Return to SecurityGuidelinesForDevelopers.html CVS log

Up to [Pegasus] / pegasus / doc

BUG#: 1336
TITLE: Request userid not propagated for provider upcalls

DESCRIPTION: Added additional security considerations to the Security Guidelines for Developers document.

<html>
<head>
<meta http-equiv="Content-Language" content="en-us">
<meta http-equiv="Content-Type" content="text/html; charset=windows-1252">
<title>PEP</title>
<link rel="stylesheet" href="http://www.opengroup.org/stylesheets/info1+margins.css">
<style><!--
-->
</style>
</head>
<body bgcolor="#FFFFFF">
<h1> <font color="#000000">Pegasus Security Implementation Guidelines</font></h1>
<hr>
<h2><font color="#000000">Problem Statement</font></h2>
<p><font color="#000000">Open Pegasus has a challenging role. It provides a portal 
  for users and programs to access a wide variety of information on a system. 
  Pegasus is responsible for user authentication and provides a framework for 
  the provider authors to authorize specific read and write operations on the 
  server. Managing the resulting &quot;trust delta&quot; (the difference between 
  what the provider <em>could </em>do in its current execution context, vs. what 
  a given user is <em>authorized</em> to do) is hard. The bigger the trust delta, 
  the greater the incentive to &quot;break in&quot; past authorizations in providers 
  to &quot;get to&quot; a super-user/administrator execution context(or to the 
  context of a user that can do something the authenticated user isn't authorized 
  to do). Though the OpenPegasus 2.5 feature, <strong>&quot;run-as-requestor,&quot;</strong> 
  does provide a way to lower the risk to a given provider that takes advantage 
  of the run-as-requestor context, there are still risks for the providers that 
  decide to run at elevated privilege (defined as when the execution context has 
  more permissions/abilities than the authorized users... hard to avoid when the 
  execution is not <em>as</em> the authorized user).</font></p>
<p><font color="#008000">Pre-2.5 implementations of OpenPegasus do not change 
  the user-context (effective-user) of the provider from that of the CIMOM. Thus, 
  when Pegasus is used on the majority of platforms (that don't provide intra-thread 
  protections and/or work in a multi-user environment) a provider can, through 
  mistake or intent, easily affect the security of the CIMOM and other providers. 
  </font></p>
<h2><font color="#000000">Requirements, Constraints or Assumptions</font></h2>
<p>Most of the risk of failing to follow the guidelines below are only present 
  when code is not run as the authenticated user (whether that code be in the 
  provider or the client) or in deployments for which the concept of authentication 
  isn't used (i.e.: SNMP-public info). Even for more limited deployments, many 
  of the problems below can still cause potential crashes/denial of service conditions.</p>
<h2><font color="#000000">Definitions:</font></h2>
<p>Elevated code: a difference between the actions that the logged in user is 
  authorized to perform, and the execution context of the running program.&nbsp; 
  This &quot;trust delta&quot; must then be managed by the code to ensure that 
  the user doesn't perform more actions than they are authorized either directly 
  or through side-effect.&nbsp; For example, a process run as the UID of the authenticated 
  user is thus said to be &quot;non-elevated.&quot;&nbsp; A process running as 
  administrator on behalf of a &quot;non-administrator&quot; user, would be called 
  elevated.</p>
<p>Privilege: The collection of actions a process or user is not prevented from 
  doing.&nbsp; An &quot;administrator&quot;/root user is said to be full privilege 
  with respect to a system since that execution context does not prevent any action 
  on the system.</p>
<p>Trust: The degree to which an actor that is interacting with&nbsp; the component 
  under consideration is believed to behave non-maliciously.&nbsp; For example, 
  an arbitrary user on the Internet has no 'trust,&quot; a junior operator or 
  administration is trusted to no attempt malicious activity, but may accidentally 
  attempt damaging actions.&nbsp; Root/Administrator code is trusted to behave 
  correctly.</p>
<p>Security Testing: Non-functional testing that centers around behavior in the 
  presence of malicious use.&nbsp; Examples includes testing for crashes or security 
  side-effects in the presence of overly long inputs, special character inputs, 
  high-system load, and network storm environments.</p>
<p>Security Side Effect: Applications often, in accomplishing their goals, perform 
  actions beyond those visible to the user.&nbsp; Examples include writing temporary 
  files, or clearing or requesting memory.&nbsp; Since this behavior is not specified 
