urho3d/Source/ThirdParty/libwebsockets/lib/secure-streams

Secure Streams

Secure Streams is a networking api that strictly separates payload from any metadata. That includes the client endpoint address for the connection, the tls trust chain and even the protocol used to connect to the endpoint.

The user api just receives and transmits payload, and receives advisory connection state information.

The details about how the connections for different types of secure stream should be made are held in JSON "policy database" initially passed in to the context creation, but able to be updated from a remote copy.

Both client and server networking can be handled using Secure Streams APIS.

Secure Streams CLIENT State lifecycle

Secure Streams are created using lws_ss_create(), after that they may acquire underlying connections, and lose them, but the lifecycle of the Secure Stream itself is not directly related to any underlying connection.

Once created, Secure Streams may attempt connections, these may fail and once the number of failures exceeds the count of attempts to conceal in the retry / backoff policy, the stream reaches LWSSSCS_ALL_RETRIES_FAILED. The stream becomes idle again until another explicit connection attempt is given.

Once connected, the user code can use lws_ss_request_tx() to ask for a slot to write to the peer, when this if forthcoming the tx handler can send a message. If the underlying protocol gives indications of transaction success, such as, eg, a 200 for http, or an ACK from MQTT, the stream state is called back with an LWSSSCS_QOS_ACK_REMOTE or LWSSSCS_QOS_NACK_REMOTE.

SS Callback return handling

SS state(), rx() and tx() can indicate with their return code some common situations that should be handled by the caller.

Constant	Scope	Meaning
LWSSSSRET_TX_DONT_SEND	tx	This opportunity to send something was passed on
LWSSSSRET_OK	state, rx, tx	No error, continue doing what we're doing
LWSSSSRET_DISCONNECT_ME	state, rx	assertively disconnect from peer
LWSSSSRET_DESTROY_ME	state, rx	Caller should now destroy the stream itself
LWSSSSRET_SS_HANDLE_DESTROYED	state	Something handled a request to destroy the stream

Destruction of the stream we're calling back on inside the callback is tricky, it's preferable to return LWSSSSRET_DESTROY_ME if it is required, and let the caller handle it. But in some cases, helpers called from the callbacks may destroy the handle themselves, in that case the handler should return LWSSSSRET_SS_HANDLE_DESTROYED indicating that the handle is already destroyed.

Secure Streams SERVER State lifecycle

You can also run servers defined using Secure Streams, the main difference is that the user code must assertively create a secure stream of the server type in order to create the vhost and listening socket. When this stream is destroyed, the vhost is destroyed and the listen socket closed, otherwise it does not perform any rx or tx, it just represents the server lifecycle.

When client connections randomly arrive at the listen socket, new Secure Stream objects are created along with accept sockets to represent each client connection. As they represent the incoming connection, their lifecycle is the same as that of the underlying connection. There is no retry concept since as with eg, http servers, the clients may typically not be routable for new connections initiated by the server.

Since connections at socket level are already established, new connections are immediately taken through CREATING, CONNECTING, CONNECTED states for consistency.

Some underlying protocols like http are "transactional", the server receives a logical request and must reply with a logical response. The additional state LWSSSCS_SERVER_TXN provides a point where the user code can set transaction metadata before or in place of sending any payload. It's also possible to defer this until any rx related to the transaction was received, but commonly with http requests, there is no rx / body. Configuring the response there may look like

		/*
		 * We do want to ack the transaction...
		 */
		lws_ss_server_ack(m->ss, 0);
		/*
		 * ... it's going to be text/html...
		 */
		lws_ss_set_metadata(m->ss, "mime", "text/html", 9);
		/*
		 * ...it's going to be 128 byte (and request tx)
		 */
		lws_ss_request_tx_len(m->ss, 128);

Otherwise the general api usage is very similar to client usage.

Convention for rx and tx callback return

Function	Return	Meaning
tx	LWSSSSRET_OK	Send the amount of buf stored in *len
tx	LWSSSSRET_TX_DONT_SEND	Do not send anything
tx	LWSSSSRET_DISCONNECT_ME	Close the current connection
tx	LWSSSSRET_DESTROY_ME	Destroy the Secure Stream
rx	>=0	accepted
rx	<0	Close the current connection

JSON Policy Database

Example JSON policy... formatting is shown for clarity but whitespace can be omitted in the actual policy.

