cntlm - authenticating HTTP(S) proxy with TCP/IP tunneling and
cntlm [ -AaBcDdFfgHhILlMPprSsTUuvw ] [ host1 port1 | host1:port1 ] ...
Cntlm is an NTLM/NTLMv2 authenticating HTTP proxy. It takes the address
of your proxy or proxies (host1..N and port1..N) and opens a listening
socket, forwarding each request to the parent proxy (moving in a
circular list if the active parent stops working). Along the way, a
connection to the parent is created anew and authenticated or, if
available, previously cached connection is reused to achieve higher
efficiency and faster responses. When the chain is set up, cntlm
should be used as a proxy in your applications. Cntlm also integrates
transparent TCP/IP port forwarding (tunneling) through the parent
(incl. authentication). Each tunnel opens a new listening socket on the
defined local port and forwards all connections to the given host:port
behind the secondary proxy. Manual page explains how to setup cntlm
properly using configuration file or command-line arguments.
Cntlm works similarly to NTLMAPS, plus full NTLM support, a bucket of
new features and none of its shortcomings and inefficiencies. It adds
support for real keep-alive (on both sides) and it caches all
authenticated connections for reuse in subsequent requests. It can be
restarted without TIME_WAIT delay, uses just a fraction of memory
compared to NTLMAPS and by orders of magnitude less CPU. Each thread is
completely independent and one cannot block another. Cntlm has many
security/privacy features like NTLMv2 support and password protection -
it is possible to substitute password hashes (which can be obtained
using -H) for the actual password or to enter the password
interactively. If plaintext password is used, it is automatically
hashed during the startup and all its traces are removed from the
In addition to lower usage of system resources, cntlm achieves higher
throughput on a given link. By caching authenticated connections, it
acts as an HTTP accelerator; This way, the 5-way auth handshake for
each connection is transparently eliminated, providing direct access
most of the time. NTLMAPS doesn’t authenticate in parallel with the
request - instead, it first connects, sends a probe and disconnects. No
sooner than that it connects again and initiates NTLM handshake. Cntlm
also doesn’t read the whole request including HTTP body into memory, in
fact, no traffic is generated except for the exchange of headers until
the client <-> server connection is fully negotiated. Only then are the
request and response bodies forwarded, directly between client and
server sockets. This way, cntlm avoids most of the TCP/IP overhead of
similar proxies. Along with the fact that cntlm is written in optimized
C, it achieves up to fifteen times faster responses. The slower the
line, the more impact cntlm has on download speeds.
An example of cntlm compared to NTLMAPS under the same conditions:
cntlm gave avg 76 kB/s with peak CPU usage of 0.3% whereas with NTLMAPS
it was avg 48 kB/s with peak CPU at 98% (Pentium M 1.8 GHz). The
extreme difference in resource usage is one of many important benefits
for laptop use. Peak memory consumption (several complex sites, 50
paralell connections/threads; values are in KiB):
VSZ RSS CMD
3204 1436 ./cntlm -f -c ./cntlm.conf -P pid
411604 6264 /usr/share/ntlmaps/main.py -c /etc/ntlmaps/server.cfg
Inherent part of the development is profiling and memory management
screening using Valgrind. The source distribution contains a file
called valgrind.txt, where you can see the report confirming zero
leaks, no access to unallocated memory, no usage of uninitialized data
- all tracked down to each CPU instruction emulated in Valgrind’s
virtual CPU during a typical production lifetime of the proxy.
Most options can be pre-set in a configuration file. Specifying an
option more than once is not an error, but cntlm ignores all occurences
except the last one. This does not apply to options like -L, each of
which creates a new instance of some feature. Cntlm can be built with a
hardcoded configuration file (e.g. /etc/cntlm.conf), which is always
loaded, if possible. See -c option on how to override some or all of
Use -h to see available options with short description.