  in the functional requirements, it is often not tested.&nbsp; This &quot;side 
  effect&quot; behavior is often the behavior that a malicious user attempts to 
  leverage when trying to gain privilege.&nbsp; Examples include exploiting race 
  conditions where a temporary file is momentarily world writeable, before it 
  is chmod-ed.&nbsp; This window is an opportunity for a malicious user to insert 
  data that can change the behavior of the application.</p>
<h2><font color="#000000">References:</font></h2>
<p>Architecture: <a href="http://www.opengroup.org/security/secarch.htm"> http://www.opengroup.org/security/secarch.htm</a></p>
<p>Books and References (not endorsed by Opengroup or partners): </p>
<ul>
  <li><a href="http://www.oreilly.com/catalog/securecdng/"> http://www.oreilly.com/catalog/securecdng/</a></li>
  <li><a href="http://www.oreilly.com/catalog/puis3/"> http://www.oreilly.com/catalog/puis3/</a></li>
  <li><a href="http://www.securityfocus.com/infocus/1596"> http://www.securityfocus.com/infocus/1596</a> 
  </li>
</ul>
<h2><font color="#000000">Proposed Solution </font></h2>
<h3><font color="#000000">General Implementation Guidelines: </font></h3>
<h4><font color="#000000">Code that doesn't adhere to the following guidelines 
  in elevated code should be considered a bug, including providers not running 
  as-requestor. it is a best-practice to follow the following guidelines for all 
  code. </font></h4>
<ol>
  <li> <u>Avoid buffer overflow vulnerabilities</u> in your code (hackers use 
    to insert arbitrary code) <br>
    Buffer overflows in network-accessible software cause the most, and some believe, 
    the majority, of software vulnerabilities. Since Pegasus is written in C/C++, 
    it is especially susceptible to buffer overflows. Strongly consider using 
    a static tool like Flawfinder or RATS to look for common problems. Dynamic 
    tools can also be used but only identify problems when the overflow actually 
    happens vs. finding potential overflows. The susceptibility stems from a lack 
    of bounds checks in C++ and C. Problematic functions include strcpy, sprintf, 
    strcat, gets, and strlcat.</li>
  <ul>
    <li> References:<br>
      <ul>
        <li> Smashing the stack for fun and profit, <a href="http://www.phrack.com/show.php?p=49">http://www.phrack.com/show.php?p=49</a> 
        <br>
        <li> Heap Overflows: <a href="http://www.phrack.org/phrack/57/p57-0x08%20">http://www.phrack.org/phrack/57/p57-0x08</a> 
        <br>
      </ul>
    </li>
  </ul>
  <li> <u>Avoid format string vulnerabilities</u> in your code (hackers use to 
    read or insert arbitrary code)<br>
    Format strings define the format and types of program variables that are substituted 
    into an input or output string. Exploitation of format strings occurs when 
    functions that require a format string are coded with a variable, and that 
    variable is not validated. For example the following is vulnerable code: printf(string_from_untrusted_user) 
    as the user can supply the format string, and read or overwrite (using %n) 
    arbitrary data. Each developer must: </li>
  <ul>
    <li> Use functions with a strict format string argument.</li>
    <li> Check the input parameter data for format strings before assigning them 
      to variables.</li>
    <li>Always validate that the input parameter data does not contain any program-specific 
      format characters before assigning the input data to variables.</li>
    <li> Always check the return codes of library functions for failure.<br>
    </li>
  </ul>
  <li><u>Adhere to the general, good programming practices</u>: 
    <ol>
      <li> Always have people other than the coder review the code.</li>
      <li> Always have people other than the coder develop tests and test the 
        end product.</li>
      <li> Check return codes from system or library calls, and handle errors 
        or exceptions gracefully.</li>
      <li>Keep your code simple: simple code decreases the risk of defects; complex 
        code increases the risk of defects.</li>
      <li> Don't use uninitialized variables.</li>
      <li> Use symbolic constants (such as #define) to minimize typos and improve 
        code maintainability.</li>
      <li> Use temporary files with care. Do not create temporary directories 
        or files from your program that are world writeable. Limit the permissions 
        to what is needed by the program. Clean up temporary files or directories 
        when you are finished using them.</li>
      <li>Do not put sensitive information in log files. For example, do not print 
        social security numbers, passwords, credit card numbers, or any other 
        sensitive or personal information for debugging purposes in the log files. 