Ordering is not critical in itself, but forward references are not allowed, things must be defined before they are allowed to be referenced later in the JSON.

{
	"release": "01234567",
	"product": "myproduct",
	"schema-version": 1,
	"retry": [{
		"default": {
			"backoff": [1000, 2000, 3000, 5000, 10000],
			"conceal": 5,
			"jitterpc": 20
		}
	}],
	"certs": [{
		"isrg_root_x1": "MIIFazCCA1OgAw...AnX5iItreGCc="
	}, {
		"LEX3_isrg_root_x1": "MIIFjTCCA3WgAwIB...WEsikxqEt"
	}],
	"trust_stores": [{
		"le_via_isrg": ["isrg_root_x1", "LEX3_isrg_root_x1"]
	}],
	"s": [{
		"mintest": {
			"endpoint": "warmcat.com",
			"port": 4443,
			"protocol": "h1get",
			"aux": "index.html",
			"plugins": [],
			"tls": true,
			"opportunistic": true,
			"retry": "default",
			"tls_trust_store": "le_via_isrg"
		}
	}]
}

Release

Identifies the policy version

Product

Identifies the product the policy should apply to

Schema-version

The minimum version of the policy parser required to parse this policy

via-socks5

Optional redirect for Secure Streams client traffic through a socks5 proxy given in the format address:port, eg, 127.0.0.1:12345.

retry

A list of backoff schemes referred to in the policy

backoff

An array of ms delays for each retry in turn

conceal

The number of retries to conceal from higher layers before giving errors. If this is larger than the number of times in the backoff array, then the last time is used for the extra delays

jitterpc

Percentage of the delay times mentioned in the backoff array that may be randomly added to the figure from the array. For example with an array entry of 1000ms, and jitterpc of 20%, actual delays will be chosen randomly from 1000ms through 1200ms. This is to stop retry storms triggered by a single event like an outage becoming synchronized into a DoS.

certs

Certificates needed for validation should be listed here each with a name. The format is base64 DER, which is the same as the part of PEM that is inside the start and end lines.

trust_stores

Chains of certificates given in the certs section may be named and described inside the trust_stores section. Each entry in trust_stores is created as a vhost + tls context with the given name. Stream types can later be associated with one of these to enforce validity checking of the remote server.

Entries should be named using "name" and the stack array defined using "stack"

s

These are an array of policies for the supported stream type names.

server

SERVER ONLY: if set to true, the policy describes a secure streams server.

endpoint

CLIENT: The DNS address the secure stream should connect to.

This may contain string symbols which will be replaced with the corresponding streamtype metadata value at runtime. Eg, if the streamtype lists a metadata name "region", it's then possible to define the endpoint as, eg, ${region}.mysite.com, and before attempting the connection setting the stream's metadata item "region" to the desired value, eg, "uk".

If the endpoint string begins with +, then it's understood to mean a connection to a Unix Domain Socket, for Linux +@ means the following Unix Domain Socket is in the Linux Abstract Namespace and doesn't have a filesystem footprint. This is only supported on unix-type and windows platforms and when lws was configured with -DLWS_UNIX_SOCK=1

SERVER: If given, the network interface name or IP address the listen socket should bind to.

port

CLIENT: The port number as an integer on the endpoint to connect to

SERVER: The port number the server will listen on

protocol

CLIENT: The wire protocol to connect to the endpoint with. Currently supported streamtypes are

Wire protocol	Description
h1	http/1
h2	http/2
ws	http/1 Websockets
mqtt	mqtt 3.1.1
raw

Raw protocol is a bit different than the others in that there is no protocol framing, whatever is received on the connection is passed to the user rx callback and whatever the tx callback provides is issued on to the connection. Because tcp can be arbitrarily fragmented by any intermediary, such streams have to be regarded as an ordered bytestream that may be fragmented at any byte without any meaning in terms of message boundaries, for that reason SOM and EOM are ignored with raw.

allow_redirects

By default redirects are not followed, if you wish a streamtype to observe them, eg, because that's how it responds to a POST, set "allow_redirects": true

tls

Set to true to enforce the stream travelling in a tls tunnel

client cert

Set if the stream needs to authenticate itself using a tls client certificate. Set to the certificate index counting from 0+. The certificates are managed using lws_sytstem blobs.

opportunistic

Set to true if the connection may be left dropped except when in use

nailed_up

Set to true to have lws retry if the connection carrying this stream should ever drop.

retry

The name of the policy described in the retry section to apply to this connection for retry + backoff

timeout_ms

Optional timeout associated with streams of this streamtype.