-A IP/mask (Allow)
Allow ACL rule. Together with -D (Deny) they are the two rules
allowed in ACL policy. It is more usual to have this in a
configuration file, but Cntlm follows the premise that you can
do the same on the command-line as you can using the config
file. When Cntlm receives a connection request, it decides
whether to allow or deny it. All ACL rules are stored in a list
in the same order as specified. Cntlm then walks the list and
the first IP/mask rule that matches the request source address
is applied. The mask can be any number from 0 to 32, where 32
is the default (that is exact IP match). This notation is also
known as CIDR. If you want to match everything, use 0/0 or an
asterix. ACLs on the command-line take precedence over those in
the config file. In such case, you will see info about that in
the log (among the list of unused options). There you can also
see warnings about possibly incorrect subnet spec, that’s when
the IP part has more bits than you declare by mask (e.g.
10.20.30.40/24 should be 10.20.30.0/24).
-a NTLMv2 | NTLM2SR | NT | NTLM | LM (Auth)
Authentication type. NTLM(v2) comprises of one or two hashed
responses, NT and LM or NTLM2SR or NTv2 and LMv2, which are
computed from the password hash. Each response uses a different
hashing algorithm; as new response types were invented, stronger
algorithms were used. When you first install cntlm, find the
strongest one which works for you (preferably using -M). Above
they are listed from strongest to weakest. Very old servers or
dedicated HW proxies might be unable to process anything but LM.
If none of those work, see compatibility flags option -F or
submit a Support Request.
IMPORTANT: Although NTLMv2 is not widely adopted (i.e.
enforced), it is supported on all Windows since NT 4.0 SP4.
That’s for a very long time! I strongly suggest you use it to
protect your credentials on-line. You should also replace
plaintext Password options with hashed Pass[NTLMv2|NT|LM]
equivalents. NTLMv2 is the most and possibly the only secure
authentication of the NTLM family.
This option enables "NTLM-to-basic", which allows you to use one
cntlm for multiple users. Please note that all security of NTLM
is lost this way. Basic auth uses just a simple encoding
algorithm to "hide" your credentials and it is moderately easy
to sniff them.
IMPORTANT: HTTP protocol obviously has means to negotiate
authorization before letting you through, but TCP/IP doesn’t
(i.e. open port is open port). If you use NTLM-to-basic and
DON’T specify some username/password in the configuration file,
you are bound to loose tunneling features, because cntlm alone
won’t know your credentials.
Because NTLM identification has at least three parts (username,
password, domain) and the basic authentication provides fields
for only two (username, password), you have to smuggle the
domain part somewhere. You can set the Domain config/cmd-line
parameter, which will then be used for all users, who don’t
specify their domain as a part of the username. To do that and
override the global domain setting, use this instead of plain
username in the password dialog: "domain\username".
Configuration file. Command-line options, if used, override its
single options or are added at the top of the list for multi
options (tunnels, parent proxies, etc) with the exception of
ACLs, which are completely overriden. Use /dev/null to disable
any config file.
-D IP/mask (Deny)
Deny ACL rule. See option -A above.
-d <domain> (Domain)
The domain or workgroup of the proxy account. This value can
also be specified as a part of the username with -u.
-F <flags> (Flags)
NTLM authentication flags. This option is rater delicate and I
do not recommend to change the default built-in values unless
you had no success with parent proxy auth and tried magic
autodetection (-M) and all possible values for the Auth option
(-a). Remember that each NT/LM hash combination requires
different flags. This option is sort of a complete "manual
override" and you’ll have to deal with it yourself.
-f Run in console as a foreground job, do not fork into background.
In this mode, all syslog messages will be echoed to the console
(on platforms which support syslog LOG_PERROR option). Though
cntlm is primarily designed as a classic UNIX daemon with
syslogd logging, it provides detailed verbose mode without
detaching from the controlling terminal; see -v. In any case,
all error and diagnostic messages are always sent to the system
-G <pattern> (ISAScannerAgent)
User-Agent matching (case insensitive) for trans-isa-scan plugin
(see -S for explanation). Positive match identifies requests
(applications) for which the plugin should be enabled without
considering the size of the download (see -S). You can use shell
wildcard characters, namely "*", "?" and "". If used without
-S or ISAScannerSize, the max_size_in_kb is internally set to
infinity, so the plugin will be active ONLY for selected User-
Agents, regardless of download size.
