Improving The Security In Web Applications Computer Science Essay

The activities of concerns, military and authoritiess rely progressively on web engineerings and applications. The easiness of execution and usage of these engineerings has made them an indispensable constituent of on-line commercial sites, intranet and extranet applications, every bit good as the cyberspace services offered by companies. Today, new applications are about consistently developed with web engineerings. Together with that, hackers have found more elusive ways to assail web applications. Harmonizing to international statistics, SQL Injection and Cross Site Scripting are most popular exposures of web applications. The effects of this type of onslaughts are rather unsafe, such as sensitive information could be stolen or authentication systems might be by-passed.

To extenuate the state of affairs, several techniques have been adopted. In this paper, a security solution is proposed utilizing Artificial Neural Network to protect web applications against this type of onslaughts.Keywords: Artificial Neural Networks ANN, SQL Injection, Cross-Site Scripting, Datasets, Web Application Firewall.


“ Google says it suspects Microsoft is making this by utilizing Internet Explorer 8 and the Bing toolbar, both of which send user informations to Microsoft, to watch how people use Google. “ [ 25 ]“ was hacked utilizing blind SQL injection by two hackers that go by “ TinKode ” and “ Ne0h ” . MySQL ‘s parent company Sun/Oracle has besides been attacked by the same hackers. Both tabular arraies and electronic mails were dumped from their databases, but no watchwords.

“ [ 26 ]“ The squad announced that the waiter of the developer wiki has been hacked by unidentified aggressors who stole history certificates. “ [ 27 ]“ Dr Mallya ‘s web site www.mallyainparliament.

com has been hacked and the Pakistani flag has been placed with a desperate message from an organisation known as the Pakistan Cyber Army. “ [ 28 ]Above illustrations show that most power full applications are besides attacked by hackers.To supply maximal security for web applications, there are specific solutions should be implemented. One of these solutions is Web Application Firewalls ( WAF ) . Most WAFs are based on filtrating incoming user petitions against a set of predefined regulations and signatures.

The ability of form matching is chiefly achieved utilizing regular looks, such as in ModSecurity the most celebrated WAF [ 23 ] . However, with the rapid development of web applications, the figure of menaces and defined onslaughts signatures is dramatically increasing. Consequently, traditional form fiting techniques ( peculiarly regular looks ) are non effectual any longer. There is an pressing demand to follow a new form matching technique that tackles the demands of the current phase of security steps.This paper will travel through the construct of unreal nervous webs and how to use it in a signifier of a web application firewall.

To concentrate on turn outing the construct of using ANN in a web application security model. Furthermore, it has to be able to cover with the dynamic nature of web application onslaughts and signatures, including its complicated forms, such as SQL injection signatures with all possible equivocation techniques. More significantly, the clip of filtrating incoming petitions should non impact the public presentation of the web application.The Open Web Application Security Project ( OWASP ) Top 10 Web Application Security Risks for 2010 are [ 19 ] :A1: InjectionA2: Cross-Site Scripting ( XSS )A3: Broken Authentication and Session ManagementA4: Insecure Direct Object ReferencesA5: Cross-Site Request Forgery ( CSRF )A6: Security MisconfigurationA7: Insecure Cryptographic StorageA8: Failure to Restrict URL AccessA9: Insufficient Transport Layer ProtectionA10: Unvalidated Redirects and ForwardsMany of above onslaughts are avoided by effectual constellation of waiter and php.

ini file.

Introduction to SQL Injection

In the OWASP Top Ten 2007 web application exposures [ 20 ] , Injection Flaw was ranked the 2nd most prevailing exposure. The precedence has jumped to figure one most critical exposure in OWASP Top Ten 2010 release [ 19 ] .

This reflects the earnestness of this type of exposures. In the Injection Flaw household, SQL Injection is peculiarly popular and can do assorted effects in compromising web applications. Basically, SQL injection onslaughts occur when web applications straight use user ‘s inputs to construct an SQL question to entree the backend database without a proper proof on the inputs [ 8 ] , [ 21 ] . To execute SQL injection, hackers can utilize different techniques.

