Implement chunk allocator on Windows.

[polipo.git] / polipo.texi

\input texinfo @c -*-texinfo-*-
@c %**start of header
@setfilename polipo.info
@settitle The Polipo Manual
@afourpaper
@c %**end of header

@dircategory Network Applications
@direntry
* Polipo: (polipo).                     The Polipo caching web proxy.
@end direntry

@copying
Copyright @copyright{} 2003 -- 2005 by Juliusz Chroboczek.
@end copying

@titlepage
@title The Polipo Manual
@author Juliusz Chroboczek
@page
@vskip 0pt plus 1fill
Polipo is a caching web proxy designed to be used as a personal
cache or a cache shared among a few users.
@vskip 0pt plus 1fill

@insertcopying

@end titlepage

@contents

@ifnottex
@node Top, Background, (dir), (dir)
@top Polipo

Polipo is a caching web proxy designed to be used as a personal
cache or a cache shared among a few users.

@ifhtml
The latest version of Polipo can be found on 
@uref{http://www.pps.jussieu.fr/~jch/software/polipo/,the Polipo web page}.
@end ifhtml

This manual was written by
@uref{http://www.pps.jussieu.fr/~jch/,,Juliusz Chroboczek}.

@end ifnottex

@menu
* Background::                  Background information.
* Running::                     Running Polipo
* Network::                     Polipo and the network.
* Caching::                     Caching.
* Memory usage::                Limiting Polipo's memory usage.
* Copying::                     Your rights and mine.
* Variable index::              Variable index.
* Concept index::               Concept index.
@end menu

@node Background, Running, Top, Top
@chapter Background

@menu
* The web::                     The web and HTTP.
* Proxies and caches::          Proxies and caches.
* Latency and throughput::      Optimise latency, not throughput.
* Network traffic::             Be nice to the net.
* Partial instances::           Don't discard data.
* POST and PUT::                Other requests
* Other HTTP proxies::          Why did I write Polipo from scratch?
@end menu

@node The web, Proxies and caches, Background, Background
@section The web and HTTP
@cindex URL
@cindex resource
@cindex instance
@cindex entity
@cindex HTTP

The web is a wide-scale decentralised distributed hypertext system,
something that's obviously impossible to achieve reliably.

The web is a collection of @dfn{resources} which are identified by
@dfn{URLs}, strings starting with @code{http://}.  At any point in
time, a resource has a certain value, which is called an
@dfn{instance} of the resource.

The fundamental protocol of the web is HTTP, a simple request/response
protocol.  With HTTP, a client can make a request for a resource to a
server, and the server replies with an @dfn{entity}, which is an
on-the-wire representation of an instance or of a fragment thereof.

@node Proxies and caches, Latency and throughput, The web, Background
@section Proxies and caches
@cindex proxy
@cindex caching

A proxy is a program that acts as both a client and a server.  It
listens for client requests and forwards them to servers, and forwards
the servers' replies to clients.

An HTTP proxy can optimise web traffic away by @dfn{caching} server
replies, storing them in memory in case they are needed again.  If a
reply has been cached, a later client request may, under some
conditions, be satisfied without going to the source again.

In addition to taking the shortcuts made possible by caching, proxies
can improve performance by generating better network traffic than the
client applications would do.

Proxies are also useful in ways unrelated to raw performance.  A proxy
can be used to contact a server that are not visible to the browser,
for example because there is a firewall in the way (@pxref{Parent
proxies}), or because the client and the server use different lower
layer protocols (for example IPv4 and IPv6).  Another common
application of proxies is to modify the data sent to servers and
returned to clients, for example by censoring headers that expose too
much about the client's identity (@pxref{Censoring headers}) or
removing advertisements from the data returned by the server
(@pxref{Forbidden}).

Polipo is a caching HTTP proxy that was originally designed as a
@dfn{personal} proxy, i.e. a proxy that is used by a single user or a
small group of users.

@node Latency and throughput, Network traffic, Proxies and caches, Background
@section Latency and throughput
@cindex throughput
@cindex latency

Most network benchmarks consider @dfn{throughput}, or the average
amount of data being pushed around per unit of time.  While important
for batch applications (for example benchmarks), average throughput is
mostly irrelevant when it comes to interactive web usage.  What is more
important is a transaction's median @dfn{latency}, or whether the data
starts to trickle down before the user gets annoyed.

Typical web caches optimise for throughput --- for example, by
consulting sibling caches before accessing a remote resource.  By
doing so, they significantly add to the median latency, and therefore
to the average user frustration.

Polipo was designed to minimise latency.

@node Network traffic, Partial instances, Latency and throughput, Background
@section Network traffic

The web was developed by people who were interested in text processing
rather than in networking and, accordingly, the first versions of the
HTTP protocol did not make very good use of network resources.  The
main problem in HTTP/0.9 and early versions of HTTP/1.0 was that a
separate TCP connection (``virtual circuit'' for them telecom people)
was created for every entity transferred.

Opening multiple TCP connections has significant performance
implications.  Most obviously, connection setup and teardown require
additional packet exchanges which increase network usage and, more
importantly, latency.

Less obviously, TCP is not optimised for that sort of usage.  TCP aims
to avoid network @dfn{congestion}, a situation in which the network
becomes unusable due to overly aggressive traffic patterns.  Thus, a
correct TCP implementation will very carefully probe the network at
the beginning of every connection.  A TCP connection is therefore very
slow during the first couple of kilobytes transferred, and only gets
up to speed later.  Because most HTTP entities are small (in the 1 to
10 kilobytes range), HTTP/0.9 uses TCP where it is most inefficient.

@menu
* Persistent connections::      Don't shut connections down.
* Pipelining::                  Send a bunch of requests at once.
* Poor Mans Multiplexing::      Split requests.
@end menu

@node Persistent connections, Pipelining, Network traffic, Network traffic
@subsection Persistent connections
@cindex Persistent connection
@cindex keep-alive connection

Later HTTP versions allow the transfer of multiple entities on a
single connection.  A connection that carries multiple entities is
said to be @dfn{persistent} (or sometimes @dfn{keep-alive}).
Unfortunately, persistent connections are an optional feature of HTTP,
even in version 1.1.

Polipo will attempt to use persistent connections on the server side,
and will honour persistent connection requests from clients.

@node Pipelining, Poor Mans Multiplexing, Persistent connections, Network traffic
@subsection Pipelining
@cindex Pipelining

With persistent connections it becomes possible to @dfn{pipeline} or
@dfn{stream} requests, i.e. to send multiple requests on a single
connection without waiting for the replies to come back.  Because this
technique gets the requests to the server faster, it reduces latency.
Additionally, because multiple requests can often be sent in a single
packet, pipelining reduces network traffic.

Pipelining is a fairly common technique@footnote{The X11 protocol
fundamentally relies on pipelining.  NNTP does support pipelining.
SMTP doesn't support pipelining, and ESMTP makes it an option.  FTP
does support pipelining on the control connection.}, but it is not
supported by HTTP/1.0.  HTTP/1.1 makes pipelining support compulsory
in every server implementation that can use persistent connections,
but there are a number of buggy servers that claim to implement
HTTP/1.1 but don't support pipelining.

Polipo carefully probes for pipelining support in a server and uses
pipelining if it believes that it is reliable.

Polipo also deeply enjoys being pipelined at by a
client@footnote{Other client-side implementations of HTTP that make
use of pipelining include @uref{http://www.opera.com/,,Opera}, recent
versions of @uref{http://www.mozilla.org,,Mozilla}, APT (the package
downloader used by @uref{http://www.debian.org,,Debian} GNU/Linux)
and LFTP.}.

@node Poor Mans Multiplexing,  , Pipelining, Network traffic
@subsection Poor Man's Multiplexing
@cindex Poor Man's Multiplexing
@cindex multiplexing

A major weakness of the HTTP protocol is its inability to share a
single connection between multiple simultaneous transactions --- to
@dfn{multiplex} a number of transactions over a single connection.  In
HTTP, a client can either request all instances sequentially, which
significantly increases latency, or else open multiple concurrent
connections, with all the problems that this implies
(@pxref{Persistent connections}).

Poor Man's Multiplexing (PMM) is a technique that simulates
multiplexing by requesting an instance in multiple segments; because
the segments are fetched in independent transactions, they can be
interleaved with requests for other resources.

Obviously, PMM only makes sense in the presence of persistent
connections; additionally, it is only effective in the presence of
pipelining (@pxref{Pipelining}).

PMM poses a number of reliability issues.  If the resource being
fetched is dynamic, it is quite possible that it will change between
segments; thus, an implementation making use of PMM needs to be able
to switch to full-resource retrieval when it detects a dynamic
resource.

Polipo supports PMM, but disables it by default (@pxref{PMM}).

