__pmProcessAddArg - Man Page

#include "pmapi.h"
#include "libpcp.h"

int __pmProcessAddArg(__pmExecCtl_t **handle, const char *arg);
int __pmProcessUnpickArgs(__pmExecCtl_t **handle, const char *command);
int __pmProcessExec(__pmExecCtl_t **handle, int toss, int wait);

cc ... -lpcp

This documentation is intended for internal Performance Co-Pilot (PCP) developer use.

These interfaces are not part of the PCP APIs that are guaranteed to remain fixed across releases, and they may not work, or may provide different semantics at some point in the future.

Within the libraries and applications of the Performance Co-Pilot (PCP) these routines are provide a convenient and safe alternative to system(3) for executing commands in a separate process.

Use __pmProcessAddArg to register the executable and command arguments in order. handle should be set to NULL before the first call to __pmProcessAddArg for a particular command execution and it will be set to an opaque pointer to data structures that are manipulated in __pmProcessAddArg, __pmProcessExec and the related __pmProcessPipe(3) routines.

When called with handle set to NULL arg is treated as the name of the command to be executed and subsequent calls (if any) are for the arguments to that command. The name of the command can be a full pathname, or the name of an executable that can be found on the current $PATH as per the rules of execvp(2) that is used by __pmProcessExec.

__pmProcessUnpickArgs is a convenience wrapper to assist conversion of code that assumes the command is a shell command that has been prepared for use with system(3) or popen(3) in existing code. The arguments in command are picked off one-by-one and used to call __pmProcessAddArg. The parser is simple, as the routine is designed for simple shell command syntax, where arguments are separated by one or more spaces but embedded spaces within an argument are allowed if the argument is enclosed in single or double quotes.  More advanced shell syntax like escape characters and input-output redirection are not recognized.

Once all the command name and arguments have been registered calling __pmProcessExec uses a fork(2) and execvp(2) sequence to execute the command.

The argument toss may be used to assign some or all of the standard I/O streams for the command to /dev/null - specifically toss is either PM_EXEC_TOSS_NONE to keep all I/O streams the same as the parent process, else the bit-wise or of PM_EXEC_TOSS_STDIN and/or PM_EXEC_TOSS_STDOUT and/or PM_EXEC_TOSS_STDERR to reassign stdin, stdout and stderr respectively. PM_EXEC_TOSS_ALL is a convenience macro equivalent to PM_EXEC_TOSS_STDIN | PM_EXEC_TOSS_STDOUT | PM_EXEC_TOSS_STDERR.

The wait argument should be PM_EXEC_WAIT if __pmProcessExec should wait for completion of the command and harvest the exit status, else PM_EXEC_NOWAIT in which case __pmProcessExec returns immediately (effectively running the command in the background).

Nested calling of __pmProcessExec and/or __pmProcessPipe(3) is not allowed.  Once __pmProcessAddArg is called with handle set to NULL to start the registration and execution sequence any attempt to start a second registration sequence will be blocked until the first one is completed by calling __pmProcessExec or __pmProcessPipe(3).

If successful __pmProcessAddArg returns 0.  Other conditions are rare (e.g. memory allocation failure) and are indicated by a return value that can be decoded using pmErrStr(3). When an error does occur __pmProcessAddArg cleans up any allocations made in the current call and uses handle to clean up any allocations made by previous calls so there is no need for the caller to worry about memory leaks, and then (re)sets handle to NULL before returning.

__pmProcessUnpickArgs returns 0 on success.  In the case of an unterminated string, a message is generated and PM_ERR_GENERIC is returned. Other return values less than 0 indicate a more serious error and the value can be decoded using pmErrStr(3). When an error does occur the clean up is similar to __pmProcessAddArg so there is no need for the caller to worry about memory leaks.

The return status from __pmProcessExec is more complicated. If either PM_EXEC_NOWAIT is specified, or the command completes with an exit status of 0, the return value is 0. Return values less than 0 indicate a more serious error and the value can be decoded using pmErrStr(3). If the command was executed, but did not exit with status of 0 then the return value is an encoding of the waitpid(2) status as follows: 2000 if something unknown went wrong, else if 1000 + signal number of the command was killed or stopped by a signal, else the exit status of the command.

execvp(2), fork(2), __pmProcessPipe(3), popen(3), system(3) and waitpid(3).

__pmProcessPipe(3).

The man pages __pmProcessExec(3) and __pmProcessUnpickArgs(3) are aliases of __pmProcessAddArg(3).