Thursday, July 27, 2017

Remote Capture

Kismet once again supports remote capture in git-master!

Remote packet capture allows multiple systems to report packets back to a single, central Kismet server.  This makes it simple to use light-weight, and usually cheap, embedded devices as remote sensors to expand physical coverage or logical channel coverage.

Remote capture uses the same binaries Kismet uses locally to capture from data sources - for instance, 'kismet_cap_linux_wifi'.  These super-lightweight capture drivers can run on almost any openwrt/lede supported system; the latest version is a 64Kb package, making it plausible to install on almost any device, including limited devices with 4MB of flash.

Remote capture binaries use a fixed amount of RAM - buffers are allocated at the start and are not dynamically adjusted - which means each radio capture uses approximately 4MB of RAM.

Remote capture sources use the same protocol as local, traditional sources - instead of an IPC channel the communication is piped over a TCP socket.  Kismet is able to track what source saw a packet, per-source GPS, using per-source or centralized timestamps, and is able to distribute channels and control channel hopping and other behaviors of remote sources.

More info is in the README:

xx. Remote Packet Capture

    Kismet can capture from a remote source over a TCP connection.

    Kismet remote packet feeds are initiated by the same tools that Kismet uses to
    configure a local source; for example if Kismet is running on a host on IP 
    192.168.1.2, to capture from a Linux Wi-Fi device on another device you could
    use:

        $ /usr/local/bin/kismet_capture_tools/kismet_cap_linux_wifi \
            --connect 192.168.1.2:3501 --source=wlan1

    Specifically, this uses the kismet_cap_linux_wifi tool, which is by default 
    installed in `/usr/local/bin/kismet_capture_tools/`, to connect to the IP
    192.168.1.2 port 3501.

    The --source=... parameter is the same as you would use in a `source=' Kismet
    configuration file entry, or as `-c' to Kismet itself.

    By default, Kismet only allows remote packet connections from the localhost IP; 
    you must either:

    1.  Set up a tunnel, for example using SSH port forwarding, to connect the remote 
        device to the host Kismet is running on.  This is very simple to do, and adds
        security to the remote packet connection:

        $ ssh someuser@192.168.1.2 -L 3501:localhost:3501

        Then in another terminal:

        $ /usr/local/bin/kismet_capture_tools/kismet_cap_linux_wifi \
            --connect localhost:3501 --source=wlan1

        The `ssh' command places SSH in the background (using `-f'), connects to 
        the host Kismet is running on, and tunnels port 3501.

        The kismet_cap_linux_wifi command is the same as the first example, but
        connects to localhost:3501 to use the SSH port forwarding.

        Other, more elegant solutions exist for building the SSH tunnel, such 
        as `autossh'.

    2.  Kismet can be configured to accept connections on a specific interface,
        or from all IP addresses, by changing the `remote_capture_listen=' line in
        kismet.conf:

        remote_capture_listen=0.0.0.0

        would enable listening on all interfaces, while

        remote_capture_listen=192.168.1.2

        would enable listening only on the given IP (again using the above example 
        of Kismet running on 192.168.1.2).

        Remote capture *should only be enabled on interfaces on a protected LAN*.

    Additional remote capture arguments

    Kismet capture tools supporting remote capture also support the following options:

    --connect=[host]:[port]

        Connects to a remote Kismet server on [host] and port [port].  When using
        `--connect=...' you MUST specify a `--source=...' options

    --source=[source definition]

        Define a source; this is used only in remote connection mode.  The source
        definition is the same as defining a local source for Kismet via `-c' or
        the `source=' config file option.

    --disable-retry

        By default, a remote source will attempt to reconnect if the connection
        to the Kismet server is lost.

    --daemonize

        Places the capture tool in the background and daemonizes it.



A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!

Tuesday, June 6, 2017

Kismet Plugins - Web-only Plugins

The new Kismet plugin architecture is beginning to solidify, and as part of that, the ability to create plugins which only affect the Web UI is now in place.

If your plugin idea doesn't require any changes to how Kismet itself functions, and only modifies or adds to the Web UI, you can now do it as a runtime-loaded add-in with no C++ coding at all, and only minimal modification of the example Makefile.

The example code is in Kismet git under plugin-demo-webonly; you can browse it at https://github.com/kismetwireless/kismet/tree/master/plugin-demo-webonly .

