Ronda Can Fatjo, 9, 1p Parc Tecnologic del Valles Cerdanyola, 08520 (SPAIN)
Phone:+34 933968800
www.voztele.com
<elias.baixas@voztele.com>
Copyright © 2006 VozTelecom Sistemas
listen_sockets
parameteras_relay_t
usageSEAS module enables OpenSER to transfer the execution logic control of a sip message to a given external entity, called the Application Server. When the OpenSER script is being executed on an incoming SIP message, invocation of the as_relay_t() function makes this module send the message along with some transaction information to the specified Application Server. The Application Server then executes some call-control logic code, and tells OpenSER to take some actions, ie. forward the message downstream, or respond to the message with a SIP repy, etc.
The module acts implements a network protocol acting as the interface between OpenSER internal API and the external Application Server entity.
There's only one relevant function, as_relay_t, exported by this module. This function receives as a parameter the name of the application server to which the message should be relaied. Every message relaied to an Application Server is automatically associated to a SIP transaction (a transaction is created for it). Just after the message is relaied to the Application Server, the script stops its execution on the message, because the control of message-processing is now in the Application Server.
In the context of SEAS module, relaying a message to an App Server, is _not_ done in SIP protocol, but in a special protocol by means of which the SEAS module and the Application Server comunicate efficiently and seamlessly. This procotol is specially designed so that a message doesn't need to be parsed again once it arrives at the Application Server. This protocol carries information regarding the internal structure of the SIP message (to avoid reparsing) and also information about the associated transaction (recall that invoking as_relay_t indirectly calls t_newtran). This way, all the SIP-Transaction machinery, and the SIP-Message parsing, is handled at the OpenSER core, while the execution of the Application Logic is carried in the Application Server.
The SEAS module and protocol provide a means by which an external entity can utilize OpenSER as a transaction-stateful SIP-stack to act on behalf of it. This means that this external entity (which we call the Application Server) is notified whenever a SIP-Request enters OpenSER, and this external entity can then order OpenSER to execute some actions, either replying the request, or generating new UAC transactions.
This version of SEAS works with VozTelecom's WeSIP Application Server. This Application Server is a SipServlet JAVA Container.
When OpenSER starts and SEAS module is loaded, a new process is spawn which listens on a server-socket (IP and port are specified as a parameter in the config script). From then on, the Application Servers can connect to that socket so that OpenSER can relay messages to them. When an Application Server connects to the socket, it sends its name through the socket, so every App Server is identified with a name. Within the OpenSER script, invoking as_relay_t() receives a string as a parameter, which specifies the name of an application server to which the message has to be sent. If that concrete application server hasn't already connected to the module, the function returns a negative value, otherwise (the Application Server is connected), the message is relaied to it.
SEAS module relies on the Transaction Module (TM module) for operation.
Using the SEAS module requires to have an Application Server running and connected to a particular instance of OpenSER.
At the moment, the only Application Server that works with SEAS is WeSIP Application Server, which can be downloaded from www.wesip.eu, and used freely for non-comercial purposes.
listen_sockets
(string)The listen_sockets string tells SEAS where to listen for incoming connections of Application Servers. It has the form: "ip:port". SEAS will open two server-sockets on that IP, at the specified port, and another at port+1. Application Servers must be configured to connect to that port.
In case this parameter is ommited, SEAS listens on the default IP which OpenSER is using, and opens the ports 5080 and 5081 to listen for Application Servers.
as_relay_t(String
name)
Creates a new transaction (if it isn't already created) and sends the SIP Request and transaction information to the Application Server specified in the parameter. Every Application Server connected to OpenSER through the SEAS module, must be identified with a different name.
This function can be used within REQUEST_ROUTE.
Example 1-2. as_relay_t
usage
... if (!as_relay_t("app_server_1")) { log("Error sending to app server"); t_reply("500","App Server not connected"); } ...
In case the Application Server is connected to OpenSER, the function does _not_ return, the Application Server is now in charge of processing the request, and it may then reply to the request, initiate new transactions, or whatever the application being executed wants.
In case the Application Server identified by the string parameter passed to as_relay_t() is not connected to OpenSER, the function returns 0, so that the script can continue processing the request.
At the moment, the only Application Server known to work with SEAS is WeSIP. You can download a copy from www.wesip.eu.
WeSIP is a converged Sip/Http Servlet Container.
Servlets are pieces of code that encapsulate the logic of an application. Servlets are deployed into an Application Server. Whenever a user requests service, the Application Server processes the request, and passes control to the servlet. The servlet then executes some logic, may it be a query to a database, the execution of a business process, the creation of customized content for the user, or whatever the service programmer could imagine. When the servlet finishes the execution, it creates a response and gives it back to the Application Server, which is in charge of making it reach back to the user. The Application Server implements the network protocol, it takes care of everything needed for a proper communication between user and server, so the servlet doesn’t have to care about these things. The servlet uses a set of resources from the Application Server, such as Session management, service routing or chaining, and request/response header composition.