@node Partial instances, POST and PUT, Network traffic, Background
@section Caching partial instances
@cindex partial instance
@cindex range request

A partial instance is an instance that is being cached but only part
of which is available in the local cache.  There are three ways in
which partial instances can arise: client applications requesting only
part of an instance (Adobe's Acrobat Reader plugin is famous for
that), a server dropping a connection mid-transfer (because it is
short on resources, or, surprisingly often, because it is buggy), a
client dropping a connection (usually because the user pressed
@emph{stop}).

When an instance is requested that is only partially cached, it is
possible to request just the missing data by using a feature of HTTP
known as a @dfn{range} request.  While support for range requests is
optional, most servers honour them in case of static data (data that
are stored on disk, rather then being generated on the fly e.g.@: by a
CGI script).

Caching partial instances has a number of positive effects.  Obviously,
it reduces the amount of data transmitted as the available data
needn't be fetched again.  Because it prevents partial data from being
discarded, it makes it reasonable for a proxy to unconditionally abort
a download when requested by the user, and therefore reduces network
traffic.

Polipo caches arbitrary partial instances in its in-memory cache.  It
will only store the initial segment of a partial instance (from its
beginning up to its first hole) in its on-disk cache, though.  In
either case, it will attempt to use range requests to fetch the
missing data.

@node POST and PUT, Other HTTP proxies, Partial instances, Background
@section Other requests
@cindex GET request
@cindex HEAD request
@cindex PUT request
@cindex POST request
@cindex OPTIONS request
@cindex PROPFIND request

The previous sections pretend that there is only one kind of request
in HTTP --- the @samp{GET} request.  In fact, there are some others.

The @samp{HEAD} request method retrieves data about an resource.  Polipo
does not normally use @samp{HEAD}, but will fall back to using it for
validation it if finds that a given server fails to cooperate with its
standard validation methods (@pxref{Cache transparency}).  Polipo will
correctly reply to a client's @samp{HEAD} request.

The @samp{POST} method is used to request that the server should do
something rather than merely sending an entity; it is usually used
with HTML forms that have an effect@footnote{HTML forms should use the
@samp{GET} method when the form has no side-effect as this makes the
results cacheable.}.

The @samp{PUT} method is used to replace an resource with a different
instance; it is typically used by web publishing applications.

@samp{POST} and @samp{PUT} requests are handled pretty much like
@samp{GET} and @samp{HEAD}; however, for various reasons, some
precautions must be taken.  In particular, any cached data for the
resource they refer to must be discarded, and they can never be
pipelined.

Finally, HTTP/1.1 includes a convenient backdoor with the
@samp{CONNECT} method.  For more information, please see
@ref{Tunnelling connections}.

Polipo does not currently handle the more exotic methods such as
@samp{OPTIONS} and @samp{PROPFIND}.

@node Other HTTP proxies,  , POST and PUT, Background
@section Other HTTP proxies
@cindex proxy

I started writing Polipo because the weather was bad.  But also
because I wanted to implement some features that other web proxies
don't have.

@menu
* Harvest and Squid::           Historic proxies.
* Apache::                      The web server has a proxy.
* WWWOFFLE::                    A personal proxy.
* Junkbuster::                  Get rid of ads.
* Privoxy::                     Junkbuster on speed.
* Oops::                        A multithreaded cache.
@end menu

@node Harvest and Squid, Apache, Other HTTP proxies, Other HTTP proxies
@subsection Harvest and Squid
@cindex Harvest
@cindex Squid

Harvest, the grandfather of all web caches, has since evolved into
@uref{http://www.squid-cache.org/,,Squid}.

Squid sports an elegant single-threaded non-blocking architecture and
multiplexes multiple clients in a single process.  It also features
almost complete support for HTTP/1.1, although for some reason it
doesn't currently advertise it.

Squid is designed as a large-scale shared proxy running on a dedicated
machine, and therefore embodies certain design decisions which make it
difficult to use as a personal proxy.  Because Squid keeps all
resource meta-data in memory, it requires a fair amount of RAM in
order to manipulate a reasonably sized cache.

Squid doesn't cache partial instances, and has trouble with instances
larger than available memory@footnote{Recent versions of Squid support
instances larger than available memory by using a hack that the
authors call a ``sliding window algorithm''.}.  If a client connection
is interrupted, Squid has to decide whether to continue fetching the
resource (and possibly waste bandwidth) or discard what it already has
(and possibly waste bandwidth).

Some versions of squid would, under some circumstances, pipeline up to
two outgoing requests on a single connection.  At the time of writing,
this feature appears to have been disabled in the latest version.

Squid's developers have decided to re-write it in C++.

@node Apache, WWWOFFLE, Harvest and Squid, Other HTTP proxies
@subsection The Apache proxy
@cindex Apache

The @uref{http://www.apache.org/,,Apache web server} includes a
complete HTTP/1.1 proxy.

The Apache web server was designed to maximise ease of programming ---
a decision which makes Apache immensely popular for deploying
web-based applications.  Of course, this ease of programming comes at
a cost, and Apache is not the most lightweight proxy available.

As cheaper caching proxies are available, Apache is not useful as a
standalone proxy.  The main application of Apache's proxy is to join
multiple web servers' trees into a single hierarchy.

The Apache proxy doesn't cache partial instances and doesn't pipeline
multiple outgoing requests.

@node WWWOFFLE, Junkbuster, Apache, Other HTTP proxies
@subsection WWWOFFLE
@cindex WWWOFFLE

@uref{http://www.gedanken.demon.co.uk/wwwoffle/,,WWWOFFLE}, an elegant
personal proxy, is the primary model for Polipo.

WWWOFFLE has more features than can be described here.  It will censor
banner ads, clean your HTML, decorate it with random colours, schedule
fetches for off-peak hours.

Unfortunately, the HTTP traffic that WWWOFFLE generates is disgusting.
It will open a connection for every fetch, and forces the client to do
the same.

WWWOFFLE only caches complete instances.

I used WWWOFFLE for many years, and frustration with WWWOFFLE's
limitations was the main reason why I started Polipo in the first
place.

@node Junkbuster, Privoxy, WWWOFFLE, Other HTTP proxies
@subsection Junkbuster
@cindex Junkbuster

@uref{http://internet.junkbuster.com/,,Junkbuster} is a simple
non-caching web proxy designed to remove banner ads and cookies.  It
was the main model for WWWOFFLE's (and therefore Polipo's) header and
ad-removing features.

Junkbuster's HTTP support is very simple (some would say broken): it
doesn't do persistent connections, and it breaks horribly if the
client tries pipelining.  Junkbuster is no longer being maintained,
and has evolved into Privoxy.

@node Privoxy, Oops, Junkbuster, Other HTTP proxies
@subsection Privoxy
@cindex Privoxy

@uref{http://www.privoxy.org/,,Privoxy} is the current incarnation of
Junkbuster.  Privoxy has the ability to randomly modify web pages
before sending them to the browser --- for example, remove
@samp{<blink>} or @samp{<img>} tags.

Just like its parent, Privoxy cannot do persistent connections.  Under
some circumstances, it will also buffer whole pages before sending
them to the client, which significantly adds to its latency.  However,
this is difficult to avoid given the kinds of rewriting it attempts to
perform.

@node Oops,  , Privoxy, Other HTTP proxies
@subsection Oops
@cindex Oops

@uref{http://zipper.paco.net/~igor/oops.eng/,,Oops} is a caching web
proxy that uses one thread (lightweight process) for every connection.
This technique does cost additional memory, but allows good
concurrency of requests while avoiding the need for complex
non-blocking programming.  Oops was apparently designed as a
wide-scale shared proxy.

Although Oops' programming model makes it easy to implement persistent
connections, Oops insists on opening a separate connection to the
server for every single resource fetch, which disqualifies it from
production usage.

@node Running, Network, Background, Top
@chapter Running Polipo

@menu
* Polipo Invocation::           Starting Polipo.
* Browser configuration::       Configuring your browser.
* Stopping::                    Stopping and refreshing Polipo.
* Local server::                The local web server and web interface.
@end menu

@node Polipo Invocation, Browser configuration, Running, Running
@section Starting Polipo
@cindex invocation

By default, Polipo runs as a normal foreground job in a terminal in
which it can log random ``How do you do?'' messages.  With the right
configuration options, Polipo can run as a daemon.

Polipo is run with the following command line:
@example
$ polipo [ -h ] [ -v ] [ -x ] [ -c @var{config} ] [ @var{var}=@var{val}... ]
@end example
All flags are optional.  The flag @option{-h} causes Polipo to print a
short help message and to quit.  The flag @option{-v} causes Polipo to
list all of its configuration variables and quit.  The flag
@option{-x} causes Polipo to purge its on-disk cache and then quit
(@pxref{Purging}).  The flag @option{-c} specifies the configuration
file to use (by default @file{~/.polipo} or
@file{/etc/polipo/config}).  Finally, Polipo's configuration can be
changed on the command line by assigning values to given configuration
variables.