All plugins, regardless of if they use C++ native code or only extend the web UI, have 2 files which must be present:
  1. Makefile, which tells the build system how to compile.  For web-only plugins this file is very simple, and the only change that needs to be made is changing the name of the plugin on the first line (PLUGIN_NAME ?= ...).  This plugin name is used to control the directory the plugin is installed into.
  2. manifest.conf, which tells Kismet about the plugin.  This is a simple config file that is very similar to the Kismet config files; it contains the name, description, author information, version information, and module information.
For a web-only plugin, the manifest only needs to define the name, version, description, and author fields.  To tell Kismet to load a JS object when the web UI is loaded, there also needs to be a 'js' parameter.

Plugins are automatically mapped into the Kismet webserver, serving content from [plugindir]/[pluginname]/httpd/ on the filesystem to the URL /plugin/[pluginname]/.  The plugin directory is automatically determined and mapped, regardless if it is installed system-wide or in a user directory; This means you can serve full content from your plugins httpd/ directory.

To dynamically load JS modules, Kismet expects a few standard conventions - the demo JS file in the example plugin shows the minimal basics, but for more information about how to automatically interface with the Kismet web UI framework, check out some of the developer docs:

  • webui_rest.md explains how to use various endpoints in Kismet to get additional data
  • webui.md explains the format of Javascript modules and how to hook into existing functions in the Kismet UI, like adding settings panels, sidebar triggers, tabbed frames, etc
  • device_data.md explains how to add additional details to the Kismet device details windows, which are shown when you click on a device.



A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!

Monday, May 22, 2017

Eliminating default passwords from Kismet

Kismet no longer uses a default password distributed in the kismet_httpd.conf config file; it will now auto-generate a random password and store it in ~/.kismet/kismet_httpd.conf

As part of this process, there are some new onboarding screens for Kismet:

Kismet will now show a local-only text alert (not sent to the browser or other message bus clients) showing the new password which has been generated.

Kismet will also show an alert that a new password has been generated, and warning existing users of the git-master code that the old config option is no longer used.


The first time you visit Kismet with your browser (or the first time for a particular browser - the setting is stored in HTML5 local storage so it is specific per browser), you'll get a welcome screen asking to take you to the settings panel.

The settings panel now describes the need to change the login.

Subsequent visits to the Kismet page will warn you if the password is invalid, and offer to take you to the settings panel.  If you're a guest on a server, or don't want to log in for some reason, these alerts can be silenced with the "Don't warn again" option.


Finally, the new settings handler can confirm the validity of the login and warn if it is not valid.

While these add a few extra steps, the added security of not having default logins potentially exposed to the Internet definitely outweighs it.

Installation-time passwords can still be set by using the httpd_username= and httpd_password= options in /usr/local/etc/kismet_httpd.conf, or in the per-user ~/.kismet/kismet_httpd.conf, and the random generated password can be changed by editing ~/.kismet/kismet_httpd.conf as well.



A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!

Monday, May 15, 2017

Fun with a new toy - Kismet on the Alftel Airbud

Alftel (http://www.alftel.com) were very kind and sent me the production rev of their Airbud platform to get Kismet running on it.

Anyone who swung by the NOC at Shmoocon might have seen the pre-production Airbud running a demo there:

The final rev is a lot more elegant!



It's an Intel platform - which means it will happily run Ubuntu, Fedora, Pentoo, and so on - with an ungodly number of mpci-e slots for radios (8x on this model):


This one is stocked with Atheros 11n 2x2 which have been the most stable so far in testing - I've had nothing but misery with the ath10k reporting bogus packets in a HT data environment, and the Intel cards can get into monitor mode but seem to have firmware issues which cause the interface to reset during tuning.


Happily, Kismet's new datasource code handles the multiple interfaces just fine, and I think it's going to ultimately be a lot more stable than the older style code.  Previously, Kismet multiplexed all the sources into a single IPC channel and controlled them from a single process; under the new model, Kismet spawns a process per interface for capture.

Some interesting things happen with this many devices - even scanning both bands and all HT channels, the coverage graph stays pretty flat - we're able to cover enough channels simultaneously that Kismet can maintain a fairly constant view of the devices:


With an estimated coverage map (which is a lot more interesting when it's live and animated) of:



With the new data source REST API it should be possible in the future, with fairly minimal coding effort, to also assign a source to lock on to specific channels when a device is highlighted  - making sure to capture as much information as possible about a specific device or AP while the rest of the interfaces continue channel hopping.

If you're interested in the Airbud HW, check out the Alftel website at http://www.alftel.com for more info.



A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!

