Mozilla

• Roadmap
• Projects
• Coding
  • Module Owners
  • Hacking
  • Get the Source
  • Build It
• Testing
  • Releases
  • Nightly Builds
  • Report A Problem
• Tools
  • Bugzilla
  • Tinderbox
  • Bonsai
  • LXR
• FAQs

Abstract:
This document provides a comprehensive discussion of the manual proxy feature "No Proxy for:", including configuration issues, testing and bugs.

Audience:
Mozilla users, mozilla developers, mozilla testers.

Overview:
As browsers rapidly grew in popularity in the mid-90's, many network administrators added proxy servers. Initially, proxy servers were used for access control and performance. Many networks had limited access to the public network via proxy servers. Because the public network was small in scope and connections were slow, a caching proxy could often improve the overall performance.

This feature was originally designed as a "blacklist" of sites or domains that was within the intranet, and should not be accessed via the proxy server. Due to various limitations, this feature should be used for only the most simple blacklist scenarios. In almost all cases, PAC gives better control and flexibility.

User interface:
"No proxy for" is an optional field, part of "Manual proxy configuration".
The field is relatively short, but can hold reasonably long (2K+ characters) entries.
New profiles contain the values "localhost, 127.0.0.1", by default.

• Preferences | Advanced |  Proxies
• control-click menu for off line-online icon (network plug)

Configuration:
The no proxy list is composed of either domain elements or IPv4 address elements.
The elements are separated by either a space (" ") or a comma (",").

To block a...	put this into "No Proxy for"	...for example use...	Limitations
domain, including sub-domains	domain suffix, starting with a dot	".mozilla.org, mozilla.org"	You must use both entries to fully block a single domain.
hostname (without domain)	hostname-only (see problems below)	"localhost"	Also blocks any possible domains that start with the entry ("www.otherdomain.localhost")
hostname (with domain)	domain name	"www.mozilla.org"	Also blocks any hostnames or domains that end in the same string (other-www.mozilla.org)
IP address	IP address	"1.2.3.4"	Does not block hostnames that resolve to the IP address ("127.0.0.1" does not block "localhost")
Network	network w/ CIDR block	"10.0.0.0/8"	Does not block hostnames that resolve to the IP address range (10.0.0.0/8 is not "no proxy for intranet hostnames")
optional - port-specific	(optional) ":" + port number	"<FILTER>:81"	Only black-lists port. Only applies to one port (no support for ranges and/or multiple ports). Port-only filters ":80" or "*:80" are not supported.

Formats that are not accepted	Example
Domain filters with interior wildcards	www.*.com
IP address string prefixes	127. or 10.0.*
IP addresses with wildcards in quads	10.*.*.*

Preferences:

Name	network.proxy.no_proxy_on
Default value	localhost, 127.0.0.1

By default "localhost" and "127.0.0.1" are excluded, since most people assume these should connect to the local system.

Communicator used "network.proxy.none"

Limitations:
No IPv6 support - The backend stores IPv4 addresses as IPv6, but IPv6 formats are not supported.
Scalability - Not usable for local domains with numerous hostnames. Not usable for large number of filters (ad blocking).
Mozilla implements this feature with significant limitations, users may find that writing a PAC file is more suitable for their needs.

Technical discussion:
The most important aspect of "no proxy for" is that it uses a right-sided string compare. This algorithm works in most cases, but is also the root cause of most broken functionality. In recent versions, more logic was added for IP addresses, but the legacy problems remain for domain names.

The main problem is rooted in the DNS syntax, which allows a hostname to exist at the node level of a domain name. In other words, for the domain "mozilla.org", the valid entries would syntactically fall into three categories:

1- Subdomains (and the contents of those subdomains) of mozilla.org ("server.office.mozilla.org")
2- hostnames in mozilla.org ("www.mozilla.org")
3- node-level entries of mozilla.org ("mozilla.org").

You cannot specify a prefix that will do a suffixed-string compare that will match all three cases precisely.

".mozilla.org" will not match "mozilla.org"
"mozilla.org" will match all three cases, but also match any domains that end in the string "mozilla.org", such as "www.amozilla.org".

Testing:
Contributors can test this feature, even without a proxy server, using a "negative proxy server test". Proxy connections that fail return an error "The proxy server you have configured  cannot be found", so configure your browser to use a non-existent HTTP proxy (hostname: "imaginary", port "80"). Test each destination in an http: URL. All proxied URLs will return errors, all non-proxied connections will be attempted normally (direct connection).

Hostnames	filter	test destination	result
basic filtering unit tests (local host)	localhost	localhost	direct
	localhost.	localhost	proxy
	127.0.0.1	local host	direct
confirm the filter uses only suffix matches (hostname unit tests)	hostname	hostname	direct
	name	hostname	direct
	host	hostname	proxy
domains with numbers	3com.com	.3com.com	direct
FQDNs
hostname.domain.com	hostname.domain.com	domain.com	proxy
	hostname.domain.com	hostname.domain.com	direct
	hostname.domain.com	host.hostname.domain.com	direct
.domain.com	.domain.com	domain.com hostname.domain.com host.hostname.domain.com	direct direct direct
*.domain.com	*.domain.com		same results as ".domain.com"
	*domain.com		same results as "domain.com"
IP address
host IP address	127.0.0.1	127.0.0.1	direct
	127.0.0.0	127.0.0.1	proxy
network range	127.0.0.0/8	127.0.0.1	direct
	127/8	127.0.0.1	proxy
	127.*.*.*	127.0.0.1	proxy
	127.	127.0.0.1	proxy
Ports	1	127.0.0.1	proxy
no port	.mozilla.org	www.mozilla.org:80 www.mozilla.org:81	direct direct
port number	.mozilla.org:80	www.mozilla.org:80 www.mozilla.org	direct direct
	:81	www.mozilla.org:81	proxy

Developer notes:
The no_proxy for logic is written in C++. PAC is written in JS, so there are potential problems with feature consistency and porting. David Baron has pointed out that the original PAC code in the "classic" tree is written in C++. The PAC in C++ has not been tested in mozilla, so porting PAC in C++ forward would not be a panacea.

The relevant code lives in nsProtocolProxyService.cpp.

 nsProtocolProxyService::PrefsChanged - loads preferences
 nsProtocolProxyService::LoadFilters - parses filters
 nsProtocolProxyService::ExamineForProxy - decides to check filters if configuration is "manual"
 nsProtocolProxyService::CanUseProxy - performs URL vs. filter comparison

Notable bugs:
[meta] Proxy: "no proxy for" items 172083
Proxy: "No Proxy" w/ form based UI 80917
Proxy: "no proxy for" default domain filtering fails w/ non-FQDN (e.g., http://web/) 91587
No proxy for: support IPv6 address literals 277378
Proxy: no proxy IP entries do not block DNS resolved IPs 136789
314712 No proxy for: "hostname.domain.com" should block only "hostname.domain.com"314712
Proxy: "bypass proxy server for local addresses" (IE pref) 72444
"No proxy for" does not use FQDN wildcards "*" like IE 260883

Bugzilla sources:

the correct separator are spaces or commas. So use this-

dogwood.state.mo.us .intra.state.mo.us dor.intranet
or
dogwood.state.mo.us, .intra.state.mo.us, dor.intranet

Note that you don't need to (read shouldn't) put a * for all hosts with that 
domain ending. Corollary- a no_proxies_on entry of netscape.com will be 
applicable to all hosts ending at netscape.com including foonetscape.com but an 
entry of .netscape.com will specifically be applicable to all hosts in the 
domain of netscape.com.

---