        Sometimes such information shows up in the web browser in case of exceptions 
        or application failure.</li>
      <li> Enforce strong password policies and a delay on failed logins. This 
        helps to prevent unauthorized access to private data. See libpam for a 
        good way to implement this.</li>
      <li> 
        <p>Validate input to the program or system before processing the input. 
          Test to see if the input is the proper type of data and in the range 
          of acceptable or expected values and test both upper and lower bounds. 
          For example, if you are reading in a year value (int), and you have 
          already checked for buffer overflow and format strings:</p>
        <blockquote> 
          <p>Correct: if 0 &lt;= year &lt;= 3000 then (accept input and process)</p>
          <p>Unsafe: if year &lt;= 3000 then (accept input and process)</p>
          <p>The vulnerability of the unsafe example is that if someone were to 
            return a value of -32769, then they could intentionally stop or corrupt 
            a procedure. If the language which you are using does not enforce 
            strong types, then a type check should also be performed before accepting 
            the input.</p>
        </blockquote>
      </li>
    </ol>
  </li>
  <li> <u>Use the principles of least and necessary privilege</u>.<br>
    Only grant the minimum set of privileges required to perform an operation, 
    and grant these privileges for the minimum required amount of time. For example, 
    if a provider needs to modify both mail queues and print spools, don't run 
    the application as root; instead, use /etc/logingroup or other facilities 
    to give the application the privileges which it needs, but not more privileges 
    than it needs. <br>
  </li>
  <li> <u>Use SSL securely</u>: 
    <p> Please refer to the OpenSSL documentation for usage. Pegasus libraries 
      help with some but not all functions necessary for certificate management 
      and usage.</li>
  <li> <u>Handle race conditions securely</u><br>
    Race conditions occur when two or more processes access a shared resource 
    in an order that was not expected by the program. Unordered access to resources 
    is common in multitasking environments, and is mostly associated with either 
    Signal Handlers or File Handling. For example, if a program &quot;A&quot; 
    checks to see if a file exists before writing, but a program &quot;B&quot; 
    creates a link after the check, but before the write, &quot;A&quot; may inadvertently 
    overwrite the link destination with the permissions associated with &quot;A&quot;. 
    This can be a security problem if &quot;A&quot; has different permissions 
    than &quot;B.&quot;</li>
  <li> <u>Use secure defaults when possible, or clearly document when they aren't 
    used</u> </li>
  <li> <u>Design Securely</u><br>
    Design your code so that as little as possible runs as a privileged user. 
    All privileged user code (especially if it listens on a network or executes 
    on behalf of other users) should be inspected very thoroughly, so it should 
    be short and simple. Each module of code should have a clean interface for 
    other modules to use and a well-defined perimeter around each module.&nbsp; 
    Additional details can be found at: <a href="http://www.joeyoder.com/papers/patterns/Security/appsec.pdf"> 
    Architectural Patterns for Enabling Application Security</a> <font size="-4">(http://www.joeyoder.com/papers/patterns/Security/appsec.pdf)</font></li>
  <li> <u>Test for security </u>(use both positive and negative tests)<br>
    &quot;Positive tests&quot;verify that the functionality of the product works as specified. 