If user code applies the lws_ss_start_timeout() api on a stream with a timeout of LWSSS_TIMEOUT_FROM_POLICY, the timeout_ms entry given in the policy is applied.

tls_trust_store

The name of the trust store described in the trust_stores section to apply to validate the remote server cert.

server_cert

SERVER ONLY: subject to change... the name of the x.509 cert that is the server's tls certificate

server_key

SERVER ONLY: subject to change... the name of the x.509 cert that is the server's tls key

swake_validity

Set to true if this streamtype is important enough for the functioning of the device that its locally-initiated periodic connection validity checks of the interval described in the associated retry / backoff selection, are important enough to wake the whole system from low power suspend so they happen on schedule.

http transport

http_method

HTTP method to use with http-related protocols, like GET or POST. Not required for ws.

http_expect

Optionally indicates that success for HTTP transactions using this streamtype is different than the default 200 - 299.

Eg, you may choose to set this to 204 for Captive Portal Detect usage if that's what you expect the server to reply with to indicate success. In that case, anything other than 204 will be treated as a connection failure.

http_fail_redirect

Set to true if you want to fail the connection on meeting an http redirect. This is needed to, eg, detect Captive Portals correctly. Normally, if on https, you would want the default behaviour of following the redirect.

http_url

Url path to use with http-related protocols

The URL path can include metatadata like this

"/mypath?whatever=${metadataname}"

${metadataname} will be replaced by the current value of the same metadata name. The metadata names must be listed in the "metadata": [ ] section.

http_auth_header

The name of the header that takes the auth token, with a trailing ':', eg

  "http_auth_header": "authorization:"

http_dsn_header

The name of the header that takes the dsn token, with a trailing ':', eg

  "http_dsn_header": "x-dsn:"

http_fwv_header

The name of the header that takes the firmware version token, with a trailing ':', eg

  "http_fwv_header": "x-fw-version:"

http_devtype_header

The name of the header that takes the device type token, with a trailing ':', eg

  "http_devtype_header": "x-device-type:"

http_auth_preamble

An optional string that precedes the auth token, eg

 "http_auth_preamble": "bearer "

auth_hexify

Convert the auth token to hex ('A' -> "41") before transporting. Not necessary if the auth token is already in printable string format suitable for transport. Needed if the auth token is a chunk of 8-bit binary.

nghttp2_quirk_end_stream

Set this to true if the peer server has the quirk it won't send a response until we have sent an END_STREAM, even though we have sent headers with END_HEADERS.

h2q_oflow_txcr

Set this to true if the peer server has the quirk it sends an maximum initial tx credit of 0x7fffffff and then later increments it illegally.

http_multipart_ss_in

Indicates that SS should parse any incoming multipart mime on this stream

http_multipart_name

Indicates this stream goes out using multipart mime, and provides the name part of the multipart header

http_multipart_filename

Indicates this stream goes out using multipart mime, and provides the filename part of the multipart header

http_multipart_content_type

The content-type to mark up the multipart mime section with if present

http_www_form_urlencoded

Indicate the data is sent in x-www-form-urlencoded form

rideshare

For special cases where one logically separate stream travels with another when using this protocol. Eg, a single multipart mime transaction carries content from two or more streams.

ws transport

ws_subprotocol

** CLIENT **: Name of the ws subprotocol to request from the server

** SERVER **: Name of the subprotocol we will accept

ws_binary

Use if the ws messages are binary

MQTT transport

mqtt_topic

Set the topic this streamtype uses for writes

mqtt_subscribe

Set the topic this streamtype subscribes to

mqtt qos

Set the QOS level for this streamtype

mqtt_keep_alive

16-bit number representing MQTT keep alive for the stream.

This is applied at connection time... where different streams may bind to the same underlying MQTT connection, all the streams should have an identical setting for this.

mqtt_clean_start

Set to true if the connection should use MQTT's "clean start" feature.

