cvs: phpdoc /en/security apache.xml cgi-bin.xml current.xml database.xml errors.xml filesystem.xml general.xml globals.xml hiding.xml index.xml
intro.xml variables.xml

betz Mon Jan 26 08:22:26 2004 EDT Added files: /phpdoc/en/security apache.xml cgi-bin.xml current.xml database.xml errors.xml filesystem.xml general.xml globals.xml hiding.xml intro.xml variables.xml Modified files: /phpdoc/en/security index.xml Log: split of monolithic index.xml

http://cvs.php.net/diff.php/phpdoc/en/security/index.xml?r1=1.66&r2=1.67&ty=u Index: phpdoc/en/security/index.xml diff -u phpdoc/en/security/index.xml:1.66 phpdoc/en/security/index.xml:1.67 --- phpdoc/en/security/index.xml:1.66 Tue Jan 6 04:22:21 2004 +++ phpdoc/en/security/index.xml Mon Jan 26 08:22:25 2004 @@ -1,5 +1,6 @@ <?xml version="1.0" encoding="iso-8859-1"?> -<!-- $Revision: 1.66 $ --> +<!-- $Revision: 1.67 $ --> +<!-- Last Revision before split: 1.66 --> <chapter id="security.index"> <title>Security</title> @@ -38,1270 +39,16 @@ coding for different levels of security. </simpara> - <sect1 id="security.general"> - <title>General considerations</title> - <simpara> - A completely secure system is a virtual impossibility, so an - approach often used in the security profession is one of balancing - risk and usability. If every variable submitted by a user required - two forms of biometric validation (such as a retinal scan and a - fingerprint), you would have an extremely high level of - accountability. It would also take half an hour to fill out a fairly - complex form, which would tend to encourage users to find ways of - bypassing the security. - </simpara> - <simpara> - The best security is often inobtrusive enough to suit the - requirements without the user being prevented from accomplishing - their work, or over-burdening the code author with excessive - complexity. Indeed, some security attacks are merely exploits of - this kind of overly built security, which tends to erode over time. - </simpara> - <simpara> - A phrase worth remembering: A system is only as good as the weakest - link in a chain. If all transactions are heavily logged based on - time, location, transaction type, etc. but the user is only - verified based on a single cookie, the validity of tying the users - to the transaction log is severely weakened. - </simpara> - <simpara> - When testing, keep in mind that you will not be able to test all - possibilities for even the simplest of pages. The input you - may expect will be completely unrelated to the input given by - a disgruntled employee, a cracker with months of time on their - hands, or a housecat walking across the keyboard. This is why it's - best to look at the code from a logical perspective, to discern - where unexpected data can be introduced, and then follow how it is - modified, reduced, or amplified. - </simpara> - <simpara> - The Internet is filled with people trying to make a name for - themselves by breaking your code, crashing your site, posting - inappropriate content, and otherwise making your day interesting. - It doesn't matter if you have a small or large site, you are - a target by simply being online, by having a server that can be - connected to. Many cracking programs do not discern by size, they - simply trawl massive IP blocks looking for victims. Try not to - become one. - </simpara> - </sect1> - - <sect1 id="security.cgi-bin"> - <title>Installed as CGI binary</title> - - <sect2 id="security.cgi-bin.attacks"> - <title>Possible attacks</title> - <simpara> - Using PHP as a <acronym>CGI</acronym> binary is an option for - setups that for some reason do not wish to integrate PHP as a - module into server software (like Apache), or will use PHP with - different kinds of CGI wrappers to create safe chroot and setuid - environments for scripts. This setup usually involves installing - executable PHP binary to the web server cgi-bin directory. CERT - advisory <ulink url="&url.cert;">CA-96.11</ulink> recommends - against placing any interpreters into cgi-bin. Even if the PHP - binary can be used as a standalone interpreter, PHP is designed - to prevent the attacks this setup makes possible: - </simpara> - <itemizedlist> - <listitem> - <simpara> - Accessing system files: <filename - role="url">http://my.host/cgi-bin/php?/etc/passwd</filename> - </simpara> - <simpara> - The query information in a url after the question mark (?) is - passed as command line arguments to the interpreter by the CGI - interface. Usually interpreters open and execute the file - specified as the first argument on the command line. - </simpara> - <simpara> - When invoked as a CGI binary, PHP refuses to interpret the - command line arguments. - </simpara> - </listitem> - <listitem> - <simpara> - Accessing any web document on server: <filename - role="url">http://my.host/cgi-bin/php/secret/doc.html</filename> - </simpara> - <simpara> - The path information part of the url after the PHP binary name, - <filename role="uri">/secret/doc.html</filename> is - conventionally used to specify the name of the file to be - opened and interpreted by the <acronym>CGI</acronym> program. - Usually some web server configuration directives (Apache: - Action) are used to redirect requests to documents like - <filename - role="url">http://my.host/secret/script.php</filename> to the - PHP interpreter. With this setup, the web server first checks - the access permissions to the directory <filename - role="uri">/secret</filename>, and after that creates the - redirected request <filename - role="url">http://my.host/cgi-bin/php/secret/script.php</filename>. - Unfortunately, if the request is originally given in this form, - no access checks are made by web server for file <filename - role="uri">/secret/script.php</filename>, but only for the - <filename role="uri">/cgi-bin/php</filename> file. This way - any user able to access <filename - role="uri">/cgi-bin/php</filename> is able to access any - protected document on the web server. - </simpara> - <simpara> - In PHP, compile-time configuration option <link - linkend="install.configure.enable-force-cgi-redirect">--enable-force-cgi-redirect</link> - and runtime configuration directives <link - linkend="ini.doc-root">doc_root</link> and <link - linkend="ini.user-dir">user_dir</link> can be used to prevent - this attack, if the server document tree has any directories - with access restrictions. See below for full the explanation - of the different combinations. - </simpara> - </listitem> - </itemizedlist> - </sect2> - - <sect2 id="security.cgi-bin.default"> - <title>Case 1: only public files served</title> - - <simpara> - If your server does not have any content that is not restricted - by password or ip based access control, there is no need for - these configuration options. If your web server does not allow - you to do redirects, or the server does not have a way to - communicate to the PHP binary that the request is a safely - redirected request, you can specify the option <link - linkend="install.configure.enable-force-cgi-redirect">--enable-force-cgi-redirect</link> - to the configure script. You still have to make sure your PHP - scripts do not rely on one or another way of calling the script, - neither by directly <filename - role="php">http://my.host/cgi-bin/php/dir/script.php</filename> - nor by redirection <filename - role="php">http://my.host/dir/script.php</filename>. - </simpara> - <simpara> - Redirection can be configured in Apache by using AddHandler and - Action directives (see below). - </simpara> - </sect2> - - <sect2 id="security.cgi-bin.force-redirect"> - <title>Case 2: using --enable-force-cgi-redirect</title> - <simpara> - This compile-time option prevents anyone from calling PHP - directly with a url like <filename - role="php">http://my.host/cgi-bin/php/secretdir/script.php</filename>. - Instead, PHP will only parse in this mode if it has gone through - a web server redirect rule. - </simpara> - <simpara> - Usually the redirection in the Apache configuration is done with - the following directives: - </simpara> - <programlisting role="apache-conf"> -<![CDATA[ -Action php-script /cgi-bin/php -AddHandler php-script .php -]]> - </programlisting> - <simpara> - This option has only been tested with the Apache web server, and - relies on Apache to set the non-standard CGI environment variable - <envar>REDIRECT_STATUS</envar> on redirected requests. If your - web server does not support any way of telling if the request is - direct or redirected, you cannot use this option and you must use - one of the other ways of running the CGI version documented - here. - </simpara> - </sect2> - - <sect2 id="security.cgi-bin.doc-root"> - <title>Case 3: setting doc_root or user_dir</title> - <simpara> - To include active content, like scripts and executables, in the - web server document directories is sometimes considered an insecure - practice. If, because of some configuration mistake, the scripts - are not executed but displayed as regular HTML documents, this - may result in leakage of intellectual property or security - information like passwords. Therefore many sysadmins will prefer - setting up another directory structure for scripts that are - accessible only through the PHP CGI, and therefore always - interpreted and not displayed as such. - </simpara> - <simpara> - Also if the method for making sure the requests are not - redirected, as described in the previous section, is not - available, it is necessary to set up a script doc_root that is - different from web document root. - </simpara> - <simpara> - You can set the PHP script document root by the configuration - directive <link linkend="ini.doc-root">doc_root</link> in the - <link linkend="configuration.file">configuration file</link>, or - you can set the environment variable - <envar>PHP_DOCUMENT_ROOT</envar>. If it is set, the CGI version - of PHP will always construct the file name to open with this - <parameter>doc_root</parameter> and the path information in the - request, so you can be sure no script is executed outside this - directory (except for <parameter>user_dir</parameter> - below). - </simpara> - <simpara> - Another option usable here is <link - linkend="ini.user-dir">user_dir</link>. When user_dir is unset, - only thing controlling the opened file name is - <parameter>doc_root</parameter>. Opening an url like <filename - role="url">http://my.host/~user/doc.php</filename> does not - result in opening a file under users home directory, but a file - called <filename role="uri">~user/doc.php</filename> under - doc_root (yes, a directory name starting with a tilde - [<literal>~</literal>]). - </simpara> - <simpara> - If user_dir is set to for example <filename - role="dir">public_php</filename>, a request like <filename - role="url">http://my.host/~user/doc.php</filename> will open a - file called <filename>doc.php</filename> under the directory - named <filename role="dir">public_php</filename> under the home - directory of the user. If the home of the user is <filename - role="dir">/home/user</filename>, the file executed is - <filename>/home/user/public_php/doc.php</filename>. - </simpara> - <simpara> - <parameter>user_dir</parameter> expansion happens regardless of - the <parameter>doc_root</parameter> setting, so you can control - the document root and user directory access - separately. - </simpara> - </sect2> - - <sect2 id="security.cgi-bin.shell"> - <title>Case 4: PHP parser outside of web tree</title> - <para> - A very secure option is to put the PHP parser binary somewhere - outside of the web tree of files. In <filename - role="dir">/usr/local/bin</filename>, for example. The only real - downside to this option is that you will now have to put a line - similar to: - <informalexample> - <programlisting> -<![CDATA[ -#!/usr/local/bin/php -]]> - </programlisting> - </informalexample> - as the first line of any file containing PHP tags. You will also - need to make the file executable. That is, treat it exactly as - you would treat any other CGI script written in Perl or sh or any - other common scripting language which uses the - <literal>#!