    &quot;Negative tests&quot;attempt to subvert the security of the system, and are often 
    overlooked when testing software. Spend some time thinking like a hacker and 
    trying to break your system. Always test boundary conditions or corner cases 
    for values of data, size of data, and type of data. Many common bugs are related 
    to this. Sometimes this type of bug may result in wrong information being 
    retrieved from the database instead of failing gracefully. Attempt to exploit 
    the system with buffer overflow and format string attacks.</li>
  <li> <u>Don't bundle private copies of security code</u><br>
    Security code (especially highly scrutinized open-source code) is likely to 
    have security bulletins issued against it. When such security bulletins are 
    inevitably issued against code you depend on, you don't want to have to issue 
    a bulletin against your product also. If you put a dependency in your code 
    to a standard distribution of a component which you need (for example OpenSSL), 
    rather than embedding a private copy, then whenever a security bulletin is 
    issued against it, you won't have to reissue the bulletin after repacking 
    the fix for your private copy. <br>
  </li>
</ol>
<h4><font color="#000000">General Coding Best-Practices:</font></h4>
<ol>
  <li> <u>Avoid implementing security functionality</u>: Making security claims 
    in your documentation (beyond the implied security claims of authentication 
    and authorization done by the operating system) can increase your risk of 
    having a security defect. This is because any of those claims that are not 
    fully implemented or enforced is by definition a security defect and requires 
    an expedited fix and a security bulletin to announce that fix. Reuse of tried-and-tested 
    code, that has been used in a security context is always a better choice. 
    Never implement a random number generator or cryptographic algorithm unless 
    you're a cryptographer by profession. You will almost certainly get it wrong.<br>
    &nbsp;&nbsp;&nbsp; You should always, however, document your security behavior. 
  </li>
  <li><u>Duplicating authorization code</u>: Related to risk #1, every provider 
    has the risk of authorization related defects because the authorization done 
    in each provider is a duplication of the kernel authorization code. However, 
    you can still decrease your risk by using common API's. For example, many 
    providers will need a way to tell if the authenticated user should have access 
    to a given file. The code which does this needs to check the user id, group 
    ids of the file in question and all of its parent directories. Any defect 
    in this code could easily be a security defect and would need to be fixed 
    in every copy of that code. For this reason it is imperative that this logic 
    exists in only one place and that your provider uses that copy. Do not try 
    to replicate this complex logic in your own provider, unless you are the single 
    owner of that code. </li>
  <li>WBEM provider/client combinations: Writing a WBEM provider that is also 
    a WBEM client (makes requests of other providers) has security risks/challenges. 
    There are two subcategories of this risk/challenge: 
    <ol>
      <li>using the connectLocal() API uses the UID of the running process to 
        do authentication. Thus, the provider initiating the request must ensure 
        authorization of the other provider's data before making the request. 
        (This is another example of Risk 2, multiple copies of authorization code) 
        One feasible way to do this is to check that the user is a privileged 
        user before calling the other provider (in which case the UID matches 
        the running process)</li>
      <li> Using the connect() API has additional complexities. Credentials must 
        be somehow passed into the provider and then handled appropriately. Also, 
        there are additional client responsibilities as far as certificate validation 
        and testing, and the consequences are more severe because the client is 
        running with elevated privileges</li>
    </ol>
  </li>
</ol>
<h3> <font color="#000000">Provider Implementation Guidelines&nbsp; </font></h3>
<h4> <font color="#000000">Code that doesn't following the following guidelines 
  in providers running at elevated-privilege should be considered a bug.<br>
  Code in providers running as-requestor can consider the following as general 
  best practice.</font></h4>
<ol>
  <li> <u>Check the username/uid and execute every method as if it was running 
    as that user</u> (i.e. had the OS kernel or authorization service done the 
    authorization).<br>
    By checking each operation they perform, and ensuring those operations, when 
    performed on behalf of a non-privileged user, do not have security side-effects. 
    Any discrepancy between OS authorizations done by the kernel and that done 
    by the provider that is not part of documented behavior is a security defect. 
