/usr/share/doc/python-lamson/bounce_detection.html is in python-lamson 1.0pre11-1.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 | <?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>LamsonProject: Bounce Detection</title>
<link rel="stylesheet" href="/styles/global.css" type="text/css" charset="utf-8" />
<link rel="stylesheet" href="/css/code.css" type="text/css" charset="utf-8" />
<!--[if IE 7]>
<style type="text/css" media="screen">
div#column_left ul.sidebar_menu li div.color{
display: none;
}
</style>
<![endif]-->
<link href="/prettify.css" type="text/css" rel="stylesheet" />
<script type="text/javascript" src="/prettify.js"></script>
</head>
<body onload="prettyPrint()">
<div id="content_centered">
<div id="header">
<h1><img id="logo" src="/images/lamson.png" alt="Lamson Project(TM) - Pipes and aliases are so 1970." /></h1>
<ul id="header_menu">
<li><a href="/">Home</a></li>
<li><a href="/blog/">News</a></li>
<li><a href="/feed.xml">Feed</a></li>
<li><a href="/download.html">Download</a></li>
<li><a href="/docs/">Documentation</a></li>
<li><a href="/docs/api/">API</a></li>
</ul>
</div>
<div id="main_content">
<h1>Bounce Detection</h1>
Lamson supports intelligent bounce detection with its
<a href="http://lamsonproject.org/docs/api/lamson.bounce-module.html">lamson.bounce</a>
module. The bounce handling is based on a probability that, depending on found
headers, the message is a bounce. It then gives you a nice clean interface to
check who it’s from, the originating <span class="caps">SMTP</span> server, the error messages, and any
human readable messages.
<h1>How Bounces Actually Work</h1>
<p>Figuring out how a bounce is actually handled is a bit difficult because the
majority of the information available is written by people who know very little
of <span class="caps">SMTP</span> server operations. When the average programmer thinks of handling a
bounce, she usually has one of these concepts in mind:</p>
<ol>
<li>The message could not be delivered, so the remote <span class="caps">SMTP</span> server sent back a bounce message.</li>
<li>The message could not be delivered, so the local <span class="caps">SMTP</span> server sent back a bounce message.</li>
<li>The message could not be delivered, so the recipient’s email client sent back a bounce message.</li>
<li>The message could not be delivered, so Lamson crafted a bounce message.</li>
</ol>
<p>Obviously logically you can’t have the recipient’s email client send you a
bounce unless the user does something weird (and incredibly annoying). This
makes sense because the message was <code>not delivered</code>. How can the email client
send back a bounce if they don’t receive the message.</p>
<blockquote>
<p>Yes, some clients do support sending bounces, but very few people use this
feature. If you do please talk about it as one more edge case to deal with.</p>
</blockquote>
<p>Next Lamson can’t craft the bounce messages for you because Lamson is simply
trying to deliver outgoing mail to a smart-host. Lamson does nothing but ask
the smart-host (your local <span class="caps">SMTP</span> server) to deliver, and then waits for a
response. That means Lamson is not sending you any kind of bounce unless you
write code to make it do that.</p>
<p>That leaves only two options for <code>who</code> is really sending the bounce message:
the remote <span class="caps">MTA</span> or the local <span class="caps">MTA</span>. The truth is it’s a little bit of both.</p>
<h2>Your Local <span class="caps">MTA</span> Is A Person</h2>
<p>How a bounce works in practice involves two <span class="caps">SMTP</span> servers: your local
smart-host, and the remote server that it tries to connect to for delivery.
The message you actually get in lamson is from the <code>local</code> server, and usually
an address <code>at</code> that local server. It does not come from the remote server,
but inside your bounce will be a message and status indicators from the remote
server indicating why it bounced.</p>
<p>What happens is your local server attempts to deliver, and the remote server
rejects it for some reason. Now your local server is supposed to try again in
certain situations, but after a certain number of rejections or failures it
crafts a bounce message. That bounce message is then returned to your <code>lamson</code>
server based on the <code>envelope</code> header of the message (more on that later).</p>
<p>Now, what’s inside this message? Well, it’s a oddly nested barely standardized
pile of random other messages. This is the frustration with bounce handling.