</literal> shell-escape mechanism for launching - itself. - </para> - <para> - To get PHP to handle <envar>PATH_INFO</envar> and - <envar>PATH_TRANSLATED</envar> information correctly with this - setup, the PHP parser should be compiled with the <link - linkend="install.configure.enable-discard-path">--enable-discard-path</link> - configure option. - </para> - </sect2> - - </sect1> - - <sect1 id="security.apache"> - <title>Installed as an Apache module</title> - <simpara> - When PHP is used as an Apache module it inherits Apache's user - permissions (typically those of the "nobody" user). This has several - impacts on security and authorization. For example, if you are using - PHP to access a database, unless that database has built-in access - control, you will have to make the database accessible to the - "nobody" user. This means a malicious script could access and modify - the database, even without a username and password. It's entirely - possible that a web spider could stumble across a database - administrator's web page, and drop all of your databases. You can - protect against this with Apache authorization, or you can design - your own access model using LDAP, &htaccess; files, etc. and include - that code as part of your PHP scripts. - </simpara> - <simpara> - Often, once security is established to the point where the PHP user - (in this case, the apache user) has very little risk attached to it, - it is discovered that PHP is now prevented from writing any files - to user directories. Or perhaps it has been prevented from accessing - or changing databases. It has equally been secured from writing - good and bad files, or entering good and bad database transactions. - </simpara> - <simpara> - A frequent security mistake made at this point is to allow apache - root permissions, or to escalate apache's abilitites in some other - way. - </simpara> - <simpara> - Escalating the Apache user's permissions to root is extremely - dangerous and may compromise the entire system, so sudo'ing, - chroot'ing, or otherwise running as root should not be considered by - those who are not security professionals. - </simpara> - <simpara> - There are some simpler solutions. By using - <link linkend="ini.open-basedir">open_basedir</link> you can control and restrict what - directories are allowed to be used for PHP. You can also set up - apache-only areas, to restrict all web based activity to non-user, - or non-system, files. - </simpara> - </sect1> - - <sect1 id="security.filesystem"> - <title>Filesystem Security</title> - <simpara> - PHP is subject to the security built into most server systems with - respect to permissions on a file and directory basis. This allows - you to control which files in the filesystem may be read. Care - should be taken with any files which are world readable to ensure - that they are safe for reading by all users who have access to that - filesystem. - </simpara> - <simpara> - Since PHP was designed to allow user level access to the filesystem, - it's entirely possible to write a PHP script that will allow you - to read system files such as /etc/passwd, modify your ethernet - connections, send massive printer jobs out, etc. This has some - obvious implications, in that you need to ensure that the files - that you read from and write to are the appropriate ones. - </simpara> - <simpara> - Consider the following script, where a user indicates that they'd - like to delete a file in their home directory. This assumes a - situation where a PHP web interface is regularly used for file - management, so the Apache user is allowed to delete files in - the user home directories. - </simpara> - <para> - <example> - <title>Poor variable checking leads to....</title> - <programlisting role="php"> -<![CDATA[ -<?php -// remove a file from the user's home directory -$username = $_POST['user_submitted_name']; -$homedir = "/home/$username"; -$file_to_delete = "$userfile"; -unlink ("$homedir/$userfile"); -echo "$file_to_delete has been deleted!"; -?> -]]> - </programlisting> - </example> - Since the username is postable from a user form, they can submit - a username and file belonging to someone else, and delete files. - In this case, you'd want to use some other form of authentication. - Consider what could happen if the variables submitted were - "../etc/" and "passwd". The code would then effectively read: - <example> - <title>... A filesystem attack</title> - <programlisting role="php"> -<![CDATA[ -<?php -// removes a file from anywhere on the hard drive that -// the PHP user has access to. If PHP has root access: -$username = "../etc/"; -$homedir = "/home/../etc/"; -$file_to_delete = "passwd"; -unlink ("/home/../etc/passwd"); -echo "/home/../etc/passwd has been deleted!"; -?> -]]> - </programlisting> - </example> - There are two important measures you should take to prevent these - issues. - <itemizedlist> - <listitem> - <simpara> - Only allow limited permissions to the PHP web user binary. - </simpara> - </listitem> - <listitem> - <simpara> - Check all variables which are submitted. - </simpara> - </listitem> - </itemizedlist> - Here is an improved script: - <example> - <title>More secure file name checking</title> - <programlisting role="php"> -<![CDATA[ -<?php -// removes a file from the hard drive that -// the PHP user has access to. -$username = $_SERVER['REMOTE_USER']; // using an authentication mechanisim - -$homedir = "/home/$username"; - -$file_to_delete = basename("$userfile"); // strip paths -unlink ($homedir/$file_to_delete); - -$fp = fopen("/home/logging/filedelete.log","+a"); //log the deletion -$logstring = "$username $homedir $file_to_delete"; -fputs ($fp, $logstring); -fclose($fp); - -echo "$file_to_delete has been deleted!"; -?> -]]> - </programlisting> - </example> - However, even this is not without it's flaws. If your authentication - system allowed users to create their own user logins, and a user - chose the login "../etc/", the system is once again exposed. For - this reason, you may prefer to write a more customized check: - <example> - <title>More secure file name checking</title> - <programlisting role="php"> -<![CDATA[ -<?php -$username = $_SERVER['REMOTE_USER']; // using an authentication mechanisim -$homedir = "/home/$username"; - -if (!ereg('^[^./][^/]*$', $userfile)) - die('bad filename'); //die, do not process - -if (!ereg('^[^./][^/]*$', $username)) - die('bad username'); //die, do not process -//etc... -?> -]]> - </programlisting> - </example> - </para> - <para> - Depending on your operating system, there are a wide variety of files - which you should be concerned about, including device entries (/dev/ - or COM1), configuration files (/etc/ files and the .ini files), - well known file storage areas (/home/, My Documents), etc. For this - reason, it's usually easier to create a policy where you forbid - everything except for what you explicitly allow. - </para> - </sect1> - - <sect1 id="security.database"> - <title>Database Security</title> - - <simpara> - Nowadays, databases are cardinal components of any web based application by - enabling websites to provide varying dynamic content. Since very sensitive - or secret information can be stored in a database, you should strongly - consider protecting your databases. - </simpara> - <simpara> - To retrieve or to store any information you need to connect to the database, - send a legitimate query, fetch the result, and close the connection. - Nowadays, the commonly used query language in this interaction is the - Structured Query Language (SQL). See how an attacker can <link - linkend="security.database.sql-injection">tamper with an SQL query</link>. - </simpara> - <simpara> - As you can surmise, PHP cannot protect your database by itself. The - following sections aim to be an introduction into the very basics of how to - access and manipulate databases within PHP scripts. - </simpara> - <simpara> - Keep in mind this simple rule: defense in depth. The more places you - take action to increase the protection of your database, the less - probability of an attacker succeeding in exposing or abusing any stored - information. Good design of the database schema and the application - deals with your greatest fears. - </simpara> - - <sect2 id="security.database.design"> - <title>Designing Databases</title> - <simpara> - The first step is always to create the database, unless you want to use - one from a third party. When a database is created, it is - assigned to an owner, who executed the creation statement. Usually, only - the owner (or a superuser) can do anything with the objects in that - database, and in order to allow other users to use it, privileges must be - granted. - </simpara> - <simpara> - Applications should never connect to the database as its owner or a - superuser, because these users can execute any query at will, for - example, modifying the schema (e.g. dropping tables) or deleting its - entire content. - </simpara> - <simpara> - You may create different database users for every aspect of your - application with very limited rights to database objects. The most - required privileges should be granted only, and avoid that the same user - can interact with the database in different use cases. This means that if - intruders gain access to your database using your applications credentials, - they can only effect as many changes as your application can. - </simpara> - <simpara> - You are encouraged not to implement all the business logic in the web - application (i.e. your script), instead do it in the database schema - using views, triggers or rules. If the system evolves, new ports will be - intended to open to the database, and you have to re-implement the logic - in each separate database client. Over and above, triggers can be used - to transparently and automatically handle fields, which often provides - insight when debugging problems with your application or tracing back - transactions. - </simpara> - </sect2> - - <sect2 id="security.database.connection"> - <title>Connecting to Database</title> - <simpara> - You may want to estabilish the connections over SSL to encrypt - client/server communications for increased security, or you can use ssh - to encrypt the network connection between clients and the database server. - If either of these is used, then monitoring your traffic and gaining - information about your database will be difficult for a would-be attacker. - </simpara> - <!--simpara> - If your database server has native SSL support, consider using <link - linkend="ref.openssl">OpenSSL functions</link> in communication between - PHP and database via SSL. - </simpara--> - </sect2> - - <sect2 id="security.database.storage"> - <title>Encrypted Storage Model</title> - <simpara> - SSL/SSH protects data travelling from the client to the server, SSL/SSH - does not protect the persistent data stored in a database. SSL is an - on-the-wire protocol. - </simpara> - <simpara> - Once an attacker gains access to your database directly (bypassing the - webserver), the stored sensitive data may be exposed or misused, unless - the information is protected by the database itself. Encrypting the data - is a good way to mitigate this threat, but very few databases offer this - type of data encryption. - </simpara> - <simpara> - The easiest way to work around this problem is to first create your own - encryption package, and then use it from within your PHP scripts. PHP - can assist you in this with several extensions, such as <link - linkend="ref.mcrypt">Mcrypt</link> and <link - linkend="ref.mhash">Mhash</link>, covering a wide variety of encryption - algorithms. The script encrypts the data before inserting it into the database, and decrypts - it when retrieving. See the references for further examples of how - encryption works. - </simpara> - <simpara> - In case of truly hidden data, if its raw representation is not needed - (i.e. not be displayed), hashing may also be taken into consideration. - The well-known example for the hashing is storing the MD5 hash of a - password in a database, instead of the password itself. See also - <function>crypt</function> and <function>md5</function>. - </simpara> - <example> - <title>Using hashed password field</title> - <programlisting role="php"> -<![CDATA[ -// storing password hash -$query = sprintf("INSERT INTO users(name,pwd) VALUES('%s','%s');", - addslashes($username), md5($password)); -$result = pg_exec($connection, $query); - -// querying if user submitted the right password -$query = sprintf("SELECT 1 FROM users WHERE name='%s' AND pwd='%s';", - addslashes($username), md5($password)); -$result = pg_exec($connection, $query); - -if (pg_numrows($result) > 0) { - echo "Welcome, $username!"; -} -else { - echo "Authentication failed for $username."; -} -]]> - </programlisting> - </example> - </sect2> - - <sect2 id="security.database.sql-injection"> - <title>SQL