Wednesday, May 10, 2017

Pcap over HTTP and other fun

Pcap over HTTP!

Kismet git-master now supports fetching pcap logs over the HTTP REST interface!

$ curl -s  http://kismet:kismet@localhost:2501/datasource/pcap/all_sources.pcapng | tshark -r -
    1   0.000000              → ImpathNe_a2:76:f2 (00:03:d8:a2:76:f2) (RA) 802.11 50 Acknowledgement, Flags=...P....C
    2   0.133960 06:18:d6:9d:0d:ac → Broadcast    802.11 252 Beacon frame, SN=1654, FN=0, Flags=........C, BI=100, SSID=UESC
    3   0.184984 16:18:d6:9d:0d:ac → Broadcast    802.11 254 Beacon frame, SN=2266, FN=0, Flags=........C, BI=100, SSID=UESC-N
    4   0.236459 06:18:d6:9d:0d:ac → Broadcast    802.11 252 Beacon frame, SN=1655, FN=0, Flags=........C, BI=100, SSID=UESC
    5   0.273332 D&MHoldi_a0:8f:26 → Broadcast    802.11 142 Probe Request, SN=9, FN=0, Flags=........C, SSID=                                

Not only that, but Kismet now supports the pcap-ng file format; this new format allows for multiple interfaces in a single capture file, as well as conflicting linktypes in a single capture.


There are a few trade-offs:  Programs written using traditional libpcap (tcpdump, many other utilities, and Kismet itself) can handle pcap-ng files with a single link type, but cannot currently handle mixed link types in the same capture file - so for instance mixing Wi-Fi and future non-Wi-Fi packets may cause a problem.

No worries - you can also capture packets from a single specific interface, by UUID:

$ curl -s http://kismet:kismet@localhost:2501/datasource/pcap/by-uuid/5fe308bd-0000-0000-0000-a0f3c10cbf1c/packets.pcapng | tcpdump -r -
reading from file -, link-type IEEE802_11_RADIO (802.11 plus radiotap header)
13:55:38.239867 913824498us tsft 24.0 Mb/s 2437 MHz 11g -86dBm signal -86dBm signal antenna 0 Acknowledgment RA:00:03:d8:a2:76:f2 (oui Unknown) 
13:55:38.263293 913845355us tsft 24.0 Mb/s 2437 MHz 11g -86dBm signal -86dBm signal antenna 0 Acknowledgment RA:00:03:d8:a2:76:f2 (oui Unknown) 
13:55:38.265179 913847499us tsft 1.0 Mb/s 2437 MHz 11b -95dBm signal -93dBm signal antenna 0 Beacon (A276EC) [1.0* 2.0* 5.5* 11.0* 6.0* 9.0 12.0* 18.0 Mbit] ESS CH: 6, PRIVACY
13:55:38.303905 913886656us tsft 1.0 Mb/s 2437 MHz 11b -89dBm signal -89dBm signal antenna 0 Beacon (UESC) [1.0* 2.0* 5.5* 11.0* 18.0 24.0* 36.0 54.0 Mbit] ESS CH: 6, PRIVACY

Saving pcaps dynamically

Pcap logs are sent as a constant HTTP stream, streaming the live packet data out of Kismet.  If you have configured your Kismet instance to use HTTPS, they'll be encrypted of course.  They can be send directly to tools via pipes as in the above examples, or they can be sent to a file:

$ wget --auth-no-challenge http://kismet:kismet@localhost:2501/data/all_packets.pcapng -O packets.pcapng
--2017-05-10 13:58:28--  http://kismet:*password*@localhost:2501/data/all_packets.pcapng
Resolving localhost (localhost)... 127.0.0.1
Connecting to localhost (localhost)|127.0.0.1|:2501... connected.
HTTP request sent, awaiting response... 200 OK
Length: unspecified
Saving to: ‘packets.pcapng’

packets.pcapng             [       <=>                 ]  15.41K  2.16KB/s

Better data!

Because pcap-ng allows us to retain the original link types, Kismet now logs the original radiotap data instead of converting it to PPI.  This preserves the entirety of the radiotap header - the MCS encoding data, the per-antenna signal data, everything held in the original record is now present in the output:


More coming soon...

Expect additional built-in pcap filtering endpoints coming soon - per-phy, per-bssid, and multi-bssid are on the list, as well as options in the UI to link to downloading pcaps directly.

As the new logging infrastructure evolves in Kismet pcapng will become the standard logging format for constant logging as well, with the option to split into multiple files recording a stream per source, or a single file combining all the sources.