You pretty much either have to use a probability mechanism (like Lamson does)
or you have to craft your bounce handling for your very specific local server
and any other possible servers you talk with. Even then you can have problems
dealing with bounces reliably.</p>
<p>Inside this description is an important concept to understand:</p>
<blockquote>
<p>Lamson does <span class="caps">NOT</span> process bounces from the remote <span class="caps">MTA</span> or the remote user in
any way. Lamson process bounces from the <code>local</code> <span class="caps">SMTP</span> server.</p>
</blockquote>
<p>This is important because if you try to use the bounce message as if it comes
from the remote user, then you’ll accidentally put your <code>local</code> server into a
state that prevents you from processing future bounce messages.</p>
<p>If you did not get that, reread this whole section again until moving forward.</p>
<h2>“Hard” vs. “Soft”</h2>
<p>Another complexity in dealing with bounces is the concept of “Hard” vs. “Soft”
bounces. My opinion is that the distinction is fairly retarded since it is
almost entirely unreliable and has no meaning to someone trying to handle a
bounce.</p>
<p>In popular terminology the main difference is this:</p>
<ul>
<li>Hard bounce means that person does not exist, so I can not sell him any more crap and need to remove him (maybe).</li>
<li>Soft bounce means that person still exists, but my marketing bullshit didn’t get through, I should try again 10 or 15 more times until he gets my important message about winning the lottery.</li>
</ul>
<p>You may be laughing, or screaming various pedantic specifics, but the above two
distinctions are both how many email services look at bounces, and how most
malicious mail users view them.</p>
<p>For the mail services, leaking a hard or soft bounce is a security problem
because it tells a malicious sender if that address is a valid person, why the
sender was blocked, and how they can work around it. This is why many of the
error messages you get back are vague and mostly lies. The major email
services just don’t want to give you any information that you can use to
circumvent their anti-spam measures.</p>
<blockquote>
<p>The services that have the strictest anti-spam measures also have the most
nasty disgusting spam on the web pages displaying user’s mail. Yahoo is both
the worst for erroneously blocking email and for showing the worst most tricky
spam all over every square inch of their mail service.</p>
</blockquote>
<p>How does this relate to your Lamson application? It basically says that you
should treat bounces as being basically soft bounces all the time, and then
tune your rules heuristically over time as you run into more and more.</p>
<h2><span class="caps">VERP</span></h2>
<p>The final topic to touch on before getting into how Lamson handles bounces is
one of <span class="caps">VERP</span>. Remember in the description of bounce handling above I said that
your <code>local</code> <span class="caps">MTA</span> sends the bounce message back to the <code>envelope from</code> not the
From in the headers (well, most ones will). Because of this you can have a
From in the envelope that is only replied to when there’s a bounce, and then
put the real From in the headers for normal processing.</p>
<p>This is effectively what <a href="http://en.wikipedia.org/wiki/Variable_envelope_return_path">Variable Envelope Return
Path</a> does to
process bounce messages. Rather than attempt to parse the body of a bounce
message, <span class="caps">VERP</span> rewrite the From address so that when a bounce is returned, you
only have to process the returned address to see what the original was.</p>
<p>Lamson could potentially support this, but it has several problems for generic
bounce handling which meant that actually parsing bounce bodies was a better
option.</p>
<h2>Using Lamson’s Bounce Handling</h2>
<p>Using Lamson’s bounce handling is fairly simple. It involves the following
process:</p>
<ol>
<li>Import a special decorator <code>bounce_to</code> from <a href="http://lamsonproject.org/docs/api/lamson.bounce-module.html">lamson.bounce</a></li>
<li>Create two (or one) handlers to deal with bounces.</li>
<li>Place the decorator on any <code>START</code> entry points to your handlers that can receive bounces, pointing them at your handlers.</li>
<li>Handle the bounces in your handlers, making sure to return back to the <code>START</code> state for the local <span class="caps">MTA</span> (remember, the local <span class="caps">MTA</span> is a person for bounce handling).</li>
</ol>
<p>Here’s some simple code that does exactly this by just ignoring bounces from
the <a href="http://myinboxisnota.tv/">myinboxisnota.tv</a> example:</p>
<pre class="code prettyprint">
from config.settings import BOUNCES
from lamson.routing import route
from lamson.bounce import bounce_to
@route(".+")
def IGNORE_BOUNCE(message):
bounces = queue.Queue(BOUNCES)
bounces.push(message)
return START
@route("start@(host)")
@bounce_to(soft=IGNORE_BOUNCE, hard=IGNORE_BOUNCE)
def START(message, host=None):
CONFIRM.send(relay, "start", message, "mail/start_confirm.msg", locals())
return CONFIRMING
</pre>
<p>This example is stripped down from the real code so you can see what’s going
on. If we walk through this you can see what happens:</p>
<ol>
<li>First we import the <code>BOUNCES</code> variable from <code>config.settings</code> which is just the queue we want to dump bounces into.</li>
<li>We then create a special handler named <code>IGNORE_BOUNCE</code> that accepts a message to anything in its <code>route</code> and just puts the message in the <code>BOUNCES</code> queue.</li>
<li>This <code>IGNORE_BOUNCE</code> handler then immediately returns <code>START</code> so we go back to the <span class="caps">START</span> state.</li>
<li>On the <code>START</code> state you’ll see that we have our <code>bounce_to</code> declaration that points for <code>hard</code> and <code>soft</code> bounces at <code>IGNORE_BOUNCE</code>.</li>
<li>This decorator wraps your handler in a little check that, if the message is a bounce, your <code>START</code> state won’t get called, and instead your <code>IGNORE_BOUNCE</code> state will.</li>
</ol>
<p>That is pretty much all you need to deal with to re-route bounces somewhere else.