Injection</title> - <simpara> - Many web developers are unaware of how SQL queries can be tampered with, - and assume that an SQL query is a trusted command. It means that SQL - queries are able to circumvent access controls, thereby bypassing standard - authentication and authorization checks, and sometimes SQL queries even - may allow access to host operating system level commands. - </simpara> - <simpara> - Direct SQL Command Injection is a technique where an attacker creates or - alters existing SQL commands to expose hidden data, or to override valuable - ones, or even to execute dangerous system level commands on the database - host. This is accomplished by the application taking user input and - combining it with static parameters to build a SQL query. The following - examples are based on true stories, unfortunately. - </simpara> - <para> - Owing to the lack of input validation and connecting to the database on - behalf of a superuser or the one who can create users, the attacker - may create a superuser in your database. - <example> - <title> - Splitting the result set into pages ... and making superusers - (PostgreSQL and MySQL) - </title> - <programlisting role="php"> -<![CDATA[ -$offset = argv[0]; // beware, no input validation! -$query = "SELECT id, name FROM products ORDER BY name LIMIT 20 OFFSET $offset;"; -// with PostgreSQL -$result = pg_exec($conn, $query); -// with MySQL -$result = mysql_query($query); -]]> - </programlisting> - </example> - Normal users click on the 'next', 'prev' links where the <varname>$offset</varname> - is encoded into the URL. The script expects that the incoming - <varname>$offset</varname> is a decimal number. However, what if someone tries to - break in by appending a <function>urlencode</function>'d form of the - following to the URL - <informalexample> - <programlisting> -<![CDATA[ -// in case of PostgreSQL -0; -insert into pg_shadow(usename,usesysid,usesuper,usecatupd,passwd) - select 'crack', usesysid, 't','t','crack' - from pg_shadow where usename='postgres'; --- - -// in case of MySQL -0; -UPDATE user SET Password=PASSWORD('crack') WHERE user='root'; -FLUSH PRIVILEGES; -]]> - </programlisting> - </informalexample> - If it happened, then the script would present a superuser access to him. - Note that <literal>0;</literal> is to supply a valid offset to the - original query and to terminate it. - </para> - <note> - <para> - It is common technique to force the SQL parser to ignore the rest of the - query written by the developer with <literal>--</literal> which is the - comment sign in SQL. - </para> - </note> - <para> - A feasible way to gain passwords is to circumvent your search result pages. - The only thing the attacker needs to do is to see if there are any submitted variables - used in SQL statements which are not handled properly. These filters can be set - commonly in a preceding form to customize <literal>WHERE, ORDER BY, - LIMIT</literal> and <literal>OFFSET</literal> clauses in <literal>SELECT</literal> - statements. If your database supports the <literal>UNION</literal> construct, - the attacker may try to append an entire query to the original one to list - passwords from an arbitrary table. Using encrypted password fields is - strongly encouraged. - <example> - <title> - Listing out articles ... and some passwords (any database server) - </title> - <programlisting role="php"> -<![CDATA[ -$query = "SELECT id, name, inserted, size FROM products - WHERE size = '$size' - ORDER BY $order LIMIT $limit, $offset;"; -$result = odbc_exec($conn, $query); -]]> - </programlisting> - </example> - The static part of the query can be combined with another - <literal>SELECT</literal> statement which reveals all passwords: - <informalexample> - <programlisting> -<![CDATA[ -' -union select '1', concat(uname||'-'||passwd) as name, '1971-01-01', '0' from usertable; --- -]]> - </programlisting> - </informalexample> - If this query (playing with the <literal>'</literal> and - <literal>--</literal>) were assigned to one of the variables used in - <varname>$query</varname>, the query beast awakened. - </para> - <para> - SQL UPDATE's are also susceptible to attack. These queries are - also threatened by chopping and appending an entirely new query to it. But - the attacker might fiddle with the <literal>SET</literal> clause. In this - case some schema information must be possessed to manipulate the query - successfully. This can be acquired by examining the form variable names, or - just simply brute forcing. There are not so many naming conventions for - fields storing passwords or usernames. - <example> - <title> - From resetting a password ... to gaining more privileges (any database server) - </title> - <programlisting role="php"> -<![CDATA[ -$query = "UPDATE usertable SET pwd='$pwd' WHERE uid='$uid';"; -]]> - </programlisting> - </example> - But a malicious user sumbits the value - <literal>' or uid like'%admin%'; --</literal> to <varname>$uid</varname> to - change the admin's password, or simply sets <varname>$pwd</varname> to - <literal>"hehehe', admin='yes', trusted=100 "</literal> (with a trailing - space) to gain more privileges. Then, the query will be twisted: - <informalexample> - <programlisting role="php"> -<![CDATA[ -// $uid == ' or uid like'%admin%'; -- -$query = "UPDATE usertable SET pwd='...' WHERE uid='' or uid like '%admin%'; --"; - -// $pwd == "hehehe', admin='yes', trusted=100 " -$query = "UPDATE usertable SET pwd='hehehe', admin='yes', trusted=100 WHERE ...;" -]]> - </programlisting> - </informalexample> - </para> - <para> - A frightening example how operating system level commands can be accessed - on some database hosts. - <example> - <title>Attacking the database hosts operating system (MSSQL Server)</title> - <programlisting role="php"> -<![CDATA[ -$query = "SELECT * FROM products WHERE id LIKE '%$prod%'"; -$result = mssql_query($query); -]]> - </programlisting> - </example> - If attacker submits the value - <literal>a%' exec master..xp_cmdshell 'net user test testpass /ADD' --</literal> - to <varname>$prod</varname>, then the <varname>$query</varname> will be: - <informalexample> - <programlisting role="php"> -<![CDATA[ -$query = "SELECT * FROM products - WHERE id LIKE '%a%' - exec master..xp_cmdshell 'net user test testpass /ADD'--"; -$result = mssql_query($query); -]]> - </programlisting> - </informalexample> - MSSQL Server executes the SQL statements in the batch including a command - to add a new user to the local accounts database. If this application - were running as <literal>sa</literal> and the MSSQLSERVER service is - running with sufficient privileges, the attacker would now have an - account with which to access this machine. - </para> - <note> - <para> - Some of the examples above is tied to a specific database server. This - does not mean that a similar attack is impossible against other products. - Your database server may be similarly vulnerable in another manner. - </para> - </note> - - <sect3 id="security.database.avoiding"> - <title>Avoiding techniques</title> - <simpara> - You may plead that the attacker must possess a piece of information - about the database schema in most examples. You are right, but you - never know when and how it can be taken out, and if it happens, - your database may be exposed. If you are using an open source, or - publicly available database handling package, which may belong to a - content management system or forum, the intruders easily produce - a copy of a piece of your code. It may be also a security risk if it - is a poorly designed one. - </simpara> - <simpara> - These attacks are mainly based on exploiting the code not being written - with security in mind. Never trust any kind of input, especially that - which comes from the client side, even though it comes from a select box, - a hidden input field or a cookie. The first example shows that such a - blameless query can cause disasters. - </simpara> - - <itemizedlist> - <listitem> - <simpara> - Never connect to the database as a superuser or as the database owner. - Use always customized users with very limited privileges. - </simpara> - </listitem> - <listitem> - <simpara> - Check if the given input has the expected data type. PHP has - a wide range of input validating functions, from the simplest ones - found in <link linkend="ref.variables">Variable Functions</link> and - in <link linkend="ref.ctype">Character Type Functions</link> - (e.g. <function>is_numeric</function>, <function>ctype_digit</function> - respectively) and onwards to the - <link linkend="ref.pcre">Perl compatible Regular Expressions</link> - support. - </simpara> - </listitem> - <listitem> - <para> - If the application waits for numerical input, consider verifying data - with <function>is_numeric</function>, or silently change its type - using <function>settype</function>, or use its numeric representation - by <function>sprintf</function>. - <example> - <title>A more secure way to compose a query for paging</title> - <programlisting role="php"> -<![CDATA[ -settype($offset, 'integer'); -$query = "SELECT id, name FROM products ORDER BY name LIMIT 20 OFFSET $offset;"; - -// please note %d in the format string, using %s would be meaningless -$query = sprintf("SELECT id, name FROM products ORDER BY name LIMIT 20 OFFSET %d;", - $offset); -]]> - </programlisting> - </example> - </para> - </listitem> - <listitem> - <simpara> - Quote each non numeric user input which is passed to the database with - <function>addslashes</function> or <function>addcslashes</function>. - See <link linkend="security.database.storage">the first example</link>. - As the examples shows, quotes burnt into the static part of the query - is not enough, and can be easily cracked. - </simpara> - </listitem> - <listitem> - <simpara> - Do not print out any database specific information, especially - about the schema, by fair means or foul. See also <link - linkend="security.errors">Error Reporting</link> and <link - linkend="ref.errorfunc">Error Handling and Logging Functions</link>. - </simpara> - </listitem> - <listitem> - <simpara> - You may use stored procedures and previously defined cursors to abstract - data access so that users do not directly access tables or views, but - this solution has another impacts. - </simpara> - </listitem> - </itemizedlist> - - <simpara> - Besides these, you benefit from logging queries either within your script - or by the database itself, if it supports logging. Obviously, the logging is unable - to prevent any harmful attempt, but it can be helpful to trace back which - application has been circumvented. The log is not useful by itself, but - through the information it contains. More detail is generally better than less. - </simpara> - </sect3> - </sect2> - </sect1> - - <sect1 id="security.errors"> - <title>Error Reporting</title> - <para> - With PHP security, there are two sides to error reporting. One is - beneficial to increasing security, the other is detrimental. - </para> - <para> - A standard attack tactic involves profiling a system by feeding - it improper data, and checking for the kinds, and contexts, of the - errors which are returned. This allows the system cracker to probe - for information about the server, to determine possible weaknesses. - For example, if an attacker had gleaned information about a page - based on a prior form submission, they may attempt to override - variables, or modify them: - <example> - <title>Attacking Variables with a custom HTML page</title> - <programlisting role="php"> -<![CDATA[ -<form method="post" action="attacktarget?username=badfoo&password=badfoo"> -<input type="hidden" name="username" value="badfoo" /> -<input type="hidden" name="password" value="badfoo" /> -</form> -]]> - </programlisting> - </example> - </para> - <para> - The PHP errors which are normally returned can be quite helpful to a - developer who is trying to debug a script, indicating such things - as the function or file that failed, the PHP file it failed in, - and the line number which the failure occured in. This is all - information that can be exploited. It is not uncommon for a php - developer to use <function>show_source</function>, - <function>highlight_string</function>, or - <function>highlight_file</function> as a debugging measure, but in - a live site, this can expose hidden variables, unchecked syntax, - and other dangerous information. Especially dangerous is running - code from known sources with built-in debugging handlers, or using - common debugging techniques. If the attacker can determine what - general technique you are using, they may try to brute-force a page, - by sending various common debugging strings: - <example> - <title>Exploiting common debugging variables</title> - <programlisting role="php"> -<![CDATA[ -<form method="post" action="attacktarget?errors=Y&amp;showerrors=1&amp;debug=1"> -<input type="hidden" name="errors" value="Y" /> -<input type="hidden" name="showerrors" value="1" /> -<input type="hidden" name="debug" value="1" /> -</form> -]]> - </programlisting> - </example> - </para> - <para> - Regardless of the method of error handling, the ability to probe a - system for errors leads to providing an attacker with more - information. - </para> - <para> - For example, the very style of a generic PHP error indicates a system - is running PHP. If the attacker was looking at an .html page, and - wanted to probe for the back-end (to look for known weaknesses in - the system), by feeding it the wrong data they may be able to - determine that a system was built with PHP. - </para> - <para> - A function error can indicate whether a system may be running a - specific database engine, or give clues as to how a web page or - programmed or designed. This allows for deeper investigation into - open database ports, or to look for specific bugs or weaknesses - in a web page. By feeding different pieces of bad data, for example, - an attacker can determine the order of authentication in a script, - (from the line number errors) as well as probe for exploits that - may be exploited in different locations in the script. - </para> - <para> - A filesystem or general PHP error can indicate what permissions - the webserver has, as well as the structure and organization of - files on the web server. Developer written error code can aggravate - this problem, leading to easy exploitation of formerly "hidden" - information. - </para> - <para> - There are three major solutions to this issue. The first is to - scrutinize all functions, and attempt to compensate for the bulk - of the errors. The second is to disable error reporting entirely - on the running code. The third is to use PHP's custom error - handling functions to create your own error handler. Depending - on your security policy, you may find all three to be applicable - to your situation. - </para> - <para> - One way of catching this issue ahead of time is to make use of - PHP's own <function>error_reporting</function>, to help you - secure your code and find variable usage that may be dangerous. - By testing your code, prior to deployment, with E_ALL, you can - quickly find areas where your variables may be open to poisoning - or modification in other ways. Once you are ready for deployment, - by using E_NONE, you insulate your code from probing. - <example> - <title>Finding dangerous variables with E_ALL</title> - <programlisting role="php"> -<![CDATA[ -<?php -if ($username) { // Not initialized or checked before usage - $good_login = 1; -} -if ($good_login == 1) { // If above test fails, not initialized or checked before usage - readfile ("/highly/sensitive/data/index.html"); -} -?> -]]> - </programlisting> - </example> - </para> - </sect1> - - <sect1 id="security.registerglobals"> - <title>Using Register Globals</title> - <para> - Perhaps the most controversial change in PHP is when the default value - for the PHP directive <link linkend="ini.register-globals"> - register_globals</link> went from ON to OFF in PHP - <ulink url="&url.php.release4.2.0;">4.2.0</ulink>. Reliance on this - directive was quite common and many people didn't even know it existed - and assumed it's just how PHP works. This page will explain how one can - write insecure code with this directive but keep in mind that the - directive itself isn't insecure but rather it's the misuse of it. - </para> - <para> - When on, register_globals will inject (poison) your scripts will all - sorts of variables, like request variables from HTML forms. This - coupled with the fact that PHP doesn't require variable initialization - means writing insecure code is that much easier. It was a difficult - decision, but the PHP community decided to disable this directive by - default. When on, people use variables yet really don't know for sure - where they come from and can only assume. Internal variables that are - defined in the script itself get mixed up with request data sent by - users and disabling register_globals changes this. Let's demonstrate - with an example misuse of register_globals: - </para> - <para> - <example> - <title>Example misuse with register_globals = on</title> - <programlisting role="php"> -<![CDATA[ -<?php -// define $authorized = true only if user is authenticated -if (authenticated_user()) { - $authorized = true; -} - -// Because we didn't first initialize $authorized as false, this might be -// defined through register_globals, like from GET auth.php?authorized=1 -// So, anyone can be seen as authenticated! -if ($authorized) { - include "/highly/sensitive/data.php"; -} -?> -]]> - </programlisting> - </example> - </para> - <para> - When register_globals = on, our logic above may be compromised. When - off, <varname>$authorized</varname> can't be set via request so it'll - be fine, although it really is generally a good programming practice to - initialize variables first. For example, in our example above we might - have first done <literal>$authorized = false</literal>. Doing this - first means our above code would work with register_globals on or off as - users by default would be unauthorized. - </para> - <para> - Another example is that of <link linkend="ref.session">sessions</link>. - When register_globals = on, we could also use - <varname>$username</varname> in our example below but again you must - realize that <varname>$username</varname> could also come from other - means, such as GET (through the URL). - </para> - <para> - <example> - <title>Example use of sessions with register_globals on or off</title> - <programlisting role="php"> -<![CDATA[ -<?php -// We wouldn't know where $username came from but do know $_SESSION is -// for session data -if (isset($_SESSION['username'])) { - - echo "Hello <b>{$_SESSION['username']}</b>"; - -} else { - - echo "Hello <b>Guest</b><br />"; - echo "Would you like to login?"; - -} -?> -]]> - </programlisting> - </example> - </para> - <para> - It's even possible to take preventative measures to warn when forging is - being attempted. If you know ahead of time exactly where a variable - should be coming from, you can check to see if the submitted data is - coming from an inappropriate kind of submission. While it doesn't - guarantee that data has not been forged, it does require an attacker to - guess the right kind of forging. If you don't care where the request - data comes from, you can use <varname>$_REQUEST</varname> as it contains - a mix of GET, POST and COOKIE data. See also the manual section on - using <link linkend="language.variables.external">variables from outside - of PHP</link>. - </para> - <para> - <example> - <title>Detecting simple variable poisoning</title> - <programlisting role="php"> -<![CDATA[ -<?php -if (isset($_COOKIE['MAGIC_COOKIE'])) { - - // MAGIC_COOKIE comes from a cookie. - // Be sure to validate the cookie data! - -} elseif (isset($_GET['MAGIC_COOKIE']) || isset($_POST['MAGIC_COOKIE'])) { - - mail("admin@example.com", "Possible breakin attempt", $_SERVER['REMOTE_ADDR']); - echo "Security violation, admin has been alerted."; - exit; - -} else { - - // MAGIC_COOKIE isn't set through this REQUEST - -} -?> -]]> - </programlisting> - </example> - </para> - <para> - Of course, simply turning off register_globals does not mean your code - is secure. For every piece of data that is submitted, it should also be - checked in other ways. Always validate your user data and initialize - your variables! To check for unitialized variables you may turn up - <function>error_reporting</function> to show - <constant>E_NOTICE</constant> level errors. - </para> - - &note.superglobals; - - </sect1> - - - <sect1 id="security.variables"> - <title>User Submitted Data</title> - <para> - The greatest weakness in many PHP programs is not inherent in the - language itself, but merely an issue of code not being written with - security in mind. For this reason, you should always take the time - to consider the implications of a given piece of code, to ascertain - the possible damage if an unexpected variable is submitted to it. - <example> - <title>Dangerous Variable Usage</title> - <programlisting role="php"> -<![CDATA[ -<?php -// remove a file from the user's home directory... or maybe -// somebody else's? -unlink ($evil_var); - -// Write logging of their access... or maybe an /etc/passwd entry? -fputs ($fp, $evil_var); - -// Execute something trivial.. or rm -rf *? -system ($evil_var); -exec ($evil_var); - -?> -]]> - </programlisting> - </example> - You should always carefully examine your code to make sure that any - variables being submitted from a web browser are being properly - checked, and ask yourself the following questions: - <itemizedlist> - <listitem> - <simpara> - Will this script only affect the intended files? - </simpara> - </listitem> - <listitem> - <simpara> - Can unusual or undesirable data be acted upon? - </simpara> - </listitem> - <listitem> - <simpara> - Can this script be used in unintended ways? - </simpara> - </listitem> - <listitem> - <simpara> - Can this be used in conjunction with other scripts in a negative - manner? - </simpara> - </listitem> - <listitem> - <simpara> - Will any transactions be adequately logged? - </simpara> - </listitem> - </itemizedlist> - By adequately asking these questions while writing the script, - rather than later, you prevent an unfortunate re-write when you - need to increase your security. By starting out with this mindset, - you won't guarantee the security of your system, but you can help - improve it. - </para> - <para> - You may also want to consider turning off register_globals, - magic_quotes, or other convenience settings which may confuse - you as to the validity, source, or value of a given variable. - Working with PHP in error_reporting(E_ALL) mode can also help warn - you about variables being used before they are checked or - initialized (so you can prevent unusual data from being - operated upon). - </para> - </sect1> - - <sect1 id="security.hiding"> - <title>Hiding PHP</title> - <para> - In general, security by obscurity is one of the weakest forms of security. - But in some cases, every little bit of extra security is desirable. - </para> - <para> - A few simple techniques can help to hide PHP, possibly slowing - down an attacker who is attempting to discover weaknesses in your - system. By setting expose_php = off in your &php.ini; file, you - reduce the amount of information available to them. - </para> - <para> - Another tactic is to configure web servers such as apache to - parse different filetypes through PHP, either with an &htaccess; - directive, or in the apache configuration file itself. You can - then use misleading file extensions: - <example> - <title>Hiding PHP as another language</title> - <programlisting role="apache-conf"> -<![CDATA[ -# Make PHP code look like other code types -AddType application/x-httpd-php .asp .py .pl -]]> - </programlisting> - </example> - Or obscure it completely: - <example> - <title>Using unknown types for PHP extensions</title> - <programlisting role="apache-conf"> -<![CDATA[ -# Make PHP code look like unknown types -AddType application/x-httpd-php .bop .foo .133t -]]> - </programlisting> - </example> - Or hide it as HTML code, which has a slight performance hit because - all HTML will be parsed through the PHP engine: - <example> - <title>Using HTML types for PHP extensions</title> - <programlisting role="apache-conf"> -<![CDATA[ -# Make all PHP code look like HTML -AddType application/x-httpd-php .htm .html -]]> - </programlisting> - </example> - For this to work effectively, you must rename your PHP files with - the above extensions. While it is a form of security through - obscurity, it's a minor preventative measure with few drawbacks. - </para> - </sect1> - - <sect1 id="security.current"> - <title>Keeping Current</title> - <simpara> - PHP, like any other large system, is under constant scrutiny and - improvement. Each new version will often include both major and - minor changes to enhance and repair security flaws, configuration - mishaps, and other issues that will affect the overall security - and stability of your system. - </simpara> - <simpara> - Like other system-level scripting languages and programs, the best - approach is to update often, and maintain awareness of the latest - versions and their changes. - </simpara> - </sect1> + &security.general; + &security.cgi-bin; + &security.apache; + &security.filesystem; + &security.database; + &security.errors; + &security.globals; + &security.variables; + &security.hiding; + &security.current; </chapter> <!