/* ** (c) COPYRIGHT MIT 1995. ** Please first read the full copyright statement in the file COPYRIGH. */
The Host class manages what we know about a remote host. This can for example be what type of host it is, and what version it is using. Notice that a host object can be used to describe both a server or a client - all information in the Host object can be shared regardless of whether it is to be used in a server application or a client application.
This module is implemented by HTHost.c, and it is a part of the W3C Sample Code Library.
#ifndef HTHOST_H
#define HTHOST_H
#ifdef __cplusplus
extern "C" {
#endif
typedef struct _HTHost HTHost;
#define HOST_HASH_SIZE HT_M_HASH_SIZE
#include "HTChannl.h"
#include "HTReq.h"
#include "HTEvent.h"
#include "HTProt.h"
#include "HTTimer.h"
The Host class contains information about the remote host, for example the type (HTTP/1.0, HTTP/1.1, FTP etc.) along with information on how the connections can be used (if it supports persistent connections, interleaved access etc.)
We keep a cache of information that we know about a remote host. This allows us to be much more detailed in generating requests. Search the host info cache for a host object or create a new one and add it. Examples of host names are
extern HTHost * HTHost_new (char * host, u_short u_port); extern HTHost * HTHost_newWParse(HTRequest * request, char * url, u_short u_port); extern int HTHost_hash (HTHost * host);
The Host Class contains an automatic garbage collection of Host objects so that we don't keep information around that is stale.
Searches the cache of known hosts to see if we already have information about this host. If not then we return NULL.
extern HTHost * HTHost_find (char * host);
Cleanup and delete the host table.
extern void HTHost_deleteAll (void);
You can use this function to see whether a host object is idle or in use. We have several modes describing how and when a host is idle. This is a function of the Transport Object
extern BOOL HTHost_isIdle (HTHost * host);
We keep track of the capabilities of the host in the other end so thatwe may adjust our queries to fit it better
Get the name of the remote host. This is set automatically when a new Host object and can be asked for at any point in time. You can not change the host name but must create a new Host object instead.
extern char * HTHost_name (HTHost * host);
Define the host class of the host at the other end. A class is a generic description of the protocol which is exactly like the access method in a URL, for example "http" etc. The host version is a finer distinction (sub-class) between various versions of the host class, for example HTTP/0.9, HTTP/1.1 etc. The host version is a bit flag that the protocol module can define on its own. That way we don't have to change this module when registering a new protocol module. The host type is a description of whether we can keep the connection persistent or not.
extern char * HTHost_class (HTHost * host); extern void HTHost_setClass (HTHost * host, char * s_class); extern int HTHost_version (HTHost * host); extern void HTHost_setVersion (HTHost * host, int version);
A server can inform a client about the supported methods using the
Public header.
extern HTMethod HTHost_publicMethods (HTHost * me); extern void HTHost_setPublicMethods (HTHost * me, HTMethod methodset); extern void HTHost_appendPublicMethods (HTHost * me, HTMethod methodset);
A server can send its server application name and version in a HTTP response. We pick up this information and add it to the Host object
extern char * HTHost_server (HTHost * host); extern BOOL HTHost_setServer (HTHost * host, const char * server);
A client can send the name of the client application in a HTTP request. We pick up this information and add it to the Host Object
extern char * HTHost_userAgent (HTHost * host); extern BOOL HTHost_setUserAgent (HTHost * host, const char * userAgent);
Since all HTTP entities are represented in HTTP messages as sequences of bytes, the concept of a byte range is meaningful for any HTTP entity. (However, not all clients and servers need to support byte-range operations.) Byte range specifications in HTTP apply to the sequence of bytes in the entity-body (not necessarily the same as the message-body). A byte range operation may specify a single range of bytes, or a set of ranges within a single entity.
You can also check whether a specific range unit is OK. We always say
YES except if we have a specific statement from the server that
it doesn't understand byte ranges - that is - it has sent "none" in a
"Accept-Range" response header
extern char * HTHost_rangeUnits (HTHost * host); extern BOOL HTHost_setRangeUnits (HTHost * host, const char * units); extern BOOL HTHost_isRangeUnitAcceptable (HTHost * host, const char * unit);
This can be used for anything that the application would like to keep tabs on.
extern void HTHost_setContext (HTHost * me, void * context); extern void * HTHost_context (HTHost * me);
Requests are queued in the Host object until we have resources to start them. The request is in the form of a Net object as we may have multiple socket requests per Request object. This is for example the case with FTP which uses two connections.
extern int HTHost_addNet (HTHost * host, HTNet * net); extern BOOL HTHost_deleteNet (HTHost * host, HTNet * net, int status); extern HTList * HTHost_net (HTHost * host);
A Channel object is an abstraction for a transport, like a TCP connection, for example. Each host object can have at most one channel object associated with it.