These techniques can be classified into five chief classs as will be explained below [ 5 ] , [ 24 ] .


In this type, the aggressors inject some SQL token into the user input and do the choice clause of an SQL question to be true all the clip:Choice * from users where username=’admin ‘ or 1=1 — ‘ and password= ” ;

Union Questions

In this type, the aggressors inject a Union question into the SQL question to acquire more informations:Select bookTitle, ISBN from books where bookID = 1 UNION Select “ drudge ” , balance from histories where accNo = 3456 — ;

Piggyback Questions

In this type, the aggressors inject extra statements to put to death for choping intent:Choice * from users where username= ” ; bead tabular array histories — and password= ”

Malformed Questions

This type is based on the mistake message returned from the web waiter to happen more information about the database:Choice * from books where bookID =

convert ( int, ( choice top 1 name from sysobjects where xtype =’u ‘ ) ) ;


This type of onslaught frequently is based on different response-time of the web waiter to detect other information about the database:Choice * from users where username=’hello1 ‘ ; choose if ( user ( ) like ‘root @ % ‘ , benchmark ( 1000000, sha1 ( ‘test ‘ ) ) , ‘false ‘ ) ; — ‘ and password= ”

Alternate Encoding

This technique is used to by-pass the defending strategy that escapes particular characters ( such as quotation marks, elans, etc. ) or some keywords:Choice * from books where bookID = 1 ; White House ( char ( 0x730065006c00650063007400200040004000760065007200730069006f006e00 ) ;This runs sp_msdropretry [ foo bead tabular array logs select * from sysobjects ] , [ saloon ] .The assorted techniques of SQL injection listed above are used by hackers to accomplish different intents: short-circuiting a login system, modifying a tabular array in a database ( utilizing some SQL questions, such as insert, delete, update, etc ) , closing down SQL Server, acquiring database information from the returned mistake message or illation, or put to deathing stored processs. Furthermore, this can take to farther amendss. For case, after acquiring the login certificates of the administrator/root of a web site through updating the database or abstracting valuable information from the mistake message, the hacker can login with the decision maker privilege and execute sensitive actions. The following subdivision will cast the visible radiation on advanced techniques used by hackers to short-circuit traditional security defence systems.

Cross-site scripting ( XSS )

Cross-site scripting ( XSS ) is a type of computing machine security exposure typically found in web applications that enables malicious aggressors to shoot client-side book into web pages viewed by other users. An exploited cross-site scripting exposure can be used by aggressors to short-circuit entree controls such as the same beginning policy. Cross-site scripting carried out on web sites were approximately 80 % of all security exposures documented by Symantec as of 2007. Their impact may run from a junior-grade nuisance to a important security hazard, depending on the sensitiveness of the informations handled by the vulnerable site, and the nature of any security extenuations implemented by the site ‘s proprietor [ 7 ] , [ 10 ] .

some illustrations are here& lt ; SCRIPTSRC=http: // & gt ; & lt ; /SCRIPT & gt ;& lt ; IMG SRC=javascript: qui vive ( ‘XSS ‘ ) & gt ;Cross-Site Scripting ( XSS ) onslaughts occur when:Data enters a Web application through an un trusted beginning, most often a web petition. The information is included in dynamic content that is sent to a web user without being validated for malicious codification.Ten onslaughts can by and large be categorized into two classs: stored and reflected. There is a 3rd, much less good known type of XSS onslaught called DOM Based XSS that is discussed individually here.

Stored XSS Attacks

Stored onslaughts are those where the injected codification is for good stored on the mark waiters, such as in a database, in a message forum, visitant log, remark field, etc. The victim so retrieves the malicious book from the waiter when it requests the stored information.

Reflected XSS Attacks

Reflected onslaughts are those where the injected codification is reflected off the web waiter, such as in an mistake message, hunt consequence, or any other response that includes some or all of the input sent to the waiter as portion of the petition. Reflected onslaughts are delivered to victims via another path, such as in an e-mail message, or on some other web waiter. When a user is tricked into snaping on a malicious nexus or subjecting a specially crafted signifier, the injected codification travels to the vulnerable web waiter, which reflects the onslaught back to the user ‘s browser. The browser so executes the codification because it came from a “ sure ” waiter.

