ChorusOS man pages section 3RPC: RPC Services

clnt_create(3RPC)

NAME

rpc_clnt_create, clnt_control, clnt_create, clnt_create_timed, clnt_create_vers, clnt_create_vers_timed, clnt_destroy, clnt_dg_create, clnt_pcreateerror, clnt_raw_create, clnt_spcreateerror, clnt_tli_create, clnt_tp_create, clnt_tp_create_timed, clnt_vc_create, rpc_createerr- library routines for dealing with creation and manipulation of CLIENT handles

API RESTRICTIONS

The function or functions documented here may not be used safely in all application contexts with all APIs provided in the ChorusOS 5.0 product.

See API(5FEA) for details.

DESCRIPTION

RPC library routines allow C language programs to make procedure calls on other machines across the network. First a CLIENT handle is created and then the client calls a procedure to send a request to the server. On receipt of the request, the server calls a dispatch routine to perform the requested service, and then sends a reply.

These routines are MT-Safe. In the case of multithreaded applications, the _REENTRANT flag must be defined on the command line at compilation time ( -D_REENTRANT ). When the _REENTRANT flag is defined, rpc_createerr becomes a macro which enables each thread to have its own rpc_createerr .

This call exists to maintain the comatibility with the Solaris Sun RPC library. However, the TLI / XTI interfaces are not supported in the ChorusOS operating system, so this function creates the network connection using the BSD socket interface.

Routines

See rpc(3RPC) for the definition of the CLIENT data structure.

#include <rpc/rpc.h>

bool_t clnt_control(CLIENT * clnt , const uint_t req , char * info );

A function macro to change or retrieve various information about a client object. req indicates the type of operation, and info is a pointer to the information. For both connectionless and connection-oriented transports, the supported values of req , their argument types and what they do are:

CLSET_TIMEOUT struct timeval * set total timeout
CLGET_TIMEOUT	struct timeval *	get total timeout

If the timeout is set using clnt_control() , the timeout argument passed by clnt_call() is ignored in all subsequent calls. If the timeout value is set to 0 , clnt_control() immediately returns RPC_TIMEDOUT . Set the timeout parameter to 0 to batch calls.

CLGET_SERVER_ADDR struct netbuf * get server's address
CLGET_SVC_ADDR	struct netbuf *	get server's address
CLGET_FD	int *	get associated file descriptor
CLSET_FD_CLOSE	void	close the file descriptor when
		destroying the client handle
		(see clnt_destroy())
CLSET_FD_NCLOSE	void	do not close the file
		descriptor when destroying
                the client handle

CLGET_VERS   rpcvers_t     get 
the RPC program's version
		number associated with the
		client handle
CLSET_VERS	rpcvers_t	set the version number of the
RPC program associated with the
		client handle.  This assumes
		that the RPC server for this
		new version is still listening
		to the address of the previous
		version.
CLGET_XID	uint32_t	get the XID of the previous
		remote procedure call
CLSET_XID	uint32_t	set the XID of the next
		remote procedure call
CLGET_PROG	rpcprog_t	get program number
CLSET_PROG	rpcprog_t	set program number

The following operations are valid for connection-oriented transports only:

CLSET_IO_MODE rpciomode_t*    set the IO mode used
 		to send one-way requests. The argument for this operation 
		can be either: 
   - RPC_CL_BLOCKING 	all sending 
		operations block until the underlying transport 
		protocol has accepted requests. If you specify this argument
 		you cannot use flush and getting and setting buffer
 		size is meaningless. 
   - RPC_CL_NONBLOCKING   sending
		operations do not block and return as soon 
		as requests enter the buffer. You can now use non-blocking I/O. 
		The requests in the buffer are then pending. 
		The requests are sent to the server as soon 
		as a two-way request is sent or a flush is 
		done. You are responsible for flushing the 
		buffer. When you choose RPC_CL_NONBLOCKING 
		argument you have a choice of flush modes as 
		specified by 
		CLSET_FLUSH_MODE.
CLGET_IO_MODE rpciomode_t* 		get the current IO mode 
CLSET_FLUSH_MODE rpcflushmode_t* 		set the flush mode. 
The flush mode can only be used in non-blocking I/O mode. 
		The argument can be either of the following: 	
	- RPC_CL_BESTEFFORT_FLUSH: All 
		flushes send requests in the buffer until the 
		transport end-point blocks. If the transport 
		connection is congested, the call returns directly. 	
	- RPC_CL_BLOCKING_FLUSH: Flush blocks 
		until the underlying transport protocol accepts
		all pending requests into the queue. 
CLGET_FLUSH_MODE rpcflushmode_t*  	get the current flush mode. 
CLFLUSH rpcflushmode_t  		flush the pending requests. 
		This command can only be used in non-blocking I/O mode. 
		The flush policy depends on which of the following 
		parameters is specified:
	- RPC_CL_DEFAULT_FLUSH, or NULL:		The flush is done 
	   according to the current flush mode policy 
		(see CLSET_FLUSH_MODE option).
	- RPC_CL_BESTEFFORT_FLUSH:		The flush tries
		to send pending requests without blocking; the call
		returns directly. If the transport connection is
		congested, this call could return without the request
		being sent.
	- RPC_CL_BLOCKING_FLUSH:		The flush sends all pending 
		requests. This call will block until all the requests 
		have been accepted by the transport layer.
CLSET_CONNMAXREC_SIZE int*    set the buffer 
		size. It is not possible to dynamically 
		resize the buffer if it contains data. The 
		default size of the buffer is 16 kilobytes.
CLGET_CONNMAXREC_SIZE int*  		sets the maximum 
		size currently allowed per request. 
CLGET_CURRENT_REC_SIZE int*  		get the current size of 
		the buffer. Use of this 
		command is only recommended when you are in non-blocking 
		I/O mode. The current size of the buffer is always zero 
		when the handle is in blocking mode as the buffer is not 
		used in this mode.