You can also redirect them to a totally different handler, which is exactly
what the <a href="http://librelist.com/">librelist.com</a> example does.</p>
<h2>How It Works</h2>
<p>How does Lamson figure out that something is a bounce? What Lamson does is it assumes bounces will have some or all of these
headers:</p>
<pre class="code prettyprint">
BOUNCE_MATCHERS = {
'Action': re.compile(r'(failed|delayed|delivered|relayed|expanded)', re.IGNORECASE | re.DOTALL),
'Content-Description': re.compile(r'(Notification|Undelivered Message|Delivery Report)', re.IGNORECASE | re.DOTALL),
'Diagnostic-Code': re.compile(r'(.+);\s*([0-9\-\.]+)?\s*(.*)', re.IGNORECASE | re.DOTALL),
'Final-Recipient': re.compile(r'(.+);\s*(.*)', re.IGNORECASE | re.DOTALL),
'Received': re.compile(r'(.+)', re.IGNORECASE | re.DOTALL),
'Remote-Mta': re.compile(r'(.+);\s*(.*)', re.IGNORECASE | re.DOTALL),
'Reporting-Mta': re.compile(r'(.+);\s*(.*)', re.IGNORECASE | re.DOTALL),
'Status': re.compile(r'([0-9]+)\.([0-9]+)\.([0-9]+)', re.IGNORECASE | re.DOTALL)
}
</pre>
<p>The problem traditionally with parsing a bounce message was that, while you
need to find all of these headers, there was no real standard as to how the
messages in the bounce message are structured. From my
<a href="http://www.postfix.org/">postfix</a> server the bounce message is a sequence of
about 6 nested attachments containing other messages, and sometimes the nesting
goes three deep.</p>
<p>Rather than rely on this structure (which changes all the time) or that these
headers are always present (they aren’t), Lamson takes a probabilistic approach
based on the number of headers and properly formatted values it finds in <code>all</code>
nested attachments. The process goes something like this:</p>
<ol>
<li>Traverse all the possible nested attachments.</li>
<li>Try to find each header in the attachment. If it’s found add a point.</li>
<li>If the header is found, use the regex associated with it (above) to try to match the value.
<ol>
<li>If the value matches, then keep the regex captures for later. Add another point.</li>
</ol></li>
<li>For each header found, and any regex captures that matched the bodies, put them into an internal dict for analyzing the bounce information.</li>
<li>Finally, calculate a probability score that is the total number of BOUNCE_MATCHERS * 2.0 / points.</li>
</ol>
<p>In general, if a message is found that has a 0.3 or higher bounce probability
then it is considered a bounce message and you can look at it. The <code>bounce_to</code>
decorator has a threshold you can adjust if you want to be more or less strict.</p>
<p>The final result of this processing (which is actually very fast) is that any
calls to <code>MailRequest.is_bounce</code> will either return True or False, and then
after you call is_bounce you can access the <code>MailRequest.bounce</code> attribute to
analyze the information. That information is captured and cooked into a
<a href="http://lamsonproject.org/docs/api/lamson.bounce.BounceAnalyzer-class.html">BounceAnalyzer</a>
object.</p>
<h2>What It Looks Like To Receive One</h2>
<p>It’s also instructive to see what it looks like when Lamson processes a bounce
message. Here’s the <a href="http://librelist.com/">librelist.com</a> server processing
a bounce message:</p>
<pre class="code">
2009-08-21 13:43:47,223 - root - DEBUG - Pulled message with key:
'1250876622.V8e00I219de0M128371' off
2009-08-21 13:43:47,231 - root - DEBUG - Message received from Peer:
'/var/mail/vhosts/librelist.com/delivery/', From: u'"SPAMMER"
<SPAMMER@hotmail.com>', to To [u'lamson@librelist.com'].