In HttpServlets, a service programmer has to implement a method in a JAVA class, which could be called doGet() or doPost(). Whenever an HTTP request arrived at the server, one of these functions was called with the request as a parameter, so the logic of the application was executed over that particular request.
HttpServlet has been extensively used over the past years, in all kinds of business and web services.
This is how a typical HttpServlet looks like:
Example 1-3. Typical example of an HttpServlet
public final class Hello extends HttpServlet { protected void doGet(HttpServletRequest request,HttpServletResponse response) throws IOException, ServletException { response.setContentType("text/html"); PrintWriter writer = response.getWriter(); writer.println("<html>"); writer.println("<head>"); writer.println("<title>Sample Application Servlet</title>"); writer.println("</head>"); writer.println("<body bgcolor=white>"); writer.println("<table border=\"0\" width=\"100%\">"); Enumeration names = request.getHeaderNames(); while (names.hasMoreElements()) { String name = (String) names.nextElement(); writer.println("<tr>"); writer.println("<th align=\"right\">"+name+":</th>"); writer.println("<td>"+request.getHeader(name)+"</td>"); writer.println("</tr>"); } writer.println("</table>"); writer.println("</body>"); writer.println("</html>"); } }
The successor of HttpServlet for SIP networks, is the SipServlet API. Making most of the success of HttpServlet, the SipServlet API follows the same programming paradigm, so that SIP application programmers can reuse their knowledge in the field.
SipServlet API works the same way as HttpServlet: an Application Server implements a SIP Stack and executes all the complex protocol logic. It receives and pre-processes the requests from the network, and at the right moment, passes control to the servlet doXxx() method, where the programmer implemented the application logic. Depending on what kind of SIP Message it was, a method or another will be executed. For example, if an INVITE is received, the doInvite() method will be invoked in the servlet.
The application can then access all the parts of the request and do its work. When the service has been executed, it passes control back to the Application Server with a response, so that it can be forwarded to the user, and the service be satisfied.
Sip Servlets can be used to implement basic SIP network functionalities (such as Proxy or Registrar servers), but their true power emerges in the implementation of value-added services, which greatly surpasses the basic service functionality of plain SIP servers.
Examples of value-added services, are Virtual PBX or IPCentrex, Attended call forwarding, Instant Messaging, etc.
This is the appearance a typical SipServlet:
Example 1-4. Typical Sip Servlet Example
public class ProxyServlet extends SipServlet { protected void doInvite(SipServletRequest req) throws ServletException, IOException { if (req.isInitial()) { Proxy proxy = req.getProxy(); proxy.setRecordRoute(false); proxy.setParallel(parallel); proxy.setSupervised(supervised); SipURI rrURI = proxy.getRecordRouteURI(); rrURI.setParameter("foo", "bar"); req.setContent("Method is INVITE", "text/plain"); proxy.proxyTo(uris); } else { log("re-INVITE"); } } protected void doAck(SipServletRequest req) throws ServletException, IOException { log("doAck " + req.getRequestURI()); if (req.isInitial()) { throw new ServletException("unexpectedly got initial ACK");
It is a very straightforward way of programming services, and the Servlet API provides very easy and powerful means to access information about the SIP-session or Http-session, about the request or response, about the state of the dialog, or whatever it is needed.
The application programmer has a rich framework of resources that allow him to focus only on the service logic, without having to worry about the underlying protocol specifics (SIP or HTTP).
The Servlet programming language is JAVA, which offers a wide spectrum of programming API’s dealing with all kinds of techniques, tools and resources, which also are available seamlessly from the Servlet context.This makes the SipServlet API very desirable for all kinds application developers.
SipServlet allows a rapid SIP application development and deployment, and also provides a reliable and secure framework of service execution (the JAVA sandbox and the Application Server execution environment).
SipServlets achieve the most of it when they can be deployed along with HttpServlets, in the same Application Server (also known as Servlet Container). This environment truly realizes the power of converged voice/data networks: Http protocol represents one of the most powerful data transmission protocols used in modern networks (think of the SOAP web-services protocol), and SIP is the protocol of choice in most of the modern and future voice over IP (VoIP) networks for the signaling part. So an Application Server capable of combining and leveraging the power of these two APIs will be the most successful.
Convergence of SIP and HTTP protocols into the same Application Server offers, amongst others, the following key advantages:
-It doesn’t require to have 2 different servers (Http and Sip) so it relieves from maintenance problems, and eases user and configuration provisioning.
-It offers great convenience to the application programmer to have all the classes related to the different protocols handled within the same code.
-As it eases development of interactive and multimedia services, realizing the power of well-known web-services and intermixing them with new voice services.
These are some simple, but suggestive examples of services that could be developed within a converged Http/Sip servlet:
-IP Centrex: through the use of the Web-interface, users could have a layout of the office in a web page, and see what phones were ringing at a given moment, so they could pick-up a call ringing in another phone in their own desktop. Or they could forward a call to another party by clicking on the web page and selecting which of the office phones it had to be transferred to.
-Voicemail: users could upload an audio file to the server through a web-page, to be used as the automatic answering message, and then also download their voicemail through the web-page, or organize the messages and remove the old ones.