    If the user does not have the privileges to perform the requested operation 
    the Provider must throw CIMAccessDeniedException.</li>
  <li> <u>Keep your design/provider simple.</u> <br>
    While this is difficult to quantify, it is important to minimize the amount 
    of code running as a privileged user. As a general guideline, if you have 
    significant lines of code running with elevated privilege, the likelihood 
    of a security defect is high. Remember that defects in elevated privilege 
    code is a potential security defect, so all of this code must be straightforward 
    and easy to review based on the principles mentioned in the General Coding 
    principles above. The likelihood of a defect not being found is proportional 
    to the amount and complexity of the code. </li>
  <li><u>Provider must not use any calls such as setuid or setting environment 
    variables (i.e. PATH) that would alter the state of the process running the 
    CIM Server.</u> <br>
    This could cause unexpected results for other providers or threads.</li>
  <li><u>Provider must document property authorizations</u>. <br>
    Specifically, the provider should describe which data elements they make available 
    for reading, which system changes they are capable of making, and which users 
    will be able to read those elements and make those changes.</li>
  <li><u>Provider must check all untrusted input for validity.</u> <br>
    While the CIM Server ensures that the input is a valid CIM request, the provider 
    is responsible for validating that the CIM request does not cause any side 
    effects by ensuring that the input strings contain only expected characters 
    and that values are within an expected range. Examples of input data that 
    must be checked include directory or file names, data within files that are 
    read by the provider, and data returned from system calls.</li>
  <li><u>Provider must execute stress tests</u>.<br>
    These include operation in the presence of multiple interacting provider requests. 
    Based on a white box analysis of your provider, identify ways in which testing 
    could stress your provider. For example, sending large input strings, a large 
    number of simultaneous requests, requests including out-of-bounds data, or 
    ensuring that every branch is covered are just a few ways that you could stress 
    your product to find potential defects. By exploring the way your provider 
    fails, you can look for side effects that might lead to &quot;infinite&quot; 
    resource requests, overwritten data, or other anomalies that could cause a 
    denial-of-service or reveal a side-effect that can be leveraged as an exploit.</li>
  <li><u>Design your provider to expect belligerent input</u>. <br>
    For example, have a common method that validates all CIM requests and ensure 
    that that method gets called for every request. The method should assume that 
    input is invalid unless it matches a specific format and specific bounds are 
    checked. Also, if your provider allocates any memory buffers or writes to 
    any file based on user input, all error conditions (out-of-memory, disk full, 
    file is a symbolic link/device file/directory instead of the expected format, 
    buffer/array too small for data, etc.) should be checked and all of this should 
    be enforced in a common place.</li>
  <li><u>Do not allow group or world-write access </u>to your shared library, 
    any other executable code, configuration files, or any parent directory of 
    any of the above. <br>
    Although only a privileged user ought to be able to create the symbolic links 
    or shortcuts to the provider shared library in the designated WBEM provider 
    library directory, the actual provider shared library can be placed in any 
    directory. A provider must ensure that their shared libraries are protected 
    in such a way that only a privileged user can modify or delete the shared 
    library or the directory where the shared library is located.</li>
  <li><u><font color="#008000">Don't bypass authentication</font></u><font color="#008000">: 
    To avoid damaging CIMOM and other-providers' security, providers should not 
    bypass CIMOM authentication steps in communicating to other providers or the 
    CIMOM (manipulating CIMOM-handle userid) and instead use CIMClient in accordance 
    with &quot;General-Client Best Practices&quot; section. In cases where the 
    Pegasus use-model does rely on platform-native inter-thread protections, a 
    future protection algorithm may have to be implemented in OpenPegasus that 
    ensures an unchanged state of the operation context. Note however that, In 
    Pegasus 2.5, if a provider uses run-as-requestor, that will ensure the provider 
    runs in the right user-context, and cleans up the interface to that provider 
    to ensure it doesn't skip inter-provider authentication. </font><br>
  </li>
</ol>
<h4><br>
  <font color="#000000">Provider Best-Practices:</font></h4>
<ol>
  <li><u>Use &quot;UserContext registration&quot; setting, present in Open Pegasus 
    2.5 and later</u>: <br>
    In Pegasus 2.5 and after, you should strongly consider registering your provider 
    to run as requestor context, or if not available, use Windows &quot;impersonation&quot; 
    or fork a correct-user-running process. For providers in versions prior to 
    2.5, you may want to consider implementing your own out of process provider, 
    to avoid the risks of running at elevated privilege. For those that must run 
    privileged: </li>
  <ul>
    <li><u>Check that the authenticated username provided by CIM matches the effective 
      user id</u> of the running process. For Pegasus 2.4 and prior, this means 
      that only the privileged user would be able to use your provider. The general 
      property is that if you are not elevating privilege (running on behalf of 
      a different user), then the likelihood of a security defect is greatly decreased. 