-- Keep this comment at the end of the file http://cvs.php.net/co.php/phpdoc/en/security/apache.xml?r=1.1&p=1 Index: phpdoc/en/security/apache.xml +++ phpdoc/en/security/apache.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.apache"> <title>Installed as an Apache module</title> <simpara> When PHP is used as an Apache module it inherits Apache's user permissions (typically those of the "nobody" user). This has several impacts on security and authorization. For example, if you are using PHP to access a database, unless that database has built-in access control, you will have to make the database accessible to the "nobody" user. This means a malicious script could access and modify the database, even without a username and password. It's entirely possible that a web spider could stumble across a database administrator's web page, and drop all of your databases. You can protect against this with Apache authorization, or you can design your own access model using LDAP, &htaccess; files, etc. and include that code as part of your PHP scripts. </simpara> <simpara> Often, once security is established to the point where the PHP user (in this case, the apache user) has very little risk attached to it, it is discovered that PHP is now prevented from writing any files to user directories. Or perhaps it has been prevented from accessing or changing databases. It has equally been secured from writing good and bad files, or entering good and bad database transactions. </simpara> <simpara> A frequent security mistake made at this point is to allow apache root permissions, or to escalate apache's abilitites in some other way. </simpara> <simpara> Escalating the Apache user's permissions to root is extremely dangerous and may compromise the entire system, so sudo'ing, chroot'ing, or otherwise running as root should not be considered by those who are not security professionals. </simpara> <simpara> There are some simpler solutions. By using <link linkend="ini.open-basedir">open_basedir</link> you can control and restrict what directories are allowed to be used for PHP. You can also set up apache-only areas, to restrict all web based activity to non-user, or non-system, files. </simpara> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/cgi-bin.xml?r=1.1&p=1 Index: phpdoc/en/security/cgi-bin.xml +++ phpdoc/en/security/cgi-bin.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.cgi-bin"> <title>Installed as CGI binary</title> <sect2 id="security.cgi-bin.attacks"> <title>Possible attacks</title> <simpara> Using PHP as a <acronym>CGI</acronym> binary is an option for setups that for some reason do not wish to integrate PHP as a module into server software (like Apache), or will use PHP with different kinds of CGI wrappers to create safe chroot and setuid environments for scripts. This setup usually involves installing executable PHP binary to the web server cgi-bin directory. CERT advisory <ulink url="&url.cert;">CA-96.11</ulink> recommends against placing any interpreters into cgi-bin. Even if the PHP binary can be used as a standalone interpreter, PHP is designed to prevent the attacks this setup makes possible: </simpara> <itemizedlist> <listitem> <simpara> Accessing system files: <filename role="url">http://my.host/cgi-bin/php?/etc/passwd</filename> </simpara> <simpara> The query information in a url after the question mark (?) is passed as command line arguments to the interpreter by the CGI interface. Usually interpreters open and execute the file specified as the first argument on the command line. </simpara> <simpara> When invoked as a CGI binary, PHP refuses to interpret the command line arguments. </simpara> </listitem> <listitem> <simpara> Accessing any web document on server: <filename role="url">http://my.host/cgi-bin/php/secret/doc.html</filename> </simpara> <simpara> The path information part of the url after the PHP binary name, <filename role="uri">/secret/doc.html</filename> is conventionally used to specify the name of the file to be opened and interpreted by the <acronym>CGI</acronym> program. Usually some web server configuration directives (Apache: Action) are used to redirect requests to documents like <filename role="url">http://my.host/secret/script.php</filename> to the PHP interpreter. With this setup, the web server first checks the access permissions to the directory <filename role="uri">/secret</filename>, and after that creates the redirected request <filename role="url">http://my.host/cgi-bin/php/secret/script.php</filename>. Unfortunately, if the request is originally given in this form, no access checks are made by web server for file <filename role="uri">/secret/script.php</filename>, but only for the <filename role="uri">/cgi-bin/php</filename> file. This way any user able to access <filename role="uri">/cgi-bin/php</filename> is able to access any protected document on the web server. </simpara> <simpara> In PHP, compile-time configuration option <link linkend="install.configure.enable-force-cgi-redirect">--enable-force-cgi-redirect</link> and runtime configuration directives <link linkend="ini.doc-root">doc_root</link> and <link linkend="ini.user-dir">user_dir</link> can be used to prevent this attack, if the server document tree has any directories with access restrictions. See below for full the explanation of the different combinations. </simpara> </listitem> </itemizedlist> </sect2> <sect2 id="security.cgi-bin.default"> <title>Case 1: only public files served</title> <simpara> If your server does not have any content that is not restricted by password or ip based access control, there is no need for these configuration options. If your web server does not allow you to do redirects, or the server does not have a way to communicate to the PHP binary that the request is a safely redirected request, you can specify the option <link linkend="install.configure.enable-force-cgi-redirect">--enable-force-cgi-redirect</link> to the configure script. You still have to make sure your PHP scripts do not rely on one or another way of calling the script, neither by directly <filename role="php">http://my.host/cgi-bin/php/dir/script.php</filename> nor by redirection <filename role="php">http://my.host/dir/script.php</filename>. </simpara> <simpara> Redirection can be configured in Apache by using AddHandler and Action directives (see below). </simpara> </sect2> <sect2 id="security.cgi-bin.force-redirect"> <title>Case 2: using --enable-force-cgi-redirect</title> <simpara> This compile-time option prevents anyone from calling PHP directly with a url like <filename role="php">http://my.host/cgi-bin/php/secretdir/script.php</filename>. Instead, PHP will only parse in this mode if it has gone through a web server redirect rule. </simpara> <simpara> Usually the redirection in the Apache configuration is done with the following directives: </simpara> <programlisting role="apache-conf"> <![CDATA[ Action php-script /cgi-bin/php AddHandler php-script .php ]]> </programlisting> <simpara> This option has only been tested with the Apache web server, and relies on Apache to set the non-standard CGI environment variable <envar>REDIRECT_STATUS</envar> on redirected requests. If your web server does not support any way of telling if the request is direct or redirected, you cannot use this option and you must use one of the other ways of running the CGI version documented here. </simpara> </sect2> <sect2 id="security.cgi-bin.doc-root"> <title>Case 3: setting doc_root or user_dir</title> <simpara> To include active content, like scripts and executables, in the web server document directories is sometimes considered an insecure practice. If, because of some configuration mistake, the scripts are not executed but displayed as regular HTML documents, this may result in leakage of intellectual property or security information like passwords. Therefore many sysadmins will prefer setting up another directory structure for scripts that are accessible only through the PHP CGI, and therefore always interpreted and not displayed as such. </simpara> <simpara> Also if the method for making sure the requests are not redirected, as described in the previous section, is not available, it is necessary to set up a script doc_root that is different from web document root. </simpara> <simpara> You can set the PHP script document root by the configuration directive <link linkend="ini.doc-root">doc_root</link> in the <link linkend="configuration.file">configuration file</link>, or you can set the environment variable <envar>PHP_DOCUMENT_ROOT</envar>. If it is set, the CGI version of PHP will always construct the file name to open with this <parameter>doc_root</parameter> and the path information in the request, so you can be sure no script is executed outside this directory (except for <parameter>user_dir</parameter> below). </simpara> <simpara> Another option usable here is <link linkend="ini.user-dir">user_dir</link>. When user_dir is unset, only thing controlling the opened file name is <parameter>doc_root</parameter>. Opening an url like <filename role="url">http://my.host/~user/doc.php</filename> does not result in opening a file under users home directory, but a file called <filename role="uri">~user/doc.php</filename> under doc_root (yes, a directory name starting with a tilde [<literal>~</literal>]). </simpara> <simpara> If user_dir is set to for example <filename role="dir">public_php</filename>, a request like <filename role="url">http://my.host/~user/doc.php</filename> will open a file called <filename>doc.php</filename> under the directory named <filename role="dir">public_php</filename> under the home directory of the user. If the home of the user is <filename role="dir">/home/user</filename>, the file executed is <filename>/home/user/public_php/doc.php</filename>. </simpara> <simpara> <parameter>user_dir</parameter> expansion happens regardless of the <parameter>doc_root</parameter> setting, so you can control the document root and user directory access separately. </simpara> </sect2> <sect2 id="security.cgi-bin.shell"> <title>Case 4: PHP parser outside of web tree</title> <para> A very secure option is to put the PHP parser binary somewhere outside of the web tree of files. In <filename role="dir">/usr/local/bin</filename>, for example. The only real downside to this option is that you will now have to put a line similar to: <informalexample> <programlisting> <![CDATA[ #!/usr/local/bin/php ]]> </programlisting> </informalexample> as the first line of any file containing PHP tags. You will also need to make the file executable. That is, treat it exactly as you would treat any other CGI script written in Perl or sh or any other common scripting language which uses the <literal>#!</literal> shell-escape mechanism for launching itself. </para> <para> To get PHP to handle <envar>PATH_INFO</envar> and <envar>PATH_TRANSLATED</envar> information correctly with this setup, the PHP parser should be compiled with the <link linkend="install.configure.enable-discard-path">--enable-discard-path</link> configure option. </para> </sect2> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/current.xml?r=1.1&p=1 Index: phpdoc/en/security/current.xml +++ phpdoc/en/security/current.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.current"> <title>Keeping Current</title> <simpara> PHP, like any other large system, is under constant scrutiny and improvement. Each new version will often include both major and minor changes to enhance and repair security flaws, configuration mishaps, and other issues that will affect the overall security and stability of your system. </simpara> <simpara> Like other system-level scripting languages and programs, the best approach is to update often, and maintain awareness of the latest versions and their changes. </simpara> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/database.xml?r=1.1&p=1 Index: phpdoc/en/security/database.xml +++ phpdoc/en/security/database.