As a Net Object doesn't necessarily know whether there is a channel up and running and whether that channel can be reused, it must do an explicit connect the the host.
extern int HTHost_connect (HTHost * host, HTNet * net, char * url); extern int HTHost_accept (HTHost * host, HTNet * net, char * url); extern int HTHost_listen (HTHost * host, HTNet * net, char * url);
As soon as we know that a channel is about to close (for example because the server sends us a Connection: close header field) then we register this informtation in the Host object:
extern BOOL HTHost_setCloseNotification (HTHost * host, BOOL mode); extern BOOL HTHost_closeNotification (HTHost * host);
Here you can find an already associated channel with a host object or you can explicitly associate a channel with a host object.
extern BOOL HTHost_setChannel (HTHost * host, HTChannel * channel); extern HTChannel * HTHost_channel (HTHost * host);
When a channel is deleted, it must be unregistered from the host object which is done by this operation:
extern BOOL HTHost_clearChannel (HTHost * host, int status);
The way a channel can be used depends on the transport and what mode the channel is in. The mode (whether we can use persistent connections, pipeline, etc.) may change mode in the middle of a connection If the new mode is lower than the old mode then adjust the pipeline accordingly. That is, if we are going into single mode then move all entries in the pipeline and move the rest to the pending queue. They will get launched at a later point in time.
extern HTTransportMode HTHost_mode (HTHost * host, BOOL * active); extern BOOL HTHost_setMode (HTHost * host, HTTransportMode mode);
There are two ways we can end up with pending requests:
This set of functions handles pending host objects and can start new requests as resources get available. The first function checks the host object for any pending Net objects and return the first of these Net objects.
extern HTNet * HTHost_nextPendingNet (HTHost * host);
The second checks the list of pending host objects waiting for a socket and returns the first of these Host objects.
extern HTHost * HTHost_nextPendingHost (void);
Start the next pending request if any. First we look for pending requests for the same host and then we check for any other pending hosts. If nothing pending then register a close event handler to have something catching the socket if the remote server closes the connection, for example due to timeout.
extern BOOL HTHost_launchPending (HTHost * host);
Controls whether pending requests should be automatically activated. The default is on, but if turned off then no pending requests are launched.
extern void HTHost_enable_PendingReqLaunch (void); extern void HTHost_disable_PendingReqLaunch (void);
We don't want more than (Max open sockets) - 2 connections to be persistent in order to avoid deadlock. You can set the max number of simultaneous open connection in the HTNet manager.
extern BOOL HTHost_setPersistent (HTHost * host, BOOL persistent,
HTTransportMode mode);
extern BOOL HTHost_isPersistent (HTHost * host);
If the server doesn't close the connection on us then we close it after a while so that we don't unnecessarily take up resources (see also how the timeouts of individual requests can be set). Libwww provides two mechanisms: an active timeout and a passive timeout. The former uses libwww timers and is the preferred mechanism, the latter passively looks at the Host object when a new request is issued in order to determine whether the existing channel can be reused. This is primariliy for non-preemptive requests which in general is deprecated.
By default we have an active timeout of 60 secs and a passive timeout of 120 secs (the latter is longer as this is less reliable). Active timeout s can be accessed using these functions:
extern BOOL HTHost_setActiveTimeout (ms_t timeout); extern ms_t HTHost_activeTimeout (void);
and passive timeouts can be accessed using these functions
extern time_t HTHost_persistTimeout (void); extern BOOL HTHost_setPersistTimeout (time_t timeout);
The following two functions are deprecated:
extern void HTHost_setPersistExpires (HTHost * host, time_t expires); extern time_t HTHost_persistExpires (HTHost * host);
Another way to detect when a connection is about to close is to count the number of requests made. For example, the (current) default bevaior by most Apache servers is to close a TCP connection after 100 requests. I don't quite think it makes sense to control the close of a connection like this but anyway, there we go.
extern void HTHost_setReqsPerConnection (HTHost * host, int reqs); extern int HTHost_reqsPerConnection (HTHost * host); extern void HTHost_setReqsMade (HTHost * host, int reqs); extern int HTHost_reqsMade (HTHost * host);
Which Net object can read and/or write? When doing pipelining, we essentially serialize requests and therefore we must keep track of who can read and who can write.