      Making your code more general may mean less work in the future when non-privilege-elevated 
      providers are able to run with the correct user-id.&nbsp; Even in the model 
      where Pegasus is run under a non-privileged user, there is a delta in &quot;trust&quot; 
      between the different users.&nbsp; This still represents some, though not 
      as much, risk as deploying a run-as-administrator Pegasus.&nbsp; There is 
      an opportunity to improve Pegasus to better support fully protecting this 
      use model, though this is less urgent than protecting the higher risk associated 
      with an administrator running CIM server.</li>
    <li><u>Recommend configuring the WBEM users group (ref: PEP 142)</u>: For 
      Pegasus versions prior to 2.5, and subsequent, customers can configure a 
      specific group of users who has access to WBEM providers. This allows customers 
      to choose a tradeoff between security risk and ease-of-setup. Since every 
      provider runs with elevated privilege, the risk of security defects is high. 
      Thus, it is advised that customers configure this group of WBEM users to 
      only allow access to users who are trusted not to be malicious. If you also 
      do not run by default, this information can be in your initial setup documentation 
      so that it gets to all of your customers. This can greatly decrease your 
      risk of having a security defect, because all malicious activities can be 
      potentially ruled out.</li>
  </ul>
  <li><u>Providers should consider the tradeoff between default installation/registration 
    and optional</u>:&nbsp; An optional installation of a component (as part of 
    an OS or software package) gives customers a choice as to whether or not to 
    limit their interface/exposure, and maintenance/patch burden.&nbsp; Your provider 
    likely meets a real need for many customers, but there are also customers 
    who do not need the functionality you provide. There are many customers who 
    would prefer less patching/update cost and decreased security risk (risk is 
    added whenever there is a new interface) versus the functionality that your 
    product provides. Although technically this doesn't decrease the risk of having 
    a security defect, it can give you more options for interim workarounds until 
    you can get a critical fix out, and fewer customers would be affected by any 
    given defect.&nbsp; Provider writers and bundlers should consider these benefits 
    and weigh those against the bundling benefits of mandatory inclusion.</li>
  <li><u>Log important events, such as unauthorized requests</u>: This can help 
    a customer track down a potential intrusion as well as debug problems. Do 
    not include confidential information, such as passwords, in the log. Ensure 
    that the confidentiality of information stored in the log is commensurate 
    with access to the log. It is recommended that you use a common logging facility, 
    such as syslog. Syslogd takes care of things like log rotation, etc. and the 
    administrator already knows where to look for your logs.</li>
  <li><u>When making system changes, use platform security checks where possible 
    vs. rewriting your own authorization code</u>: Duplicating authorization code 
    at least doubles the work and is more error-prone.</li>
</ol>
<h3> <font color="#000000">Client Implementation Guidelines:</font></h3>
<p> Note: In general, these are the responsibility of the applications invoking 
  CIM client libraries to the extent that the client libraries don't yet provide 
  the direct support.</p>
<h4> <font color="#000000">Client code that doesn't follow these guidelines should 
  be considered a bug:</font></h4>
<ol>
  <li><u>Use SSL as<a> follows in your remote production client. Though WBEM does 
    provide libraries to help, client behavior is the client's responsibility:</a></u></li>
  <li><u>Protect the Keystore and Truststore for remote production clients</u>: 
    <ul>
      <li>Use proper file and directory permissions to protect keystore and truststore 
        files. </li>
      <li>If your applications are importing the servers' certificate to a truststore, 
        you must ensure that the user validates the certificates received before 
        adding them to a truststore or keystore.</li>
      <li>Do not use less than 1024 bit keysize to create keystores.</li>