@menu
* Configuring Polipo::          Plenty of options.
* Daemon::                      Running in the background.
* Logging::                     Funnelling status messages.
@end menu

@node Configuring Polipo, Daemon, Polipo Invocation, Polipo Invocation
@subsection Configuration
@cindex runtime configuration
@cindex variable
@cindex configuration file

There is a number of variables that you can tweak in order to
configure Polipo, and most are described in this manual
(@pxref{Variable index}).  You can display the complete, most
up-to-date list of configuration variables by using the @option{-v}
command line flag or by accessing the ``current configuration'' page
of Polipo's web interface (@pxref{Web interface}).  Configuration
variables can be set either on the command line or else in the
configuration file given by the @option{-c} command-line flag.

Configuration variables are typed, and @option{-v} will display their
types.  The type can be of one of the following:
@itemize @bullet
@item
@samp{integer} or @samp{float}: a numeric value;

@item
@samp{boolean}: a truth value, one of @samp{true} or @samp{false};

@item
@samp{tristate}: one of @samp{false}, @samp{maybe} or @samp{true};

@item
@samp{4-state}, one of @samp{false}, @samp{reluctantly},
@samp{happily} or @samp{true};

@item
@samp{5-state}, one of @samp{false}, @samp{reluctantly}, @samp{maybe},
@samp{happily} or @samp{true};

@item
@samp{atom}, a string written within double quotes @samp{"});

@item
@samp{list}, a comma-separated list of strings;

@item
@samp{intlist}, a comma-separated list of integers and ranges of
integers (of the form `@var{n}--@var{m}').
@end itemize

The configuration file has a very simple syntax.  All blank lines are
ignored, as are lines starting with a hash sign @samp{#}.  Other lines
must be of the form
@example
@var{var} = @var{val}
@end example
where @var{var} is a variable to set and @var{val} is the value to set
it to.

It is currently not possible to change the configuration of a running
Polipo without restarting it.

@node Daemon, Logging, Configuring Polipo, Polipo Invocation
@subsection Running as a daemon
@cindex daemon
@cindex terminal
@cindex pid
@vindex daemonise
@vindex pidFile
@vindex logLevel

If the configuration variable @code{daemonise} is set to true, Polipo
will run as a daemon: it will fork and detach from its controlling
terminal (if any).  The variable @code{daemonise} defaults to false.

When Polipo is run as a daemon, it can be useful to get it to
atomically write its pid to a file.  If the variable @code{pidFile} is
defined, it should be the name of a file where Polipo will write its
pid.  If the file already exists when it is started, Polipo will
refuse to run.

@node Logging,  , Daemon, Polipo Invocation
@subsection Logging
@cindex logging
@vindex logFile

When it encounters a difficulty, Polipo will print a friendly message.
The location where these messages go is controlled by the
configuration variable @code{logFile}.  If it is empty, messages go to
the error output of the process (normally the terminal); otherwise, it
is the name of a file where all output will accumulate.

The variable @code{logFile} defaults to empty if @code{daemonise} is
false, and to @samp{/var/log/polipo} otherwise.

The amount of logging is controlled by the variable @code{logLevel}.
Please see the file @samp{log.h} in the Polipo sources for the
possible values of @code{logLevel}.

@node Browser configuration, Stopping, Polipo Invocation, Running
@section Configuring your browser
@cindex browser configuration
@cindex user-agent configuration

Telling your user-agent (browser) to use Polipo is an operation that
depends on the browser.  Many user-agents will transparently use
Polipo if the environment variable @samp{http_proxy} points at it;
e.g.@:
@example
$ export http_proxy=http://localhost:8123/
@end example
Netscape Navigator, Mozilla and probably other browsers require that
you configure them manually through their @emph{Preferences} menu.

If your user-agent sports such options, tell it to use persistent
connections when speaking to proxies, to speak HTTP/1.1 and to use
HTTP/1.1 pipelining.

@node Stopping, Local server, Browser configuration, Running
@section Stopping Polipo and getting it to reload
@cindex signals
@cindex shutting down
@cindex stopping

Polipo will shut down cleanly if it receives @code{SIGHUP},
@code{SIGTERM} or @code{SIGINT} signals; this will normally happen
when a Polipo in the foreground receives a @code{^C} key press, when
your system shuts down, or when you use the @code{kill} command with
no flags.  Polipo will then write-out all its in-memory data to disk
and quit.

If Polipo receives the @code{SIGUSR1} signal, it will write out all
the in-memory data to disk (but won't discard them), reopen the log
file, and then reload the forbidden URLs file (@pxref{Forbidden}).

Finally, if Polipo receives the @code{SIGUSR2} signal, it will write
out all the in-memory data to disk and discard as much of the memory
cache as possible.  It will then reopen the log file and reload the
forbidden URLs file.

@node Local server,  , Stopping, Running
@section The local web server
@vindex localDocumentRoot
@vindex disableProxy
@vindex disableLocalInterface
@cindex web server
@cindex local server

Polipo includes a local web server, which is accessible on the same
port as the one the proxy listens to.  Therefore, by default you can
access Polipo's local web server as @samp{http://localhost:8123/}.

The data for the local web server can be configured by setting
@code{localDocumentRoot}, which defaults to
@file{/usr/share/polipo/www/}.  Setting this variable to @samp{""}
will disable the local server.

Polipo assumes that the local web tree doesn't change behind its back.
If you change any of the local files, you will need to notify Polipo
by sending it a @code{SIGUSR2} signal (@pxref{Stopping}).

If you use polipo as a publicly accessible web server, you might want
to set the variable @code{disableProxy}, which will prevent it from
acting as a web proxy.  (You will also want to set
@code{disableLocalInterface} (@pxref{Web interface}) and run Polipo in
a @emph{chroot} jail.)

@menu
* Web interface::               The web interface.
@end menu

@node Web interface,  , Local server, Local server
@subsection The web interface
@cindex runtime configuration
@cindex web interface
@vindex disableLocalInterface

The subtree of the local web space rooted at
@samp{http://localhost:8123/polipo/} is treated specially: URLs under
this root do not correspond to on-disk files, but are generated by
Polipo on-the-fly.  We call this subtree Polipo's @dfn{local web
interface}.

The page @samp{http://localhost:8123/polipo/config?} contains the
values of all configuration variables, and forms for setting some of
those.

The page @samp{http://localhost:8123/polipo/status?} provides a summary
status report about the running Polipo, currently mainly memory usage.

The page @samp{http://localhost:8123/polipo/servers?} contains the list
of known servers, and the statistics maintained about them
(@pxref{Server statistics}).

The pages starting with @samp{http://localhost:8123/polipo/index?}
contain indices of the disk cache.  For example, the following page
contains the index of the cached pages from the server of some random
company:
@example
http://localhost:8123/polipo/index?http://www.microsoft.com/
@end example

The pages starting with
@samp{http://localhost:8123/polipo/recursive-index?} contain recursive
indices of various servers.

The local interface is disabled if @code{disableLocalInterface} is true.

@node Network, Caching, Running, Top
@chapter Polipo and the network

@menu
* Client connections::          Speaking to clients
* Contacting servers::          Contacting servers.
* HTTP tuning::                 
* Offline browsing::            Browsing with poor connectivity.
* Server statistics::           Polipo keeps statistics about servers.
* PMM::                         Poor Man's Multiplexing.
* Forbidden::                   You can forbid some URLs.
* DNS::                         How Polipo finds hosts.
* Parent proxies::              Fetching data from other proxies.
* Tuning POST and PUT::         Tuning POST and PUT requests.
* Tunnelling connections::      Tunnelling foreign protocols and https.
@end menu

@node Client connections, Contacting servers, Network, Network
@section Client connections

@vindex proxyAddress
@vindex proxyPort
@vindex proxyName
@cindex address
@cindex port
@cindex IPv6
@cindex proxy loop
@cindex loop
@cindex proxy name
@cindex via
@cindex loopback address
@cindex security

There are three fundamental values that control how Polipo speaks to
clients.  The variable @code{proxyAddress}, defines the IP address on
which Polipo will listen; by default, its value is the @dfn{loopback
address} @code{"127.0.0.1"}, meaning that Polipo will listen on the
IPv4 loopback interface (the local host) only.  By setting this
variable to a global IP address or to one of the special values
@code{"::"} or @code{"0.0.0.0"}, it is possible to allow Polipo to
serve remote clients.  This is likely to be a security hole unless you
set @code{allowedClients} to a reasonable value (@pxref{Access control}).

Note that the type of address that you specify for @code{proxyAddress}
will determine whether Polipo listens to IPv4 or IPv6.  Currently, the
only way to have Polipo listen to both protocols is to specify the
IPv6 unspecified address (@code{"::"}) for
@code{proxyAddress}@footnote{This doesn't work on OpenBSD, which
breaks the IPv6 API in a misguided attempt at security.}.

The variable @code{proxyPort}, by default 8123, defines the TCP port
on which Polipo will listen.

The variable @code{proxyName}, which defaults to the host name of the
machine on which Polipo is running, defines the @dfn{name} of the
proxy.  This can be an arbitrary string that should be unique among
all instances of Polipo that you are running.  Polipo uses it in error
messages and optionally for detecting proxy loops (by using the
@samp{Via} HTTP header, @pxref{Censoring headers}).