-Instant Messaging: users could continue a voice call by starting or joining a new Instant Messaging session carried over a web-page.
-Click-to-dial: users could initiate SIP sessions only by clicking a link on a web page, without the need of the Web-Browser being SIP-aware nor needing even a SIP phone: the server could handle all the logic so the user who clicked could receive a call from the server’s SIP network.
The WeSIP Application Server configuration file is based on the Apache Tomcat configuration system: It is an XML-formatted file, in which the different components of the server are specified.
The default config file that comes with the WeSIP distribution package should be suitable for most of the deployment configurations.
>
The topmost element in the XML configuration file is the "server" which has 2 xml attributes, called "port" and "shutdown". The former specifies a port on which the WeSIP AS will listen for the shutdown command, and the latter is the magic word that will make the server shutdown.
if you send the magic word "SHUTDOWN" to the port 8005 of the localhost, the server will stop cleanly.
>
Nested within the Server element, must be a "Service" element, with an attribute called "name" which specifies the name for the service. This attribute is not very relevant, you can call it whatever you like.
Within the Service element must be two or more elements: the connectors and the engines. A connector is the instance that will receive messages from the network. You can specify HTTP connectors and/or SIP connectors. Every connector needs an attribute called "className" which specifies which class will be responsible for receiving the messages from the network. For HTTP connectors, the classname must be "org.apache.catalina.connector.http.HttpConnector" and for SIP connectors "com.voztele.sipservlet.connector.SipConnector".
The SIP Connector uses 4 attributes:
className="com.voztele.sipservlet.connector.SipConnector"
specifies the classname of the connector.
minProcessors="5"
specifies the minimum number of SIPprocessor instances (and threads in the pool) to process incoming SIP messages. More processors should allow more load to be processed. This is the minimum number of instances, even if they are spare and not working.
maxProcessors="10"
specifies the maximum number of SIP processors used (a negative value specifies that there is no limit).
addresses="localhost:5060"
Specifies the SIP address and port in which the Application Server from which the Application Server will process the SIP messages. This Addres is where OpenSER listens for the messages, so in fact, OpenSER is listening on them, but OpenSER passes the messages to WeSIP, so WeSIP must be aware of this IP/port.
this attribute MUST match one of the listening points declared within OpenSER in the "listen" parameters. For example in openser.cfg: listen = tcp:localhost:5060 listen = udp:localhost:5060 |
Within the SIP Connector element there must be an ExtraProperties element, containing nestes Property elements. Each property element specifies a parameter for the SIP Stack. Each property is specified by a key and a value. The most significant keys are:
com.voztele.javax.sip.SER_ADDRESS
This specifies the IP and port in which the OpenSER is listening for Application Servers to connect and register.This specifies the IP and port in which the OpenSER is listening for Application Servers to connect and register.
This needs to match the
modparam("seas", "listen_sockets","127.0.0.1:5080") |
javax.sip.STACK_NAME
Specifies the name identifying this instance of the Application Server.
This is the name you will set in the OpenSER configuration script when you invoke the WeSIP Application Server, by calling the as_relay_t function. This is the name you pass as the parameter of the function. If you have different WeSIP instances all connecting to the same OpenSER, they must each one have a different STACK_NAME", and within OpenSER you can call each of them by invoking as_relay_t() with a different name. Example: <Property key="javax.sip.STACK_NAME" value="app_server_one" /> |
com.voztele.javax.sip.THREAD_POOL_SIZE (integer)
Specifies the number of threads there must be in the pool to process incoming SIP messages. If unspecificed, the default is "infinity".
com.voztele.javax.sip.SPIRAL_HDR
This property tells WeSIP and SEAS that every SipRequest and UAC transaction generated from WeSIP, must spiral through SER, and will be added a special Header called "X-WeSIP-SPIRAL: true" this will make all the outgoing messages pass again through the OpenSER script, so that they can be accounted or whatever the configurator wants. For example, the configuration script could go:
route{ if(is_present_hf("X-WeSIP-SPIRAL")){ /* account, log, register, or whatever */ t_relay(); }else{ as_relay_t("app_server_1"); } }
>
The Engine must also be nested within the Server element, along with the Connectors. It must have a "name" attribute with whatever name you feel like. It needs to have another attribute called "defaultHost" which will be the default host to which to pass the incoming request (in HTTP/1.0 the requests dont have a Host header, so they will be passed to this default host, in SIP, this attribute doesn't have a meaning.). In order to have this Engine handling also SIP messages, the "className" attribute of the Engine must be "com.voztele.sipservlet.core.ConvergedEngine".
Within the Engine, there can be one or more Hosts, each one specified within a "Host" element nested in the engine.
A mapper is used to map an incoming request to one or another SIP or HTTP host. In case it is a SIP request, the mapping is done based on the sip.xml deployment descriptor rules. The classname of the SIP mapper MUST BE "com.voztele.sipservlet.core.EngineSipMapper". The "mapper" element must also have a "protocol" attribute, specifying which protocol this mapper handles. In case of the SIP mapper it must be "SIP/2.0". The HTTP mapper's classname must be "org.apache.catalina.core.StandardEngineMapper" and the protocol attribute "HTTP/1.1"
The authentication in HTTP is performed in Apache-Tomcat through Realms. The memory realm is (textual copy from the Apache-Tomcat javadoc"): "Simple implementation of Realm that reads an XML file to configure the valid users, passwords, and roles."
