Revision 3, 11-Jan-2015
The Snarl Network Protocol (SNP) provides simple access to both local and remote instances of Snarl. It is TCP-based and thus the sending computer need not be running Snarl (or Windows) - so long as it can connect via TCP/IP to a remote computer running Snarl, it will be able to send notifications.
Note: this document refers to the current version of SNP, for previous versions of SNP, see the legacy SNP page.
SNP 3.0 introduces the following benefits over earlier versions of the protocol:
Communication takes place using TCP via either port 9887 or port 5233. Starting with Snarl R4.0, communication can take place over any valid TCP port, although applications may still expect communication to work across a specific port.
Typically it is down to the client to manage the connection and terminate it when no further communication is required, however in some cases Snarl may terminate the connection (due to malformed messages, security concerns, etc.).
Note: To use TCP port 5233, the destination computer must be running Snarl R2.5 or later.
The communication process is as follows:
Callback events and notifications forwarded as the result of a subscription will also be received by the client.
The client may choose to create a new socket for each request, or re-use an existing socket. If the client is expecting to handle callbacks then it should ensure the socket remains open at least until the callback is received. If a client has subscribed for notifications from a remote computer, it must leave the socket open in order to receive notifications sent by the remote computer.
An SNP 3.0 message consists of:
Additionally, the following:
Two types of message are currently defined: requests and responses.
An SNP 3.0 request is sent from a client (which may be Snarl but more likely will part of an application) to computer running Snarl. An SNP 3.0 request consists of:
The header, all request lines, and the terminator must all end with a CR/LF pair.
id/version is always required and must be "SNP/3.0"; request_type, hash_type and cypher_type are optional, however if cypher_type is specified then hash_type must also be specified.
See Security for details of supported hashing and encryption algorithms.
The following request types are defined:
request_type is blank. A standard request typically consists of a registration action, one or more event additions and one or more notification actions.
request_type is "FORWARD". A forwarded notification is one sent from a computer to another computer on the network. Forwarding does not typically involve application registration, instead the forwarding computer effectively acts as a proxy for the notification it has sent.
Each action must be on a separate line and there must be at least one action contained within the message (otherwise you'll receive a SNARL_ERROR_BAD_PACKET response).
Actions follow the standard format, however there are some specific points to note:
With SHA-256 hashing:
An SNP 3.0 response is received by a client from Snarl. An SNP 3.0 response consists of:
The header, all content lines, and the terminator all end with a CR/LF pair. Unlike an SNP 3.0 request, an SNP 3.0 response follows a MIME style format which each line of content a key/value pair separated with a colon and a space character. There are three types of SNP 3.0 response: the first two (success and failure) are typically received immediately after a request is sent; the third type (notification response) can be received at any time.
The SNP 3.0 response header is as follows:
A failure response provides the same content as a success response. In some cases, a "hint" line will indicate more specific information about the error that occurred.
Status codes are defined here.
A notification response provides the feedback to the application concerning a notification it has generated.
Authorisation ensures only applications that share a common secret (in this case, a password) can communicate with Snarl. The password is entered on the remote computer running Snarl and is used in the construction of the key and key hash on the client, consequently it is never transmitted in clear text between the client and remote computer.
Currently, three forms of hashing are supported:
Due to deficiencies identified in both MD5 and SHA-1, it is recommended that SHA-256 should always be used when communication may occur over a WAN or across the Internet in an unsecured channel.
The authorisation type used is determined by the sender and is included in the SNP 3.0 header, along with the key hash and salt value used. A new salt value should be used for each request.
The key hash is computed as follows:
Using a password of "abcdef" and random salt of "1A2B3C4D5E6F" and MD5 and the hashing algorithm:
Resulting SNP 3.0 header:
Subscribing and forwarding achieve a similar end result but are subtly different.
As the name implies, a subscription is an open-ended connection initiated by a client to a known remote computer, both running Snarl. Some examples of subscriptions:
To summarise: a subscription takes place between two instances of Snarl, is initiated by the client computer and remains in place until the client unsubscribes.
Conversely, forwarding is effectively a wrapper for an SNP 3.0 register/notify action pair. Forwarding is still initiated by a client, however it's the client that is wishing to '''send''' notifications to an instance of Snarl, not receive them. The instance of Snarl may be running on the same computer, but more often than not it will be running on a remote computer. The client may use Snarl to send the register/notify message, although it's more likely it will be an application which creates the message itself. Some examples of forwarding:
The subscriber-name parameter is optional but its use is highly recommended as it gives the remote computer (that is, the one you're subscribing to) a better indication of who you are. Future revisions of the SNP protocol may require this parameter to be specified.If the subscription request completes successfully, you'll receive a standard SNP 3.0 success response.
Depending on the remote source you're subscribing to, you may immediately start receiving notifications (e.g. from a store of buffered messages, a welcome message, etc.) - the implementation depending wholly on the intention of the remote source.
All notifications will be received to the TCP socket you issued the [subscribe] command from; closing the socket ceases the connection with the remote source.