@menu
* Access control::              Deciding who can connect.
@end menu

@node Access control,  , Client connections, Client connections
@subsection Access control
@vindex proxyAddress
@vindex authCredentials
@vindex authRealm
@vindex allowedClients
@cindex access control
@cindex authentication
@cindex loopback address
@cindex security
@cindex username
@cindex password

By making it possible to have Polipo listen on a non-routable address
(for example the loopback address), the variable @code{proxyAddress}
provides a very primitive form of @dfn{access control}: the ability to
decide which hosts are allowed to connect.

A finer form of access control can be implemented by specifying
explicitly a number of client addresses or ranges of addresses
(networks) that a client is allowed to connect from.  This is realised
by the variable @code{allowedClients}, which is a list of networks
that are allowed to connect.

Every entry in @code{allowedClients} can be an IP address, for example
@samp{134.157.168.57} or @samp{::1}.  It can also be a network
address, i.e.@: an IP address and the number of bits in the network
prefix, for example @samp{134.157.168.0/24} or
@samp{2001:660:116::/48}.  Typical uses of @samp{allowedClients}
variable include
@example
allowedClients = 127.0.0.1, ::1, 134.157.168.0/24, 2001:660:116::/48
@end example
or, for an IPv4-only version of Polipo,
@example
allowedClients = 127.0.0.1, 134.157.168.0/24
@end example

A different form of access control can be realised by requiring each
client to @dfn{authenticate}, i.e.@: to prove its identity before
connecting.  Polipo currently only implements the most insecure form
of authentication, @dfn{HTTP basic authentication}, which sends
usernames and passwords in clear over the network.  HTTP basic
authentication is used when the variable @code{authCredentials} is non
null; its value should be of the form @samp{username:password}.

Note that both IP-based authentication and HTTP basic authentication
are insecure: the former is vulnerable to IP address spoofing, the
latter to replay attacks.  If you need to access Polipo over the
public Internet, the only secure option is to have it listen over the
loopback interface only and use an ssh tunnel (@pxref{Parent
proxies})@footnote{It is not quite clear to me whether HTTP digest
authentication is worth implementing.  On the one hand, if implemented
correctly, it appears to provide secure authentication; on the other
hand, and unlike ssh, it doesn't make any attempt at ensuring privacy,
and its optional integrity guarantees are impossible to implement
without significantly impairing latency.}.

@node Contacting servers, HTTP tuning, Client connections, Network
@section Contacting servers

@cindex multiple addresses
@cindex IPv6

A server can have multiple addresses, for example if it is
@dfn{multihomed} (connected to multiple networks) or if it can speak
both IPv4 and IPv6.  Polipo will try all of a hosts addresses in turn;
once it has found one that works, it will stick to that address until
it fails again.

@menu
* Allowed ports::               
@end menu

@node Allowed ports,  , Contacting servers, Contacting servers
@subsection Allowed ports

@cindex Allowed ports
@cindex Forbidden ports
@cindex ports
@vindex allowedPorts

A TCP service is identified not only by the IP address of the machine
it is running on, but also by a small integer, the TCP @dfn{port} it
is @dfn{listening} on.  Normally, web servers listen on port 80, but
it is not uncommon to have them listen on different ports; Polipo's
internal web server, for example, listens on port 8123 by default.

The variable @code{allowedPorts} contains the list of ports that
Polipo will accept to connect to on behalf of clients; it defaults to
@samp{80-86, 1024-65535}.  Set this variable to @samp{1-65535} if your
clients (and the web pages they consult!) are fully trusted.  (The
variable @code{allowedPorts} is not considered for tunnelled
connections; @pxref{Tunnelling connections}).

@node HTTP tuning, Offline browsing, Contacting servers, Network
@section Tuning at the HTTP level
@cindex HTTP
@cindex headers

@menu
* Tuning the HTTP parser::      
* Censoring headers::           
@end menu

@node Tuning the HTTP parser, Censoring headers, HTTP tuning, HTTP tuning
@subsection Tuning the HTTP parser
@vindex laxHttpParser
@vindex bigBufferSize

As a number of HTTP servers and CGI scripts serve incorrect HTTP
headers, Polipo uses a @emph{lax} parser, meaning that incorrect HTTP
headers will be ignored (with a warning).  If the variable
@code{laxHttpParser} is not set (it is set by default), Polipo will
use a @emph{strict} parser, and refuse to serve an instance unless it
could parse all the headers.

When the amount of headers exceeds one chunk's worth (@pxref{Chunk
memory}), Polipo will allocate a @dfn{big buffer} in order to store
the headers.  The size of big buffers, and therefore the maximum
amount of headers Polipo can parse, is specified by the variable
@code{bigBufferSize} (32@dmn{kB} by default).

@node Censoring headers,  , Tuning the HTTP parser, HTTP tuning
@subsection Censoring headers
@cindex privacy
@cindex anonymity
@cindex Referer
@cindex cookies
@vindex censorReferer
@vindex censoredHeaders
@vindex proxyName
@vindex disableVia

Polipo offers the option to censor given HTTP headers in both client
requests and server replies.  The main application of this feature is
to very slightly improve the user's privacy by eliminating cookies and
some content-negotiation headers.

It is important to understand that these features merely make it
slightly more difficult to gather statistics about the user's
behaviour.  While they do not actually prevent such statistics from
being collected, they might make it less cost-effective to do so.

The general mechanism is controlled by the variable
@code{censoredHeaders}, the value of which is a case-insensitive list
of headers to unconditionally censor.  By default, it is empty, but I
recommend that you set it to the list consisting of @samp{Set-Cookie},
@samp{Cookie}, @samp{Cookie2}, @samp{From} and @samp{Accept-Language}.
You should probably not censor @samp{User-Agent}, as many sites use it
in order to customise their pages for a given implementation.

The case of the @samp{Referer}@footnote{HTTP contains many mistakes
and even one spelling error.} header is treated specially because many
sites will refuse to serve pages when it is not provided.  If
@code{censorReferer} is @code{false} (the default), @samp{Referer}
headers are passed unchanged to the server.  If @code{censorReferer}
is @code{maybe}, @samp{Referer} headers are passed to the server only
when they refer to the same host as the resource being fetched.  If
@code{censorReferer} is @code{true}, all @samp{Referer} headers are
censored.  I recommend setting @code{censorReferer} to @code{maybe}.

Another header that can have privacy implications is the @samp{Via}
header; by default, Polipo does not use it.  In order to enable its
use, set the variable @code{disableVia} @code{false}, in which case
you should probably set the @code{proxyName} variable to some random
string (@pxref{Client connections}).

@menu
* Censor Accept-Language::      Why Accept-Language is evil.
@end menu

@node Censor Accept-Language,  , Censoring headers, Censoring headers
@subsubsection Why censor Accept-Language
@cindex negotiation
@cindex content negotiation
@cindex Accept-Language

Recent versions of HTTP include a mechanism known as @dfn{content
negotiation} which allows a user-agent and a server to negotiate the
best representation (instance) for a given resource.  For example, a
server that provides both PNG and GIF versions of an image will serve
the PNG version to user-agents that support PNG, and the GIF version
to Internet Explorer.

Content negotiation requires that a client should send with every
single request a number of headers specifying the user's cultural and
technical preferences.  Most of these headers do not expose sensitive
information (who cares whether your browser supports PNG?).  The
@samp{Accept-Language} header, however, is meant to convey the user's
linguistic preferences; in some cases, this information is sufficient
to pinpoint with great precision the user's origins and even her
political or religious opinions, as for example in the case of
@samp{Accept-Language: yi}.

At any rate, @samp{Accept-Language} is not useful.  Its design is
based on the assumption that language is merely another representation
for the same information, and @samp{Accept-Language} simply carries a
prioritised list of languages, which is not enough to usefully
describe a literate user's preferences.  A typical French user, for
example, will prefer an English-language original to a French
(mis-)translation, while still wanting to see French language texts
when they are original.  Such a situation cannot be described by the
@samp{Accept-Language} header.

@node Offline browsing, Server statistics, HTTP tuning, Network
@section Offline browsing
@vindex proxyOffline
@vindex relaxTransparency
@cindex offline browsing
@cindex browsing offline
@cindex connectivity
@cindex warning
@cindex shift-click

In an ideal world, all machines would have perfect connectivity to the
network at all times and servers would never crash.  In the real
world, the values of the variables @code{proxyOffline} and
@code{relaxTransparency} can be used to moderate Polipo's eagerness to
contact remote machines.

Setting @code{proxyOffline} to @code{true}, prevents Polipo from
contacting remote servers, no matter what.  This setting is suitable
when you have no network connection whatsoever.

If @code{proxyOffline} is false, Polipo's behaviour is controlled by
the variable @code{relaxTransparency}.  If @code{relaxTransparency} is
@code{false} (the default), all stale instances are validated
(@pxref{Cache transparency}), and failures to connect are reported to
the client.  This is the default mode of operation of most other
proxies, and the least likely to surprise the user.