The classname must be "org.apache.catalina.realm.MemoryRealm"
A "pathname" attribute can be specified to tell the Realm which file contains the usernames, passwords and roles. If not specified, it is "conf/wesip-users.xml"
A Host represents a VirtualHost in HTTP/1.1 servers, so the requests will be dispatched to one or another virtual host depending on the Host: header. In SIP this doesn't make much sense, because there's no such Host: header, and virtual hosting is not done in this way. Every host must have a "name" attribute which specifies the name of the virtual host, it must also have a "nameSip" attribute which MUST MATCH the IP or hostname _and_ port" specified in OpenSER listen parameters and in the Sip Connector the hostname and the port must be separated with an underscore. for example: nameSip="localhost_5060" or nameSip="192.168.1.1_5060" The next important attribute that must have the Host element is "appBase" which declares the directory where the WEB and SIP applications reside. It usually is a directory called apps in the directory from which the server runs. The attribute "unpackWARs" says the WeSIP Application Server to unpack the Web or Sip Application Archives (.war or .sar extensions) found inside the appBase directory. It should usually be set to "true". The "port" attribute specifies the port where this host is going to receive SIP messages . This only has to do with the SIP protocol, not with HTTP. It must be the same as the port specified in OpenSER parameter "listen_sockets" (for the seas module). The "autoDeploy" attribute tells the host to monitor the "appBase" directory for new application archives (.sar or .war) so they can automatically be deployed. This parameter should be set to "true". The "className" used for the Host _must_be_ "com.voztele.sipservlet.core.ConvergedHost"
Hosts must also have a nested Mapper element, but when the mapper is inside a Host (and not in an Engine) the classnames must be "com.voztele.sipservlet.core.SipHostMapper" for the "SIP/2.0" protocol and "org.apache.catalina.core.HttpHostMapper" for the "HTTP/1.1" protocol. (2 mappers must be nested inside the Host).
In general, you can configure WeSIP to work with your OpenSER in two ways: have 2 OpenSER instances, the first acting as Proxy/Registrar/Redirect and the second cooperating with WeSIP to act as the Application Server. This is the preferred deployment layout, as the first OpenSER works as usual, and the requests that need special services are relaied to another OpenSER which acts on behalf of the WeSIP AS. This configuration profile distributes load (call-routing logic in one instance, and Application Services in the other), and is also more fault-tolerant. On the other hand, you can have all your call-routing logic and Application Server on the same OpenSER, having one script handle all the logic, and then invoking the App Server at any point.
>
debug=3 # debug level (cmd line: -dddddddddd) fork=yes log_stderror=no # (cmd line: -E) check_via=no # (cmd. line: -v) dns=no # (cmd. line: -r) rev_dns=no # (cmd. line: -R) port=5060 children=4 loadmodule "/usr/local/lib/openser/modules/sl.so" loadmodule "/usr/local/lib/openser/modules/tm.so" loadmodule "/usr/local/lib/openser/modules/rr.so" loadmodule "/usr/local/lib/openser/modules/maxfwd.so" loadmodule "/usr/local/lib/openser/modules/usrloc.so" loadmodule "/usr/local/lib/openser/modules/registrar.so" loadmodule "/usr/local/lib/openser/modules/textops.so" loadmodule "/usr/local/lib/openser/modules/seas.so" loadmodule "/usr/local/lib/openser/modules/mi_fifo.so" modparam("mi_fifo", "fifo_name", "/tmp/openser_fifo") modparam("usrloc", "db_mode", 0) modparam("rr", "enable_full_lr", 1) modparam("seas", "listen_sockets", "127.0.0.1:5080"); route{ if (!mf_process_maxfwd_header("10")) { sl_send_reply("483","Too Many Hops"); exit; }; if (msg:len >= 2048 ) { sl_send_reply("513", "Message too big"); exit; }; if (!method=="REGISTER") record_route(); if (loose_route()) { append_hf("P-hint: rr-enforced\r\n"); route(1); }; if (uri==myself) { if (method=="REGISTER") { save("location"); exit; }; lookup("aliases"); if (!uri==myself) { append_hf("P-hint: outbound alias\r\n"); route(1); }; if (!lookup("location")) { sl_send_reply("404", "Not Found"); exit; }; append_hf("P-hint: usrloc applied\r\n"); }; route(1); } route[1] { if(!as_relay_t("app_server_one")){ t_reply("500","Application Server error"); } }
debug=9 # debug level (cmd line: -dddddddddd) fork=yes log_stderror=yes # (cmd line: -E) check_via=no # (cmd. line: -v) dns=no # (cmd. line: -r) rev_dns=no # (cmd. line: -R) port=5060 children=4 reply_to_via=1 listen = tcp:localhost:5060 listen = udp:localhost:5060 mpath="/home/elias/src/sipservlet/seas" loadmodule "modules/tm/tm.so" loadmodule "modules/seas/seas.so" loadmodule "modules/mi_fifo/mi_fifo.so" modparam("mi_fifo", "fifo_name", "/tmp/openser_fifo") modparam("seas", "listen_sockets","127.0.0.1:5080") route{ if(!as_relay_t("app_server_1")){ t_reply("500","Application Server error"); } }
