<title>Defense Against Robots</title>
A typical Fossil website can have billions and billions of pages,
and many of those pages (for example diffs and annotations and tarballs)
can be expensive to compute.
If a robot walks a Fossil-generated website,
it can present a crippling bandwidth and CPU load.
A "robots.txt" file can help, but in practice, most robots these
days ignore the robots.txt file, so it won't help much.
A Fossil website is intended to be used
interactively by humans, not walked by robots. This article
describes the techniques used by Fossil to try to welcome human
users while keeping out robots.
<h2>Defenses Are Enabled By Default</h2>
In the latest implementations of Fossil, most robot defenses are
enabled by default. You can probably get by with standing up a
public-facing Fossil instance in the default configuration. But
you can also customize the defenses to serve your particular needs.
<h2>Customizing Anti-Robot Defenses</h2>
Admin users can configure robot defenses on the
"Robot Defense Settings" page (/setup_robot).
That page is accessible (to Admin users) from the default menu bar
by click on the "Admin" menu choice, then selecting the
"Robot-Defense" link from the list.
<h2>The Hyperlink User Capability</h2>
Every Fossil web session has a "user". For random passers-by on the internet
(and for robots) that user is "nobody". The "anonymous" user is also
available for humans who do not wish to identify themselves. The difference
is that "anonymous" requires a login (using a password supplied via
a CAPTCHA) whereas "nobody" does not require a login.
The site administrator can also create logins with
passwords for specific individuals.
Users without the <b>[./caps/ref.html#h | Hyperlink]</b> capability
do not see most Fossil-generated hyperlinks. This is
a simple defense against robots, since [./caps/#ucat | the "nobody"
user category] does not have this capability by default.
Users must log in (perhaps as
"anonymous") before they can see any of the hyperlinks. A robot
that cannot log into your Fossil repository will be unable to walk
its historical check-ins, create diffs between versions, pull zip
archives, etc. by visiting links, because there are no links.
A text message appears at the top of each page in this situation to
invite humans to log in as anonymous in order to activate hyperlinks.
But requiring a login, even an anonymous login, can be annoying.
Fossil provides other techniques for blocking robots which
are less cumbersome to humans.
<h2>Automatic Hyperlinks Based on UserAgent and Javascript</h2>
Fossil has the ability to selectively enable hyperlinks for users
that lack the <b>Hyperlink</b> capability based on their UserAgent string in the
HTTP request header and on the browsers ability to run Javascript.
The UserAgent string is a text identifier that is included in the header
of most HTTP requests that identifies the specific maker and version of
the browser (or robot) that generated the request. Typical UserAgent
strings look like this:
<ul>
<li> Mozilla/5.0 (Windows NT 6.1; rv:19.0) Gecko/20100101 Firefox/19.0
<li> Mozilla/4.0 (compatible; MSIE 8.0; Windows_NT 5.1; Trident/4.0)
<li> Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)
<li> Wget/1.12 (openbsd4.9)
</ul>
The first two UserAgent strings above identify Firefox 19 and
Internet Explorer 8.0, both running on Windows NT. The third
example is the robot used by Google to index the internet.
The fourth example is the "wget" utility running on OpenBSD.
Thus the first two UserAgent strings above identify the requester
as human whereas the second two identify the requester as a robot.
Note that the UserAgent string is completely under the control
of the requester and so a malicious robot can forge a UserAgent
string that makes it look like a human. But most robots want
to "play nicely" on the internet and are quite open
about the fact that they are a robot. And so the UserAgent string
provides a good first-guess about whether or not a request originates
from a human or a robot.
The [/help?cmd=auto-hyperlink|auto-hyperlink] setting, shown as
"<b>Enable hyperlinks based on User-Agent and/or Javascript</b>" on
the Robot Defense Settings page,
can be set to "UserAgent only" or "UserAgent and Javascript" or "off".
If the UserAgent string looks like a human and not a robot, then
Fossil will enable hyperlinks even if the <b>Hyperlink</b> capability
is omitted from the user permissions. This setting gives humans easy
access to the hyperlinks while preventing robots
from walking the billions of pages on a typical Fossil site.
If the setting is "UserAgent only" (2), then the hyperlinks are simply
enabled and that is all. But if the setting is "UserAgent and Javascript" (1),
then the hyperlinks are not enabled directly.
Instead, the HTML code that is generated contains anchor tags ("<a>")
with "href=" attributes that point to [/honeypot] rather than the correct
link. JavaScript code is added to the end of the page that goes back and
fills in the correct "href=" attributes of
the anchor tags with the true hyperlink targets, thus enabling the hyperlinks.
This extra step of using JavaScript to enable the hyperlink targets
is a security measure against robots that forge a human-looking
UserAgent string. Most robots do not bother to run JavaScript and
so to the robot the empty anchor tag will be useless. But all modern
web browsers implement JavaScript, so hyperlinks will show up
normally for human users.
If the [/help?cmd=auto-hyperlink|"auto-hyperlink"] setting is (2)
"<b>Enable hyperlinks using User-Agent and/or Javascript</b>",
then there are now two additional sub-settings that control when
hyperlinks are enabled.
The first new sub-setting is a delay (in milliseconds) before setting
the "href=" attributes on anchor tags. The default value for this
delay is 10 milliseconds. The idea here is that a robots will try to
interpret the links on the page immediately, and will not wait for delayed
scripts to be run, and thus will never enable the true links.
