECT url FROM request WHERE html LIKE '%upload%' COLLATE NOCASE
Htcap features an algorithm able to crawl ajax-based pages in a recursive manner.
The algorithm works by capturing ajax calls, mapping DOM changes to them and repeat the process
recursively against the newly added elements.
When a page is loaded htcap starts by triggering all events and filling input values in the aim to
to trigger ajax calls. When an ajax call is detected, htcap waits until it is completed and the
relative callback is called; if, after that, the DOM is modified, htcap runs the same algorithm
against the added elements and repeats it until all the ajax calls have been fired.
_________________
| |
|load page content|
'--------,--------'
|
|
|
________V________
| interact with |
| new content |<-----------------------------------------+
'--------,--------' |
| |
| |
| | YES
______V______ ________________ ______l_____
/ AJAX \ YES | | / CONTENT \
{ TRIGGERED? }-------->| wait ajax |----->{ MODIFIED? }
\ ______ ______ / '----------------' \ ______ _____ /
| NO | NO
| |
| |
________V________ |
| | |
| return |<-----------------------------------------+
'-----------------'
$ htcap crawl -h
usage: htcap [options] url outfile
Options:
-h this help
-w overwrite output file
-q do not display progress informations
-v be verbose
-m MODE set crawl mode:
- passive: do not intract with the page
- active: trigger events
- aggressive: also fill input values and crawl forms (default)
-s SCOPE set crawl scope
- domain: limit crawling to current domain (default)
- directory: limit crawling to current directory (and subdirecotries)
- url: do not crawl, just analyze a single page
-D maximum crawl depth (default: 100)
-P maximum crawl depth for consecutive forms (default: 10)
-F even if in aggressive mode, do not crawl forms
-H save HTML generated by the page
-d DOMAINS comma separated list of allowed domains (ex *.target.com)
-c COOKIES cookies as json or name=value pairs separaded by semicolon
-C COOKIE_FILE path to file containing COOKIES
-r REFERER set initial referer
-x EXCLUDED comma separated list of urls to exclude (regex) - ie logout urls
-p PROXY proxy string protocol:host:port - protocol can be 'http' or 'socks5'
-n THREADS number of parallel threads (default: 10)
-A CREDENTIALS username and password used for HTTP authentication separated by a colon
-U USERAGENT set user agent
-t TIMEOUT maximum seconds spent to analyze a page (default 300)
-u USER_SCRIPT inject USER_SCRIPT into any loaded page
-S skip initial checks
-G group query_string parameters with the same name ('[]' ending excluded)
-N don't normalize URL path (keep ../../)
-R maximum number of redirects to follow (default 10)
-I ignore robots.txt
-O dont't override timeout functions (setTimeout, setInterval)
Htcap supports three scan modes: passive, active and aggressive.
When in passive mode, htcap do not interacts with the page, this means that no events are triggered
and only links are followed. In this mode htcap acts as a very basic web crawler that collects only
the links found in the page (A tags). This simulates a user that just clicks on links.
The active mode behaves like the passive mode but it also triggers all discovered events. This
simulates a user that interact with the page without filling input values.
The aggressive mode makes htcap to also fill input values and post forms. This simulates a user
that performs as many actions as possible on the page.
Crawl http://www.target.local trying to be as stealth as possible
$ htcap/htcap.py crawl -m passive www.target.local target.db
Htcap limits the crawling process to a specific scope. Available scopes are: domain, directory and
url.
When scope is set to domain, htcap will crawl the domain of the taget only, plus the
allowed_domains (-d option).
If scope is directory, htcap will crawl only the target directory and its subdirectories and if the
scope is url, htcap will not crawl anything, it just analyzes a single page.
The excluded urls (-x option) are considered out of scope, so they get saved to database but not
crawled.
Crawl all discovered subdomains of http://target.local plus http://www.target1.local starting from
http://www.target.local
$ htcap/htcap.py crawl -d '*.target.local,www.target1.local' www.target.local target.db
Crawl the directory admin and never go to the upper directory level
$ htcap/htcap.py crawl -s directory www.target.local/admin/ target.db
It's possible to exclude some urls from crawling by providing a comma separated list of regular
expression. Excluded urls are considered out of scope.
$ htcap/htcap.py crawl -x '.*logout.*,.*session.*' www.target.local/admin/ target.db
htcap is designed to limit the crawl depth to a specific threshold.
By default there are two depth limits, one for general crawling (-D) and the other for sequential
post request (-P).
Cookies can be specified both as json and as string and can be passed as commandline option or
inside a file.
The json must be set as follow, and only the 'name' and 'value' properties are mandatory.