XSS Attack Consequences

The effect of an XSS onslaught is the same regardless of whether it is stored or reflected ( or DOM Based ) . The difference is in how the warhead arrives at the waiter.

Make non be fooled into believing that a “ read merely ” or “ booklet ware ” site is non vulnerable to serious reflected Ten onslaughts. Ten can do a assortment of jobs for the terminal user that scope in badness from an irritation to finish history via media. The most terrible Ten onslaughts involve revelation of the user ‘s session cooky, leting an aggressor to commandeer the user ‘s session and take over the history. Other damaging onslaughts include the revelation of terminal user files, installing of Trojan Equus caballus plans, redirect the user to some other page or site, or modify presentation of content. An XSS exposure leting an aggressor to modify a imperativeness release or intelligence point could impact a company ‘s stock monetary value or lessen consumer assurance.

XSS exposure on a pharmaceutical site could let an aggressor to modify dose information ensuing in an overdose.

Related work

As the figure of web application security exposures and incidents additions twenty-four hours by twenty-four hours, there have been some solutions to extenuate the state of affairs. These solutions are frequently in a signifier of a Web Application Scanner ( WAS ) and a Web Application Firewall ( WAF ) .

A WAS [ 9 ] is computing machine package that hunt for web applications ‘ exposures before these web applications are published on-line [ 9 ] . Since a WAS is non meant to work as a real-time filtering mechanism for incoming traffic, it does non impact the public presentation of the web applications. However, a WAS can non protect the web applications on the fly and requires alteration on the codification of the web applications which is frequently arduous and boring.

Besides, if the beginning of the web applications is non accessible when the trial is running after printing the web application, so the detected exposures might non be mitigated. To protect web applications on the fly, web application firewalls are used. A web application firewall ( WAF ) [ 11 ] is an contraption, server circuit board, or filter that applies a set of regulations to an HTTP conversation.

By and large, these regulations cover common onslaughts such as Cross-site Scripting ( XSS ) and SQL Injection. By custom-making the regulations to your application, many onslaughts can be identified and blocked. The attempt to execute this customization can be important and needs to be maintained as the application is modified.

Positive V. Negative Security theoretical accounts of WAF

The two attacks fig 1 to security most frequently mentioned in the context of application security positive and negative are diametrically opposed in all of their characteristic behaviours, but they are structured really likewise. Both positive and negative security attacks [ 9 ] , [ 12 ] operate harmonizing to an established set of regulations. Access Control Lists ( ACL ‘s ) and signatures are two execution illustrations of positive and negative security regulations, severally. Positive security moves off from “ blocked, ” terminal of the spectrum, following an “ allow merely what I know ” methodological analysis.

Every regulation added to a positive security theoretical account increases what is classified as known behaviour, and therefore allowed, and decreases what is blocked, or what is unknown. Therefore, a positive security theoretical account with nil defined should barricade everything and relax ( i.e.

, allow broader entree ) as the acceptable content contexts are defined. At the opposite terminal of the spectrum, negative security moves towards “ blocked what I know is bad, ” intending it denies entree based on what has antecedently identified as content to be blocked, running face-to-face to the known/allowed positive theoretical account. Every regulation added to the negative security policy increases the barricading behaviour, thereby diminishing what is both unknown and allowed as the policy is tightened. Therefore, a negative security policy with nil defined would allow entree to everything, and be tightened as feats are discovered. Although negative security does retain some facet of known informations, negative security cognition comes from a list of really specific depositories of fiting forms. As information is passed through a negative security policy, it is evaluated against single known “ bad ” forms. If a known form is matched, the information is rejected ; if the information fluxing through the policy is unidentifiable, it is allowed to go through. Negative security policies do non take into history how the application works, they merely notice what accesses the application and if that entree violates any negative security forms.

