/usr/share/doc/pyro/html/9-security.html is in pyro-doc 1:3.16-2.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 | <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html>
<head>
<meta http-equiv="Content-type" content="text/html;charset=UTF-8">
<title>PYRO - Security</title>
<link rel="stylesheet" type="text/css" href="pyromanual_print.css" media="print">
<link rel="stylesheet" type="text/css" href="pyromanual.css" media="screen">
</head>
<body>
<div class="nav">
<table width="100%">
<tr>
<td align="left"><a href="8-example.html"><previous</a> | <a href="index.html">contents</a> | <a href=
"10-errors.html">next></a></td>
<td align="right">Pyro Manual</td>
</tr>
</table>
<hr></div>
<h2>9. Security</h2>
This chapter discusses the security aspects of Pyro, the features you can use to control
security, and some important warnings.
<ul>
<li><a href="#warning"><em>Imporant Security Warning</em></a></li>
<li><a href="#validator">Authenticating using Connection Validators</a></li>
<li><a href="#customvalidator">Customizing authentication using custom Validator</a></li>
<li><a href="#nameserverplugins">Name Server security plugins</a></li>
<li><a href="#mobile">Mobile objects and Code Validators</a></li>
<li><a href="#firewalls">Firewalls</a></li>
<li><a href="#pickle">The pickle trojan security problem, and XML pickling</a></li>
<li><a href="#ssl">SSL (secure socket layer) support</a></li>
</ul>
<h3><a name="warning" id="warning"></a>Imporant Security Warning</h3><strong>Read this carefully:</strong> Pyro is a
technology that may easily expose private data to the world, if used incorrectly. While Pyro has some security
related functions such as connection validators, it is imporant to understand that exposing a remote object interface
in any way (with Pyro, with XMLRPC, or whatever) on an untrusted network (like the internet) possibly creates a big
security risk. The risk could be because of a hole in Pyro itself or because of security issues in the used libraries
(such as pickle), Python version, or even operating system. <em>Be sure to know what you are doing when using Pyro
outside a trusted network and outside trusted applications! Pyro has never been truly designed to provide a secure
communication mechanism, nor has it had a security review or -test by a security expert. Read the Pyro software
license and <a href="LICENSE">DISCLAIMER</a>.</em>
<h3><a name="validator" id="validator"></a>Authenticating using Connection Validators</h3>To guard against unwanted
or unauthorized connections, Pyro uses so-called <em>new connection validators</em>. These are objects that are
called from the Pyro Daemon to check whether a Pyro server may or may not accept a new connection to that daemon that
a client tries to make. By default, there is only one built-in check; the number of connections was limited to a
certain amount specifed in the PYRO_MAXCONNECTIONS config item). This check is done by the <em>default connection
validator</em> <code>Pyro.protocol.DefaultConnValidator</code>. The fun is that you can supply your own validator
object, and that you can therefore implement much more complex access checks. For instance, you might want to check
if the client's site is authorized to connect. Or perhaps you require a password to connect.
<p>The default validator already supports passphrase protection as authentication validation. This means that a
client that wants to connect to your Pyro server needs to supply a valid authentication passphrase, or the connection
is denied. The check takes place automatically (it is performed by the default connection validator), at connect
time. The following items are important:</p>
<ul>
<li>Unless you tell Pyro otherwise, no authentication check is done. Clients can connect anonymously, unless you
deny access based on max. connections or client host address.</li>
<li>No passphrases are stored by Pyro, not on disk, not in memory. Pyro uses secure hmac-md5 hashes to compare
passphrases.</li>
<li>Server sends an authentication challenge to the client that must be used to generate the actual auth ident
(digest of passphrase+challenge). Because the challenge changes for each connection attempt, and thus the required
auth ident changes, this scheme is safe against eavesdroppers (auth idents are not reusable).</li>
<li>Once a client is accepted and connected, no further checks are done.</li>
<li>The same passphrases are valid for all objects connected to a certain Pyro Daemon.</li>
<li>Clients can use a different passphrase for each object.</li>
<li>There is no config item to specify the passphrase (because this would be very insecure). Code your own
passphrase requester/supplier.</li>
</ul>The Name Server and the Event Server can both be instructed to require authentication too.
<p>Look at the "denyhosts", "authenticate" and "user_passwd_auth" examples
to see how you can use the connection validators.</p>
<h4>How to use the default connection Validator</h4>You can specify the maximum number of connections that Pyro
accepts by setting the <code>PYRO_MAXCONNECTIONS</code> configuration item. This limit is <em>always</em> checked
when a new client connects.