[
{
"name":"session",
"value":"eyJpdiI6IkZXV1J",
"domain":"target.local",
"secure":false,
"path":"/",
"expires":1453990424,
"httponly":true
},
{
"name":"version",
"value":"1.1",
"domain":"target.local",
"secure":false,
"path":"/",
"expires":1453990381,
"httponly":true
}
]
The string format is the classic list of name=value pairs separated by a semicolon:
session=eyJpdiI6IkZXV1J; version=1.1
A quick note about encoding: if cookies are passed as string their value gets url-decoded. This
means that to put, for example, a semicolon into the cookie value it must be urlencoded.
$ htcap/htcap.py crawl -c 'session=someBetter%3BToken; version=1' www.target.local/admin/ target.db'
$ htcap/htcap.py crawl -c '[{name:"session",value:"someGood;Token"}]' www.target.local/admin/ target.db'
$ htcap/htcap.py crawl -C cookies.json www.target.local/admin/ target.db'
Htcap's database is composed by the tables listed below
CRAWL_INFO Crawl informations
REQUEST Contains the requests discovered by the crawler
REQUEST_CHILD Relations between requests
ASSESSMENT Each scanner run generates a new assessment
VULNERABILITY Vulnerabilities discovered by an assessment
The CRAWL_INFO table contains the informations about the crawl and, since each crawl has its own
database, it contains one row only. It's composed by the following fields:
HTCAP_VERSION Version of htcap
TARGET Target URL
START_DATE Crawl start date
END_DATE Crawl end date
COMMANDLINE Crawler commandline options
USER_AGENT User-agent set by the crawler
The REQUEST table contains all the requests discovered by the crawler.
It's composed by the following fields:
ID Id of the request
ID_PARENT Id of the parent request
TYPE The type of the request
METHOD Request method
URL Request URL
REFERER Referer URL
REDIRECTS Number of redirects ahead of this page
DATA POST data
COOKIES Cookies as json
HTTP_AUTH Username:password used for basic http authentication
OUT_OF_SCOPE Equal to 1 if the URL is out of crawler scope
TRIGGER The html element and event that triggered the request
CRAWLED Equal to 1 if the request has been crawled
CRAWLER_ERRORS Array of crawler errors as json
HTML HTML generated by the page
The parent request is the request from wich the main request has been generated. For example, each
request inherits the cookies from the parent.
Consider that the crawler follows just one path, this means that if page A is linked from page B
and page C, the crawler will load page A as if the navigation comes from page B but not from page
C. To save all the connections between pages, the crawler uses a separate table.
This table, called REQUEST_CHILD, contains the following fields:
ID_REQUEST Id of the parent request
ID_CHILD Id of the child request
By combining these two tables it's possible to rebuild the whole site structure.
Each scaner run generates a new record in this table to save the scanning informations, so
basically an assessment is considered the execution of a scanner module.
It contains the following fields:
ID Id of the assessment
SCANNER The name of the scanner
START_DATE Scan start date
END_DATE Scan end date
The VULNERABILITY table contains all the vulnerabilities discovered by the various assessments. It
contains the following fields:
ID Id of the vulnerability
ID_ASSESSMENT Id of the assessment to which it belongs to
ID_REQUEST Id of the request that has been scanned
TYPE vulnerability type (see vulntypes)
DESCRIPTION Description of the vulnerability
Each scanner module is a python class that extends the BaseScanner class and overrides the
following methods:
- init
- get_settings
- get_cmd
- scanner_executed
Basically the execution process of a scanner is as follow:
- Scanner is initalized by the parent class by calling the get_settings and init methods
- The parent class calls the get_cmd method to get the command to execute for the given request
- Once the command returns, the parent class passes the the output to the scanner module by calling
the scanner_executed method. - The scanner_executed method parses the command output and saves the result to the database by
calling the save_vulnerability method.
[...]
from core.scan.base_scanner import BaseScanner
class Curl(BaseScanner):
def init(self, argv):
pass
def get_settings(self):
return dict(
request_types = "link,redirect",
num_threads = 10,
process_timeout = 20,
scanner_exe = "/usr/bin/env curl"
)
def get_cmd(self, request, tmp_dir):
cmd = ["-I", request.url]
return cmd
def scanner_executed(self, request, out, err, tmp_dir, cmd):
if not re.search("^X-XSS-Protection\:", out, re.M):
type = "xss-portection-missing"
descr = "X-XSS-Protection header is not set"
self.save_vulnerability(request, type, descr)
To add a scanner module called, for example, myscanner follow these steps:
- Create a file named myscanner.py inside core/scan/scanners/
- Inside that file create Myscanner class that overrides BaseScanner
- Override methods and adjust settings
To execute myscanner run the following command:
$ htcap/htcap.py scan myscanner target.db
One of the biggest problem when scanning a webapplication is how to determine the uniqueness of a
page. Modern web applications usually use the same page to generate different contents based to the
url parameters or patterns. A vulnerability scanner don't need to analyze the same page just
because its content has changed.