A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!

Wednesday, April 19, 2017

Data sources - new Kismet capture code!

I've just pushed a large change into the git-master branch for Kismet:  Data sources.

Data sources replace the Kismet packet capture system (previously internally called 'capsources', truly, the most original of names).  This reworks the whole privilege separation, IPC, capture, channel control, and source definition systems.

At a high level, datasources split the capture into an independent process per interface: this allows the capture process to be written in any language which can speak a simple IPC API, and opens the door for capturing using Python (specifically for support with GNU Radio and other tools).  High-CPU capture methods (like SDR capture) will no longer interfere with Kismet in general by taking too much time in capture loops, plugins can install suid-root capture tools to do proper privilege separation, and in general adding new capture code should become much much simpler.

As part of this rewrite, the 'kismet_capture' binary is now gone - previously, this was a C++ binary launched as root by Kismet which wrapped all the C++ capture code and fed packets to Kismet.  In it's place are independent binaries for each data source type; currently kismet_cap_pcapfile which handles replaying previously logged data, and kismet_cap_linux_wifi which implements the mac80211 and legacy wireless ioctl driver interfaces.

Splitting the capture binaries up allows each binary to enable root only as needed - pcapfile does not, linux wifi does - and minimizes the code running in the root binary.

The rewrite of the Linux Wi-Fi capture engine also solves a number of lurking problems:  the ability to tune to HT and VHT channels (HT40+/- 802.11n and 80 and 160MHz 802.11ac), the rfkill system preventing us from bringing up interfaces, very long interface names on some distributions breaking the monitor vif creation, and NetworkManager reconfiguring the interface out from under us.

The new channel tuning automatically detects HT40, 80, and 160MHz channels by comparing the list of supported frequencies and band capabilities reported by mac80211; doing proper tuning to these channels helps with data capture on 11n and 11ac devices.

All of these changes have required more modifications to the config file - before upgrading to the latest git-master code, check out the README at:

https://github.com/kismetwireless/kismet/blob/master/README

which, while being rewritten, includes directions on how to prep the install and the new configuration options for sources.

The internals of the new datasource protocol are being documented at:

https://github.com/kismetwireless/kismet/blob/master/docs/dev/datasource.md

and the pure-C capture framework at

https://github.com/kismetwireless/kismet/blob/master/docs/dev/datasource_capframework.md

both docs are under heavy dev so check back for more updates soon.

There will certainly be some bugs exposed in the new system, and the GUI is still being built which will allow picking interfaces, reconfiguring them, and adding new ones while Kismet is running - more on that to come!  In the meantime, swing by IRC (#kismet on irc.freenode.net), Discord (link to the right), or ping bugs to @kismetwireless on twitter!

Over the next few days missing features (re-opening sources in error, per-source channel lists and exclusions, and similar) will be restored in data sources.

If you're interested in helping bring Kismet support for a non-linux platform up to date (BSD and OSX come to mind!) please, ping via one of the above channels and we can talk about the new capture API; with each platform using its own binary for capture, maintaining platform ports should be much simpler.



A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!

Friday, March 31, 2017

Scripting Against The Web Interface, Part 2

Different Ways To Get Devices

These examples should work with any language, but for now we'll look at using the Python class provided with Kismet; no particular reason, but I needed Python for some other tasks, it's used in GNURadio, and it's a just generally a popular language.

(If you'd like to implement a similar utility class in Ruby or any other language, please do, and consider sending it my way!)

JSON vs Msgpack

Nearly all the interfaces in Kismet support multiple types of output serialization - currently JSON and Msgpack, which is a binary packed format which can represent the same style of data.

The Python API favors the msgpack variants because pymsgpack is easy to use and quickly returns complete Python objects; there's no reason not to use JSON however.  Both will return identical data in identical structures, and can be turned into native language objects.

All of the example URLs should work on either - substitute ".json" as the extension instead of ".msgpack" if you're not using the Python KismetRest interface code.

Listing devices

There are several ways to get a list of devices from Kismet.  In general they all allow selecting specific fields, groups of fields, or simple matching.

The simplest method for fetching devices is to simply ask for a list of all of them; this will enumerate every device Kismet knows about and return a list:

/devices/all_devices.json

for instance will return all devices in the system, formatted as JSON, with full objects for all devices.