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.database"> <title>Database Security</title> <simpara> Nowadays, databases are cardinal components of any web based application by enabling websites to provide varying dynamic content. Since very sensitive or secret information can be stored in a database, you should strongly consider protecting your databases. </simpara> <simpara> To retrieve or to store any information you need to connect to the database, send a legitimate query, fetch the result, and close the connection. Nowadays, the commonly used query language in this interaction is the Structured Query Language (SQL). See how an attacker can <link linkend="security.database.sql-injection">tamper with an SQL query</link>. </simpara> <simpara> As you can surmise, PHP cannot protect your database by itself. The following sections aim to be an introduction into the very basics of how to access and manipulate databases within PHP scripts. </simpara> <simpara> Keep in mind this simple rule: defense in depth. The more places you take action to increase the protection of your database, the less probability of an attacker succeeding in exposing or abusing any stored information. Good design of the database schema and the application deals with your greatest fears. </simpara> <sect2 id="security.database.design"> <title>Designing Databases</title> <simpara> The first step is always to create the database, unless you want to use one from a third party. When a database is created, it is assigned to an owner, who executed the creation statement. Usually, only the owner (or a superuser) can do anything with the objects in that database, and in order to allow other users to use it, privileges must be granted. </simpara> <simpara> Applications should never connect to the database as its owner or a superuser, because these users can execute any query at will, for example, modifying the schema (e.g. dropping tables) or deleting its entire content. </simpara> <simpara> You may create different database users for every aspect of your application with very limited rights to database objects. The most required privileges should be granted only, and avoid that the same user can interact with the database in different use cases. This means that if intruders gain access to your database using your applications credentials, they can only effect as many changes as your application can. </simpara> <simpara> You are encouraged not to implement all the business logic in the web application (i.e. your script), instead do it in the database schema using views, triggers or rules. If the system evolves, new ports will be intended to open to the database, and you have to re-implement the logic in each separate database client. Over and above, triggers can be used to transparently and automatically handle fields, which often provides insight when debugging problems with your application or tracing back transactions. </simpara> </sect2> <sect2 id="security.database.connection"> <title>Connecting to Database</title> <simpara> You may want to estabilish the connections over SSL to encrypt client/server communications for increased security, or you can use ssh to encrypt the network connection between clients and the database server. If either of these is used, then monitoring your traffic and gaining information about your database will be difficult for a would-be attacker. </simpara> <!--simpara> If your database server has native SSL support, consider using <link linkend="ref.openssl">OpenSSL functions</link> in communication between PHP and database via SSL. </simpara--> </sect2> <sect2 id="security.database.storage"> <title>Encrypted Storage Model</title> <simpara> SSL/SSH protects data travelling from the client to the server, SSL/SSH does not protect the persistent data stored in a database. SSL is an on-the-wire protocol. </simpara> <simpara> Once an attacker gains access to your database directly (bypassing the webserver), the stored sensitive data may be exposed or misused, unless the information is protected by the database itself. Encrypting the data is a good way to mitigate this threat, but very few databases offer this type of data encryption. </simpara> <simpara> The easiest way to work around this problem is to first create your own encryption package, and then use it from within your PHP scripts. PHP can assist you in this with several extensions, such as <link linkend="ref.mcrypt">Mcrypt</link> and <link linkend="ref.mhash">Mhash</link>, covering a wide variety of encryption algorithms. The script encrypts the data before inserting it into the database, and decrypts it when retrieving. See the references for further examples of how encryption works. </simpara> <simpara> In case of truly hidden data, if its raw representation is not needed (i.e. not be displayed), hashing may also be taken into consideration. The well-known example for the hashing is storing the MD5 hash of a password in a database, instead of the password itself. See also <function>crypt</function> and <function>md5</function>. </simpara> <example> <title>Using hashed password field</title> <programlisting role="php"> <![CDATA[ // storing password hash $query = sprintf("INSERT INTO users(name,pwd) VALUES('%s','%s');", addslashes($username), md5($password)); $result = pg_exec($connection, $query); // querying if user submitted the right password $query = sprintf("SELECT 1 FROM users WHERE name='%s' AND pwd='%s';", addslashes($username), md5($password)); $result = pg_exec($connection, $query); if (pg_numrows($result) > 0) { echo "Welcome, $username!"; } else { echo "Authentication failed for $username."; } ]]> </programlisting> </example> </sect2> <sect2 id="security.database.sql-injection"> <title>SQL Injection</title> <simpara> Many web developers are unaware of how SQL queries can be tampered with, and assume that an SQL query is a trusted command. It means that SQL queries are able to circumvent access controls, thereby bypassing standard authentication and authorization checks, and sometimes SQL queries even may allow access to host operating system level commands. </simpara> <simpara> Direct SQL Command Injection is a technique where an attacker creates or alters existing SQL commands to expose hidden data, or to override valuable ones, or even to execute dangerous system level commands on the database host. This is accomplished by the application taking user input and combining it with static parameters to build a SQL query. The following examples are based on true stories, unfortunately. </simpara> <para> Owing to the lack of input validation and connecting to the database on behalf of a superuser or the one who can create users, the attacker may create a superuser in your database. <example> <title> Splitting the result set into pages ... and making superusers (PostgreSQL and MySQL) </title> <programlisting role="php"> <![CDATA[ $offset = argv[0]; // beware, no input validation! $query = "SELECT id, name FROM products ORDER BY name LIMIT 20 OFFSET $offset;"; // with PostgreSQL $result = pg_exec($conn, $query); // with MySQL $result = mysql_query($query); ]]> </programlisting> </example> Normal users click on the 'next', 'prev' links where the <varname>$offset</varname> is encoded into the URL. The script expects that the incoming <varname>$offset</varname> is a decimal number. However, what if someone tries to break in by appending a <function>urlencode</function>'d form of the following to the URL <informalexample> <programlisting> <![CDATA[ // in case of PostgreSQL 0; insert into pg_shadow(usename,usesysid,usesuper,usecatupd,passwd) select 'crack', usesysid, 't','t','crack' from pg_shadow where usename='postgres'; -- // in case of MySQL 0; UPDATE user SET Password=PASSWORD('crack') WHERE user='root'; FLUSH PRIVILEGES; ]]> </programlisting> </informalexample> If it happened, then the script would present a superuser access to him. Note that <literal>0;</literal> is to supply a valid offset to the original query and to terminate it. </para> <note> <para> It is common technique to force the SQL parser to ignore the rest of the query written by the developer with <literal>--</literal> which is the comment sign in SQL. </para> </note> <para> A feasible way to gain passwords is to circumvent your search result pages. The only thing the attacker needs to do is to see if there are any submitted variables used in SQL statements which are not handled properly. These filters can be set commonly in a preceding form to customize <literal>WHERE, ORDER BY, LIMIT</literal> and <literal>OFFSET</literal> clauses in <literal>SELECT</literal> statements. If your database supports the <literal>UNION</literal> construct, the attacker may try to append an entire query to the original one to list passwords from an arbitrary table. Using encrypted password fields is strongly encouraged. <example> <title> Listing out articles ... and some passwords (any database server) </title> <programlisting role="php"> <![CDATA[ $query = "SELECT id, name, inserted, size FROM products WHERE size = '$size' ORDER BY $order LIMIT $limit, $offset;"; $result = odbc_exec($conn, $query); ]]> </programlisting> </example> The static part of the query can be combined with another <literal>SELECT</literal> statement which reveals all passwords: <informalexample> <programlisting> <![CDATA[ ' union select '1', concat(uname||'-'||passwd) as name, '1971-01-01', '0' from usertable; -- ]]> </programlisting> </informalexample> If this query (playing with the <literal>'</literal> and <literal>--</literal>) were assigned to one of the variables used in <varname>$query</varname>, the query beast awakened. </para> <para> SQL UPDATE's are also susceptible to attack. These queries are also threatened by chopping and appending an entirely new query to it. But the attacker might fiddle with the <literal>SET</literal> clause. In this case some schema information must be possessed to manipulate the query successfully. This can be acquired by examining the form variable names, or just simply brute forcing. There are not so many naming conventions for fields storing passwords or usernames. <example> <title> From resetting a password ... to gaining more privileges (any database server) </title> <programlisting role="php"> <![CDATA[ $query = "UPDATE usertable SET pwd='$pwd' WHERE uid='$uid';"; ]]> </programlisting> </example> But a malicious user sumbits the value <literal>' or uid like'%admin%'; --</literal> to <varname>$uid</varname> to change the admin's password, or simply sets <varname>$pwd</varname> to <literal>"hehehe', admin='yes', trusted=100 "</literal> (with a trailing space) to gain more privileges. Then, the query will be twisted: <informalexample> <programlisting role="php"> <![CDATA[ // $uid == ' or uid like'%admin%'; -- $query = "UPDATE usertable SET pwd='...' WHERE uid='' or uid like '%admin%'; --"; // $pwd == "hehehe', admin='yes', trusted=100 " $query = "UPDATE usertable SET pwd='hehehe', admin='yes', trusted=100 WHERE ...;" ]]> </programlisting> </informalexample> </para> <para> A frightening example how operating system level commands can be accessed on some database hosts. <example> <title>Attacking the database hosts operating system (MSSQL Server)</title> <programlisting role="php"> <![CDATA[ $query = "SELECT * FROM products WHERE id LIKE '%$prod%'"; $result = mssql_query($query); ]]> </programlisting> </example> If attacker submits the value <literal>a%' exec master..xp_cmdshell 'net user test testpass /ADD' --</literal> to <varname>$prod</varname>, then the <varname>$query</varname> will be: <informalexample> <programlisting role="php"> <![CDATA[ $query = "SELECT * FROM products WHERE id LIKE '%a%' exec master..xp_cmdshell 'net user test testpass /ADD'--"; $result = mssql_query($query); ]]> </programlisting> </informalexample> MSSQL Server executes the SQL statements in the batch including a command to add a new user to the local accounts database. If this application were running as <literal>sa</literal> and the MSSQLSERVER service is running with sufficient privileges, the attacker would now have an account with which to access this machine. </para> <note> <para> Some of the examples above is tied to a specific database server. This does not mean that a similar attack is impossible against other products. Your database server may be similarly vulnerable in another manner. </para> </note> <sect3 id="security.database.avoiding"> <title>Avoiding techniques</title> <simpara> You may plead