Gateway mode, cntlm listens on all network interfaces. Default
is to bind just loopback. That way, only local processes can
connect to cntlm. In the gateway mode though, cntlm listens on
all interfaces and is accessible to other machines on the
network. Please note that with this option the command-line
order matters when specifying proxy or tunnel local (listening)
ports. Those positioned before it will bind only loopback; those
after will be public.
IMPORTANT: All of the above applies only to local ports for
which you didn’t specify any source address. If you did, cntlm
tries to bind the given port only on the specified interface (or
rather IP address).
-H Use this option to get hashes for password-less configuration.
In this mode, cntlm prints the results and exits. You can just
copy & paste right into the config file. You ought to use this
option with explicit -u and -d, because some hashes include the
username and domain name in the calculation. Do see -a for
-h Display help (available options with a short description) and
-I Interactive password prompt. Any password settings from the
command line or config file is ignored and a password prompt is
issued. Use this option only from shell.
-L [<saddr>:]<lport>:<rhost>:<rport> (Tunnel)
Tunnel specification. The syntax is the same as in OpenSSH’s
local forwarding (-L), with a new optional prefix, saddr - the
source IP address to bind the lport to. Cntlm will listen for
incomming connections on the local port lport, forwarding every
new connection through the parent proxy to the rhost:rport
(authenticating on the go). This option can be used multiple
times for unlimited number of tunnels, with or without the saddr
option. See -g for the details concerning local port binding
when saddr is not used.
Please note that many corporate proxies do not allow connections
to ports other than 443 (https), but if you run your target
service on this port, you should be safe. Connect to HTTPS is
"always" allowed, otherwise nobody would be able to browse
https:// sites. In any case, first try if you can establish a
connection through the tunnel, before you rely on it. This
feature does the same job as tools like corkscrew(1), but
instead of communicating over a terminal, cntlm keeps it TCP/IP.
-l [<saddr>:]<lport> (Listen)
Local port for the cntlm proxy service. Use the number you have
chosen here and the hostname of the machine running cntlm
(possibly localhost) as proxy settings in your browser and/or
the environment. Most applications (including console) support
the notion of proxy to connect to other hosts. On POSIX, set the
following variables to use e.g. wget(1) without any trouble
(fill in the actual address of cntlm):
$ export ftp_proxy=http://localhost:3128
$ export http_proxy=$ftp_proxy
$ export https_proxy=$ftp_proxy
You can choose to run the proxy service on more than one port,
in such case just use this option as many times as necessary.
But unlike tunnel specification, cntlm fails to start if it
cannot bind all of the proxy service ports. Proxy service port
can also be bound selectively. Use saddr to pick source IP
address to bind the lport to. This allows you, for example, to
run the service on different ports for subnet A and B and make
it invisible for subnet C. See -g for the details concerning
local port binding when saddr is not used.
Run magic NTLM dialect detection. In this mode, cntlm tries some
known working presets against your proxy. Probe requests are
made for the specified testurl, with the strongest hashes going
first. When finished, settings for the most secure setup are
printed. Although the detection will tell you which and how to
use Auth, Flags and password-hash options, you have to configure
at least your credentials and proxy address first. You can use
-I to enter your password interactively.
-O [<saddr>:]<port_number> (SOCKS5Proxy)
Enable SOCKS5 proxy and make it listen on local port port_number
(source IP spec is also possible, as with all options). By
default, there will be no restrictions as to who can use this
service. Some clients don’t even support SOCKS5 authentication
(e.g. almost all browsers). If you wish to enforce
authentication, use -R or its equivalent option, SOCKS5User. As
with port tunneling, it is up to the parent proxy whether it
will allow connection to any requested host:port. This feature
can be used with tsocks(1) to make most TCP/IP applications go
thru the proxy rather than directly (only outgoing connections
will work, obviously). To make apps work without DNS server, it
is important that they don’t resolve themselves, but using
SOCKS. E.g. Firefox has this option available through URI
"about:config", key name network.proxy.socks_remote_dns, which
must be set to true. Proxy-unaware tsocksified apps, will have
to be configured using IP addresses to prevent them from DNS
Create a PID file pidfile upon startup. If the specified file
exists, it is truncated and overwritten. This option is
intended for use with start-stop-daemon(8) and other servicing
mechanisms. Please note that the PID file is created AFTER the
process drops its privileges and forks. When the daemon finishes
cleanly, the file is removed.