The following operations are valid for connectionless transports only:

CLSET_RETRY_TIMEOUT  struct timeval *    set the retry timeout
CLGET_RETRY_TIMEOUT  struct timeval *    get the retry timeout

The retry timeout is the time that RPC waits for the server to reply before retransmitting the request.

clnt_control() returns TRUE on success and FALSE on failure.

CLIENT *clnt_create(const char * host , const rpcprog_t prognum , const rpcvers_t versnum , const char * nettype );

Generic client creation routine for program prognum and version versnum . host identifies the name of the remote host where the server is located. nettype indicates the class of transport protocol to use. The transports are tried in left to right order in the NETPATH variable or in top to bottom order in the netconfig database.

clnt_create() tries all the transports of the nettype class available from the NETPATH environment variable and the netconfig database, and chooses the first successful one. A default timeout is set and can be modified using clnt_control() . This routine returns NULL if it fails. The clnt_pcreateerror() routine can be used to print the reason for failure.

CLIENT *clnt_create_timed(const char * host , const rpcprog_t prognum , const rpcvers_t versnum , const char * nettype , const struct timeval * timeout );

Generic client creation routine which is similar to clnt_create() but which also has the additional parameter timeout that specifies the maximum amount of time allowed for each transport class tried. In all other respects, the clnt_create_timed() call behaves exactly like the clnt_create() call.

CLIENT *clnt_create_vers(const char * host , const rpcprog_t prognum , rpcvers_t * vers_outp , const rpcvers_t vers_low , const rpcvers_t vers_high , char * nettype );

Generic client creation routine which is similar to clnt_create() but which also checks for the version availability. host identifies the name of the remote host where the server is located. nettype indicates the class transport protocols to be used. If the routine is successful it returns a client handle created for the highest version between vers_low and vers_high that is supported by the server. vers_outp is set to this value. That is, after a successful return vers_low <= *vers_outp <= vers_high . If no version between vers_low and vers_high is supported by the server then the routine fails and returns NULL. A default timeout is set and can be modified using clnt_control() . This routine returns NULL if it fails. The clnt_pcreateerror() routine can be used to print the reason for failure.

Note: clnt_create() returns a valid client handle even if the particular version number supplied to clnt_create() is not registered with the rpcbind service. This mismatch will be discovered by a clnt_call later (see rpc_clnt_calls(3RPC) ). However, clnt_create_vers() does this for you and returns a valid handle only if a version within the range supplied is supported by the server.

CLIENT *clnt_create_vers_timed(const char * host , const rpcprog_t prognum , rpcvers_t * vers_outp , const rpcvers_t vers_low , const rpcvers_t vers_high , char * nettype const struct timeval * timeout );

Generic client creation routine similar to clnt_create_vers() but with the additional parameter timeout , which specifies the maximum amount of time allowed for each transport class tried. In all other respects, the clnt_create_vers_timed() call behaves exactly like the clnt_create_vers() call.

void clnt_destroy(CLIENT * clnt );

A function macro that destroys the client's RPC handle. Destruction usually involves deallocation of private data structures, including clnt itself. Use of clnt is undefined after calling clnt_destroy() . If the RPC library opened the associated file descriptor, or CLSET_FD_CLOSE was set using clnt_control() , the file descriptor will be closed.

The caller should call auth_destroy( clnt -> cl_auth) (before calling clnt_destroy() ) to destroy the associated AUTH structure (see rpc_clnt_auth(3RPC) ).