extern HTNet * HTHost_firstNet (HTHost * host); extern HTNet * HTHost_getReadNet (HTHost * host); extern HTNet * HTHost_getWriteNet (HTHost * host);
extern HTInputStream * HTHost_getInput (HTHost * host, HTTransport * transport, void * param, int mode); extern HTOutputStream * HTHost_getOutput (HTHost * host, HTTransport * tp, void * param, int mode);
Because of the push streams, the streams must keep track of how much data actually was consumed by that stream.
extern int HTHost_read(HTHost * host, HTNet * net); extern BOOL HTHost_setConsumed(HTHost * host, size_t bytes); extern BOOL HTHost_setRemainingRead(HTHost * host, size_t remainaing); extern size_t HTHost_remainingRead (HTHost * host);
When possible, we try to pipeline requests onto the same connection as this saves a lot of time and leads to much higher throughput.
Use these functions to set the max number of requests that can be pipelined at any one time on a single, persistent connection. The higher the number, the more we have to recover if the server closes the connection prematurely. The default is about 50 requests which is enough to fill most links.
extern BOOL HTHost_setMaxPipelinedRequests (int max); extern int HTHost_maxPipelinedRequests (void);
You can query how many Het objects (essentially requests) are outstanding or pending on a host object using these methods:
extern int HTHost_numberOfOutstandingNetObjects (HTHost * host); extern int HTHost_numberOfPendingNetObjects (HTHost * host);
Pipelines normally run by themselves (requests are issued and responses recieved). However, it may be necessry to either prematurely abort a pipeline or to recover a broken pipeline due to communication problems with the server. In case a pipeline is broken then we have to recover it and start again. This is handled automatically by the host object, so you do not have to call this one explicitly.
extern BOOL HTHost_recoverPipe (HTHost * host); extern BOOL HTHost_doRecover (HTHost * host);
Call this function to terminate all requests (pending as well as active) registered with a host object. This is typically the function that handles timeout, abort (user hits the red button, etc). You can also use the HTNet object kill method which in terms call this function.
extern BOOL HTHost_killPipe (HTHost * host);
These functions are used to register and unregister events (read, write, etc.) so that the host object knows about it.
extern int HTHost_register(HTHost * host, HTNet * net, HTEventType type); extern int HTHost_unregister(HTHost * host, HTNet * net, HTEventType type); extern int HTHost_tickleFirstNet(HTHost * host, HTEventType type); extern SockA * HTHost_getSockAddr(HTHost * host);
Events can be assigned a timeout which causes the event to be triggered if the timeout happens before other action is available on the socket. You can assign a global timeout for all host object using the following methods. The default is no timeout.
extern int HTHost_eventTimeout (void); extern void HTHost_setEventTimeout (int millis);
These methods can control how long we want to wait for a flush on a pipelined channel. The default is 30ms which is OK in most situations.
extern BOOL HTHost_setWriteDelay (HTHost * host, ms_t delay); extern ms_t HTHost_writeDelay (HTHost * host); extern int HTHost_findWriteDelay(HTHost * host, ms_t lastFlushTime, int buffSize);
It is also possible to explicitly require a flush using the following method. This can also be set directly in the request object for a single request.
extern int HTHost_forceFlush(HTHost * host);
You can also set the global value so that all new host objects (and therefore all new requests) will inherit this value instead of setting it individually.
extern BOOL HTHost_setDefaultWriteDelay (ms_t delay); extern ms_t HTHost_defaultWriteDelay (void);
We keep track of hosts with multiple IP addresses - socalled multi-homed hosts. This is used for two things: finding the fastest IP address of that host and as a backup if one or more of the hosts are down. This is handled in connection with the DNS manager
extern BOOL HTHost_setHome (HTHost * host, int home); extern int HTHost_home (HTHost * host); extern BOOL HTHost_setRetry (HTHost * host, int retry); extern int HTHost_retry (HTHost * host); extern BOOL HTHost_decreaseRetry (HTHost * host);
A new callback plugged to the activation of a request which allows an application to know when a request has become active.
typedef int HTHost_ActivateRequestCallback (HTRequest * request); extern void HTHost_setActivateRequestCallback (HTHost_ActivateRequestCallback * cbf);
#ifdef __cplusplus } #endif #endif /* HTHOST_H */