-p <password> (Password, PassNT, ...)
Proxy account password. Cntlm deletes the password from the
memory, to make it invisible in /proc or with inspection tools
like ps(1), but the preferable way of specifying password is the
configuration file. To that end, you can use Password option
(for plaintext, human readable format), or "encrypt" your
password via -H and then use PassNTLMv2, PassNT and/or PassLM.
-R <username>:<password> (SOCKS5User)
If SOCKS5 proxy is enabled, this option can make it accessible
only to those who have been authorized. It can be used several
times, to create a whole list of accounts (allowed user:pass
-S <max_size_in_kb> (ISAScannerSize)
Enables the plugin for transparent handling of the dreaded ISA
AV scanner, which returns an interactive HTTP page (displaying
the scanning progress) instead of the file/data you’ve
requested, every time it feels like scanning the contents. This
presumptuous behavior breaks every automated downloader, updater
and basically EVERY application relying on downloads (e.g. wget,
The parameter max_size_in_kb allows you to choose maximum
download size you wish to handle by the plugin (see below why
you might want that). If the file size is bigger than this,
cntlm forwards you the interactive page, effectively disabling
the plugin for that download. Zero means no limit. Use
-G/ISAScannerAgent to identify applications for which
max_size_in_kb should be ignored (forcing the plugin). It works
by matching User-Agent header and is necessary for e.g. wget,
apt-get and yum, which would fail if the response is some HTTP
page instead of requested data.
How it works: the client asks for a file, cntlm detects ISA’s
bullshit response and waits for the secret link to ISA’s cache,
which comes no sooner than the file is downloaded and scanned by
ISA. Only then can cntlm make the second request for the real
file and forward it along with correct headers to the client.
The client doesn’t timeout while waiting for it, b/c cntlm is
periodically sending an extra "keepalive" header, but the user
might get nervous not seeing the progress bar move. It’s of
course purely psychological matter, there’s no difference if
cntlm or your browser requests the scanned file - you must wait
for ISA to do it’s job and download then. You just expect to see
some progress indicator move, which is all what the ISA’s page
does: it shows HTML countdown.
If the plugin cannot parse the interactive page for some reason
(unknown formatting, etc.), it quits and the page is forwarded
to you - it’s never "lost".
The keepalive header is called ISA-Scanner and shows ISA’s
HTTP/1.1 200 OK
ISA-Scanner: 1000 of 10000
ISA-Scanner: 2000 of 10000
-r "<name>: <value>" (Header)
Header substitution. Every client’s request will be processed
and any headers defined using -r or in the configuration file
will be added to it. In case the header is already present, its
value will be replaced.
-s Serializes all requests by not using concurrent threads for
proxy (tunneling still works in parallel). This has a horrible
impact on performance and is available only for debugging
purposes. When used with -v, it yields nice sequential debug
log, where requests take turns.
Used in combination with -v to save the debug output into a
trace file. It should be placed as the first parameter on the
command line. To prevent data loss, it never overwrites an
existing file. You have to pick a unique name or manually delete
the old file.
When executed as root, do the stuff that needs such permissions
(read config, bind ports, etc.) and then immediately drop
privileges and change to uid. This parameter can be either
number or system username. If you use a number, both uid and
gid of the process will be set to this value; if you specify a
username, uid and gid will be set according to that user’s uid
and primary gid as defined in /etc/passwd. You should use the
latter, possibly using a dedicated cntlm account. As with any
daemon, you are strongly advised to run cntlm under a non-
-u <user>[@<domain>] (Username)
Proxy account/user name. Domain can be be entered as well.
-v Print debugging information. Automatically enables (-f).
-w <workstation> (Workstation)
Workstation NetBIOS name. Do not use full domain name (FQDN)
here. Just the first part. If not specified, cntlm tries to get
the system hostname and if that fails, uses "cntlm" - it’s
because some proxies require this field non-empty.