CLIENT *clnt_dg_create(const int fildes , const struct netbuf * svcaddr , const rpcprog_t prognum , const rpcvers_t versnum , const uint_t sendsz , const uint_t recvsz );

This routine creates an RPC client for the remote program prognum and version versnum ; the client uses a connectionless transport. The remote program is located at address svcaddr . The fildes parameter is an open and bound file descriptor. This routine will resend the call message at intervals of 15 seconds until a response is received or until the call times out. The total time for the call to time out is specified by clnt_call() (see clnt_call() rpc_clnt_calls(3RPC) ). The retry time out and the total time out periods can be changed using clnt_control() . The user may set the size of the send and receive buffers with the parameters sendsz and recvsz ; values of 0 choose suitable defaults. This routine returns NULL if it fails.

void clnt_pcreateerror(const char * s );

Print a message to standard error indicating why a client RPC handle could not be created. The message is prepended with the string s and a colon, and appended with a newline.

CLIENT *clnt_raw_create(const rpcprog_t prognum , const rpcvers_t versnum );

This routine creates an RPC client handle for the remote program prognum and version versnum . The transport used to pass messages to the service is a buffer within the process's address space, so the corresponding RPC server must be located in the same address space (see svc_raw_create() in rpc_svc_create(3RPC) ). This allows simulation of RPC and measurement of RPC overheads, such as round trip times, without any microkernel or networking interference. This routine returns NULL if it fails. clnt_raw_create() should be called after svc_raw_create() .

char *clnt_spcreateerror(const char * s );

Similar to clnt_pcreateerror() , except that it returns a string instead of printing to the standard error. A newline is not appended to the message in this case.

Warning: returns a pointer to a buffer that is overwritten on each call. In multithread applications, this buffer is implemented as thread-specific data.

CLIENT *clnt_tli_create(const int fildes , const struct netconfig * netconf , const struct netbuf * svcaddr , const rpcprog_t prognum , const rpcvers_t versnum , const uint_t sendsz , const uint_t recvsz );

This routine creates an RPC client handle for the remote program prognum and version versnum . The remote program is located at address svcaddr . If svcaddr is NULL and it is connection-oriented, it is assumed that the file descriptor is connected. For connectionless transports, if svcaddr is NULL , RPC_UNKNOWNADDR error is set. fildes is a file descriptor which may be open, bound and connected. If it is RPC_ANYFD , it opens a file descriptor on the transport specified by netconf . If fildes is RPC_ANYFD and netconf is NULL , an RPC_UNKNOWNPROTO error is set. If fildes is unbound, then it will attempt to bind the descriptor. The user may specify the size of the buffers with the parameters sendsz and recvsz ; values of 0 choose suitable defaults. Depending upon the type of the transport (connection-oriented or connectionless), clnt_tli_create() calls appropriate client creation routines. This routine returns NULL if it fails. The clnt_pcreateerror() routine can be used to print the reason for failure. The remote rpcbind service (see rpcbind(1CC) ) is not consulted for the address of the remote service.

CLIENT *clnt_tp_create(const char * host , const rpcprog_t prognum , const rpcvers_t versnum , const struct netconfig * netconf );

Like clnt_create() except clnt_tp_create() tries only one transport specified through netconf .

clnt_tp_create() creates a client handle for the program prognum , the version versnum , and for the transport specified by netconf . Default options are set, which can be changed using clnt_control() calls. The remote rpcbind service on the host host is consulted for the address of the remote service. This routine returns NULL if it fails. The clnt_pcreateerror() routine can be used to print the reason for failure.

CLIENT *clnt_tp_create_timed(const char * host , const rpcprog_t prognum , const rpcvers_t versnum , const struct netconfig * netconf , const struct timeval * timeout );

Like clnt_tp_create() except clnt_tp_create_timed() has the extra parameter timeout which specifies the maximum time allowed for the creation attempt to succeed. In all other respects, the clnt_tp_create_timed() call behaves exactly like the clnt_tp_create() call.

CLIENT *clnt_vc_create(const int fildes , const struct netbuf * svcaddr , const rpcprog_t prognum , const rpcvers_t versnum , const uint_t sendsz , const uint_t recvsz );

This routine creates an RPC client for the remote program prognum and version versnum ; the client uses a connection-oriented transport. The remote program is located at address svcaddr . The parameter fildes is an open and bound file descriptor. The user may specify the size of the send and receive buffers with the parameters sendsz and recvsz ; values of 0 choose suitable defaults. This routine returns NULL if it fails.

The address svcaddr should not be NULL and should point to the actual address of the remote program. clnt_vc_create() does not consult the remote rpcbind service for this information.

struct rpc_createerr rpc_createerr;

A global variable whose value is set by any RPC client handle creation routine that fails. It is used by the routine clnt_pcreateerror() to print the reason for the failure.

In multithreaded applications, rpc_createerr becomes a macro which enables each thread to have its own rpc_createerr .

RESTRICTIONS

For the clnt_tli_create() function, file descriptors passed as arguments must be related to sockets.

ATTRIBUTES

See attributes(5) for descriptions of the following attributes:

ATTRIBUTE TYPE	ATTRIBUTE VALUE
MT-Level	MT-Safe