If @code{relaxTransparency} is @code{maybe}, all stale instances are
still validated, but a failure to connect is only reported as an error
if no data is available in the cache.  If a connection fails and stale
data is available, it is served to the client with a suitable HTTP/1.1
@samp{Warning} header.  Current user-agents do not provide visible
indication of such warnings, however, and this setting will typically
cause the browser to display stale data with no indication that
anything went wrong.  This setting is useful when you are consulting a
live web site but don't want to be bothered with failed revalidations.

If @code{relaxTransparency} is @code{true}, missing data is fetched
from remote servers, but stale data are unconditionally served with no
validation.  Client-side @samp{Cache-Control} directives are still
honoured, which means that you can force an end-to-end revalidation
from the browser's interface (typically by shift-clicking on
``reload'').  This setting is useful if you have very bad network
connectivity or are consulting a very slow web site or one that
provides wrong cache control
information@footnote{@code{www.microsoft.com} comes to mind.} and are
willing to manually revalidate pages that you suspect are stale.

@node Server statistics, PMM, Offline browsing, Network
@section Server statistics
@vindex serverExpireTime
@vindex smallRequestTime
@vindex replyUnpipelineTime
@vindex replyUnpipelineSize
@vindex maxPipelineTrain
@cindex server statistics
@cindex round-trip time
@cindex transfer rate
@cindex small request
@cindex large request
@cindex breaking pipelines

In order to decide when to pipeline requests (@pxref{Pipelining}) and
whether to perform Poor Man's Multiplexing 
(@pxref{Poor Mans Multiplexing}), Polipo needs to keep statistics
about servers.  These include the server's ability to handle
persistent connections, the server's ability to handle pipelined
requests, the round-trip time to the server, and the server's transfer
rate.  The statistics are accessible from Polipo's web interface
(@pxref{Web interface}). 

The variable @samp{serverExpireTime} (default 1 day) specifies how
long such information remains valid.  If a server has not been
accessed for a time interval of at least @code{serverExpireTime},
information about it will be discarded.

As Polipo will eventually recover from incorrect information about a
server, this value can be made fairly large.  The reason why it exists
at all is to limit the amount of memory used up by information about
servers.

The main use of server information is to decide whether to pipeline
additional requests on a connection that already has in-flight
requests.  This is controlled by the variable
@code{pipelineAdditionalRequests}; if it is @code{false}, no
additional requests will be pipelined.  If it is @code{true},
additional requests will be pipelined whenever possible.  If it is
@code{maybe} (the default), additional requests will only be pipelined
following @dfn{small} requests, where a small request one whose
download is estimated to take no more than @code{smallRequestTime}
(default 5@dmn{s}).

Sometimes, a request has been pipelined after a request that prompts a
very large reply from the server; when that happens, the pipeline
needs be broken in order to reduce latency.  A reply is @dfn{large}
and will cause a pipeline to be broken if either its size is at least
@code{replyUnpipelineSize} (default one megabyte) or else the server's
transfer rate is known and the body is expected to take at least
@code{replyUnpipelineTime} to download (default 15@dmn{s}).

The variable @code{maxPipelineTrain} defines the maximum number of
requests that will be pipelined in a single write (default 10).
Setting this variable to a very low value might (or might not) fix
interaction with some unreliable servers that the normal heuristics
are unable to detect.

@node PMM, Forbidden, Server statistics, Network
@section Poor Man's Multiplexing
@cindex Poor Man's Multiplexing
@cindex multiplexing
@vindex pmmSize
@vindex pmmFirstSize

By default, Polipo does not use Poor Man's Multiplexing (@pxref{Poor
Mans Multiplexing}).  If the variable @code{pmmSize} is set to a
positive value, Polipo will use PMM when speaking to servers that are
known to support pipelining.  It will request resources by segments of
@code{pmmSize} bytes.  The first segment requested has a size of
@code{pmmFirstSize}, which defaults to twice @code{pmmSize}.

PMM is an intrinsically unreliable technique.  Polipo makes heroic
efforts to make it at least usable, requesting that the server disable
PMM when not useful (by using the @samp{If-Range} header) and
disabling it on its own if a resource turns out to be dynamic.
Notwithstanding these precautions, unless the server
cooperates@footnote{More precisely, unless CGI scripts cooperate.},
you will see failures when using PMM, which will usually result in
blank pages and broken image icons; hitting @emph{Reload} on your
browser will usually cause Polipo to notice that something went wrong
and correct the problem.

@node Forbidden, DNS, PMM, Network
@section Forbidden and redirected URLs
@cindex forbidden
@cindex redirect
@cindex web counter
@cindex counter
@cindex web bug
@cindex bug
@cindex advertisement
@cindex web ad
@cindex banner ad

The web contains advertisements that a user-agent is supposed to
download together with the requested pages.  Not only do
advertisements pollute the user's brain, pushing them around takes
time and uses up network bandwidth.

Many so-called content providers also track user activities by using
@dfn{web bugs}, tiny embedded images that cause a server to log where
they are requested from.  Such images can be detected because they are
usually uncacheable (@pxref{Cache transparency}) and therefore logged
by Polipo by default.

Polipo can be configured to prevent certain URLs from reaching the
browser, either by returning a @emph{forbidden} error message to the
user, or by @emph{redirecting} such URLs to some other URL.

@menu
* Internal forbidden list::     
* External redirectors::        
@end menu

@node Internal forbidden list, External redirectors, Forbidden, Forbidden
@subsection Internal forbidden list
@cindex forbidden
@cindex redirect
@vindex forbiddenFile
@vindex forbiddenUrl
@vindex forbiddenRedirectCode

The file pointed at by the variable @code{forbiddenFile} (defaults to
@file{~/.polipo-forbidden} or @file{/etc/polipo/forbidden}, whichever
exists) specifies the set of URLs that should never be fetched.  If
@code{forbiddenFile} is a directory, it will be recursively searched
for files with forbidden URLs.

Every line in a file listing forbidden URLs can either be a domain
name --- a string that doesn't contain any of @samp{/}, @samp{*} or
@samp{\} ---, or a POSIX extended regular expression.  Blank lines are
ignored, as are those that start with a hash sign @samp{#}.

By default, whenever it attempts to fetch a forbidden URL, the browser
will receive a @emph{403 forbidden} error from Polipo.  Some users
prefer to have the browser display a different page or an image.

If @code{forbiddenUrl} is not null, it should represent a URL to which
all forbidden URLs will be redirected.  The kind of redirection used
is specified by @code{forbiddenRedirectCode}; if this is 302 (the
default) the redirection will be marked as temporary, if 301 it will
be a permanent one.

@node External redirectors,  , Internal forbidden list, Forbidden
@subsection External redirectors
@cindex forbidden
@cindex redirect
@cindex redirector
@cindex Squid-style redirector
@vindex redirector
@vindex redirectorRedirectCode

Polipo can also use an external process (a @dfn{Squid-style
redirector}) to determine which URLs should be redirected.  The name
of the redirector binary is determined from the variable
@code{redirector}, and the kind of redirection generated is specified
by @code{redirectorRedirectCode}, which should be 302 (the default) or
301.

@node DNS, Parent proxies, Forbidden, Network
@section The domain name service
@cindex DNS
@cindex name server
@cindex gethostbyname
@cindex resolver
@cindex IPv6
@vindex dnsMaxTimeout
@vindex dnsUseGethostbyname
@vindex dnsNameServer
@vindex dnsNegativeTtl
@vindex dnsGethostbynameTtl
@vindex dnsQueryIPv6

The low-level protocols beneath HTTP identify machines by IP
addresses, sequences of four 8-bit integers such as
@samp{199.232.41.10}@footnote{Or sequences of eight 16-bit integers if
you are running IPv6.}.  HTTP, on the other hand, and most application 
protocols, manipulate host names, strings such as @samp{www.polipo.org}.

The @dfn{domain name service} (DNS) is a distributed database that
maps host names to IP addresses.  When an application wants to make
use of the DNS, it invokes a @dfn{resolver}, a local library or
process that contacts remote name servers.

The Unix interface to the resolver is provided by the
@code{gethostbyname}(3) library call (@code{getaddrinfo}(3) on recent
systems), which was designed at a time when a host lookup consisted in
searching for one of five hosts in a @samp{HOSTS.TXT} file.  The
@code{gethostbyname} call is @dfn{blocking}, meaning that all activity
must cease while a host lookup is in progress.  When the call
eventually returns, it doesn't provide a @dfn{time to live} (TTL)
value to indicate how long the address may be cached.  For these
reasons, @code{gethostbyname} is hardly useful for programs that need
to contact more than a few hosts@footnote{Recent systems replace
@code{gethostbyname}(3) by @code{getaddrinfo}(3), which is reentrant.
While this removes one important problem that multi-threaded programs
encounter, it doesn't solve any of the other issues with
@code{gethostbyname}.}.