<p>To enable passphrase authentication, you must tell the Pyro Daemon a list of accepted passphrases. Do this by
calling the <code>setAllowedIdentifications(ids)</code> method of the daemon, where <code>ids</code> is a list of
passphrases (strings). If you use <code>None</code> for this, the authentication is disabled again. <em>Note that the
ID list is a shared resource and that you will have to use thread locking if you change it from different
threads.</em> To specify for your client what passphrase to use for a specific object, call the
<code>proxy._setIdentification(id)</code> method of the Pyro proxy, where <code>id</code> is your passphrase
(string). Use <code>Null</code> to disable authentication again. Call the method right after you obtained the proxy
using <code>getProxyForURI</code> or whatever.</p>
<p>If a connection is denied, Pyro will raise a <code>ConnectionDeniedError</code>, otherwise the connection is
granted and your client proxy can invoke any methods it likes, untill disconnected.</p>
<h4>The default SSL connection Validator</h4>
For SSL connections, the <code>Pyro.protocol.BasicSSLValidator</code> is
used by default. This is an extension to the normal validator, it also checks if the client has supplied
a SSL certificate. See the "ssl" example for details.
<h3><a name="customvalidator" id="customvalidator"></a>Customizing authentication using custom Validator</h3>All
authentication logic is contained in the Connection Validator object. By writing your own specialization of the
<code>DefaultConnValidator</code>, you can control all logic that Pyro uses on the client-side and server-side for
authenicating new connections. You are required to make a subclass (specialization) of the default connection
validator <code>Pyro.protocol.DefaultConnValidator</code>. There are two methods that you can use to set your own
validator object:
<dl>
<dt>Client side, on the Proxy:</dt>
<dd><code>_setNewConnectionValidator(validator)</code></dd>
<dt>Server side, on the Daemon:</dt>
<dd><code>setNewConnectionValidator(validator)</code></dd>
</dl>In both cases, you have to pass an instance object of the validator that you want to use. Don't forget that you
still have to use the two methods already mentioned above:
<dl>
<dt>Client side, on the Proxy:</dt>
<dd><code>_setIdentification(ident)</code><br>
<strong>Note:</strong> the ident that you provide doesn't have to be a single string (or passphrase). It can be any
Python object you want, for instance a login/password tuple! It is passed unchanged into the connection validator
(see below) that creates a protocol token from it.</dd>
<dt>Server side, on the Daemon:</dt>
<dd><code>setAllowedIdentifications(idents)</code></dd>
</dl>
The "denyhosts" and "user_passwd_auth" examples show two possible ways to use a custom
connection validator.
<p>Below you see the meaning of the different methods that are used in the connection validator class (and that you
can override in your custom validator):</p>
<dl>
<dt><code>class MyCustomValidator(Pyro.protocol.DefaultConnValidator):</code></dt>
<dd><em>...has to be inherited...</em></dd>
</dl>
<blockquote>
<dl>
<dt><code>def __init__(self):</code></dt>
<dd><code>Pyro.protocol.DefaultConnValidator.__init__(self)</code> <em>...required...</em></dd>
<dt><code>def acceptHost(self,daemon,connection):</code></dt>
<dd><em>...called first, to check the client's origin. Arguments are the current Pyro Daemon, and the connection
object (<code>Pyro.protocol.TCPConnection</code> object). The client's socket address is in
<code>connection.addr</code>. You can check the client's IP address for instance, to see if it is in a trusted
range. The default implementation of this method checks if the number of active connections has not reached the
limit. (<code>Pyro.config.PYRO_MAXCONNECTIONS</code>) <strong>See table below for return codes</strong></em></dd>
<dt><code>def acceptIdentification(self, daemon, connection, token, challenge):</code></dt>
<dd><em>...called to verify the client's identification token (check if the client supplied a correct
authentication passphrase). The arguments are: daemon and connection same as above, client's token object (that
was created by <code>createAuthToken</code> below), server challenge object that was sent to the client. The
default implementation uses <code>createAuthToken</code> to create a secure hash of the auth id plus the
challenge to compare that to the client's token. Effectively, it checks if the client-supplied hash is among the
accepted passphrases of the daemon (hash of passphrase+challenge) -- if any are specified, otherwise it is just
accepted. <strong>See table below for return codes</strong>. NOTE: you can use the <code>connection</code>
argument to store the authentication token (which might be a username). A Pyro object may access this again by
getting the connection object <code>self.getLocalStorage().caller</code> and getting the authentication token
from there.</em></dd>
<dt><code>def createAuthToken(self, authid, challenge, peeraddr, URI, daemon):</code></dt>
<dd><em>...called from both client (proxy) and server (daemon) to create a token that is used to validate the
connection. The arguments are: identification string (comes from <code>mungeIdent</code> below), challenge from
server, socket address of the other party, Pyro URI of the object that is to be accessed, current Pyro Daemon.