Web Application SecurityPositive Model Negative ModelWhat is AllowedWhat is Denied

Fig. 1 Models of WAF

Artificial Neural Networks

An Artificial Neural Network ( ANN ) is a massively parallel distributed processor consists of a set of nerve cells interconnected to each other [ 13 ] . Like a human encephalon, an ANN has the ability to larn through a preparation procedure to obtain cognition and makes that cognition available for subsequently usage.The basic constituent of an ANN is the nerve cell. Each nerve cell has three of import constituents ( fig. 2 ) : a set of synaptic connexions ( which are represented by a set of synaptic weights and prejudice ) ; a extension map ( I? ) which is a additive combination between the input elements modified by the set of synaptic weights and prejudice ; and an activation map ( I† ) which takes the end product of the extension map as its input and generates the end product of the nerve cell.

It is the set of synaptic weights and prejudice that shops the cognition acquired during the learning stage.






I† ( . )






Fig. 2 Model of A Neuron [ 5 ]

The manner nerve cells are connected to one another will specify the architecture of an ANN.

In this research, Multilayer Feed frontward Networks ( MLN ) was used. The architecture of MLNs is demonstrated in fig. 3.With the acquisition ability, ANN can be trained to execute different technology undertakings. Some of the undertakings that can be identified are: pattern acknowledgment, pattern association, map estimate, control systems, filtering, and beam forming [ 15 ] . Among these different acquisition undertakings, pattern acknowledgment is the 1 of involvement in this research. Pattern acknowledgment is a procedure in which a form or input is assigned to one of a predefined class or a category.

There are some algorithms that can be used to develop an ANN for a form acknowledgment undertaking, such as: Back Propagation, Radial-basis Function, and Support Vector Learning, etc. Among them, Back Propagation is the algorithm that is specifically devised to develop a multilayer perceptron. The algorithm has been implemented in Matlab [ 15 ] , which is a popular tool to develop ANNs. MLNs trained with Back Propagation have been used in different Fieldss such as Intrusion Detection Systems [ 16 ] and Image Processing [ 17 ] . The categorization truths in all these applications are higher than 90 % , particularly the application of MLN in an invasion sensing attack has an truth of 99.

25 % [ 18 ] .

Fig.3 A Multilayer Feedforward Network ( MLN ) [ 15 ]

The success of ANN in invasion sensing systems has motivates this papre to look into a new solution for the ambitious restrictions of Web Application Firewalls ( WAF ) . More significantly, with ANNs the reply for a scalable solution for WAFs can be found based on some of the replete characteristics of the ANNs [ 14 ] : The ability to larn and hive away the empirical cognition ; the nonlinearity of the ANN ; the ability to generalise the solutions ; the ability to accommodate when the context alterations ; the computational public presentation ; and the massively parallel construction of the ANN.

Proposed Solution

In this proposed system we design an Artificial Neural Network with the tool MatLab fig 4. Artificial Neural Network constellation should be like below table.

1Training FunctionTRAINLMAdaption Learning FunctionLEARNGDMPerformance FunctionMSENumber Of Layers5Number of Input Neurons25Entire Number of Neurons101Transportation FunctionPURELIN

Table 1 Settings of Artificial Neural Network in MatLab

Fig.4 Snapshot of Designed Artificial Neural Network in MatLab

The Artificial Neural Network is trained with figure of assailing keywords of web applications. Now this ANN system act as a web application firewall and it can filtrate the attacking keywords from informations which is entered by user in web signifiers. The below fig.

5 shows the basic theoretical account our proposed system.Securely Configured Server which can protect from many simple onslaughtsClientWaiterDatabaseSubmitUserName:Password:Datas preprocessingArtificial Neural Network ComponentDecision devising

Fig.5 Model of Proposed System

To implement this system we use XAMPP Package in Windows 7 operating system. XAMPP contain the undermentioned unfastened beginning tools as package tabular array 2.WaiterApacheVersion 2.2.