This is applied at connection time... where different streams may bind to the same underlying MQTT connection, all the streams should have an identical setting for this.

mqtt_will_topic

Set the topic of the connection's will message, if any (there is none by default).

This is applied at connection time... where different streams may bind to the same underlying MQTT connection, all the streams should have an identical setting for this.

mqtt_will_message

Set the content of the connect's will message, if any (there is none by default).

This is applied at connection time... where different streams may bind to the same underlying MQTT connection, all the streams should have an identical setting for this.

mqtt_will_qos

Set the QoS of the will message, if any (there is none by default).

This is applied at connection time... where different streams may bind to the same underlying MQTT connection, all the streams should have an identical setting for this.

mqtt_will_retain

Set to true if the connection should use MQTT's "will retain" feature, if there is a will message (there is none by default).

This is applied at connection time... where different streams may bind to the same underlying MQTT connection, all the streams should have an identical setting for this.

Loading and using updated remote policy

If the default, hardcoded policy includes a streamtype fetch_policy, during startup when lws_system reaches the POLICY state, lws will use a Secure Stream of type fetch_policy to download, parse and update the policy to use it.

The secure-streams-proxy minimal example shows how this is done and fetches its real policy from warmcat.com at startup using the built-in one.

Applying streamtype policy overlays

This is intended for modifying policies at runtime for testing, eg, to force error paths to be taken. After the main policy is processed, you may parse additional, usually smaller policy fragments on top of it.

Where streamtype names in the new fragment already exist in the current parsed policy, the settings in the fragment are applied over the parsed policy, overriding settings. There's a simple api to enable this by giving it the override JSON in one string

int
lws_ss_policy_overlay(struct lws_context *context, const char *overlay);

but there are also other apis available that can statefully process larger overlay fragments if needed.

An example overlay fragment looks like this

	{ "s": [{ "captive_portal_detect": {
		"endpoint": "google.com",
		"http_url": "/",
		"port": 80
	}}]}

ie the overlay fragment completely follows the structure of the main policy, just misses out anything it doesn't override.

Currently ONLY streamtypes may be overridden.

You can see an example of this in use in minimal-secure-streams example where --force-portal and --force-no-internet options cause the captive portal detect streamtype to be overridden to force the requested kind of outcome.

Captive Portal Detection

If the policy contains a streamtype captive_portal_detect then the type of transaction described there is automatically performed after acquiring a DHCP address to try to determine the captive portal situation.

		"captive_portal_detect": {
                        "endpoint": "connectivitycheck.android.com",
                        "port": 80,
                        "protocol": "h1",
                        "http_method": "GET",
                        "http_url": "generate_204",
                        "opportunistic": true,
                        "http_expect": 204,
			"http_fail_redirect": true
                }

Stream serialization and proxying

By default Secure Streams expects to make the outgoing connection described in the policy in the same process / thread, this suits the case where all the participating clients are in the same statically-linked image.

In this case the lws_ss_ apis are fulfilled locally by secure-streams.c and policy.c for policy lookups.

However it also supports serialization, where the SS api can be streamed over another transport such as a Unix Domain Socket connection. This suits the case where the clients are actually in different processes in, eg, Linux or Android.

In those cases, you run a proxy process (minimal-secure-streams-proxy) that listens on a Unix Domain Socket and is connected to by one or more other processes that pass their SS API activity to the proxy for fulfilment (or onward proxying).

Each Secure Stream that is created then in turn creates a private Unix Domain Socket connection to the proxy for each stream.

In this case the proxy uses secure-streams.c and policy.c as before to fulfil the inbound proxy streams, but uses secure-streams-serialize.c to serialize and deserialize the proxied SS API activity. The proxy clients define LWS_SS_USE_SSPC either very early in their sources before the includes, or on the compiler commandline... this causes the lwsss apis to be replaced at preprocessor time with lwssspc equivalents. These serialize the api action and pass it to the proxy over a Unix Domain Socket for fulfilment, the results and state changes etc are streamed over the Unix Domain Socket and presented to the application exactly the same as if it was being fulfilled locally.