When in the client (proxy), daemon is always None. When in the server (daemon), URI is always None. The default
implementation returns a secure hmac-md5 hash of the ident string and the challenge.</em></dd>
<dt><code>def createAuthChallenge(self, tcpserver, conn):</code></dt>
<dd><em>...called in the server (daemon) when a new connection comes in. It must return a challenge string that
is to be sent to the client, to be used in creating the authentication token. By default it returns a secure hash
of server IP, process ID, timestamp and a random value. Currently it is <strong>required</strong> that the
challenge string is exactly <strong>16</strong> bytes long! (a md5 hash is 16 bytes).</em></dd>
<dt><code>def mungeIdent(self, ident):</code></dt>
<dd><em>utility method to change a clear-text ident string into something that isn't easily recognised.
By default it returns the secure hash of the ident string. This is used to store the authentication
strings more securely (<code>setAllowedIdentifications</code>). The ident object that is
passed is actually free to be what you want, for instance you could use <code>obj._setIdentification(
("login", "password") )</code> to use a
login/password tuple. You have to use a custom connection validator to handle this, of course.</em></dd>
<dt><code>def setAllowedIdentifications(self, ids):</code></dt>
<dd><em>To tell the Daemon what identification strings are valid (the allowed secure passphrases).</em></dd>
<dt><br>
<em>(the following only if you subclass from <code>Pyro.protocol.BasicSSLValidator</code> (for SSL
connections):</em><br>
<code>def checkCertificate(self, cert):</code></dt>
<dd><em>...checks the SSL certificate. The client's SLL certificate is passed as an argument.
<strong>Note:</strong> this method is called from the <code>acceptHost</code> method, so you must leave that one
as-is or call the <code>BasicSSLValidator</code> base class implementation of that method if you override it.
<strong>See table below for return codes</strong></em></dd>
</dl>
</blockquote>The three check methods <code>acceptHost</code>, <code>acceptIdentification</code> and
<code>checkCertificate</code> must return <code>(1,0)</code> if the connection is accepted, or <code>(0,code)</code>
when the connection is refused, where <code>code</code> is one of the following:
<table>
<tr>
<th>Deny Reason Code</th>
<th>Description</th>
</tr>
<tr>
<td><code>Pyro.constants.DENIED_UNSPECIFIED</code></td>
<td>unspecified</td>
</tr>
<tr>
<td><code>Pyro.constants.DENIED_SERVERTOOBUSY</code></td>
<td>server too busy (too many connections)</td>
</tr>
<tr>
<td><code>Pyro.constants.DENIED_HOSTBLOCKED</code></td>
<td>host blocked</td>
</tr>
<tr>
<td><code>Pyro.constants.DENIED_SECURITY</code></td>
<td>security reasons (general)</td>
</tr>
</table>
<p>Pyro will raise the appropriate <code>ConnectionDeniedError</code> on the client when you deny a new connection.
On the server, you'll have to log the reason in the Pyro logfile yourself, if desired. When you accept a connection,
the daemon will log an entry for you.</p>
<h3><a name="nameserverplugins" id="nameserverplugins"></a>Name Server security plugins</h3>The Name Server supports
security plugins, to facilitate access control to the Name Server. Different options are available:
<ul>
<li>A validator for incoming broadcast server requests (such as 'what is your location' and 'shutdown'). You can
now decide what or when such commands are accepted or denied, for instance, based on client IP address.</li>
<li>A new connection validator for the NS server itself (which is a regular Pyro object, and this is the general
validator mentioned earlier). With this you can implement very coarse access control to the NS, for instance, deny
certain clients based on their IP address.</li>
</ul>You'll have to write a Python module that contains the following:
<ul>
<li><code>BCGuard()</code> function that returns a BC request validator object, or <code>None</code>.</li>
<li><code>NSGuard()</code> function that returns a NS new conn validator object, or <code>None</code>.</li>
<li>A class that implements a BC request validator. This class must inherit from
<code>Pyro.naming.BCReqValidator</code>. You must override the two methods that check for each command if it is
allowed or if it is refused. These are <code>acceptLocationCmd(self)</code> and
<code>acceptShutdownCmd(self)</code>, and they return 0 or 1 (accept or deny). You can access
<code>self.addr</code> to have the client's address (ip,port). You can call <code>self.reply('message')</code> to
send a message back to the client. This may be polite, to let it know why you refused the command.</li>
<li>A class that implements a NS new conn validator. See the documentation on connection validators, above, to see
how you must implement this.</li>
</ul>
When you start the NS using the '-s' switch, it will read your module and call the two functions mentioned
above to get your validator objects. Make sure your module is in your Python import path. The NS prints
the names of the plugins to show that it's using them and then starts. Have a look at the "NS_sec_plugins"
example to see how things are done.