<Server port="8005" shutdown="SHUTDOWN" debug="0"> <Service name="WeSIP-Standalone"> <Connector className="org.apache.catalina.connector.http.HttpConnector" port="8080" minProcessors="5" maxProcessors="75" enableLookups="true" address="localhost" acceptCount="10" debug="10" /> <Connector className="com.voztele.sipservlet.connector.SipConnector" minProcessors="5" maxProcessors="75" addresses="localhost:5060" > <ExtraProperties> <Property key="com.voztele.javax.sip.SER_ADDRESS" value="127.0.0.1:5080" /> <Property key="javax.sip.STACK_NAME" value="app_server_one" /> <Property key="com.voztele.javax.sip.THREAD_POOL_SIZE" value="10" /> </ExtraProperties> </Connector> <Engine name="Standalone" defaultHost="localhost" debug="10" className="com.voztele.sipservlet.core.ConvergedEngine"> <Logger className="org.apache.catalina.logger.SystemOutLogger" timestamp="true"/> <Mapper className="org.apache.catalina.core.StandardEngineMapper" protocol="HTTP/1.1"/> <Mapper className="com.voztele.sipservlet.core.EngineSipMapper" protocol="SIP/2.0"/> <Realm className="org.apache.catalina.realm.MemoryRealm" /> <Host name="localhost" nameSip="localhost_5060" debug="10" appBase="webapps" unpackWARs="true" port="5060" autoDeploy="true" className="com.voztele.sipservlet.core.ConvergedHost"> <Mapper className="com.voztele.sipservlet.core.SipHostMapper" protocol="SIP/2.0"/> <Mapper className="org.apache.catalina.core.HttpHostMapper" protocol="HTTP/1.1"/> </Host> </Engine> </Service> </Server>
The SEAS module runs within the Open Sip Express Router aka. OpenSER. OpenSER uses a pool of processes to execute the script logic on every new message received. These are called the worker processes. One of these processes will be selected to process the script, and at some point it will find a function invoking the relay of the SIP message to one of the Application Servers registered. This function has been called as_relay_t, which stands for Application Server relay (the _t stands for TransactionStatefully), and receives as the only parameter the name of the application server to be invoked.
The process will execute the as_relay_t function, which looks up in a table if there is a registered Application Server with that name. If there is one, the process will craft the SEAS header for the SIP message being handled, put it in a shared memory segment, and write the address of that segment to a pipe (4 bytes pointer in IA32).
This way, we will have all the OpenSER processes composing the SEAS header along with the SIP message, and putting its shared memory address into that pipe. This technique of inter-process communication avoids race conditions because writing to a pipe is granted to be an atomic operation if the data to write is less than _POSIX_PIPE_BUF, which usually is 512 bytes.
At the initialization of OpenSER, the SEAS module creates the discussed pipe, so that all the OpenSER worker processes inherit the file descriptor associated to the pipe. Then it spawns a new process, which will be the one to open two server sockets, and wait for the Application Servers to connect and register.
Each Application Server wishing to receive events from OpenSER, will have to open a socket to the module (the port and IP of the socket are defined at start time in the script). After connection, it has to print its identification name. The SEAS process (from now on, called event dispatcher) will then register it in its internal structures, so that the OpenSER processes can push events for it. The following picture, shows the internals of the SEAS Event dispatcher process:
Within the SER server, the flowing of SIP Messages and control flow, is depicted in the following diagram:SIP is a very flexible protocol. It can be very easily extended with new features, and SIP entities have a high level of freedom in composing the SIP messages, for example setting IPs or hostnames in URIs, reordering header fields, folding headers, aggregating/scattering headers, etc.
This flexibility, though, makes it difficult to implement efficiently, because parsing of text headers requires a lot of state.
OpenSER implements a very efficient parsing mechanism and SIP-transaction machinery. The goal of the SEAS protocol is to keep all this information that has been already extracted at OpenSER, so that it can be reused at the Application Server.
The SEAS protocol is a layer of information regarding the internal structure of a SIP message that is added whenever SEAS sends a SIP event to the Application Servers. The protocol is used for communication between OpenSER and the Application Servers.
Once an incoming SIP message has reached the worker process within OpenSER, it copies its content into a private memory area (which is, a memory chunk not shared across processes). In this point, the message first line is parsed to know whether it is a SIP request or response.
OpenSER uses a technique called lazy-parsing, which consists in delaying the parse of headers until some piece of the code requires it.
As the SIP message goes traversing functions and the script code, a function called parse_msg() gets called again and again, and the SIP message gets parsed further and further. Each call to parse_msg passes an integer value argument (32 bits) in which every bit signals a header to be parsed, if they are already parsed (because a previous invocation of parse_msg), the function returns immediately, otherwise, the SIP message is scanned and parsed until all the headers requested get parsed.