In order to avoid the @code{gethostname}(3)'s issues, Polipo usually
tries to speak the DNS protocol itself rather than using the system
resolver.  Its precise behaviour is controlled by the value of
@code{dnsUseGethostbyname}.  If @code{dnsUseGethostbyname} is
@code{false}, Polipo never uses the system resolver.  If it is
@code{reluctantly} (the default), Polipo tries to speak DNS and falls
back to the system resolver if a name server could not be contacted.
If it is @code{happily}, Polipo tries to speak DNS, and falls back to
the system resolver if the host couldn't be found for any reason.
Finally, if @code{dnsUseGethostbyname} is @code{true}, Polipo never
tries to speak DNS itself and uses the system resolver straight away.

If the internal DNS support is used, Polipo must be given a recursive
name server to speak to.  By default, this information is taken from
the @samp{/etc/resolv.conf} file; however, if you wish to use a
different name server, you may set the variable @code{dnsNameServer}
to an IP address@footnote{While Polipo does its own caching of DNS
data, I strongly recommend that you run a local caching name server.
I am very happy with
@uref{http://home.t-online.de/home/Moestl/,,@code{pdnsd}},
notwithstanding its somewhat bizarre handling of TCP connections.}.

When the reply to a DNS request is late to come, Polipo will retry
multiple times using an exponentially increasing timeout.  The maximum
timeout used before Polipo gives up is defined by @code{dnsMaxTimeout}
(default 60@dmn{s}); the total time before Polipo gives up on a DNS
query will be roughly twice @code{dnsMaxTimeout}.

The variable @code{dnsNegativeTtl} specifies the time during which
negative DNS information (information that a host @emph{doesn't}
exist) will be cached; this defaults to 120@dmn{s}.  Increasing this
value reduces both latency and network traffic but may cause a failed
host not to be noticed when it comes back up.

The variable @code{dnsQueryIPv6} specifies whether to query for IPv4
or IPv6 addresses.  If @code{dnsQueryIPv6} is @code{false}, only IPv4
addresses are queried.  If @code{dnsQueryIPv6} is @code{reluctantly},
both types of addresses are queried, but IPv4 addresses are preferred.
If @code{dnsQueryIPv6} is @code{happily} (the default), IPv6 addresses
are preferred.  Finally, if @code{dnsQueryIPv6} is @code{true}, only
IPv6 addresses are queried.

If the system resolver is used, the value @code{dnsGethostbynameTtl}
specifies the time during which a @code{gethostbyname} reply will be
cached (default 5 minutes).

@node Parent proxies, Tuning POST and PUT, DNS, Network
@section Parent proxies

Polipo will usually fetch instances directly from source servers as
this configuration minimises latency.  In some cases, however, it may
be useful to have Polipo fetch instances from a @dfn{parent} proxy.

Polipo can use two protocols to speak to a parent proxy: HTTP and
SOCKS.

@menu
* HTTP parent proxies::         Using an HTTP parent proxy.
* SOCKS parent proxies::        Using a SOCKS4a parent proxy.
@end menu

@node HTTP parent proxies, SOCKS parent proxies, Parent proxies, Parent proxies
@subsection HTTP parent proxies
@vindex parentProxy
@vindex parentAuthCredentials
@cindex parent proxy
@cindex upstream proxy
@cindex firewall
@cindex authentication

The variable @code{parentProxy} specifies the hostname and port number
of an HTTP parent proxy; it should have the form @samp{host:port}.

If the parent proxy requires authorisation, the username and password
should be specified in the variable @code{parentAuthCredentials} in
the form @samp{username:password}.  Only @emph{Basic} authentication
is supported, which is vulnerable to replay attacks.

The main application of the parent proxy support is to cross
firewalls.  Given a machine, say @code{trurl}, with unrestricted
access to the web, the following evades a firewall by using an
encrypted compressed @code{ssh} link:
@example
$ ssh -f -C -L 8124:localhost:8123 trurl polipo
$ polipo parentProxy=localhost:8124
@end example

@node SOCKS parent proxies,  , HTTP parent proxies, Parent proxies
@subsection SOCKS parent proxies
@cindex SOCKS
@vindex socksParentProxy
@vindex socksUserName

The variable @code{socksParentProxy} specifies the hostname and port
number of a SOCKS4a parent proxy; it should have the form
@samp{host:port}.

The user name passed to the SOCKS proxy is defined by the variable
@code{socksUserName}.

@node Tuning POST and PUT, Tunnelling connections, Parent proxies, Network
@section Tuning POST and PUT requests
@cindex POST
@cindex PUT
@vindex expectContinue

The main assumption behind the design of the HTTP protocol is that
requests are idempotent: since a request can be repeated by a client,
a server is allowed to drop a connection at any time.  This fact, more
than anything else, explains the amazing scalability of the protocol.

This assumption breaks down in the case of POST requests.  Indeed, a
POST request usually causes some action to be performed (a page to be
printed, a significant amount of money to be transferred from your
bank account, or, in Florida, a vote to be registered), and such a
request should not be repeated.

The only solution to this problem is to reserve HTTP to idempotent
activities, and use reliable protocols for action-effecting
activities.  Notwithstanding that, HTTP/1.1 makes a weak attempt at
making POST requests slightly more reliable and efficient than they
are in HTTP/1.0.

When speaking to an HTTP/1.1 server, an HTTP client is allowed to
request that the server check @emph{a priori} whether it intends to
honour a POST request.  This is done by sending @dfn{an expectation},
a specific header with the request, @samp{Expect: 100-continue}, and
waiting for either an error message or a @samp{100 Continue} reply
from the server.  If the latter arrives, the client is welcome to send
the rest of the POST request@footnote{This, of course, is only part of
the story.  Additionally, the server is not required to reply with
@samp{100 Continue}, hence the client must implement a timeout.
Furthermore, according to the obsolete RFC@tie{}2068, the server is
allowed to spontaneously send @samp{100 Continue}, so the client must
be prepared to ignore such a reply at any time.}.

Polipo's behaviour w.r.t.@: client expectations is controlled by the
variable @code{expectContinue}.  If this variable is false, Polipo
will never send an expectation to the server; if a client sends an
expectation, Polipo will fail the expectation straight away, causing
the client (if correctly implemented) to retry with no expectation.
If @code{expectContinue} is @code{maybe} (the default), Polipo will
behave in a standards-compliant manner: it will forward expectations
to the server when allowed to do so, and fail client expectations
otherwise.  Finally, if @code{expectContinue} is @code{true}, Polipo
will always send expectations whenever allowed to do so; this violates
the relevant standards, but might decrease network traffic under some
circumstances.

@node Tunnelling connections,  , Tuning POST and PUT, Network
@section Tunnelling connections
@cindex tunnel
@cindex tunnelling proxy
@cindex https
@cindex HTTP/SSL
@cindex rsync
@cindex CONNECT
@vindex tunnelAllowedPorts

Polipo is an HTTP proxy; it proxies HTTP traffic, and clients using
other protocols should establish a direct connection to the server or
use an @emph{ad hoc} proxy.

Due to wide-spread brain-damage@footnote{Sometimes known as NAT.},
however, there are some circumstances in which establishing a direct
connection to the server is not possible.  In such situations, it is
possible to have Polipo behave as a @emph{tunnelling} proxy --- a
proxy that merely forwards traffic between the client and the server
without understanding it.  Polipo enters tunnel mode when the server
requests it by using the HTTP @samp{CONNECT} method.

Most web browsers will use this technique for HTTP over SSL (sometimes
known as `https') if configured to use Polipo as their `https proxy'.
Recent versions of ssh can also use it to cross mis-configured
firewalls.

The variable @code{tunnelAllowedPorts} specifies the set of ports that
Polipo will accept to tunnel traffic to.  It defaults to @samp{22, 80,
443, 873}, meaning that Polipo will only tunnel ssh, HTTP, https and
rsync traffic.

@node Caching, Memory usage, Network, Top
@chapter Caching

@menu
* Cache transparency::          Fresh and stale data.
* Memory cache::                The in-memory cache.
* Disk cache::                  The on-disk cache.
@end menu

@node Cache transparency, Memory cache, Caching, Caching
@section Cache transparency and validation
@cindex transparent cache
@cindex cache transparency
@cindex out-of-date instances
@cindex validation
@cindex revalidation
@cindex expire
@cindex stale
@cindex fresh

If resources on a server change, it is possible for a cached instance
to become out-of date.  Ideally, a cache would be perfectly
@dfn{transparent}, meaning that it never serves an out-of-date
instance; in a universe with a finite speed of signal propagation,
however, this ideal is impossible to achieve.

If a caching proxy decides that a cached instance is new enough to
likely still be valid, it will directly serve the instance to the
client; we then say that the cache decided that the instance is
@dfn{fresh}.  When an instance is @dfn{stale} (not fresh), the cache
will check with the upstream server whether the resource has changed;
we say that the cached instance is being @dfn{revalidated}.