2009-08-21 13:43:47,251 - routing - DEBUG - Matched u'lamson@librelist.com'
against START.
2009-08-21 13:43:47,332 - routing - DEBUG - Message to
set([u'lamson@librelist.com']) was handled by app.handlers.admin.START
2009-08-21 13:43:57,367 - root - DEBUG - Pulled message with key:
'1250876627.V8e00I219661M719350' off
2009-08-21 13:43:57,381 - root - DEBUG - Message received from Peer:
'/var/mail/vhosts/librelist.com/delivery/', From:
u'MAILER-DAEMON@librelist.com (Mail Delivery System)', to To
[u'lamson-confirm-74e2ca94b24a4be18da277f4666a6494@librelist.com'].
2009-08-21 13:43:57,410 - routing - DEBUG - Matched
u'lamson-confirm-74e2ca94b24a4be18da277f4666a6494@librelist.com' against START.
2009-08-21 13:43:57,431 - routing - DEBUG - Message to
set([u'lamson-confirm-74e2ca94b24a4be18da277f4666a6494@librelist.com']) was
handled by app.handlers.admin.START
</pre>
<p>These log messages show the following interaction:</p>
<ol>
<li>SPAMMER@hotmail.com tried to spam the lamson@librelist.com mailing list.</li>
<li>They were required to subscribe, so Lamson sent them a confirmation mail.</li>
<li>That message bounced, so Postfix sent back a bounce message from MAILER-DAEMON@librelist.com to Lamson.</li>
<li>This message from <span class="caps">MAILER-DAEMON</span> is a bounce, so the librelist code handled it on the <span class="caps">START</span> state, <span class="caps">NOT</span> the CONFIRMING_SUBSCRIBE state.</li>
<li>Internally, librelist looked up the target user and just zapped them.</li>
</ol>
<p>That shows how the bounce doesn’t come from SPAMMER@hotmail.com nor any server
at hotmail.com, but instead from MAILER-DAEMON@librelist.com. You could also
get messages from a remote <span class="caps">MTA</span>, but if they were found to be bounce messages
then that remote <span class="caps">MTA</span> would be treated like your own <span class="caps">MTA</span>.</p>
<h2>Gettting Bounce Information From BounceAnalyzer</h2>
<p>The <a href="http://lamsonproject.org/docs/api/lamson.bounce.BounceAnalyzer-class.html">BounceAnalyzer</a>
does the work of figuring out additional useful information you can use to
determine what to do with the bounce. It looks at the final headers that are
scanned in the above process and pulls out important information everyone
needs. The list of information you can get is:</p>
<ul>
<li>primary_status — The main status number that determines hard vs soft.</li>
<li>secondary_status — Advice status.</li>
<li>combined_status — the 2nd and 3rd number combined gives more detail.</li>
<li>remote_mta — The <span class="caps">MTA</span> that you sent mail to and aborted.</li>
<li>reporting_mta — The <span class="caps">MTA</span> that was sending the mail and has to report to you.</li>
<li>diagnostic_codes — Human readable codes usually with info from the provider.</li>
<li>action — Usually 'failed’, and turns out to be not too useful.</li>
<li>content_parts — All the attachments found as a hash keyed by the type.</li>
<li>original — The original message, if it’s found.</li>
<li>report — All report elements, as lamson.encoding.MailBase raw messages.</li>
<li>notification — Usually the detailed reason you bounced.</li>
</ul>
<p>But, refer to the documentation for more accurate listings. An important
feature is that the status codes are parsed and converted into a standard list
available in <code>lamson.bounce</code> based on their numeric values. Rather than parse
the details given by the remote <span class="caps">MTA</span>, you just use the number codes to get a
human readable output.</p>
<p>The best way to see all that’s possible is to take a glance at the Lamson unit
test for the BounceAnalyzer:</p>
<pre class="code prettyprint">
def test_bounce_analyzer_on_bounce():
bm = mail.MailRequest(None,None,None, open("tests/bounce.msg").read())
assert bm.is_bounce()
assert bm.bounce
assert bm.bounce.score == 1.0
assert bm.bounce.probable()
assert_equal(bm.bounce.primary_status, (5, u'Permanent Failure'))
assert_equal(bm.bounce.secondary_status, (1, u'Addressing Status'))
assert_equal(bm.bounce.combined_status, (11, u'Bad destination mailbox address'))
assert bm.bounce.is_hard()
assert_equal(bm.bounce.is_hard(), not bm.bounce.is_soft())
assert_equal(bm.bounce.remote_mta, u'gmail-smtp-in.l.google.com')
assert_equal(bm.bounce.reporting_mta, u'mail.zedshaw.com')