In each call to parse_msg, different parts of the message are analyzed, and different SIP header-specific structures get filled. Each one of this structures, give quick access to each of the parts of a SIP message header.
For example, a Via header struct is called via_body, and has these members: name, version, transport, host, proto, port, port_str, params, comment, received, rport, etc. each of these members gives quick access to each of the parts of the header. For example, a via header like this: “Via: SIP/2.0/UDP 192.168.1.64:5070;branch=z9hG4bK-c02c60cc” would have the member proto pointing to the “U” of “UDP”, and a length of 3, the host member would be pointing to “192.168.1.64” and have a length of 12, the branch member would be pointing to “z9hG4bK-c02c60cc” and a length of 16, and so on.
This structure is the result of the parsing. All this meta-information regarding the SIP message structure, is stored in a sip_msg structure, using dynamically-allocated memory segments.
OpenSER defines different structure types describing different SIP headers, such as via_body, to_body, cseq_body, via_param, and so on. These structures are generally composed of another kind of structure called str.
The str structure is a key component of OpenSER's high performance. In the C programming language, a string's length is known because a '0' (null-character) is found at the end of it. This forces each of the string manipulation functions to keep looking for a '0' in the byte stream, which is quite processor consuming. Instead of this, OpenSER defines a structure composed of a char pointer and an integer. The char points to the start of a string, and the integer gives its length, thus avoiding the '0' lookup problem, and giving a significant performance boost.
This structure has been quite useful to the design of the SEAS protocol, because it enables the description of the SIP message anatomy by giving pointers to each of its fields, and integers describing each of its lengths.
Knowing that a SIP header does not usually occupy more than a few characters (always less than 256), the pointer in the structure has been relativized to the beginning of the SIP message or the beginning of the SIP header, and the integer giving the length, has been casted to an unsigned byte (256 values, so 256 characters maximum length).
When messages get transferred from OpenSER to the Application Server, it is optimum to keep this worthy meta-information regarding the SIP message, so that it can be used at the AS part. For this to be possible, it is needed to store the pointers to each of the syntactic structures and their length.
In general, pointers are variables that point to a region in the memory of a computer. The region of the memory is counted from the 0x00000000 address in IA32 architectures (from the beginning).
C provides functionality to do any kind of arithmetic operations over pointers (add, subtract, multiply and divide), so that the euclidean distance over the one-dimension address space can be calculated just by subtracting a base address from another pointer.
These pointers will have to be transmitted through the network, along with the SIP message, so for the pointers to keep their meaning, they need to be relativized to a known point, and the most meaningful known point in a SIP message is its start.
So making the pointers relative to the message start, gives two important features: first, it makes the pointers still valid when they arrive at another computer (because they are relative to the beginning of the message), and they occupy far less memory, because from a 4-byte pointer (in IA32) it gets translated to a 1 or 2 byte index, because an important amount of redundant information is elicited (we already know that each of the parts of the message belong to the message, so why carry the message begin address in each of the pointers ?).
The SIP messages are composed of protocol headers and a payload. The headers section don't usually surpass the 1500 byte limit, amongst other reasons, because the usual Maximum Transmission Unit in Ethernet networks is 1500 bytes and the protocol was initially designed to work on UDP. For that reason, 11 bits should be enough to address a particular region within the SIP message, because it yields 2048 positions. The closest greater value to 11 bits multiple of a byte (the basic TCP network transport unit) is 16 bits, or 2 bytes, which makes it possible to address 65536 positions from the beginning.
For the SEAS protocol to be extensible and platform-independent, all the 2-byte pointers or indexes to each of the message regions are sent in network-byte-order, or big endian. This is also useful in the JAVA part to retrieve the indexes, because the JAVA natively uses a big-endian representation of integers, regardless the architecture on which it runs.
For each kind of standard SIP header (this is, the headers referred to in the SIP specification) there is a code specification, regarding the composition of the header. Each one of its parts points to one the several components of the header. For example, a From header always has a SipURI and may have several parameters, amongst others, a tag. Then, the From header code has a field indicating where the URI starts, a codification of the URI, and several pointers that point to each one of the parameter names and values. This is the codification of the From header. All the other headers have a similar codification.
Every header codification, regardless it is known to the server or not, begins with a 2-byte unsigned integer, which points to the beginning of that header counted from the SIP message begin (a SIP message start based pointer to the header). Following these two bytes is another byte giving the length of the name, and another byte giving the length of the entire header (including name and value).
For example:As the SIP URI is one of the most used types in a SIP message, a special structure has been defined to describe the contents of it. A URI is always included inside a SIP header, or may be in the first line of a SIP Request (as the request URI).
The codification of any URI is as follows:
What follows is an example of a SIP URI codification with the SEAS protocol.
The first byte in the encoded-URI structure, gives the index where the URI starts, counting from the beginning of the SIP header where it appears. The next two bytes are flags indicating known fields present in the URI (such as port, host, user, etc.).All the following bytes are uri-start based pointers to the fields that are present in the URI, as specified by the flags. They must appear in the same order shown in the flags, and only appear if the flag was set to 1.