In HTTP/1.1, responsibility for revalidation is shared between the
client, the server and the proxy itself.  The client can override
revalidation policy by using the @samp{Cache-Control}
header@footnote{Or the obsolete @samp{Pragma} header.}; for example,
some user-agents will request end-to-end revalidation in this way when
the user shift-clicks on @emph{reload}.  The server may choose to
specify revalidation policy by using the @samp{Expires} and
@samp{Cache-Control} headers.  As to the proxy, it needs to choose a
revalidation policy for instances with neither server- nor client-side
cache control information.  Of course, nothing (except the HTTP/1.1
spec, but that is easily ignored) prevents a proxy from overriding the
client's and server's cache control directives.

@menu
* Tuning validation::           Tuning Polipo's validation behaviour.
* Tweaking validation::         Further tweaking of validation.
@end menu

@node Tuning validation, Tweaking validation, Cache transparency, Cache transparency
@subsection Tuning validation behaviour
@cindex age
@vindex maxAge
@vindex maxAgeFraction
@vindex maxExpiresAge
@vindex maxNoModifiedAge

Polipo's revalidation behaviour is controlled by a number of
variables.  In the following, an resource's @dfn{age} is the time since
it was last validated, either because it was fetched from the server
or because it was revalidated.

The policy defining when cached instances become stale in the absence
of server-provided information is controlled by the variables
@code{maxAge}, @code{maxAgeFraction}, @code{maxExpiresAge} and
@code{maxNoModifiedAge}.  If an instance has an @samp{Expires} header,
it becomes stale at the date given by that header, or when its age
becomes larger than @code{maxExpiresAge}, whichever happens first.  If
an instance has no @samp{Expires} header but has a @samp{LastModified}
header, it becomes stale when its age reaches either
@code{maxAgeFraction} of the time since it was last modified or else
the absolute value @code{maxAge}, whichever happens first.  Finally,
if an instance has neither @samp{Expires} nor @samp{Last-Modified}, it
will become stale when its age reaches @code{maxNoModifiedAge}.

@node Tweaking validation,  , Tuning validation, Cache transparency
@subsection Further tweaking of validation behaviour
@cindex uncachable
@cindex vary
@vindex cacheIsShared
@vindex mindlesslyCacheVary
@vindex uncachableFile

If @code{cacheIsShared} is false (the default), Polipo will ignore the
server-side @samp{Cache-Control} directives @samp{s-maxage} and
@samp{proxy-must-revalidate}.  This is highly desirable behaviour, but
happens to violate RFC@tie{}2616 if the cache is shared among multiple
users.

If @code{mindlesslyCacheVary} is true, the presence of a Vary header
(which indicates that content-negotiation occurred, @pxref{Censor
Accept-Language}) is ignored, and cached negotiated instances are
mindlessly returned to the client.  If it is false (the default),
negotiated instances are revalidated on every client request.

Cache transparency can be further relaxed by using the variables that
control offline browsing (@pxref{Offline browsing}).

A number of servers incorrectly mark variable resources as cachable;
such servers can be worked around in polipo by manually marking given
categories of objects as uncachable.  If @code{dontCacheCookies} is
true, all pages carrying HTTP cookies will be treated as uncachable.
If @code{dontCacheRedirects} is true, all redirects (301 and 302) will
be treated as uncachable.  Finally, if everything else fails, a list
of uncachable URLs can be given in the file specified by
@code{uncachableFile}, which has the same format as the
@code{forbiddenFile} (@pxref{Internal forbidden list}).  If not
specified, this file defaults to @samp{~/.polipo-uncachable} or
@samp{/etc/polipo/uncachable}, whichever exists.

@node Memory cache, Disk cache, Cache transparency, Caching
@section The in-memory cache

The memory cache consists of a list of HTTP and DNS objects maintained
in least-recently used order.  An index to the in-memory cache is
maintained as a (closed) hash table.

When the in-memory cache grows beyond a certain size (controlled by a
number of variables, @pxref{Memory usage}), or when a hash table
collision occurs, resources are written out to disk.

@node Disk cache,  , Memory cache, Caching
@section The on-disk cache
@cindex filesystem
@cindex NFS
@vindex diskCacheRoot
@vindex maxDiskEntries
@vindex diskCacheWriteoutOnClose
@vindex diskCacheFilePermissions
@vindex diskCacheDirectoryPermissions

The on-disk cache consists in a filesystem subtree rooted at a
location defined by the variable @code{diskCacheRoot}, by default
@code{"/var/cache/polipo/"}.  This directory should normally be
writeable, readable and seekable by the user running Polipo.  It
should normally point at a local filesystem; however, using NFSv3
should be safe in most implementations, but NFSv2 is definitely
not@footnote{Polipo assumes that @samp{open(O_CREAT | O_EXCL)} works
reliably.}.

Polipo will accept to use a read-only cache.  Note, however, that
using Polipo in this manner will prevent it from updating usage
information of on-disk objects, which will make purge control
problematic (@pxref{Purging}).

If @code{diskCacheRoot} is an empty string, no disk cache is used.

The value @code{maxDiskEntries} (32 by default) is the absolute
maximum of file descriptors held open for on-disk objects.  When this
limit is reached, Polipo will close descriptors on a
least-recently-used basis, which will cause no harm other than a
decrease in performance.  This value should be set to be larger than
the number of resources that you expect to be live at a single time;
defining the right notion of liveness is left as an exercise for the
reader.

The value @code{diskCacheWriteoutOnClose} (32@dmn{kB} by default) is
the amount of data that Polipo will write out when closing a disk
file.  Writing out data when closing a file can avoid subsequently
reopening and validating it, but causes unnecessary work if the
instance is later superseded.

The integers @code{diskCacheDirectoryPermissions} and
@code{diskCacheFilePermissions} are the Unix filesystem permissions
with which files and directories are created in the on-disk cache.

@menu
* Asynchronous writing::        Writing out data when idle.
* Purging::                     Purging the on-disk cache.
* Disk format::                 Format of the on-disk cache.
@end menu

@node Asynchronous writing, Purging, Disk cache, Disk cache
@subsection Asynchronous writing
@vindex idleTime
@vindex maxObjectsWhenIdle
@vindex maxWriteoutWhenIdle

When Polipo runs out of memory (@pxref{Limiting memory usage}), it
will start discarding instances from its memory cache.  If a disk
cache has been configured, it will write out any instance that it
discards.  Any memory allocation that prompted the purge must then
wait for the write to complete.

In order to avoid the latency hit that this causes, Polipo will
preemptively write out instances to the disk cache whenever it is
idle.  The integer @code{idleTime} specifies the time during which
Polipo will remain idle before it starts writing out random objects to
the on-disk cache; this value defaults to 30@dmn{s}.  You may want to
decrease this value for a busy cache with little memory, or increase
it if your cache is often idle and has a lot of memory.

The value @code{maxObjectsWhenIdle} (default 32) specifies the
maximum number of instances that an idle Polipo will write out without
checking whether there's any new work to do.  The value
@code{maxWriteoutWhenIdle} specifies the maximum amount of data
(default 64@dmn{kB}) that Polipo will write out without checking for new
activity.

@node Purging, Disk format, Asynchronous writing, Disk cache
@subsection Purging the on-disk cache
@cindex purging
@vindex diskCacheUnlinkTime
@vindex diskCacheTruncateTime
@vindex diskCacheTruncateSize
@vindex preciseExpiry

Polipo never removes a file in its on-disk cache except when it finds
that the instance that it represents has been superseded by a newer
version.  In order to keep the on-disk cache from growing without
bound, it is necessary to @dfn{purge} it once in a while.  Purging the
cache typically consists in removing some files, truncating large
files (@pxref{Partial instances}) or moving them to off-line storage.

Polipo itself can be used to purge its on-disk cache; this is done by
invoking Polipo with the @option{-x} flag.  However, as resources are
stored on disk in plain files (@pxref{Disk format}), any utility that
can parse the files written by Polipo can be used.

Purging the on-disk cache can be safely done when Polipo is running as
long as no file is ever modified in place; in particular, it is
@emph{not} safe to truncate a file in-place.  It @emph{is} safe to
unlink (remove, delete) files in the disk cache, or to add new files
to the cache (in which case you should perform an exclusive open).
Polipo's behaviour when invoked with the flag @option{-x} does obey
these restrictions.

For a purge to be effective, it is necessary to cause Polipo to
write-out its in-memory cache to disk (@pxref{Stopping}).
Additionally, Polipo will not necessarily notice the changed files
until it attempts to access them; thus, you will want it to discard
its in-memory cache after performing the purge.  The safe way of
performing a purge is therefore:
@example
$ kill -USR1 @var{polipo-pid}
$ sleep 1
$ polipo -x
$ kill -USR2 @var{polipo-pid}
@end example

The behaviour of the @option{-x} flag is controlled by three
configuration variables.  The variable @code{diskCacheUnlinkTime}
specifies the time during which an on-disk entry should remain unused
before it is eligible for removal; it defaults to 32 days.  