Configuration file has the same syntax as OpenSSH ssh_config. It
comprises of whitespace delimited keywords and values. Comment begins
with a hash ’#’ and can begin anywhere in the file. Everything after
the hash up until the EOL is a comment. Values can contain any
characters, including whitespace. Do not quote anything. For detailed
explanation of keywords, see appropriate command-line options.
Following keywords are available:
ACL allow rule, see -A.
Auth NTLMv2 | NTLM2SR | NT | NTLM | LM
Select any possible combination of NTLM hashes using a single
ACL deny rule, see -A.
Proxy account domain/workgroup name.
NTLM authentication flags. See -F for details.
Gateway mode. In the configuration file, order doesn’t matter.
Gateway mode applies the same to all tunnels.
Header <headername: value>
Header substitution. See -r for details and remember, no
Wildcard-enabled (*, ?, ) case insensitive User-Agent string
matching for the trans-isa-plugin. If you don’t define
ISAScannerSize, it is internally set to infinity, i.e. disabling
the plugin for all downloads except those agent-matched ones.
Enable trans-isa-scan plugin. See -S for more.
Local port number for the cntlm’s proxy service. See -l for
Proxy account password.
PassNTLMv2, PassNT, PassLM <password>
Hashes of the proxy account password (see -H and -a). When you
want to use hashes in the config (instead of plaintext
password), each Auth settings requires different options:
Settings | Requires
Auth NTLMv2 | PassNTLMv2
Auth NTLM2SR | PassNT
Auth NT | PassNT
Auth NTLM | PassNT + PassLM
Auth LM | PassLM
Parent proxy, which requires authentication. The same as proxy
on the command-line, can be used more than once to specify
unlimited number of proxies. Should one proxy fail, cntlm
automatically moves on to the next one. The connect request
fails only if the whole list of proxies is scanned and (for each
request) and found to be invalid. Command-line takes precedence
over the configuration file.
Enable SOCKS5 proxy. See -O for more.
Create a new SOCKS5 proxy account. See -R for more.
Enable/disable NTLM-to-basic authenticatoin. See -B for more.
Tunnel specification. See -L for more.
Proxy account name, without the possibility to include domain
name (’at’ sign is interpreted literally).
The hostname of your workstation.
The optional location of the configuration file is defined in the
Makefile, with the default for 1) deb/rpm package, 2) traditional
"make; make install" and 3) Windows installer being:
Cntlm has been successfully compiled and tested on both little and big
endian machines (Linux/i386 and AIX/PowerPC). For compilation details,
see README in the source distribution. Porting to any POSIX conforming
OS shouldn’t be more than a matter of the Makefile rearrangement. Cntlm
uses strictly POSIX.1-2001 interfaces with ISO C99 libc (snprintf(3)),
it is also compliant with SUSv3. Since version 0.33, cntlm supports
Windows using POSIX emulation layer Cygwin.
In the much needed NTLM-proxy departement, cntlm aims to be a drop-in
replacement for NTLMAPS. But please note that NTLM WWW auth (that is
auth to HTTP servers), when it is running without any parent proxy as a
standalone proxy server in itself, won’t probaly be implemented ever.
Even though the tasks share common NTLM authentication, they are
different things. Also, I’ve never seen any access-protected HTTP
server requiring solely NTLM without any alternative. Such a narrow-
spectrum tool can be written in Perl in a few minutes. I strive to keep
the code of cntlm simple and efficient.
This software is still BETA, so there are probably many bugs for you to
uncloak even though I’m testing every new piece of code AMAP and use
cntlm daily. I’ll be happy to fix all of them, but if you can manage,
patches would be better. ;)
To report a bug, enable the debug output, save it to a file and submit
on-line along with a detailed description of the problem and how to
reproduce it. The link can be found on the homepage.
To generate the debug tracefile correctly, first run cntlm from the
shell (command line) and make sure you can reproduce the bug. When you
will have verfied that, stop cntlm (hit Ctrl-C) and insert the
following parameters at the beginning of the command line, preserving
their order. Example:
cntlm[.exe] -T cntlmtrace.log -v -s ... the rest ...
Written by David Kubicek <dave (o) awk.cz>
Copyright © 2007 David Kubicek
Cntlm uses DES, MD4, MD5 and HMAC-MD5 routines from gnulib and Base64
routines from mutt(1).