      <li>Keystores/truststores should not be readable or writeable by anyone 
        other than the user who owns them.</li>
    </ul>
  </li>
  <li><u>General programming standards</u> 
    <ul>
      <li>Do not use world-writeable files or directories (including /tmp and 
        /var/tmp).&nbsp; Make sure all credentials (passwords/certificates) are 
        readable only by their owner.</li>
      <li>Do not cache passwords unless directed to do so by the user.&nbsp; The 
        user should be aware that their password is being stored permanently on 
        the client machine.</li>
      <li>Do not pass passwords as an option on the command-line in non-windows 
        clients.&nbsp; Command-lines are visible to all users on the system in 
        some operating systems.</li>
    </ul>
  </li>
</ol>
<h4><font color="#000000">General client best-practices:</font></h4>
<ol>
  <li><u>Limit access to client data</u>: Each user of a WBEM client should have 
    his/her own WBEM client instance.&nbsp; The WBEM client process should run 
    as the correct user on the client machine.</li>
  <li><u>Local vs. Remote Requests and Username/Password Authentication</u>: Use 
    the connectLocal() API call to connect to the CIM server whenever possible.&nbsp; 
    To use this API call properly, the process must run with the correct userid 
    <br>
    <br>
    <font face="Courier">Warning: &nbsp;For Pegasus earlier than 2.5, doing client 
    operations from a CIM provider significantly increases your security risk 
    if the initial client requester was not running as root.&nbsp; This is due 
    to the implementation which runs the provider in the CIM Server process space 
    with a single, often privileged, user so the provider it connects to will 
    be unable to use built-in authentication.&nbsp; Providers issuing WBEM client 
    operations must adequately address the security risk.&nbsp; A few alternatives 
    to address the security concern are: 1) ensure (either at design time or at 
    runtime in the provider) that the user is authorized to access the data being 
    requested from the second provider, and 2) the provider could launch another 
    process and issue the request to the second provider as the intended user.</font> 
    <p>Background on connectLocal():&nbsp; <br>
      A local connection mechanism exists for clients to communicate with the 
      CIM Server on the same system. The connectLocal() function is used for this 
      purpose, and does not take any arguments. In the case where PEGASUS_LOCAL_DOMAIN_SOCKET 
      is defined, (default on all but Windows, as currently the Windows connectLocal 
      authentication is not functional as of 2.5) the user ID passed to the provider 
      is that of the process in which the client program is running. The CIM Server 
      verifies that the user ID of the request is indeed that of the requesting 
      process. Namespace authorization, if enabled, is still performed.&nbsp; 
      When the client must be able to connect to a CIM Server on a remote system, 
      or when it must be able to specify a different user than that of the process, 
      it must use the connect() function. This function allows a hostname and 
      port number to be specified, as well as a username and password.&nbsp; If 
      you need to use the connect() API, the WBEM client has several responsibilities 
      to ensure correct authentication and to protect confidential information.&nbsp;&nbsp; 
      Because connectLocal() does not use SSL, these guidelines only apply to 
      the connect() interface. Using connectLocal() bypasses these requirements 
      except where PEGASUS_LOCAL_DOMAIN_SOCKET is not defined.&nbsp; In that case, 
      it behaves like connect(), using HTTPS and/or HTTP as defined in Pegasus 
      settings.<br>
      &nbsp;</li>
  <li><u>General programming standards</u> 
    <ul>
      <li>Design for belligerent input.&nbsp; A separate module should be responsible 
        for validating all input before taking any action.&nbsp; Invalid input 
        should be discarded.&nbsp; If you client has high availability requirements, 
        deal with invalid input quickly to avoid Denial of Service attacks.</li>
      <li>Use a strongly-typed language if possible (i.e. Java).&nbsp; If your 
        client is in C++, then use a security scanner such as RATS (<a href="http://www.securesoftware.com/resources/download_rats.html">http://www.securesoftware.com/resources/download_rats.html</a>) 