In the aim to solve this problem, or at least reduce its impact, htcap implements an algorithm for
url comparision that can let scanner modules skip "duplicated" urls.
Inside the get_cmd method it's possbile to implement this feature using the code below.
if self.is_request_duplicated(request):
return False
The algorithm is extreamely simple, it just removes the values from parameters and it sorts
them alphabetically; for example http://www.test.local/a/?c=1&a=2&b=3 becames
http://www.test.local/a/?a=&b=&c= .
A good idea would be the use of the SimHash algorithm but lots of tests are needed.
In case of POST requests the same algorithm is also applied to the following payloads:
- URL-Encoded
- XML
- JSON
[...]
from core.scan.base_scanner import BaseScanner
class CustomScanner(BaseScanner):
def init(self, argv):
"""
custom initializer
the first argument is an array of command line arguments (if any) passed to the scanner
"""
if argv[0] == "-h":
print "usage: ...."
self.exit(0)
def get_settings(self):
"""
scanner settings
"""
return dict(
request_types = "xhr,link,redirect,form,jsonp", # request types to analyze
num_threads = 10, # number of parallel commands to execute
process_timeout = 180, # command execution timeout
scanner_exe = "/usr/bin/customscanner"
)
def get_cmd(self, request, tmp_dir):
"""
this method is called by the parent class to get the command to execute
the first argument is the Request object containing the url, the cookies ecc
the second argument is the path to the temporary dir used to store output files ecc
"""
if self.is_request_duplicated(request):
return False
out_file = tmp_dir + "/output"
cmd = [
"--url", request.url,
"--out", out_file
]
# return False to skip current request
#return False
return cmd
def scanner_executed(self, request, out, err, tmp_dir, cmd):
"""
this method is called when the execution of a command is completed
the first argument is the Request object used to generate the command
the second and the third are the output and the error returned by the command
the forth argumnt is the path to the temporary dir used by the command
the fifth argument is the command executed
"""
out_file = tmp_dir + "/output"
with open(out_file,'r') as file:
output = file.read()
# parse output
......
for vulnerability in report:
# request, type, description
self.save_vulnerability(request, "sqli", vulnerability)
Scanner modules analyze only in-scope requests. If the crawler scope is set to "url", any
discovered request will be considered out of scope, including ajax requests, jsonp ecc.
For example if target.local/foo/bar.php has been crawled with scope set to "url" and it contains
ajax request to target.local/foo/ajax.php, they won't be scanned. With a simple query it's possible
to make those request visible to scanners.
UPDATE request set out_of_scope=0 where type in ('xhr','websocket','jsonp')
Htcap allows the user to script the page analysis by using the so called User Script (-u option).
When htcap analyzes a page it triggers some hooks to let the user customize the analysis behaviour.
For example an hook is called before any XHR request and the user can deceide if it must be performed or cancelled.
A reference to the UI object is passed to all hooks, the purposes of that object are:
- Let the user store variables inside the ui.vars object
- Let the user access the request id via ui.id
- Let the user call some methods to interact with the page. Actually the available methods are:
- ui.render(path_to_file) - save a screenshot of the page current state
- ui.triggerEvent(element, event) - trigger an event
Available hooks are:
- onInit - called when the page is initialized
- onBeforeStart - called before the analysis starts
- onTriggerEvent - called before triggering an avent
- onXhr - called before XHR requests
- onAllXhrsCompleted - called when all XHRs are completed
- onDomModified - called when the DOM is modified
- onEnd - called before exit
{
onInit: function(ui){
// override natove methods
window.prompt = function(){ return "AAA" };
// init local variables
ui.vars.cnt = 0;
},
onBeforeStart: function(ui){},
onBeforeTriggerEvent: function(ui, element, event){
// cancel trigger if element has calss kill-all
if(element.matches(".kill-all")) return false;
},
onTriggerEvent: function(ui, element, event){},
onXhr: function(ui, request){
// cancel XHR request if url matches XXX
if(request.url.match(/XXX/))
return false
},
onAllXhrsCompleted: function(ui){},
onDomModified: function(ui, rootElements, allElements){
// save a screenshot on every DOM change
ui.render(ui.id + "-screen-" + ui.vars.cnt + ".png");
ui.vars.cnt++;
},
onEnd: function(ui){}
}