The variable @code{diskCacheTruncateTime} specifies the time for which
an on-disk entry should remain unused before it is eligible for
truncation; it defaults to 4 days and a half.  The variable
@code{diskCacheTruncateSize} specifies the size at which files are
truncated after they have not been accessed for
@code{diskCacheTruncateTime}; it defaults to 1@dmn{MB}.

Usually, Polipo uses a files modification time in order to determine
whether it is old enough to be expirable.  This heuristic can be
disabled by setting the variable @code{preciseExpiry} to true.

@node Disk format,  , Purging, Disk cache
@subsection Format of the on-disk cache
@vindex DISK_CACHE_BODY_OFFSET
@cindex on-disk file
@cindex on-disk cache

The on-disk cache consists of a collection of files, one per instance.
The format of an on-disk resource is similar to that of an HTTP
message: it starts with an HTTP status line, followed by HTTP headers,
followed by a blank line (@samp{\r\n\r\n}).  The blank line is
optionally followed by a number of binary zeroes.  The body of the
instance follows.

The headers of an on-disk file have a few minor differences with HTTP
messages.  Obviously, there is never a @samp{Transfer-Encoding} line.
A few additional headers are used by Polipo for its internal
bookkeeping:
@itemize
@item 
@samp{X-Polipo-Location}: this is the URL of the resource stored in this
file.  This is always present.

@item
@samp{X-Polipo-Date}: this is Polipo's estimation of the date at which
this instance was last validated, and is used for generating the
@samp{Age} header of HTTP messages.  This is optional, and only stored
if different from the instance's date.

@item
@samp{X-Polipo-Access}: this is the date when the instance was last
accessed by Polipo, and is used for cache purging (@pxref{Purging}).
This is optional, and is absent if the instance was never accessed.

@item
@samp{X-Polipo-Body-Offset}: the presence of this line indicates that
the blank line following the headers is followed by a number of zero
bytes.  Its value is an integer, which indicates the offset since the
beginning of the file at which the instance body actually starts.
This line is optional, and if absent the body starts immediately after
the blank line.

@end itemize

@node Memory usage, Copying, Caching, Top
@chapter Memory usage
@cindex memory

Polipo uses two distinct pools of memory, the @dfn{chunk pool} and
the @dfn{malloc pool}.

@menu
* Chunk memory::                Chunk memory.
* Malloc memory::               Malloc memory.
* Limiting memory usage::       Limiting Polipo's memory usage.
@end menu

@node Chunk memory, Malloc memory, Memory usage, Memory usage
@section Chunk memory

@vindex CHUNK_SIZE
@vindex MALLOC_CHUNKS
@cindex chunk
@cindex memory

Most of the memory used by Polipo is stored in chunks, fixed-size
blocks of memory; the size of a chunk is defined by the compile-time
constant @code{CHUNK_SIZE}, and defaults to 4096 bytes.  Chunks are
used for storing object data (bodies of instances) and for temporary
I/O buffers.  Increasing the chunk size increases performance
somewhat, but at the cost of larger granularity of allocation and
hence larger memory usage.

By default, Polipo uses a hand-crafted memory allocator based on
@code{mmap}(2) for allocating chunks; this is very slightly faster
than the stock memory allocator and limits memory fragmentation.  It
is possible to disable the chunk allocator, and use @code{malloc}(3)
for all memory allocation, by defining @code{MALLOC_CHUNKS} at compile
time; this is probably only useful for debugging.

There is one assumption made about @code{CHUNK_SIZE}:
@code{CHUNK_SIZE} multiplied by the number of bits in an
@code{unsigned int} (actually in a @code{ChunkBitmap} --- see
@file{chunk.c}) must be a multiple of the page size, which is 4096 on
most systems (8192 on Alpha).

As all network I/O will be performed in units of one to two chunks,
@code{CHUNK_SIZE} should be at least equal to your network interface's
MTU (typically 1500 bytes).  Additionally, as much I/O will be done at
@code{CHUNK_SIZE}-aligned addresses, @code{CHUNK_SIZE} should ideally
be a multiple of the page size.

In summary, 2048, 4096, 8192 and 16384 are good choices for
@code{CHUNK_SIZE}.

@node Malloc memory, Limiting memory usage, Chunk memory, Memory usage
@section Malloc allocation
@cindex malloc
@cindex memory

Polipo uses the standard @code{malloc}(3) memory allocator for
allocating small data structures (up to 100 bytes), small strings and
atoms (unique strings).

@node Limiting memory usage,  , Malloc memory, Memory usage
@section Limiting Polipo's memory usage
@cindex limiting memory
@cindex memory

Polipo is designed to work well when given little memory.  However,
because it can be used in different environments, you need to inform
it of the amount of memory it can use.

@menu
* Limiting chunk usage::        Discard objects when low on chunks.
* Limiting object usage::       Limit the number of objects.
* OS usage limits::             Don't impose OS limits.
@end menu

@node Limiting chunk usage, Limiting object usage, Limiting memory usage, Limiting memory usage
@subsection Limiting chunk usage

@vindex chunkHighMark
@vindex chunkCriticalMark
@vindex chunkLowMark
@vindex CHUNK_SIZE
@cindex memory
@cindex chunk

You can limit Polipo's usage of chunk memory by setting
@code{chunkHighMark} and @code{chunkLowMark}.

The value @code{chunkHighMark} is the absolute maximum number of bytes
of allocated chunk memory.  When this value is reached, Polipo will
try to purge objects from its in-memory cache; if that fails to free
memory, Polipo will start dropping connections.

When chunk usage falls back below @code{chunkLowMark}, Polipo will
stop discarding in-memory objects.  The value
@code{chunkCriticalMark}, which should be somewhere between
@code{chunkLowMark} and @code{chunkHighMark}, specifies the value
above which Polipo will make heroic efforts to free memory, including
punching holes in the middle of instances, but without dropping
connections.

Unless set explicitly, both @code{chunkLowMark} and
@code{chunkCriticalMark} are computed automatically from
@code{chunkHighMark}.

@node Limiting object usage, OS usage limits, Limiting chunk usage, Limiting memory usage
@subsection Limiting object usage

@vindex objectHighMark
@vindex publicObjectLowMark
@vindex objectHashTableSize

Besides limiting chunk usage, it is possible to limit Polipo's memory
usage by bounding the number of objects it keeps in memory at any given
time.  This is done with @code{objectHighMark} and
@code{publicObjectLowMark}.

The value @code{objectHighMark} is the absolute maximum of objects
held in memory (including resources and server addresses).  When the
number of in-memory public objects falls below
@code{publicObjectLowMark}, Polipo will stop writing out objects to
disk (private objects are discarded as soon as possible).

On 32-bit architectures, every object costs 108 bytes of memory, plus
storage for every globally unique header that is not handled specially
(hopefully negligible) plus an overhead of one word (4 bytes) for
every chunk of data in the object.

You may also want to change @code{objectHashTableSize}.  This is the
size of the hash table used for holding objects; it should be a power
of two and defaults to eight times @code{objectHighMark}.  Increasing
this value will reduce the number of objects being written out to disk
due to hash table collisions.  Every hash table entry costs one word.

@node OS usage limits,  , Limiting object usage, Limiting memory usage
@subsection OS usage limits
@cindex usage limit
@cindex ulimit
@cindex OOM killer

Many operating systems permit limiting a process' memory usage by
setting a @dfn{usage limit}; on most Unix-like systems, this is done
with the @option{-v} option to the @command{ulimit} command.
Typically, the effect is to cause calls to the @code{malloc} and
@code{mmap} library functions to fail.

Polipo will usually react gracefully to failures to allocate
memory@footnote{There are exactly three places in the code where Polipo
will give up and exit if out of memory; all three are extremely unlikely to
happen in practice.}.  Nonetheless, you should avoid using OS limits
to limit Polipo's memory usage.  Indeed, when it hits an OS limit,
Polipo cannot allocate the memory needed to schedule recovery from the
out-of-memory condition, and has no choice other than to drop a
connection.

Unfortunately, some operating system kernels (notably certain Linux
releases) fail to fail an allocation if no usage limit is given;
instead, they either crash when memory is exhausted, or else start
killing random processes with no advance warning.  On such systems,
imposing a (unrealistically large) usage limit on Polipo is the safe
thing to do.

@node Copying, Variable index, Memory usage, Top
@unnumbered Copying
Polipo is distributed under the following terms.  In short, you are
allowed to do anything you wish with Polipo as long as you don't deny
my right to be recognised as the author and you don't blame me if
anything goes wrong.

@quotation
Copyright @copyright{} 2003--2006 by Juliusz Chroboczek

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), 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:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", 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.
@end quotation

@node Variable index, Concept index, Copying, Top
@unnumbered Variable index
@printindex vr

@node Concept index,  , Variable index, Top
@unnumbered Concept index
@printindex cp

@bye