The downside to the simple method is that it enumerates all devices in the system.  This works fine for dozens or hundreds of devices, but if you're querying a system with thousands or tens of thousands of devices, it can cause significant lag and visible issues - dropped packets, an unresponsive server, etc.  (Internally, this happens because Kismet needs to lock the device list while it processes all the devices, which means new packets can't update devices as they arrive).  In the worst case this can cause Kismet to abort if it takes more than 5 seconds to process the list; on a slow system with tens of thousands of devices it can take some time.

The simplest query

Using the KismetRest client, you can get a list of devices with the device_summary() function:

#!/usr/bin/python
import KismetRest
import time

kr = KismetRest.KismetConnector("http://localhost:2501")
devices = kr.device_summary()
while 1:
    print devices
    time.sleep(1)

This will print all the devices Kismet knows about (as python objects, not the prettiest output in the world) once a second.

This method will ONLY work well for very small numbers of devices - tens to hundreds.  If your Kismet server can see thousands of devices or more, this will likely have a noticeable performance hit.

Making it better: Query by time

But how do we make this better?  The first, and easiest, way to reduce the impact is to reduce the number of devices we look at.  We can accomplish this with the device_summary_since(ts) function, which calls /devices/last-time/[ts]/devices.msgpack (or devices.json of course) endpoint with a timestamp; Kismet will only return devices which have been created or modified after the timestamp provided.

#!/usr/bin/python
import KismetRest

kr = KismetRest.KismetConnector("http://localhost:2501")

# Initialize our timestamp to 0
ts = 0

while 1:
    summary = kr.device_summary_since(ts)

    # Get our timestamp from the response!
    ts = summary['kismet.devicelist.timestamp']

    print summary['kismet.device.list']

    time.sleep(1)

By starting with a timestamp of 0 we request every device Kismet knows about (since all have been modified since the start of time).  After that first request, we save the current server timestamp (from the 'kismet.devicelist.timestamp' field) and use it to make our queries in the future; from then on we'll only get devices which have changed since the last time we looked, no matter how long we sleep in between.

This approach still has some problems; the initial query can take a LOT of processing power if there are a huge number of devices, because it still has to look up every single device.

Make it better: Simplify what we're asking for

The next way to reduce the impact on the server is to apply a simplification of the fields: This returns only the fields specified, instead of every field (and sub-field) in the device object, which can reduce the load of serialization by a factor of hundreds.

If you're familiar with SQL, this is the difference between doing a select * and doing a select field1, field2, ... ; By excluding data you don't care about you can greatly reduce the impact.

Fields can be specified by name or by path:

  • "kismet.device.base.channel" would select the channel field from the device object
  • "kismet.device.base.signal" would select the entire signal object from the device object and include it.
  • "kismet.device.base.signal/kismet.common.signal.last_signal_dbm" would select the last signal record from the signal object.  Fields selected by path are always placed in the result as the final field in the path, so our object would contain kismet.common.signal.last_signal_dbm
Fields can also be specified as pairs, where a field is renamed - if you need two fields in different paths, of the same name, they can be specified as an array of 2 elements: ["kismet.device.base.signa/kismet.common.signal.last_signal_dbm", "last_base_signal"] would rename the last_signal_dbm field.

Field simplification is done by POSTing to /devices/summary/devices.json or to /devices/last-time/[ts]/devices.json (or, of course, .msgpack).

In the Kismet Rest API, you can use the smart_summary_since(...) function:

fields = [
        "kismet.device.base.channel", # Standard channel field
        "unknown_field", # Requesting an unknown field will return a field containing integer 0
        "dot11.device/dot11.device.last_beaconed_ssid", # Complex path specification
        ["kismet.device.base.key", "dev.key"], # Fields may be renamed
        "kismet.device.base.packets.rrd" # Fields can be complex objects
        ];

while 1:
    time.sleep(1)

    # Get summary of devices
    devices = kr.smart_summary_since(sincets, fields)

Installing the Kismet Rest code

The Python utility class used by all of the python examples is in rest_examples/KismetRest and operates like a normal Python module - to install it, install like you would any other module:

$ cd rest_examples/KismetRest
$ python2 ./setup.py build
$ sudo python2 ./setup.py install

While development is ongoing it's probably a good idea to keep an eye on the changelog for the Python KismetRest interface in git and upgrade it; I'll try to make any API-breaking changes pretty explicit in the commit logs.

Next time...

That should be enough to get started with some simple scripting interfaces; the next post will talk about how to filter results with regular expressions, get info about the general state of the system, and some other fun tricks.







A huge thanks to all who support Kismet on Patreon - if you'd like to help, you can become a patron here!