that the attacker must possess a piece of information about the database schema in most examples. You are right, but you never know when and how it can be taken out, and if it happens, your database may be exposed. If you are using an open source, or publicly available database handling package, which may belong to a content management system or forum, the intruders easily produce a copy of a piece of your code. It may be also a security risk if it is a poorly designed one. </simpara> <simpara> These attacks are mainly based on exploiting the code not being written with security in mind. Never trust any kind of input, especially that which comes from the client side, even though it comes from a select box, a hidden input field or a cookie. The first example shows that such a blameless query can cause disasters. </simpara> <itemizedlist> <listitem> <simpara> Never connect to the database as a superuser or as the database owner. Use always customized users with very limited privileges. </simpara> </listitem> <listitem> <simpara> Check if the given input has the expected data type. PHP has a wide range of input validating functions, from the simplest ones found in <link linkend="ref.variables">Variable Functions</link> and in <link linkend="ref.ctype">Character Type Functions</link> (e.g. <function>is_numeric</function>, <function>ctype_digit</function> respectively) and onwards to the <link linkend="ref.pcre">Perl compatible Regular Expressions</link> support. </simpara> </listitem> <listitem> <para> If the application waits for numerical input, consider verifying data with <function>is_numeric</function>, or silently change its type using <function>settype</function>, or use its numeric representation by <function>sprintf</function>. <example> <title>A more secure way to compose a query for paging</title> <programlisting role="php"> <![CDATA[ settype($offset, 'integer'); $query = "SELECT id, name FROM products ORDER BY name LIMIT 20 OFFSET $offset;"; // please note %d in the format string, using %s would be meaningless $query = sprintf("SELECT id, name FROM products ORDER BY name LIMIT 20 OFFSET %d;", $offset); ]]> </programlisting> </example> </para> </listitem> <listitem> <simpara> Quote each non numeric user input which is passed to the database with <function>addslashes</function> or <function>addcslashes</function>. See <link linkend="security.database.storage">the first example</link>. As the examples shows, quotes burnt into the static part of the query is not enough, and can be easily cracked. </simpara> </listitem> <listitem> <simpara> Do not print out any database specific information, especially about the schema, by fair means or foul. See also <link linkend="security.errors">Error Reporting</link> and <link linkend="ref.errorfunc">Error Handling and Logging Functions</link>. </simpara> </listitem> <listitem> <simpara> You may use stored procedures and previously defined cursors to abstract data access so that users do not directly access tables or views, but this solution has another impacts. </simpara> </listitem> </itemizedlist> <simpara> Besides these, you benefit from logging queries either within your script or by the database itself, if it supports logging. Obviously, the logging is unable to prevent any harmful attempt, but it can be helpful to trace back which application has been circumvented. The log is not useful by itself, but through the information it contains. More detail is generally better than less. </simpara> </sect3> </sect2> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/errors.xml?r=1.1&p=1 Index: phpdoc/en/security/errors.xml +++ phpdoc/en/security/errors.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.errors"> <title>Error Reporting</title> <para> With PHP security, there are two sides to error reporting. One is beneficial to increasing security, the other is detrimental. </para> <para> A standard attack tactic involves profiling a system by feeding it improper data, and checking for the kinds, and contexts, of the errors which are returned. This allows the system cracker to probe for information about the server, to determine possible weaknesses. For example, if an attacker had gleaned information about a page based on a prior form submission, they may attempt to override variables, or modify them: <example> <title>Attacking Variables with a custom HTML page</title> <programlisting role="php"> <![CDATA[ <form method="post" action="attacktarget?username=badfoo&password=badfoo"> <input type="hidden" name="username" value="badfoo" /> <input type="hidden" name="password" value="badfoo" /> </form> ]]> </programlisting> </example> </para> <para> The PHP errors which are normally returned can be quite helpful to a developer who is trying to debug a script, indicating such things as the function or file that failed, the PHP file it failed in, and the line number which the failure occured in. This is all information that can be exploited. It is not uncommon for a php developer to use <function>show_source</function>, <function>highlight_string</function>, or <function>highlight_file</function> as a debugging measure, but in a live site, this can expose hidden variables, unchecked syntax, and other dangerous information. Especially dangerous is running code from known sources with built-in debugging handlers, or using common debugging techniques. If the attacker can determine what general technique you are using, they may try to brute-force a page, by sending various common debugging strings: <example> <title>Exploiting common debugging variables</title> <programlisting role="php"> <![CDATA[ <form method="post" action="attacktarget?errors=Y&amp;showerrors=1&amp;debug=1"> <input type="hidden" name="errors" value="Y" /> <input type="hidden" name="showerrors" value="1" /> <input type="hidden" name="debug" value="1" /> </form> ]]> </programlisting> </example> </para> <para> Regardless of the method of error handling, the ability to probe a system for errors leads to providing an attacker with more information. </para> <para> For example, the very style of a generic PHP error indicates a system is running PHP. If the attacker was looking at an .html page, and wanted to probe for the back-end (to look for known weaknesses in the system), by feeding it the wrong data they may be able to determine that a system was built with PHP. </para> <para> A function error can indicate whether a system may be running a specific database engine, or give clues as to how a web page or programmed or designed. This allows for deeper investigation into open database ports, or to look for specific bugs or weaknesses in a web page. By feeding different pieces of bad data, for example, an attacker can determine the order of authentication in a script, (from the line number errors) as well as probe for exploits that may be exploited in different locations in the script. </para> <para> A filesystem or general PHP error can indicate what permissions the webserver has, as well as the structure and organization of files on the web server. Developer written error code can aggravate this problem, leading to easy exploitation of formerly "hidden" information. </para> <para> There are three major solutions to this issue. The first is to scrutinize all functions, and attempt to compensate for the bulk of the errors. The second is to disable error reporting entirely on the running code. The third is to use PHP's custom error handling functions to create your own error handler. Depending on your security policy, you may find all three to be applicable to your situation. </para> <para> One way of catching this issue ahead of time is to make use of PHP's own <function>error_reporting</function>, to help you secure your code and find variable usage that may be dangerous. By testing your code, prior to deployment, with E_ALL, you can quickly find areas where your variables may be open to poisoning or modification in other ways. Once you are ready for deployment, by using E_NONE, you insulate your code from probing. <example> <title>Finding dangerous variables with E_ALL</title> <programlisting role="php"> <![CDATA[ <?php if ($username) { // Not initialized or checked before usage $good_login = 1; } if ($good_login == 1) { // If above test fails, not initialized or checked before usage readfile ("/highly/sensitive/data/index.html"); } ?> ]]> </programlisting> </example> </para> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/filesystem.xml?r=1.1&p=1 Index: phpdoc/en/security/filesystem.xml +++ phpdoc/en/security/filesystem.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.filesystem"> <title>Filesystem Security</title> <simpara> PHP is subject to the security built into most server systems with respect to permissions on a file and directory basis. This allows you to control which files in the filesystem may be read. Care should be taken with any files which are world readable to ensure that they are safe for reading by all users who have access to that filesystem. </simpara> <simpara> Since PHP was designed to allow user level access to the filesystem, it's entirely possible to write a PHP script that will allow you to read system files such as /etc/passwd, modify your ethernet connections, send massive printer jobs out, etc. This has some obvious implications, in that you need to ensure that the files that you read from and write to are the appropriate ones. </simpara> <simpara> Consider the following script, where a user indicates that they'd like to delete a file in their home directory. This assumes a situation where a PHP web interface is regularly used for file management, so the Apache user is allowed to delete files in the user home directories. </simpara> <para> <example> <title>Poor variable checking leads to....</title> <programlisting role="php"> <![CDATA[ <?php // remove a file from the user's home directory $username = $_POST['user_submitted_name']; $homedir = "/home/$username"; $file_to_delete = "$userfile"; unlink ("$homedir/$userfile"); echo "$file_to_delete has been deleted!"; ?> ]]> </programlisting> </example> Since the username is postable from a user form, they can submit a username and file belonging to someone else, and delete files. In this case, you'd want to use some other form of authentication. Consider what could happen if the variables submitted were "../etc/" and "passwd". The code would then effectively read: <example> <title>... A filesystem attack</title> <programlisting role="php"> <![CDATA[ <?php // removes a file from anywhere on the hard drive that // the PHP user has access to. If PHP has root access: $username = "../etc/"; $homedir = "/home/../etc/"; $file_to_delete = "passwd"; unlink ("/home/../etc/passwd"); echo "/home/../etc/passwd has been deleted!"; ?> ]]> </programlisting> </example> There are two important measures you should take to prevent these issues. <itemizedlist> <listitem> <simpara> Only allow limited permissions to the PHP web user binary. </simpara> </listitem> <listitem> <simpara> Check all variables which are submitted. </simpara> </listitem> </itemizedlist> Here is an improved script: <example> <title>More secure file name checking</title> <programlisting role="php"> <![CDATA[ <?php // removes a file from the hard drive that // the PHP user has access to. $username = $_SERVER['REMOTE_USER']; // using an authentication mechanisim $homedir = "/home/$username"; $file_to_delete = basename("$userfile"); // strip paths unlink ($homedir/$file_to_delete); $fp = fopen("/home/logging/filedelete.log","+a"); //log the deletion $logstring = "$username $homedir $file_to_delete"; fputs ($fp, $logstring); fclose($fp); echo "$file_to_delete has been deleted!"; ?> ]]> </programlisting> </example> However, even this is not without it's flaws. If your authentication system allowed users to create their own user logins, and a user chose the login "../etc/", the system is once again exposed. For this reason, you may prefer to write a more customized check: <example> <title>More secure file name checking</title> <programlisting role="php"> <![CDATA[ <?php $username = $_SERVER['REMOTE_USER']; // using an authentication mechanisim $homedir = "/home/$username"; if (!ereg('^[^./][^/]*$', $userfile)) die('bad filename'); //die, do not process if (!ereg('^[^./][^/]*$', $username)) die('bad username'); //die, do not process //etc... ?