        to identify problem areas and follow the recommendations.&nbsp; (Note: 
        code scanners such as these tend to make a lot of recommendations, so 
        plan on adequate time for manual analysis and focus on your input validation 
        module.)</li>
      <li>Do not use world-writeable files or directories (including /tmp and 
        /var/tmp).&nbsp; Make sure all credentials (passwords/certificates) are 
        readable only by their owner.</li>
      <li>Do not cache passwords unless directed to do so by the user.&nbsp; The 
        user should be aware that their password is being stored permanently on 
        the client machine.</li>
      <li>Do not pass passwords as an option on the command-line on non-windows 
        systems.&nbsp; Command-lines on non-windows systems are visible to all 
        users on the system.</li>
      <li>If possible, do not make any server-initiated changes on the client 
        system.&nbsp; Doing so increases the risk of security vulnerabilities 
        in your client, and a security reviewer should be consulted.</li>
      <li>If possible, log events of interest, including certificate warning messages 
        and invalid responses sent from the server. Doing so increases the ability 
        of a user or system administrator to track down unauthorized actions.&nbsp; 
        Use either a user-specific logfile or syslog.&nbsp; Be sure to check for 
        corner cases like disk-space limitations.</li>
      <li>HTTP Indications should only be used to send confidential information 
        in environments where the risk of exposure to man-in-the-middle type attacks 
        is low (e.g. where a rogue CIM Listener could intercept indications).&nbsp; 
        If your listener expects to receive confidential information, be sure 
        to document that this information will be visible to anyone on the network 
        clearly to the customer initiating the subscription.</li>
    </ul>
  </li>
  <li><u>Security Testing Guidelines</u> 
    <ul>
      <li>Run the following tests, and ensure that your client gives a useful 
        error message and does not crash.&nbsp; Crashes on strange and unexpected 
        input are, at a minimum, a denial-of-service, and often represent buffer 
        or format-string vulnerabilities.: 
        <ul>
          <li>CIM server you are connecting to is not available (disabled or network 
            problems)</li>
          <li>CIM server responds with an extremely large response</li>
          <li>CIM server or provider responds with invalid characters or garbage 
            in the response</li>
          <li>CIM server returns 'access denied'</li>
        </ul>
      </li>
    </ul>
  </li>
</ol>
<h2><font color="#000000">Platform Considerations</font></h2>
<p><font color="#000000
">The coding guidelines may not help, but will not hurt implementations where 
  Pegasus and its providers are not run at elevated privilege. Examples of this 
  include environments with only one user or where Pegasus itself is executed 
  as the requesting user.</font></p>
<hr>
<p><i><font size="2">Copyright (c) 2005 EMC Corporation; Hewlett-Packard Development 
  Company, L.P.; IBM Corp.; The Open Group; VERITAS Software Corporation</font><br>
  <br>
  <font size="1">Permission is hereby granted, free of charge, to any person obtaining 
  a copy&nbsp; of this software and associated documentation files (the &quot;Software&quot;), 
  to deal in the Software without restriction, including without limitation the 
  rights to use, copy, modify, merge, publish, distribute, sublicense, and/or 
  sell copies of the Software, and to permit persons to whom the Software is furnished 
  to do so, subject to the following conditions:</font><br>
  <font size="2"><br>
  </font> <font size="1">THE ABOVE COPYRIGHT NOTICE AND THIS PERMISSION NOTICE 
  SHALL BE INCLUDED IN ALL COPIES OR SUBSTANTIAL PORTIONS OF THE SOFTWARE. THE 
  SOFTWARE IS PROVIDED&nbsp; &quot;AS IS&quot;, WITHOUT WARRANTY OF ANY KIND, 
  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
  FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
  AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, 
  WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR 
  IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.</font></i></p>
<hr>
</body>
</html>