The end of the field, will be the place where the following pointer points to, minus one (note that all the fields present in a URI are preceded by 1 character, ie sip[:user][:passwod][@host][:port][;param1=x][;param2=y][?hdr1=a][&hdr2=b]) it will also be necessary to have a pointer at the end, pointing two past the end of the URI, so that the length of the last header can be computed.
The reason to have the “other parameters” and headers flags at the beginning (just after the strictly URI stuff), is that it will be necessary to know the length of the parameters section and the headers section. The parameters can appear in an arbitrary order, they won't be following the convention of transport-ttl-user-method-maddr-lr, so we can't rely on the next pointer to compute the length of the previous pointer field, as the ttl parameter can appear before the transport parameter. So the parameter pointers must have 2 bytes: pointer+length.
To and From headers follow the same structure, so the same codification structure has been used to describe both. The structure is depicted in the drawing:
The contact header is one of those SIP headers that can be combined, which means that if several headers of the same type are present in the message, they can be aggregated in a single header, having the header values separated by a comma. Thus, a single Contact header can contain more than one contact-value. For this reason, the Contact codification is composed of a several Contact codifications concatenated, and a byte at the beginning telling how much Contact codifications are present. The code is depicted in the following drawing:
Both Route and Record-Route headers follow an identical structure, and it is also permitted to combine several headers into one, with their bodies (or header values) separated by commas. In this case, both kinds of headers follow the same structure, defined as follows:
These two kinds of headers carry mime type and subtype definitions in the form “type/subtype” (ie. text/xml, application/sdp or whatever). For internal handling of this headers, SER codifies the known types and subtypes into a single 32 bit integer, with the highest two bytes giving the mime type, and the lowest two bytes giving the subtype.
The difference is that Accept header can also be combined, carrying more than one header value in a single header row. Thus the Accept header has a leading byte giving the number of mime type/subtype integers present, while the Content-Type only uses 4 bytes (a 32-bit integer) giving the type/subtype.
SIP has inherited the authentication scheme from HTTP, which is based on a digest scheme. There are several headers regarding these authorization scheme, namely Proxy-Authenticate, WWW-Authenticate, Authorization and Proxy-Authorization. All of them can be codified using the same schema, which is as follows:
For each field present, there are 2 bytes, one pointing the place where it starts, the next giving how long this field is. The URI is a special case, and is composed of 1 byte telling how long is the URI structure, and then the encoded URI structure.
Allow headers carry request methods that a user agent or proxy understands or is willing to accept. In SER, request methods are codified into a 32-bit integer, each of its bits signals a different kind of header. The Allow header is codified copying that integer into the payload of the header.
The content-disposition is encoded within 2 bytes: the first is a header-start based pointer to where the content-disposition value starts, and the second is its length. If there are parameters present, each of them uses 1 byte pointing to where the parameter name starts, and 1 byte pointing to where the parameter value starts. From these two values, the parameter name and value lengths can be inferred.
The content length header is codified as a 4-byte unsigned integer, in network byte order.
The Cseq header is codified using 9 bytes. The first one is a number corresponding to the internal value that SER assigns to that request method (the method ID). The following 4 bytes are an unsigned 32-bit integer according to the Cseq number. The next two bytes are the header based pointer to the beginning of the Cseq number and its length, and two more bytes pointing to the beginning of the method name and its length.
The expires header is composed of 6 bytes. The first four bytes are an unsigned 32-bit integer with the parsed value of the header (which is the number of seconds before a request expires). Then follows 1 byte pointing to the beginning of the header value (the expires value as a string) and a byte giving the length of the value.
In SER, not only the headers are parsed with a high degree of optimization, but also the first line is. So for the SEAS protocol to realize this improvement, a codification for the first line of every SIP messages has also been defined.
The first two bytes of the codification are a 2-byte unsigned integer. If its value is equal or greater than 100, then this is a response, and the integer represents its status code. If its value is smaller than 100, then it is a request, and the integer represents the method of the request being transported.
The next two bytes are an unsigned integer which is a pointer to where the actual SIP message starts, beginning from the start of the codified payload.
The next two bytes are also an unsigned integer giving the SIP message length.
The next bytes differ on the meaning depending on whether the message is a SIP Request or Response.
In case it is a Request:
The next two bytes, are a SIP-message-start based pointer to where the method begins, and the method length.
The next two bytes, are a SIP-message-start based pointer to where the Request URI begins, and the request URI length.
The next two bytes, are a SIP-message-start based pointer to where the version identifier begins, and the version identifier length.
In case it was a Response:
The next two bytes, are a SIP-message-start based pointer to where the response code begins, and the response code length.
The next two bytes, are a SIP-message-start based pointer to where the reason phrase begins, and the reason phrase length.
The next two bytes, are a SIP-message-start based pointer to where the version identifier begins, and the version identifier length.
In case the message is a SIP response, the following bytes correspond to the Request URI codification. The first byte is the length of the URI codification, followed by the URI code.