To demonstrate this, some minimal examples, eg, minimal-secure-streams and mimimal-secure-streams-avs build themselves both ways, once with direct SS API fulfilment and once with Unix Domain Socket proxying and -client appended on the executable name. To test the -client variants, run minimal-secure-streams-proxy on the same machine.

Complicated scenarios with secure streams proxy

As mentioned above, Secure Streams has two modes, by default the application directly parses the policy and makes the outgoing connections itself. However when configured at cmake with

-DLWS_WITH_SOCKS5=1 -DLWS_WITH_SECURE_STREAMS=1 -DLWS_WITH_SECURE_STREAMS_PROXY_API=1 -DLWS_WITH_MINIMAL_EXAMPLES=1

and define LWS_SS_USE_SSPC when building the application, applications forward their network requests to a local or remote SS proxy for fulfilment... and only the SS proxy has the system policy. By default, the SS proxy is on the local machine and is connected to via a Unix Domain Socket, but tcp links are also possible. (Note the proxied traffic is not encrypyed by default.)

Using the configuration above, the example SS applications are built two ways, once for direct connection fulfilment (eg, ./bin/lws-minimal-secure-streams), and once with LWS_SS_USE_SSPC also defined so it connects via an SS proxy, (eg, ./bin/lws-minimal-secure-streams-client).

Testing an example scenario with SS Proxy and socks5 proxy

 [ SS application ] --- tcp --- [ socks 5 proxy ] --- tcp --- [ SS proxy ] --- internet

In this scenario, everything is on localhost, the socks5 proxy listens on :1337 and the SS proxy listens on :1234. The SS application connects to the socks5 proxy to get to the SS proxy, which then goes out to the internet

1 Start the SS proxy

Tell it to listen on lo interface on port 1234

$ ./bin/lws-minimal-secure-streams-proxy -p 1234 -i lo

2 Start the SOCKS5 proxy

$ ssh -D 1337 -N -v localhost

The -v makes connections to the proxy visible in the terminal for testing

3 Run the SS application

The application is told to make all connections via the socks5 proxy at 127.0.0.1:1337, and to fulfil its SS connections via an SS proxy, binding connections to 127.0.0.1 (ipv4 lo interface, -1), to 127.0.0.1:1234 (-a/-p).

socks_proxy=127.0.0.1:1337 ./bin/lws-minimal-secure-streams-client -p 1234 -i 127.0.0.1 -a 127.0.0.1

You can confirm this goes through the ssh socks5 proxy to get to the SS proxy and fulfil the connection.

Using static policies

If one of your targets is too constrained to make use of dynamic JSON policies, but using SS and the policies is attractive for wider reasons, you can use a static policy built into the firmware for the constrained target.

The secure-streams example "policy2c" (which runs on the build machine, not the device)

https://libwebsockets.org/git/libwebsockets/tree/minimal-examples/secure-streams/minimal-secure-streams-policy2c

accepts a normal JSON policy on stdin, and emits a C code representation that can be included directly in the firmware.

https://libwebsockets.org/git/libwebsockets/tree/minimal-examples/secure-streams/minimal-secure-streams-staticpolicy/static-policy.h

Using this technique it's possible to standardize on maintaining JSON policies across a range of devices with different contraints, and use the C conversion of the policy on devices that are too small.

The Cmake option LWS_WITH_SECURE_STREAMS_STATIC_POLICY_ONLY should be enabled to use this mode, it will not build the JSON parser (and the option for LEJP can also be disabled if you're not otherwise using it, saving an additional couple of KB).

Notice policy2c example tool must be built with LWS_ROLE_H1, LWS_ROLE_H2, LWS_ROLE_WS and LWS_ROLE_MQTT enabled so it can handle any kind of policy.

HTTP and ws serving

All ws servers start out as http servers... for that reason ws serving is handled as part of http serving, if you give the ws_subprotocol entry to the streamtype additionally, the server will also accept upgrades to ws.

To help the user code understand if the upgrade occurred, there's a special state LWSSSCS_SERVER_UPGRADE, so subsequent rx and tx can be understood to have come from the upgraded protocol. To allow separation of rx and tx handling between http and ws, there's a ss api lws_ss_change_handlers() which allows dynamically setting SS handlers.

Since the http and ws upgrade identity is encapsulated in one streamtype, the user object for the server streamtype should contain related user data for both http and ws underlying protocol identity.