assert_equal(bm.bounce.final_recipient,
u'asdfasdfasdfasdfasdfasdfewrqertrtyrthsfgdfgadfqeadvxzvz@gmail.com')
assert_equal(bm.bounce.diagnostic_codes[0], u'550-5.1.1')
assert_equal(bm.bounce.action, 'failed')
assert 'Content-Description-Parts' in bm.bounce.headers
assert bm.bounce.error_for_humans()
</pre>
<p>Here you can see that you can figure out if the bounce is hard vs. soft, get a
human description, access status codes of various flavors, get the remote MTA’s
name, the reporting <span class="caps">MTA</span> (your local), who it was originally for
(final_recipient), and even access the raw <code>bounce.headers</code> if that’s not good
enough.</p>
<h2>Augmenting The Matchers</h2>
<p>Another advantage of this method of processing the bounces is that if your <span class="caps">SMTP</span>
server crafts something that hasn’t been handled, then you can augment the
matchers being used. Simply update the <code>lamson.bounce.BOUNCE_MATCHERS</code> dict
with your new ones and make sure to update <code>lamson.bounce.BOUNCE_MAX</code> to be 2
times that.</p>
<p>The status codes are also available in the same way. Refer to the source for
more information.</p>
<p>One tricky part of Lamson’s bounce handling is that it does assume a certain
structure for the BounceAnalyzer to get at the internal details. This
structure is the one used by Postfix, and it should be the same for other
servers. However, if you run into a structural difference, report the results
back so the handling can be improved.</p>
<h2>A More Complete Example</h2>
<p>Finally, the <a href="http://librelist.com/">librelist.com</a> example code has a much more
complete example of using bounces to disable users and shift their state.
Rather than describe it in detail here, I’ll simply point you at the <a href="/releases/">source
releases</a> so you can see it for yourself. Look in
<code>examples/librelist/app/handlers/bounce.py</code> to see how it all works.</p>
<p>In fact, studying how this is triggered from the rest of the librelist example
is a great way to learn how to use Lamson in an advanced fashion. Study well.</p>
<h2>Conclusion</h2>
<p>Lamson bounce handling is very advanced and can deal with a wide range of
scenarios. It should work with a wide range of bounce styles and other
servers, but feel free to report your own experiences and differences.</p>
</div>
<div id="column_left">
<ul class="sidebar_menu">
<li>
<div class="item">
<div class="color" style="background-color: #ff0000;"> </div>
<a href="/blog/">Latest News</a>
</div>
</li>
<li>
<div class="item">
<div class="color" style="background-color: #ff9900;"> </div>
<a href="/download.html">Download the Gear</a>
</div>
</li>
<li>
<div class="item">
<div class="color" style="background-color: #99cc00;"> </div>
<a href="/docs/getting_started.html">Getting Started</a>
</div>
</li>
<li>
<div class="item">
<div class="color" style="background-color: #3399ff;"> </div>
<a href="/docs/">Documentation</a>
</div>
</li>
<li>
<div class="item">
<div class="color" style="background-color: #ff3399;"> </div>
<a href="/docs/faq.html">Frequently Asked Questions</a>
</div>
</li>
<li>
<div class="item">
<div class="color" style="background-color: #006699;"> </div>
<a href="/about.html">About Lamson</a>
</div>
</li>
<li>
<div class="item">
<div class="color" style="background-color: #0099cc;"> </div>
<a href="/contact.html">Getting Help with Lamson</a>
</div>
</li>
</ul>
<div class="sidebar_item">
<h3>Quick Start</h3>
<p>See the download instructions for information on getting lamson, and read the getting started instructions to start your own application in less than 10 minutes.</p>
</div>
<br/>
<div class="sidebar_item">
<h3>Mailing Lists</h3>
<p>Lamson hosts its own <a href="/lists/">mailing lists</a> as well as provides a free open mailing list
service for anyone who needs one. Simply send an email to the list you want @librelist.com and it will
get you started.</p>
</div>
</div>
<div id="footer">
<div class="footer_content">
Lamson Project(TM) and all material on this site Copyright © 2009 <a href="http://zedshaw.com/" title="Zed Shaw's blog">Zed Shaw</a> unless otherwise stated.<br/>
Website Designed by <a href="http://kenkeiter.com/">Kenneth Keitner</a> and donated to the LamsonProject.
</div>
</div>
<!-- end:centered_content -->
</div>
</body>
</html>
|