The last byte in this set, is the number of headers present in the SIP message. After this byte, goes a section, called the Message Headers Index, which gives quick access to each of the headers and their codifications present in the message.
As it has been already discussed, the aim of SEAS project is to achieve as high a performance as possible. One of the techniques enabling high performance in text-based servers is the so called lazy parsing. To enable the laziest possible parsing at the Application Server endpoint, a mechanism has been used so that access to a requested SIP header can be delayed until the application requests it, and the access can be direct to that header, without parsing the former headers present in the SIP message. Recall that one of the performance drawbacks of the SIP protocol is that headers of any type can be spread all along the header section, not having the constraint of putting the most critical sip-specific headers at the beginning and ordered (which would be, in fact, very desirable).
For this to be possible, there is a section right after the beginning of the payload (the general message information section) which is a kind of hash table, giving quick access to the codes (as explained in the previous sections) of each of the headers present in the message.
This sort of hash table, is composed of triplets of bytes. The first byte of each three is a code indicating which kind of header it points to (whether it is a From, To, Call-ID, Route header, etc). Then follows a 2 byte network-byte-order integer that points to a section in the codified-header where the body of this header is more specifically described.
This gives really fast access to any of the headers. For example, if all the Route Headers were requested by the application, then a lookup in this table would be necessary, looking for the value '9' (corresponding to the Route header) in each of the positions multiple of 3 (0,3,6,9,12, etc). This can be done in a extremely fast and easy way, as this snipped of pseudo code explains:
for(int j=0,int i=0;i<table_length;i+=3){
if(payload[i]==9)
results[j++]=i;
}
this would let in the “results” array all the indexes in the headers table that refer to a Route header. Then, the Route codification for each of the headers could be reached thanks to the two-byte unsigned integer that follows each of the header identifiers.
So a SIP message codified by the SEAS protocol, has the following layout:SIP Messages are a fundamental part of the protocol, but they are not the only one. Transaction play a very important role in the SIP protocol, within SER and in any JAIN-SIP implementation. For this reason, the SEAS protocol also needs to define and implement some semantics regarding transaction handling. The events related to a transaction are: Incoming Request, Outgoing Request, Incoming Response, Outgoing Response, Timeout and Transport Error.
So the SEAS protocol defines a specific format for each one of these events. Internally, SER stores the transactions in a hash table. This hash table generates an integer for each transaction applying a hash function to its Via branch parameter, this integer is the hash index, and it identifies in which slot within the hash table the transaction is stored. The transaction table usually uses 65536 entries, so the hash collision is pretty unlikely. Anyway, every hash entry is in reality a linked list of transactions, so in the case a hash collision (two transactions being assigned to the same hash slot) the transactions are added to the same slot, each one being identified by another integer called the label. The label within a hash slot, is initially generated randomly, and then increased by one each time a transaction falls in the same slot. So every transaction is identified by a hash index and a label.
For incoming SIP requests, a transaction is generated at SER, and the SEAS module gets that transaction identifier (hash index + label), then grabs the source and destination IP, port and transport from every message, and crafts a SEAS RequestIn event. This kind of event carries all this information within it.
In order to send Responses out for the Server Transactions, JAIN can send a type of Action messages, that order SER to send them to the network. These messages follow a structure very similar to that of RequestIn events: they start with the Action length in bytes, then follows a byte giving the type of action, then follows the Hash Index and the Label associated with the transaction that is being replied, and finally the SIP Message in raw format. It doesn’t use the SEAS codification described above, because SER can easily parse the JAIN provided Response to process it and send it out, so the pre-parsing is not needed in that direction.
In order to generate Client Transactions, that is, sending SIP Requests out, JAIN utilizes another kind of action called Seas Request Action. In this case, when JAIN generates the Request to be sent out, it doesn’t have any means to know the transaction identifier (hash index and label) that will be assigned to it by SER, so a new mechanism has bee implemented to correlate JAIN requests to SER transactions. Basically, JAIN-SIP assigns a unique identifier (an integer) that is incremented by one for each new Client Transaction generated. This identifier is passed to SER along with the SIP Request, so when a SIP Response arrives to SER regarding that transaction, SER sends a ResponseIn event to the JAIN stack, containing both the initial integer identifying the transaction at JAIN and the hash index and label that have been assigned to the transaction. This way, JAIN can correlate its own identifiers with the identifiers used within SER.
In case there is a Transaction Timeout, it is notified to the JAIN SIP Stack by passing it a Seas Incoming Response with a flag called Faked Reply, and a Response code number 408 (Request Timeout).
Take a look at http://openser.org/
Take a look at http://www.wesip.eu/
First at all check if your question was already answered on one of our mailing lists:
User Mailing List - http://openser.org/cgi-bin/mailman/listinfo/users
Developer Mailing List - http://openser.org/cgi-bin/mailman/listinfo/devel
E-mails regarding any stable OpenSER release should be sent to
<users@openser.org>
and e-mails regarding development
versions should be sent to <devel@openser.org>
.
If you want to keep the mail private, send it to
<team@openser.org>
.
Please follow the guidelines provided at: http://sourceforge.net/tracker/?group_id=139143.