<h3><a name="mobile" id="mobile"></a>Mobile objects and Code Validators</h3>
The mobile code support of Pyro is very
powerful but also dangerous, because the server is running code that comes in over the wire. Any code
can enter over the wire, correct, buggy, but also evil code (Trojans). It's obvious that loading and
running arbitrary code is dangerous. That's why you should set a <em>codeValidator</em> for each Pyro
object that might load mobile code (mobile objects). The default validator offers no protection: it
accepts all code. <em>Be aware that a simple check
on the name of uploaded code is not enough to make things safe; the client may supply it's own evil
version of the module you thought was perfectly safe.</em> Currently, there is no mechanism to guarantee
that the code is safe (for instance using some form of "code signing").
<p>This codeValidator is a function (or callable object) that takes three arguments: the name of the module, the code
itself, and the address of the client (usually a (IP,port) tuple). It should return 0 or 1, for 'deny' and 'accept'.
<code>Pyro.core.ObjBase</code>, the base class of all Pyro objects, has a <code>setCodeValidator(v)</code> method
that you must call with your custom validator function (or callable object). You can set a different validator for
each Pyro object that your server has.</p>
<p>The codeValidator is used for both directions; it checks if code is allowed from clients into the
server, but also if code is allowed to be sent from the server to clients. In the first case, all
three parameters have a value as mentioned above. In the second case (code from server to client),
only the <em>name</em> has a value, the other two
are None. For example, the code validator shown below is taken from the "agent2" example. It checks
if incoming code is from the "agent.ShoppingAgent" module, and outgoing code is from the "objects"
package:</p>
<pre>
def codeValidator(name,module,address):
if module and address:
return name=='agent.ShoppingAgent' # client uploads to us
else:
return name.startswith("objects.") # client downloads from us
.
.
.
mall.setCodeValidator(codeValidator)
</pre>
<p>Notice that a client doesn't have a code validator. If you're using 2-way mobile code (you've enabled
<code>PYRO_MOBILE_CODE</code> on the client), you will silently receive everything you need from the server. This is
because the clients usually trust the server... otherwise they wouldn't be calling it, would they?</p>
<h3><a name="firewalls" id="firewalls"></a>Firewalls</h3>Using a firewall to protect your network has nothing to do
with security in Pyro, but it may affect Pyro. Ofcourse the firewall can be used to fully protect your network for
systems outside the firewall; it can make it impossible for those systems to connect to your Pyro servers. But if you
want to access Pyro objects from outside the firewall, you may have to take some additional steps. Because they have
to do with configuring Pyro, and not with security, they are described in detail in the <a href=
"7-features.html#dnsip">Freatures chapter</a>.
<h3><a name="pickle" id="pickle"></a>The pickle trojan security problem, and XML pickling</h3>
<h4>Security warning: possible trojan attack</h4>By default, Pyro uses the native Python <code>pickle</code> protocol
to pass calls to remote objects. There is a security problem with <code>pickle</code>: it is possible to execute
arbitrary code on the server by passing an artificially constructed pickled string message. The standard Python
<code>Cookie</code> module also suffers from this problem. At the moment of writing, the Python documentation is not
clear on this subject. The problem is known to various people. <em>Using Pyro over the internet could expose your
server to this vulnerability!!!!</em>
<p>Using the (safe) <code>marshal</code> module is no option for Pyro because we lose the ability to
serialize user defined objects. But, if you accept a performance penalty of an order of a magnitude,
and more required bandwith (2-4 times more), you can choose to use the safe XML pickling.