17Data BaseMySQLVersion 5.5.8Dynamic Programming LanguagePHPVersion 5.3.5

Table 2 Main tools installed in Server System

Server Side Settings to Avoid some simple onslaughts

The Server is configured in such a manner that it can avoid many of little onslaughts, many of these are PHP constellation scenes [ 3 ] .

The Settings will be shown below.register_globals set to murdersafe_mode set to murderdisplay_errors set to murderDisable these maps: system ( ) , exec ( ) , passthru ( ) , shell_exec ( ) , proc_open ( ) , popen ( ) .Disable permissions like delete, bead to user on mysql.open_basedir set for /tmp and /htdocs or /wwwexpose_php set to murderallow_url_fopen set to murderallow_url_include set to murdermagic_quotes_gpc set to on

Datas Preprocessing

Make user entered signifier informations into suited format which informations can be used as input to Artificial Neural Network. Data preprocessing basic maps are

Remove C-like remark

Remove the commented stings in the user entered informations.

Ex-husband:UNI/*anything */ON/* anything */ SE/* anything */LE/* anything*/CTToUnion SELECT

Remove twine concatenation

Remove the concatenation symbols in user entered informations.Ex-husband:EXEC ( “ IN ” + “ SERT ” + “ IN ” + “ TOaˆ¦ ” aˆ¦ ) ”ToEXEC ( “ INSERT INTO.. ” )


Divide the biting into words and convert those words into Decimal format because denary format is more suited for Artificial Neural Network input.Ex-husband: “ SELECT ” is Converted to “ 83 69 76 69 67 84 ”

Artificial Neural Network Component

The Artificial Neural Network is trained in such manner that it can filtrate the attacking keywords from user entered informations. The Artificial Neural Network is trained with dataset contain different keywords the below tabular array 4 show the some sample keywords [ 1 ] , [ 2 ] , [ 4 ] of informations set.


Table 4 Sample preparation informations set

Decision Making

If Artificial Neural Network find assailing keywords [ 6 ] so without rejecting the information, Decision doing component make alterations to informations with alternate words [ 22 ] which is safe to hive away in database. Alteration of informations is done in the undermentioned manner table 5Restricted wordAlternate wordsSelectChooseUpdateUpdatesDropDroppenArticulationCombine

Table 5 sample mapping tabular array of restricted words and Alternative Words

The below diagram fig 6 shows the basic flow of informationsUser enter Data As:“ Choice name FROM STUDENT ” ;YesNoEnd product: 0Decision MakingConvert Decimal to StringingUsing mapping table replace restricted word with Alternative word83 69 76 69 67 84 to ChooseSELECT to ChooseArtificial Neural Network CompontFor Input: 83 69 76 69 67 84Output:0Datas Preprocessing ” 34 92SELECT 83 69 76 69 67 84name aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦ .FROM aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦ .Student aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦ .

” ; aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦aˆ¦ .

Due to magic_quotes_gpc set to on informations transformed into ” SELECT name FROM STUDENT ” ;Execute the bid with user entered informations illustration storing in database etc.


6 basic flow of informations

In this attack, the experiments besides give assuring consequences on the preparation informations set with some constellation gives 100 % truth on both preparation informations and proof informations.


Artificial Neural Network can separating between normal and malicious content based on the preparation informations. This attack gives assuring consequence of both the truth and the processing clip. The Artificial Neural Network have ability to observe new bad forms even those are non trained.

The cardinal point is that the ANN can be re-trained overtime to integrate more “ cognition ” into the ANN.The solution besides has some restrictions. The quality of a trained ANN frequently depends on its architecture and the manner the ANN is trained. More significantly, the quality of the trained ANN besides depends on the quality of the preparation informations used and the characteristics that are extracted from the informations.

Future Work

The solution can be extended to include user inputs from any possible HTTP petition ( non merely in the petition line the petition organic structure ) , such as headings to hold more control over session handling. Besides, more protocols can be considered other than HTTP to generalise the solution, such as accepting input from Web Services protocols, like SOAP and Develop a engineering which can filtrate the images uploaded by user.