The second sub-setting waits to run the
JavaScript that sets the "href=" attributes on anchor tags until after
at least one "mousedown" or "mousemove" event has been detected on the
<body> element of the page. The thinking here is that robots will not be
simulating mouse motion and so no mouse events will ever occur and
hence the hyperlinks will never become enabled for robots.
See also [./loadmgmt.md|Managing Server Load] for a description
of how expensive pages can be disabled when the server is under heavy
load.
<h2>Do Not Allow Robot Access To Certain Pages</h2>
The [/help?cmd=robot-restrict|robot-restrict setting] is a comma-separated
list of GLOB patterns for pages for which robot access is prohibited.
The default value is:
<blockquote><pre>
timelineX,diff,annotate,fileage,file,finfo,reports
</pre></blockquote>
Each entry corresponds to the first path element on the URI for a
Fossil-generated page. If Fossil does not know for certain that the
HTTP request is coming from a human, then any attempt to access one of
these pages brings up a javascript-powered captcha. The user has to
click the accept button the captcha once, and that sets a cookie allowing
the user to continue surfing without interruption for 15 minutes or so
before being presented with another captcha.
Some path elements have special meanings:
* <b>timelineX →</b>
This means a subset of /timeline/ pages that are considered
"expensive". The exact definition of which timeline pages are
expensive and which are not is still the subject of active
experimentation and is likely to change by the time you read this
text. The idea is that anybody (including robots) can see a timeline
of the most recent changes, but timelines of long-ago change or that
contain lists of file changes or other harder-to-compute values are
prohibited.
* <b>zip →</b>
The special "zip" keyword also matches "/tarball/" and "/sqlar/".
* <b>zipX →</b>
This is like "zip" in that it restricts access to "/zip/", "/tarball"/
and "/sqlar/" but with exceptions:<ol type="a">
<li><p> If the [/help?cmd=robot-zip-leaf|robot-zip-leaf] setting is
true, then tarballs of leaf check-ins are allowed. This permits
URLs that attempt to download the latest check-in on trunk or
from a named branch, for example.
<li><p> If a check-in has a tag that matches the GLOB list in
[/help?cmd=robot-zip-tag|robot-zip-tag], then tarballs of that
check-in are allowed. This allow check-ins tagged with
"release" or "allow-robots" (for example) to be downloaded
without restriction.
</ol>
The "zipX" restriction is not in the default robot-restrict setting.
This is something you might want to add, depending on your needs.
* <b>diff →</b>
This matches /vdiff/ and /fdiff/ and /vpatch/ and any other page that
is primarily about showing the difference between two check-ins or two
file versioons.
* <b>annotate →</b>
This also matches /blame/ and /praise/.
Other special keywords may be added in the future.
The default [/help?cmd=robot-restrict|robot-restrict]
setting has been shown in practice to do a good job of keeping
robots from consuming all available CPU and bandwidth while will
still allowing humans access to the full power of the site without
having to be logged in.
One possible enhancement is to add "zipX" to the
[/help?cmd=robot-restrict|robot-restrict] setting,
and enable [help?cmd=robot-zip-leaf|robot-zip-leaf]
and configure [help?cmd=robot-zip-tag|robot-zip-tag].
Do this if you find that robots downloading lots of
obscure tarballs is causing load issues on your site.
<h2>Anti-robot Exception RegExps</h2>
The [/help?cmd=robot-exception|robot-exception setting] under the name
of <b>Exceptions to anti-robot restrictions</b> is a list of
[/re_rules|regular expressions], one per line, that match
URIs that will bypass the captcha and allow robots full access. The
intent of this setting is to allow automated build scripts
to download specific tarballs of project snapshots.
The recommended value for this setting allows robots to use URIs of the
following form:
<blockquote>
<b>https://</b><i>DOMAIN</i><b>/tarball/release/</b><i>HASH</i><b>/</b><i>NAME</i><b>.tar.gz</b>
</blockquote>
The <i>HASH</i> part of this URL can be any valid
[./checkin_names.wiki|check-in name]. The link works as long as that
check-in is tagged with the "release" symbolic tag. In this way,
robots are permitted to download tarballs (and ZIP archives) of official
releases, but not every intermediate check-in between releases. Humans
who are willing to click the captcha can still download whatever they
want, but robots are blocked by the captcha. This prevents aggressive
robots from downloading tarballs of every historical check-in of your
project, once per day, which many robots these days seem eager to do.
For example, on the Fossil project itself, this URL will work, even for
robots:
<blockquote>
https://fossil-scm.org/home/tarball/release/version-2.27/fossil-scm.tar.gz
</blockquote>
But the next URL will not work for robots because check-in 3bbd18a284c8bd6a
is not tagged as a "release":
<blockquote>
https://fossil-scm.org/home/tarball/release/3bbd18a284c8bd6a/fossil-scm.tar.gz
</blockquote>
The second URL will work for humans, just not robots.
<h2>The Ongoing Struggle</h2>
Fossil currently does a good job of providing easy access to humans
while keeping out troublesome robots. However, robots
continue to grow more sophisticated, requiring ever more advanced
defenses. This "arms race" is unlikely to ever end. The developers of
Fossil will continue to try improve the robot defenses of Fossil so
check back from time to time for the latest releases and updates.
Readers of this page who have suggestions on how to improve the robot
defenses in Fossil are invited to submit your ideas to the Fossil Users
forum:
[https://fossil-scm.org/forum].