To enable this, set the <code>PYRO_XML_PICKLE</code> config item to the appropriate value.
You need to have the appropriate library installed otherwise Pyro won't start. The server
will answer in XML pickled messages also, regardless of the server's <code>PYRO_XML_PICKLE</code> setting.
So make sure that the correct XML packages are installed on both ends of the communication. If the
server is configured to use <code>PYRO_XML_PICKLE</code>, it will
<em>only</em> accept XML pickled requests! This means that if you set this option, your server is safe
against pickling attacks.</p>
<p>Please note that at least since Python 2.2 a few pickle security flaws appear to have been removed, and the
obvious trojan exploit with pickle no longer works on Python 2.2+. But still, do you trust pickle? ;-) Use
<code>PYRO_XML_PICKLE</code> if you want to be safe.</p>
<p>If you decide to use the 'gnosis' XML Pickler, there is an additional config item to think about: <code>PYRO_GNOSIS_PARANOIA</code>.
It sets the 'paranoia' level that will be used for the Gnosis XML pickler. Higher=more secure.
The default setting (0) prevents automatic imports of modules during unpickling, because
this is potentially unsafe. However, it creates problems when you are sending arbitrary
user-defined types across the wire. The receiving side may not be able to fully
reconstruct the data types that were sent. You could explicitly import the needed
modules on the receiving side, or you could consider to set this config item to -1,
which enables automatic imports of user defined modules in the Gnosis pickler.
Note that setting it to a higer value than 0 breaks Pyro altogether because the pickler
will operate in a too strict way. The only sensible values at this time are 0 and -1.
When you want to use mobile code with Gnosis XML pickler, you need to set this to -1 as well.
Note that you have to use the same Gnosis XML library version everywhere. You can't mix
older versions with newer versions.</p>
<h3><a name="ssl" id="ssl"></a>SSL (secure socket layer) support</h3>
Pyro supports communication over secure sockets
(SSL). See the "ssl" example. Because Python doesn't support server-side SSL out-of-the-box, you'll
need the following add-on libraries to enable SSL support:
<ul>
<li><a href="http://chandlerproject.org/bin/view/Projects/MeTooCrypto">M2Crypto</a></li>
<li><a href="http://www.openssl.org">OpenSSL</a> (needed by M2Crypto)</li>
</ul>
<p>To start using SSL, you need to tell your Pyro daemon that it must use SSL instead of regular sockets. Do that by
passing a <code>prtcol</code> parameter when you create a daemon, as follows:</p>
<pre>
daemon = Pyro.core.Daemon(prtcol='PYROSSL')
</pre>(the <code>prtcol</code> defaults to 'PYRO' ofcourse). All Pyro objects connected to this daemon will get
registered in the Name Server using the special PYROSSL protocol, that tells Pyro to use SSL instead of regular
sockets. You may also want to add a special SSL connection validator on your daemon that checks the client certificate.
The client programs don't need any changes because Pyro knows automatically how to deal with the PYROSSL protocol.
There are a few configuration items that deal with the SSL configuration, look for <code>PYROSSL_CERTDIR</code> and the
other items starting with <code>PYROSSL</code>. See the <a href="http://chandlerproject.org/bin/view/Projects/MeTooCrypto">M2Crypto
homepage</a> or <a href="http://www.openssl.org">OpenSSL documentation</a> for instructions on how to create your own
Certificate Authority- and server/client certificates. There's also a nice guide <a href="http://sial.org/howto/openssl/ca/">here.</a>
<p>There are a few config items to tweak how Pyro uses SSL, where it can find your cert files, key files, and so on.
Have a look in the Configuration chapter to see what they are (they all have 'SSL' in their name)
</p>
<p><strong>Note:</strong> it is very likely that you have to set PYRO_DNS_URI to True, because Pyro uses IP addresses by default while SSL
certificates will usually contain hostnames. If you don't set this config item the SSL certificate will be rejected
because the name won't match.</p>
<p>Please note that there is a <a href="http://mail.python.org/pipermail/python-list/2007-January/593716.html">known bug</a> in M2Crypto that makes it impossible to use socket timeouts (socket.setdefaulttimeout) when using SSL.</p>
<div class="nav">
<hr>
<table width="100%">
<tr>
<td align="left"><a href="8-example.html"><previous</a> | <a href="index.html">contents</a> | <a href=
"10-errors.html">next></a></td>
<td align="right">Pyro Manual</td>
</tr>
</table></div>
</body>
</html>
|