> ]]> </programlisting> </example> </para> <para> Depending on your operating system, there are a wide variety of files which you should be concerned about, including device entries (/dev/ or COM1), configuration files (/etc/ files and the .ini files), well known file storage areas (/home/, My Documents), etc. For this reason, it's usually easier to create a policy where you forbid everything except for what you explicitly allow. </para> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/general.xml?r=1.1&p=1 Index: phpdoc/en/security/general.xml +++ phpdoc/en/security/general.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.general"> <title>General considerations</title> <simpara> A completely secure system is a virtual impossibility, so an approach often used in the security profession is one of balancing risk and usability. If every variable submitted by a user required two forms of biometric validation (such as a retinal scan and a fingerprint), you would have an extremely high level of accountability. It would also take half an hour to fill out a fairly complex form, which would tend to encourage users to find ways of bypassing the security. </simpara> <simpara> The best security is often inobtrusive enough to suit the requirements without the user being prevented from accomplishing their work, or over-burdening the code author with excessive complexity. Indeed, some security attacks are merely exploits of this kind of overly built security, which tends to erode over time. </simpara> <simpara> A phrase worth remembering: A system is only as good as the weakest link in a chain. If all transactions are heavily logged based on time, location, transaction type, etc. but the user is only verified based on a single cookie, the validity of tying the users to the transaction log is severely weakened. </simpara> <simpara> When testing, keep in mind that you will not be able to test all possibilities for even the simplest of pages. The input you may expect will be completely unrelated to the input given by a disgruntled employee, a cracker with months of time on their hands, or a housecat walking across the keyboard. This is why it's best to look at the code from a logical perspective, to discern where unexpected data can be introduced, and then follow how it is modified, reduced, or amplified. </simpara> <simpara> The Internet is filled with people trying to make a name for themselves by breaking your code, crashing your site, posting inappropriate content, and otherwise making your day interesting. It doesn't matter if you have a small or large site, you are a target by simply being online, by having a server that can be connected to. Many cracking programs do not discern by size, they simply trawl massive IP blocks looking for victims. Try not to become one. </simpara> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/globals.xml?r=1.1&p=1 Index: phpdoc/en/security/globals.xml +++ phpdoc/en/security/globals.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.registerglobals"> <title>Using Register Globals</title> <para> Perhaps the most controversial change in PHP is when the default value for the PHP directive <link linkend="ini.register-globals"> register_globals</link> went from ON to OFF in PHP <ulink url="&url.php.release4.2.0;">4.2.0</ulink>. Reliance on this directive was quite common and many people didn't even know it existed and assumed it's just how PHP works. This page will explain how one can write insecure code with this directive but keep in mind that the directive itself isn't insecure but rather it's the misuse of it. </para> <para> When on, register_globals will inject (poison) your scripts will all sorts of variables, like request variables from HTML forms. This coupled with the fact that PHP doesn't require variable initialization means writing insecure code is that much easier. It was a difficult decision, but the PHP community decided to disable this directive by default. When on, people use variables yet really don't know for sure where they come from and can only assume. Internal variables that are defined in the script itself get mixed up with request data sent by users and disabling register_globals changes this. Let's demonstrate with an example misuse of register_globals: </para> <para> <example> <title>Example misuse with register_globals = on</title> <programlisting role="php"> <![CDATA[ <?php // define $authorized = true only if user is authenticated if (authenticated_user()) { $authorized = true; } // Because we didn't first initialize $authorized as false, this might be // defined through register_globals, like from GET auth.php?authorized=1 // So, anyone can be seen as authenticated! if ($authorized) { include "/highly/sensitive/data.php"; } ?> ]]> </programlisting> </example> </para> <para> When register_globals = on, our logic above may be compromised. When off, <varname>$authorized</varname> can't be set via request so it'll be fine, although it really is generally a good programming practice to initialize variables first. For example, in our example above we might have first done <literal>$authorized = false</literal>. Doing this first means our above code would work with register_globals on or off as users by default would be unauthorized. </para> <para> Another example is that of <link linkend="ref.session">sessions</link>. When register_globals = on, we could also use <varname>$username</varname> in our example below but again you must realize that <varname>$username</varname> could also come from other means, such as GET (through the URL). </para> <para> <example> <title>Example use of sessions with register_globals on or off</title> <programlisting role="php"> <![CDATA[ <?php // We wouldn't know where $username came from but do know $_SESSION is // for session data if (isset($_SESSION['username'])) { echo "Hello <b>{$_SESSION['username']}</b>"; } else { echo "Hello <b>Guest</b><br />"; echo "Would you like to login?"; } ?> ]]> </programlisting> </example> </para> <para> It's even possible to take preventative measures to warn when forging is being attempted. If you know ahead of time exactly where a variable should be coming from, you can check to see if the submitted data is coming from an inappropriate kind of submission. While it doesn't guarantee that data has not been forged, it does require an attacker to guess the right kind of forging. If you don't care where the request data comes from, you can use <varname>$_REQUEST</varname> as it contains a mix of GET, POST and COOKIE data. See also the manual section on using <link linkend="language.variables.external">variables from outside of PHP</link>. </para> <para> <example> <title>Detecting simple variable poisoning</title> <programlisting role="php"> <![CDATA[ <?php if (isset($_COOKIE['MAGIC_COOKIE'])) { // MAGIC_COOKIE comes from a cookie. // Be sure to validate the cookie data! } elseif (isset($_GET['MAGIC_COOKIE']) || isset($_POST['MAGIC_COOKIE'])) { mail("admin@example.com", "Possible breakin attempt", $_SERVER['REMOTE_ADDR']); echo "Security violation, admin has been alerted."; exit; } else { // MAGIC_COOKIE isn't set through this REQUEST } ?> ]]> </programlisting> </example> </para> <para> Of course, simply turning off register_globals does not mean your code is secure. For every piece of data that is submitted, it should also be checked in other ways. Always validate your user data and initialize your variables! To check for unitialized variables you may turn up <function>error_reporting</function> to show <constant>E_NOTICE</constant> level errors. </para> &note.superglobals; </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/hiding.xml?r=1.1&p=1 Index: phpdoc/en/security/hiding.xml +++ phpdoc/en/security/hiding.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.hiding"> <title>Hiding PHP</title> <para> In general, security by obscurity is one of the weakest forms of security. But in some cases, every little bit of extra security is desirable. </para> <para> A few simple techniques can help to hide PHP, possibly slowing down an attacker who is attempting to discover weaknesses in your system. By setting expose_php = off in your &php.ini; file, you reduce the amount of information available to them. </para> <para> Another tactic is to configure web servers such as apache to parse different filetypes through PHP, either with an &htaccess; directive, or in the apache configuration file itself. You can then use misleading file extensions: <example> <title>Hiding PHP as another language</title> <programlisting role="apache-conf"> <![CDATA[ # Make PHP code look like other code types AddType application/x-httpd-php .asp .py .pl ]]> </programlisting> </example> Or obscure it completely: <example> <title>Using unknown types for PHP extensions</title> <programlisting role="apache-conf"> <![CDATA[ # Make PHP code look like unknown types AddType application/x-httpd-php .bop .foo .133t ]]> </programlisting> </example> Or hide it as HTML code, which has a slight performance hit because all HTML will be parsed through the PHP engine: <example> <title>Using HTML types for PHP extensions</title> <programlisting role="apache-conf"> <![CDATA[ # Make all PHP code look like HTML AddType application/x-httpd-php .htm .html ]]> </programlisting> </example> For this to work effectively, you must rename your PHP files with the above extensions. While it is a form of security through obscurity, it's a minor preventative measure with few drawbacks. </para> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 --> http://cvs.php.net/co.php/phpdoc/en/security/intro.xml?r=1.1&p=1 Index: phpdoc/en/security/intro.xml +++ phpdoc/en/security/intro.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> http://cvs.php.net/co.php/phpdoc/en/security/variables.xml?r=1.1&p=1 Index: phpdoc/en/security/variables.xml +++ phpdoc/en/security/variables.xml <?xml version="1.0" encoding="iso-8859-1"?> <!-- $Revision: 1.1 $ --> <!-- splitted from ./index.xml, last change in rev 1.66 --> <sect1 id="security.variables"> <title>User Submitted Data</title> <para> The greatest weakness in many PHP programs is not inherent in the language itself, but merely an issue of code not being written with security in mind. For this reason, you should always take the time to consider the implications of a given piece of code, to ascertain the possible damage if an unexpected variable is submitted to it. <example> <title>Dangerous Variable Usage</title> <programlisting role="php"> <![CDATA[ <?php // remove a file from the user's home directory... or maybe // somebody else's? unlink ($evil_var); // Write logging of their access... or maybe an /etc/passwd entry? fputs ($fp, $evil_var); // Execute something trivial.. or rm -rf *? system ($evil_var); exec ($evil_var); ?> ]]> </programlisting> </example> You should always carefully examine your code to make sure that any variables being submitted from a web browser are being properly checked, and ask yourself the following questions: <itemizedlist> <listitem> <simpara> Will this script only affect the intended files? </simpara> </listitem> <listitem> <simpara> Can unusual or undesirable data be acted upon? </simpara> </listitem> <listitem> <simpara> Can this script be used in unintended ways? </simpara> </listitem> <listitem> <simpara> Can this be used in conjunction with other scripts in a negative manner? </simpara> </listitem> <listitem> <simpara> Will any transactions be adequately logged? </simpara> </listitem> </itemizedlist> By adequately asking these questions while writing the script, rather than later, you prevent an unfortunate re-write when you need to increase your security. By starting out with this mindset, you won't guarantee the security of your system, but you can help improve it. </para> <para> You may also want to consider turning off register_globals, magic_quotes, or other convenience settings which may confuse you as to the validity, source, or value of a given variable. Working with PHP in error_reporting(E_ALL) mode can also help warn you about variables being used before they are checked or initialized (so you can prevent unusual data from being operated upon). </para> </sect1> <!-- Keep this comment at the end of the file Local variables: mode: sgml sgml-omittag:t sgml-shorttag:t sgml-minimize-attributes:nil sgml-always-quote-attributes:t sgml-indent-step:1 sgml-indent-data:t indent-tabs-mode:nil sgml-parent-document:nil sgml-default-dtd-file:"../../manual.ced" sgml-exposed-tags:nil sgml-local-catalogs:nil sgml-local-ecat-files:nil End: vim600: syn=xml fen fdm=syntax fdl=2 si vim: et tw=78 syn=sgml vi: ts=1 sw=1 -->